Pragmatist Semantics: A Use-Based Approach to Linguistic Representation 9780192874757

Citation preview

Pragmatist Semantics

Pragmatist Semantics A Use-Based Approach to Linguistic Representation JOSÉ L. ZALABARDO

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © José L. Zalabardo 2023 The moral rights of the author have been asserted All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2023935136 ISBN 978–0–19–287475–7 DOI: 10.1093/oso/9780192874757.001.0001 Printed and bound in the UK by Clays Ltd, Elcograf S.p.A. Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

Contents Preface

ix

1. Representational Discourse 1.1 Propositional Representation and Meaning Grounds 1.2 A Version of Representationalism 1.3 Reference Failure and Representation 1.4 Descriptive Semantics and Foundational Semantics 1.5 Conclusion

1 1 3 8 10 13

2. The Open-Question Argument in Ethics 2.1 Open Questions and Synonymy 2.2 The Naturalist Strikes Back 2.3 The Argument Revised 2.4 Moral Twin-Earth 2.5 Moral Motivation 2.6 Moral Intuitionism 2.7 Conclusion

14 14 17 20 28 33 33 35

3. The Open-Question Argument in Semantics 3.1 Truth Ascriptions 3.2 Propositional Attitudes 3.3 Meaning 3.4 Guidance 3.5 Conclusion

37 37 45 49 57 59

4. Some Reactions 4.1 Pragmatist Representationalism

60 60

4.1.1 Ethical Subjectivism 4.1.2 Protagorean Relativism 4.1.3 Interpretivism

4.2 The Flight from Representation 4.2.1 Ethical Non-cognitivism 4.2.2 Deflationism about Truth 4.2.3 Instrumentalism about the Attitudes 4.2.4 Kripke’s Sceptical Solution 4.3 Representation Regained 4.4 Conclusion

61 63 63

68 69 72 74 74 77 81

vi



5. Pragmatist Meaning Grounds 5.1 Assertibility Conditions 5.2 Justification? 5.3 Wittgenstein on Meaning and Use 5.4 Assertion and Acceptance 5.5 Pragmatist Meaning Grounds 5.6 Pragmatism and Its Fellow Travellers 5.6.1 Pragmatist Representationalism vs. Genuine Pragmatism 5.6.2 Pragmatism vs. Non-cognitivism 5.6.3 Rationalist Pragmatism

5.7 The ‘Explanation’ Explanation 5.8 Conclusion

6. Belief and Desire 6.1 Belief, Desire, Behaviour 6.2 The Intentional Stance 6.3 Indeterminacy and Charity 6.4 The Ontogenesis of the Intentional Stance 6.5 The Curse of Knowledge 6.6 The Hybrid Policy and the Meaning Grounds of Belief and Desire Ascriptions 6.7 Other Conditions 6.7.1 6.7.2 6.7.3 6.7.4 6.7.5

Causation Representation Mind Reading Self-Interpretations Linguistic Evidence

6.8 Conclusion

7. Meaning and Truth 7.1 Radical Translation/Interpretation 7.2 Compositionality 7.3 Meaning and Belief 7.4 Charity 7.5 Permutations 7.6 Familiarity 7.7 Reference and Causation 7.8 Projection 7.9 Linguistic Interpretation and Non-linguistic Belief Ascription 7.10 Self-Interpretation 7.11 Consequences of Pragmatism 7.12 Truth Ascriptions 7.13 Representational Discourse 7.14 Wright’s Discipline: Cognitive Command 7.15 Ideal Conditions 7.16 Conclusion

83 83 86 88 91 93 95 95 97 101

104 106

107 107 109 114 115 120 121 124 124 125 125 126 128 129

130 130 132 135 136 137 140 141 143 144 146 147 150 151 158 160 161



vii

8. Harmony and Abstraction 8.1 The Problem of Harmony 8.2 An Idea of Quine’s 8.3 Abstraction 8.4 Two Readings of Abstraction Principles 8.5 Abstraction and Predicate Reference 8.6 The Two-Way Approach 8.7 Some Consequences of the Two-Way Approach 8.8 Abstraction and the Problematic Discourses 8.9 Aufhebung? 8.10 Conclusion

163 163 166 168 170 172 174 175 178 180 181

9. The Primacy of Practice 9.1 Theoretical Science 9.2 Lewisian Humility 9.3 Epistemological Challenges to Lewis’s Argument 9.4 Quidditism and Combinatorialism 9.5 Ramseyan Structuralism 9.6 Other Options 9.7 Pragmatist Meaning Grounds for Theoretical Science 9.8 A Pragmatist Progress 9.9 Pragmatist Meaning Grounds and Genuine Representation 9.10 A Regress Argument 9.11 Tu quoque? 9.12 Conclusion

183 183 185 188 190 193 198 199 201 205 207 210 211

Epilogue: The Meaning Grounds of Meaning-Ground Specifications References Index

213 219 231

Preface The Introduction to a recent collective volume starts with the following sentence: There is a story about the function of large parts of our language, which is, in its broad outlines, intuitive and straightforward. On this view, much of our verbal activity is representational. (Gross, Tebben, and Wiliams 2015: 3) I think the story is indeed intuitive and straightforward—much of our verbal activity presupposes that language has the power to represent things as being a certain way. However, in contemporary philosophy, the intuitive story has come under attack. Challenges to the intuitive story are the subject matter of the collection from which the quote is taken. Some challenges to the intuitive story concern specific regions of discourse that we might be inclined to regard as representational. Challengers contend that, contrary to initial appearances, the target discourse doesn’t actually succeed in representing the world. This kind of challenge comes in two forms. According to one, the target discourse has indeed the function of representing the world, but is incapable of performing this function. According to the other, the representational surface of the discourse disguises the fact that its real function is nonrepresentational. Other challenges to the intuitive story are more wide-ranging—they call into question the very idea that language is anywhere representational. Representing the world, on these views, is not among the things language is capable of doing, or even among the things that language is supposed to do. These challenges to the intuitive story have a wide variety of sources, but I think we can discern a common pattern in many of them: they rest on the assumption that a sentence can represent the world only if its meaning can receive a particular kind of explanation—one that makes reference to semantic relations between the sentence and the bits of the world that the sentence represents. The link between the power to represent the world and explanations of meaning based on languageworld connections is often treated as self-evident. After presenting the intuitive story in the passage quoted above, the authors go on to characterize representational theories of meaning as those views which accept the story, adding that,

x



according to these theories, ‘semantic categories, such as truth and reference, have important explanatory roles to play’ (Gross, Tebben, and Wiliams 2015: 3).¹ With this link in place, the representational character of a discourse is hostage to our ability to provide an explanation of the meaning of its sentences in which semantic notions, such as truth or reference, play a central role. I think this is the main route through which the intuitive story is called into question—a specific discourse, or language as a whole, can’t represent the world, the thought goes, because the requisite explanation of the meanings of the target sentences is not to be had. Discussion of these issues has been particularly intense with respect to some specific discourses. Ethical discourse is one of them. In a familiar version of the problem, it goes roughly like this: You might think that the sentence ‘killing one to save five is morally right’ represents things as being a certain way with an action type, but in order for the sentence to achieve this, our explanation of its meaning has to invoke a connection between the predicate ‘is morally right’ and a property playing the role of its referent—the property that we represent an action as instantiating with a sentence that ascribes the predicate to the action. But there are powerful arguments suggesting that there is no plausible candidate for this job—no property that the predicate could be credibly said to refer to. Another discourse for which the difficulty arises is truth ascriptions. In order for the sentence ‘ “the cat is on the mat” is true’ to represent things as being a certain way with the sentence ‘the cat is on the mat’, the meaning of the former sentence has to be explained in terms of a relation between the predicate ‘is true’ and a property playing the role of its referent, and every candidate for this job exhibits seemingly fatal shortcomings. Similar problems arise for other semantic discourses, including belief ascriptions, like ‘Lorna believes that there’s milk in the fridge’, and meaning ascriptions, like, ‘ “Tisch”, as understood by Kurt, means table’. Difficulties of this kind have been raised for several other ostensibly representational discourses, but my focus in the present book will be on these four—moral discourse plus the three semantic discourses just mentioned—ascriptions of truth, belief, and meaning. In fact, my real focus will be on the last three. I will also consider moral discourse for its historical importance in the development of some of the central ideas in this area, and to highlight the fact that the difficulties under discussion do not arise from the semantic nature of the target discourses, but I won’t consider in much detail the specific features of this case. A major theme of the present book is that these challenges to the intuitive story don’t succeed, because the link they presuppose between representation and

¹ In the abstract to the online edition of the volume we read: ‘Much contemporary thinking about language is animated by the idea that the core function of language is to represent how the world is and that therefore the notion of representation should play a fundamental explanatory role in any explanation of language and language use’ (my italics).



xi

meaning-explanations in terms of language-world connections should be rejected. I’m going to argue that other types of explanation of the meaning of a sentence are perfectly compatible with the thought that the sentence succeeds in representing the world. I will contend that this goes, in particular, for explanations of the meanings of sentences in terms of how they are used. I will use the term pragmatist to refer to explanations of this kind.² My point then is that adopting a pragmatist explanation of the meaning of a sentence doesn’t prevent us from treating the sentence as successfully representing the world. I’m going to apply this approach in the first instance to ethical discourse and the three semantic discourses mentioned above. I will argue that the meanings of their sentences should be explained along pragmatist lines, without thereby giving up on the idea that they represent things as being a certain way. Later on, I will explore other applications of the pragmatist approach, arguing that our grasp of language-world connections across the board might rest on pragmatist foundations. I propose to understand the task of explaining the meaning of a sentence as aiming at specifying its meaning ground—the fact by virtue of which the sentence has the meaning it has. Explanations of meaning in terms of language-world connections can be construed as following a particular strategy for specifying the meaning grounds of declarative sentences. On this strategy, what makes a sentence have the meaning it has is a semantic relation to the state of affairs it represents as obtaining, or to the items whose combination would produce this state of affairs. On this approach, the sentence ‘the cat is on the mat’ has the meaning it has by virtue of a relation to the state of affairs of the cat being on the mat, or to the cat, the mat and the on-top-of relation. I refer to this strategy for specifying the meaning ground of a sentence as the representationalist approach. My proposal is that the threat to the representational status of ethical and semantic discourses rests on the assumption that propositional representation requires representationalist meaning grounds—that a declarative sentence can discharge the function of representing the world only if its meaning ground is representationalist. The problem for the target discourses, on my construal, is that attempts to provide representationalist meaning grounds for their sentences face formidable obstacles. In Chapter 1, I present what I regard as the most plausible account of what the representationalist meaning ground of a declarative sentence would have to look like. In Chapter 2 I develop a version of the open-question argument in support of the conclusion that ethical sentences can’t have representational meaning grounds. On my construal of the argument, the problem it poses for representationalism arises from our intuition concerning the circumstances under which we are prepared to take someone else as meaning by one of their predicates what we

² For the idea of offering a pragmatist explanans for a semantic explanandum, see (Price 2011a).

xii



mean by our ethical predicates. In Chapter 3 I consider how the argument can be deployed to the same effect for ascriptions of truth, meaning, and propositional attitudes. If the difficulties presented in Chapter 2 and Chapter 3 are real, we need to decide how to treat the sentences of the discourses that are affected by them. Chapter 4 is devoted to reviewing some of the strategies that have been proposed for dealing with this situation. First, I consider the idea that the difficulties can be overcome with versions of representationalism that single out the bits of the world from which sentences obtain their meaning in terms of the procedures employed by speakers for regulating their acceptance. I argue that this strategy either fails to solve the problem or produces manifestly implausible characterizations of the meanings of the target sentences. Then I turn to positions that accept the conclusion that the target discourses can’t discharge the function of representing the world because we can’t provide representationalist meaning grounds for their sentences, and react to this outcome by giving up the idea that the function of these discourses is representational. On this approach, the target discourses actually have, or should be construed as having, an alternative nonrepresentational function, that can be discharged by sentences without representational meaning grounds. I end by considering the view that the ascription of a non-representational function to a discourse is perfectly compatible with the view that its sentences succeed in representing the world, once we understand that there are, in fact, no substantive conditions that a declarative sentence needs to satisfy in order to count as discharging the function of representing the world. I argue that the vindication of the representational character of the target discourses achieved by this line of reasoning is unsatisfactory, since it preserves the contrast between sentences that have no function other than representing the world, and those whose primary function is non-representational, although they can be characterized as having the subsidiary function of representing the world. In Chapter 5 I begin to articulate the construal I propose to adopt for the target discourses. I reject the assumption that propositional representation requires representationalist meaning grounds, maintaining, to the contrary, that a sentence with non-representationalist meaning grounds is in principle perfectly capable of discharging the task of representing the world. The non-representationalist meaning grounds that I propose to adopt for the target discourses are pragmatist. On this proposal, the sentences of the target discourses have their meanings by virtue of the fact that they are used in a certain way, more specifically by virtue of the fact that their acceptance is regulated in a certain way. I introduce the proposal by considering in the first instance positions according to which the target sentences obtain their meanings from their conditions of warranted assertibility. I argue that this approach should be modified by eliminating its reference to the epistemic notion of the conditions under which assertion is justified, and by replacing assertion with the mental notion of acceptance. On the resulting approach, the



xiii

meaning ground of a declarative sentence is provided by the procedure actually employed by speakers for regulating its acceptance. This is the general template that I propose to apply to the sentences of the target discourses. I spell out the differences between my proposal and other approaches for which it might be easily mistaken. In Chapter 6 I take on the task of specifying pragmatist meaning grounds for sentences that ascribe beliefs and desires. I take as my starting point Daniel Dennett’s Intentional Stance, construed as a procedure for regulating the acceptance of belief and desire ascriptions. I argue that some aspects of the resulting position are unsatisfactory. In particular, a pragmatist specification of meaning grounds for belief and desire ascriptions cannot make reference, without circularity, to the notion of what an agent ought to believe or desire. I concentrate on the other main aspect of the Intentional Stance—the role that belief and desire ascriptions play in the prediction of behaviour. This role sustains an acceptance procedure for belief and desire ascriptions—to accept or reject them on the basis of the success of the behaviour predictions they generate. My provisional proposal is that belief and desire ascriptions have the meaning they have as a result of our employment of this acceptance procedure for them. This proposal has one important shortcoming: it doesn’t register our preference for the ascription of true beliefs. I argue that this preference can be explained in terms of ideas from developmental psychology concerning the procedure we actually employ for predicting behaviour. I argue that our default is to predict behaviour that would be as a matter of fact (i.e. by our lights) conducive to the promotion of the goals we’ve ascribed to the agent. It’s only when this procedure produces unsatisfactory results that we switch to the more complex method of predicting that the agent will behave in ways that would be conducive to the promotion of its goals if the (false) beliefs we have ascribed to it were true. The default status of the simpler predictive strategy is the source of our reluctance to ascribe false belief. In Chapter 7 I articulate proposals for pragmatist meaning grounds for ascriptions of meaning and truth. For meaning ascriptions (interpretations) I take as my starting point the ideas defended by W. V. O. Quine and Donald Davidson under the label Radical Interpretation (/Translation). I discuss the role that compositionality plays in our meaning ascriptions and I introduce the idea of a doxastic procedure for selecting interpretations. A procedure is doxastic when it is based on the belief ascriptions we are committed to by our interpretation of sentences the speaker accepts. The charity criterion endorsed by Quine and Davidson is a doxastic criterion—it recommends interpretations that result in the ascription of true beliefs. I argue that charity by itself doesn’t generate a plausible characterization of our interpretative procedure. I suggest that our interpretative procedure employs, in addition, a familiarity criterion—a preference for ascribing referents that can be easily defined in terms of our concepts. I then argue that charity should be replaced with another doxastic criterion—projection—that recommends

xiv



interpretations that result in the ascription of the beliefs we would have if we found ourselves in the speaker’s epistemic situation. I contend, in addition, that our interpretative procedure also takes into account non-linguistic evidence bearing on belief ascription—if we have evidence against ascribing a belief to the speaker, we avoid interpretations that result in the ascription of this belief. My proposal is that meaning ascriptions obtain their meaning from the fact that we use this procedure to regulate their acceptance—a combination of the projection and familiarity criteria supplemented, as appropriate, with non-verbal evidence for belief ascription. For truth ascriptions my proposal is that they obtain their meaning from the fact that we regulate their acceptance using a criterion based on the disquotational character of the truth predicate: We accept an ascription of the predicate ‘is true’ to one of our sentences just in case we accept the sentence, and we accept its ascription to someone else’s sentence just in case we have interpreted the sentence as representing a state of affairs that we believe obtains. I then apply the pragmatist approach to ascriptions of representational character—sentences that ascribe to other sentences the function of representing the world. My proposal for the target discourses involves maintaining that their sentences succeed in representing the world even though they have pragmatist (i.e. nonrepresentationalist) meaning grounds. Thus my proposal is openly opposed to the assumption that representation requires representationalist meaning grounds. This assumption is not without support. In Chapter 8 I present what strikes me as the most powerful argument for the assumption. My strategy for defusing this argument is a major ingredient of the position I’m advocating. The argument is based on the intuitive idea that if a sentence represents the world, there has to be a possible state of affairs that the sentence represents as obtaining, and if a predicate is capable of representing the world, there has to be a property that a sentence ascribing the predicate to an object represents the object as instantiating. This is not a reformulation of the assumption that representation requires representationalist meaning grounds. The thought here is that there has to be a state of affairs that a representational sentence represents as obtaining, whether or not this state of affairs plays a role in the meaning ground of the sentence. But there’s a problem with the idea that there is a state of affairs that a sentence with a nonrepresentationalist meaning ground represents as obtaining. The problem is that its connection with the state of affairs it represents has to be a necessary condition for the sentence to have the meaning it has, but its non-representationalist meaning ground has to be a sufficient condition for the sentence to have the meaning it has. These two requirements are incompatible unless its meaning ground is a sufficient condition for the sentence to represent the state of affairs it represents, and it’s hard to see how a non-representationalist meaning ground can satisfy this requirement. I call this the problem of harmony. I regard it as the main source of intuitive support for the assumption that representation requires



xv

representationalist meaning grounds. Chapter 8 develops a strategy for dealing with the problem. My basic idea, for predicates, is that their referents can be identified using abstraction principles based on the synonymy conditions specified by their pragmatist meaning grounds. Similarly, for sentences, the states of affairs they represent as obtaining would be identified by abstraction principles generated by their meaning-grounding acceptance procedures—the state of affairs a sentence represents as obtaining is the state of affairs represented by every sentence whose acceptance is regulated by the relevant procedure. I discuss how this picture can be made compatible with the thought that the states of affairs represented by sentences with representationalist meaning grounds can be identified by explicit definitions. The ideas I develop up to this point sustain a revision of a standard account of linguistic representation, according to which all representational sentences have to have representationalist meaning grounds, consisting in semantic links with states of affairs that can be in principle identified using the language of science. My discussion so far has defended the view that we can have, in addition, representational sentences that don’t satisfy this requirement—sentences that represent the world but have pragmatist meaning grounds. Chapter 9 presents an application of the pragmatist template to representational sentences of the first kind. That model of propositional representation rests on the implicit assumption that the expressions from the language of science with which we specify representationalist meaning grounds have themselves representationalist meaning grounds. However, the idea that theoretical scientific terms have representationalist meaning grounds is problematic. I contend that David Lewis’s argument for Humility shows that if the meaning grounds of theoretical scientific terms are representationalist, we can’t have cognitive access to them, and the same goes, a fortiori, for the expressions whose meaning grounds are specified using these terms. Hence, representationalist meaning grounds in general are beyond our cognitive grasp. I reach this result from a discussion of theoretical terms in science, but I argue that it can also be reached by a more direct route, from the reflection that the representationalist method for specifying meaning grounds is not self-sufficient— its application rests on the use of some other method for the expressions with which representationalist meaning grounds are ultimately specified. I argue that this difficulty can be overcome by adopting a pragmatist account of the meaning grounds of theoretical terms in science. On the resulting picture, our cognitive access to the referents of our terms and the states of affairs represented by our sentences is universally mediated by abstraction principles generated by the procedures that govern the ascription of our terms and the acceptance of our sentences. The thoroughgoing pragmatist picture presented in Chapter 9 still leaves some expressions untouched—those with which we specify the meaning-grounding features of our procedures for ascribing predicates and accepting sentences. In

xvi



the Epilogue I consider briefly what would be involved in applying the pragmatist template to these expressions. I argue that, for the pragmatist, claims concerning which features of the acceptance procedure of a sentence ground its meaning would have to be adjudicated by reference to the acceptance procedures of meaning-ascribing sentences—i.e. the interpretative procedures described in Chapter 7. This goes, in particular, for claims about the meaning-grounding features of our acceptance procedures for meaning ascriptions. My engagement with the ideas presented here can be traced back to the 1990s, when I was trying to articulate a middle-ground position between realism and anti-realism (Zalabardo 1996, 2000b, 2001). That project ran its course and I moved on to other issues, but I remained convinced that its basic insights were correct and important, and merited further development. I saw the opportunity to return to them around ten years ago, when I discovered the work of Huw Price. Price’s approach provided me with a new framework in which to develop my old insights in more fruitful directions. The present volume is the result. I am grateful to Price for this and for his generous support of the project. I am indebted to many other people for conversations, correspondence, advice or feedback. They include: Renée Baillargeon, Tim Button, Justin D’Arms, Josh Gert, Javier González de Prado Salas, David Macarthur, Paolo Mancosu, Lisa Miracchi, Johannes Roessler, Daniel Rothschild, François Schroeter, Laura Schroeter, Mat Simpson, and two excellent referees for OUP. I first presented these ideas, in embryonic form, in my inaugural lecture at UCL in 2015. I have since presented versions of this material at the following venues: University of Oviedo, Universidad Nacional Autónoma de México, University of Valencia, University of Bristol, University of Murcia, University of Sao Paulo, University of Oslo, University of Luxembourg, University of York, Ontario, University of Sydney, the College of William and Mary, the Saul Kripke Center of the City University of New York, the University of Concepción, Chile, and the 24th World Congress of Philosophy, held in Beijing. I am grateful to these audiences. I worked on this book as an Anderson Visiting Fellow at the Philosophy Department of the University of Sydney. I am grateful to the Department, and especially to my host, David Macarthur, for their hospitality. I’ve presented this material in my research seminars at UCL. I am grateful to the students who took part in these sessions. I learnt a lot from their reactions. Chapter 6 uses material from (Zalabardo 2019a), Chapter 7 uses material from (Zalabardo forthcoming-b), and Chapter 9 uses material from (Zalabardo forthcoming-a). I am grateful to the publishers of these pieces for their permission to reproduce this material here.

1 Representational Discourse 1.1 Propositional Representation and Meaning Grounds I’m walking to the tube station in the morning, on my usual route to work. I come across a neighbour walking in the opposite direction. She greets me and, without stopping, utters the words ‘The Piccadilly Line is suspended’. Her words, as I understand them, represent things as being a certain way—they represent a certain underground line as being out of service. I believe that my neighbour understands the sentence the way I do (we speak the same language). I also believe that she believes that things are as this sentence, as she understands it, represents them as being (she is sincere). And I believe that things are likely to be, in this respect, as she believes them to be (she is reliable). So I form the belief that things are as her words, as I understand them, represent them as being—that the Piccadilly line is suspended. I immediately start thinking of alternative ways of getting to work. I decide to take the 134 bus and start walking in the direction of the bus-stop. Words, as understood by speakers, represent the world. Declarative sentences, like my neighbour’s ‘The Piccadilly Line is suspended’ as understood by me, represent things as being a certain way. I’m going to use for this form of linguistic representation the standard label: propositional representation. Propositional representation in language is the subject matter of this book. Later on I will address the question of what we mean when we say of a sentence that it represents things as being a certain way, but for the time being I’m going to rely on our intuitive understanding of the idea, as applied to uncontroversial cases like my neighbour’s sentence.¹ The main questions I want to ask about propositional representation take a different form. I won’t be asking in the first instance what we mean when we say of a sentence that it represents the world, but what makes the sentences that succeed in representing the world have the meanings they have. Questions of this kind are based on the assumption that the meaning of a linguistic expression is not a fundamental property of it—that a linguistic expression has the meaning it has by virtue of something else being the case about it. This strikes me as a plausible assumption and it will govern my discussion here. On this assumption, the ¹ We will also discuss (Chapter 6) what we mean when we say of someone that they believe that things are thus and so.

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0001

2

 

sentence ‘The Piccadilly Line is suspended’, as I understand it, has the meaning it has by virtue of the obtaining of some other fact about it, and about me and possibly my environment. I’m going to refer to the fact that plays this role as the meaning ground of the sentence. I’m going to approach the phenomenon of propositional representation through the task of identifying the meaning grounds of the sentences that achieve it. At this point we face complex questions concerning how a fact would have to be related to the meaning of a sentence in order for the fact to qualify as the meaning ground of the sentence. I’m not going to provide a full account of this relation, but I want to highlight a necessary condition for a fact to count as a meaning ground.² We can think of a fact about a sentence as a property that the sentence instantiates or a condition it satisfies. If a property ϕ that a sentence S instantiates is the meaning ground of S, then S has the meaning it has by virtue of its instantiation of ϕ. I’m going to assume that for this to be the case, instantiating ϕ has to be a sufficient condition for a sentence to have the meaning S has, i.e. for every sentence X, if X instantiates ϕ then X has the same meaning as S. It is natural to assume that the converse doesn’t hold in general—that there may be different properties whose presence in a sentence would make the sentence have the same meaning as S. I think this is right, but we can turn the instantiation of ϕ into a necessary condition for having the same meaning as S by taking as ϕ the disjunction of all the properties that can confer on a sentence the meaning of S.³ It will simplify matters to think of meaning grounds in these terms. Then the claim that ϕ is the meaning ground of S will have the following consequence: For every sentence X, X has the same meaning as S if and only if ϕ(X). I’m going to assume that this is a necessary condition for ϕ to be the meaning ground of S. The main plot of the book is the contest between two strategies for specifying the meaning ground of a declarative sentence, for which I’ll use the labels representationalism and pragmatism. According to representationalism, the meaning ground of a declarative sentence consists in relations between the sentence and the bits of the world that the sentence represents.⁴ According to pragmatism, the meaning ground of a declarative sentence consists in features of the way the sentence is used. Pragmatism will be our hero; representationalism, our villain. I’m going to argue that the pragmatist strategy can be

² For the grounding relation in general, see (Fine 2012). The notion of the meaning ground of a declarative sentence that I’m using here is very similar to the notion in (Fine 2001). ³ This move might be problematic if the list of properties that can confer on a sentence the meaning of S is open-ended, but for the cases that we will focus on this problem won’t arise. ⁴ The term is used in this way in (Williams 2013).

 

3

employed in cases that might have seemed to be the exclusive preserve of the representationalist strategy. The obstacle to the adoption of pragmatism that will focus most of my attention is the impression that a pragmatist account of the meaning ground of a declarative sentence is incompatible with the claim that the sentence successfully discharges the function of representing things as being a certain way. Not all declarative sentences have this function, and there’s generally no major resistance to the view that, when a declarative sentence has some other function instead, its meaning ground can be specified in pragmatist terms. However, for sentences that have the function of representing things as being a certain way, and succeed in discharging this function, the representationalist account is widely seen as mandatory. Showing that this assumption can be rejected will be one of the main components of my defence of pragmatism. I’m going to argue that we can ascribe a pragmatist meaning ground to a declarative sentence without abandoning the claim that the sentence successfully discharges the function of representing things as being a certain way. Some declarative sentences represent the world even though they obtain their meaning from features of the way they are used.

1.2 A Version of Representationalism My first goal in this chapter is to introduce the representationalist strategy for specifying the meaning grounds of declarative sentences, and the general philosophical outlook that results from the assumption that a declarative sentence can succeed in representing the world only if its meaning ground can be specified in representationalist terms. I’m going to focus on what I regard as a particularly plausible version of the representationalist strategy. My claim will be that if this version of the representationalist model can’t handle the meaning grounds of a given discourse, then the discourse can’t have representationalist meaning grounds. I’m going to provide some motivation for the choices that result in the version I’m going to focus on, but my goal is not to offer a full defence of this over other possible ways of implementing the representationalist strategy. The choice would be relevant for my purposes only if other versions of representationalism could overcome the difficulties that I’m going to raise for the one I’ll focus on. I’m going to argue that this is not the case—the objections I’m going to raise against my chosen version of representationalism can be overcome only by abandoning the representationalist strategy altogether. Consider the sentence ‘Fido barks’. This sentence, as I understand it, has meaning, and it is part of its meaning that it has the function of representing things as being a certain way. It follows from our assumptions that the sentence has to have a meaning ground—there’s got to be a fact by virtue of which the sentence has the meaning it has. What is this fact? According to the

4

 

representationalist strategy, this fact has to be found among relations between the sentence and the bits of the world that it represents. What the representationalist needs to provide is a specific account of how the meaning of the sentence arises from these language-world relations. One particularly straightforward option here is to maintain that the meaning of the sentence arises from a relation that the sentence bears to a fact—the fact that Fido barks. The world contains facts, and Fido’s barking is one of them. There’s no immediate reason why a sentence could not bear a relation to this fact and obtain its meaning as a result of this relation. This is only the skeleton of an account of the meaning ground of ‘Fido barks’. A full account would have to specify the particular sentence-fact relation from which the sentence obtains its meaning. However, there’s no point in searching for a satisfactory account of this relation. The meaning of declarative sentences cannot arise from a relation between sentences and facts. The reason is simple. A sentence can represent things as being the way they are, as ‘Fido barks’ does. For these sentences, the proposal might conceivably work. But a sentence can also represent things as being the way they are not, as, e.g., ‘Fido meows’ does. For these sentences, the proposal is a non-starter. ‘Fido meows’ would have to obtain its meaning from a relation to the fact that Fido meows, but the world contains no such fact. In general, the world doesn’t contain facts to which false sentences bear the relation that ‘Fido barks’ bears to Fido’s barking. The account of meaning grounds as a relation between sentences and facts doesn’t work for false sentences.⁵ One strategy for overcoming this obstacle consists in expanding our ontology with the postulation of a category of items to which false sentences are related in the same way in which true sentences are related to the facts they represent as obtaining. This is the strategy that Bertrand Russell adopted in his dual-relation theory of judgement (Russell 1903: 363; 1907).⁶ He referred to the items that would play this role as objective non-facts, and he used the term proposition to refer to both facts and objective non-facts. On this proposal, a sentence, true or false, has the meaning it has by virtue of a relation between the sentence and a proposition. ‘Fido barks’ means what it means as a result of a relation to the true proposition (i.e. the fact) of Fido’s barking. Likewise, ‘Fido meows’ means what it means as a result of a relation to the false proposition (i.e. the objective non-fact) of Fido’s meowing. As Russell himself soon came to accept, this proposal faces important difficulties (Russell 1910, 1912). One major problem is the implausibility of postulating ⁵ When we reach the problem through this route, it is natural to locate the source of the difficulty in the need to make room for false representations. This is the way in which Russell and early Wittgenstein saw the problem, but the thought is already present in Plato. See (Plato 1963: 236d–237a). ⁶ See (Zalabardo 2015: ch. 1), for a presentation of Russell’s ideas on this point.

 

5

an objective non-fact for every false declarative sentence anyone might conceivably understand. A related problem arises from the need for our account of the meaning ground of a sentence to be independent of its truth value—the thought that what makes a sentence mean what it means is independent of whether things are as the sentence represents them as being. What we want to avoid is having two separate accounts of the meaning grounds of sentences—one for true sentences and one for false sentences. Hence, if true sentences obtained their meaning from a relation to a fact, and false sentences obtained their meaning from a relation to an objective non-fact, facts and objective non-facts would have to have the same ontological status. Fido’s meowing would have to be the same kind of thing as Fido’s barking, and what makes the latter, but not the former, a fact would have to be understood in terms of a fundamental, irreducible property that is instantiated by some propositions and not by others. Another option one could take at this point rests on a different piece of ontological profligacy—the modal realist view that non-actual possible worlds are just as real as the actual world (Lewis 1986). On this picture, reality includes possible worlds in which Fido meows, which are just as real as those, including the actual world, in which he doesn’t. We can use these resources to provide a representationalist account of the meaning grounds of declarative sentences. On this account, a declarative sentence has the meaning it has by virtue of a relation between the sentence and a class of possible worlds. And a sentence represents things as being a certain way by representing the actual world as contained in the class of possible worlds from which the sentence obtains its meaning.⁷ Thus, both ‘Fido barks’ and ‘Fido meows’ obtain their meaning from a relation between the sentence and a class of possible worlds, but whereas the class to which the former is related includes the actual world, the class to which the latter is related doesn’t. This proposal will not hold any appeal for those who are not prepared to accept a modal realist ontology, and even for modal realists the proposal faces important difficulties. A modal realist would not have any problem accepting that the meaning of a declarative sentence connects the sentence with a class of possible worlds—those worlds in which things are as the sentence represents them as being. In fact, even on an actualist construal of possible-world talk this claim would not face any major difficulties. Notice however that what this claim asserts is that the meaning of a declarative sentence will generate a relation between the sentence and a class of possible worlds. This is the reverse of the order of priority

⁷ The idea of specifying the semantic properties of sentences in terms of sets of possible worlds was developed by Robert Stalnaker (Stalnaker 1984). See also (Lewis 1970). Notice, though, that Stalnaker is not proposing that sentences obtain their meanings from a relation to sets of possible worlds. His idea is that by pairing sentences with sets of possible worlds we can provide a perspicuous representation of some features of sentence meanings in terms of features of sets of possible worlds, in the same way in which we can represent, say, features of people’s heights in terms of features of numbers, by pairing each person with a number representing his or her height.

6

 

required by the proposal under discussion. According to this proposal, a relation between a sentence and a class of possible worlds is what makes the sentence have the meaning it has. The problem with this is that it is hard to see how the class of possible worlds to which a sentence is related could be singled out without having already determined how the sentence represents things as being. If we can assume that ‘Fido barks’ represents Fido as barking, we can unproblematically pair the sentence with the class of possible worlds in which Fido barks, but without making any assumptions about how the sentence represents things as being there’s no obvious way of deciding, for a given possible world, whether it is contained in the class of worlds from which the sentence obtains its meaning. The issue is, then, the relative priority of the way a sentence represents things as being and the class of possible worlds in which things are as the sentence represents them as being. On the proposal under discussion, the class of possible worlds has to be treated as the primary notion, with the way the sentence represents things as being defined in terms of this class—as what the worlds in the class have in common. I’m arguing that this takes things the wrong way round. We should treat the way the sentence represents things as being as the primary notion, and define the class of possible worlds singled out by the sentence as containing the worlds in which things are like that. Many representationalists take a different approach. Its starting point is a simple piece of metaphysical analysis: facts are produced by the combination of more fundamental items. Thus, for example, the fact of Fido’s barking consists in the instantiation of a property, barking, by an individual, Fido. If Fido meowed, this fact would consist in the instantiation by the individual Fido of the property of meowing. We can exploit this idea in an account of the language-world relation from which a declarative sentence obtains its meaning. On this account, a declarative sentence has the meaning it has as a result of a relation between the sentence and the items that would have to be combined with one another in order for things to be as the sentence represents them as being.⁸ ‘Fido barks’ obtains its meaning from a relation between the sentence, on the one hand, and the individual, Fido, and the property of barking, on the other. Likewise, ‘Fido meows’ obtains its meaning from a relation to the individual, Fido, and the property of meowing. This is the basic idea of Russell’s multiple-relation theory of judgement (Russell 1910, 1912, 1984). This account of the meaning grounds of declarative sentences provides us with a strategy for explaining how they represent the world. In simple cases like our examples, the account is relatively straightforward. ‘Fido meows’ represents the world as containing the fact that would result from the combination of the items ⁸ As we’ll see in Chapter 7, this account of meaning grounds can also be endorsed by those who reject the metaphysical picture on which it’s based in this presentation.

 

7

in the world with which the sentence is related—the individual, Fido, and the property of meowing. Other cases require a more sophisticated treatment. In Russell’s famous example, the sentence ‘Desdemona loves Casio’ can’t be said to represent the world as containing the fact that would result from the combination of the items in the world to which the sentence is related. Two different facts would answer to this description—Desdemona loving Casio and Casio loving Desdemona—and the sentence represents only one of them as obtaining. A more basic issue affects even the simpler cases: the meaning of a declarative sentence would have to determine not only which items in the world the sentence represents as combined with one another but also how the sentence represents these items as combined—what mode of combination the sentence represents the items as exemplifying. The meaning of ‘Fido meows’ has to entail not only that the sentence represents Fido and the property of meowing as combined with one another but also that it represents these items as combined in a specific way—i.e. the former as instantiating the latter.⁹ This move is usually combined with the idea that linguistic representation is compositional: the representational properties of a sentence arise from the representational properties of the terms that figure in it. The result is a picture according to which a sentence represents the world by virtue of a relation between the terms that figure in it and items in the world that the sentence represents as combined with one another. ‘Fido meows’ represents Fido as meowing as a result of a pairing between the singular term ‘Fido’ and the individual Fido, and the predicate ‘meows’ and the property of meowing. I’m going to use the term reference for this relation between linguistic terms and worldly items as a result of which, on this picture, sentences come to represent the world. I will think of it as a function and designate the image of a term under the reference function as the referent of the term.¹⁰ This is the version of representationalism on which I’m going to focus: a declarative sentence represents the world as a result of referential links between the terms that figure in it and items in the world that the sentence represents as combined with one another. Notice that this characterization of the view leaves many important questions open, and as a result substantially different views will count as versions of the position under discussion. One crucial point on which the characterization is silent is the nature of the relation pairing terms with their referents. In traditional versions of the approach, the term-referent link is mediated by a phenomenal item—a concept or idea. In contemporary naturalistic versions the link goes through items in the speaker’s nervous system, as ⁹ In (Russell 1984), Russell tried to address this issue by appeal to forms. See (Zalabardo 2015: 23–36) for an interpretation of Russell’s strategy on this point. ¹⁰ This model doesn’t register the difference in meaning between a predicate, e.g., ‘barks’, and a singular term referring to the property that the predicate ascribes, e.g., ‘barking’, as in ‘barking deters intruders’. I’m not going to discuss this issue here.

8

 

characterized by neuroscience or cognitive science (Fodor 1987). Some versions include an additional intermediary between term and referent besides the speaker’s mental conception—the items that Gottlob Frege called senses or modes of presentation (Frege 1980b). This move has familiar advantages. It enables us to explain that the sentences ‘Superman flies’ and ‘Clark Kent flies’, as understood by Lois Lane, have different meanings, even though both sentences represent the same individual as instantiating the property of flying. Other versions bypass intermediaries of any kind, treating reference as a direct relation between the term and the item that plays the role of its referent, as in causal-historical accounts that connect a term through a chain of communication to an initial act of baptism (Kripke 1980). None of this will be directly relevant to the issues I want to discuss, and in what follows I will make no assumptions about the nature of the referential link postulated by the version of representationalism on which I’m going to focus.¹¹

1.3 Reference Failure and Representation The representationalist strategy for specifying the meaning grounds of linguistic expressions is not universally applicable. There are many linguistic expressions that don’t seem to obtain their meaning from semantic relations to the world. Expressions such as ‘Come on, England!’, ‘Cheers’, ‘Hello’, ‘Thank you’, or ‘What a shame’ don’t seem to obtain their meanings along the lines of the representationalist model. However, it is widely assumed that the representationalist model is mandatory for sentences that perform the function of representing things as being a certain way. I’m going to refer to this as the RR assumption: RR

A sentence that performs the function of representing things as being a certain way must have a representationalist meaning ground.

Representation, according to RR, requires representationalism. One of my central goals in this book is to reject RR—to show that a sentence can perform the task of representing the world even if its meaning ground doesn’t

¹¹ If we wanted to avoid using properties as the semantic values of predicates, we could pursue any of the standard alternatives to Platonism. One possibility would be to pair predicates with concepts, as mental items that have their instantiation conditions intrinsically, not as a result of their association with properties. Another approach would pair each predicate with a set—the set of particulars that satisfy the predicate. A third approach would be to construe the semantic value of predicates in terms not of a function pairing each predicate with the item playing the role of its referent (be it a property, a concept, or a set) but of a relation pairing each predicate directly with the particulars that satisfy it. I’m not going to discuss these approaches, but the difficulties that I’m going to raise for representationalism will not be solved by adopting them.

 

9

follow the representationalist template. But for now I want to explore some important consequences of RR. The fact that we treat a sentence as representing the world doesn’t guarantee that the sentence sustains the links with the world that are required, on the representationalist model, for the sentence to succeed in performing this function. On the version of representationalism on which I’m focusing, a sentence with the function of representing the world will fail to perform this function if the terms that occur in it fail to have referents. This is in line with our intuitions. We believe that empty terms cannot figure as subjects or predicates in sentences that represent the world.¹² Notice that in order for a sentence to represent the world it’s not enough that the speaker believes that its terms, as she understands them, have referents. Speakers can be wrong about this—a term, as understood by a speaker, might have no referent even if the speaker believes it has one. And a sentence in which such a term figures will fail to represent the world even if the speaker is convinced that it succeeds in doing so. Take the singular term ‘Vulcan’, introduced by Urbain Le Verrier in the nineteenth century to refer to a postulated planet whose gravitational force he used to explain some peculiarities in the orbit of Mercury. Le Verrier believed that the sentence ‘Vulcan has craters’ represented things as being a certain way—the planet he thought ‘Vulcan’ referred to as instantiating the property of having craters. However, this planet turned out not to exist. Hence, contrary to what Le Verrier thought, the term ‘Vulcan’, as he understood it, had no referent, and his sentences with this term as subject failed to represent things as being a certain way. We find a similar phenomenon with predicates. Take the predicate ‘is phlogisticated’,¹³ used by the eighteenth-century chemist Georg Stahl, among others, to refer to the property of containing a substance he thought was present in combustible bodies and dissipated in air during combustion. Stahl would have believed that the sentence ‘this fluid is phlogisticated’ represented things as being a certain way—a fluid as instantiating the property that ‘is phlogisticated’ refers to. However, phlogiston theory turned out to be wrong. There isn’t a substance present in combustible bodies that is lost in combustion—there is no such thing as phlogiston and there is no property that ‘is phlogisticated’ could be plausibly said to refer to. Hence, contrary to what Stahl thought, his sentence ‘this fluid is phlogisticated’ didn’t represent things as being a certain way, because there isn’t a property that the sentence represents as instantiated by the fluid.¹⁴

¹² I’m not concerned here with the specific problems posed by the interpretation of fictional discourse. ¹³ Or rather ‘ist phlogistisiert’, but let’s pretend Stahl spoke English. ¹⁴ Notice that saying that a sentence fails to represent things as being a certain way is in principle compatible with saying that the sentence has a truth value. We might want to say that the sentence ‘this fluid is phlogisticated’ is false, so long as we understand that our reason for saying this is not that things are not as the sentence represents them as being, but that there isn’t a way the sentence represents things as being.

10

 

We might try to save the representational character of these sentences by ascribing to them a logical form that doesn’t correspond to their surface grammar. For singular terms, this is achieved by treating them as synonymous with definite descriptions, and using something like Russell’s theory of descriptions to characterize the logical form of the sentences in which they figure as subjects (Russell 1905). And sentences with empty predicates could receive a similar treatment. However, there are good reasons for thinking that, even if this strategy works for some terms, there are many cases in which it fails.¹⁵ In these cases, the subjectpredicate structure has to be taken at face value, and reference failure entails that the sentence fails to represent. The threat of reference failure has played an important role in debates concerning the analysis of philosophically important discourses, including the ones we are focusing on. According to RR, a sentence succeeds in representing the world only if it has a representationalist meaning ground. Having a representationalist meaning ground, on the version of the model on which I’m focusing, requires that the terms that occur in the sentence have referents. If a sentence doesn’t satisfy this requirement, it doesn’t have a representationalist meaning ground and, according to RR, it cannot discharge the task of representing the world. Several familiar philosophical arguments are naturally construed as aiming to show that a certain declarative discourse exemplifies this situation—its sentences appear to represent the world, but they don’t, because the central terms in the discourse fail to refer. There’s a wide range of discourses for which this kind of argument has been deployed. In the present book our primary focus will be on semantic discourses—on ascriptions of meaning, truth, and propositional attitudes. I will also consider ethical discourse, because some of the positions and arguments that we will discuss were originally presented in this connection, but I’m not going to look into the specific problems raised by this discourse in any detail. Readers with an interest in meta-ethics will have to adjust their expectations accordingly. Discussing ethical discourse alongside the semantic discourses that will be my primary focus will also highlight the fact that my concerns with the semantics of these discourses are in principle unrelated to the fact that they have a semantic subject matter, although I will also be interested, towards the end, in the consequences of this fact.

1.4 Descriptive Semantics and Foundational Semantics Many philosophers of language see the discipline as organized around a contrast between two levels of enquiry: descriptive semantics and foundational semantics

¹⁵ On this point, see Paul Boghossian’s argument that the standard externalist account of naturalkind terms is incompatible with a descriptivist construal of empty natural-kind terms (Boghossian 1997).

 

11

(also known as meta-semantics). Robert Stalnaker provides a canonical characterization of the distinction. A descriptive-semantic theory, Stalnaker writes, is a theory that says what the semantics for the language is without saying what it is about the practice of using that language that explains why that semantics is the right one. A descriptive-semantic theory assigns semantic values to the expressions of the language, and explains how the semantic values of the complex expressions are a function of the semantic values of their parts. (Stalnaker 1997: 535)

Foundational semantics, by contrast, asks questions about what the facts are that give expressions their semantic values, or more generally, about what makes it the case that the language spoken by a particular individual or community has a particular descriptive semantics. (Stalnaker 1997: 535)¹⁶

It’s not easy to see how the question of the meaning ground of a linguistic expression can be located in this framework. The representationalist approach, in the version on which I’m concentrating, sounds like a view on the nature of the semantic values that representational predicates need to be assigned—they have to be properties. This suggests that a rival to the representationalist approach would have to put forward an alternative proposal on this point—on the kind of item that can play the role of the semantic value of a predicate. But the pragmatist approach doesn’t aim to do that. The way a predicate is used is not a plausible candidate for the role of its semantic value. The way a predicate is used is more naturally invoked for explaining how its semantic value is singled out. But then we seem forced to conclude that the representationalist approach and the pragmatist approach are not really in competition with one another, since the former is a view in descriptive semantics, while the latter concerns foundational semantics. In fact, the two views seem perfectly compatible. We can plausibly maintain that the semantic value of a predicate is a property and that the property that plays this role is singled out by how the predicate is used.¹⁷ The question concerning the meaning ground of a predicate is intended to fall, broadly speaking, on the foundational side of Stalnaker’s divide, but it doesn’t quite follow the template that Stalnaker proposes for questions in this area. Stalnaker presents foundational semantics as following an agenda dictated by

¹⁶ See also (Kaplan 1989a: 573–4). ¹⁷ We seem to have made the mistake that David Lewis warns us against concerning the distinction between descriptive and foundational semantics: ‘Only confusion comes of mixing these two topics’ (Lewis 1970: 19).

12

 

descriptive semantics. If descriptive semantics tells us, e.g., that predicates take properties as semantic values, then the goal of foundational semantics, according to Stalnaker, is to identify the facts that determine which property plays this role for a given predicate.¹⁸ Questions concerning the meaning grounds of linguistic expressions, as I propose to discuss them, do not aim at identifying the facts that determine the pairings of expressions with the items they need to have as semantic values in a descriptive-semantic theory. They aim at identifying the facts that make an expression have the meaning it has, without making any assumptions as to what kind of item, if any, these facts need to pair with the expression as its semantic value. We will be looking for features that we think of as necessary and sufficient for an arbitrary expression to have the same meaning as the target expression, taking no notice, in the first instance, of the consequences that our choice of meaning grounds might have for descriptive semantics. Now, in some cases, we will find that the feature of a target predicate that’s necessary and sufficient for the predicate to have the meaning it has is the fact that it refers to a specific property—that a predicate has the same meaning as the target predicate just in case it has that property as its referent. Predicates that behave like this have representationalist meaning grounds. However, representationalism doesn’t carry a commitment to the view that the pairing of a predicate with its referent is a primitive fact. There will be other facts about the predicate, including, possibly, the way the predicate is used, that will single out this property as playing the role of referent for the target predicate. But for predicates with representationalist meaning grounds, it’s the identity of the referent, and not the way it’s singled out, that makes the predicate have the meaning it has. Predicates in which the same referent is singled out by other facts will have the same meaning as the target predicate, and predicates in which the same facts single out a different referent will have a different meaning. We might also find that for some predicates no property can play this role— that there is no property of which we can plausibly say that referring to it is necessary and sufficient for an arbitrary predicate to have the same meaning as the target predicate. Predicates that behave like this don’t have representationalist meaning grounds. What makes them have the meaning they have must be a different fact about them. In some of these cases, we might find that the way the predicate is used is what makes it have the meaning it has—that being used in that way is necessary and sufficient for an arbitrary predicate to have the same meaning as the target predicate. Predicates that behave like this have pragmatist meaning grounds. ¹⁸ On this point, see (Yalcin 2018: 352): ‘insofar as “foundational semantics” has a clear agenda, it would be constrained by “descriptive semantics” inasmuch as the latter states the semantic value facts that the former project is charged with grounding.’

 

13

In the next two chapters I’m going to argue that the central predicates of the discourses I’m interested in can’t have representationalist meaning grounds. Then, in Chapters 4–7 I will articulate and defend a pragmatist account of the meaning grounds of these predicates—what makes them have the meaning they have is features of the way they are used. I will argue that ascribing pragmatist meaning grounds to these predicates is perfectly compatible with the view that they succeed in performing the function of representing the world. This claim might seem to clash with the intuition that in order for a predicate to perform the function of representing the world it needs to have a referent—there has to be a property that we represent an object as instantiating when we ascribe the predicate to it. In Chapter 8 I will present a strategy for removing this obstacle, by showing that predicates with pragmatist meaning grounds can also refer to properties. Hence, in the end, the choice between the representationalist approach and the pragmatist approach to the meaning ground of a predicate will have no consequences at the level of descriptive semantics.

1.5 Conclusion The subject matter of the present book is the meaning grounds of sentences that discharge the function of representing the world. In this chapter I have introduced the issue and two main approaches to the task of specifying the meaning grounds of representational sentences—representationalism and pragmatism. Representationalism specifies the meaning ground of a sentence in terms of semantic relations between the sentence and the bits of the world it represents. Pragmatism specifies the meaning ground of a sentence in terms of how it’s used. I have then presented what I regard as the most plausible version of the representationalist strategy, which treats as the basic notion of its specification of meaning grounds a reference relation pairing terms with items that sentences represent as combined with one another—individuals, properties, etc. I have presented the assumption (RR) that the sentences of a discourse can succeed in performing the function of representing the world only if they have representationalist meaning grounds—specifically, if the central predicates of the discourse have referents. In the next two chapters I’m going to develop a line of reasoning for the conclusion that the discourses that interest us can’t satisfy this condition—no property can play the role of referent for their central predicates. In Chapter 2, I’m going to present the argument in connection with ethical discourse, for which it was originally developed. In Chapter 3, I will consider the application of this line of reasoning to the three semantic discourses that interest us.

2 The Open-Question Argument in Ethics In the previous chapter I argued that on the most promising version of the representationalist approach, the language-world connections from which sentences obtain their meanings include a function pairing each predicate with the property playing the role of its referent. In the present chapter I’m going to outline a powerful line of reasoning for the conclusion that for ethical predicates there can’t be properties playing this role. In the next chapter I will consider how the argument can be adapted for drawing a similar conclusion for the semantic discourses we are interested in.

2.1 Open Questions and Synonymy Consider the sentence ‘killing one to save five is morally right’. It is a subjectpredicate sentence that seems to many of us to have the function of representing the world. If we take its surface grammar at face value, this requires, according to representationalism, that the terms that figure in the sentence have referents—that ‘killing one to save five’ refers to a specific action type and ‘is morally right’ refers to a property that an action may or may not instantiate. The problems we are going to discuss concern the referent of the predicate: which property could play this role? Many philosophers believe that the range of candidates is restricted to natural properties—very roughly, and for our immediate purposes, properties whose instantiation by an action can be ascertained by standard empirical means. Let’s use as an example a simplistic utilitarian proposal, according to which ‘is morally right’ refers to the property of maximizing overall utility. On this proposal, the sentence represents the action of killing one to save five as maximizing overall utility. There’s of course huge scope for debate on whether the role should be assigned to this property or to some other natural property. I’m not going to be concerned with this. I’m going to consider objections to the proposal that would undermine, if successful, any view according to which ‘is morally right’ refers to a natural property. One prominent objection to naturalistic accounts of the referents of ethical terms is what’s known as the open-question argument. The argument was originally presented by G. E. Moore against the possibility of defining or analysing good. Moore writes:

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0002

 -   

15

The hypothesis that disagreement about the meaning of good is disagreement with regard to the correct analysis of a given whole, may be most plainly seen to be incorrect by consideration of the fact that, whatever definition be offered, it may be always asked, with significance, of the complex so defined, whether it is itself good. (Moore 1903: 15)

Moore is happy to accept that there may be necessary and sufficient conditions for something being good. What he doesn’t accept is that these conditions could be taken as defining or analysing good. The reason that the quoted passage gives for this is that we can meaningfully ask whether these conditions are good. However, the premise of the argument can be profitably reformulated as the claim that we can meaningfully ask whether satisfaction of these conditions is sufficient or, presumably, necessary for something being good, and hence that we are prepared to countenance a negative answer. In Moore’s presentation, the target of the argument is a view concerning the intra-linguistic relationship between an ethical term and a non-ethical (e.g. natural) term—that the two terms are synonymous, but it can be easily adapted to target the view we are interested in, concerning a language-world relation between an ethical predicate and a natural property—that the property is the referent of the predicate. The central thought of the argument is best appreciated if we assume that the actions that we take to satisfy ‘is morally right’ are precisely those of which we believe that they maximize overall utility. On this assumption, for us, the question whether an action that satisfies ‘is morally right’ maximizes overall utility, or whether an action that doesn’t satisfy ‘is morally right’ fails to maximize overall utility, will always receive an affirmative answer. Nevertheless, the Moorean thought goes, the question is always open—a negative answer is, in some sense, never ruled out. I propose to understand this thought as concerning the conditions under which we count someone as meaning by ‘is morally right’ what we mean by this predicate. The thought would then be that someone who answers these questions in the negative can still count as meaning by ‘is morally right’ what we mean by it. If someone takes an action to satisfy ‘is morally right’ even though they believe that the action doesn’t maximize overall utility, or vice versa, their ascriptions of the predicate can still count as representing actions in the same way in which our own ascriptions of the predicate represent them. I’m going to refer to this thought as the Moorean premise. The Moorean premise can be employed as the basis of an argument against the claim that ‘is morally right’, as understood by us, refers to the property of maximizing overall utility. The argument can be articulated as follows: 1

Someone can mean by ‘is morally right’ what we mean by it even though they take an action to satisfy ‘is morally right’ but believe that the action doesn’t maximize overall utility, or take an action not to satisfy ‘is morally right’ but believe that the action maximizes overall utility. (The Moorean premise)

16

 

2

If someone takes an action to satisfy ‘is morally right’ but believes that the action doesn’t maximize overall utility, or takes an action not to satisfy ‘is morally right’ but believes that the action maximizes overall utility, then ‘is morally right’, as understood by this speaker, doesn’t refer to the property of maximizing overall utility. (Premise)

3

Someone can mean by ‘is morally right’ what we mean by it even though the predicate, as understood by them, doesn’t refer to the property of maximizing overall utility. (from 1 and 2)

4

Someone can’t mean by ‘is morally right’ what we mean by it if the predicate has different referents as understood by them and as understood by us. (Premise)

Therefore: 5

‘is morally right’, as we understand it, doesn’t refer to the property of maximizing overall utility. (from 3 and 4)

Is this argument sound? It is unquestionably valid. Hence resisting the conclusion would require taking issue with at least one of the premises. I think the Moorean premise rests on firm ground. We have a strong intuition that even if we take the instantiation of a natural property as necessary and sufficient for the satisfaction of ‘is morally right’ by an action, acceptance of this point by other speakers, or by ourselves at a different time, is not required for meaning by the predicate what we mean by it. We will of course take someone who rejects this predicate-property connection to be wrong—but the point is that we may take them to be expressing views about the same subject matter that we seek to represent with our ascriptions of the predicate. We could take the difference between them and us as a factual disagreement, rather than as a difference in the meaning that each of us attaches to the predicate. Notice that the claim is not that the intuition that supports the Moorean premise is infallible.¹ We need to be open to the possibility of rejecting the intuition if it is in conflict with the all-things-considered best account of the semantics of ‘is morally right’. The claim is simply that, other things being equal, an account of the semantics of the predicate that satisfies the intuition is to be preferred to one that doesn’t. Premise 4 is also fairly uncontroversial. There might be concepts of meaning on which co-reference is not a necessary condition for synonymy, but we can simply

¹ If the intuitions that support the Moorean premise are treated as infallible, the argument can attract the charge of begging the question. See (Frankena 1939). For the more modest approach adopted here, see (Darwall, Gibbard, and Railton 1992).

 -   

17

stipulate that, for the type of meaning we are talking about, this condition is in place. How you represent things as being can coincide with how I represent things as being only if the predicates you and I use for this purpose have the same referents, as meant by each of us. Premise 2, however, is much more problematic. It’s tantamount to the thought that the reference of a term, as understood by a speaker, is transparent to her. This might once have seemed to be a tenable point, but work on the semantics of natural kind terms pioneered in the 1970s by Saul Kripke and Hilary Putnam (Kripke 1980; Putnam 1975) showed conclusively that, for at least some terms, reference transparency doesn’t hold. And if it doesn’t hold for all terms, we cannot assume, as Premise 2 does, that it holds for ‘is morally right’.

2.2 The Naturalist Strikes Back This line of resistance to the open-question argument has been embraced by many naturalists. The general strategy is clearly expressed by Gilbert Harman: [ . . . ] as it stands the open question argument is invalid. An analogous argument could be used on someone who was ignorant of the chemical composition of water to ‘prove’ to him that water is not H₂O. This person will agree that it is not an open question whether water is water but it is an open question, at least for him, whether water is H₂O. Since this argument would not show that water is not H₂O, the open question argument in ethics cannot be used as it stands to show that for an act to be an act that ought to be done is not for it to have some natural characteristic C. (Harman 1977: 19)

Harman’s point can be easily adapted to bear on the version of the argument we are discussing. Harman’s general point is that if the argument for the claim that ‘is morally right’ doesn’t refer to the property of maximizing overall utility is regarded as sound, we will also have to accept as sound a parallel argument for the conclusion that ‘is water’ doesn’t refer to the property of containing mostly H₂O molecules. But this conclusion is false, so the argument can’t be sound, and if the argument concerning ‘is water’ is unsound, the argument concerning ‘is morally right’ will be under suspicion of exhibiting the same shortcoming. If we adapt the argument for ‘is morally right’ to the case of ‘is water’, we obtain the following: W1

Someone can mean by ‘is water’ what we mean by it even though they take a substance to satisfy ‘is water’ but believe that the substance doesn’t contain

18

  H₂O molecules, or take a substance not to satisfy ‘is water’ but believe that the substance contains mostly H₂O molecules. (Premise)

W2

If someone takes a substance to satisfy ‘is water’ but believes that the substance doesn’t contain H₂O molecules, or takes a substance not to satisfy ‘is water’ but believes that the substance contains mostly H₂O molecules, then ‘is water’, as understood by this speaker, doesn’t refer to the property of containing mostly H₂O molecules. (Premise)

W3

Someone can mean by ‘is water’ what we mean by it even though the predicate, as understood by them, doesn’t refer to the property of containing mostly H₂O molecules. (from W1 and W2)

W4

Someone can’t mean by ‘is water’ what we mean by it if the predicate has different referents as understood by them and as understood by us. (Premise)

Therefore: W5

‘is water’, as we understand it, doesn’t refer to the property of containing mostly H₂O molecules. (from W3 and W4)

The conclusion of this argument is false, but the argument is valid. Hence at least one of the premises has to be false. Premise W4 is just as plausible as Premise 4, so long as we focus on the kind of meaning involved in how we represent things as being with our sentences. I think Premise W1 is also unproblematic. It is widely accepted that someone can mean by ‘is water’ what we mean by it even if they take the predicate to be satisfied by a substance of which they don’t believe that it consists of H₂O molecules, or if they take the predicate not to be satisfied by a substance of which they don’t believe that it doesn’t consist of H₂O molecules. For one thing, having these beliefs would require having concepts of molecular structure, and speakers who lack these concepts can mean by ‘is water’ what we mean by it. I want to argue that it’s equally intuitive to accept, as Premise W1 maintains, that someone can mean by ‘is water’ what we mean by it even if they take the predicate to be satisfied by a substance of which they believe that it doesn’t consist of H₂O molecules, or if they take the predicate not to be satisfied by a substance of which they believe that it consists mostly of H₂O molecules. We think it’s perfectly possible for someone to have false beliefs about what a substance has to be like in order to satisfy the predicate ‘is water’, as meant by them. Speakers in preLavoisier times had all sorts of false beliefs about this, and they still could count as meaning by the predicate what we mean by it. Furthermore, there’s no reason to rule out this kind of mistake in subjects who have the concepts of molecular structure. Imagine an incompetent chemist who as a result of some errors in his

 -   

19

testing comes to the conclusion that the stuff in our lakes, etc. consists of H₂O₂ molecules, while the stuff in antiseptic bottles consists of H₂O molecules. This speaker will take ‘is water’ to be satisfied by substances of which he believes that they don’t consist of H₂O molecules, and he will take the predicate not to be satisfied by substances of which he believes that they consist of H₂O molecules. And yet it’s perfectly natural to think of him as meaning by ‘is water’ what we mean by it, and making a factual mistake about what a substance has to be like in order to satisfy the predicate, as meant both by him and by us. Premise W2, however, is unacceptable. The reason is that the referent of ‘is water’ is not fixed by speakers’ beliefs. The predicate refers to the unique property, if there is one, that satisfies a certain objective condition. Let me borrow from Hilary Putnam the term operational definition to refer to the array of criteria that a speaker (or a community) uses for applying a natural-kind term (Putnam 1975). On the standard account of the semantics of a natural-kind predicate, the property to which it refers is singled out as the underlying natural property present in most samples in the speaker’s environment that satisfy her operational definition for the term. ‘is water’, in particular, refers to the underlying property present in our environment in most samples that are transparent, tasteless liquids that quench thirst, etc. It refers to the property of containing mostly H₂O molecules because this property satisfies this condition. And this is so irrespective of whether speakers have beliefs to this effect. It follows that ‘is water’, as understood by our hapless chemist, refers to the property of containing mostly H₂O molecules, since that’s the underlying natural property present in most samples in his environment satisfying the operational definition that he associates with the predicate. This is so in spite of his beliefs to the contrary. Hence, from the fact that he means by the predicate what we mean by it (Premise W1), and that consequently the predicate has the same referent as understood by us and by him (Premise W4), it doesn’t follow that the predicate, as meant by us, doesn’t refer to the property of containing mostly H₂O molecules—this property is the referent of the predicate, as understood both by him and by us. But if Premise W2 of the ‘is water’ argument is rejected, we need to ask whether Premise 2 of the ‘is morally right’ argument shouldn’t suffer the same fate. Suppose that the referent of ‘is morally right’ is singled out in the same way as the referent of ‘is water’. On this model, there will be a certain condition C playing for ‘is morally right’ the role played for ‘is water’ by being the underlying property present in our environment in most samples that satisfy its operational definition. The referent of ‘is morally right’ will be the unique property, if there is one, satisfying condition C. Suppose that the property of maximizing overall utility is the unique satisfier of condition C. Then, on this model, ‘is morally right’ will refer to the property of maximizing overall utility. However, the fact that this property is the unique satisfier of C is an empirical fact of which speakers may well be unaware. In this situation, they can be expected to take the predicate to be satisfied

20

 

by actions of which they believe that they don’t maximize overall utility or to take it not to be satisfied by actions of which they believe that they maximize overall utility, even if they accept that the predicate is satisfied by the instances of the unique C-satisfier. However, in spite of this, the predicate, as understood by these speakers, will still refer to the property of maximizing overall utility. If ‘is morally right’ follows the natural-kind term template, Premise 2 is false. Naturalists who endorse this model have nothing to fear from the open-question argument.

2.3 The Argument Revised I think it’s undeniable that Harman and other naturalists have identified a serious gap in the open-question argument. If ‘is morally right’ follows the same semantic model as ‘is water’, the argument doesn’t provide adequate support for its conclusion. I want to suggest, however, that the advocate of the open-question argument still has some room for manoeuvre. Not every predicate follows the natural-kind model. It’s not the case for every predicate P that it refers to the underlying natural property present in most items in the speaker’s environment that (uniquely) satisfies P’s operational definition. What conditions need to be satisfied in order for P to follow this model? I think we can plausibly claim that P, as understood by a speaker, wouldn’t follow the natural-kind model unless the speaker’s dispositions concerning the ascription of P satisfied certain conditions. For example, suppose a speaker is not prepared to accept that it’s in principle possible that P fails to be satisfied by an object that satisfies its operational definition, and vice versa, that P is satisfied by an object that fails to satisfy its operational definition. It seems to me that this circumstance would give us a powerful reason for rejecting the idea that P, as understood by this speaker, follows the natural-kind model. It would be much more plausible to say that P, as understood by the speaker, refers to a property defined in terms of the operational definition of P, i.e. of surface features, not to an underlying substance. I want to suggest that there’s another condition that a speaker’s dispositions concerning the ascription of P need to satisfy in order for P, as understood by her, to follow the natural-kind model. Suppose a speaker is prepared to take P to be satisfied by an object of which she believes that it doesn’t instantiate the underlying property present in most instances of P’s operational definition, or to take P not to be satisfied by an object of which she believes that it instantiates the underlying property present in most instances of P’s operational definition. In these conditions, the claim that P follows the natural-kind model would involve ascribing to the speaker not the kind of empirical ignorance involved in not knowing which underlying property is present in most instances of P’s operational definition, but semantic confusion. We would have to claim that the speaker is confused about the meaning of P, as she understands it. There might be

 -   

21

circumstances under which this is the least bad thing to do, but in general, if our account of the semantics of a predicate, as understood by a speaker, forces us to ascribe to her semantic confusion, this should be treated as a powerful reason for thinking that our account of the semantics of the predicate is mistaken. This point needs to be qualified to take into account the social dimension of language (Burge 1979). Ascribing semantic confusion to a speaker might be reasonable if the semantics of her terms are not fixed exclusively by her dispositions, but also by those of other members of the community. If the community as a whole takes P to be satisfied by instances of the underlying property present in most instances of P’s operational definition, it might be the right thing to say that P follows the natural-kind model even as understood by members of the community who would count on this interpretation as exhibiting semantic confusion. But there are cases in which the attribution of semantic confusion cannot be justified in this way—for example, if it would affect the whole linguistic community, or the experts whose linguistic dispositions our semantic theory needs to be faithful to. In these situations, if our account of the meaning of a term forces us to ascribe semantic confusion to the speakers, we should take this as a reason for abandoning our account. In my discussion here I’m going to assume that, in the cases we’ll be considering, the attribution of semantic confusion cannot be justified by appeal to the social dimension of meaning. Now, if the predicate ‘is morally right’ followed the natural-kind model, then, as we’ve seen, there would have to be a condition C such that ‘is morally right’ refers to the unique satisfier, if there is one, of C. The fact that the referent of the predicate is singled out in this way would be a semantic fact. Suppose now that a speaker takes an action to satisfy ‘is morally right’ but believes that the action doesn’t instantiate the unique satisfier of condition C, or takes an action not to satisfy ‘is morally right’ but believes that the action instantiates the unique satisfier of condition C. Then maintaining that the referent of the predicate, as understood by this speaker, is singled out as the unique satisfier of C would force us to attribute to her semantic confusion. And, as we’ve seen, if an account of the semantics of a term, as understood by a speaker, forces us to attribute to her semantic confusion, and this attribution cannot be justified in terms of the social dimension of meaning, we should take this as a powerful reason for rejecting this semantic account. These reflections sustain a revised version of the open-question argument— targeting the view that ‘is morally right’ follows the natural-kind model:² 1*

A speaker can mean by a predicate what we mean by ‘is morally right’ even though she takes an action to satisfy the predicate but believes that the action

² I first presented a version of this line of reasoning in (Zalabardo 2012b).

22

  doesn’t instantiate the unique satisfier of condition C, or takes an action not to satisfy the predicate but believes that the action instantiates the unique satisfier of condition C. (Premise)

2*

If a speaker takes a predicate to be satisfied by an action of which she believes that it doesn’t instantiate the unique satisfier of condition C or takes the predicate not to be satisfied by an action of which she believes that it instantiates the unique satisfier of condition C, then the referent of the predicate, as understood by the speaker, is not singled out as the unique satisfier of condition C. (Premise)

3*

A speaker can mean by a predicate what we mean by ‘is morally right’ even though the referent of the predicate, as understood by the speaker, is not singled out as the unique satisfier of condition C. (from 1* and 2*)

4*

If a speaker means by a predicate what we mean by ‘is morally right’ and the referent of the predicate, as understood by the speaker, is not singled out as the unique satisfier of condition C, then the referent of ‘is morally right’, as understood by us, is not singled out as the unique satisfier of condition C. (Premise)

Therefore: 5*

The referent of ‘is morally right’, as understood by us, is not singled out as the unique satisfier of condition C. (from 3* and 4*)

It follows from our discussion just now that someone who maintains that the referent of ‘is morally right’ is singled out along the lines of the natural-kind model cannot object to Premise 2*, subject to our assumption that the social dimension of meaning doesn’t affect the situation. Hence, if 1* and 4* are accepted, the naturalistic account of the referent of ‘is morally right’ will have to be rejected. Defending 1* would require showing that someone could mean by a predicate what we mean by ‘is morally right’ even if she takes the predicate to be satisfied by an action that doesn’t instantiate the unique satisfier of C, or takes the predicate not to be satisfied by an action that instantiates the unique satisfier of C. What condition could play the role of C? One popular line for naturalists at this point is to invoke the higher-order property of being the natural property that causally regulates our use of the predicate (Boyd 1988).³ If we focus on this proposal, defending the premise will require showing that someone could mean by a

³ This is the option discussed by Terence Horgan and Mark Timmons (Horgan and Timmons 1991). I think the same argument can be run for the functionalist account of how moral terms obtains their referents defended in (Jackson and Pettit 1995).

 -   

23

predicate what we mean by ‘is morally right’ even though she takes the predicate to be satisfied by an action she believes not to instantiate the natural property that causally regulates our ascriptions of the predicate, or takes the predicate not to be satisfied by an action she believes to instantiate the natural property that causally regulates our ascriptions of the predicate. It seems to me that the intuition in favour of this is as strong as for Premise 1 of the original version of the open-question argument. In both cases, the intuition is generated by the nature of the procedures that we employ for determining who means by a predicate what we mean by ‘is morally right’. These procedures postulate sufficient conditions for sameness of meaning that are compatible with wide differences in beliefs regarding how the satisfaction of the target predicate by an action is correlated with how things stand with the action in other respects. Later on, I’m going to consider a specific proposal concerning the character of these sufficient conditions, but at this point we don’t need to be specific. We can take them to include familiar features of how the predicate is used such as, for example, the fact that speakers decide on the ascription of the predicate on the basis of their sense of moral approval, the fact that speakers feel motivated to perform or promote the actions to which they ascribe the predicate, the fact that we take satisfaction of the predicate by an action to be connected with whether the action promotes our well-being, etc. I’m claiming that there are combinations of features of this kind, such that if they are present in someone else’s use of a predicate, we will take them to mean by the predicate what we mean by ‘is morally right’, irrespective of whether they agree with us on the correlation between satisfaction of the predicate by an action and how things stand with the action in other respects. Specifically, in the presence of these features, we will take other speakers to mean by their predicate what we mean by ‘is morally right’ even if they disagree with us on whether the predicate is satisfied precisely by actions that maximize overall utility, as Premise 1 dictates, and even if they disagree with us on whether the predicate is satisfied precisely by actions that instantiate the natural property that causally regulates our ascriptions of the predicate, as Premise 1* dictates. And if, as I suspect, the intuition is equally strong for any plausible account of C, we will be able to adapt Premise 1* to any specific proposal as to how the referent of ‘is morally right’ is singled out. Premise 4* also needs some defence. It rests on the assumption that if the referent of a predicate P, as understood by a speaker S, and the referent of a predicate P0 , as understood by a speaker S0 , are singled out in different ways, then P, as understood by S and P0 as understood by S0 cannot have the same meaning. It could be argued, however, that this assumption runs counter to a plausible account of synonymy conditions for natural-kind terms. Consider the situation in Putnam’s famous thought experiment, of a planet in which our operational definition of ‘is water’ is satisfied not by H₂O, but by the different chemical compound XYZ (Putnam 1975). These speakers have a

24

 

predicate, ‘is twater’, associated with the same operational definition as our predicate ‘is water’, but since the operational definition picks out different referents in the two environments, ‘is twater’, as meant by these speakers, doesn’t have the same meaning as ‘is water’, as meant by us. Now suppose that, unlike in Putnam’s thought experiment, H₂O is actually present on Twin Earth, as a very rare, yellow, bitter liquid present at the bottom of some mines. Suppose also that Twin English has, in addition to ‘is twater’, a less familiar predicate, ‘is swater’, associated with the operational definition that singles out H₂O in their environment—being a very rare, yellow, bitter liquid found at the bottom of some mines. Clearly, ‘is swater’, as understood by these speakers, has the same referent as ‘is water’ as understood by us. But do the two predicates also have the same meaning? If they do, then they offer a counterexample to the assumption that two predicates can have the same meaning only if their referents are singled out in the same way. The referent of our predicate ‘is water’ is singled out as the underlying property present in most instances of our operational definition of our predicate, while the referent of ‘is swater’ is singled out as the underlying property present in most instance of our twins’ very different operational definition of their predicate. The naturalist could use these observations to undermine Premise 4*. She could argue that if ‘is morally right’ follows the natural-kind model, then it’s possible that its referent is singled out as the unique satisfier of condition C even if it has the same meaning as another predicate whose referent is not singled out in this way. If this possibility is accepted, Premise 4* won’t hold, and the remaining premises of the argument won’t yield a conclusion about how the referent of ‘is morally right’ is singled out. I think the advocate of the argument has some strategies at her disposal for overcoming this objection. One option, of course, is to reject the claim that two predicates whose referents are singled out in different ways can nevertheless have the same meaning, and hence that our predicate ‘is water’ has the same meaning as our twins’ ‘is swater’. But even if this claim is accepted it might be possible to save the argument. The basic idea of the construal of the semantics of ‘is water’ on which the objection is based is that there isn’t a unique operational definition D of which a speaker needs to believe that the satisfaction conditions of one of her predicates are in line with the instantiation conditions of the property present in most instances of D in order to mean by the predicate what we mean by ‘is water’. We might concede this and still maintain that, for each speaker, there is an operational definition playing this role—possibly different operational definitions for different speakers. A speaker in our environment won’t mean by a predicate what we mean by ‘is water’ unless she accepts the connection between the predicate and our operational definition of ‘is water’. A speaker on Twin Earth won’t mean by a predicate what our twins mean by ‘is swater’—i.e. what we mean

 -   

25

by ‘is water’—unless she accepts the connection between the predicate and their operational definition of ‘is swater’. I think that this weaker link between the predicate’s meaning and operational definitions is enough to uphold the new version of the open-question argument. There are no speakers for which intuition would dictate that they don’t mean by ‘is morally right’ what we mean by this predicate unless they accept that its satisfaction conditions are in line with the instantiation conditions of the property that satisfies some non-trivial description. Pursuing this strategy would involve postulating a function pairing each speaker S with the condition CS that has to single out the referent of a predicate in order for the predicate, as understood by her, to have the same meaning as ‘is morally right’, as understood by us. If we now replace C with CS in our formulation of the argument, the analogue of Premise 4* is not undermined by the water/swater point. The analogue of Premise 2* remains true, and the considerations we’ve adduced in support of Premise 1* would also support the new version. The second strategy for resisting the objection would be to call into question the analogy between ‘is morally’ right and ‘is water’, if the semantics of the latter is construed on the model that the objection assumes. On this model, there are no restrictions on the operational definition with which a predicate needs to be associated in order to have the same meaning as ‘is water’, as we understand it, so long as the operational definition picks out the same referent in the relevant environment. But when applied to ‘is morally right’, this line is problematic. Assume for the sake of the argument that as a matter of fact our use of ‘is morally right’ is causally regulated by the property of maximizing overall utility. Suppose now that there’s a community for which actions that maximize overall utility cause hilarity, and nothing else does. Suppose these speakers have a predicate that has this kind of mirth as its operational definition. Since this operational definition picks out the same property as the description that singles out the referent of ‘is morally right’, we would have to conclude, on the model under discussion, that these speakers mean by their predicate what we mean by ‘is morally right’. I find this consequence highly implausible. A predicate that is ascribed to an action on the basis of whether the action makes the speaker laugh can’t possibly have the same meaning as ‘is morally right’, as we understand this predicate. The right account of the meaning of this predicate can’t entail synonymy conditions that impose no constraints on the procedure that regulates ascription of the predicate.⁴ This completes my defence of the new version of the open-question argument. I think that its three premises stand on firm ground, and that they validly entail the conclusion that the referent of ‘is morally right’ is not singled out along the ⁴ The point can also be made using the science-fiction example presented in (Wedgwood 2007: 63–4).

26

 

lines of the natural-kind term template. I think, more generally, that this line of reasoning undermines any representationalist specification of the meaning ground of ‘is morally right’. According to the representationalist, the predicate obtains its meaning from its relation to the property playing the role of its referent. On this picture, there would be a condition that singles out the property playing this role—the referent of the predicate will be the unique satisfier, if there is one, of this condition. Providing a representationalist meaning ground for the predicate would consist in identifying the condition playing this role. And the fact that the predicate refers to the unique satisfier of this condition will be a semantic fact. Hence the reasoning that we’ve presented for the natural-kind model will apply to any representationalist specification of the meaning ground of ‘is morally right’. Notice that in this case a version of the bad-company objection levelled by Harman against the original argument is not available. To be sure, the analogue for ‘is water’ of Premise 2* will have to be accepted as true: W2*

If a speaker takes a predicate to be satisfied by a substance of which she believes that it doesn’t instantiate the underlying natural property present in most samples of the operational definition of ‘is water’, or takes the predicate not to be satisfied by a substance of which she believes that it instantiates the underlying natural property present in most samples of the operational definition of ‘is water’, then the referent of the predicate, as understood by this speaker, isn’t singled out as the underlying natural property present in most samples in her environment of the operational definition of the predicate.

If a speaker thinks that a predicate can be satisfied by a substance that doesn’t instantiate the underlying natural property present in the transparent, tasteless, thirst-quenching liquid in our environment, or vice versa, and she is not semantically confused, then the predicate, as meant by this speaker, doesn’t refer to this underlying natural property.⁵ However, the analogue of Premise 1* is not true: W1*

A speaker can mean by a predicate what we mean by ‘is water’ even though she takes a substance to satisfy the predicate but believes that the substance doesn’t instantiate the underlying natural property present in most samples of the operational definition of ‘is water’, or takes a substance not to satisfy the predicate but believes that the substance instantiates the underlying natural property present in most samples in her environment of the operational definition of ‘is water’.

⁵ The reasoning could be easily adapted to accommodate the water/swater point.

 -   

27

For speakers who are not semantically confused, meaning by ‘is water’ what we mean by it is incompatible with thinking that the satisfaction conditions of the predicate differ from the instantiation conditions of the underlying natural property present in the transparent, tasteless, thirst-quenching liquid in our environment (or of the property that satisfies the condition playing the same role in the speaker’s environment). This is the point at which ‘is water’ and ‘is morally right’ come apart. It follows that treating our revised open-question argument for ‘is morally right’ as sound doesn’t commit us to treating as sound a parallel argument for the false conclusion that the referent of ‘is water’ is not singled out as the underlying natural property present in most samples in our environment of the operational definition of the predicate. Let’s take stock. We are assessing the idea that ‘is morally right’ refers to a natural property, using the property of maximizing overall utility as an example. The first version of the open-question argument we’ve considered derives the rejection of this proposal from the thought (the Moorean premise) that someone can mean by ‘is morally right’ what we mean by it even if they think the satisfaction conditions of the predicate are not in line with the instantiation conditions of the property. I’ve contended that the naturalist could defend her position from this attack by challenging another premise (Premise 2) required by the anti-naturalist argument. They would contend that, contrary to what this premise claims, ‘is morally right’, as understood by a speaker, can refer to the property of maximizing overall utility even if the speaker thinks that the satisfaction conditions of the predicate are not in line with the instantiation conditions of the property. I’ve then argued that the open-question argument might be saved by adjusting its target. The new version is based on a revised Moorean premise, according to which someone can mean by ‘is morally right’ what we mean by it even if they think the satisfaction conditions of the predicate are not in line with the instantiation conditions of the unique satisfier of the condition C which singles out the referent of the predicate, on the naturalist picture. I’ve argued that the naturalist strategy for blocking the original argument no longer works here. ‘is morally right’, as understood by a speaker, can refer to the property of maximizing overall utility even if the speaker thinks that the satisfaction conditions of the predicate are not in line with the instantiation conditions of the property. However, ‘is morally right’, as understood by a speaker, can’t refer to the property that uniquely satisfies condition C if the speaker thinks that the satisfaction conditions of the predicate are not in line with the instantiation conditions of the unique C-satisfier. This version of the open-question argument does pose a serious challenge to the naturalist account of the reference of ethical predicates, and to any representationalist account of their meaning grounds.

28

 

2.4 Moral Twin-Earth The version of the open-question argument I’ve developed here has important points of contact with the version developed by Terence Horgan and Mark Timmons (Horgan and Timmons 1992). Their version, like mine, is aimed at overcoming the attempt to block the argument by construing the semantics of ethical terms along the lines of the natural-kind term paradigm. Horgan and Timmons develop their argument on the basis of a thought experiment concerning what they call Moral Twin Earth, in analogy with the Twin Earth thought experiments employed in discussion of the semantics of natural-kind terms. Moral Twin Earth is like Earth in most respects. Specifically, Twin English speakers have a vocabulary that works much like the human moral vocabulary, whose terms are orthographically identical to our moral terms. And these terms behave on the surface in the same way as our moral terms: In particular, the terms are used to reason about considerations bearing on Moral Twin Earthling well-being; Moral Twin Earthlings are normally disposed to act in certain ways corresponding to judgments about what is ‘good’ and ‘right’; they normally take considerations about what is ‘good’ and ‘right’ to be especially important, even of overriding importance in most cases, in deciding what to do, and so on. (Horgan and Timmons 1992: 164)

There is, however, one important difference between their use of these terms and our use of moral terms. Let’s assume, with Horgan and Timmons, that our own use of moral terms is causally regulated by a utilitarian property. Assume, specifically that our ascription of ‘is morally right’ to an action is causally regulated by whether the action maximizes overall utility. Moral Twin Earth differs from Earth on this point. The ascription of ‘is morally right’ to an action by speakers of Twin English is causally regulated instead by whether the action instantiates some deontological property. The question now is whether we should interpret the predicate ‘is morally right’ in Twin English as having the same meaning as our own predicate. Horgan and Timmons argue convincingly that the naturalist is committed to answering this question in the negative. If ‘is morally right’, as meant by a speaker, refers to the property that causally regulates her uses of the predicate, then their predicate and ours have different referents, and a fortiori different meanings. On this view, when we assert ‘killing one to save five is morally right’ and they reject this sentence, we are not really disagreeing. The property we are claiming the action to exemplify is different from the property they are claiming the action not to exemplify. It’s not a disagreement in moral belief, but a disagreement in meaning. However, according to Horgan and Timmons, this verdict is in open conflict with our intuitions. We would be strongly inclined to maintain:

 -   

29

that moral and twin-moral terms do not differ in meaning or reference, and hence that any apparent moral disagreements that might arise between Earthlings and Twin Earthlings would be genuine disagreements—i.e., disagreements in moral belief and in normative moral theory, rather than disagreements in meaning. (Horgan and Timmons 1992: 165)

Accommodating this intuition requires rejecting the moral naturalist view that ‘is morally right’ refers to the property that causally regulates our ascription of the predicate. Furthermore, Horgan and Timmons maintain, a similar reasoning could be developed to undermine any other account of how the natural property to which the predicate refers is singled out. Moral Twin Earth is not merely a thought experiment, but a recipe for constructing thought experiments: For any potential version of SSN [Synthetic Semantic Naturalism] that might be proposed, according to which (i) moral terms bear some relation R to certain natural properties that collectively satisfy some specific normative moral theory T, and (ii) moral terms supposedly refer to the natural properties to which they bear this relation R, it should be possible to construct a Moral Twin Earth scenario suitably analogous to the one constructed above—i.e., a scenario in which twin-moral terms bear the same relation R to certain natural properties that collectively satisfy some specific normative theory T0 , incompatible with T. (Horgan and Timmons 1992: 167)

In other words, their version of the argument can be deployed against any account of the semantics of ‘is morally right’, as understood by us, according to which the predicate refers to the unique natural property, if there is one, that satisfies some condition C. Hence their argumentative strategy can be deployed to establish a conclusion of the same generality as the version of the argument I’ve developed. I’m interested in considering how my version of the open-question argument compares to the version developed by Horgan and Timmons, when taken as targeting the view that ‘is morally right’, as understood by us, refers to the natural property that causally regulates our ascriptions of the predicate. For this purpose, I propose to construe their argument as having the following structure: HT1

Twin Earthlings mean by ‘is morally right’ what we mean by it. (Premise)

HT2

Someone can’t mean by ‘is morally right’ what we mean by it if the predicate has different referents as understood by them and as understood by us. (Premise)

HT3

‘is morally right’ doesn’t have different referents as understood by us and as understood by Twin Earthlings. (from HT1 and HT2)

30

 

HT4

Twin Earthlings’ ascriptions of ‘is morally right’ are not causally governed by the same natural property as our ascriptions of this predicate. (Premise)

HT5

It’s not the case that ‘is morally right’, as understood by us, and ‘is morally right’, as understood by Twin Earthlings, both refer to the natural property that causally governs ascriptions of the predicate. (from HT3 and HT4)

HT6

If ‘is morally right’, as understood by us, refers to the natural property that causally regulates our ascriptions of the predicate, then ‘is morally right’, as understood by Twin Earthlings, refers to the natural property that causally regulates their ascriptions of the predicate. (Premise)

Therefore: HT7

‘is morally right’, as understood by us, doesn’t refer to the natural property that causally regulates our ascription of the predicate. (from HT5 and HT6)

It seems to me that Premises HT2, HT4, and HT6 are fairly uncontroversial. HT2 can be taken as a stipulation of the kind of meaning we are concerned with. HT4 is a direct consequence of how moral Twin Earth is described. And HT6 follows from the reflection that there’s no reason to think that our predicate and the Twin Earthlings’ predicate don’t follow the same semantic pattern. The only promising line for the naturalist to resist the argument is to object to HT1. Critics have questioned the extent to which our intuitions about Moral Twin Earth can be taken to lend legitimate support to HT1. Stephen Laurence, Eric Margolis, and Angus Dawson (1999) have argued that our intuitions regarding the case are invalidated by important disanalogies between Moral Twin Earth and the original Twin Earth thought experiment. They observe that, whereas the fictional compound XYZ of Putnam’s original case is not a plausible candidate for the role of referent of our predicate ‘is water’, the deontological property that’s supposed to regulate the Twin Earthlings’ use of their predicate is a plausible referent for our predicate ‘is morally right’. This circumstance, they claim, is going to bias the case towards the intuition that their predicate has the same meaning as ours (Laurence, Margolis, and Dawson 1999: 156). They also point out that, whereas H₂O is not present on Putnam’s Twin Earth, utilitarian properties such as maximizing overall utility are surely exemplified on Moral Twin Earth, and we should expect its inhabitants to have terms for them (Laurence, Margolis, and Dawson 1999: 160). Jorn Sonderholm (2013) has objected to the argument on the grounds that, ‘there, at present, is a significant amount of disagreement among experts on the question of what property it is that uniquely, causally regulates the use of “right” ’, but ‘Timmons builds into his twin-earth example that this disagreement has completely vanished and that we all know what property it is that plays this regulative role on earth’. Consequently, Sonderholm contends, ‘it is difficult for us,

 -   

31

through the workings of our imagination, to get a proper grasp of what it would be like to be in a position in which one had the knowledge stipulated by Timmons’ (Sonderholm 2013: 85).⁶ I’m not going to discuss these attacks on Horgan and Timmons’s argument here.⁷ I only want to point out that none of these concerns affect my version of the argument. The speakers that are supposed to instantiate my Premise 1* are not inhabitants of Moral Twin Earth. They are among us, people like you and me whose ascriptions of ‘is morally right’ are not in line with their beliefs on which actions instantiate the natural property that causally regulates their ascriptions of the predicate. One way this might happen is if they have false beliefs about which property plays the causal-regulation role, or which actions instantiate the property that plays this role, or both. Notice that this circumstance is compatible with their ascriptions being causally regulated by the same property that regulates our ascriptions. I want to concentrate instead on another possible line of attack that highlights, in my view, the superior dialectical effectiveness of my version of the argument over Horgan and Timmons’s. Both versions of the argument are based on the thought that other speakers could mean by a predicate what we mean by ‘is morally right’ in spite of differences between these speakers and us that would force the naturalist to reject this verdict. I think Horgan and Timmons and I would adduce very similar considerations to support our synonymy verdicts. The features of our use of ‘is morally right’ that I have invoked for this purpose are very close to what Horgan and Timmons describe as ‘the “formal” marks that we take to characterize moral vocabulary and moral practice’ (Horgan and Timmons 1992: 164). The main difference between my version of the argument and theirs is the nature of the differences that we take the synonymy verdicts to be compatible with. In Horgan and Timmons’s version, Twin Earthlings are claimed to mean by ‘is morally right’ what we mean by it in spite of the fact that their use of the predicate and our use of the predicate are causally regulated by different natural properties. In my version, the claim is that a speaker can mean by ‘is morally right’ what we mean by it in spite of the fact that their ascriptions of the predicate are not in line with their beliefs on which actions instantiate the natural property that causally regulates our ascriptions of the predicate. I think this difference affects the plausibility of the charge of begging the question that the naturalist might level against our respective versions of the argument.⁸

⁶ Sonderholm is discussing the version of the argument presented in (Timmons 1998). ⁷ For a reply to Laurence, Margolis, and Dawson, see (Rubin 2008). For a reply to Sonderholm, see (Horgan and Timmons 2015). ⁸ On the possibility that Horgan and Timmons’s version of the argument begs the question against the naturalist see (Miller 2013: 170).

32

 

Consider Horgan and Timmons’s version first. They are asking the naturalist to accept that the differences between the causal regulation of our use of the predicate and of the Twin Earthlings’ use don’t undermine the view that they mean by the predicate what we mean by it. Hence, in effect, Horgan and Timmons base their argument on the contention that which natural property causally regulates a speaker’s ascriptions of ‘is morally right’ is irrelevant when it comes to deciding who means by the predicate what we mean by it. The problem is that, for the naturalist, whether a speaker means by a predicate what we mean by ‘is morally right’ is defined in terms of which natural property causally regulates their ascriptions of the predicate. It is a direct consequence of their view that the differences in causal regulation should lead us to conclude that the Twin Earthlings don’t mean by the predicate what we mean by it, and that our standard synonymy criteria yield the wrong result in this case. Unless these criteria are implausibly treated as infallible, the naturalist can maintain that they yield the wrong results whenever they sanction synonymy between moral terms whose use is causally regulated by different properties. Consider now my version of the argument, as applied to the brand of naturalism that Horgan and Timmons target. What I’m asking the naturalist to accept, in the first instance, is that the fact that someone’s ascriptions of ‘is morally right’ are not in line with their beliefs on which actions instantiate the property that causally regulates their ascriptions of the predicate shouldn’t undermine the view that they mean by the predicate what we mean by it. This places me in what I regard as a more favourable dialectical position vis à vis the naturalist. Her view is that whether a speaker means by a predicate what we mean by ‘is morally right’ is not determined by the speaker’s beliefs, but by actual facts of causal regulation. Hence accepting that speakers’ beliefs shouldn’t be taken into account for deciding whether they mean by a predicate what we mean by ‘is morally right’ doesn’t immediately force the naturalist to reject her account of the reference of the predicate. Acceptance of this comes into conflict with her account of the reference of the predicate only through the mediation of the independently motivated Premise 2*. Furthermore, the naturalist would accept that whether someone means by ‘is morally right’ what we mean by it is not generally affected by how their verdicts on the satisfaction of the predicate by an action are aligned with their beliefs on how things stand with the action in other respects. If our ascriptions of ‘is morally right’ are causally regulated by the property of maximizing overall utility, someone can still mean by the predicate what we mean by it although they ascribe the predicate to actions which they believe not to maximize overall utility, or vice versa. In order to defend her view, the naturalist would have to argue that there is an exception to this general principle in the case of speakers’ beliefs on which actions instantiate the property that causally regulates their ascriptions of the predicate. And this places the onus on the naturalist to convince us that this exception is justified, without begging the question in favour of her view.

 -   

33

2.5 Moral Motivation A related line of reasoning against naturalist accounts of the reference of ethical terms is based on the idea that ethical judgement necessarily includes an element of motivation. On this view, the meaning of the sentence ‘killing one to save five is morally right’ has to be such that you don’t count as believing that things are as the sentence represents them as being unless you exhibit some motivation to promote, facilitate, or engage in the action of killing one to save five. Accommodating this thought within the representationalist paradigm would seem to require picking as the referent of ‘is morally right’ an intrinsically motivating property—a property such that anyone who believes it is instantiated by an action will ipso facto be motivated to promote, facilitate, or engage in this action. However, the thought goes, no natural property is intrinsically motivating. Someone might believe, for example, that an action maximizes overall utility without exhibiting any positive motivation with respect to the action. In general, for every natural property P, it is perfectly possible for someone to believe that an action instantiates P without being motivated to promote, facilitate, or engage in the action. Therefore, the argument concludes, the predicate ‘is morally right’ can’t refer to the property of maximizing overall utility, or to any other natural property.⁹ This line of reasoning has generated extensive debate, and assessing it adequately would require discussing in detail the strategies developed by naturalists for trying to undermine it. One option for them is to reject the major premise of the argument, maintaining, to the contrary, that it is in principle possible to believe that an action is morally right without exhibiting any positive motivation towards the action (Railton 1986; Boyd 1988; Brink 1989). This is of course compatible with the existence of a contingent connection between moral belief and motivation, which would explain the intuitive plausibility of the premise. Other strategies are open to the naturalist.¹⁰ I’m not going to discuss their prospects here. My goal has been only to outline the challenge that the link between moral belief and motivation poses for naturalist accounts of the reference of ethical terms.

2.6 Moral Intuitionism Many philosophers believe that natural properties are the only properties the world contains, and hence the only candidates for the job of predicate referents. If this view is correct, it follows that if ‘is morally right’ can’t refer to a natural

⁹ For presentations of this line of reasoning, see (Mackie 1977; Smith 1994). For critical discussion of Smith’s version of the argument see (Shafer-Landau 1998). ¹⁰ One strategy is presented by (Tresan 2006). See (Björklund et al. 2012) for a survey of extant approaches.

34

 

property, then it can’t have a referent at all. Then, according to representationalism, the sentence ‘killing one to save five is morally right’ will fail to represent things as being a certain way. These conclusions no longer follow if we are happy to accept that the world also contains non-natural properties. Notice, though, that, as Moore already noted, accepting non-natural properties as potential referents of ethical terms doesn’t by itself remove the threat of the open-question argument (Moore 1903: 91). The reasoning that we’ve presented against the view that ‘is morally right’ refers to the property of maximizing overall utility can be easily adapted to target the view that the predicate refers, say, to the property of being liked by God.¹¹ The only way to avoid the problem is to take ‘is morally right’ to refer to a (nonnatural) sui generis property, which can be specified only with a nominalization of the predicate itself, or some closely related terms.¹² This sui generis character renders the proposal immune to the open-question argument, as it leaves no scope for believing that an action instantiates the property but doesn’t satisfy the predicate, or satisfies the predicate but doesn’t instantiate the property. The problem of moral motivation can also be addressed if we maintain that the sui generis referent of the predicate has the intrinsic motivational force that natural properties appear to lack. The view is not without problems, even if we are in principle prepared to accept the existence of non-natural properties with intrinsic motivating force. One major source of difficulties is the epistemology of the beliefs that moral sentences would express on this account. Let’s grant that there’s a property of moral rightness that is present in some actions and not in others, and let’s grant that sentences ascribing the predicate ‘is morally right’ to an action represent the action as instantiating this property. Now we need to ask, how do we know which actions instantiate this property? This is not a generic form of scepticism that would question our knowledge of the instantiation of any property. For natural properties we have a story to tell concerning how our cognitive devices can detect their presence in or absence from objects. What this proposal would need to supply is a parallel story about how we detect the presence in or absence from an action of the property of moral rightness. The standard strategy is to appeal to the feelings of moral approval and disapproval that we routinely use to decide whether things are as a sentence ascribing the predicate ‘is morally right’ to an action represents them as being. These feelings are unquestionably real, but their existence is not enough to meet the challenge. This requires construing these feelings as a cognitive faculty that enables us to detect the property of moral rightness.¹³ But if this faculty is real, ¹¹ On this point, see (Darwall, Gibbard, and Railton 1992). ¹² The view is associated with G. E. Moore (Moore 1903). For a more recent defence see (ShaferLandau 2003). ¹³ See (McDowell 1979) for insightful discussion of the epistemological consequences of a nonnaturalistic conception of ethical facts.

 -   

35

there will have to be a causal mechanism through which the presence in or absence from an action of the property makes a difference to our cognitive state.¹⁴ Without an adequate answer to this epistemological challenge, the claim that ‘is morally right’ refers to a sui generis non-natural property wouldn’t hold much appeal. Another source of difficulties for the view is the intuition that the ethical supervenes on the natural—two states of the world can’t differ from each other in moral respects unless they also differ from each other in some natural respects. If the referents of ethical predicates are natural properties, this phenomenon can be easily explained, but if the referents of ethical predicates are non-natural, explaining supervenience will force us to postulate necessary connections between natural and non-natural properties, and it would seem hard to argue that this move is not entirely ad hoc. For our purposes, the most salient shortcoming of this approach is its questionable value for an account of the meaning ground of the target predicate. According to the representationalist template, what makes the predicate ‘is morally right’ have the meaning it has is its relation to the property that plays the role of its referent. However, if we think of this as a sui generis property, only specifiable with a nominalization of the predicate—as the property of being morally right, it’s hard to avoid the impression of an explanatory circle. The predicate has the meaning it has by virtue of its relation to a property about which we can only say that it is the property to which the predicate refers. It’s not clear that we can learn in this way what makes the predicate have the meaning it has.

2.7 Conclusion I have presented the challenge posed by a version of the open-question argument to the project of identifying the referents of ethical terms. On my construal, the crucial thought is that someone can mean by an ethical predicate what we mean by it even though their ascriptions of the predicate are not in line with their views on the instantiation of any property the representationalist might want to treat as the referent of the predicate. I have argued that the standard version of the argument can be blocked by a naturalist conception of the reference of ethical terms, but that the argument can be modified to overcome this obstacle. I have discussed the parallels and differences between my version of the argument and the Moral Twin Earth argument developed by Terence Horgan and Mark Timmons. I have outlined the challenges faced by the attempt to solve these

¹⁴ See (Benacerraf 1965) for a similar challenge to Platonism in mathematics.

36

 

difficulties by embracing moral intuitionism. In the next chapter I’m going to consider how the version of the open-question argument I have developed for ethical predicates can be deployed with regard to ascriptions of truth, meaning, and propositional attitudes, to argue that the central predicates of these discourses can’t have referents, and hence that their sentences can’t have representationalist meaning grounds.

3 The Open-Question Argument in Semantics The open-question argument was originally developed for ethical discourse, but more recently it has also played an important role in discussions of semantic discourses. My goal in the present chapter is to consider how the version of the argument developed for ethics in the previous chapter can be deployed to the same effect for ascriptions of truth, propositional attitudes, and meanings—to undermine the claim that the central predicates of these discourses have referents, and that their sentences have representationalist meaning grounds.

3.1 Truth Ascriptions Consider the sentence: T

‘ “snow is white” is true’

On the face of it, it is a subject-predicate sentence. Its subject is a singular term, ‘ “snow is white” ’, referring to a sentence, ‘snow is white’, produced by enclosing the sentence in quotation marks. T appears to represent things as being a certain way with ‘snow is white’—to represent the sentence as instantiating a property— the property that the predicate ‘is true’ refers to. Which property could play this role? Traditional theories of truth can be formulated as accounts of the referent of the truth predicate. Thus, for example, the correspondence theory would give us the claim that ‘is true’ refers to the property of corresponding to the facts, the coherence theory could be formulated as the claim that ‘is true’ refers to the property of being a member of a maximally coherent system of sentences, and from a pragmatic theory of truth we obtain the claim that ‘is true’ refers to the property of being accepted at the end of enquiry. I’m not going to discuss here the relative virtues of these proposals. What I want to consider is a general line of reasoning against the possibility of treating any property as the referent of ‘is true’. As with ethical predicates, I’m going to focus on a specific account of the referent of ‘is true’, but the problem I’m going to discuss will apply equally to

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0003

38

 

any other proposal. Thus consider the pragmatic proposal that ‘is true’ refers to the property of being accepted at the end of enquiry.¹ On this proposal, T represents the sentence ‘snow is white’ as having the property of being accepted at the end of enquiry. In order for the proposal to count as a version of representationalism, being accepted at the end of enquiry has to be treated, not just as an empty label, but as an objective property instantiated by some sentences and not by others. If the notion is made sufficiently precise, we can in principle have procedures for determining whether a given sentence instantiates this property, presumably by studying how enquiry evolves, how it will end and what sentences will be accepted at this point. I’m not suggesting that advocates of this account of truth would accept this construal of the proposal. My point is that this has to be regarded as possible in order for these ideas to generate a representationalist account of the meaning grounds of truth ascriptions. The problem we considered in the previous chapter for representationalist accounts of the meaning grounds of ethical sentences arose from the possibility that someone who thinks that there are actions that maximize overall utility but don’t satisfy ‘is morally right’, or actions that don’t maximize overall utility but satisfy ‘is morally right’, might still be counted as meaning by the sentence ‘killing one to save five is morally right’ what we mean by it. I want to suggest that the pragmatic account of truth faces a parallel problem. Consider someone who has undertaken a serious study of the evolution of enquiry and has come to the conclusion that at the end of enquiry the sentence ‘snow is white’ won’t be accepted. Suppose, however, that this speaker believes that ‘snow is white’ satisfies ‘is true’, as she understands the predicate. She has this belief because she has the belief expressed by ‘snow is white’, and her ascription of ‘is true’ is regulated by her beliefs in this way: she ascribes the predicate to a sentence just in case it expresses a belief of hers, although she is prepared to accept that this procedure can produce wrong results—the predicate might not be satisfied by a sentence expressing a belief of hers, or it might be satisfied by a sentence expressing a proposition she doesn’t believe. I want to suggest that in these circumstances we would want to count the speaker as meaning by T what we mean by it, in spite of the fact that her belief that ‘snow is white’ will not be accepted at the end of inquiry coexists with her belief that ‘snow is white’ satisfies the predicate ‘is true’, as she understands it. However, the argument goes, if ‘is true’, as we understand the predicate, referred to the property of being accepted at the end of enquiry, it would follow that someone who accepts T in these circumstances cannot mean by the predicate what we mean

¹ In spite of the label, it should be clear that this position is a version of representationalism, and it’s not the position I will be defending here.

 -   

39

by it. For this would be tantamount to believing that a sentence satisfies the predicate, as she understands it, and that it doesn’t instantiate its referent. These ideas generate an argument against the pragmatic account of the referent of ‘is true’ along the same lines as the open-question argument against the utilitarian account of the referent of ‘is morally right’: T1

Someone can mean by ‘is true’ what we mean by it even though they take a sentence to satisfy ‘is true’ but believe that the sentence won’t be accepted at the end of enquiry, or take a sentence not to satisfy ‘is true’ but believe that the sentence will be accepted at the end of enquiry. (Premise)

T2

If someone takes a sentence to satisfy ‘is true’ but believes that the sentence won’t be accepted at the end of enquiry, or takes a sentence not to satisfy ‘is true’ but believes that the sentence will be accepted at the end of enquiry, then ‘is true’, as understood by this speaker, doesn’t refer to the property of being accepted at the end of enquiry. (Premise)

T3

Someone can mean by ‘is true’ what we mean by it even though the predicate, as understood by them, doesn’t refer to the property of being accepted at the end of enquiry. (from T1 and T2)

T4

Someone can’t mean by ‘is true’ what we mean by it if the predicate has different referents as understood by them and as understood by us. (Premise)

Therefore: T5

‘is true’, as we understand it, doesn’t refer to the property of being accepted at the end of enquiry. (from T3 and T4)

At this point the advocate of the pragmatic account of the reference of ‘is true’ could replicate the move made by the advocate of the utilitarian account of the referent of ‘is morally right’. She could argue that, as witnessed by the case of ‘is water’, a predicate, as understood by a speaker, can refer to a property even though the speaker doesn’t realize that it does. Specifically, ‘is true’, as we understand it, can refer to the property of being accepted at the end of enquiry even if we don’t realize that it does. The situation I’ve outlined can fall under this description: the speaker believes that ‘snow is white’ satisfies ‘is true’ even though it fails to instantiate the property of being accepted at the end of enquiry because she doesn’t realize that this property is the referent of the predicate. But this is perfectly compatible with the hypothesis that ‘is true’, as she understands it, refers, as a matter of fact, to the property of being accepted at the end of enquiry. It follows from these considerations that Premise T2 has to be rejected. Taking this line involves subscribing to the view that ‘is true’, like ‘is water’, refers to the unique property, if there is one, that satisfies a certain condition TC,

40

 

where it’s an empirical question which property, if any, satisfies TC. The situation we are discussing arises, on this diagnosis, because being accepted at the end of enquiry might satisfy TC even if speakers are not aware that it does. I don’t know whether advocates of a representationalist account of ‘is true’ would find this picture at all attractive. Be this as it may, I want to argue that the proposal falls prey to a revised version of the open-question argument along the lines of what I presented in my discussion of ethical discourse. On this proposal, speakers might fail to be aware of the fact that ‘is true’ refers to the property of being accepted at the end of enquiry because the property acquires the status of the referent of the predicate by virtue of the fact that it satisfies TC, and this is an empirical fact of which speakers might fail to be aware. But the situation is different for the fact that ‘is true’ refers to the unique property, if there is one, that satisfies TC. Unlike the fact that the property of being accepted at the end of enquiry satisfies TC, this is a semantic fact. The claim that speakers are wrong about this would involve treating them as semantically confused. And other things being equal we should reject an account of the meaning of a term, as understood by a speaker, that forces us to ascribe to her semantic confusion.² It follows that if a speaker S believes that a sentence instantiates the unique TC-compliant property but doesn’t satisfy one of her predicates, ϕ, or that a sentence doesn’t instantiate the unique TC-compliant property but satisfies ϕ, we have to conclude that the referent of ϕ, as understood by S, isn’t singled out as the unique satisfier of condition TC. Hence, if we want to maintain that the referent of ‘is true’, as understood by us, is singled out as the unique TC-satisfier, we have to say that S doesn’t mean by ϕ what we mean by ‘is true’. I want to suggest that, as in the ethical case, there are circumstances under which this outcome would be unacceptable. Suppose S has the offending beliefs but has a policy of ascribing ϕ to one of her sentences if and only if she has the belief expressed by the sentence, but is prepared to accept that the satisfaction conditions of the predicate might not be correlated with which sentences express her beliefs. I want to claim that in these circumstances we would be perfectly happy to say that S means by ϕ what we mean by ‘is true’. And if this is so, it follows that the referent of ‘is true’, as we understand it, cannot be singled out as the unique TC-compliant property, if there is one, and a fortiori that ‘is true’ cannot refer to the property of being accepted at the end of enquiry, even if this property uniquely satisfies TC. This reasoning has, once more, the same structure as the argument we provided in the ethical case: T1*

A speaker can mean by a predicate what we mean by ‘is true’ even though she takes a sentence to satisfy the predicate but believes that the sentence doesn’t instantiate the unique satisfier of condition TC, or takes a sentence

² The qualifications to this claim introduced in Chapter 2 also apply here.

 -   

41

not to satisfy the predicate but believes that the sentence instantiates the unique satisfier of condition TC. (Premise) T2*

If a speaker takes a predicate to be satisfied by a sentence of which she believes that it doesn’t instantiate the unique satisfier of condition TC or takes the predicate not to be satisfied by a sentence of which she believes that it instantiates the unique satisfier of condition TC, then the referent of the predicate, as understood by the speaker, is not singled out as the unique satisfier of condition TC. (Premise)

T3*

A speaker can mean by a predicate what we mean by ‘is true’ even though the referent of the predicate, as understood by the speaker, is not singled out as the unique satisfier of condition TC. (from T1* and T2*)

T4*

If a speaker means by a predicate what we mean by ‘is true’ and the referent of the predicate, as understood by the speaker, is not singled out as the unique satisfier of condition TC, then the referent of ‘is true’, as understood by us, is not singled out as the unique satisfier of condition TC. (Premise)

Therefore: T5*

The referent of ‘is true’, as understood by us, is not singled out as the unique satisfier of condition TC. (from T3* and T4*)

I’ve argued that T2*, unlike T2, has to be accepted by those who construe the reference of ‘is true’ along the lines of the natural-kind-term model. Embracing the natural-kind model doesn’t save the pragmatic account of the reference of ‘is true’ from the open-question argument. Many advocates of the view that there’s a property that ‘is true’ refers to have abandoned the hope of specifying the property that plays this role with an explicit definition. They have lowered their expectations from the original goal of specifying: A condition C such that, for every sentence S, S satisfies ‘is true’ if and only if C(S) to the more modest goal of specifying: For every sentence S, a condition C such that S satisfies ‘is true’ if and only if C(S), where it is assumed that different conditions will play this role for different sentences. The more modest goal is expected to be achieved with the help of what is known as a Tarskian definition of truth. As we shall see later on, this formal resource has been championed by philosophers who don’t endorse a

42

 

representationalist account of the meaning grounds of truth ascriptions. Here I’m only interested in its potential for serving this purpose—for identifying the property that ‘is true’ refers to.³ A Tarskian definition of truth uses the inductive structure of the set of sentences of a language in order to define by recursion a function pairing each sentence with a truth value.⁴ Like any recursive definition, it proceeds in two steps. First, it specifies the image under the function of the items in the base of the inductive definition, in our case the atomic sentences of the language. Second, it specifies how the image of the output of each inductive clause is determined by the images of the input—how the truth value of a molecular sentence depends on the truth values of its immediate components.⁵ As readers who are familiar with standard fist-order semantics will realize, this basic picture has to be made more complex to deal with quantified sentences, which force us to assign semantic values to open formulas as well as sentences.⁶ However, none of this needs to be considered for our purposes. Nor do we need to consider how molecular sentences of any form are treated. We can concentrate exclusively of how the approach deals with atomic sentences, and to simplify matters even more we can concentrate on subject-predicate sentences. All we need to consider is how the approach proposes to pair each subject-predicate sentence with a unique truth value. The strategy with which this is achieved is very much in the spirit of representationalism. It invokes a function f pairing each term with its referent—each singular term with a particular and each predicate with a property.⁷ This enables us to specify the conditions under which a subject-predicate sentence ‘Pa’ will be true, which, for the purposes of providing a representationalist account of the meaning grounds of the predicate ‘is true’, can be specified as the following condition for its satisfaction: ‘Pa’ satisfies ‘is true’ if and only if f(‘a’) instantiates f(‘P’). The Tarskian definition of truth will provide a biconditional of this form for every sentence of the language. These biconditionals will accomplish the more modest goal of the approach to the referent of ‘is true’ under discussion. I want to argue that this proposal is liable to the same objection as the project of providing an explicit definition of the referent of ‘is true’. To see this, it’s important to realize that ‘f ’ isn’t just an abbreviation for the reference function. ³ This is in line with the way in which the Tarskian framework is used in (Field 1972). Harty Field later abandoned this approach. ⁴ See (Tarski 1936) for Tarski’s own version of the proposal. ⁵ For details, see, for example, (Zalabardo 2000a: 93–101). ⁶ Tarski’s main formal contribution in this connection was a technique for dealing with this problem. ⁷ Formal applications Tarski’s ideas follow his original presentation in pairing predicates, not with properties, but with sets of particulars. This difference won’t be relevant for our purposes.

 -   

43

In order for the theory to provide an account of how ascriptions of ‘is true’ represent things as being, f has to be a real functional relation between terms and their referents. And for a naturalist this means that it has to be a natural relation.⁸ So what the naturalist is proposing is that there is a natural relation f pairing each term with the item in the world that plays the role of its referent, and that f plays a crucial role in specifying how ascriptions of ‘is true’ represent things as being. Thus, with respect to T, the proposal is that this sentence represents f(‘snow’) (i.e. the individual substance to which the term ‘snow’ bears natural relation f) as instantiating f(‘is white’) (i.e. the property to which the predicate ‘is white’ bears natural relation f). Now consider a speaker who has identified relation f and has investigated which items it connects her terms with. Suppose that she comes to the conclusion that f(‘snow’) is grass and f(‘is white’) is the property of redness. She will then believe that f(‘snow’) doesn’t instantiate f(‘is white’). As we’ve just seen, on the proposal under discussion, T represents f(‘snow’) as instantiating f(‘is white’). Hence, our speaker believes things are not as T represents them as being, on this proposal. Suppose, however that, in spite of this, she believes that ‘snow is white’, satisfies ‘is true’ as she understands it. On the version of the open-question argument I’m now considering, the naturalist would have to conclude from this that the speaker doesn’t mean by ‘is true’ what we mean by it, since she believes that the predicate is satisfied by ‘snow is white’ even though she believes things are not as they would have to be in order for the sentence to satisfy the predicate, as we understand it. However, as before, there are perfectly normal circumstances in which we would want to say, to the contrary, that the speaker means by the predicate what we mean by it. This would be the natural thing to say if her ascriptions of ‘is true’ are regulated in the same way as ours. These considerations yield two premises with which we can replace T1 and T2 in the original argument to adapt the reasoning to the Tarskian approach: TT1

Someone can mean by ‘is true’ what we mean by it even though they take ‘snow is white’ to satisfy ‘is true’ but believe that f(‘snow’) doesn’t instantiate f(‘is white’), or take ‘snow is white’ not to satisfy ‘is true’ but believe that f(‘snow’) instantiates f(‘is white’).

TT2

If someone takes ‘snow is white’ to satisfy ‘is true’ but believes that f(‘snow’) doesn’t instantiate f(‘is white’), or takes ‘snow is white’ not to satisfy ‘is true’ but believes that that f(‘snow’) instantiates f(‘is white’), then ‘is true’, as understood by this speaker, doesn’t refer to a property that’s instantiated by a subject-predicate sentence ‘Pa’ iff f(‘a’) instantiates f(‘P’).

⁸ The proposal is much more plausible for mental items, terms of the language of thought, than for linguistic items. See (Field 1978).

44

 

As before, the naturalist might try to avoid the difficulty with the suggestion that reference might in fact be f even if speakers don’t realize that this is so. On this proposal, reference is the unique natural function, if there is one, satisfying condition TTC, and reference is f because f uniquely satisfies TTC. But the fact that f uniquely satisfies TTC is an empirical fact of which speakers might be unaware. On this approach, the reason why our speaker believes that ‘snow is white’ satisfies ‘is true’ although she believes that f(‘snow’) doesn’t instantiate f(‘is white’) is that she doesn’t realize that f uniquely satisfies TTC, and a fortiori that f(‘snow’) instantiating f(‘is white’) is what things would have to be like in order for ‘snow is white’ to satisfy ‘is true’. However, as before, the open-question argument can be adapted in reaction to this move. If the naturalist account of how the referent of ‘is true’ is singled out is correct, then a speaker S who means by a predicate ϕ what we mean by ‘is true’ can take ‘snow is white’ to satisfy ϕ even though she believes that f(‘snow’) doesn’t instantiate f(‘is white’), or take ‘snow is white’ not to satisfy ϕ even though she believes that f(‘snow’) instantiates f(‘is white’). We would get this situation if S doesn’t realize that f uniquely satisfies TTC. However, on the naturalist picture, S can’t mean by ϕ what we mean by ‘is true’ if she takes ‘snow is white’ to satisfy ϕ but believes that the image of ‘snow’ under the unique TTC-compliant function doesn’t instantiate the image of ‘is white’ under the unique TTC-compliant function, or takes ‘snow is white’ not to satisfy ϕ but believes that the image of ‘snow’ under the unique TTC-compliant function instantiates the image of ‘is white’ under the unique TTC-compliant function. This would involve treating S as semantically confused, and, as we saw in the previous chapter, we should in principle aim to avoid meaning ascriptions that have this consequence. The problem for the naturalist is that we think there are circumstances under which S should count as meaning by ϕ what we mean by ‘is true’, even if she has the offending beliefs. This will be the situation if the procedure she employs for deciding on the ascription of ϕ matches our procedure for deciding on the ascription of ‘is true’. We can use these ideas to generate a version of the open-question argument against the Tarskian version of the representationalist account of ‘is true’, by replacing T1* and T2* with: TT1*

Someone can mean by a predicate what we mean by ‘is true’ even though they take ‘snow is white’ to satisfy the predicate but believe that the image of ‘snow’ under the unique TTC-compliant function doesn’t instantiate the image of ‘is white’ under the unique TTC-compliant function, or take ‘snow is white’ not to satisfy the predicate but believe that the image of ‘snow’ under the unique TTC-compliant function instantiates the image of ‘is white’ under the unique TTC-compliant function.

 -    TT2*

45

If someone takes ‘snow is white’ to satisfy a predicate but believes that the image of ‘snow’ under the unique TTC-compliant function doesn’t instantiate the image of ‘is white’ under the unique TTC-compliant function, or takes ‘snow is white’ not to satisfy the predicate but believes that the image of ‘snow’ under the unique TTC-compliant function instantiates the image of ‘is white’ under the unique TTC-compliant function, then the referent of the predicate, as understood by this speaker, is not singled out as a property that’s instantiated by a subject-predicate sentence ‘Pa’ iff the image of ‘a’ under the unique TTC-compliant function instantiates the image of ‘P’ under the unique TTC-compliant function.

TT1* is as plausible as T1* or TT1, and for the reasons given above, adapting the natural-kind account of the reference of ‘is true’ to the Tarskian approach doesn’t license the rejection of TT2*. The resulting argument successfully establishes that the referent of ‘is true’, as understood by us, is not singled out as a property that’s instantiated by a subject-predicate sentence ‘Pa’ iff the image of ‘a’ under the unique TTC-compliant function instantiates the image of ‘P’ under the unique TTC-compliant function.

3.2 Propositional Attitudes I want to consider next sentences that ascribe beliefs, desires, and other propositional attitudes to subjects, as, for example: John believes that there’s milk in the fridge. Mary desires that the trains are running on time. These are declarative sentences that appear to represent things as being a certain way—to represent how things stand with John or Mary. They can be naturally parsed as having a subject-predicate structure, although the predicates—‘believes that there’s milk in the fridge’ and ‘desires that the trains are running on time’— might ultimately have to be construed relationally. On a representationalist account of the meaning ground of a propositionalattitude ascription, there has to be a property playing the role of referent with respect to its predicate. The ascription would then represent the referent of its subject (e.g. John, Mary) as instantiating the referent of its predicate. Which properties could play this role? I’m going to concentrate on an influential approach to this question, although the problems I’m going to raise for it will apply more widely. The approach I’m going to focus on is the view known as the Representational Theory of Mind

46

 

(RTM). RTM postulates the existence of a language-like medium of mental representation (the Language of Thought, or Mentalese) defined in neurological/ cognitive/computational/functional terms (Fodor 1975). According to RTM, propositional attitudes such as belief or a desire consist in the tokening in specific ways of Mentalese sentences. For our purposes we can concentrate on the syntactically simplest case of a belief that an object I’m perceptually presented with instantiates a property. According to RTM, I form the belief that the object I’m presented with is a table by ‘tokening in the belief mode’ a Mentalese term that refers to the property of being a table. Suppose we accept the RTM story—that forming a belief about an object I’m presented with consists in tokening a Mentalese predicate in the belief mode. This story calls for an account of how Mentalese terms obtain their semantic properties—specifically, what determines which property each Mentalese predicate refers to. The standard strategy for accomplishing this task is known as Information-Theoretic Semantics (ITS). According to ITS, the referent of a Mentalese predicate is singled out as the bearer of a causal/nomological relation to the tokening of the predicate in the belief mode. A Mentalese predicate P will refer to a property ∏ just in case it is a law that P is tokened in the belief mode under certain conditions if and only if the subject is presented with an instance of ∏. Notice that the link between a Mentalese predicate and its referent requires only that the predicate would be tokened in the belief mode if the subject were presented with an instance of its referent under certain conditions. Relativizing the link between predicates and referents to specific conditions is unavoidable. The reason is, in a nutshell, the possibility of false belief. A subject forms a false belief when she tokens in the belief mode a Mentalese predicate referring to a property that’s not instantiated by the object she’s presented with. To borrow an example from Jerry Fodor, when I form the false belief that the cow I’m looking at is a horse, I token in the belief mode a predicate that refers to the property of being a cow, even though the object I’ve been presented with doesn’t instantiate this property. And, presumably, this tokening will be covered by laws. We should expect that it will be a law that the predicate is tokened in the belief mode whenever I’m presented with a horsey-looking cow. Relativizing the link between predicates and referents to specific conditions is the standard strategy for making room for the possibility of a Mentalese predicate whose referent is not instantiated by some of the objects that trigger its tokening in the belief mode—e.g. a predicate that refers to the property of being a horse even though it’s tokened in the belief mode when the subject is presented, among other things, with a horsey-looking cow. The idea is that there will be conditions under which the tokening of a Mentalese predicate in the belief mode would only be caused by instances of its referent, even if, when the conditions don’t obtain, the tokening in the belief mode of the predicate can be caused by other objects.

 -   

47

Hence, according to ITS, there has to be a set of conditions C such that P refers to ∏ just in case P would be tokened in the belief mode under conditions C if and only if the subject were presented with an instance of ∏. Identifying conditions that can play this role is seen by advocates of ITS as the main obstacle confronting the approach. Several proposals have been considered, but they all seem to be subject to counterexamples.⁹ The RTM-ITS account of propositional attitudes rests on substantial empirical hypotheses. Belief and desire are widely seen as involved in the causation of behaviour. In a standard example of their role, my walking towards the fridge is caused by my desire to drink some water and my belief that there’s water in the fridge. On the RTM-ITS story, this account of the causes of my walking towards the fridge requires that my behaviour was caused by the tokening in the belief mode of a Mentalese sentence F that would be tokened in the belief mode under C conditions just in case there was water in the fridge and the tokening in the desire mode of a Mentalese sentence W that would be tokened in the belief mode under C conditions just in case I was drinking water.¹⁰ This story entails that the causes of my behaviour have a structure that can be described in these terms. It requires, for example that my cognitive mechanism contains an item F and a property B (the tokening of a Mentalese sentence in the belief mode) such that F instantiating B plays a causal role in my walking towards the fridge, and in precisely those other cases in which my behaviour is caused by the belief that there’s water in the fridge. Some philosophers have argued that these hypotheses are likely to be wrong (Churchland 1981; Stich 1983). But even if we concede that these hypotheses are correct, the RTM-ITS account of propositional attitudes faces additional difficulties. I’m going to argue that it can be challenged by adapting the version of the open-question argument I’ve levelled against representationalist accounts of the meaning grounds of ethical sentences and truth ascriptions. Suppose that we’ve reached the point where neuroscientists have located the items that RTM-ITS identifies with Mentalese predicates, the property B whose instantiation by one of these items RTM-ITS characterizes as its tokening in the belief mode and the conditions C under which, according to RTM-ITS, a Mentalese predicate would be tokened in the belief mode precisely by instances of its referent. Consider now Laura, a neuroscientist who is familiar with the items that RTM-ITS identifies with Mentalese predicates, and can determine when one of ⁹ In (Fodor 1992a), Jerry Fodor provides a critical survey of proposals put forward by other authors. In (Fodor 1992b) he presents his own proposal. For critical discussion, see (Rupert 2008). I raise some difficulties for ITS in (Zalabardo 1995). ¹⁰ As Paul Boghossian explains, this approach ascribes a fundamental role to belief in the explanation of content: ‘an informational semantics has to understand the fixation of content through its role in the fixation of belief’ (Boghossian 2003: 44).

48

 

these items, P, has property B and which property ∏ is such that P would instantiate B under C-conditions just in case the subject were presented with an instance of ∏. Consider the predicate ‘believes that’s a horse’, whose ascription to a subject represents her as believing that a visually presented object is a horse. According to RTM-ITS, the referent of this predicate can be specified as the instantiation of B by an item in the subject’s Mentalese vocabulary that would instantiate B under C-conditions just in case the subject was visually presented with a horse. Call this property Φ. Suppose that Laura observes cowboy Peter’s behaviour and comes to the conclusion that when he was looking at a horse the previous evening he satisfied the predicate ‘believes that’s a horse’. Suppose, in addition, that Laura has scans of Peter’s brain activity, which indicate to her very clearly that, at the time, B was not instantiated by any item that would instantiate B if Peter were presented with a horse under C conditions. In other words, she believes that Peter didn’t instantiate at the time the property, Φ, that RTM-ITS identifies as the referent of ‘believes that’s a horse’. Nevertheless, she is convinced that Peter satisfied the predicate. Hence, the argument goes, if we accept the RTM-ITS-account of the reference of ‘believes that’s a horse’, as we understand the predicate, we will have to conclude that Laura doesn’t mean by the predicate what we mean by it. However, it seems to me that we might be disinclined to draw this conclusion about Laura. Suppose that her procedures for ascribing ‘believes that’s a horse’ are wholly in line with ours— e.g. she ascribes it to subjects that find themselves in situations in which she would believe that’s a horse. She also expects the subjects to which she ascribes the predicate to behave in ways that would be conducive to the achievement of their goals if the object they are presented with were a horse. In particular, her ascription of the predicate to cowboy Peter satisfied this description. I want to suggest that in these circumstances we would have no inclination to give up the idea that Laura means by ‘believes that’s a horse’ what we mean by this predicate, even though she believes that it is satisfied by a subject of which she also believes that he doesn’t instantiate the property Φ that RTM-ITS designates as the referent of ‘believes that’s a horse’. According to the open-question argument, vindicating this verdict would require rejecting the RTM-ITS-account of the reference of ‘believes that’s a horse’. As in the applications of this line of reasoning we have considered previously, the advocate of RTM-ITS could protest that Φ might actually be the referent of ‘believes that’s a horse’, as meant by Laura, even if she doesn’t realize that the property plays this role—in the same way in which ‘is water’ might refer to the property of consisting mainly of H₂O molecules as understood by speakers who are not aware of this fact. Hence the fact that Laura believes that Peter satisfies ‘believes that’s a horse’ but doesn’t instantiate Φ is compatible with the property being the referent of the predicate, as RTM-ITS proposes, as meant both by us and by Laura.

 -   

49

But once again the challenge can be reinstated by shifting the target of the argument. The contingency of the link between predicate and property is achieved with an account according to which the referent of the predicate is singled out as the property, if there is one, that uniquely satisfies a certain condition X. If ‘believes that’s a horse’ refers to Φ, that’s because Φ uniquely satisfies X, but this is a contingent fact of which Laura may well be unaware. And in that case her belief that Peter satisfies ‘believes that’s a horse’ but doesn’t instantiate Φ is perfectly compatible with the predicate, as meant by her, actually referring to the property. However, the situation is different with respect to the claim that ‘believes that’s a horse’ refers to the property, if there is one, that uniquely satisfies X. Unlike the claim that ‘believes that’s a horse’ refers to Φ, this isn’t an empirical claim that may be true even if the subject doesn’t realize that it is. The claim that ‘believes that’s a horse’, as meant by Laura, refers to the unique satisfier of X can’t be plausibly accepted if Laura believes that Peter satisfies ‘believes that’s a horse’ and that he doesn’t instantiate the unique satisfier of X, as this would involve ascribing to Laura semantic confusion. However, we could easily describe circumstances in which Laura would have the proscribed combination of beliefs but we wouldn’t be inclined to abandon our conviction that she means by the predicate what we mean by it. And since it follows from the proscribed belief combination that the referent of ‘believes that’s a horse’, as Laura understands it, is not singled out as the unique satisfier of X, we would have to conclude that the referent of the predicate, as we understand it, is not singled out in this way either, and hence that the RTM-ITS account of the reference of the predicate is incorrect. We could easily extract from these considerations an instance of the argument form presented in previous cases.

3.3 Meaning I want to consider next sentences that ascribe meanings to linguistic expressions, as, for example: N

‘ “La neige est blanche”, as understood by Pierre, means that snow is white’

or K

‘ “Tisch”, as understood by Kurt, means table’.

I’m going to refer to sentences of this kind, ascribing a meaning to a linguistic expression, as understood by a speaker, as interpretations. Interpretations are declarative sentences which appear to many of us to represent things as being a certain way. They appear to represent how things stand with a linguistic

50

 

expression, as understood by a speaker. We can think of them as subject-predicate sentences. In our examples, the subjects would be ‘ “La neige est blanche” as understood by Pierre’ and ‘ “Tisch” as understood by Kurt’, and the predicates ‘means that snow is white’ and ‘means table’. I’m parsing them in this way in order to simplify matters, but as we’ll see when we consider them in more detail, they are more usefully analysed as having a relational form—representing an expression, as understood by a speaker, as related to some item in a certain way. According to RR,¹¹ an interpretation will succeed in representing the world only if the terms that figure in it have referents. Its predicate, in particular, will have to refer to a property. Then the interpretation will represent the referent of its subject as instantiating the referent of its predicate. N will represent ‘La neige est blanche’, as understood by Pierre, as instantiating the property that ‘means that snow is white’ refers to; K will represent ‘Tisch’, as understood by Kurt, as instantiating the property that ‘means table’ refers to. Which properties could play this role? One possibility endorsed by some early modern philosophers is to find these properties among items in the speaker’s phenomenology. Thus, for K, on which I’ll focus, one version of the approach would dictate that the property of meaning table resides in a connection between the term and a mental image.¹² Then K would represent the word ‘table’ as triggering in Kurt a mental image of a table. As we’ll see later on, this approach has important advantages, but it also faces fatal objections. It comes under sustained attack in the work of the later Wittgenstein. One of Wittgenstein’s lines of attack is based on the idea that the instantiation by ‘Tisch’, as understood by Kurt, of the property of meaning table has to determine which objects satisfy ‘Tisch’, as understood by Kurt. But a mental image, Wittgenstein argues, cannot achieve this. Suppose that the word ‘Tisch’ does trigger a table-shaped mental image. If it seems to Kurt that a chair is one of the objects the image depicts, there’s no obvious sense in which his verdict can be said to be wrong. The image certainly resembles the chair, in some respect. And in general, the image will resemble any object in some respect (Wittgenstein 2009: §139). Wittgenstein considers a possible retort to the effect that even though the mental image by itself doesn’t determine which objects satisfy the associated term, a mental image under a certain interpretation does achieve the requisite determination (Wittgenstein 2009: §141).¹³ But Wittgenstein rejects this move as ineffectual. Suppose that the interpretation of the image consists in a phenomenal item connecting the image with tables. But what form could this take? Suppose it’s ¹¹ See section 1.3, above. ¹² The view is generally associated with John Locke (Locke 1975: III, ii, 2). Some scholars have questioned the identification of Locke’s ideas with mental images, e.g. (Yolton 1996), but that’s certainly how they were understood by his critics, notably Berkeley (Berkeley 1988). ¹³ The notion of interpretation that figures in Wittgenstein’s argument is more general than the notion of a meaning ascription, for which I’m using the term interpretation here.

 -   

51

a composite mental image, including an image of the image being interpreted and an image of a table. Clearly the same problem arises here. If it seems to Kurt that a chair is one of the objects that the composite image depicts as depicted by the original image, there’s no clear sense in which Kurt could be said to be mistaken. And continuing on this route, with an interpretation of the interpretation, doesn’t seem to help matters. This brief summary of Wittgenstein’s rich discussion of the phenomenal approach to meaning properties falls well short of a cogent refutation, but it seems to me that Wittgenstein’s concerns can be developed in ways that would ultimately render the approach untenable.¹⁴ Another approach to the task of identifying the property that ‘means table’ refers to is to locate it in features of the way a term is used. On this approach, K would represent Kurt’s use of ‘Tisch’ as exhibiting this feature. Much of what Wittgenstein says about meaning might seem to recommend this approach, although I’ll argue later on that there’s a more promising way of understanding the connection that Wittgenstein sees between meaning and use. In order to develop this approach, we would need to identify the kind of feature of the use of a term that can be plausibly treated as a meaning property. What feature would have to be present in Kurt’s use of ‘Tisch’ in order for the term, as understood by Kurt, to mean table? A term like ‘Tisch’ can be expected to be used in a wide variety of contexts—e.g. in a sentence in which Kurt says what his favourite piece of furniture is, a sentence with which Kurt asks his roommate to clear the table, or even in a piece of poetry. However, accounts of the meaning of general terms based on the way they are used tend to focus on one particular aspect of their use—their ascription to objects in subject-predicate sentences, including those with demonstrative subjects. One straightforward way of developing this approach is to focus on the objects the speaker has actually ascribed the term to, and to say that the property of meaning table is the property of having ascribed the term precisely to tables. However, this proposal is inadequate. One problem it faces is that it entails that a term, as understood by a speaker, will have multiple incompatible meaning properties. We can dramatize the point using a concept introduced by Saul Kripke in his discussion of these ideas. Kripke defines the concept of a tabair as ‘anything that is a table not found at the base of the Eiffel Tower, or a chair found there’ (Kripke 1982: 19). Suppose that we say that the property of meaning M by term T is the property of having ascribed T precisely to Ms. Then, if Kurt has so far ascribed ‘Tisch’ precisely to the tables he’s encountered, but he hasn’t been under the Eiffel Tower, then ‘Tisch’, as understood by Kurt, has both the property of meaning table, and the property of meaning tabair. But ‘Tisch’, as understood by

¹⁴ For discussion of Wittgenstein’s ideas in this area, see my (Zalabardo 2003, 2009).

52

 

Kurt, can’t have both properties, as it can’t be both satisfied and not satisfied by a chair under the Eiffel Tower. One promising strategy for overcoming this difficulty is to shift from actual ascriptions of the term to the speaker’s disposition to ascribe it it—from the objects to which she has actually ascribed the term to the objects to which she is disposed to ascribe it. On this proposal, ‘Tisch’, as meant by Kurt, means table just in case Kurt is disposed to ascribe the term precisely to tables. This move seems to have the potential for solving the problem. Even if Kurt has never been under the Eiffel Tower, it’s still the case that, if he went there and saw a chair, he either would ascribe ‘Tisch’ to it or he wouldn’t. If it is the case that he wouldn’t, then ‘Tisch’, as meant by Kurt, would not have the property of meaning tabair. An influential version of the dispositional account arises from the application to meaning of the RTM-ITS approach to propositional attitudes discussed in the previous section. On this account each public-language term is associated with a Mentalese term from which it inherits its semantic properties. The association will consist in the (sincere) ascription of the public-language term being regulated by the tokening in the belief mode of the associated Mentalese term.¹⁵ Hence, in our example, ‘means table’ will refer to the following property: Having its ascription regulated by the tokening in the belief mode of a Mentalese term that would be tokened in the belief mode under C conditions just in case the subject were presented with a table. This account of the reference of ‘means table’ is unquestionably a species of the dispositional approach.¹⁶ ‘Tisch’, as understood by Kurt, will instantiate the referent of ‘means table’ just in case he is disposed to ascribe it to tables he is presented with under C conditions. For some philosophers, the dispositional account faces fundamental difficulties. One objection that has been raised at this level arises from the claim that meaning is a normative notion. The point was forcefully presented by Saul Kripke in his (Kripke 1982), in the course of assessing an account of what I mean by the symbol ‘+’ in terms of the responses I am disposed to give to questions of the form ‘m + n?’.¹⁷ Kripke presents the objection I want to focus on in the following passages: The dispositionalist gives a descriptive account of this relation [between the meaning I ascribed to ‘+’ in the past and how I am disposed to answer the question ‘68 + 57 = ?’]: if ‘+’ meant addition, then I will answer ‘125’. But this is

¹⁵ Another approach would derive the meaning of linguistic expressions from the content of Mentalese terms in terms of speakers’ audience-directed intentions (Grice 1957). ¹⁶ Paul Boghossian makes this point (Boghossian 1989: 527–8). ¹⁷ For the interpretation of Kripke’s reasoning on this point, see my (Zalabardo 1997).

 -   

53

not the proper account of the relation, which is normative, not descriptive. The point is not that, if I meant addition by ‘+’, I will answer ‘125’, but rather that, if I intend to accord with my past meaning of ‘+’, I should answer ‘125’. [ . . . ] the relation of meaning and intention to future action is normative, not descriptive. (Kripke 1982: 37) A candidate for what constitutes the state of my meaning one function, rather than another, by a given function sign, ought to be such that, whatever in fact I (am disposed to) do, there is a unique thing that I should do. Is not the dispositional view simply an equation of performance and correctness? (Kripke 1982: 24)

On the face of it, Kripke’s point is based on the general injunction against deriving an ought from an is, as presented in this famous passage of Hume’s Treatise: In every system of morality, which I have hitherto met with, I have always remark’d, that the author proceeds for some time in the ordinary ways of reasoning, and establishes the being of a God, or makes observations concerning human affairs; when all of a sudden I am surpriz’d to find, that instead of the usual copulations of propositions, is, and is not, I meet with no proposition that is not connected with an ought, or an ought not. This change is imperceptible; but is however, of the last consequence. For as this ought, or ought not, expresses some new relation or affirmation, ’tis necessary that it shou’d be observ’d and explain’d; and at the same time that a reason should be given; for what seems altogether inconceivable, how this new relation can be a deduction from others, which are entirely different from it. (Hume 1978: 469)

If the difficulty is real, it will invalidate the RTM-ITS approach, but advocates of the view have failed to see any merit in the objection. They accept that there are important obstacles for their project of providing an extensionally accurate semantics for Mentalese terms—one that pairs each predicate with the property it actually refers to. The need to make room for the normativity of meaning is usually construed as an additional difficulty, that would be faced even by an extensionally accurate semantics.¹⁸ Advocates of RTM-ITS reject this idea. They maintain that if the goal of an extensionally accurate semantics were achieved, the normative character of meaning would not pose an additional hurdle. Here’s how Fodor puts the point: [ . . . ] requiring normativity to be grounded suggests that there is more to demand of a naturalized semantics than that it provide a reduction of such

¹⁸ See (Boghossian 1989: 530–4).

54

  notions as, say, extension. But what could that ‘more’ amount to? To apply a term to a thing in its extension is to apply the term correctly; once you’ve said what it is that makes the tables the extension of ‘table’s, there is surely no further question about why it’s correct to apply a ‘table’ to a table. It thus seems that if you have a reductive theory of semantic relations, there is no job of grounding normativity left to do. In short, I’m not clear how—or whether—‘open question’ arguments can get a grip in the present case. I am darkly suspicious that the Kripkensteinian worry about the normative force of meaning is either a nonissue or just the reduction issue over again; anyhow, that it’s not a new issue. (Fodor 1992b: 135–6)

If the problem is posed in these terms, it seems to me that the advocates of RTMITS have the upper hand. Once we concede that they have provided an account that pairs each predicate with the property that as a matter of fact it refers to, I can’t see in what way the account falls short of specifying which predicate ascriptions are correct—to ascribe a predicate correctly, in the relevant sense, is surely nothing but to ascribe it to instances of its referent. I want to suggest, however, that there is a more promising way of construing the concern. As Fodor suggests, the objection can be seen as a version of the open-question argument, and we can easily adapt to this case the version of the argument that I outlined for the RTM-ITS account of propositional attitudes. Assume, as before, that neuroscience has located the items that RTM-ITS identifies with Mentalese predicates, the property B whose instantiation by one of these items RTM-ITS characterizes as its tokening in the belief mode and the conditions C under which, according to RTM-ITS, a Mentalese predicate would be tokened in the belief mode precisely when the subject is presented with an instance of its referent. Suppose that we have also identified the function f which, according to the RTM-ITS account of meaning, pairs each public-language term with the Mentalese term from which it inherits its semantic properties. Hence, according to RTM-ITS, ‘means table’ refers to the property: Having as its image under f an item that would instantiate B under C conditions just in case the subject were presented with a table. Call this property Ψ. Consider again Laura, our neuroscientist from the previous section. Suppose that she decides to practice her skills by conducting some experiments on Kurt, and comes to the conclusion that his term ‘Tisch’ is paired by f with item m, and that m would instantiate B under C conditions when Kurt is presented with a table not under the Eiffel tower or when he is presented with a chair found there. In this situation Laura believes that ‘Tisch’, as meant by Kurt, doesn’t instantiate the referent of ‘means table’, on the RTM-ITS account of what we mean by

 -   

55

this predicate—she believes that it instantiates instead the referent of ‘means tabair’, on the RTM-ITS account. Hence, the open-question argument goes, if Laura believes that ‘Tisch’, as understood by Kurt, satisfies the predicate ‘means table’, it will follow from the RTM-ITS account that Laura doesn’t mean by ‘means table’ what we mean by this predicate. I want to argue, to the contrary, that there are circumstances under which we would find this outcome completely unacceptable. Suppose that Laura has reached her conviction through extensive successful communicative interaction with Kurt involving his term ‘Tisch’, and in general the procedures that she employs for deciding on the ascription of ‘means table’ are wholly in line with the ones we use. I suggest that in these circumstances we would have no hesitation in treating Laura as meaning by ‘means table’ what we mean by this predicate. If RTM-ITS entails that this is the wrong verdict, we will conclude that RTM-ITS is incorrect. These reflections would support the analogue for ‘means table’ of the Moorean premise of the version of the open-question argument we’ve been considering: M1

Someone can mean by ‘means table’ what we mean by it even though they take a term, as understood by a speaker, to satisfy ‘means table’ but believe that the term doesn’t instantiate Ψ, or take a term, as understood by a speaker, not to satisfy ‘means table’ but believe that the term instantiates Ψ.

As in the applications of the open-question argument that we’ve considered previously, the reasoning can be resisted with the suggestion that ‘means table’ could refer to Ψ even if speakers don’t realize that this is the case. The fact that ‘means table’ refers to Ψ is knowable only through empirical research that speakers might not have undertaken. Hence from the fact that Laura believes that ‘Tisch’, as meant by Kurt, satisfies ‘means table’ and that it doesn’t instantiate Ψ, it doesn’t follow that ‘means table’, as understood by Laura, doesn’t refer to Ψ. ‘means table’, as she understands it, might refer to Ψ even though she doesn’t realize that it does.¹⁹ To put it in terms of the argument template we’ve been considering, in order to undermine the RTM-ITS account of the referent of ‘means table’, Premise M1 would have to be combined with: M2

If someone takes a term, as understood by a speaker, to satisfy ‘means table’ but believes that the term doesn’t instantiate Ψ, or takes a term, as understood by a speaker, not to satisfy ‘means table’ but believes that the term instantiates Ψ, then ‘means table’, as understood by this speaker, doesn’t refer to Ψ.

The advocate of the RTM-ITS picture can resist the reasoning by rejecting this premise.

¹⁹ Scott Soames uses this line of reasoning in (Soames 1998).

56

 

However, as on previous occasions, we can revive the challenge of the openquestion argument by adjusting its target. This rejoinder is based on a picture of the semantics of ‘means table’ according to which the predicate refers to the unique property, if there is one, that satisfies a certain condition X. Ψ qualifies as the referent of ‘means table’ because it uniquely satisfies X, and there’s no reason why Laura should know that this is the case. However, the challenge can be reinstated by running the argument replacing Laura’s belief that ‘Tisch’, as understood by Kurt, doesn’t instantiate Ψ with her belief that ‘Tisch’, as understood by Kurt, doesn’t instantiate the unique satisfier of X. Once again, we can describe situations in which Laura would believe that ‘Tisch’, as understood by Kurt, satisfies ‘means table’ but doesn’t instantiate the unique satisfier of X, and we would still want to say that Laura means by ‘means table’ what we mean by it. And now the advocate of RTM-ITS wouldn’t be able to object to the argument in the same way. This would provide support for the following premise: M1*

Someone can mean by a predicate what we mean by ‘means table’ even though they take a term, as understood by a speaker, to satisfy the predicate but believe that the term doesn’t instantiate the unique satisfier of X, or take a term, as understood by a speaker, not to satisfy the predicate but believe that the term instantiates the unique satisfier of X.

This would generate an argument against the RTM-ITS approach when combined with the premise: M2*

If a subject S takes a term, as understood by a speaker, to satisfy a predicate ϕ but believes that the term doesn’t instantiate the unique satisfier of X, or takes a term, as understood by a speaker, not to satisfy ϕ but believes that the term instantiates the unique satisfier of X, then the referent of ϕ, as understood by S, is not singled out as the unique satisfier of X.

And as in previous applications of the argument template, the advocate of the RTM-ITS approach will have to accept this premise. The fact that Ψ is the unique satisfier of condition X is an empirical fact of which speakers might be unaware, but the fact, if it is a fact, that ‘means table’ refers to the unique property that satisfies X, if there is one, cannot be treated as empirical—it can only be a semantic fact, and an interpretation that renders a speaker mistaken about this fact would involve ascribing to her semantic confusion. And this, as discussed, should other things being equal be treated as a refutation of the interpretation. I’ve claimed that Laura can mean by ‘means table’ what we mean by this predicate even if she believes that ‘Tisch’, as understood by Kurt, satisfies ‘means table’ and that it doesn’t instantiate the unique X-compliant property. In the presence of Premise M2*, it follows from this that even if Ψ uniquely satisfies X, ‘means table’, as understood by Laura and by us, cannot refer to Ψ.

 -   

57

3.4 Guidance Kripke’s discussion of dispositional accounts of meaning facts contains a line of reasoning that he sometimes appears to present as a version of the normativity argument, but should, in my view, be treated independently. It is presented in the following passage: So it does seem that a dispositional account misconceives the sceptic’s problem— to find a past fact that justifies my present response. As a candidate for a ‘fact’ that determines what I mean, it fails to satisfy the basic condition on such a candidate [ . . . ] that it should tell me what I ought to do in each new instance. Ultimately, almost all objections to the dispositional account boil down to this one. (Kripke 1982: 24)

The emphasis here is not on the contrast between the descriptive and the normative, but on an accessibility constraint. In the normativity argument, the leading thought is that the fact that bestows meaning on ‘Tisch’, as understood by Kurt, (the fact that would determine whether ‘Tisch’, as understood by Kurt, satisfies ‘means table’) has to determine which objects Kurt ought to apply the term to. In this passage the main idea seems to be different—that the fact that bestows meaning on ‘Tisch’, as understood by Kurt, has to tell him which objects he ought to apply the term to. This would seem to require that Kurt’s decision whether to ascribe ‘Tisch’ to an object can be guided by his conscious access to the fact that makes the term, as he understands it, mean what it means. If we accept this constraint, the referent of ‘means table’ has to be a property whose instantiation by ‘Tisch’, as understood by Kurt, can provide Kurt with conscious guidance for his ascriptions of ‘Tisch’. The approach that identifies meanings with items in the speaker’s phenomenology would seem to have the resources for satisfying this constraint. If the meaning of ‘Tisch’, as understood by Kurt, were a table-shaped mental image of his, it would be possible for him to consult it in order to decide whether to ascribe the predicate to an object—by considering whether the object resembles the mental image. Hence, if ‘means table’ referred to the property of a predicate being associated by the speaker with a table-shaped mental image, the guidance constraint would seem to be satisfied. If speakers’ dispositions concerning predicate ascription were construed in terms of what they feel inclined to do on each occasion on which the question arises, the dispositionalist might also have a story to tell about how predicate ascriptions can be guided by the meaning-determining property. However, if dispositions are construed in this way, identifying the referent of ‘means table’ with a disposition would leave no room for ascriptions of the term to be wrong— for the speaker to ascribe the term to an object that doesn’t satisfy it or to refrain from ascribing it to an object that does satisfy it. Making room for the possibility

58

 

of error forces the dispositionalist to construe dispositions in terms of how speakers would be inclined to respond under certain conditions. This is of course the strategy used by ITS to deal with the problem of false belief in its account of the semantic properties of Mentalese terms. But once dispositions are construed along these lines, it becomes hard to explain how Kurt’s decision on whether to ascribe ‘Tisch’ can be guided by his disposition. Even if there are conditions under which his responses are guaranteed to align with whether or not the term is satisfied, Kurt would normally have no idea what these conditions are, so his ascriptions of the term cannot possibly be guided by awareness of how he would respond under those conditions. The thought is particularly salient for the RTM-ITS story. On this approach, whether ‘Tisch’, as understood by Kurt, is satisfied by an object is determined by whether the Mentalese term with which ‘Tisch’ is paired by f would have property B under conditions C if Kurt were presented with the object. Kurt has no idea what f, B, or C are, let alone that they play the role that RTM-ITS assigns to them. Hence, it’s simply not possible for his ascriptions of ‘Tisch’ to be consciously guided by this fact. And if the referent of ‘means table’ is identified with a dispositional property of this kind, the guidance constraint won’t be satisfied. The dispositionalist has two strategies at her disposal for trying to overcome this difficulty. First, she could reject the guidance constraint that drives the objection, arguing that there is no reason to expect that meaning facts can guide speakers’ decisions on the ascription of terms.²⁰ The thought could be supported with the observation that there are uncontroversial cases, e.g. natural-kind terms, where speakers are not usually aware of the meaning-determining facts or of what decision they recommend on each particular case. Many English speakers are not aware of the facts about the molecular structure of a substance that determine whether it satisfies the term ‘is water’, and it would seem wrong to demand that speakers base their decisions on the ascription of the term on molecular analysis. Another possibility would be to invoke Wittgenstein’s rich discussion of the guidance requirement, which appears to suggest that the requirement is incoherent (Wittgenstein 2009). Alternatively, the dispositionalist could argue that when the guidance requirement is construed correctly, there’s no reason to deny that her account satisfies it. The proposal here would be to construe guidance in terms of causal influence. On this construal, our everyday applications of ‘is water’ can be said to be guided by a substance’s molecular structure, insofar as molecular structure is causally responsible for the surface criteria we usually employ to apply the term. In this sense of guidance, the advocate of ITS could argue, we might be able to say that Kurt’s applications of ‘Tisch’ to an object are guided by whether the Mentalese term with

²⁰ See (Wikforss 2001) for an attack on the guidance constraint.

 -   

59

which ‘Tisch’ is paired by f would have property B if Kurt were presented with the object under conditions C.

3.5 Conclusion In Chapter 1 I considered how the open-question argument can be deployed against the claim that ethical predicates have referents, and consequently that ethical sentences have representationalist meaning grounds. In this chapter I have considered how this line of reasoning can be adapted to the three semantic discourses we are interested in, to argue in each case that the central predicates of the discourse can’t have referents, and that its sentences can’t have representationalist meaning grounds. In the next few chapters, we are going to be concerned with the situation we face if this obstacle to the representationalist model can’t be overcome. In the presence of the RR assumption, this outcome would force us to abandon the idea that the sentences of our chosen discourses succeed in representing the world. After considering one last attempt to overcome the problems I’ve raised for representationalist meaning grounds, the next chapter will focus on treatments of our target discourses that accept that their sentences don’t have representationalist meaning grounds, and avoid falling foul of RR by maintaining that these sentences don’t have representation as their main function. After that, my main task will be to articulate the idea that, contrary to RR, the sentences of these discourses can have non-representationalist (i.e. pragmatist) meaning grounds and still succeed in representing the world.

4 Some Reactions In the previous two chapters I have considered some standard approaches to the provision of representationalist meaning grounds for the sentences of four philosophically important discourses. I have argued that these approaches face serious difficulties. Suppose that some or all of these approaches had to be abandoned. How could we react to this outcome? My goal in the present chapter is to consider some alternative approaches to these discourses that are compatible with the RR assumption—the claim that sentences that represent the world must have representationalist meaning grounds.

4.1 Pragmatist Representationalism The difficulties I raised in the previous two chapters for representationalist accounts of the meaning grounds of the sentences of the target discourses were presented as versions of the open-question argument. On my construal, the argument was based on the possibility that speakers attach to one of the target predicates the same meaning that we attach to it even though they believe that its satisfaction conditions are not aligned with the instantiation conditions of the property that the representationalist wants to treat as its referent. This is so because our judgements on who attaches to these predicates the same meaning that we attach to them appear to be governed by features of how the predicates are used. Thus, for example, if a speaker decides whether to ascribe ‘is morally right’ to an action on the basis of her feeling of moral approval and feels motivated to promote actions to which she ascribes the predicate, we will say that she means by the predicate what we mean by it, whether or not her ascriptions of the predicate are in line with her views on the instantiation of any independently defined property. The representationalist could try to save her approach from this line of attack by using the aspects of the predicate’s use that govern our synonymy verdicts to define the referent of the predicate. In this section I want to outline how this move would work for ethical discourse and the semantic discourses that interest us, and to consider the challenges faced by this approach.

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0004

 

61

4.1.1 Ethical Subjectivism As I’ve mentioned just know, my version of the open-question argument against naturalist accounts of the referent of ‘is morally right’ was based on the intuition that, so long as a speaker decides whether to ascribe ‘is morally right’ to an action on the basis of her moral sense and feels motivated to promote actions to which she ascribes the predicate, we will say that she means by the predicate what we mean by it. This is so irrespective of whether her ascriptions of the predicate are in line with her beliefs on the instantiation of any property that the naturalist might want to single out as the referent of ‘is morally right’. The naturalist might reply that there is a property to which the argument doesn’t apply—the property of triggering moral approval and motivation, or being endorsed, for short. Any speaker who, according to the argument, means by ‘is morally right’ what we mean by this predicate cannot fail to ascribe the predicate in line with (her views on) the instantiation conditions of this property. Hence, if we take the predicate to refer to the property of being endorsed, the open-question argument will no longer apply. One source of difficulties for this proposal is the fact that different actions are endorsed by different persons, and even by one person at different times. In other words, the property of being endorsed by me and the property of being endorsed by you are different properties, and the same goes for the property of being endorsed by me now and the property of being endorsed by me last year. Now, the property that the ascriptions of ‘is morally right’ by subject S at time t are bound to be in line with is the property of being endorsed by S at t.¹ However, the problem can’t be solved by taking ‘is morally right’, as understood by S at t, to refer to the property of being endorsed by S at t. On this proposal no two speakers ascribe the same meaning to the predicate, even if they use it in ways that make us say that they mean by the predicate what we mean by it. This outcome is clearly unacceptable. All apparent cases of moral disagreement would have to be taken instead as cases in which we talk at cross purposes. When I ascribe ‘is morally right’ to killing one to save five and you don’t, we are not really disagreeing—I’m representing the action as instantiating a property (being endorsed by me) and you are refusing to represent it as instantiating some other property (being endorsed by you). In the same way, all apparent changes in our moral opinions would have to be construed instead as changes in what we mean by our moral terms. The situation is not significantly improved by treating ‘is morally right’ as an implicitly relational predicate, ‘x is morally right for y’, where x is an action and y a moral assessor, and assuming that sentences in which the assessor is not mentioned treat the speaker implicitly as occupying that position: ‘Killing one to save

¹ I’m assuming that facts about which actions I endorse now are perfectly tracked by my beliefs.

62

 

five is morally right’ would have to be understood, on this proposal, as representing the action as bearing to the speaker the binary relation denoted by ‘is morally right’. We could now say that the predicate refers to the binary relation of x being endorsed by y. In this way, the predicate would refer to the same relation for you and me even if we endorse different actions. Nevertheless, the problem hasn’t been solved. The predicate may now have the same meaning as understood by you and by me, but it’s still the case that if I ascribe it to killing one to save five and you don’t, we are not disagreeing with each other. And when I go from ascribing the predicate to the action to not doing so, I’m still not changing my moral views.² Familiar moves for overcoming these difficulties include taking the predicate to refer to the property of being endorsed by most of us, or by the best of us, or under ideal conditions, but these proposals face equally familiar difficulties. It seems to be part of our understanding of the predicate ‘is morally right’ that most of us and even the best among us could be wrong about its instantiation. The point can be framed as a version of the open-question argument: a speaker who ascribes ‘is morally right’ to the actions she endorses will count as meaning by the predicate what we mean by it even if she believes that the predicate is satisfied by an action that is not endorsed by most of us or the best of us. And the appeal to ideal conditions faces a dilemma: if, on the one hand, whether conditions count as ideal is specified independently of which actions we endorse when the conditions obtain, then once again it’s part of our understanding of the predicate that we might reach the wrong verdicts in those conditions. A speaker who ascribes ‘is morally right’ to the actions she endorses will count as meaning by the predicate what we mean by it even if she believes of an action that it satisfies the predicate and that we wouldn’t endorse it under those conditions, or vice versa. If, on the other hand, we specify ideal conditions as those in which we would endorse instances of the referent of the predicate, we can’t non-circularly invoke these conditions to specify the referent of the predicate. Needless to say, these brief remarks don’t amount to a refutation of the approach, but it seems to me that the problems I’ve outlined can be developed into fatal objections.³ ² A similar approach applies to ethical discourse the framework developed by David Kaplan for indexical expressions (Kaplan 1989b). This involves taking the meaning of ‘is morally right’ to be a Kaplanian character—a function pairing contexts with referents, treating as the relevant feature of the context, in each case, the speaker’s sense of moral approval. For this contextualist approach, see (Glanzberg 2007). On this approach, if your sense of moral approval and mine are paired with different referents by the character of the predicate, differences between which actions you and I ascribe the predicate to will not correspond to disagreements about how things stand with the actions—the problem of disagreement has not been solved. Relativist views would seek to overcome the difficulty by maintaining that the sentence ‘killing one to save five is morally right’ expresses the same proposition independently of the sense of moral approval of each speaker, but this proposition has truth values relatively, e.g. to the speaker’s sense of moral approval. For relativist treatments of other discourses see (MacFarlane 2007; Egan 2006). James Dreier (Dreier 2009) has argued persuasively that this move doesn’t offer a successful solution to the problem of disagreement. ³ See (Blackburn 1993) for an interest discussion of the difficulties sketched here.

 

63

4.1.2 Protagorean Relativism The application of the open-question argument to truth ascriptions was based on the intuition that when a speaker ascribes the predicate ‘is true’ to a sentence just in case she has the belief expressed by the sentence, she will count as meaning by the predicate what we mean by it, even if her ascriptions of the predicate are knowingly out of line with the instantiation conditions of any sentence-property that the representationalist might designate as the referent of the predicate. This suggests that the problem could be avoided by picking as the referent of ‘is true’, as understood by S, the property of expressing one of S’s beliefs. This referent choice might seem to solve the problem, as speakers who mean by ‘is true’ what we mean by this predicate are bound to ascribe it in line with their beliefs about the instantiation conditions of this property. But this is an empty victory. On this proposal, we all attach different meanings to the predicate, and differences between the sentences to which the predicate is ascribed by different speakers, or by a speaker at different times, won’t count as substantive disagreements or changes of opinion, but as merely verbal disagreements or changes in meaning. As in the case of ethics, we could try to avoid the problem by taking the predicate to refer to the property of expressing a belief held by most of us or by the best of us, or a belief that we would hold under ideal conditions, but these proposals would face the same difficulties.

4.1.3 Interpretivism In Section 3.2 I considered the project of providing a representationalist specification of the meaning grounds of propositional-attitude ascriptions based on the account of propositional attitudes put forward by the Representational Theory of Mind and Information-Theoretic Semantics. Then in Section 3.3 I considered the application of these ideas to the case of meaning ascriptions. I argued that these proposals face versions of the open-question argument. In the example I used for belief ascriptions, neuroscientist Laura, who believes that cowboy Peter satisfies ‘believes that’s a horse’, as she understands it, and that he doesn’t instantiate the property treated by RTM-ITS as the referent of this predicate, can nevertheless count as meaning by ‘believes that’s a horse’ what we mean by this predicate. This will be so if her use of the predicate is relevantly similar to ours—e.g. if she decides on its ascription on the basis of what she would believe if she found herself in the subject’s situation and expects the subjects to which she ascribes it to behave in ways that would be conducive to the satisfaction of their goals if there was a horse in front of them. Similarly, I argued that Laura can believe that ‘Tisch’, as understood by Kurt, satisfies ‘means table’ and that Kurt’s term doesn’t instantiate the property designated by RTM-ITS as the referent of ‘means table’, and still

64

 

count as meaning by ‘means table’ what we mean by this predicate. We will think that this is the right verdict if the procedures that she employs for deciding on the ascription of ‘means table’ are broadly in line with the ones we use. In both cases, our conviction that Laura means by these predicates what we mean by them seems to be based on the similarity between the way she uses these predicates and the way we use them—between the procedures that she employs for deciding on the ascription of the predicates and the ones we employ. This suggests that the representationalist could try once more to save her approach by defining the referents of ‘believes that’s a horse’ and ‘means table’ in terms of the features of the predicates’ use that govern our decision to count Laura as meaning by the predicates what we mean by them. What are these features? We could invoke at this point two influential proposals concerning the procedures that we employ for ascribing propositional attitudes and linguistic meaning. I have in mind, for propositional attitudes, the ideas presented by Daniel Dennett under the label Intentional Stance, and for linguistic meaning, the procedure described by W. V. O. Quine and Donald Davidson under the label Radical Interpretation (Radical Translation, in Quine’s version). Dennett presents the Intentional Stance as a strategy that we employ for predicting the behaviour of an object. Belief and desire ascriptions have to be understood, according to Dennett, in terms of the role that they play in this strategy. Here’s how Dennett characterizes the strategy: Here is how it works: first you decide to treat the object whose behaviour is to be predicted as a rational agent; then you figure out what beliefs that agent ought to have, given its place in the world and its purpose. Then you figure out what desires it ought to have, on the same considerations, and finally you predict that this rational agent will act to further its goals in the light of its beliefs. A little practical reasoning from the chosen set of beliefs and desires will in many—but not all—instances yield a decision about what the agent ought to do; that is what you predict the agent will do. (Dennett 1987b: 17)

Radical Interpretation is a procedure for ascribing meanings to someone else’s linguistic expressions on the basis of evidence that seems frightfully limited, but is, according to Quine and Davidson, the only evidence ultimately at our disposal for all our meaning ascriptions. For the radical interpreter, Davidson writes, the evidence available is just that speakers of the language to be interpreted hold various sentences to be true at certain times and under specific circumstances. (Davidson 1973: 322)

And on the basis of this evidence the interpreter begins the process of assigning meanings using the procedure illustrated by Quine’s famous example:

 

65

A rabbit scurries by, the native says ‘Gavagai’, and the linguist notes down the sentence ‘Rabbit’ (or ‘Lo, a rabbit’) as a tentative translation, subject to testing in further cases. (Quine 1960: 29)

These ideas will play an important role in the accounts I’m going to defend of ascriptions of propositional attitudes and meanings, and we will discuss them in some detail in due course. For now, my goal is to consider in very general terms their potential for sustaining representationalist accounts of propositionalattitude and meaning ascriptions that overcome the difficulties posed by the open-question argument. Consider first Dennett’s Intentional Stance. Suppose that our verdicts on whether someone is ascribing beliefs and desires are based on whether they are applying the Intentional Stance, and in particular, we would count Laura as meaning by ‘believes that’s a horse’ what we mean by this predicate so long as her ascriptions of the predicate are governed by the rules that define the Intentional Stance. Then the representationalist could try to use the Intentional Stance to define the states of affairs that belief and desire ascriptions represent as obtaining and the properties that predicates such as ‘believes that’s a horse’ refer to. On this proposal, ‘believes that’s a horse’ would refer to property ∏, defined as follows: x instantiates ∏ just in case applying the Intentional Stance to x recommends ascribing to x the belief that that’s a horse. If the representationalist could treat ‘believes that’s a horse’ as referring to this property, the open-question argument would no longer seem to be a threat. Someone who believes that a subject satisfies ‘believes that’s a horse’ but doesn’t instantiate ∏ would not be applying the Intentional Stance, and there would be no reason for counting them as meaning by the predicate what we mean by it. We can use Davidson’s Radical Interpretation to generate a similar proposal for meaning ascriptions. Suppose that our verdicts on who counts as ascribing meanings to linguistic expressions are as a matter of fact based on whether the ascriptions follow the Radical Interpretation strategy. Thus, Laura would count as meaning by ‘means table’ what we mean by it just in case her decision whether to apply the predicate to a term, as understood by a speaker, is ultimately governed by the principles of Radical Interpretation. Then the representationalist could try to use Radical Interpretation to define the states of affairs that meaning ascriptions represent as obtaining and the properties referred to by meaning-ascribing predicates. On this proposal, ‘means table’ would refer to the property Φ defined as follows: x instantiates Φ just in case applying Radical Interpretation to x recommends interpreting x as meaning table.

66

 

Once again, this approach would enable the representationalist to circumvent the open-question challenge. If Laura believes that ‘Tisch’, as meant by Kurt, satisfies ‘means table’ but doesn’t instantiate Φ, she won’t be applying Radical Interpretation, and, subject to our assumptions, we would have no inclination to treat her as meaning by ‘means table’ what we mean by it. A serious assessment of these proposals would require presenting them in much more detail than I have provided. Here my goal is to discuss their prospects in broad outline. One source of difficulties is the claim that our verdicts on whether someone counts as ascribing propositional attitudes or meanings are really governed by our views on whether they are applying the Intentional Stance or Radical Interpretation. This presupposes that our procedure for ascribing propositional attitudes is actually governed by the Intentional Stance and our procedure for ascribing meanings is governed by Radical Interpretation. However, these assumptions have received important challenges, both in the case of the Intentional Stance (Stich 1981; Goldman 1989) and in the case of Radical Interpretation (Fodor and Lepore 1994). But even if we accept that the Intentional Stance and Radical Interpretation offer broadly accurate characterizations of the procedures that we employ for ascribing propositional attitudes and meanings, the proposals exhibit a shortcoming that I regard as fatal. The problem, in very general terms, is that neither the Intentional Stance nor Radical Interpretation offers characterizations of our procedure for ascribing propositional attitudes or meanings that are sufficiently specific to sustain the definitions that the representationalist requires. Consider Radical Interpretation first. The problem is that different interpreters who are applying the principles of Radical Interpretation, and hence, under our assumption, count as ascribing meanings, could end up supporting different and incompatible meaning ascriptions. This circumstance, in the case of translation, is the source of Quine’s claim that Radical Translation is indeterminate: manuals for translating one language into another can be set up in divergent ways, all compatible with the totality of speech dispositions, yet incompatible with one another. 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. (Quine 1960: 27)

Adapted to interpretation, the thought is that the principles of Radical Interpretation fail to single out a unique interpretation as correct for every term of every speaker’s language. There’s room for debate on the extent of the phenomenon. Quine’s arguments for indeterminacy have been subjected to intense scrutiny and have often been found wanting (Hylton 2007: ch. 8). Davidson expects the problem to be much

 

67

less widespread than Quine suggests (Davidson 1973: 327). However, I think it’s accepted by all parties that we shouldn’t expect that Radical Interpretation will be fully determinate. The reason for this is that the procedure takes as input features of the interpreter’s cognitive make-up—e.g. her sense of salience, simplicity, and projectability—that vary from interpreter to interpreter, and will affect the interpretations that the procedure yields as output. As a rough illustration of the point, compare a normal interpreter, for whom a chair under the Eiffel Tower is very different from any table, with an interpreter for whom a chair under the Eiffel Tower is obviously the same kind of thing as any table. Suppose now that they both have seen Kurt ascribe ‘Tisch’ to a variety of tables. The normal interpreter can be expected to interpret ‘Tisch’ as understood by Kurt, as meaning table, while the exotic interpreter can be expected to interpret the term as meaning tabair.⁴ And the disagreement won’t be resolved by seeing how Kurt reacts to a chair under the Eiffel Tower. If, on the one hand, Kurt ascribes ‘Tisch’ to the chair, the normal interpreter will take this as providing some evidence against her interpretation, but instead of abandoning it, she could treat the case as a mistake on Kurt’s part, without thereby departing from the methodology of Radical Interpretation. If, on the other hand, Kurt refrains from ascribing ‘Tisch’ to the chair under the Eiffel Tower, the exotic interpreter can legitimately hold on to her interpretation in the same way. The immediate consequence of this circumstance for ‘means table’ is that property Φ is not well defined. Whether applying Radical Interpretation to x recommends interpreting x as meaning table depends on who is doing the interpreting. Overcoming this difficulty would require including in the definition of the referent of ‘means table’ those aspects of the interpreter’s cognitive make-up that affect the interpretations she ends up with when she applies the procedure of Radical Interpretation. And here the representationalist faces a dilemma. On the one hand, she could do this independently for each interpreter, using in each case that interpreter’s cognitive make-up. Thus ‘means table’, as understood by Laura, would be satisfied by those terms for which applying Radical Interpretation would recommend to Laura interpreting them as meaning table. But clearly taking this route would result in an unappealing form of relativism— subjects with relevantly different cognitive make-ups would count as understanding ‘means table’ in different ways. When the normal interpreter and the exotic interpreter disagree on whether ‘means table’ is satisfied by ‘Tisch’, as understood by Kurt, this won’t count as a difference of opinion, but as a merely verbal disagreement. The normal interpreter will be representing ‘Tisch’, as understood by Kurt, as instantiating a property, and the exotic interpreter will be representing ‘Tisch’, as understood by Kurt, as not instantiating a different property.

⁴ See Section 3.3, above.

68

 

On the other hand, the definition of the referent of ‘means table’ as property Φ could be completed using features of a cognitive make-up that are regarded as ideal, where what counts as ideal is specified non-circularly—i.e. not in terms of whether it produces the right meaning ascriptions. But taking this route would reinstate the open-question challenge, so long as our verdicts on whether someone counts as ascribing meanings are still governed by whether they apply the method of Radical Interpretation, independently of whether their cognitive make-up is ideal in the relevant sense. If Laura’s ascriptions of ‘means table’ are governed by Radical Interpretation, but her cognitive make-up is non-ideal, she could believe (a) that ‘Tisch’, as understood by Kurt, satisfies ‘means table’ and (b) that the application of Radical Interpretation by the ideal interpreter would not recommend interpreting the term as meaning table—i.e. that the term doesn’t instantiate the referent of ‘means table’, under the current proposal. However, since her ascriptions of ‘means table’ are governed by Radical Interpretation, she would still count as meaning by the predicate what we mean by it. A similar situation applies for the representationalist account of the referent of ‘believes that’s a horse’ based on the Intentional Stance. Once again, there’s room for debate on the extent to which the application of the Intentional Stance by different interpreters can result in different and incompatible belief and desire ascriptions. Dennett himself expects it to be an improbable but real possibility (Dennett 1987b: 29). One reason for expecting some level of indeterminacy is that the Intentional Stance, like Radical Interpretation, requires input from the cognitive make-up of the interpreter that will affect which belief and desire ascriptions the application of the Intentional Stance will end up recommending to her. As in the case of ‘means table’, this means that the definition of the referent of ‘believes that’s a horse’ as property ∏ is incomplete—whether applying the Intentional Stance to x recommends ascribing to x the belief that that’s a horse will depend on who’s doing the interpreting. Once again, solving the problem would require building the requisite aspects of the interpreter’s make-up into the definition of the referent of ‘believes that’s a horse’, and once again the representationalist faces a dilemma: if on the one hand she does this independently for each interpreter, interpreters with different cognitive make-ups will ascribe different meanings to ‘believes that’s a horse’; if, on the other hand, she uses features of the cognitive make-up of an ideal interpreter, the resulting position is once again vulnerable to a version of the open-question argument.

4.2 The Flight from Representation In Chapters 2 and 3 and in Section 4.1 we have considered four instances of the following situation—discourses whose sentences appear to have the function of representing the world but the attempt to find referents for some of the central

 

69

predicates that figure in these sentences faces serious difficulties. Suppose that for some or all of these discourses we accept the claim that their sentences contain predicates that cannot refer. What conclusions should we draw from this? One immediate consequence is that the meaning grounds of these sentences cannot be specified along the lines of the version of representationalism on which I’ve focused. Suppose no other version of representationalism can improve on this. Then we’ll have to conclude that these sentences don’t have representationalist meaning grounds. In Chapter 1 I presented the assumption (RR) that a sentence can succeed in performing the task of representing the world only if its meaning ground is representationalist. With RR in place, accepting that the problematic sentences don’t have representationalist meaning grounds would force us to conclude that they are not capable of performing the task of representing the world. So long as we maintain that the function of the target sentences is to represent the world, we won’t be able to avoid the conclusion that these sentences fail to perform their function.⁵ If we endorse RR, but we think that a discourse can’t have representationalist meaning grounds, we can still avoid concluding that the discourse fails to perform its function by giving up the idea that its function is to represent the world and assigning to it an alternative, non-representational function. This move can be framed in two different ways. On the one hand, one could argue that we were wrong in thinking that the discourse had the function of representing the world— that in spite of its representational appearance, the discourse actually has a different function. On the other hand, we could accept that the discourse has the function of representing the world, but since it is incapable of discharging it successfully, it should be revised by assigning to it a non-representational function. If we reach a non-representational construal of a discourse by either of these routes, we will join those who were never under the impression that the discourse had the function of representing the world. In this section I want to present briefly some proposals for how to construe each of the four discourses we have been considering as having a non-representational function.

4.2.1 Ethical Non-cognitivism Non-representational construals of ethical discourse have been enormously influential since the 1930s. One of the early advocates was A. J. Ayer. Ayer uses a version of the open-question argument to reject the idea that ethical predicates ⁵ Some authors are prepared to accept this conclusion for some of the discourses we are interested in. This is the line taken by eliminative materialists concerning ascriptions of propositional attitudes (Churchland 1981; Stich 1983).

70

 

refer to natural properties, including properties defined in terms of feelings of approval (Ayer 1936: 104–5). He then goes on to reject ‘absolutist’ views, according to which ethical terms refer to irreducible, sui-generis properties, on the grounds that they make ethical knowledge impossible (Ayer 1936: 106). From the demise of these views, he concludes that ethical statements don’t represent the world. If I say ‘Stealing money is wrong’, he writes: I produce a sentence which has no factual meaning—that is, expresses no proposition that can be either true or false. [ . . . ] in saying that a certain type of action is right or wrong, I am not making any factual statement, not even a statement about my own state of mind. (Ayer 1936: 107)

However, Ayer doesn’t conclude from this that ethical statements are idle. The representational function they can’t perform is not their real function. Their function is to express certain feelings: in every case in which one would be commonly said to be making an ethical judgement, the function of the relevant ethical word is purely ‘emotive’. It is used to express feeling about certain objects, but not to make any assertion about them. (Ayer 1936: 108)

In addition, ethical statements have another non-representational function. Ethical terms, Ayer writes, ‘are calculated also to arouse feeling, and so to stimulate action’ (Ayer 1936: 108). In sum, he concludes: we may define the meanings of the various ethical words in terms both of the different feelings they are ordinarily taken to express, and also of the different responses they are calculated to provoke. (Ayer 1936: 108)⁶

Since the 1930s a wide range of non-representational accounts of ethical language have been developed. On one influential position, for example, the function of a statement ascribing an ethical predicate to an action is to express an attitude, not towards the action, but towards rules permitting or forbidding the action (Gibbard 1990). Another approach is to treat ethical statements as having a prescriptive function, similar to the function of imperatives (Carnap 1937; Hare 1952). In any case, my goal here is not to assess the relative merits of different nonrepresentational accounts of ethical language. Ethics here, and throughout this book, is only playing an illustrative role, as a prop for the general discussion of how to deal with discourses for which representationalist meaning grounds can’t

⁶ For the emotivist approach, see also (Barnes 1933; Stevenson 1937, 1944).

 

71

be provided. In-depth discussion of specific discourses will be restricted to the semantic cases—truth, meaning, and propositional attitudes. Nevertheless, two substantial challenges faced by ethical expressivism are worth mentioning here. The first, known as the Frege-Geach problem, concerns the treatment that the expressivist would give of the occurrence of ethical statements in non-asserted contexts (Geach 1960, 1965; Searle 1962). According to the expressivist, when I assert the sentence ‘killing one to save five is morally right’, I’m not representing things as being a certain way, but simply giving expression to my attitude of moral approval towards the killing of one to save five. However, ethical sentences also appear in non-asserted contexts, including truth-functional compounds such as: C

‘if killing one to save five is morally right, then killing one to save six is also morally right’.

The problem is that in its occurrence as the antecedent of C ‘killing one to save five is morally right’ is not giving expression to the speaker’s attitude of moral approval towards the action. Asserting C does not require having this attitude towards the action. The expressivist needs to provide an account of the meaning of ethical sentences in non-asserted contexts such as C, and this task is far from straightforward. One salient challenge is the need to preserve the logical relations between self-standing occurrences of moral statements and compound sentences in which they are embedded, as, for example, the logical entailment from ‘killing one to save five is morally right’ and C to ‘killing one to save six is morally right’. Expressivists have offered ingenious solutions to the problem.⁷ This is not the place to assess their prospects. In any case, it should be clear that an expressivist account of ethical discourse will be incomplete until a satisfactory treatment is offered of the meaning of ethical statement in non-asserted contexts. The second problem that I want to mention concerns the attitude in terms of which the expressivist seeks to specify the meaning grounds of ethical sentences. Take Ayer’s feelings of approval towards actions. One way to introduce the problem is to reflect that not all feelings of approval towards an action are aptly described as moral. No doubt I feel a certain kind of approval towards, say, Zinedine Zidane’s 2002 Glasgow volley, but this approval is not moral. Someone who decides on the ascription of a predicate on the basis of this kind of approval doesn’t mean by the predicate what we mean by ‘is morally right’. This poses a challenge for the expressivist. In Justin D’Arms and Daniel Jacobson’s words:

⁷ See e.g. (Blackburn 1984: 189–96; 1988; Horgan and Timmons 2006). See (Schroeder 2008) for a survey.

72

  as ‘good’, ‘right’, ‘beautiful’, et al. are not synonyms, expressivism must distinguish between the different varieties of evaluative sentiment. The expressivist needs both to identify a family of mental states expressed in value judgment and to single out those family members implicated in specifically moral evaluation. (D’Arms and Jacobson 1994: 740)⁸

The main challenge for the expressivist here is to avoid a circular account. Whether the approval that governs a speaker’s ascription of a predicate is of the moral kind cannot be made to depend on whether the predicate, as understood by the speaker, has the same meaning as ‘is morally right’, as understood by us.

4.2.2 Deflationism about Truth Truth ascriptions have also received non-representationalist construals. One approach follows the same basic pattern of expressivist accounts of ethical language, treating truth ascriptions as giving expression to the attitude of agreement or endorsement (Strawson 1949). Here I want to concentrate on a slightly different approach, known as deflationism. One of the central themes of the deflationist approach is the idea that the presence of the predicate ‘is true’ in a sentence doesn’t make any contribution to its representational value. As Ayer puts it: in all sentences of the form ‘p is true’, the phrase ‘is true’ is logically superfluous. When, for example, one says that the proposition ‘Queen Anne is dead’ is true, all that one is saying is that Queen Anne is dead. (Ayer 1936: 88)

A truth ascription, according to deflationism, doesn’t represent the item to which the truth predicate is applied as instantiating a property—its representational value is identical to that of the item to which the predicate is applied. From a representational point of view, the truth predicate is entirely redundant. But if ‘is true’ doesn’t play a representational role, what function, if any, does it have? Here deflationists invoke its utility for expressing attitudes towards propositions whose content we cannot specify. Paul Horwich, a leading proponent of the view, explains: In fact, the truth predicate exists solely for the sake of a certain logical need. On occasion we wish to adopt some attitude towards a proposition—for example, believing it, assuming it for the sake of argument, or desiring that it be the

⁸ See (Foot 2002b, 2002a; Wiggins 1998).

 

73

case—but find ourselves thwarted by ignorance of what exactly the proposition is. We might know it only as ‘what Oscar thinks’ or ‘Einstein’s principle’; perhaps it was expressed, but not clearly or loudly enough, or in a language we don’t understand; or—and this is especially common in logical and philosophical contexts—we may wish to cover infinitely many propositions (in the course of generalizing) and simply can’t have all of them in mind. In such situations the concept of truth is invaluable. For it enables the construction of another proposition, intimately related to the one we can’t identify, which is perfectly appropriate as the alternative object of our attitude. (Horwich 1998a: 2–3)⁹

Performing this logical function doesn’t require the existence of a property playing the role of the predicate’s referent. All that’s required is that the truth predicate abides by the disquotational schema: ‘p’ is true if an only if p. This enables us to get from: What Jonny said is true and What Jonny said = ‘we are out of milk’ to: We are out of milk. However, not all versions of deflationism can be plausibly construed as advocating a non-representational account of truth ascriptions. Horwich, in particular, explicitly rejects the non-representational view. His version of deflationism associates a definite propositional constituent with the truth predicate, which is involved in the meaning of ‘ “snow is white” is true’, but not of ‘snow is white’. On his view, the truth predicate is used ‘to give descriptions or make statements about the things to which it is applied’ (Horwich 1998a: 40). On the other hand, Horwich doesn’t seem to see representation as the fundamental function of truth ascriptions. Concerning the version of deflationism he endorses he writes: ‘This view is best presented together with a plausible account of the raison d’être of our notion of truth: namely that it enables us to compose generalizations of a special sort which

⁹ See also (Ramsey 1927).

74

 

resist formulation by means of the usual devices—the words “all” and “every” and, in logical notation, the universal quantifier’ (Horwich 1995: 358–9).

4.2.3 Instrumentalism about the Attitudes Propositional-attitude ascriptions lend themselves to a similar treatment. Ascriptions of belief and desire are routinely used for predicting behaviour. If I ascribe to you the desire to drink water and the belief that there’s water in the fridge, I will predict that you will walk towards the fridge. As we saw in Section 4.1.3, this predictive role is a central ingredient of Dennett’s Intentional Stance. We considered there the possibility of using the Intentional Stance to define the states of affairs that propositionalattitude ascriptions represent as obtaining. An alternative construal of the ideas deployed by Dennett would be to abandon the claim that propositional-attitude ascriptions have a representational function, and treat them instead as tools for the prediction of behaviour, to be assessed exclusively on the basis of the accuracy of the behavioural predictions they generate. On this proposal, by ascribing to you the belief that there’s water in the fridge I wouldn’t be representing how things stand with you in any respect, but simply engaging a tool that would enable me to predict your behaviour. Dennett himself rejects instrumentalist readings of his views, but for some critics his attempts to distance himself from instrumentalism do not succeed (Lycan 1988; Stich 1981). If propositional-attitude ascriptions are construed in this way, the unavailability of representationalist meaning grounds for them is no longer a problem.

4.2.4 Kripke’s Sceptical Solution A non-representational construal has also been proposed for meaning-ascribing discourse. The thought that meaning ascriptions don’t represent things as being a certain way, but have a different, non-representational function, is a central ingredient of the account of the discourse that Kripke characterizes as a sceptical solution to the sceptical problem posed by Wittgenstein’s rule-following considerations (Kripke 1982). The sceptical problem is this: there are no facts concerning what a speaker means by a linguistic expression. Hence there’s no state of affairs that the sentence ‘ “Tisch”, as understood by Kurt, means table’ represents as obtaining and no property that the predicate ‘means table’ refers to. One immediate consequence of this outcome is that it’s not possible to provide a representationalist account of the meaning grounds of meaningascribing discourse. Hence, if we want to save the idea that meaning ascriptions are meaningful, we need to provide an alternative, non-representationalist account of their meaning grounds. Kripke’s sceptical solution is such an account.

 

75

He presents the situation in terms of the contrast between the strategies for specifying the meaning grounds endorsed by the early Wittgenstein in the Tractatus Logico-Philosophicus (Wittgenstein 1974) and by the late Wittgenstein in the Philosophical Investigations (Wittgenstein 2009). In the Tractatus, according to Kripke, Wittgenstein defends a representationalist account of the meaning ground of a declarative sentence: The simplest, most basic idea of the Tractatus can hardly be dismissed: a declarative sentence gets its meaning by virtue of its truth conditions, by virtue of its correspondence to facts that must obtain if it is true. For example, ‘the cat is on the mat’ is understood by those speakers who realize that it is true if and only if a certain cat is on a certain mat; it is false otherwise. The presence of the cat on the mat is a fact or condition-in-the-world that would make the sentence true (express a truth) if it obtained. (Kripke 1982: 72)

If we identify the fact that must obtain if a sentence is true with the fact that the sentence represents as obtaining, this is a clear statement of the representationalist approach. It follows from the sceptical problem that this approach to specifying the meaning ground of a declarative sentence is not applicable to meaning ascriptions. If we want to resist the conclusion that meaning ascriptions are meaningless, we need a different account of their meaning grounds. Kripke suggests that a strategy that might work in this case can be found in Wittgenstein’s Philosophical Investigations. Quoting Michael Dummett, Kripke claims that the Investigations contains an implicit rejection of the representationalist approach of the Tractatus. Whereas the old strategy seeks to specify the meaning ground of a declarative sentence in terms of the state of affairs it represents as obtaining, the new strategy proposes to specify the meaning ground of a declarative sentence by answering two questions: first, ‘Under what conditions may this form of words be appropriately asserted (or denied)?’; second, given an answer to the first question, ‘What is the role, and the utility, in our lives of our practice of asserting (or denying) the form of words under these conditions?’ (Kripke 1982: 73)

Thus, on the new strategy, the meaning ground of a declarative sentence is given by (a) its assertibility conditions and (b) the role played in our lives by the resulting assertoric practice. Kripke’s proposal is to apply this strategy to the task of specifying the meaning grounds of meaning ascriptions. Thus, for the sentence ‘Jones means addition by “plus” ’, he offers first a specification of its assertibility conditions, which are different for Jones himself and for other speakers:

76

  Jones is entitled, subject to correction by others, provisionally to say, ‘I mean addition by “plus”,’ whenever he has the feeling of confidence—‘now I can go on’—that he can give ‘correct’ responses in new cases. (Kripke 1982: 90) Smith will judge Jones to mean addition by ‘plus’ only if he judges that Jones’s answers to particular addition problems agree with those he is inclined to give, or, if they occasionally disagree, he can interpret Jones as at least following the proper procedure. (Kripke 1982: 91)

Then he provides an account of the role played in our lives by an assertoric practice with these rules—one in which ‘Jones means addition by “plus” ’ is correctly asserted by other speakers when Jones’s answers to addition problems agree with ours. He illustrates this role in terms of an interaction between a customer and a grocer: The customer, when he deals with the grocer and asks for five apples, expects the grocer to count as he does, not according to some bizarre non-standard rule; and so, if his dealings with the grocer involve a computation, such as ‘68+57’, he expects the grocer’s responses to agree with his own. Indeed, he may entrust the computation to the grocer. (Kripke 1982: 92)

Asserting ‘the grocer means addition by “plus”’ in the conditions in which, on Kripke’s account, it is correctly asserted enables interactions of this kind to take place. The aspect of this approach that I want to highlight here is the fact that it involves ascribing a non-representational role to the sentences to which it is applied. Notice that it is in principle possible to endorse a strategy for specifying the meaning ground of a declarative sentence consisting only of the first of the two components of Kripke’s sceptical solution. On this strategy, the meaning ground of a declarative sentence would be provided by its conditions of warranted assertibility. This is the strategy advocated by Michael Dummett, on one reading of his ideas, as well as by philosophers in the 1960s and ’70s who sought to build a semantic theory on Wittgenstein’s notion of a criterion. We will discuss these views in some detail in the next chapter. Here I just want to point out that on these views the meaning ground of a sentence does not include a specification of its function. One can maintain that the meaning ground of a sentence consists in its assertibility conditions without having yet taken a stance on whether the function of the sentence is representational or non-representational, and specifically without ruling out that its function is to represent the world, or that it succeeds in performing this function. The point is clearly expressed by Crispin Wright, in reference to Dummett’s position: it is consistent with anti-realism in this sense to retain the idea that a discourse is representational, and answers to states of affairs which, on at least some proper understanding of the term, are independent of us. (Wright 1992: 5)

 

77

Kripke, by contrast, appears to think that specifications of meaning grounds in terms of assertibility conditions alone are bound to be incomplete. He writes: granted that our language game permits a certain ‘move’ (assertion) under certain specifiable conditions, what is the role in our lives of such permission? Such a role must exist if this aspect of the language game is not to be idle. (Kripke 1982: 75)

Notice that the threat of idleness only arises if the representational function has been ruled out. Kripke thinks that the threat is real, because the sceptical problem prevents us from treating meaning ascriptions as performing the function of representing the world. Hence, for Kripke, avoiding idleness requires ascribing to the discourse some other non-representational function.

4.3 Representation Regained We’ve been considering four regions of declarative discourse for which the provision of representationalist meaning grounds faces substantial difficulties. If these difficulties can’t be overcome, it follows from RR that the sentences of these discourses cannot successfully discharge the function of representing the world. One way one can react to this situation is to abandon the idea that the discourse in question has the function of representing the world, and ascribe to it a different, non-representational function that can be successfully discharged by sentences for which representationalist meaning grounds cannot be supplied. In Section 4.2 we have reviewed some proposals for taking this line with respect to each of the discourses we’ve been considering. I haven’t discussed the prospects of these positions in any detail. Some or all of them might face insurmountable obstacles. For our purposes, their most salient feature is the fact that they construe the target sentences as not having the function of representing the world. I regard this as a major disadvantage of these views. It seems to me that on our intuitive conception of these discourses, they aim at representing the world, and they succeed in doing so, no less than the discourses for which representationalist meaning grounds can be provided. In my view, other things being equal, we should aim to vindicate this intuitive conception. One of my main goals in the present book is to develop a strategy for achieving this. I’m not claiming that this intuition is sacrosanct, and that any position that can’t accommodate it should be summarily dismissed. It is perfectly possible that philosophical argument forces us to accept that abandoning this intuition is the least bad option. My main goal up to this point has been to present a powerful line of reasoning for this conclusion. Unless the argument is shown to be unsound, the outcome will have to be accepted. All I’m claiming is that if there is a position

78

 

that enables us to construe these discourses as successfully representational, then other things being equal this position should be preferred to alternatives lacking this feature. I realize that this intuition is not universally shared. One might think that a non-representational construal of some or all of these discourses should be treated as a desideratum, not as a conclusion that’s forced upon us by philosophical reasoning. I’m not going to try to defend the representational intuition against this line. In what follows I’m going to treat it as an assumption. My question will be: assuming that we would like to construe these discourses as successfully performing the function of representing the world, how, if at all, could this be achieved once we have accepted that their sentences don’t have representationalist meaning grounds? I’m going to close this chapter by considering a line of reasoning to the effect that construing a discourse as having a non-representational function doesn’t prevent us from treating its sentences as representing the world. This line of reasoning has been championed by Simon Blackburn, who uses for the resulting position the label quasi-realism.¹⁰ The line of reasoning takes as its starting point the deflationist construal of semantic language, especially of truth ascriptions. As we saw in Section 4.2.2, according to deflationism, truth ascriptions don’t have any representational content other than what belongs to the item to which the truth predicate is ascribed. The sentence A.

‘ “the cat is on the mat” is true’

doesn’t have any representational content other than what’s already present in ‘the cat is on the mat’. A doesn’t represent things as being a certain way regarding the sentence ‘the cat is on the mat’. It represents things as being a certain way regarding the cat and the mat. Hence, so long as we are happy to assert ‘the cat is on the mat’, we should be equally happy to assert ‘ “the cat is on the mat” is true’. But if we accept this construal of truth ascriptions, there’s no obvious reason for refusing to accord the same treatment to sentences that appear to be interchangeable with ‘ “the cat is on the mat” is true’, as, for example: B.

‘things are as “the cat is on the mat” represents them as being’.

Once again, the deflationist will want to say that the representational content of this sentence is identical to the representational content of ‘the cat is on the mat’, and anyone who is happy to assert the latter should have no hesitation in asserting ¹⁰ See e.g. (Blackburn 1980). James Dreier provides another eloquent presentation of the line of reasoning for the case of ethical discourse (Dreier 2004). See also (Gibbard 2012).

 

79

the former. Furthermore, the following seems to be a straightforward logical consequence of B: C.

‘ “the cat is on the mat” represents things as being a certain way’.

So long as the truth predicate can be felicitously ascribed to a sentence, the deflationist seems committed to treating the sentence as performing the function of representing the world.¹¹ Now, the truth predicate can’t be meaningfully or even grammatically ascribed to every sentence. It can’t be ascribed to, say, interrogatives or imperatives. Even for some declarative sentences its applicability is controversial. For declarative sentences expressing what we regard as subjective taste, the truth predicate doesn’t seem entirely appropriate. Someone who is happy to assert ‘Éric Rohmer’s films are dull’ may understandably balk at ‘ “Eric Rohmer’s films are dull” is true’. However, in the discourses we’ve been considering, ascription of the truth predicate is entirely felicitous. Take the sentences: ‘Killing one to save five is morally right’, ‘ “Tisch”, as understood by Kurt, means table’, ‘Lorna believes there’s water in the fridge’, or even ‘ “the cat is on the mat” is true’.¹² The question whether one of these sentences is true is intuitively appropriate and appears perfectly intelligible. If truth ascriptions had representationalist meaning grounds, these intuitions might turn out to be wrong. This would be so if these sentences were not among the possible instances of the referent of ‘is true’. However, for the deflationist this is not a possibility. The deflationist doesn’t have any grounds for excluding from the range of applicability of the truth predicate sentences like these to which we are intuitively happy to apply it.¹³ But then we can run with any of these sentences ¹¹ Clearly, C could be derived from ‘ “the cat is on the mat” is false’ in the same way as from A. ¹² I think that this form of vindication of the representational character of truth ascriptions is very much in the spirit of Horwich’s endorsement of this point. See Section 4.2.2, above. ¹³ As Frank Jackson, Graham Oppy, and Michael Smith have argued, this step in the quasi-realist argument depends on a deflationist conception, not of truth, but of truth aptness, and the former doesn’t entail the latter (Jackson, Oppy, and Smith 1994). Jackson, Oppy, and Smith have argued against deflationism about truth aptness by invoking the platitude that a sentence normally counts as truth apt only if it can be used to give the content of a belief, and endorsing a non-deflationist conception of belief. If their argument worked, the quasi-realist would not succeed in establishing that discourses with primarily non-representational functions can also perform the function of

80

 

the argument that we presented for ‘the cat is on the mat’, and conclude that all of them represent things as being a certain way—that they have the function of representing the world and successfully discharge it.¹⁴ My complaint against non-representational construals of the problematic discourses is that they fail to accommodate the intuition that their sentences succeed in representing the world. But now it seems that on the deflationist picture sentences with non-representational functions qualify for the task of representing the world to the same extent as sentences with representationalist meaning grounds. The deflationist conditions for representing the world are equally satisfied by both kinds of sentence. Does this line of reasoning invalidate my complaint? I want to argue that my complaint is not adequately addressed by the quasirealist position, as it fails to do full justice to the representational intuition. On the quasi-realist construal of a region of discourse, its sentences perform two functions. On the one hand, they perform the non-representational function they are assigned in order to deal with the consequences of the impossibility of specifying representationalist meaning grounds for them. On the other hand, they perform the function of representing the world, since, on the deflationist picture, there are no substantive conditions for performing this function that the problematic sentences might fail to satisfy. But the two functions that these sentences perform, on the quasi-realist picture, are not on a par. The non-representational function is a condition for the sentences to have the meaning they have—no sentence that fails to perform this function can have the same meaning at the target sentences. In other words, the non-representational function is treated as part of the meaning grounds of the sentences. Their representational function, by contrast, is entirely adventitious. It’s a function that they count as performing as a result of the fact that on the deflationist account of truth ascriptions there are no substantive conditions that a declarative sentence has to satisfy in order to count as representing the world, so long as speakers are happy about applying the truth predicate to it. However, this isn’t the situation across the whole of language. Representationalist meaning grounds don’t include the performance of a non-representational function among their conditions. Hence sentences with representationalist meaning grounds have representing the world as their sole function. It follows that the adoption of deflationism still leaves an important difference in the way in which the two types of sentence are related to the representational function. Sentences with representationalist meaning grounds have representation as their representing the world. I argue below that even if their argument fails the quasi-realist strategy doesn’t succeed in vindicating the representational character of these discourses. For cogent criticisms of Jackson, Oppy, and Smith’s arguments, see (Hawthorne and Price 2011). ¹⁴ Sometimes, Blackburn suggests that the quasi-realist should reject the idea that ethical sentences represent. See (Blackburn 2006: 160). As Jamin Asay indicates, it’s hard to see how this move could be justified (Asay 2013: 218).

 

81

only function.¹⁵ Representing the world is all they do. For sentences with a quasirealist construal, by contrast, the representational function is secondary. First and foremost they have a non-representational function without which they wouldn’t have the meaning they have. Their representational function, by contrast, doesn’t have this status. Two metaethicists who ascribe to ethical sentences the same nonrepresentational role will agree on the meaning of these sentences, even if one is a deflationist about truth who thinks that ethical sentences play, in addition, a representational role and the other thinks that there are substantive conditions on representation that ethical sentences fail to satisfy. I’m claiming that, so long as a non-representational function is treated as part of the meaning ground of a sentence, we haven’t done justice to the intuition that its involvement with representation is on a par with that of a sentence whose meaning ground doesn’t include the performance of any non-representational function. Genuine representation, I maintain, requires meaning grounds that don’t include the performance of a non-representational function. Therefore, a quasi-realist construal of a discourse doesn’t allow us to treat its sentences as genuinely representational.¹⁶

4.4 Conclusion In Chapter 2 and Chapter 3 I argued against some standard approaches to the provision of representationalist meaning grounds for the sentences of our four target discourses. In the present chapter I have considered some strategies we might adopt if we had to conclude that these approaches are unsuccessful but wanted to hold on to the assumption (RR) that representation requires representationalist meaning grounds. I have considered first the project of defining the referents of the central predicates of the target discourses in terms of the features of their use on which we base our verdicts regarding when someone means by their predicates what we mean by these. I have argued that this approach fails to provide satisfactory construals of the target discourses. Then I have considered views that seek to assign to the sentences of the target discourses alternative non-representational functions whose performance does not require representationalist meaning grounds. I have then discussed the idea that if we adopt a deflationist account of

¹⁵ Notice that the view that ethical statements have no function other than representing the world is not in conflict with the anti-Humean claim that ethical beliefs can motivate (Nagel 1970; McDowell 1979; Shafer-Landau 1998). One might hold that an ethical statement has as its only function representing a moral state of affairs and that belief in the obtaining of this state of affairs has the power to produce motivation. ¹⁶ My complaint also applies, I think, to the combination of semantic minimalism and functional pluralism advocated by Huw Price (Price 2011c).

82

 

truth, sentences with non-representational functions can still be treated as representing the world, as on the deflationist picture there are no substantive conditions that a declarative sentence would have to satisfy in order to count as succeeding in representing the world. I have argued that this strategy fails to provide an effective vindication of the representational character of the target sentences. I have argued that in order to treat a declarative sentence as fully representational its meaning ground can’t include the performance of a non-representational function. Representationalist meaning grounds fit the bill, but we are assuming that for the sentences of our target discourses, representationalist meaning grounds are not to be had. Hence, vindicating the intuition that these sentences represent the world would require ascribing to them meaning grounds that are not representationalist but don’t include the performance of a non-representational function. Showing how this can be achieved will be my goal in the next chapter.

5 Pragmatist Meaning Grounds Over the previous three chapters we’ve considered some powerful reasons for thinking that it won’t be possible to specify representationalist meaning grounds for the sentences of our four target discourses. We’ve then discussed a reaction to this outcome that consists in abandoning the idea that these discourses have the function of representing the world, and assigning to them some other nonrepresentational function. We’ve then seen that advocates of this line can invoke the deflationist account of truth to argue that assigning a non-representational function to a discourse is perfectly compatible with assigning to it, in addition, the function of representing the world. I’ve complained that this line fails to do justice to our intuition that the function of these sentences is genuinely representational. Sentences that are treated in this way will have a non-representational function as part of their meaning grounds. By contrast, representationalist meaning grounds do not include a nonrepresentational function. This invites the impression that this position only enables us to speak of the problematic sentences as if they performed the function that sentences with representationalist meaning grounds genuinely perform. Under the uniform representational surface we still have the contrast between genuinely representational sentences and sentences that are fundamentally in some other business. Avoiding this situation would require finding meaning grounds for the problematic sentences that are not representationalist but don’t ascribe to them a nonrepresentational function. My main goal over the next four chapters will be to articulate an approach to the meaning grounds of the problematic sentences that fits this description, and to present a strategy for overcoming the main obstacle faced by this approach.

5.1 Assertibility Conditions One extant strategy for specifying the meaning grounds of declarative sentences appears to fit the bill—non-representationalist meaning grounds that don’t ascribe to the target sentences a non-representational function. It consists in specifying meaning grounds in terms of conditions of warranted assertibility. The idea of including in a sentence’s meaning ground the conditions in which it is correctly asserted is already familiar to us, as it is part of the approach that Kripke applies to meaning ascriptions in his sceptical solution. However, as we saw in Section 4.2.4, Kripke proposes a two-fold strategy, with the meaning

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0005

84

 

ground of the target sentence specified in terms of its assertibility conditions and a non-representational function for the practice of asserting the sentence in those conditions. What we are interested in here is the possibility of dropping the second component of Kripke’s approach and specifying the meaning ground of a sentence exclusively in terms of its assertibility conditions. Proposals along these lines were put forward by Michael Dummett and his followers, under the label antirealist semantics, as well as by other philosophers who took inspiration for their semantic theory from Wittgenstein’s concept of a criterion.¹ I’m going to use the label justificationism to refer to the approach.² The best way to introduce the jusificationist approach is to take as our starting point the verificationist strategy for specifying the meaning ground of a sentence, with which the jusificationist approach is sometimes confused. According to verificationism, the state of affairs that a sentence represents as obtaining is defined in terms of the procedures that speakers use for deciding on the assertion of the sentence—it’s the state of information that would make speakers assert the sentence. Thus, in the example we considered in Section 4.1.1, the sentence ‘killing one to save five is morally right’ would represent the state of affairs of the speaker (or her community) feeling moral approval towards the killing of one to save five. This, as we saw, is a version of representationalism. It only differs from more standard versions in the identity of the state of affairs in terms of which the meaning ground of the sentence is specified. The jusificationist approach would also specify the meaning grounds of sentences in terms of the procedures employed by speakers for deciding on their assertion. In our example, the meaning ground of ‘killing one to save five is morally right’ would be specified in terms of the state of affairs of the speaker feeling moral approval towards the killing of one to save five. The difference with the verificationist approach consists in the relation it postulates between the sentence and the state of affairs. According to verificationism, the speaker feeling moral approval towards the action is the state of affairs that the sentence represents as obtaining, or its truth condition. According to justificationism, by contrast, this state of affairs is not the truth condition of the sentence, but its assertibility condition, the condition under which the sentence is warrantedly assertible: asserting the sentence will be warranted just in case the speaker feels moral approval towards the action. Of course, sentences with representationalist meaning grounds will also have assertibility conditions, but their status will depend on a contingent relation to the

¹ For Dummett’s assertibility-based semantics, see e.g. (Dummett 1976). In some of his writings, Dummett appears to have no quarrel with a truth-conditional account of meaning, so long as the notion of truth is epistemically constrained. Crispin Wright highlighted the connections between antirealist ideas and Wittgenstein’s notion of a criterion in (Wright 1982), but later distanced himself from this approach (Wright 1984). For views based on Wittgenstein’s notion of a criterion, see (Pollock 1967; Lycan 1971). ² I’m borrowing the term from Michael Dummett (Dummett 2005).

  

85

truth conditions of the sentences—on whether there exists an objective evidential link to the obtaining of the truth condition. For the jusificationist approach, by contrast, assertibility conditions don’t acquire their status in this way. The fact that assertion of a sentence is warranted in those conditions is treated as part of the meaning of the sentence. In our terminology, the proposal is that the assertibility condition constitutes the meaning ground of the sentence. What makes the sentence have the meaning it has is the fact that its assertion in those conditions is warranted. The same strategy can be applied to other problematic discourses. Thus, for example, the meaning ground of ‘“Tisch”, as understood by Kurt, means table’ would consist in the fact that asserting the sentence is warranted just in case applying Radical Interpretation to ‘Tisch’, as understood by Kurt, recommends interpreting it as meaning table. Similarly, the meaning ground of ‘“snow is white” is true’ would consist in the fact that asserting the sentence is warranted just in case asserting ‘snow is white’ is warranted. And the meaning ground of ‘Peter believes that’s a horse’ would consist in the fact that asserting the sentence is warranted just in case the Intentional Stance recommends ascribing to Peter the belief that that’s a horse. The jusificationist approach has some important advantages. Notice, first, that the open-question line of reasoning doesn’t seem to pose a problem here. To apply this argumentative strategy to the jusificationist approach, we would need to defend the possibility of speakers who use the same procedure that we use for deciding on the assertion of the target sentences, with the conditions of warranted assertibility of their sentences different from ours. It’s hard to see how one could construct a cogent case for the possibility of this kind of situation, since the conditions under which assertion of a sentence is warranted can be expected to be determined, on this approach, by speakers’ inclinations concerning when to assert the sentence. Second, the proposal avoids the difficulties encountered by verificationist versions of representationalism that define the state of affairs represented by a sentence in terms of the procedures employed by speakers for deciding on its assertion. Thus, for example, in the ethical case, we saw that, on the view that ‘killing one to save five is morally right’ represents the killing of one to save five as triggering moral approval, the sentence will represent different states of affairs as understood by different speakers who use the same procedure for deciding on its assertion. The jusificationist approach, by contrast, doesn’t yield this consequence. Even if I feel moral approval towards the killing of one to save five but you don’t, it’s still the case that, at the relevant level of abstraction, the sentence is warrantedly assertible in the same conditions for both of us—i.e. when our assertions are in line with our respective senses of moral approval. The possibility that asserting the sentence is warranted for one of us, but not for the other, is perfectly compatible with the sentence having the same meaning, as understood by each of us. The proposal also has the resources for overcoming the problem posed for interpretivism by the fact that the characterization of our interpretative practice offered by Radical Interpretation is not specific enough to single out a unique

86

 

interpretation for each term. Radical Interpretation might recommend to you, but not to me, interpreting ‘Tisch’, as understood by Kurt, as meaning table. But on the justificationist approach it doesn’t follow from this that ‘ “Tisch”, as understood by Kurt, means table’ has different meanings as understood by you and as understood by me, so long as for both of us the sentence is warrantedly assertible when it is recommended to us by Radical Interpretation. The same point applies to the version of this issue concerning interpretivism about propositional-attitude ascriptions. Finally, jusificationist meaning grounds do not include the ascription of a nonrepresentational function to the target sentence. Hence the objection that I raised against quasi-realism doesn’t apply to the jusificationist approach. There’s no reason in principle to treat a sentence with jusificationist meaning grounds as less genuinely involved with the function of representing the world than a sentence with representationalist meaning grounds.

5.2 Justification? I’m going to argue that an approach in this general region holds some promise, but the proposal, as it stands, is unsatisfactory. Advocates of jusificationist accounts of meaning regard as an essential component of the view that the connection between assertion of the target sentences and the obtaining of the conditions put forward as meaning grounds is epistemic—the obtaining of the conditions provides good reasons for or justifies the assertion. This point is clearly highlighted by Crispin Wright, one of the original advocates of the view, which he subsequently abandoned: It is important to remember that assertibility in this philosophical context is intended as a purely epistemic notion: the assertibility conditions of a sentence are conditions under which its assertion would be epistemically justified [ . . . ] (Wright 2012: 251)

This aspect of the position is essential for the anti-sceptical use to which some of its advocates have sought to put it.³ The general thought is that a sentence’s meaning ground consists in conditions under which asserting the sentence accrues some positive epistemic status. Let’s use justification as the designated epistemic status for the purposes of our discussion, but nothing will turn on this choice. Then, a jusificationist account of the meaning ground of ‘killing one to save five is morally right’ will have the following form:

³ See (Pollock 1967; Wright 1982).

  

87

‘killing one to save five is morally right’ has the meaning it has by virtue of the fact that a speaker is justified in asserting it just in case she feels moral approval towards the killing of one to save five. I think this proposal faces a serious difficulty. On this view, what makes it the case that the target sentence has the meaning it has is the fact that a speaker is justified in asserting it in certain conditions. However, it is hard to see how a nonrepresentationalist account of the meaning ground of a sentence could be built on these justification facts. The only plausible reading in this context of the idea that the obtaining of a condition justifies the assertion of a sentence will involve an objective connection between the satisfaction of the condition and the obtaining of the state of affairs that the sentence represents as obtaining. If this is right, then what the proposal designates as meaning grounds are facts that presuppose the states of affairs that the target sentences represent as obtaining. And this circumstance undermines the proposal’s aspiration to offer a genuine alternative to the representationalist approach.⁴ There are of course notions of justification under which assertion of a sentence can be justified in conditions that are not objectively connected with the state of affairs the sentence represents as obtaining—the sense in which an envatted brain counts as justified in asserting ‘I have hands’. But even if justification is understood along these lines the problem persists. Asserting a sentence is justified in certain conditions, in this sense, just in case there would be an objective connection between the obtaining of these conditions and the obtaining of the state of affairs that the sentence represents as obtaining if things were, in the relevant respects, as the subject believes them to be (e.g. in not being an envatted brain). One might try to circumvent the problem by specifying the meaning ground of the sentence not in terms of the conditions in which its assertion is actually justified, but in terms of the conditions in which the speaker believes that its assertion is justified. This would produce the following account: ‘killing one to save five is morally right’ has the meaning it has by virtue of the fact that speakers believe that they are justified in asserting it just in case they feel moral approval towards the killing of one to save five. I think this is a step in the right direction, but problems remain. It’s hard to see how the sentence having the meaning it has can require that speakers have views about the epistemic status of their assertions. And even for speakers who have these views, one could argue that a sceptical speaker, who doesn’t think that her feeling moral approval towards the action justifies her assertion of the sentence,

⁴ I think this is a version of the problem raised in (Wright 2012: 251–2).

88

 

nevertheless counts as meaning by the sentence what we mean by it, so long as she continues to base her assertion on her feelings of moral approval, contrary to what her scepticism would recommend. I think the best way to deal with the difficulties created by the role of justification in the account is to dispense with it—to specify the meaning ground of a sentence not in terms of the conditions under which its assertion is justified or the conditions in which speakers believe it’s justified, but in terms of the conditions under which speakers actually assert the sentence, construed as the procedure by which they regulate its assertion.⁵ The proposal then is that the meaning ground of ‘killing one to save five is morally right’ should be specified as follows: ‘killing one to save five is morally right’ has the meaning it has by virtue of the fact that speakers assert it just in case they feel moral approval towards the killing of one to save five. On this proposal, then, the meaning ground of the sentence will consist in the procedure actually employed by speakers for regulating its assertion. This is the central idea of the strategy for specifying meaning grounds that I’m going to develop and apply in the remainder.

5.3 Wittgenstein on Meaning and Use Before I develop the proposal in more detail, I want to pause for a minute to vindicate its Wittgensteinian credentials. One of the later Wittgenstein’s headline claims is that there is some kind of constitutive link between the meaning of a linguistic expression and the way it’s used: For a large class of cases of the employment of the word “meaning”—though not for all—this word can be explained in this way: the meaning of a word is its use in the language. (Wittgenstein 2009: §43)

Translated to our terminology, the claim appears to be that the meaning grounds of many linguistic expressions consist in the way they are used. For vast areas of language the thought is hardly controversial. It’s unquestionable that what makes, say, the expression ‘hello’ have the meaning it has is the fact that it is used as a salutation or greeting, or to begin a phone conversation. If you know that ‘hello’ is used like that, you know the meaning of the word, and any expression that is used in that way has the same meaning as ‘hello’. But Wittgenstein clearly thought that ⁵ Since our goal is to provide an alternative to representationalist meaning grounds, the procedure playing this role will have to be specifiable without invoking the state of affairs the sentence represents.

  

89

the approach should be extended to other areas of language where its application is more problematic, including declarative sentences and, in particular, those that appear to have the function of representing the world. If we want to provide a use-based account of the meaning grounds of representational declarative sentences, we need to identify the aspects of the way these sentences are used that can be regarded as essential for the sentences to have the meanings they have. Assertibility-conditional approaches are often seen as contributing to the Wittgensteinian programme on this point. Thus, Crispin Wright characterizes the idea that ‘to know the meaning of a declarative sentence is to know under what conditions its assertion is justified, and under what conditions it is not’ as ‘a possible interpretation of one aspect of Wittgenstein’s slogan that meaning is use’ (Wright 2012: 250–1). The conditions in which a sentence is warrantedly assertible are seen as the aspect of the way the sentence is used that makes the sentence have the meaning it has. I want to suggest that insofar as our goal is to identify aspects of the way a sentence is used that make it have the meaning it has, there’s no obvious reason for bringing an epistemic notion into the picture. The conditions under which we are justified in asserting a sentence are not naturally seen as an aspect of how the sentence is used. What does belong squarely to the way a sentence is used is the conditions under which it’s actually asserted. An account of the meaning ground of a sentence in terms of the conditions under which it’s asserted, or the procedures by which its assertion is regulated, has a much clearer claim to count as an implementation of the Wittgensteinian idea of explaining meaning in terms of use.⁶ It could be argued that Wittgenstein’s notion of a criterion has an epistemic dimension built into it, and hence that an account of meaning grounds based on this notion is bound to inherit this feature. This is indeed the standard account of Wittgenstein’s elusive concept at the time the Investigations were composed. According to Rogers Albritton, this is a central feature of Wittgenstein’s concept at that time: A criterion for a given thing’s being so is something that can show the thing to be so and show by its absence that the thing is not so; it is something by which one may be justified in saying that the thing is so and by whose absence one may be justified in saying that the thing is not so. (Albritton 1959: 854)

It may well be that this is the dominant strand in Wittgenstein’s use of the term in the Investigations. I want to argue, however, that in at least some passages of the book, it’s not obvious that criteria have anything to do with justification. I want to ⁶ However, I’m not proposing to read Wittgenstein as endorsing a pragmatist approach to meaning grounds. He seemed to be generally opposed to the idea of meaning grounds: ‘The mistake is to say that there is anything that meaning something consists in’ (Wittgenstein 1981: §16).

90

 

concentrate on one of the central passages in which the notion figures in the Investigations: The fluctuation in grammar between criteria and symptoms makes it look as though there were nothing at all but symptoms. We say, for example, ‘Experience teaches that there is rain when the barometer falls, but it also teaches that there is rain when we have certain feelings of wet and cold, or such-and-such visual impressions.’ As an argument in support of this, one says that these sense impressions can deceive us. But here one overlooks the fact that their deceiving us precisely about rain rests on a definition. (Wittgenstein 2009: §354)

This passage presents some uncontroversial features of the Wittgensteinian concept of a criterion. The connection between a criterion and what it is a criterion for is defeasible. However, contrary to what its defeasibility might indicate, the connection is not established by experience, but based on a definition. Wittgenstein uses as an example of the criterial relation the connection between rain-like sense impressions and rain. The presence of the first feature of criteria in this case is undeniable: rain-like sense impressions can occur when it’s actually not raining. This circumstance might make one think that the connection between rain-like sense impressions and rain has to be established by experience. However, Wittgenstein rejects this claim. The connection, he maintains, rests on a definition. The crucial question for our purposes is the specific form of the connection that rests on a definition, according to Wittgenstein. On the epistemic construal of criteria, the answer is something like this: what rests on a definition is that if I have rain-like sense impressions I’m still justified in believing it’s raining, even if it’s not actually raining. I want to suggest, however, that this would be a gratuitous distortion of the text. Nothing that Wittgenstein says here supports a connection with justification or any other epistemic notion. What rests on a definition, Wittgenstein is saying, is that if I form a belief as a result of rain-like sense impressions, the belief will have the content that it’s raining, even if it’s not actually raining. The connection to which Wittgenstein is ascribing definitional status is not between rain-like sense impressions and being justified in believing it’s raining, but between rain-like sense impressions and believing it’s raining. His point is about the content of my belief, not about its epistemic status. And his point about the content of the belief is that it is connected by a definition not with the conditions under which the belief is justified but with the conditions under which the belief is formed—when one has rain-like sense impressions. Justification simply plays no role in Wittgenstein’s thought. If we translate these ideas from the content of beliefs to the meaning of sentences, we obtain a claim about what makes the sentence ‘it’s raining’ have the meaning it has—about its meaning ground. The claim is that part of what

  

91

makes the sentence have the meaning it has is the fact that its assertion is regulated by whether or not speakers have rain-like sense impressions.⁷ An account of the meaning ground of the sentence along these lines has a perfectly legitimate claim to count as an implementation of the thought expressed in the quoted passage that rain-like sense impressions constitute a criterion for rain.

5.4 Assertion and Acceptance We are working towards an account of the meaning grounds of declarative sentences that doesn’t follow the representationalist template but refrains from ascribing to them a non-representational function. My leading thought is Wittgenstein’s claim that meaning is somehow constitutively connected with use. My goal is to articulate an account of the meaning ground of a declarative sentence on which what makes the sentence have the meaning it has is features of the way the sentence is used. I’m going to refer to meaning grounds that follow this general idea as pragmatist meaning grounds. Clearly not every aspect of the way a sentence is used can plausibly enter in its meaning ground. Your use of a sentence can differ from mine in many respects while we both still count as understanding the sentence in the same way. Articulating a pragmatist approach to meaning grounds requires identifying aspects of how a sentence is used that can be regarded as essential to its meaning—aspects of its use that are individually necessary and jointly sufficient for the sentence to have the meaning it has. Jusificationist accounts can be seen as advancing a proposal on this point: the aspect of how a sentence is used that is essential to its meaning is the conditions under which its assertion is justified. I have argued that this proposal is unsatisfactory, and I’ve suggested that we should focus instead, not on the conditions under which assertion of the sentence is justified, but on the conditions under which the sentence is actually asserted, or, equivalently, the procedures by which its assertion is actually regulated. Besides dispensing with epistemic notions, I want to propose another modification of the justificationist approach. It concerns the involvement of the notion of assertion. Assertion is a voluntary public action, in which the speaker presents herself as convinced that things are as the asserted sentence represents them as being. However, the connection between assertion and conviction is by no means straightforward. There are all sorts of circumstances in which a speaker might assert a sentence even though she is not convinced that things are as the sentence represents them as being. As Crispin Wright points out, ‘[a] lie, a facetious ⁷ I’m assuming that we can provide an independent specification of which sense impressions count as rain-like, not in terms of whether they produce the belief that it’s raining.

92

 

remark, a joke and a piece of pure speculation may each take an assertoric form’ (Wright 2012: 251). We don’t want to treat the conditions under which speakers assert a declarative sentence for any of these ‘extraneous’ reasons as part of its meaning ground. We want to filter out these spurious assertions and focus on those that are accompanied by conviction. In fact, our interest in assertion is only instrumental—as evidence of an inner involuntary state of acceptance that things are as the sentence represents them as being. What I have in mind is closely related to the phenomenon that Hume identifies with belief in the Appendix to the Treatise. He writes: belief consists merely in a certain feeling or sentiment; in something, that depends not on the will, but must arise from certain determinate causes and principles, of which we are not masters. When we are convinc’d of any matter of fact, we do nothing but conceive it, along with a certain feeling, different from what attends the mere reveries of the imagination. (Hume 1978: 624)

I’m not proposing to endorse the account of belief presented in this passage. What we call belief is certainly not a feeling or sentiment, and is related to conscious episodes only in complicated ways. But the feeling or sentiment that Hume wrongly identifies with belief is a real phenomenon, to which I’m going to refer as acceptance.⁸ Acceptance of a sentence, as I’ll use the term, is a conscious, involuntary re-identifiable attitude towards the sentence consisting in the conviction that things are as the sentence represents them as being. I’m going to use rejection for the negative correlate consisting in the conviction that things are not as the sentence represents them as being. I want to emphasize that I’m thinking of acceptance and rejection as feelings. Acceptance doesn’t ascribe a property or concept to a sentence or to its mental representative,⁹ nor is it the undertaking of a commitment of any kind. It is simply an involuntary feeling provoked by some sentences, as we understand them.¹⁰ Conceiving of acceptance and rejection along these lines doesn’t require assuming that it has a particularly rich phenomenology. There doesn’t have to be a collection of phenomenal features that are present precisely in those conscious episodes that involve acceptance or rejection. All that’s required is that the subject has the ability to re-identify these feelings. Their type-identity conditions can then be defined in terms of the subject’s verdicts. My proposal, then, is to replace assertion with acceptance in the meaninggrounding characterization of how a sentence is used. When a sentence has a

⁸ In (Zalabardo 2016) I used for this phenomenon the term conviction. ⁹ Hume considers and rejects this option, as the view that ‘belief is some new idea, such as that of reality or existence, which we join to the simple conception of an object’ (Hume 1978: 623). ¹⁰ See in this connection Terence Horgan and Mark Timmons’s discussion of the phenomenological dimension of what they call occurrent beliefs in (Horgan and Timmons 2006). See also Jonathan Cohen’s notion of credal feelings (Cohen 1992). See also Paul Horwich’s discussion of the notion for which he also uses the term acceptance (Horwich 1998b: 94–6).

  

93

pragmatist meaning ground, on my proposal, it will have the meaning it has as a result of the way in which its acceptance and rejection are regulated. Likewise, when a predicate has a pragmatist meaning ground, it will have the meaning it has as a result of the procedures that regulate its ascription to objects, with ascription understood, like acceptance, in terms of the conviction that the way the predicate represents an object as being is the way things stand with the object.¹¹

5.5 Pragmatist Meaning Grounds I have presented a pragmatist template for specifying the meaning grounds of sentences, according to which a sentence, as understood by a speaker, has the meaning it has as a result of the procedures that the speaker employs for regulating its acceptance (and rejection). Similarly, on a pragmatist account of the meaning ground of a predicate, as understood by a speaker, it will have the meaning it has by virtue of the fact that the speaker regulates its ascription in a certain way. I’m not going to argue that all declarative sentences have pragmatist meaning grounds. This is certainly not the case. As we will see in due course, there are prominent regions of discourse for which the model is unsuitable. My claim in the first instance is going to be that the pragmatist model can be successfully applied to some families of declarative sentences with which the representationalist model has trouble, including the four discourses we’ve been considering.¹² In order to apply the pragmatist approach to these discourses, we need to start by identifying features of the procedures that regulate the acceptance of their sentences and/or the ascription of their predicates that can be regarded as essential for these expressions to have the meanings they have. I want to suggest that, in broad outline, this job has already been done for us. If we look at the non-cognitivist and verificationist accounts of these discourses, we find that they treat aspects of the procedures by which the acceptance of their sentences or the ascription of their predicates is regulated as essential to the meanings of these expressions. Consider, first, the expressivist account of ethical discourse we discussed in Section 4.2.1. One of the functions that the account assigned to sentences ascribing the predicate ‘is morally right’ was to express the speaker’s feeling of moral approval towards the object of predication. But clearly the sentences won’t play this role unless speakers regulate their ascription of the predicate by whether they ¹¹ I think that pragmatist meaning grounds for predicates, on this construal, are similar to the conditions for concept possession described by Christopher Peacocke (Peacocke 1992). However, the account of predicate reference that I’ll offer in Chapter 8 is very distant from Peacocke’s views on how the possession conditions of a concept determine its semantic value. ¹² Pragmatist meaning grounds are closely related to Paul Horwich’s acceptance properties (Horwich 1998b), but whereas Horwich aims to provide a general account of the meaning of linguistic expressions in terms of acceptance properties, my application of the pragmatist approach to the specification of meaning grounds will be restricted to the discourses we’ve been focusing on.

94

 

feel moral approval towards the object of predication. What I’m proposing is that we drop the expressivist’s claim about the function of these sentences but keep her implicit claim about how the ascription of the predicate has to be regulated in order for it to have the meaning it has: ‘is morally right’ has the meaning it has by virtue of the fact that its ascription is regulated by the presence in the speaker of the feeling of moral approval. We find a similar situation with deflationist accounts of truth ascriptions. Once again, in addition to assigning to truth ascriptions a non-representational role, deflationists specify conditions under which the truth predicate would have to be ascribed in order for it to perform this non-representational role: speakers must ascribe ‘is true’ to a sentence S of their language just in case they accept S. As before, we can drop the deflationists’ claim about the non-representational function of truth ascriptions and keep their idea about how ascription of the predicate has to be regulated in order for the predicate to have the meaning it has: ‘is true’ has the meaning it has by virtue of the fact that its ascription to a sentence of the speaker’s language is regulated by the disquotational schema. With regard to propositional-attitude ascriptions, we’ve considered two accounts of their meaning grounds based on Dennett’s Intentional Stance. According to interpretivism, the predicate ‘believes that’s a horse’ refers to the property instantiated by those subjects to whom the Intentional Stance recommends ascribing the belief that that’s a horse. According to instrumentalism, ascriptions of ‘believes that’s a horse’ do not aim to represent things as being a certain way—they are simply tools in the strategy for predicting behaviour described by the Intentional Stance. I want to suggest that the Intentional Stance can be employed in a different way to provide a pragmatist account of the meaning grounds of propositional-attitude ascriptions. The idea is to treat the Intentional Stance simply as a procedure for regulating the ascription of propositional-attitude predicates. In this light, the Intentional Stance can be seen as offering two seemingly independent systems for deciding on the ascription of propositional attitudes. On the one hand, it recommends ascribing the beliefs and desires that the agent ought to have, ‘given its place in the world and its purpose’.¹³ But since these ascriptions are then supposed to be employed for predicting the agent’s behaviour, presumably our belief and desire ascriptions should be revised in light of their predictive success. These two procedures will no doubt issue, on occasion, conflicting recommendations, but presumably the idea is that we should ascribe beliefs and desires with the general aim of maximizing their joint satisfaction. What I’m proposing is that we can obtain from the Intentional Stance a pragmatist account of the meaning grounds of belief and desire ascriptions. On this account, what makes it the case that

¹³ (Dennett 1987b: 17), quoted in Section 4.1.3, above.

  

95

predicates of the form ‘believes that p’ and ‘desires that q’ have the meaning they have is the fact that their ascription is regulated by the Intentional Stance. Notice, however, that I’m presenting this pragmatist account of the meaning grounds of belief and desire ascriptions for the purpose of illustrating the application of the pragmatist approach to this discourse. The proposal, as it stands, has important shortcomings. The pragmatist account of the meaning grounds of belief and desire ascriptions I’m going to defend will employ some central ideas of the Intentional Stance, but will be substantially different from the proposal outlined here. This issue will be discussed in detail in Chapter 6. We can take a similar line with respect to meaning ascriptions, using in this case Radical Interpretation as a characterization of a procedure for ascribing meanings, without trying to extract from it a representationalist specification of the states of affairs that meaning ascriptions represent as obtaining. On the resulting pragmatist account of the meaning grounds of meaning ascriptions, what makes it the case that meaning ascriptions have the meaning they have is the fact that their acceptance is regulated by Radical Interpretation. As with propositional-attitude ascriptions, this brief characterization of a pragmatist account of meaning ascriptions is intended as an illustration of how the pragmatist strategy can be applied to this case. As it stands, the proposal faces serious obstacles that we will discuss in detail in Chapter 7, where I will propose some modifications. I have provided brief outlines of how the pragmatist model could be applied to each of our four problematic discourses. At this point these proposals are only meant as illustrations of how the model could work in the discourses that interest us. Later on I will introduce some modifications of the general template that will address some difficulties of the outlined proposals. And in subsequent chapters I will consider in detail the precise shape that the proposal should take with respect semantic discourses—ascriptions of truth, meanings, and propositional attitudes.

5.6 Pragmatism and Its Fellow Travellers In the previous section I have outlined pragmatist accounts of the meaning grounds of the sentences in our target discourses. These have substantial points of contact with treatments of these sentences that we considered and dismissed earlier on, and it is important to keep in focus how the pragmatist proposal differs from these and other related positions.

5.6.1 Pragmatist Representationalism vs. Genuine Pragmatism One general approach with which the pragmatist proposal is easily confused is the version of representationalism that seeks to define the states of affairs that the

96

 

target sentences represent as obtaining in terms of the procedures employed by speakers for regulating their acceptance. The pragmatist approach specifies the meaning grounds of these sentences in terms of these very same procedures, but they are used in a different way. The pragmatist is not seeking to specify the states of affairs represented by these sentences in terms of the associated procedures. In fact, she is generally not trying to specify in any way the states of affairs represented by these sentences. Her goal is to specify the meaning grounds of these sentences, the facts that make them have the meaning they have. If we are working within the representationalist paradigm, specifying the meaning ground of a sentence requires identifying the state of affairs it represents as obtaining. Hence, if we take the representationalist model as indispensable, an account of meaning grounds in terms of acceptance procedures can only be understood as specifying in terms of these procedures the state of affairs represented by the sentence. But the pragmatist approach is emphatically not a version of representationalism. It offers an alternative strategy for specifying the meaning ground of a sentence—one that does not require identifying the state of affairs the sentence represents as obtaining. For the pragmatist, what makes a sentence have the meaning it has is not its relation to the state of affairs it represents as obtaining, but the way its acceptance is regulated. It is exclusively in this capacity that acceptance procedures figure in pragmatist meaning grounds. By virtue of this difference, the pragmatist approach is immune to some of the objections that I levelled against the attempt to define the states of affairs represented by sentences in terms of acceptance procedures. Thus, consider the view discussed in Section 4.1.1 that the sentence ‘killing one to save five is morally right’ represents the killing of one to save five as triggering approval. I argued that this position leads to an implausible subjectivism. Triggering the approval of speaker A and triggering the approval of speaker B are different properties. Hence, on this proposal, ‘killing one to save five is morally right’ as understood by A and as understood by B will mean different things. The pragmatist approach, by contrast, doesn’t have this consequence. If A and B regulate their acceptance of the sentence on the basis of whether they feel moral approval towards the killing of one to save five, they employ the same acceptance procedure, at the level of abstraction at which procedures are categorized for the purposes of the pragmatist account.¹⁴ Therefore, for the pragmatist, they count as

¹⁴ The relevant level of abstraction gets fixed by what we regard as essential for our sentence to have the meaning it has, or for another sentence to have the same meaning. In this case, I don’t think that having its acceptance regulated by my own sense of moral approval is essential to my sentence having the meaning it has. Someone else’s sentence will have the same meaning if they regulate their acceptance of the sentence by their own sense of moral approval towards the killing of one to save five.

  

97

ascribing the same meaning to the sentence, even if, as it happens, A feels moral approval towards the killing of one to save five and B doesn’t. In the same way, the pragmatist account of the meaning grounds of truth ascriptions doesn’t generate the Protagorean relativism discussed in Section 4.1.2. Being accepted by A and being accepted by B are different sentence-properties. But if both A and B regulate the ascription of ‘is true’ to a sentence on the basis of whether they accept the sentence, then, for the pragmatist, they count as using the same procedure for ascribing the predicate. Hence, on the pragmatist account of the meaning ground of ‘is true’, the predicate has the same meaning as understood by A and as understood by B. Similarly, the lack of specificity of Radical Interpretation and the Intentional Stance highlighted in Section 4.1.3 is not a problem for the pragmatist. The fact that Radical Interpretation is compatible with different meaning ascriptions renders it unsuitable for the task of specifying the states of affairs represented by meaning-ascribing sentences. However, as I’ve explained, this is not how the pragmatist is proposing to use Radical Interpretation. The pragmatist’s claim is that what makes a meaning-ascribing sentence have the meaning it has is the fact that its acceptance is regulated on the basis of Radical Interpretation. Radical Interpretation can play this role even if it produces different and incompatible meaning ascriptions when applied by different speakers. A similar situation obtains with respect to the pragmatist employment of the Intentional Stance.¹⁵

5.6.2 Pragmatism vs. Non-cognitivism It is equally important to appreciate the difference between the pragmatist approach and the non-cognitivist proposals considered in Section 4.2. The pragmatist co-opts some central features of the non-representational functions that non-cognitivists assign to the problematic discourses. However, these ideas are put to a new use—to specify the acceptance procedures from which sentences obtain their meaning grounds—while explicitly refraining from following the non-cognitivist in assigning to these sentences a non-representational function. On the pragmatist approach, the meaning grounds of the target sentences are completely silent on the function they perform. This removes the difficulty I raised for quasi-realism concerning the possibility of treating the problematic sentences as performing the function of representing the world. Of course, an advocate of the RR assumption will argue that by endorsing nonrepresentationalist meaning grounds for the problematic sentences, the pragmatist is implicitly committed to rejecting the view that they successfully perform the ¹⁵ I think the difficulties outlined in this section afflict response-dependence accounts of the problematic discourses. See (Johnston 1993; Pettit 1991; Wright 1988).

98

 

function of representing the world. So long as RR is in place, this conclusion is unavoidable. However, a pragmatist who is prepared to challenge RR will be able to maintain that assigning pragmatist meaning grounds to a sentence doesn’t in any way affect its ability to perform the function of representing the world. Concerning ethical discourse, one obvious advantage of the pragmatist approach over non-cognitivism is avoiding the Frege-Geach problem.¹⁶ For the pragmatist, what makes ‘killing one to save five is morally right’ have the meaning it has is the fact that its acceptance as an unembedded sentence is regulated by the speaker’s feelings of moral approval. But the meaning the sentence acquires in this way is representational—by virtue of these facts about acceptance regulation, the sentence comes to represent things as being a certain way. And this meaning can be unproblematically carried over to embedded occurrences, as in: ‘If killing one to save five is morally right, then killing one to save six is also morally right’. The pragmatist can unproblematically interpret this sentence as asserting that if things are as represented by the antecedent, then things are also as represented by the consequent. As a result, the logical relations between the unembedded sentence and the compounds in which it occurs don’t pose a problem for the pragmatist. With respect to the second problem for expressivism mentioned in Section 4.2.1, the pragmatist approach has no clear advantage over its rival. If the meaning ground of ‘killing one to save five’ consists in the fact that its acceptance is regulated by the speaker’s feeling of moral approval, then a noncircular account is needed of the nature of this attitude. I see no reason why this cannot be provided, perhaps in terms of the connection between the attitude and motivation. However, I’m not going to pursue this issue here. Ethical discourse has served its purpose as a prop for us, and we will soon be leaving it behind to concentrate exclusively on semantic discourses. Notice that the pragmatist is not committed to the view that discourses with pragmatist meaning grounds have no function other than representing the world. Discourses with pragmatist meaning grounds can play all sorts of nonrepresentational roles in our lives, but exactly the same can be said for discourses with representationalist meaning grounds. Furthermore, the non-representational role played by a discourse, with either kind of meaning grounds, may well be part of an enlightening causal or genealogical explanation of how we came to have the discourse.

¹⁶ See Section 4.2.1, above.

  

99

The issue between the pragmatist and the non-cognitivist is not whether the target discourse has a non-representational function, but whether a non-representational function is included in the meaning grounds of the discourse—whether the sentences of the discourse have the meaning they have by virtue, in part, of the fact that they play this non-representational role. Noncognitivists give an affirmative answer to this question, pragmatists give a negative answer. The contrast can be usefully highlighted by reference to the discussion offered by Sebastian Köhler in a recent paper of the view to which he refers as neopragmatist meta-semantics: This meta-semantics has two components. First, a ‘role component’: an account of a term’s role in the larger linguistic framework—its conceptual or inferential role. Second, a ‘function component’: an account of the word’s function for creatures like us—what it is used for, what important thing it allows us to do, its utility or practical significance. Both components are meant to capture part of the idea that ‘meaning is use’ (Williams 2013: 135): the first captures how words are used, the second what they are used for. (Köhler forthcoming)

The view that I’m articulating here differs in two important respects from the meta-semantics described by Köhler. First, the second component—the ‘function component’, what words are used for—plays no role in my strategy for specifying the meaning grounds of linguistic expressions. On my approach, the meaning grounds of linguistic expressions are specified in terms of the first component—of how words are used. As I pointed out just now, taking this line doesn’t require denying that the target discourse has a utility or practical significance. All it requires is to refrain from including these factors in the meaning grounds of the sentences of the discourse. The attitude towards the function of the discourse that my approach recommends is accurately characterized by one of the readings of the explanatory role of functions distinguished by Köhler, using the example of the term ‘true’, with the ‘role component’ (how the term is used) characterized following a proposal by Michael Williams (as I-T and E-T) (Williams 2013: 134–5): we can hold that a term’s role component completely gives its content. For example, ‘true’ has its content solely in virtue of the role given by I-T and E-T. This means that any linguistic community possesses a term with the same content as ‘true’, if they have a term characterized by I-T and E-T. The function component only explains why we have such a term. Thus, the function component is irrelevant for individuating the content of ‘true’. Other communities could have terms with the same content, where something else explains their having them. (Köhler forthcoming)

100

 

For our immediate purposes we don’t need to concern ourselves with the specific characterization of the role component that Köhler employs in his example (I-T and E-T). The point that I want to highlight is that the attitude towards the functional component that Köhler describes in this passage is precisely the one I propose to adopt, and this is the point at which my proposal comes apart from non-cognitivist positions. Non-cognitivists adhere to the other reading of the explanatory role of functions presented by Köhler: On this reading, other communities could not have terms with the same content as ‘true’, if those terms performed different functions for them—content is tied essentially to function. (Köhler forthcoming)

I argued in Section 4.3 that once we assign this role to the non-representational function of a discourse it becomes impossible to construe the discourse as also performing the function of representing the world. Notice, also, that although my brand of pragmatism is compatible with the ascription of a non-cognitive function to the target discourse, my approach, unlike non-cognitivism, also makes room for discourses with pragmatist meaning grounds that have no discernible utility or practical significance, or even for discourses that are detrimental for our practical goals. We might prefer not to have discourses of this kind, but there’s no reason to think that we don’t have them or that, if we do have them, their practical insignificance renders them incapable of performing the function of representing the world. Excluding non-representational functions from the approach to meaning grounds I’m defending is not without cost. Köhler provides a good explanation of the work they do for the non-cognitivist (Köhler’s neo-pragmatist): to dispel theoretical worries about certain domains, neo-pragmatists appeal to their functions: they argue that these domains’ core terms play some ‘theoretically harmless’ non-representational function and that, therefore, metaphysical and epistemological worries about them do not arise. It seems, though, that this appeal presupposes the second reading. After all, it is unclear how claims about functions that do not impact content could dispel the relevant worries. (Köhler forthcoming)¹⁷

I think the metaphysical and epistemological worries that Köhler has in mind are perfectly legitimate, and my brand of pragmatism can’t address them by reference to the non-representational functions of discourses. In Chapter 8 and Chapter 9,

¹⁷ For a good example of how non-representational functions are put to this use, see (Price 2011a, 2011b).

  

101

below, I provide a construal of these worries and develop a strategy for addressing them. The second respect in which Köhler’s neo-pragmatist differs from my pragmatist concerns the character of the role component. Köhler’s neo-pragmatist, following Michael Williams, on whose ideas Köhler’s discussion is based, offers a characterization of the semantically relevant aspects of how a linguistic expression is used in terms of its ‘role in the larger linguistic framework—its conceptual or inferential role’. For Köhler’s neo-pragmatist, as well as for Williams, the meaning-grounding aspects of how a linguistic expression is used are generally intra-linguistic factors, concerning the rules governing the inferential transitions between the target expressions and other expressions in the language. My pragmatist doesn’t acknowledge this restriction. Meaning-grounding acceptance procedures may occasionally be characterizable in these terms, but in general they concern how speakers regulate the acceptance and rejection of sentences on the basis of whether certain objective conditions obtain.

5.6.3 Rationalist Pragmatism One of the most influential versions of the pragmatist approach to meaning grounds can be found in the work of Robert Brandom. Here’s his characterization of the pragmatist approach to conceptual content: An account of the conceptual might explain the use of concepts in terms of a prior understanding of conceptual content. Or it might pursue a complementary explanatory strategy, beginning with a story about the practice or activity of applying concepts, and elaborating on that basis an understanding of conceptual content. The first can be called a platonist strategy, and the second a pragmatist (in this usage, a species of functionalist) strategy. (Brandom 2000: 4)

In earlier work, he presents a similar idea concerning the meaning of truth ascriptions. He characterizes his approach as: [ . . . ] the approach whose leading idea is that the special linguistic roles of truth ascriptions are to be explained in terms of features of the ascribings of truth, rather than of what is ascribed. The explanatory emphasis placed on the act of calling something true, as opposed to its descriptive content, qualifies theories displaying this sort of strategic commitment as ‘pragmatic’ theories of truth, by contrast to ‘semantic’ ones. (Brandom 1988: 75)

In these passages and throughout his work, Brandom clearly presents himself as advocating pragmatist accounts of meaning grounds and opposing representationalist

102

 

accounts. On this point, my approach is very close to his. There is, however, one important respect in which my views differ from Brandom’s. It concerns his account of the conditions that a practice has to satisfy in order for the associated discourse to count as representational. For Brandom, who develops on this point ideas from Wilfrid Sellars, in order for something to have a conceptual content at all, it needs to be involved in ‘the game of giving and asking for reasons’— to have ‘an inferential role, as premise and conclusion of inferences’ (Brandom 2000: 57).¹⁸ Only by virtue of our participation in this practice do our mental and linguistic products acquire their representational character. He spells out the view in the following passage: Saying or thinking that things are thus-and-so is undertaking a distinctive kind of inferentially articulated commitment: putting it forward as a fit premise for further inferences, that is authorising its use as such a premise, and undertaking responsibility to entitle oneself to that commitment, to vindicate one’s authority, under suitable circumstances, paradigmatically by exhibiting it as the conclusion of an inference from other such commitments to which one has become entitled. (Brandom 2000: 11)

I want to concentrate on the responsibility that, according to Brandom, we need to undertake in order to take part in the practice that will confer on our judgements and assertions the character of representations: we need to be prepared to ‘vindicate our authority’ regarding what we are saying or thinking, ‘paradigmatically by exhibiting it as the conclusion of an inference from other such commitments to which one has become entitled.’ In order to take part in this practice, we need to be prepared to provide reasons for what we are convinced of—to defend it with other things we are convinced of. Richard Rorty highlights a similar feature of our cognitive practice as the basis of his pragmatist characterization of semantic notions—namely ‘the need to justify our beliefs and desires to ourselves and to our fellow agents’ (Rorty 1998: 26) or the hope ‘to justify our belief to as many and as large audiences as possible’ (Rorty 1998: 39). Thus, both for Rorty and for Brandom, the representational character of our mental and linguistic episodes is grounded in our acceptance of a commitment to offer reasons for what we believe, either to ourselves or to others. If we rejected this commitment, then we wouldn’t participate in the practice that confers on mental and linguistic episodes the character of representations. Our own linguistic or mental output could not be characterized as ‘saying or thinking that things are thus and so’. Brandom refers to this specific

¹⁸ For discussion of the idea in Sellars, see (Sellars 1997).

  

103

implementation of the pragmatist explanatory strategy as ‘rationalist pragmatism’ (Brandom 2000: 11).¹⁹ Brandom’s ideas on this point address a genuine need of the pragmatist approach. Brandom maintains that a sentence with a pragmatist meaning ground can successfully discharge the function of representing the world, but he doesn’t also subscribe to the implausible view that every (declarative) sentence with a pragmatist meaning ground has the function of representing the world and succeeds in discharging it. It follows that he needs to offer an account of when pragmatist meaning grounds bestow on sentences the function of representing the world and the ability to discharge it. His rationalist pragmatism is his strategy for addressing this challenge. My brand of pragmatism faces the same challenge as Brandom’s. I agree with Brandom that some discourses with pragmatist meaning grounds can successfully discharge the function of representing the world, but I also don’t want to be saddled with the view that every discourse with pragmatist meaning grounds has the function of representing the world and succeeds in discharging it. Hence I, like Brandom, owe an account of the conditions under which the practice that sustains a discourse supports treating its sentences as representing the world. I will attempt to meet this challenge in Chapter 7, and the full picture of the contrast between my views and Brandom’s won’t emerge until then. Nevertheless, I can already register my rejection of Brandom’s proposals on this point. For Brandom, representing the world requires undertaking the commitment to offer reasons in support of our claim that things are as the sentences we accept represent them as being. The pragmatist approach I am defending rejects the idea that undertaking this commitment is a necessary condition for a discourse to count as representational. On my model, if a sentence has a pragmatist meaning ground, a speaker will see its meaning-grounding acceptance procedure as providing her with reasons for the verdicts it recommends. As we’ll see in Chapter 7, treating the sentence as representing the world will also require that she treats diverging verdicts by others, or by her former self, as incorrect. But for Brandom this is not enough for the sentence to discharge the function of representing the world. For this, according to Brandom, the speaker needs, in addition, to undertake the commitment to identify reasons that would vindicate her verdicts in the eyes of fellow speakers. For Brandom, and for Rorty, a speaker who doesn’t undertake this commitment is not representing the world with her sentences.²⁰

¹⁹ Rorty has used for this aspect of his position and Brandom’s the label ‘conversationalism’ (Rorty 2000a: 237). ²⁰ Brandom does contemplate the possibility of ‘bare assertions’, ‘without the implicit claim to entitlement that is demonstrable should someone become entitled to challenge it’ (Brandom 1994: 229). However, he thinks that these claims ‘are intelligible only as exceptions against a background of practices in which claims typically have the significance of claims whose authority is redeemable by demonstration of warrant. The possibility of bare assertion is parasitic on the possibility of assertions

104

 

I see no reason to impose this requirement. Interest in offering reasons that would vindicate your verdicts in the eyes of others may be a commendable trait, but its absence in a speaker shouldn’t undermine the representational status of her sentences. If you and I disagree, representation requires that I regard you as wrong. It doesn’t require, in addition, that I undertake the enterprise of finding reasons that would save you from your error. Furthermore, it would be wrong to assume that this is a feasible enterprise in all discourses that we want to regard as representational. In ethical discourse, for example, there are fundamental disagreements for which no reasons offered by one party could convince the other party that they are in the wrong.²¹ But recognizing that we are in this situation shouldn’t be treated as undermining the idea that both parties are representing things as being a certain way, and that only one party is representing things as they are in actuality.

5.7 The ‘Explanation’ Explanation In an important paper (Dreier 2004), James Dreier has proposed a solution to what he calls the problem of creeping minimalism. The problem consists in explaining the contrast between two positions in metaethics, for which he uses the labels realism and irrealism. Traditionally the contrast was drawn in terms of a series of theses accepted by realists and rejected by their opponents. They included commitment to the existence of objective, mind-independent moral facts and properties, and a construal of moral discourse as representational. But drawing the contrast in these terms is no longer possible. The realist tenets involve concepts that can receive minimalist construals, such as fact, property, representation or even mind-independence. Adopting these minimalist construals enables philosophers we would want to classify as irrealists to join the realists in endorsing what we originally regarded as their distinctive claims. We saw in Section 4.3 how Blackburn’s quasi-realism exemplifies this situation. As Dreier explains, Horgan and Timmons’s cognitive expressivism also falls in this category (Horgan and Timmons 2006). Dreier’s problem, then, is to explain the contrast between genuine realism and views such as those of Blackburn and Horgan and Timmons, whose endorsement of realist theses is reached through the adoption of minimalist construals of concepts that figure in these theses.

that implicitly involve undertaking a conditional task-responsibility to demonstrate the asserter’s entitlement to the commitments undertaken by the performance of speech acts of that kind’ (Brandom 1994: 229). See also (Brandom 1983: 643). I find the claim that bare assertions have to be exceptional as implausible as the claim that they are impossible. A linguistic community whose members were uniformly uninterested in vindicating the claims expressed by their assertions should still count as representing the world with these. ²¹ On this point, see (Rorty 2000b).

  

105

Dreier’s proposal is that the contrast should be drawn in terms of how each position ‘seeks to explain various phenomena of ordinary moral (and other normative) thought and talk’ (Dreier 2004: 37), as, for example, the fact that Edith said that abortion is wrong or the fact that Julia believes that knowledge is intrinsically good. The kind of explanation of these facts that’s relevant for Dreier’s contrast is a specification of that in virtue of which it is true to say that these facts obtain (Dreier 2004: 38), i.e. in the terminology that Dreier borrows from Kit Fine, of what grounds these facts. For realists, on the one hand, what makes it the case that these facts obtain will include a relation between subjects and moral properties—between Edith and wrongness, or between Julia and intrinsic goodness. For the sophisticated expressivist, on the other hand, what makes it the case that these facts obtain concerns a non-cognitive functional role of Edith’s assertion and Julia’s belief. Dreier labels his proposal as the ‘explanation’ explanation.²² Notice that if the contrast is drawn in this way, an account of moral discourse along the lines of the brand of pragmatism I’m defending will not qualify as either realist or irrealist. Consider the fact: L

Sentence S, as understood by Lorna, means that killing one to save five is morally right.

Dreier’s realist will specify the ground of L in terms of a relation between S and, among other things, moral rightness. For his irrealist, by contrast, the ground of L will consist in a non-cognitive (e.g. expressive) functional role played by S. It should be clear that my pragmatist would not follow either of these routes. She explicitly rejects the representationalist approach endorsed by Dreier’s realist, but she is equally opposed to including a non-representational functional role in the meaning ground of S. In order to place my brand of pragmatism in Dreier’s classification, we would need to decide whether not mentioning a relation between S and moral rightness in the ground of L is a sufficient condition for counting as an irrealist, or, on the contrary, not including a non-representational role for S in the ground of L is a sufficient condition for counting as a realist. If we take the first option, my brand of pragmatism will qualify as irrealist; if we take the second option, it will count as realist. It seems to me that the first option is closer to Dreier’s intentions, and probably to our intuitive understanding of the terms. Be this as it may, I don’t think it’s hugely important which of these options we take. In fact, I think it would be preferrable to discuss the relative merits of rival accounts of the ground of L without taking into consideration traditional associations between the

²² For insightful discussion of Dreier’s ideas, see (Simpson 2018).

106

 

competing proposals and the positions in metaphysics that the terms ‘realism’ and ‘irrealism’ suggest. Furthermore, as we’ll see in Chapter 9, these associations might turn out to be illegitimate.

5.8 Conclusion In this chapter I have presented the main ideas of the approach to the meaning grounds of declarative sentences that I’m going to explore in the rest of the book. I’ve taken as my starting point the justificationist view, that locates the meaning ground of a declarative sentence in the conditions under which the sentence is warrantedly assertible. The approach that I’m endorsing replaces the conditions in which a sentence is warrantedly assertible with the conditions in which it is actually asserted, and assertion with acceptance. On the resulting approach, the meaning ground of a sentence consists in the procedure employed by speakers for regulating acceptance of the sentence. This is the version of the pragmatist strategy for specifying the meaning ground of a declarative sentence that I’m going to develop and defend in the remainder. I have argued that this is a plausible construal, for the case of declarative sentences, of Wittgenstein’s idea that the meaning of a linguistic expression is constitutively linked to the way the expression is used. I have then outlined accounts of the meaning grounds of the sentences of our four problematic discourses, and I have discussed the differences between my proposal and other positions deploying similar ideas. The outlines I have provided of pragmatist meaning grounds for our four problematic discourses were meant as illustrations of the general strategy, not as fully defensible proposals. This is where we’ll leave things for ethical discourse, but my next item of business is to replace these outlines with plausible accounts of the meaning grounds of the sentences of semantic discourses. I’ll deal with belief ascriptions in Chapter 6, and with ascriptions of meaning and truth in Chapter 7.

6 Belief and Desire In Section 5.5 I proposed that the meaning grounds of the sentences of the problematic discourses should be specified along pragmatist lines—in terms of features of the procedures employed by speakers for regulating their acceptance. Concerning belief and desire ascriptions, I made the provisional proposal of specifying their meaning-grounding acceptance procedures in terms of Daniel Dennett’s Intentional Stance. On this proposal, our belief and desire ascriptions would have the meaning they have by virtue of the fact that their acceptance is regulated by the Intentional Stance. My goal in the present chapter is to develop this idea. I’m going to argue that there are aspects of the Intentional Stance that are unsuitable for this purpose, but that its central idea can be used as the core of a promising account of the meaning grounds of belief and desire ascriptions. I’m going to concentrate exclusively on belief and desire ascriptions, although I won’t have a lot to say here about the contrast between these and other cognitive and conative attitudes—e.g. the difference between believing and guessing. This involves a huge simplification of the complex network of concepts that we employ to characterize our mental life. Our ascriptions of beliefs and desires in real life are inseparable from our ascriptions of other mental phenomena, including emotions, tendencies, moods, attention levels, etc. My hope is that by concentrating on belief and desire ascriptions we will be able to isolate features of their meaning grounds that will still play an important role in a more sophisticated account of our ascriptions of mental states that includes the whole range of mental phenomena.

6.1 Belief, Desire, Behaviour Belief and desire ascriptions have, in general terms, a very simple syntax. They are, or appear to be, relational sentences, in which one of two binary predicates connects the subject to which the attitude is being ascribed with the possible state of affairs that is being put forward as the content of the attitude.¹ The content can be represented by a that-clause, but other constructions are also possible, as in ‘she desires to be left alone’ or ‘he believes the witness’s testimony’. I will represent ¹ Possible states of affairs might be too coarse-grained to serve as relata in propositional-attitude ascriptions, in light of the intensional character of the notion. I’m not going to consider this complication here.

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0006

108

 

them as having the structure ‘aBp’ and ‘bDq’, with ‘a’ and ‘b’ standing for the subjects to which the attitudes are ascribed and ‘p’ and ‘q’ for the possible states of affairs postulated as the contents of the ascribed attitudes. The pragmatist’s working hypothesis is that belief and desire ascriptions have the meaning they have as a result of some features of how their acceptance is regulated. The task of identifying these features can be understood in the following terms: Consider all possible discourses involving sentences with this syntax— declarative sentences of the forms ‘aBp’ and ‘bDq’, in which ‘a’ and ‘b’ stand for human beings and possibly other individuals, and ‘p’ and ‘q’ stand for possible states of affairs. Call these BD discourses. According to the pragmatist, what turns the sentences of a BD discourse into belief and desire ascriptions is the fact that their acceptance is regulated in a certain way. On her view, there will be features of the way in which the acceptance of belief and desire ascriptions is regulated whose presence in or absence from a BD discourse will determine whether accepting ‘aBp’ (B-pairing a with p, as I will put it sometimes) counts as ascribing to a the belief that p obtains, and whether accepting ‘bDq’ (D-pairing b with q) counts as ascribing to b the desire that q obtains. One feature of our belief and desire ascriptions that has a claim to being considered essential for their meaning is the role they play in a familiar strategy for predicting behaviour, as in: Lorna wants to drink water and believes that there’s water in the fridge, so I predict that she will open the fridge. Notice that this employment has direct consequences for how the acceptance of belief and desire ascriptions is regulated. Correct behaviour predictions will support the belief and desire ascriptions that generate them; incorrect behaviour predictions will have the opposite effect. The central idea of my proposal is that what makes belief and desire ascriptions have the meaning they have is the fact that their acceptance is regulated in this way—by reference to the success of the behaviour predictions they produce. My main job in this chapter will be to provide a characterization of the predictive strategy for which this claim can be plausibly made—of the strategy that we deploy when we use belief and desire ascriptions to predict behaviour. I’m going to take as my starting point the characterization of this predictive strategy that Daniel Dennett has developed under the label Intentional Stance. I’m going to argue that some aspects of Dennett’s characterization are inadmissible for our purposes, but that his central idea can be deployed as the core of a different characterization that satisfies our requirements.²

² The central ideas of this chapter are drawn, with some major changes, from my (Zalabardo 2019a).

  

109

6.2 The Intentional Stance The idea of explicating belief and desire ascriptions in terms of the role they play in a strategy for predicting behaviour has been championed by Daniel Dennett. He refers to the predictive strategy that he employs for this purpose as the Intentional Stance. He outlines the strategy in the following passage, that we already quoted in Chapter 4:³ Here is how it works: first you decide to treat the object whose behaviour is to be predicted as a rational agent; then you figure out what beliefs that agent ought to have, given its place in the world and its purpose. Then you figure out what desires it ought to have, on the same considerations, and finally you predict that this rational agent will act to further its goals in the light of its beliefs. A little practical reasoning from the chosen set of beliefs and desires will in many—but not all—instances yield a decision about what the agent ought to do; that is what you predict the agent will do. (Dennett 1987b: 17)

Dennett clearly intends to use the Intentional Stance to specify the meaning grounds of belief and desire ascriptions. However, it’s not entirely clear how he proposes to do this. Some of his critics have assumed that his goal is to provide an instrumentalist account of the discourse—according to which the function of its sentences is not to represent things as being a certain way, but to serve as tools in a behaviour-prediction enterprise. Against this, Dennett emphasizes that on his position belief and desire ascriptions do represent states of affairs as obtaining— even if these states of affairs can only be discerned when we adopt the Intentional Stance. He openly disagrees with standard representationalist accounts of belief and desire ascriptions, that locate the states of affairs they represent in the internal causes of behaviour, along the lines of the RTM-ITS paradigm. However, he seems to think that the Intentional Stance will enable us to specify the states of affairs belief and desire ascriptions represent as obtaining: all there is to really being a true believer is being a system whose behavior is reliably predictable by the intentional strategy, and hence all there is to really and truly believing that p (for any proposition p) is being an intentional system for which p occurs as a belief in the best (most predictive) interpretation. (Dennett 1987b: 29)

This would support a representationalist reading of his ideas, according to which belief and desire ascriptions obtain their meanings from a semantic relation to ³ Other philosophers have exploited the link with behaviour in their account of propositional attitudes, without emphasizing the predictive aspect. See (Ramsey 1927; Stalnaker 1984; Whyte 1990).

110

 

states of affairs concerning the accuracy of the behaviour predictions they generate. I won’t try to determine here which interpretation of Dennett’s ideas is most faithful to his intentions. My goal is to use his ideas to provide a pragmatist account of the meaning grounds of belief and desire ascriptions, independently of the character of Dennett’s own proposal. My goal is not to use our behaviourpredicting strategy to specify what has to be the case in order for someone (an interpretee) counts as believing or desiring something. My goal is instead to use our behaviour-predicting strategy to specify what has to be the case in order for someone (an interpreter) counts as ascribing beliefs and desires—what has to be the case for a BD discourse to achieve this. I think Dennett’s Intentional Stance can be usefully characterized as involving two components—one concerning which beliefs and desires to ascribe and another concerning how to use the beliefs and desires thus ascribed for predicting behaviour: IS1

Ascribe to the agent the beliefs and desires it ought to have.

IS2

Predict that the agent will act to satisfy the desires you’ve ascribed to it in light of the beliefs you’ve ascribed to it.⁴

Presumably the idea is that the ascriptions produced under IS1 are open to revision in light of their predictive success when used according to IS2, and the goal will be to strike a balance between ascribing the beliefs and desires the agent ought to have and ascribing those that result in accurate predictions. The use of IS1 in an account of the meaning grounds of belief and desire ascriptions is problematic. The difficulty was pointed out by William Lycan: A threat of circularity. If beliefs are characterized in terms of what subjects ought to be believing, and desires are characterized in terms of what subjects ought to be wanting, what is a belief or a desire in the first place? Dennett’s view is complex, and I am far from insisting that the apparent circularity is real, vicious, and decisive; but I would like to hear exactly why it is not vicious. (Lycan 1988: 519)⁵

I want to concentrate on how the threat of circularity might affect the prospects for using the Intentional Stance in a pragmatist account of the meaning grounds of belief and desire ascriptions. The problem concerns IS1—the proposed strategy for selecting the desires and beliefs we should ascribe. The pragmatist’s goal is to ⁴ As far as I can see, the ascription of rationality to the agent plays no role in the strategy beyond the assertion that the agent will act to further its goals in the light of its beliefs. ⁵ Gilbert Harman presents a similar concern (Harman 1988: 515).

  

111

specify when a BD discourse produces belief and desire ascriptions in terms of the procedures used by speakers for regulating the acceptance of sentences of the form ‘aBp’ and ‘bDq’. Concentrating on belief, what IS1 dictates is that ‘aBp’ will count as expressing a belief ascription only if its acceptance is regulated by whether a ought to believe p. We can see why this condition cannot be legitimately included in the meaning grounds of belief ascriptions if we consider a different proposal: IS1*

Ascribe to the agent the beliefs and desires it has.

According to IS1*, ‘aBp’, as meant by a speaker, will ascribe to a the belief that p just in case she decides whether to accept ‘aBp’ on the basis of whether she believes that a believes that p. While this claim is probably true, I want to argue that it cannot legitimately figure in a pragmatist account of the meaning ground of ‘aBp’. The problem is that this approach would explain what has to be the case in order for ‘aBp’, as understood by S, to mean that a believes that p in terms of facts about when S believes that a believes that p. In other words, it provides an account of the meaning grounds of linguistic belief ascriptions that presupposes a notion of mental belief ascriptions. This would result in a much less interesting proposal than we might have expected from the pragmatist. It is generally accepted that if we have a suitable explication of mental content, we will have taken a major step towards explicating linguistic meaning, and explicating linguistic meaning in terms of mental content is unlikely to require the resources of the pragmatist approach. The pragmatist’s goal is not to provide an account of the meaning of sentences in the problematic discourses in terms of the content of the corresponding mental states. In particular, she is not trying to explain the meaning of linguistic belief ascriptions in terms of the content of mental belief ascriptions. Her goal is more ambitious—to provide an account of linguistic belief ascriptions that doesn’t presuppose the notion of mental belief ascriptions. Including IS1* in the meaning grounds of belief ascriptions would undermine this aspiration. I want to argue that the reasons just given for not using IS1* in a pragmatist account of the meaning grounds of belief ascriptions carry over smoothly to IS1. In this case, the pragmatist would be explaining what has to be the case in order for ‘aBp’, as understood by S, to mean that a believes that p in terms of facts about when S believes that a ought to believe that p. But if we are presupposing facts as to whether S believes that a ought to believe that p, we must be also presupposing facts about beliefs of S that include in their content the concept of belief in other more simple contexts, including the belief that a believes that p. If, as I’ve argued, IS1* is unacceptable for the pragmatist, the same goes for IS1. In his reply to Lycan (Dennett 1988: 543), Dennett invokes the proximity between the concept of what you ought to believe and the concept of what I would believe in your circumstances. I don’t think this observation makes IS1 more acceptable to the pragmatist. Explaining what has to be the case in order for

112

 

‘aBp’, as understood by S, to mean that a believes that p in terms of facts about when S believes that in a’s circumstances she would believe that p suffers from the same problem as doing it in terms of facts about when S believes that a ought to believe that p. Once again, by presupposing facts about when S believes that in a’s circumstances she would believe that p we would also be presupposing facts about beliefs of S that include in their content the concept of belief in other more simple contexts, including the belief that a believes that p. Even with this gloss, IS1 is of no use to a pragmatist account of the meaning grounds of belief and desire ascriptions. IS2 doesn’t suffer from these difficulties, and I propose to use it as the starting point for a pragmatist account of the meaning grounds of belief and desire ascriptions. Some aspects of the predictive strategy described by IS2 need to be clarified. First, the notion of acting to satisfy your desires ‘in light of your beliefs’. I propose to unpack this as acting in a way that would be conducive to the satisfaction of your desires if your beliefs were true. When we apply IS2, we predict that the agent will display behaviour that would be causally efficacious in the satisfaction of the desires we’ve ascribed to it if the beliefs we’ve ascribed to it were true.⁶ If we ascribe false beliefs, a behaviour might satisfy this condition even if in actuality it isn’t causally efficacious in the satisfaction of the desires we’ve ascribed. In fact, though, the reliability of this kind of prediction depends on an implicit reference to the range of behaviours available to the agent. Lorna can only be expected to open the fridge if (a) this is something that she is able to do (e.g. she hasn’t locked herself out of the house) and (b) even if there were water in the fridge there would be no better way of satisfying her desire available to her (e.g. she doesn’t have a bottle of water in her hand). This suggests that IS2 can be reformulated as: IS2*

Predict that the agent will display a behaviour that would be most conducive (no less so than any other behaviour available to it) to bringing about the states of affairs that are the contents of the desires you’ve ascribed to it if the states of affairs that are the contents of the beliefs you’ve ascribed to it obtained.⁷

The strategy can be usefully reformulated in terms of conditional probabilities. If G is the content of a desire we’ve ascribed to the agent, S is the content of a belief ⁶ I prefer to speak of behaviour, rather than action, to emphasize the continuity between the Intentional Stance and other strategies we might adopt for predicting the behaviour of a complex system. Very crudely, behaviour becomes action as a result of its link with the beliefs and desires with which we connect it in our applications of the Intentional Stance. ⁷ In real-life behaviour prediction, the selection of the behaviour we predict is likely to involve an element of cost-benefit analysis. I won’t consider here this complication.

  

113

we’ve ascribed to the agent, and B is a behaviour available to the agent, let’s refer to the value of p(G|B&S)—the probability of G conditional on B and S—as B’s degree of G-efficacy modulo S.⁸ And let’s say that B is G-optimal modulo S just in case no other behaviour available to the agent has a higher degree of G-efficacy modulo S. Then we can reformulate IS2* in the following terms: If you have ascribed to the agent a desire with content G and a belief with content S, predict that it will display a behaviour that’s G-optimal modulo S. Notice that a behaviour prediction using this model won’t follow directly from your ascription of beliefs and desires to the agent. You will need additional information about the world in order to determine which of the behaviours at the agent’s disposal would be most conducive to satisfying those desires if those beliefs were true. This includes information about which behaviours are available to the agent and about the causal efficacy of these behaviours with respect to the desires you’ve ascribed to it in situations in which the beliefs you’ve ascribed to it are true (or the probability of the contents of the desires conditional on the contents of the beliefs and the behaviours). Even if we ascribe to the agent a desire for p and the belief that a certain behaviour would be most conducive to bringing about p,⁹ the prediction of this behaviour will have to appeal to information to the effect that the behaviour in question is as a matter of fact available to the agent. Even if we ascribe to Lorna the desire to drink water and the belief that opening the fridge is (part of ) the behaviour most conducive to her drinking water, the reliability of the prediction that she’ll open the fridge rests on the assumption that she is actually right in thinking that she can make her way to the fridge (that the kitchen door is not locked, etc.). So far we’ve been working with formulations of IS2 that include the concepts of belief and desire. On these formulations, you wouldn’t be applying the predictive strategy described in IS2 unless your predictions were based on belief and desire ascriptions. This feature seems to render IS2 unsuitable to figure in the meaning grounds of belief and desire ascriptions for the same reasons that we have deployed to reject IS1. In this case, however, mention of belief and desire ascriptions is dispensable, and the behaviour-predicting strategy can be formulated for an arbitrary BD discourse, independently of whether it produces belief and desire ascriptions. For this purpose we need to think of IS2* in abstraction from the fact that the states of affairs on which the prediction is based are postulated as the contents of

⁸ This magnitude cannot be used to compare the efficacy of behaviour with respect to different goals, as it hasn’t been calibrated to the probability of the goals. ⁹ This is not the typical scenario. Normally we base our behaviour prediction on the ascription of beliefs about features of the subject’s environment, not about the causal efficacy of possible behaviours.

114

 

the agent’s beliefs and desires. When considered in this way, the strategy will take the following form: Predict that the agent will display a behaviour that would be most conducive to bringing about certain states of affairs X if certain states of affairs Y obtained. Then we can say that a BD discourse will be involved in the application of this strategy when it uses the states of affairs with which it D-pairs the agent to fill the X argument place in this schema and the states of affairs with which it B-pairs the agent to fill the Y argument place. In other words, a BD discourse will be involved in the application of the behaviour-predicting strategy expressed by IS2* just in case behaviour prediction follows this template: Predict that the agent will display a behaviour that would be most conducive to bringing about the states of affairs with which you have D-paired it if the states of affairs with which you’ve B-paired it obtained. Or: If you have D-paired the agent with G and you have B-paired it with S, predict that it will display a behaviour that’s G-optimal modulo S. I’m going to refer to this predictive strategy as the minimal intentional (MI) strategy. We can now look at an arbitrary BD discourse and determine whether it is involved in the application of the MI strategy independently of whether or not its B- and D-pairings express belief and desire ascriptions. Now, if a BD discourse is involved in the application of the MI strategy, acceptance of sentences of the form ‘aBp’ and ‘bDq’ will be regulated in a certain way—in terms of the accuracy of the behaviour predictions they generate. And this acceptance procedure could in principle be incorporated in a pragmatist account of the meaning grounds of belief and desire ascriptions. On this proposal, part of what makes it the case that a BD discourse produces belief and desire ascriptions is that the acceptance of its sentences is regulated in this way. This proposal will need to be modified in important ways to avoid some problems, but once these modifications are introduced, the proposal will provide, I submit, the basis for a plausible pragmatist specification of the meaning grounds of belief and desire ascriptions.

6.3 Indeterminacy and Charity A familiar feature of the predictive strategy described by IS2 is that any given behaviour can be predicted by innumerable belief-desire combinations.

  

115

Specifically, for any behaviour and any desire, the behaviour can be predicted by ascribing the desire, so long as we are allowed to make adjustments in the beliefs we ascribe—the behaviour would be most conducive to bringing about satisfaction of the desire if the beliefs were true. Lorna’s opening the fridge door can be predicted by ascribing to her the desire to drink water and the belief that there’s water in the fridge, but also by ascribing to her the desire to build the Death Star and the belief that the blueprint is kept in the fridge. This is not a problem in the presence of IS1, since in normal circumstances it is more likely that Lorna ought to desire to drink water and believe that there’s water in the fridge than that she ought to desire to build the Death Star and believe that the blueprint is kept in the fridge. In general, as Dennett points out, IS1 has the consequence that we end up ascribing mostly true beliefs: ‘An implication of the intentional strategy, then, is that true believers mainly believe truths’ (Dennett 1987b: 19). This is, he explains, ‘an elaboration and further specification of the fundamental rule: attribute those beliefs the system ought to have’ (Dennett 1987b: 19–20). However, once IS1 is discarded, as I am proposing, we are left with a behaviour-predicting strategy that shows no preference for the ascription of true belief. This is the situation with the MI strategy. A BD discourse can be involved in the application of the MI strategy even if it is indifferent to the truth value of the states of affairs with which it B-pairs agents. If the pragmatist thinks that our preference for ascribing true belief is an essential feature of our practice, she will want this preference to be a consequence of her account of the meaning grounds of belief and desire ascriptions. She will want to maintain that a BD discourse shouldn’t count as producing belief and desire ascriptions unless the behaviour predictions it generates exhibit this preference in some form. As we’ve just seen, she won’t secure this result by specifying the meaning grounds of belief and desire ascriptions in terms of the acceptance procedure generated by the MI strategy. One way of achieving this would be to include in the meaning grounds of belief and desire ascriptions the acceptance procedure generated by versions of the MI strategy that exhibit a preference for B-pairing agents with obtaining states of affairs. However, it would be preferable if we could find a criterion that renders our preference for ascribing true beliefs intelligible. I’m going to argue that the pragmatist can achieve this by taking a closer look at the way in which our belief and desire ascriptions are involved in the prediction of behaviour.

6.4 The Ontogenesis of the Intentional Stance We know that humans can apply the MI strategy at least from the age of 4. This is the age at which children reliably pass classical displacement tests for false-belief understanding, and passing these tests requires applying the MI strategy. In a

116

 

famous instance of this kind of test, children listen to a narration illustrated with dolls. They are told that Sally has a basket and Anne has a box. Sally puts a marble in her basket and leaves the scene, leaving the basket, with the marble, behind. In Sally’s absence Anne moves the marble from the basket to the box. Then Sally returns to the scene. Where will Sally look for her marble? (Baron-Cohen, Leslie, and Frith 1985).¹⁰ Children under the age of 4 tend to give the wrong answer: Sally will look for her marble in the box, where the marble was placed in her absence. However, children from the age of 4 reliably give the correct answer: Sally will look for her marble in the basket. On the most natural description of how they achieve this, these children are applying the MI strategy: predicting that Sally will look in the basket because this behaviour would be most conducive to satisfying her desire to retrieve the marble if her (false) belief that the marble is in the basket were true. We know, then, that the ability to implement the MI strategy is present by age 4. In fact, recent research strongly suggests that this ability is present at a much earlier age—probably from the second year of life. This is not manifested in elicited-response tests, like the Sally-Anne test, until the age of 4, but much younger children reliably pass spontaneous-response tests that involve the application of the MI strategy. A typical experiment of this kind, using the violation-ofexpectation (VOE) method, is presented in (Onishi and Baillargeon 2005). The experiment involves a scene with a green box and a yellow box. An agent hides a toy inside the green box and later, in her absence, the toy is moved from the green box to the yellow box.¹¹ The agent returns and reaches either inside the yellow box or inside the green box. 15-month-olds reliably looked longer when the agent reached inside the yellow box than when she reached inside the green box, showing that they expected (predicted) that she would reach inside the green box. The infants’ prediction is naturally described as following the MI template: they predict that the agent will reach inside the green box because doing so would be the behaviour most conducive to satisfying her desire to obtain the toy if her belief that the toy is in the green box were true. I’d like to emphasize that by describing the subjects who succeed in these tasks as applying the MI strategy I’m not providing a full account of how their behaviour predictions are produced. Subjects who use the MI strategy to predict that Sally will look for her marble in the basket will single out this behaviour as one that would be most conducive to satisfying Sally’s desire if her belief were true. Hence in order to apply the strategy, subjects first need to decide what Sally desires and what she believes. The MI strategy does not dictate how to achieve this. It needs to be supplemented with procedures for deciding which beliefs and desires ¹⁰ The experimental model was first introduced by (Wimmer and Perner 1983). For a review see (Wellman, Cross, and Watson 2001). ¹¹ The method is based on the assumption that situations that depart from what subjects expect will produce surprise. In infants, the level of surprise produced by a visual input is manifested by how long they look at the input.

  

117

to ascribe, even if the ascriptions produced in this way are subsequently assessed and revised in light of their behaviour-predicting success under the MI-strategy. Several researchers have suggested that prior to the stage at which children manifest their mastery of the MI strategy with their successful performance in false-belief tests, they employ a more rudimentary strategy for predicting behaviour that shares some features with the MI strategy. I’m going to focus here on the account of this early behaviour-predicting strategy developed by Gergely Csibra, György Gergely, and others, who refer to it as the teleological stance. In a series of VOE experiments, these researchers have demonstrated the use of this more rudimentary behaviour-predicting strategy from the age of 9–12 months. In one of their experiments, 12-month-old infants were shown on a screen a habituation sequence consisting of a small circle and a big circle separated by an obstacle—a rectangular figure, with the small circle jumping over the obstacle and approaching the big circle (Gergely et al. 1995). The subjects were then presented with two kinds of test events with the small and big circles placed as before, but with the obstacle removed. In the first, the small circle followed exactly the same trajectory as in the habituation sequence, even though the obstacle was no longer there. In the second, the small circle approached the big circle following a straight trajectory. The subjects consistently displayed longer dishabituation times (i.e. looked longer) with the first test event, in which the small circle ‘jumps over’ a non-existent obstacle, than with the second test event, in which the small circle takes the most direct route to the big circle. This was taken to show that the infants expected the small circle to take the most direct route, and were surprised when it didn’t. It is natural to describe the situation as a case of behaviour prediction: the infants predict that the small circle will take the straight route. It is also natural to say that this prediction is reached through the attribution of a goal—the goal of approaching the big circle through the shortest route. The infants predict that when the obstacle is removed the small circle will follow the straight trajectory towards the big circle because this is the behaviour most conducive to achievement of this goal. In these respects, the strategy employed by the infants resembles the MI strategy employed by older children and adults. However, Csibra and Gergely (1998: 256) contend that there are two major respects in which the two strategies differ from each other. They characterize our procedure as involving two separate aspects, that we could label predictive and ascriptive. The predictive aspect consists in the way in which we select the behaviour we predict as a function of certain states of affairs. We predict that the agent will display a behaviour that would be most conducive to bringing about a state of affairs if certain other states of affairs obtained. We predict, for example, that Sally will display a behaviour that would be most conducive to getting the marble if the marble was in the basket. The ascriptive aspect consists in ascribing to the agent representational mental states with these states of affairs as their contents, playing a causal role in the production of the behaviour. In the

118

 

Sally-Anne example, we ascribe to Sally mental states that represent her obtaining the marble and the marble being in the basket, and we accord to these states a causal role in the production of the predicted behaviour. According to Csibra and Gergely, the 12-month-olds’ predictive routine differs from ours in both these respects. On the one hand, their predictive strategy appears to employ a function from states of affairs to behaviour that is formally simpler than the one we employ—a unary function in place of our binary function. We predict that the agent will display a behaviour that would be most conducive to bringing about A if B obtained. The 12-month-olds, by contrast, appear to predict that the agent will display a behaviour that will be as a matter of fact, i.e. by their lights, most conducive to bringing about A.¹² On the other hand, according to Csibra and Gergely, in the 12-month-olds the ascriptive aspect is simply missing. They don’t ascribe to the agent a mental state representing the relevant goal as a desired state of affairs. Let’s leave the ascriptive aspect for later and concentrate on the difference in prediction. Csibra and Gergely argue persuasively that the 12-month-olds really employ the simpler predictive strategy, rather than a limited version of ours, in which the second argument-place is restricted to obtaining states of affairs. They refer to it as the teleological stance. It can be formulated in the following terms: Predict that the agent will display a behaviour that is most conducive to bringing about the goal you’ve ascribed. If we define the degree of G-efficacy of a behaviour B as the value of p(G|B), and say that a behaviour is G-optimal if no other behaviour available to the subject has a higher degree of G-efficacy, the strategy can be reformulated as follows: If you’ve ascribed to the agent goal G, predict that it will display a G-optimal behaviour. Notice that subjects who fail the Sally-Anne test give the answer required by the teleological strategy. Looking in the box is the behaviour that is as a matter of fact most conducive to bringing about the goal of finding the marble.

¹² Many researchers have endorsed the idea that motivational concepts are mastered before informational concepts, phylogenetically as well as ontogenetically. Thus in their seminal study on mindreading in chimpanzees, David Premack and Guy Woodruff write: ‘Of all possible guesses, we find the most compelling one to be that inferences about motivation will precede those about knowledge, both across species and across developmental stages. Not even the chimpanzee will fail tests that require him to impute wants, purposes, or affective attitudes to another individual, but he may fail when required to impute states of knowledge’ (Premack and Woodruff 1978: 526). Henry Wellman has defended a similar view specifically with respect to human infants: ‘I suggest that children younger than three, say two-year olds, fail to understand belief-desire psychology. They utilize instead, a simple desire psychology’ (Wellman 1991: 19).

  

119

The MI strategy is clearly more powerful than the teleological strategy. Everything that can be predicted using the teleological strategy can also be predicted by the MI strategy. A behaviour that is most conducive to bringing about A is a behaviour that would be most conducive to bringing about A if B obtained, for any obtaining B. And so long as the second argument place in the MI predictive function is restricted to obtaining states affairs, the MI strategy achieves no more than the teleological strategy. The MI strategy comes into its own when this restriction is lifted. Then the MI strategy can produce predictions for which the teleological strategy offers no match. But clearly there is a price to be paid for this increased power in terms of computational resources. We need to go from considering the actual causal efficacy of possible behaviours with respect to a goal state to considering their hypothetical causal efficacy with respect to a goal state if things differed from actuality in certain respects. These considerations motivate an intriguing hypothesis about our adult predictive practice. On this account, the teleological strategy remains our default procedure for predicting behaviour. We continue to use it whenever there is no advantage to switching to the more involved MI strategy. This is reserved for hard cases in which the teleological strategy can’t produce the right predictions. In these hard cases, and only in these, we call upon the more sophisticated resources of the MI strategy.¹³ And when we do, we aim to minimize the difference between the actual situation (by our lights) and the conception of the situation that we ascribe to the subject to generate our prediction. This would make sense on the assumption that the further from actuality a situation is located, the harder it would be to compute the relative causal efficacy in that situation of the available behaviours with respect to the ascribed goal. In this way, we reserve the more computationally demanding procedure for cases in which it is really needed, reverting to the more economical approach everywhere else (Csibra and Gergely 1998: 258; Gert 2012: 117). Let’s refer to this overall approach to behaviour prediction, using the teleological strategy as a default and deploying the MI strategy only in the hard cases, as the hybrid policy.¹⁴

¹³ I think that this two-procedure hypothesis is in principle compatible with one-system accounts of adult mindreading (Carruthers 2017). On this picture the same system that produces intentional predictions would default to the pared-down teleological operation when the full resources of the system are not required. ¹⁴ Ian Apperly and Stephen Butterfill have defended ‘the existence of two types of system for belief reasoning: one that is cognitively efficient but limited and inflexible and another that is flexible but demanding of general cognitive resources’ (Apperly and Butterfill 2009: 956). However, they don’t think that the teleological strategy offers a plausible way of achieving cognitive efficiency (Apperly and Butterfill 2009: 961).

120

 

6.5 The Curse of Knowledge So far, I have motivated my hypothesis that we follow the hybrid policy in our adult behaviour predictions using the developmental story presented by Csibra and Gergely. However, the developmental story can only lend fairly weak support to the hypothesis. Even if things develop exactly as Csibra and Gergely claim before the MI strategy makes its appearance, once it is in place it might entirely displace the teleological strategy. In addition, the existence of a period before the MI strategy appears, in which only a more rudimentary predictive strategy is available, is coming under increasing pressure as new empirical results continue to bring forward the age by which infants can employ the MI strategy.¹⁵ And even if a pre-MI strategy existed, it might not have the features that Csibra and Gergely ascribe to it.¹⁶ However, we have reason to believe that the teleological strategy continues to play some role in our behaviour prediction after the MI strategy has made its appearance. It seems increasingly likely that we acquire the ability to apply the MI strategy in the second year of life at the latest, and yet the performance of 3-yearolds in elicited-response false-belief tests seems to suggest that they are still inclined to predict behaviour along the lines of the teleological strategy. This inclination might well be due to the processing difficulties involved in applying the MI strategy, but what matters for our purposes is that when the MI strategy can’t be applied, what children offer instead is the performance that you would expect from subjects who are applying the teleological strategy.¹⁷ Does this inclination disappear by the age of 4? There’s some empirical evidence suggesting that the inclination to default to the teleological strategy survives into adulthood. I think this is a natural way to interpret the phenomena studied in the literature on the socalled curse of knowledge.¹⁸ The curse of knowledge is the label used by Susan Birch and Paul Bloom to refer to ‘a tendency to be biased by one’s own knowledge when attempting to appreciate

¹⁵ See (Scott and Baillargeon 2017) for a review of recent results. ¹⁶ On one alternative (Baillargeon, Scott, and He 2010; Onishi and Baillargeon 2005), infant behaviour prediction is based, not on the subject’s own conception of the situation, but on true beliefs ascribed to the agent, which might not include all of the subject’s beliefs. Renée Baillargeon (personal communication) has now abandoned the view that the emergence of the MI strategy is preceded by a period in which infants use a more rudimentary strategy for behaviour prediction. On another alternative, infant behaviour prediction is based on the subject’s conception of the situation as well as her conception of which goals should be pursued (Perner and Roessler 2010; Roessler and Perner 2013). Perner and Roessler use for this view the label teleological account, and refer to Csibra and Gergely’s picture as the hybrid account. As they explain (Perner and Roessler 2010: 227), their approach also differs from Csibra and Gergely’s in that they are more concerned with the explanation of actions that with the prediction of behaviour. ¹⁷ Perner and Roessler have emphasized the relevance in this connection of the fact that the performance of 3-year-olds in these tasks, although poor, is far from random (Perner and Roessler 2010: 200; Roessler and Perner 2013: 44). ¹⁸ Thanks to Alex Jackson for alerting me to the relevance of this phenomenon.

  

121

a more naive or uninformed perspective’ (Birch and Bloom 2004: 256).¹⁹ Several studies show a tendency in adults as well as children to overestimate the knowledge of others in a variety of contexts.²⁰ As Birch and Bloom point out, young children’s difficulties with traditional false-belief tasks can be described as manifestations of the curse-of-knowledge bias. Children who say that Sally will look for the marble in the box are using their own knowledge to predict the actions of uninformed Sally. I have followed Csibra and Gergely in describing these cases as applications of the teleological strategy, and in general we can expect applications of the teleological strategy to produce instances of the curse-of-knowledge bias. Hence if as adults we continue to use the teleological strategy as our default procedure for predicting behaviour, as I’m suggesting, we should expect the persistence of the bias into adulthood, and this is precisely what we find. In fact, with ingenious modifications of the displacementtask template, Birch and Bloom have shown an underlying tendency in adults to give the wrong answer in false-belief tests (Birch and Bloom 2007). Their results are open to interpretations under which they wouldn’t support my case, but so long as these authors are right in seeing their results as manifestations of the curseof-knowledge bias, the results can be taken to lend some support to the hypothesis that adults follow the hybrid policy for predicting behaviour, using the teleological strategy as a default and shifting to the MI strategy only in cases that the teleological strategy can’t handle well.

6.6 The Hybrid Policy and the Meaning Grounds of Belief and Desire Ascriptions In Section 6.2 I suggested that the pragmatist could try to specify the meaning grounds of belief and desire ascriptions in terms of the acceptance procedure for B and D pairings generated by the MI strategy. In Section 6.3 I argued that this wouldn’t result in treating our preference for the ascription of true beliefs as an essential feature of our practice of ascribing beliefs and desires. In Section 6.4 I have presented some empirical evidence for the claim that our behaviour predictions based on the ascription of beliefs and desires don’t follow in general the MI strategy, but the hybrid policy, in which the use of the MI strategy is restricted to cases in which the teleological default doesn’t generate the right predictions. What I want to argue now is that the pragmatist could succeed in treating our preference for ascribing true beliefs as essential to the practice by

¹⁹ They borrow the label from (Camerer, Loewenstein, and Weber 1989). ²⁰ See (Birch and Bloom 2004; Ghrear, Birch, and Bernstein 2016) for references. For related results, see (Mitchell et al. 1996; Epley et al. 2004).

122

 

specifying the meaning grounds of belief and desire ascriptions in terms, not of the MI strategy, but of the hybrid policy. On this proposal, a BD discourse would generate belief and desire ascriptions only if it employed the acceptance procedure associated with the hybrid policy. This acceptance procedure is similar to the one generated by the MI strategy. It involves selecting B and D pairings on the basis of the accuracy of the behaviour predictions they generate, with a preference for minimizing the distance between actuality and the states of affairs with which agents are B-paired. Cases in which this distance is reduced to zero will correspond to applications of the teleological strategy, and when these produce perfectly accurate predictions, they will trump any alternative B and D pairings. The MI-strategy will kick in when the teleological strategy produces less than perfectly accurate behaviour predictions, and accuracy can be increased by B-pairing the agent with non-obtaining states of affairs. In general, when several sets of pairings produce equally accurate behaviour predictions, preference will be given to those which B-pair the agent with states of affairs that are closest to actuality. It’s even possible in principle that less accurate behaviour predictions are accepted for the sake of reducing the distance from actuality of the states of affairs the agent is B-paired with, if this is seen as outweighing the loss in accuracy, If the use of B and D is regulated in this way, agents will be B-paired with nonobtaining states of affairs only in exceptional circumstances. Lorna’s trip to the fridge can be predicted using the teleological strategy, as a behaviour as a matter of fact most conducive to bringing about the goal of drinking some water. Hence Lorna will be D-paired with the state of affairs consisting in her drinking water by default, with no need to B-pair her with a non-obtaining state of affairs. To be sure, her behaviour could also be predicted using the MI strategy by D-pairing her with the state of affairs of her building the Death Star, which would require B-pairing her with the non-obtaining state of affairs of the fridge containing the blueprint, but this prediction is trumped, under the hybrid policy, by the successful teleological default. The proposal then is to include the acceptance procedure generated by the hybrid policy in the meaning grounds of belief and desire ascriptions. This amounts to treating this acceptance procedure as a necessary condition for the B- and D-pairings of a BD discourse to count as belief and desire ascriptions. As we’ve just seen, it follows from this proposal that false belief will be ascribed only in exceptional circumstances—when the agent’s behaviour can’t be successfully predicted without taking this step. If regulation according to the hybrid policy is an essential feature of belief and desire ascriptions, our preference for ascribing true belief (or rather, our reluctance to ascribe false belief ) will be rendered intelligible. Notice that it’s a feature of this approach that the regulation of belief and desire ascriptions by the hybrid policy never calls for the ascription of true belief. This suggests a picture of interpretation that focuses on the ways that the agent’s

  

123

conception of the situation differs from the interpreter’s. Interpretation, on this approach, starts from the background assumption of a shared conception of the environment between interpreter and agent—or, from the interpreter’s point of view, the assumption that the agent sees things as they really are. This background assumption is overridden only when doing so leads to better predictions of the agent’s behaviour.²¹ The way in which predictive success regulates belief and desire ascriptions is much more complex than this brief outline might make it sound. One factor that complicates matters is that we don’t attribute beliefs and desires one by one. At any one time my interpretation of a typical subject will consist of a vast array of belief and desire ascriptions. If my behaviour predictions are based on only some of the beliefs and desires I’ve ascribed, and the predicted behaviour is not forthcoming, I might save the belief-desire ascription by taking the behaviour to answer to other beliefs or desires in my overall interpretation of the subject that weren’t considered for the prediction. I’ve attributed to John the belief that there’s a slice of chocolate cake in the fridge and the desire to eat cake so I predict that he will walk towards the fridge, but he doesn’t. I might be able to save these belief and desire ascriptions by invoking other beliefs and desires I might have also ascribed to him, e.g. the desire to lose weight or the belief that dinner will soon be ready. When this move is unavailable, as in cases of weakness of the will, the pressure towards revising my belief and desire ascriptions will be harder to resist. But even then we might be able to hold on to our existing ascriptions. The minimal intentional strategy is a powerful predictive tool, but many of the things we do are inevitably beyond its scope. This category includes what Rosalind Hursthouse has labelled as arational actions, explained by occurrent emotion, as, in one of Hursthouse’s examples, ‘throwing an “uncooperative” tin opener on the ground or out of the window’ (Hursthouse 1991: 58). The fact that these actions are not predictable by our current belief and desire ascriptions should not lead us to revise these ascriptions, even when the action would prevent satisfaction of the desires we’ve attributed if the beliefs we’ve attributed were true. As Hursthouse points out: ‘If I throw the only tin opener out of the window, I certainly shall not be able to open the tin and may have to go hungry to bed’ (Hursthouse 1991: 66). The problem posed by these cases doesn’t concern the specific beliefs and desires I’ve ascribed. They are forms of human behaviour that the intentional strategy can’t handle well. These complexities would have to be accommodated in a full account of how predictive success under the hybrid policy has to regulate B and D pairings for these to count as belief and desire ascriptions.

²¹ We have considered situations in which we ascribe belief in states of affairs that we believe not to obtain. We also sometimes ascribe belief in states of affairs that we don’t believe to obtain and we don’t believe not to obtain. In addition, we sometimes ascribe ignorance on matters on which we hold beliefs. A full development of the hybrid policy would have to make room for these possibilities.

124

 

6.7 Other Conditions I have proposed that the pragmatist should include in the meaning grounds of belief and desire ascriptions the acceptance procedure generated by their role in the prediction of behaviour, as described by the hybrid policy. I have argued that this approach succeeds in treating our preference for the ascription of true belief as an essential feature of the concept. In this section I want to review briefly other conditions that the pragmatist might want to include in the meaning grounds of belief and desire ascriptions.

6.7.1 Causation We can think of the hybrid policy as a procedure for indexing subjects with states of affairs, and the leading thought of the approach I’m exploring is that this indexing is what the ascription of beliefs and desires fundamentally consists in. The thought that the ascription of propositional attitudes has this indexing character is familiar from the literature on the measurement analogy (Churchland 1979: 105; Stalnaker 1984: 8–11; Matthews 2007), and the position I’m exploring should be seen as a (pragmatist) version of this approach, in which agents are indexed with the states of affairs on which interpreters base their predictions of the agents’ behaviour according to the hybrid policy. However, there are features of the way we ascribe beliefs and desires that we might regard as essential to the phenomenon but are not covered by the link with our predictive strategy. One is the thought that beliefs and desires cause the behaviour they help to predict. One might hold that a BD discourse that doesn’t acknowledge this causal dimension of its indexings of agents with states of affairs should not count as ascribing beliefs and desires. I want to suggest that the pragmatist might be able to accommodate this feature of the concepts of belief and desire with the demand that in order to count as belief and desire ascriptions, B- and D-pairings should be treated as incorporating a difference-making claim.²² Suppose that you predict that the agent will display the behaviour that would be most conducive to bringing about q if p obtained, but you think that the agent would display this behaviour even if it wasn’t the behaviour that would be most conducive to bringing about q if p obtained. We might want to say, as a first stab, that in order for your B- and D-pairings to count as belief and desire ascriptions, in these circumstances you shouldn’t B-pair the agent with p and D-pair it with q. ²² For difference-making accounts of causation, see (Lewis 1973, 2000; Woodward 2003; Woodward 2011; Andreas and Günther 2021). Perner and Roessler have highlighted the potential of differencemaking conceptions of causation in this context, although their own concern is to invoke them to vindicate the causal efficacy of that they call objective reasons (Perner and Roessler 2010: 208–9; Roessler and Perner 2013: 36–7).

  

125

If the causal dimension of belief and desire ascription could be incorporated into our pragmatist account of the meaning grounds of the discourse in this or some other way, the pragmatist would have the resources for accommodating the thought that belief and desire ascriptions can be used to explain, as well as predict, an agent’s behaviour, without coming into conflict with the idea that intentional explanation is a species of causal explanation (Davidson 1963).

6.7.2 Representation Another aspect of our conception of the practice that the pragmatist might want to treat as essential is the thought that beliefs and desires represent things as being a certain way. One way in which one might try to construe this intuition is in terms of counterfactual connections between belief, desire, and behaviour. The thought is that believing p and desiring q requires not only displaying the behaviour that would be most conducive to bringing about q if p obtained. In addition, it has to be the case that for any state of affairs x, if you desired x you would display the behaviour that would be most conducive to bringing about x if p obtained, and for any state of affairs y, if you believed y, you would display the behaviour that would be most conducive to bringing about q if y obtained. One might argue that this thought grasps the representational character of belief and desire.²³ Then the pragmatist would be able to accommodate the representational character of beliefs and desires by including in the meaning grounds of belief and desire ascriptions this aspect of our acceptance procedures: B-pair a with p only if, for every state of affairs x, if you D-paired a with x, you would predict that a would display a behaviour that would be most conducive to bringing about x if p obtained. D-pair b with q only if, for every state of affairs x, if you B-paired b with x, you would predict that b would display a behaviour that would be most conducive to bringing about q if x obtained.

6.7.3 Mind Reading The acceptance procedure based on the predictive success of belief and desire ascriptions is unlikely to be the only procedure we employ for this purpose. The reason is that its application needs to wait until the behavioural predictions generated by belief and desire ascriptions have been confirmed or refuted by the

²³ I think this is congenial to the approach in (Ramsey 1927).

126

 

agent’s actual behaviour. This means that interpreters will need to use some other procedure to generate in the first place the belief and desire ascriptions that will later be tested by the predictions they generate. The acceptance procedure generated by the hybrid policy is compatible with any procedure the interpreter might use for this purpose.²⁴ The question is, then, not whether the acceptance procedure generated by the hybrid policy needs to be supplemented by some other procedure—it does. The question is whether the procedure we use for this purpose should be included in the meaning grounds of belief and desire ascriptions or, on the contrary, our choice of procedure has no consequences for whether we count as ascribing beliefs and desires. I favour the second option. Different interpreters might employ different ascription heuristics in different circumstances, and this diversity should not in principle affect the status of their B- and D-pairings as belief and desire ascriptions, so long as they then assess their pairings in light of the accuracy of the behaviour predictions they generate. If this is right, then it follows that the pragmatist should refrain from building a commitment to a specific account of the character of mindreading into the meaning grounds of belief and desire ascriptions.²⁵ The meaning grounds of the discourse should not dictate how we ought to read minds. The pragmatist should only be claiming that however we achieve this, our ‘readings’ acquire the status of belief and desire ascriptions only by virtue of the fact that they are subsequently regulated by the accuracy of the behaviour predictions they generate.

6.7.4 Self-Interpretations We’ve discussed so far the procedures that regulate our acceptance of ascriptions of beliefs and desires to other subjects, and how these procedures can figure in the meaning grounds of the ascriptions. But we also ascribe beliefs and desires to ourselves, and these self-ascriptions typically employ a procedure that is not available to us when we are interpreting other people. For the case of belief, Gareth Evans offers an insightful characterization of the standard procedure for self-ascription: in making a self-ascription of belief, one’s eyes are, so to speak, or occasionally literally, directed outward—upon the world. If someone asks me ‘Do you think there is going to be a third world war?’, I must attend, in answering him, to precisely the same outward phenomena as I would attend to if I were answering the question ‘Will there be a third world war?’ I get myself in a position to answer ²⁴ See (Csibra et al. 2003: 129; Csibra and Gergely 2007). ²⁵ See (Carruthers and Smith 1996) for an overview of the debate.

  

127

the question whether I believe that p by putting into operation whatever procedure I have for answering the question whether p. (Evans 1982: 225)

In our terms, we can characterize the procedure as accepting a sentence that ascribes a belief to myself just in case I accept the sentence with which the content of the belief is specified. For example, I accept ‘I believe there’s water in the fridge’ just in case I accept ‘there’s water in the fridge’. Following Alex Byrne (Byrne 2011: 203), I’m going to refer to this as the transparent procedure of belief self-ascription.²⁶ Our standard procedure for self-ascription of desires follows a similar pattern. We characterized acceptance as the phenomenon consisting in considering how a sentence represents things as being and feeling convinced that that’s how things are.²⁷ We can describe our transparent procedure for desire self-ascription in terms of a parallel phenomenon, consisting in considering how a sentence represents things as being and feeling convinced that that’s how we’d like things to be. Let me refer to episodes of this kind as conations. Our transparent procedure for desire self-ascription can now be described as accepting a sentence that ascribes a desire to myself just in case I feel conation towards the sentence with which the content of the desire is specified. For example, I accept ‘I desire to go sailing this weekend’ just in case I feel conation towards ‘I’ll go sailing this weekend’.²⁸ Leaving the details of my construal to one side, I think it’s unquestionable that we employ transparent procedures for ascribing beliefs and desires to ourselves. The ascriptions that we generate in this way can then be used to predict our own behaviour along the lines of the MI strategy: I can predict that I will behave in ways that would promote satisfaction of my transparently self-ascribed desires if my transparently self-ascribed beliefs were true. These predictions are usually accurate, but they occasionally fail, and when they do, they might lead us to revise our transparent self-ascriptions. Using the transparent procedure, I ascribe to myself the belief that my neighbour is honest, but when I find myself refusing his offer to keep a set of keys to my house, I might conclude that I don’t actually have the belief. Or I might transparently ascribe to myself the desire to cycle to work, but when I realize that I’ve repeatedly passed up on opportunities to cycle to work, I might conclude that I don’t actually have the desire. For our purposes, the main question regarding the transparent procedure is whether its use should be included in the meaning grounds of belief and desire ascriptions. The question is, in other words, whether someone who didn’t employ the procedure could count as ascribing beliefs and desires. This is a very difficult

²⁶ On this topic, see also (Moran 2001; Shah 2003; Shah and Velleman 2005). ²⁷ Section 5.4, above. ²⁸ Jordi Fernández presents a slightly different construal of our transparent procedure for selfascribing desires, to which he refers as the bypass procedure (Fernández 2007).

128

 

question that I’m not going to try to settle here. All I want to do is to highlight the connection between the question faced by the pragmatist on this point and a related issue that has been raised in a different connection. A subject who ascribed beliefs and desires to herself and others but didn’t use the transparent method of self-ascription would suffer from what Sydney Shoemaker has called self-blindness: A self-blind creature would be one which has the conception of the various mental states, and can entertain the thought that it has this or that belief, desire, intention, etc., but which is unable to become aware of the truth of such a thought except in a third-person way. In other words, a self-blind creature could frame and understand ascriptions to itself of various mental states, but would be incapable of knowing by self-acquaintance whether such self- ascriptions were true. (Shoemaker 1988: 189)

According to Shoemaker, self-blindness is a conceptual impossibility.²⁹ I think it’s perfectly possible for a BD discourse to be regulated by the hybrid policy with no involvement of the transparent procedure. Hence, for our purposes, Shoemaker should be understood as claiming that a BD discourse that’s regulated in this way doesn’t count as producing belief and desire ascriptions. It follows, on this view, that the transparent procedure should be included in the meaning grounds of belief and desire ascriptions.³⁰ Maintaining, to the contrary, that the transparent procedure should not be included in the meaning grounds of belief and desire ascriptions enjoins a commitment to the possibility of self-blindness. On this view, BD discourses that are regulated by the hybrid policy, without employing the transparent procedure, could still count as generating belief and desire ascriptions. The participants in these practices would be self-blind (self-)interpreters.

6.7.5 Linguistic Evidence The hybrid policy is a procedure for regulating ascriptions of beliefs and desires that doesn’t take into account linguistic evidence—except indirectly, when the behaviour that we predict on the basis of our belief and desire ascriptions is linguistic behaviour. However, for linguistic creatures whose language we understand, by far the richest source of evidence for what they believe and desire is what they say. In the case of belief, if a speaker accepts a sentence, then if we interpret the sentence, as understood by the speaker, as representing state of affairs S, we ²⁹ See also (Boyle 2009). ³⁰ Evans appears to suggest a further constraint on who counts as ascribing beliefs: ‘a willingness to recognise, as relevant to the ascription of the predicate [“ζ believes that p”] to others, evidence of their having executed the same procedure—making a judgement as to whether p—which underlies his own self-ascription’ (Evans 1982: 226).

  

129

will ascribe to the speaker the belief that S obtains. This procedure massively expands the range of states of affairs we can ascribe beliefs about.³¹ If we interpret a sentence that a speaker accepts as representing the state of affairs consisting in there being no uncountable cardinals smaller than the continuum, we will be able to ascribe to her the belief that this state of affairs obtains. It’s hard to imagine how the hybrid policy could ever lead to the ascription of this belief. The interaction between belief ascription and linguistic interpretation also works in the opposite direction. If the hybrid policy sanctions the ascription of a belief with state of affairs S as its content, this should be treated as evidence against interpreting a sentence that the speaker rejects as representing S as obtaining, or a sentence she accepts as representing S as not obtaining. We will come back to this connection in the next chapter.

6.8 Conclusion In Chapter 5 I presented a general strategy for specifying the meaning grounds of declarative sentences in terms of the acceptance procedures that speakers employ for them and I proposed in outline applications of this strategy to our four problematic discourses. For belief and desire ascriptions, in particular, I proposed specifying their meaning grounds in terms of the acceptance procedure generated by Dennett’s Intentional Stance. My goal in the present chapter has been to develop this proposal for belief and desire ascriptions. I have argued first that we should dispense with Dennett’s appeal to the beliefs and desires the agent ought to have, and concentrate exclusively on the acceptance procedure generated by the role of belief and desire ascriptions in the prediction of behaviour—the MI strategy. I have then argued that the behaviour-predicting procedure in which our belief and desire ascriptions are involved should be characterized as employing not the MI strategy, but the hybrid policy, treating the more rudimentary teleological strategy as a default and invoking the MI strategy only for cases that the teleological strategy can’t handle well. This move enables us to build into the meaning grounds of belief and desire ascriptions our preference for the ascription of true beliefs. I have then considered other aspects of our acceptance procedure for belief and desire ascriptions that we might want to include in their meaning grounds. In the next chapter I concentrate on ascriptions of meaning and truth. My goal will be to articulate a plausible pragmatist account of the meaning grounds of these discourses.

³¹ Likewise for desire, as Dennett puts it, ‘The capacity to express desires in language opens the floodgates of desire attribution’ (Dennett 1987b: 20).

7 Meaning and Truth In Section 5.5 I proposed that the meaning grounds of meaning ascriptions (interpretations) could receive a pragmatist specification, using for this purpose a characterization of our acceptance procedure for interpretation based on the idea of radical translation/interpretation put forward by W. V. O. Quine and Donald Davidson. I also suggested that truth ascriptions could receive a similar treatment, with a characterization of our acceptance procedure in which the disquotational schema plays a central role. My goal in the present chapter is to develop these proposals in some detail. I’m going to offer a pragmatist account of the meaning grounds of interpretations based, with some modifications, on the central ideas of the characterization of our interpretative procedures put forward by Quine and Davidson.¹ On this account, interpretations will have the meanings they have by virtue of the fact that speakers employ this procedure for regulating their acceptance. I will then consider the precise role that the disquotational schema should play in the meaning-grounding characterization of our acceptance procedure for truth ascriptions. I will close by considering how the pragmatist should understand discourse concerning which discourses are representational.

7.1 Radical Translation/Interpretation An interpretation is a sentence ascribing a meaning to a linguistic expression, as understood by a speaker. As we saw in Section 3.3, meanings can be ascribed both to sentences, as in: N

‘ “La neige est blanche”, as understood by Pierre, means that snow is white’

and to subsentential expressions, as in: K

‘ “Tisch”, as understood by Kurt, means table’.

One of the central tenets of the Quine-Davidson approach is that the process of interpreting a language starts with the interpretation of whole sentences, as it is at this level, according to them, that evidence for interpretations is to be found. The ¹ I’ve presented the main ideas of my proposal in (Zalabardo forthcoming-b).

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0007

  

131

characterization of our interpretative practice I’m going to offer will follow this approach, focusing on the interpretation of declarative, representational sentences. In Section 3.3 I provisionally analysed interpretations of sentences as subject-predicate sentences, in which a predicate, e.g., ‘means that snow is white’ is ascribed to a sentence as understood by a speaker. Here I’m going to adopt the more standard analysis of interpretations as relational sentences, representing a sentence, as understood by a speaker, as bearing the binary relation referred to by the dyadic predicate ‘means’ to the item that the interpretation represents as the meaning of the sentence. What items play this role? This is an important point of disagreement between Quine and Davidson. Quine’s focus on translation entails that what he is interested in is pairings of sentences of the language to be translated, the object language, with sentences of the interpreter’s language. But Davidson takes issue with this approach: When interpretation is our aim, a method of translation deals with a wrong topic, a relation between two languages, where what is wanted is an interpretation of one (in another, of course, but that goes without saying since any theory is in some language). (Davidson 1973: 317)

Interpretation doesn’t aim to connect sentences of two languages with one another. It aims to connect sentences of one language, the object language, with the world—using for this purpose the interpreter’s understanding of the sentences of her language with which the object-language sentences are interpreted. One salient alternative to Quine’s translational approach is to take sentence interpretations to pair each object-language sentence with a possible state of affairs—with the state of affairs that the object-language sentence represents as obtaining, according to the interpretation.² This is the approach that I’m going to adopt.³ An interpretation of a declarative sentence, on this construal, will represent a relational state of affairs. Its first relatum will be a declarative sentence, as understood by a speaker. The second will be a possible state of affairs. The interpretation will then represent the following semantic state of affairs—that the sentence, as understood by the speaker, represents the state of affairs with which it’s paired by the interpretation.⁴ ² This is in fact not what Davidson is proposing. See (Davidson 2001b). ³ It may turn out that the two approaches are ultimately equivalent, as it can be argued that an interpreter can pair a sentence with a state of affairs only by pairing it with one of her sentences that represents the state of affairs as obtaining. The consequences of this are important and complicated. For the sentence-sentence approach, see (Carnap 1956; Davidson 2001b; Field 2017). ⁴ Pairing a sentence with the state of affairs it represents as obtaining might not amount to a full interpretation of the sentence, as sentence meanings are more fine-grained than states of affairs due to familiar Fregean considerations (Frege 1980b). I’m not going to be concerned with this aspect of the problem here.

132

 

A representationalist account of the meaning grounds of interpretations would have to identify the state of affairs that an interpretation represents as obtaining, and this can be expected to involve identifying the relation that the interpretation represents the object-language sentence as bearing to a state of affairs. The pragmatist approach I’m going to develop here doesn’t take this route. For the pragmatist, what makes an interpretation have the meaning it has is not a semantic relation between the interpretation and a state of affairs or its constituents. What makes an interpretation have the meaning it has is features of the procedure employed by speakers for regulating its acceptance. A pragmatist account of the meaning grounds of interpretations needs to identify the meaning-grounding features of their acceptance procedure. My goal here is to identify the features of our interpretative procedure that play this role by adapting some of the main ideas of Quine and Davidson’s characterization of radical translation/interpretation.

7.2 Compositionality We interpret declarative sentences, as in N, above, and we also interpret terms, as in K. Subject to some plausible assumptions it is reasonable to maintain that interpretation should start with terms, and that the interpretation of sentences should be the result of the interpretation of the terms that figure in them. One line of reasoning for this conclusion goes as follows. A declarative sentence represents things as being a certain way by representing a state of affairs as obtaining. On a standard metaphysical picture, states of affairs are produced by the combination of more simple items.⁵ Thus, for example, the state of affairs of Fido barking, on this picture, is produced when the individual, Fido, and the property of barking are combined with one another in the way that we call (monadic) instantiation— when the individual instantiates the property. Then all the states of affairs involving Fido will have a common constituent, and the same will go for all the states of affairs involving barking.⁶ This metaphysical picture is not mandatory. One might hold instead that states of affairs are ultimate, irreducible units, rather than the result of combining more elementary items. On this alternative picture, state-of-affair ‘constituents’, such as the individual, Fido, or the property of barking, should be regarded as abstractions based on similarities between states of affairs—between the states of affairs that we describe as concerning Fido, or those that we describe as involving barking.⁷ ⁵ See Section 1.2, above. ⁶ In (Zalabardo 2017), I refer to this view as the Combinatorial Account of Facts. ⁷ I’ve argued that Wittgenstein’s Tractatus Logico-Philosophicus (Wittgenstein 1974) advances this position (Zalabardo 2015: ch. 4; 2018). For this attribution, see also (Skyrms 1981; McCarty 1991). The position is also defended in (Armstrong 1997).

  

133

These two metaphysical pictures offer different accounts of the relationship between the state of affairs of Fido barking, on the one hand, and the individual Fido and the property of barking, on the other. On the first picture, individual and property are the fundamental items, and the state of affairs is construed as produced by the combination of these. On the second picture, the state of affairs is fundamental, and individual and property are construed in terms of similarity relations between states of affairs. However, both approaches are compatible with the following claim: CONSTITUENTS: the relationship between states of affairs and their constituents is essential to the identity of states of affairs. According to CONSTITUENTS, Fido barking would not be the state of affairs it is if it didn’t involve Fido and barking, whether we construe these as fundamental items or as resulting from relations of similarity between states of affairs. I’m going to assume that CONSTITUENTS is correct. Of course, as we move away from the simplest cases it becomes increasingly hard to identify the constituents of states of affairs, but here we are going to leave these difficulties to one side. Now, a system of propositional representation could have the following features: a. The items with which states of affairs are represented as obtaining (e.g. sentences, or thoughts) are built from a common stock of constituents (or exhibit common features). b. These constituents are paired with the constituents of the states of affairs that the system represents as obtaining. c. The pairing of representational items with states of affairs is derived from the pairing of constituents of the former with constituents of the latter: a representational item represents as obtaining the state of affairs whose constituents are the items paired with the constituents of the representational items.⁸ Let’s say that a system of propositional representation is compositional when it exhibits these features. It’s clear that our representations of states of affairs in English and other natural languages are compositional in this sense. The sentence ‘Fido barks’ represents Fido as barking as a result of a pairing between the name, ‘Fido’, and the dog, Fido, and between the predicate, ‘barks’, and the property of ⁸ This formulation assumes that each set of constituents can form (/be present in) at most one state of affairs. This assumption doesn’t hold universally. The referents of the terms that figure in the sentence ‘Desdemona loves Casio’ can produce two different states of affairs (Desdemona loving Casio and Casio loving Desdemona), but the sentence represents only one of them. This phenomenon is a source of important difficulties that will not concern us here.

134

 

barking. The state of affairs the sentence represents as obtaining has as its constituents the items paired with the constituents of the sentence. It has been argued that all human languages have to be compositional, or even that any system of propositional representation has to be compositional.⁹ I find these claims very plausible, and I’m going to restrict my discussion of our procedure for selecting interpretations to the interpretation of compositional languages. My conclusions will only apply to the interpretation of compositional languages, whether or not these exhaust the range of actual or possible languages. Our procedures for selecting interpretations of sentences typically exploit the compositional structure of these. We think that Pierre’s sentence represents snow as being white because we think that ‘la neige’, as understood by Pierre, refers to snow, and ‘est blanche’, as understood by Pierre, refers to the property of being white (i.e. because we think that all of Pierre’s sentences with ‘la neige’ as their subject represent states of affairs concerning how things stand with snow and all of his sentences with ‘est blanche’ as their predicate represent states of affairs consisting in something being white). Let’s say that an interpretation procedure for a compositional language is compositional when its interpretations are selected in this way. It’s an interesting and difficult question whether an interpretation of a compositional language has to be compositional. Michael Dummett gives an example of what a non-compositional interpretation of a compositional language would look like. He asks us to imagine that he hears a Basque sentence and is told that it means that the pigeons have returned to the dovecote, even though he cannot segment the sentence into components that he can recognize in other sentences (Dummett 1981: 308–9). According to Dummett, the knowledge of the meaning of the Basque sentence that we obtain in this way doesn’t count as real understanding. I’m very sympathetic to Dummett’s claim here, and to the general thought that a practice of pairing sentences with states of affairs has to be compositional in order to count as producing genuine interpretations. However, rather than offering support for this claim, I’m going to treat it as an assumption, restricting my attention to compositional interpretations (of compositional languages). I’m going to describe our procedure for selecting interpretations of compositional languages compositionally. If all interpretations of compositional languages have to be compositional, this will count as our universal procedure for interpreting compositional languages. If, in addition, all systems of propositional representation have to be compositional, the procedure I’m going to describe will be our universal procedure for interpreting propositional representations.

⁹ Donald Davidson has provided an argument for the view (Davidson 2001c). Wittgenstein’s Tractatus appears to defend the stronger claim that compositionality is essential to the very idea of propositional representation: ‘In a proposition there must be exactly as many distinguishable parts as in the situation that it represents’ (4.04). On this point see (Bronzo 2011).

  

135

7.3 Meaning and Belief Since we are focusing on compositional interpretations of compositional languages, we might seem to have committed ourselves to a characterization of our interpretative procedures according to which interpretation starts with the interpretation of terms, and considers the sentences in which these figure only at a later stage. However, Donald Davidson has rejected this entailment. He maintains that linguistic interpretation has to be compositional, but he argues that we can select term interpretations only by considering the interpretations they produce for the sentences in which they figure. He makes the point in the following passage: Theories of another kind start by trying to connect words rather than sentences with non-linguistic facts. This is promising because words are finite in number while sentences are not, and yet each sentence is no more than a concatenation of words: this offers the chance of a theory that interprets each of an infinity of sentences using only finite resources. But such theories fail to reach the evidence, for it seems clear that the semantic features of words cannot be explained directly on the basis of non-linguistic phenomena. The reason is simple. The phenomena to which we must turn are the extra-linguistic interests and activities that language serves, and these are served by words only in so far as the words are incorporated in (or on occasion happen to be) sentences. But then there is no chance of giving a foundational account of words before giving one of sentences. (Davidson 1973: 315)

One line of reasoning for the conclusion that whole sentences take priority in interpretation is based on the thought that our interpretation procedures take as their starting point the connection between meaning and belief. As we mentioned in Section 6.7.5, meaning ascriptions and belief ascriptions are inextricably linked. The link is established by the attitude of sentence acceptance. If a speaker accepts a declarative sentence, then by interpreting the sentence, as understood by her, as representing state of affairs S, we are committing ourselves to ascribing to her the belief that S obtains, subject to the slack between acceptance and belief. This link sustains a possible general approach for deciding on the interpretation of sentences. If we make some assumptions about the kinds of belief we should be ascribing, we could select our sentence interpretations on the basis of the extent to which they lead, through sentence acceptance, to ascribing beliefs of the right kind. I’m going to refer to acceptance procedures for interpretation that follow this approach as doxastic procedures. I’m going to argue, following Davidson and Quine, that our interpretation procedures are doxastic, in this sense. We select term interpretations on the basis of the extent to which the interpretations they generate for the sentences in which they figure that the speakers accept ascribe to them beliefs of the kind we think we should ascribe.

136

 

7.4 Charity The view that we use a doxastic procedure in interpretation is a central component of the accounts of translation/interpretation advanced by Quine and Davidson.¹⁰ The specific doxastic procedure that Quine and Davidson find in our interpretative practice is based on the principle of charity: we select interpretations with the goal of maximizing truth in the beliefs ascribed to the speaker as a result of each interpretation.¹¹ Satisfaction of the charity criterion is generally a matter of interpreting sentences that the speaker accepts with states of affairs that obtain and sentences that the speaker rejects with states of affairs that don’t obtain. However, the ordering of interpretations according to the degree to which they satisfy the criterion is complicated. It’s not simply a matter of how many of the sentences that the speaker accepts or rejects receive charitable interpretations. Some beliefs are more important than others. A set of beliefs A could provide a more accurate representation of the world than a set of beliefs B even if B contains more true beliefs and fewer false beliefs than A. Also, the ordering can only be partial. There are lots of cases in which neither of two sets of beliefs provides a more accurate representation of the world than the other. It will be interesting to see how the charity criterion works in an artificially simple scenario. Let’s suppose we are interpreting a very rudimentary language, all of whose sentences have a subject-predicate structure, with a predicate ascribed to a singular term. Let’s suppose that all combinations of a singular term with a predicate produce meaningful subject-predicate sentences in this language. A compositional interpretation of this language will pair each singular term with an individual and each predicate with a property. As a result, each sentence of the language will be paired with a possible state of affairs. To apply the charity criterion to an interpretation, we would consider whether it pairs the sentences the speaker accepts with obtaining states of affairs and the sentences she rejects with non-obtaining states of affairs. To see how this would work in more detail, suppose we have interpreted all the predicates of the language, and we need to select an interpretation for singular term ‘a’. To apply the charity criterion, we would consider the ‘a’-involving sentences that the speaker accepts and those that she rejects, and the properties with which the predicates in these sentences have been interpreted. We would then pick as the referent of ‘a’ the individual that is most accurately represented as ¹⁰ See (Quine 1960: ch. 2; Davidson 1973). ¹¹ Notice that the charity criterion is only effective for compositional interpretations. For a noncompositional interpretation the criterion can be maximally satisfied in each case by pairing every sentence that the speaker accepts with the state of affairs of snow being white and every sentence the speaker rejects with the state of affairs of snow being black (although for the sentences that Quine calls occasion sentences (Quine 1960: 35–36), this would force us to interpret different tokens of the same sentence with different states of affairs. Thanks to Javier Valdés on this point). With the compositionality constraint in place these trivializing interpretations are no longer available.

  

137

instantiating the properties that the interpretation pairs with predicates in the accepted sentences and as failing to instantiate the properties paired with the predicates in the rejected sentences. The points we made above about the ordering of interpretations generated by the charity criterion apply in this restricted scenario. The degree of satisfaction of the criterion by an interpretation of ‘a’ won’t be merely a matter of counting the ‘a’-involving accepted sentences interpreted with obtaining states of affairs and the ‘a’-involving rejected sentences interpreted with non-obtaining states of affairs. Also, we cannot assume that there will be a winner in each case. We can’t rule out cases in which instantiating the properties paired with the predicates in the ‘a’involving accepted sentences and failing to instantiate the properties paired with the predicates in the ‘a’-involving rejected sentences provides a maximally accurate description of more than one individual. A similar procedure would be employed to select the interpretation of a predicate ‘P’, assuming now that all the singular terms have already been interpreted. To apply the charity criterion, we would consider the ‘P’-involving sentences that the speaker accepts and those she rejects, and the individuals with which the singular terms in these sentences have been interpreted. We would pick as the referent of ‘P’ the property whose extension comes closest to including the individuals paired with the singular terms in the ‘P’-involving accepted sentences and excluding the individuals paired with the singular terms in the ‘P’-involving rejected sentences. As before, this won’t be simply a matter of how many of the relevant individuals each interpretation places in the right category. Also, we need to be open to the possibility that two or more properties satisfy the criterion to the maximum degree. But these simplifications are artificial. We can’t apply the charity criterion to the interpretation of predicates without applying it to the interpretation of singular terms and vice versa. We can apply the criterion only to an interpretation of all terms at once. In order to assess an interpretation according to the charity criterion, we would consider the states of affairs that would be represented, on that interpretation, by the sentences that the speaker accepts, and those that would be represented by the sentences she rejects. The degree to which the interpretation satisfies the criterion would be determined by how accurately the world is represented as involving the obtaining of the former states of affairs and the non-obtaining of the latter.

7.5 Permutations The claim that our interpretative procedure is based on the principle of charity has received many objections. One prominent line of attack is most easily presented with respect to an account of the interpretation of predicates according to which

138

 

they are paired, not with properties, but with sets of individuals—the individuals the predicate is true of.¹² Consider the simple language introduced in the previous section, and take an arbitrary assignment of referents to the singular terms of the language that pairs different terms with different referents. Now, if ‘P’ is a predicate of the language, consider the ‘P’-involving sentences that the speaker accepts, and interpret ‘P’ with the set of objects paired by the interpretation with the singular terms that figure in these sentences. If we do this for every predicate of the language, we get an interpretation on which all the sentences that the speaker accepts come out true and all the sentences she rejects come out false. The interpretation satisfies the charity criterion perfectly.¹³ Clearly our procedure for selecting interpretations doesn’t generally favour candidates produced in this way. An interpretation of this kind can be built on any interpretation of the singular terms, but we do think that some interpretations of the singular terms are better than others. Furthermore, we routinely favour interpretations that result in the ascription of some false beliefs over those that produce only true-belief ascriptions. One device that we use for selecting interpretations of singular terms is ostension. With an ostensive gesture, a speaker can point to the location of the referent of a singular term.¹⁴ Our simple language could be adapted to accommodate ostension by adding to its vocabulary the singular term ‘this’, to be deployed on one side of the identity sign with another singular term on the other. The procedure is not completely unambiguous—e.g. pointing at the statue is indistinguishable from pointing at the lump of clay—but it can rule out huge numbers of candidates. Notice also that we might occasionally accept an interpretation that pairs a singular term with an individual that wasn’t in the location pointed at by an ostensive explanation. It’s perfectly possible for a speaker to occasionally misidentify an object as the referent of one of her terms. This feature of our practice can be accommodated by the charity criterion. This would be achieved by treating ostensive explanations as on a par with other sentences the speaker accepts. Other things being equal, an interpretation that makes ostensive explanations come out true would be preferred to one that doesn’t, but the best overall interpretation could make ostensive explanations come out false. Ostensive explanations of predicates are also possible. An ostensive explanation of a predicate will come out true on an interpretation when the ostended point contains an instance of the property with which the predicate is interpreted. These can be formulated in our toy language with subject-predicate sentences with ‘this’ as their subject.

¹² On this approach, see (Hochberg 1967). ¹³ A version of this argument can be found in (Quine 1969). It was later used by Putnam in support of his conclusion that the ideal empirical theory must be true (Putnam 1978). See also (Field 1975; Wallace 1977; Putnam 1981). ¹⁴ See Quine’s discussion of ostension in (Quine 1969: 39–41).

  

139

However, the cull of candidate interpretations that can be achieved by taking ostension into account will not completely solve the problem. For any interpretation of the singular terms that makes the ostensive explanations come out true, we can still use the procedure described to pair predicates with extensions in such a way that all the accepted sentences and ostensive explanations of predicates come out true and all the rejected sentences come out false. Does the problem afflict also positions according to which predicates are interpreted with properties rather than sets of individuals? The answer depends on our view on which properties exist. Clearly, if we hold that for every set of individuals there is a property with that set as its extension, moving from extensions to properties won’t improve our prospects. Even if we limit our commitment to finite sets of individuals the problem will persist. Beyond this, further restrictions on which properties are eligible as predicate referents will open the possibility of cases in which an interpretation that satisfies the charity criterion to a greater extent than any other will make some of the accepted sentences come out false. David Lewis has developed an influential version of this approach. According to Lewis, properties are ordered by their degree of naturalness, and this should be taken to determine their degree of eligibility as predicate referents (Lewis 1983, 1984). On one reading of the proposal, all properties are assumed to be definable in terms of an elite class of perfectly natural (microphysical) properties, and the degree of naturalness of a property is determined by the syntactic complexity of its simplest definition in these terms (Williams 2007). We could apply this idea is to the task of characterizing our selection procedure for interpretations. On this proposal, our criterion for selecting interpretations would combine charity with the eligibility of the properties that each interpretation uses as referents. On this approach, when two interpretations satisfy the charity criterion to the same extent, one will be preferable to the other if it pairs predicates with more eligible properties. Furthermore, an interpretation that loses out on the charity criterion might still be preferable overall, if its charity deficit is compensated by a sufficient gain in the eligibility of its predicate referents. Even if we concede that the category of perfectly natural microphysical properties is legitimate, and that all properties can receive finite definitions in terms of these, one could call into question the claim that adding eligibility to charity generates a plausible characterization of our criterion for selecting interpretations. There’s no guarantee that the referents assigned by intuitively preferable interpretations will have a higher degree of eligibility than those assigned by artificially generated alternatives. In fact, as Robert Williams has argued, it is possible to describe situations in which ‘crazy’ referent assignments will be more eligible than those that our interpretative intuitions would favour (Williams 2007: 388–91). And independently of the details of these arguments, there is no plausibility to the idea that our procedures for selecting

140

 

interpretations favour the assignment of referents with simpler definitions in terms of microphysical properties.¹⁵

7.6 Familiarity I want to propose a different account of why we sometimes favour interpretations that don’t satisfy the charity criterion to a greater extent than available alternatives. On this account, we prefer interpretations that assign referents with simple definitions, not in terms of perfectly natural microphysical properties, but in terms of properties referred to by our atomic concepts. In general, we rank interpretations according to how easily we can define concepts for the individuals and properties they employ in terms of concepts we have. We can illustrate how this consideration interacts with the charity criterion with an example discussed by Andrew Woodfield (Woodfield 1982: 276–7). As Woodfield explains, Spanish speakers are typically inclined to ascribe the predicate ‘rubio’ not only to blond hair but also to hair that is much too dark to count as blond. In light of this, the charity criterion would favour interpreting the predicate ‘rubio’ as denoting the property of being either blond or light brown over interpreting it as denoting the property of being blond. Nevertheless, we may well prefer the latter interpretation even though it fares less well with respect to the charity criterion. I’m claiming that this is due to the fact that we have a concept denoting the property of being blond, whereas a concept for the property of being blond or light brown has to be constructed out of other concepts we do have. Notice that this preference would be hard to explain in terms of a notion of objective eligibility of properties, as the contrast between blond and not blond doesn’t seem to be more natural or objective than the contrast between the hair shades to which Spanish speakers apply the term ‘rubio’ and those to which they don’t. On a construal of our interpretative practice on which this factor is taken into account, the problem considered in the previous section doesn’t arise. The artificially constructed interpretations that guarantee maximal satisfaction of the charity criterion typically use as predicate referents properties for which we have no concepts, and constructing concepts for these properties in terms of concepts we have would involve highly complex definitions. This is the reason why they lose out to other interpretations that fare less well on the charity criterion. So my proposal at this point is that our interpretative practice should be characterized as employing an additional criterion alongside charity, to which I’m going to refer as familiarity:

¹⁵ On this point see (Shaw forthcoming).

  

141

Select interpretations on the basis of how easy it is to define in terms of our atomic concepts the properties and individuals that each interpretation uses as referents.¹⁶ Thus, interpreting ‘rubio’ as blond does worse than interpreting it as blond or light brown on the charity criterion, but better on the familiarity criterion.¹⁷ The need to weigh up two criteria to select interpretations introduces another possible source of incommensurability between interpretations. When one interpretation does better than another interpretation on one criterion but worse on the other, it might not always be clear which of the two interpretations should be preferred all things considered.

7.7 Reference and Causation There’s an important family of cases that pose a problem for the charity criterion and can’t be solved in terms of familiarity. Consider an otherwise normal English speaker who is unaware of the relevant biological facts and accepts sentences ascribing the predicate ‘is a fish’ to singular terms that we have interpreted as referring to all types of swimming creatures, including both fish and marine mammals. Consider now the contest between interpreting ‘is a fish’, as meant by her, as referring to the property of being a fish and interpreting it as referring to the property of being a swimming creature. Clearly, interpreting it as referring to the property of being a swimming creature does better on the charity criterion. It might do slightly worse on the familiarity criterion, but it’s hard to see how this could outweigh the significant charity dividend. I don’t think we can avoid the conclusion that if our interpretative practice was governed by the charity and familiarity criteria we would prefer interpreting the predicate with the property of being a swimming creature to interpreting it with the property of being a fish. And yet, that might not be the right interpretation. We can easily fill in the details of the case in such a way that it seems clear to us that the predicate should be interpreted with the property of being a fish and that we should ascribe to the speaker as a result false beliefs to the effect that marine mammals are fish (suppose, e.g., that she has the disposition to revise her verdict upon learning the relevant biological facts).

¹⁶ Notice that the procedure employed by Williams to describe situations in which Lewis’s eligibility criterion is maximally satisfied by crazy referent assignments cannot be used for the alternative I’m proposing. The complexity of the definition of a property in terms of my atomic concepts is not contingent on the microphysical structure of the universe. ¹⁷ The familiarity criterion is related, I think, to the principle to which Davidson refers as reverse charity: ‘there is a principle of reverse charity that judges a theory better the more of its own resources it reads into the language for which it is a theory’ (Davidson 1979: 8).

142

 

A similar example can be obtained by adapting a case discussed by Saul Kripke. Kripke asks us to imagine that, contrary to what most people think, Gödel was not in fact the author of the incompleteness theorem: A man named ‘Schmidt’, whose body was found in Vienna under mysterious circumstances many years ago, actually did the work in question. His friend Gödel somehow got hold of the manuscript and it was thereafter attributed to Gödel. (Kripke 1980: 84)

Suppose we know these facts, but most other people don’t. Consider now a speaker who is not aware of these facts and accepts the sentence ‘Kurt Gödel discovered the incompleteness of arithmetic’. Suppose we have interpreted the predicate ‘discovered the incompleteness of arithmetic’ as referring to the property of having discovered the incompleteness of arithmetic. Consider now the contest between interpreting the singular term ‘Gödel’, as understood by this speaker, as referring to Gödel and interpreting it as referring to Schmidt. It is clear that the details of the case can be filled in in such a way that the latter interpretation is superior from the point of view of charity: the ‘Gödel’-involving sentence that the speaker accepts would come out true on the Schmidt interpretation but false on the Gödel interpretation, and there is no difference between the two interpretations with respect to the familiarity criterion. And yet, as Kripke argues, it would be wrong to interpret the speaker as referring to Schmidt by the term ‘Gödel’. ‘Gödel’, as meant by her, refers to Gödel, and the belief we attribute to her as a result of the ‘Gödel’-involving sentence she accepts is the false belief that Gödel discovered the incompleteness of arithmetic. Once again, the charity and familiarity criteria give the wrong results.¹⁸ Cases of this kind are usually considered in connection with the project of identifying representationalist meaning grounds for interpretations—by specifying what has to be the case in order for a sentence to represent a certain state of affairs as obtaining or for a term to have a certain referent. In this context, these cases are usually invoked in support of causal accounts of reference. On a familiar account of the reference of natural-kind terms, ‘fish’ refers to the biological kind that’s causally responsible for the surface features on the basis of which we decide to apply the predicate. And on an influential account of the reference of names, the referent of ‘Gödel’ is determined by an act of baptism, in which a man received that name, to which our use of the name can be traced back through a chain of communication. However, even if we don’t think these ideas will ultimately sustain a successful representationalist account of the meaning grounds of interpretations, one might

¹⁸ For another example of this phenomenon, see (Grandy 1973: 445).

  

143

think that cases like these force us to adopt an account of our interpretative procedure according to which we assign referents to terms on the basis of causal relations between terms and referents. This would involve abandoning not only the charity criterion but also the more general idea that we select interpretations on the basis of a doxastic criterion—in terms of the belief ascriptions generated by each interpretation, via the sentences that speakers accept or reject. On the resulting account of interpretation, our procedure for pairing, say, ‘is a fish’ or ‘Gödel’ with their referents would not take into account which sentences involving these terms speakers accept or reject. I’m going to argue that the advocate of doxastic criteria doesn’t need to concede defeat at this point. These cases do show that the charity criterion doesn’t provide an accurate characterization of our interpretative procedure, but they can be successfully accommodated by a slightly different doxastic criterion.

7.8 Projection Advocates of the charity criterion often emphasize the practical indistinguishability between charity and another doxastic criterion: Select interpretations on the basis of the extent to which the beliefs attributed as a result of each interpretation agree with the beliefs of the interpreter.¹⁹ The charity criterion and the agreement criterion are not equivalent. If the interpreter has false beliefs, the two criteria might generate different orderings of interpretations. However, interpreters will always obtain the same results with both criteria, since in order to determine the extent to which the beliefs attributed by an interpretation agree with how things stand in the world, all the interpreter has to go on is her own beliefs about how things stand in the world. It follows that replacing the charity criterion with the agreement criterion would not give us any advantage in dealing with the cases we considered in the previous section. I’m going to argue, however, that a slight modification of the agreement criterion produces a doxastic criterion with the potential for accommodating the problematic cases. The proposal I want to explore is an instance of what Daniel Dennett has labelled projective principles. According to projective principles, Dennett writes, one should attribute to a creature in its circumstances ‘the propositional attitudes one supposes one would have oneself in those

¹⁹ Davidson explains that the method of interpretation he describes proceeds by ‘assigning truth conditions to alien sentences that make native speakers right as often as plausibly possible, according, of course, to our own view of what is right’ (Davidson 1973: 324). See also (Williams 1999) on this point.

144

 

circumstances’ (Dennett 1987a: 342–3).²⁰ This approach readily suggests a doxastic criterion for selecting interpretations. According to the agreement criterion, as we’ve seen, we should select interpretations on the basis of the extent to which the beliefs attributed as a result of each interpretation agree with the beliefs of the interpreter. A minor modification of this gives rise to the projection criterion: Select interpretations on the basis of the extent to which the beliefs attributed as a result of each interpretation agree with the beliefs the interpreter would have if she found herself in the speaker’s epistemic situation. We can easily see that the projection criterion gives the right results for the cases considered in the previous section. An interpreter who knows that dolphins aren’t fish can easily recognize that if she found herself in the speaker’s epistemic situation she would believe that dolphins are fish. Hence the projection criterion will favour interpretations that result in the ascription of this belief. The same goes for the Gödel/Schmidt case. An interpreter who knows that Schmidt proved the incompleteness of arithmetic realizes that she would believe that Gödel did if she found herself in the speaker’s epistemic situation. Hence, other things being equal, an interpretation that ascribes to the speaker the false belief that Gödel proved the result would do better by the projection criterion than one that ascribes the true belief that Schmidt did. It is interesting to consider the relationship between the projection criterion and the agreement criterion. When interpreter and speaker find themselves in the same epistemic situation, both criteria produce the same ordering of interpretations, since the beliefs I would have in the speaker’s epistemic situation will be the beliefs I actually have. However, when speaker and interpreter find themselves in different epistemic situations the two criteria will produce different results, and the more the two situations come apart, the more the interpretations favoured by the two criteria will differ from one another.

7.9 Linguistic Interpretation and Non-linguistic Belief Ascription I think that the combination of the projection and familiarity criteria provides a fairly accurate characterization of our interpretative procedure. There is, however, one aspect of how we interpret sentences that this characterization doesn’t register. It concerns another consequence of the connection between meaning ascriptions and belief ascriptions sustained by the phenomenon of sentence ²⁰ The approach can be traced back to (Quine 1960: 219). Versions of the view have been defended by (Grandy 1973; Stich 1981, 1983).

  

145

acceptance/rejection. Linguistic episodes are the most powerful tool for belief ascription to humans, but, as we saw in the preceding chapter, we have a procedure for ascribing beliefs in the absence of linguistic evidence—to regulate our acceptance of belief (and desire) ascriptions by their success in predicting the subject’s behaviour under the hybrid policy. This means that for a restricted but important range of states of affairs, we have non-linguistic procedures for determining whether a subject believes in their obtaining. If the subject picks her umbrella when she leaves the house, this supports ascribing to her the belief that it’s going to rain. This support will be conditional on hypotheses concerning her desires (e.g. that she doesn’t want to get wet) but we can leave this complication aside for our present purposes. Now, suppose there are two rival interpretations of a sentence the subject accepts: according to one, the sentence represents the weather as being dry; according to the other, the sentence represents the weather as being wet. Her picking up the umbrella will give us a reason for favouring the second interpretation over the first—or if the first interpretation has considerable independent support, for giving up the claim that the speaker accepts the sentence. When we proceed in this way, we aim to adjust our linguistic interpretation to the beliefs we have ascribed to the speaker on the basis of non-linguistic evidence. I think that this feature of our procedure is essential to it, and should therefore be included in the meaning grounds of interpretations.²¹ We select interpretations by seeking to maximize the combined degree of satisfaction of the familiarity, projection, and non-verbal criteria. And the fact that we select interpretations in this way is what makes them have the meaning they have—a practice of pairing sentences with states of affairs counts as producing interpretations just in case acceptance of its pairings is regulated by this procedure. We can now summarize the overall picture of how our meaning-grounding procedures for ascribing propositional attitudes and linguistic meanings interact with one another. As we saw in the previous chapter, we have a procedure for regulating the ascription of propositional attitudes that doesn’t rest on meaning ascriptions—we select belief and desire ascriptions on the basis of their success in predicting behaviour under the hybrid policy. These propositional-attitude ascriptions can then be used as evidence for the ascription of linguistic meanings, as we’ve just explained—when we have non-linguistic evidence for ascribing to a speaker the belief that a state of affairs obtains (/doesn’t obtain), we try to avoid interpreting a sentence that the speaker rejects (/accepts) as representing this state of affairs.

²¹ David Lewis rightly accused W. V. O. Quine of failing to take account of this factor in his construal of our interpretative practice: ‘Too much emphasis goes to language as a vehicle for manifestation of belief and belief as manifest in language; not enough either to language as a social practice or to belief as manifest in non-linguistic behavior’ (Lewis 1974: 341).

146

 

However, the influence of independently supported propositional-attitude ascriptions on linguistic interpretation is relatively marginal. Our overall procedure for ascribing meanings, as described in the present chapter, combines the evidence obtained in this way with the projection and familiarity criteria. Projection rests on an assumption concerning speakers’ beliefs—that they believe what we would believe if we found ourselves in their epistemic situation. This is not an independently supported hypothesis, but an assumption that defines the acceptance procedure that we call linguistic interpretation. However, the assumption can be revised if this is required for maximizing the combined degree of satisfaction of our three criteria. The sentence interpretations that we reach in this way can then be used to massively expand the scope of our belief ascriptions, by ascribing belief in the states of affairs with which we’ve interpreted the sentences that the speaker accepts. In many cases, the beliefs that we ascribe in this way don’t have a significant influence on how the subject would need to behave in order to satisfy the desires we’ve ascribed to her. Hence, these belief ascriptions cannot be independently assessed by their success in predicting behaviour under the hybrid policy.

7.10 Self-Interpretation The acceptance procedure for interpretations I have described in the preceding sections are clearly intended for the interpretation of the linguistic expressions of other speakers. But I can ascribe meanings to my own linguistic expressions as well as to those of others. It’s possible in principle to apply the third-person procedure to the interpretation of my own linguistic expressions. This would involve accepting self-interpretations that maximize satisfaction of the projection, familiarity, and non-verbal evidence criteria. But this is not how we usually proceed. In self-interpretation, homophony trumps all other considerations. We accept as a matter of course interpretations of our own linguistic expressions that use those very same expressions to specify their own meanings, as in: ‘Snow is white’ as understood by me, means that snow is white. ‘Table’, as understood by me, means table. Other incompatible interpretations might conceivably do as well as these, or even better, with respect to the third-person criteria.²² However, it is a feature of our

²² Quine contemplates this possibility for the interpretation of fellow speakers of our language: ‘Our advantage with a compatriot is that with little deviation the automatic or homophonic (§13) hypothesis

  

147

overall acceptance procedure for interpretations that homophonic interpretations are always accepted, and their rivals are always rejected. It’s hard to imagine what an interpretative practice lacking this feature would look like—one that applied to self-interpretation the same rules as to the thirdperson case. There might be reasons for thinking that if someone pairs sentences with states of affairs (or predicates with properties) in this way, they shouldn’t count as interpreting—as meaning by ‘means’ what we mean by it. This would recommend including homophonic self-interpretation in the meaning grounds of interpretations. On the other hand, it might be that someone could be counted as interpreting even if they didn’t attach particular value to homophonic selfinterpretations. Then the procedure should not be included in the meaning grounds of interpretations. I’m not going to take sides on this debate here.

7.11 Consequences of Pragmatism Here, as in other applications of the pragmatist approach, it’s important not to lose sight of the contrast between pragmatist accounts of the target sentences and representationalist accounts built on the same resources. On the representationalist approach, the meaning grounds of interpretations would be specified by identifying the states of affairs they represent as obtaining. One might try to find states of affairs that would play this role among the factors we have discussed. The charity criterion could be put to this use. On the resulting proposal, the state of affairs represented by N would consist in this: The compositional interpretation of Pierre’s language that maximizes satisfaction of the charity criterion pairs ‘la neige est blanche’ with the state of affairs of snow being white. This proposal, as it stands, doesn’t succeed, as it falls prey to the problems discussed in Section 7.5. For any given interpretation of the singular terms (that makes the ostensive explanations come out true), we can manufacture an interpretation of the predicates that satisfies the charity criterion perfectly, but the sentence-state of affairs pairs that this interpretation produces do not seem to pair the sentences with the states of affairs they represent as obtaining. One might try to solve the problem with an objective eligibility ranking of potential referents. Now N would represent the following state of affairs:

of translation fills the bill. If we were perverse and ingenious we could scorn that hypothesis and devise other analytical hypotheses that would attribute unimagined views to our compatriot, while conforming to all his dispositions to verbal response to all possible stimulations’ (Quine 1960: 70).

148

 

The compositional interpretation of Pierre’s language that maximizes satisfaction of a criterion composed of charity and referent eligibility pairs ‘la neige est blanche’ with the state of affairs of snow being white. I’m not going to assess this approach here. What I want to emphasize is that it is fundamentally different from my proposal. Notice that as we replace charity with agreement and then projection, and objective eligibility with familiarity, the resulting representationalist account loses whatever plausibility it might have enjoyed. Satisfaction of the charity plus eligibility criterion is an objective matter. The ranking of interpretations it generates is in principle independent of the identity of the interpreter. By contrast, the ranking of interpretations generated by the agreement or projection criterion is radically dependent on who is doing the interpreting. If you and I have different beliefs, and would have different beliefs if we found ourselves in the speaker’s epistemic situation, agreement and projection will favour different interpretations for each of us. The same goes for familiarity, for interpreters who have different conceptual repertoires. It follows that a representationalist account of the meaning grounds of interpretations based on these criteria would result in a radically relativistic account of meaning. The same sentence would represent different states of affairs for different interpreters, and the question, which state of affairs a sentence represents as obtaining, could be made sense of only when relativized to an interpreter.²³ I regard this outcome as tantamount to a refutation of the view. But this is not what I’m proposing. I’m not using features of our interpretative practice to specify the state of affairs that an interpretation represents as obtaining. I’m using them to specify what makes it the case that interpretations have the meaning they have—what makes it the case that pairing a sentence with a state of affairs has the character of an interpretation of the former as representing the latter. Given that this is our goal, there’s no reason why we should expect that whenever these conditions are satisfied we will end up with the same interpretations. Interpreters can disagree with one another on which interpretations are correct and still count as interpreting. Specifying the meaning grounds of interpretations requires specifying who counts as interpreting. It doesn’t generally require, in addition, specifying which interpretations are correct. If we adopt the representationalist approach, our specification of who counts as interpreting will rest on a specification of which interpretations are correct. However, the pragmatist approach avoids this link. It specifies who counts as interpreting without specifying which interpretations are correct. Therefore, the fact that different interpretations might maximally satisfy the projection and familiarity criteria ²³ Davison seems to have contemplated overcoming this obstacle by reference to the charitable interpretations that would be produced by an omniscient interpreter (Davidson 2001a).

  

149

for different interpreters doesn’t force the pragmatist to accept that they are all correct relative to the interpreters in question. They are all correct in one sense— they are the sentence-state of affairs pairings that interpreters need to accept in order to count as interpreting. However, this is compatible with maintaining that there’s another, absolute notion of correctness under which at most one of two rival interpretations is correct.²⁴ As we’ll see in Section 7.13, below, this absolute sense of correctness is required for interpretations to count as representational. Notice also that, for a given interpreter, satisfying the conditions for counting as interpreting doesn’t always single out a unique interpretation as the one she should endorse. If we were in the business of providing a representationalist account of the meaning grounds of interpretations, we would have to conclude from this that, even when relativized to a particular interpreter, interpretation is indeterminate—there is no fact of the matter as to which of several interpretations is correct. However, since we are operating within the pragmatist template, this conclusion doesn’t follow. The pragmatist is not committed to the claim that if you satisfy the conditions that turn your pairings of sentences and states of affairs into interpretations, whatever interpretations you end up with will be correct. The fact that we are not trying to specify the states of affairs that interpretations represent as obtaining also defuses a line of objection against the projection criterion. The problem is mentioned by Quine when he discusses what we’ve called projection principles: ‘Casting our real selves thus in unreal roles, we do not generally know how much reality to hold constant’ (Quine 1960: 219). In order to apply the projection criterion, an interpreter needs to consider what beliefs she would have if she were in the speaker’s epistemic situation. To perform this exercise, the interpreter needs to decide which aspects of her actual cognitive make up she should include in the hypothetical scenario in which she finds herself in the speaker’s epistemic situation. It seems likely that in many cases the question won’t have a determinate answer, and any indeterminacy at this stage will result in indeterminacy concerning the extent to which different interpretations satisfy the projection criterion. Once again, this would be a disaster if we were in the business of providing a representationalist account of the meaning grounds of interpretations. For the pragmatist approach, by contrast, this outcome is not at all problematic. In order to determine whether someone counts as interpreting, we need to consider whether she is selecting interpretations on the basis of the beliefs she would have in the speaker’s epistemic situation. So long as she does this, she will count as applying the projection criterion, independently of which aspects of her actual cognitive make-up she regards as included in the hypothetical scenario. This will affect which interpretations she endorses, but not whether she counts as

²⁴ This conception of interpretation can make room for cases in which two or more interpretations of a linguistic expression are equally correct, but it doesn’t need to treat in this way all cases in which our interpretative procedures support different interpretations when applied by different interpreters.

150

 

interpreting, and the pragmatist’s answer to the latter question does not depend on an answer to the former.

7.12 Truth Ascriptions A pragmatist account of the meaning grounds of truth ascriptions needs to specify features of the procedure that regulates our ascription of the predicate ‘is true’ from which these ascriptions obtain their meaning.²⁵ For sentences of the speaker’s own language, the feature of our procedure for ascribing the truth predicate that stands out as potentially meaning grounding results from our acceptance of the disquotational schema. It is the fact that I ascribe the predicate ‘is true’ to sentences I accept, and reject ascriptions of the predicate to sentences I reject. My proposal is to include this ascription procedure in the meaning ground of the predicate: ‘is true’ has the meaning it has by virtue of the fact that we regulate its ascription to a sentence S on the basis of whether we accept S. Notice that this ascription procedure for the predicate differs from the ascription procedure that the disquotational schema would yield on Crispin Wright’s construal of the principle. According to Wright, it follows from the disquotational schema that ‘reason to regard a sentence as T is reason to regard it as warrantedly assertible, and conversely’ (Wright 1992: 18). It is perfectly possible, even if, perhaps, reprehensible, to accept a sentence when we don’t think we have justification for the belief the sentence expresses, and, vice versa, not to accept a sentence for which we take ourselves to have justification that reaches the threshold for rational acceptance. In these cases, the meaning-grounding procedure I’m proposing would dictate ascriptions of ‘is true’ that are at odds with our views on the epistemic status of the beliefs expressed by sentences. Warrant and other epistemic notions play no role in my characterization of the meaning grounds of truth ascriptions.²⁶ The ascription procedure generated by the disquotational schema is only applicable to sentences of the speaker’s own language, but I want to propose that the meaning ground of ‘is true’ should also include an ascription procedure for other sentences. We now have the tools for describing the procedure to which I propose to assign this role—a procedure I employ for regulating my ascription of ‘is true’ to a sentence A as understood by a speaker S. The procedure is based on the notion of interpretation: I ascribe ‘is true’ to A, as understood by S, if I interpret A, as understood by S, as representing an obtaining state of affairs. And I reject ascriptions of ‘is true’ to A, as understood by S, if I interpret A, as understood by S, as representing a non-obtaining state of affairs. I propose to treat ²⁵ I discuss the pragmatist approach to truth ascriptions in (Zalabardo 2016). ²⁶ On this point, see section 5.2, above.

  

151

this feature of the procedure that regulates our ascriptions of the truth predicate to the sentences of others as part of their meaning grounds. This is in line with the role that interpretation plays, according to Davidson, in the concept of truth (Davidson 1990: 295–6). John Collins offers an insightful summary of Davidson’s line of reasoning on this point: Where truth is characterised for our own language, then the determination of truth conditions is, as it were, built into our very understanding of the language, i.e., the translation of object language into metalanguage realised in instances of (T) is ‘merely syntactical’ (the identity function: the object language is mapped into itself ). However, in general, the requirement of translation cannot be syntactical, where, e.g., we characterise truth for a language previously unknown to us. In such cases, interpretation—anthropology—is required. Thus, since we want a general concept of truth with empirical application to actual speakers, interpretation is a necessary part of the concept’s characterisation. (Collins 2002: 520)

The pragmatist’s proposal then is that the sentences in which we ascribe the predicate ‘is true’ have the meaning they have by virtue of the fact that our acceptance of these sentences is regulated by this procedure. As a result, someone will mean by a predicate what we mean by ‘is true’ just in case their acceptance of sentences in which the predicate is ascribed is regulated in this way. Notice that, as in previous applications of the pragmatist template, specifying the meaning grounds of the target sentences doesn’t require identifying the states of affairs that these sentences represent as obtaining. The position doesn’t involve endorsing any substantive theory of truth, not does it involve accepting the traditional deflationist’s negative claim, that the sentence ‘ “snow is white” is true’ doesn’t represent how things stand with the sentence ‘snow is white’, but only how things stand with snow. On this proposal, ascribing the truth predicate to a sentence represents things as being a certain way with the sentence, even though the meaning grounds of truth ascriptions are not specified by identifying the states of affairs they represent as obtaining.²⁷

7.13 Representational Discourse One of the central claims of this book is that ascribing to a family of sentences pragmatist meaning grounds is perfectly compatible with treating them as successfully discharging the task of representing the world. Over the last two chapters, ²⁷ On this point, the pragmatist account of the truth predicate is wholly in line with Paul Horwich’s version of deflationism (Horwich 1998a: 39–40).

152

 

I have outlined pragmatist accounts of the meaning grounds of ascriptions of propositional attitudes, meanings, and truth values. I maintain that endorsing these meaning grounds is perfectly compatible with treating these sentences as representing the world. This claim faces substantial obstacles. Addressing these problems will be my goal in the next chapter. In this section I want to concentrate on a different point. In claiming that pragmatist meaning grounds are compatible with representation, I am not claiming that the former entails the latter—that all sentences with pragmatist meaning grounds have the function of representing the world, let alone succeed in discharging this function. I’m claiming that some discourses with pragmatist meaning grounds represent the world, but there are also other discourses with pragmatist meaning grounds that are not in the business of representing how things stand, including discourses whose sentences have declarative form. My goal in this section is to outline an account of this contrast—between sentences that represent the world and those that don’t. For the representationalist, the contrast doesn’t pose serious problems: a sentence represents the world just in case there’s a possible state of affairs to which the sentence is related in such a way that the latter represents the former as obtaining. So long as the sentence-state of affairs relation that plays this role can be identified, the representationalist will have an account of the contrast between representational and non-representational discourse. The pragmatist can’t take this line. As we will see in the next chapter, the pragmatist will be able to say of representational sentences with pragmatist meaning grounds that there are states of affairs they represent as obtaining. However, on the pragmatist position, the existence of a state of affairs playing this role for a sentence will be construed as a consequence of the fact that the sentence represents the world, and can’t be invoked, without circularity, to explain what makes the sentence representational. A more congenial approach for the pragmatist would be to explain the contrast in terms of features of the practices in which discourses are embedded. When we introduced the notion of acceptance in section 5.4, the phenomenon was restricted to sentences that speakers regard as representational. Now it will be useful to remove this restriction and consider a wider notion, including alongside representational sentences other declarative sentences that the speaker doesn’t think of as playing this role. This wider notion will include among its instances both the attitude towards the sentence ‘the cat is on the mat’ of an English speaker who is convinced that the cat is on the mat and the attitude towards the sentence ‘night sailing is fun’ of a speaker who is convinced that night sailing is fun, even if she thinks of the sentence as no more than an expression of her attitude towards night sailing. The pragmatist will draw the contrast between representational and nonrepresentational discourses in terms of features of the practice of accepting and rejecting their sentences, in this wider sense.

  

153

The feature that I want to focus on is the existence of a standard of correctness for sentence acceptance and rejection that’s absolute in the following sense: if I accept a sentence now, then I regard as correct its acceptance at any time by any speaker who at that time means by the sentence what I mean by it now, and I regard as incorrect its rejection at any time by any speaker who at that time means by the sentence what I mean by it now. And if I reject a sentence now, then I regard as correct its rejection at any time by any speaker who at that time means by the sentence what I mean by it now, and I regard as incorrect its acceptance at any time by any speaker who at that time means by the sentence what I mean by it now. Our verdicts on a sentence with respect to this absolute standard of correctness are what we express by ascribing the predicates ‘is true’ and ‘is false’. This absolute standard of correctness is different from the speaker-relative notion of correctness that applies to every discourse with pragmatist meaning grounds. If the sentences of a discourse have pragmatist meaning grounds, there is an acceptance procedure whose employment makes the sentence have the meaning it has. A speaker will mean by a sentence what we mean by it just in case she employs this procedure for regulating her acceptance of the sentence. This generates a clear sense in which a speaker is right in accepting a sentence according to its meaning-grounding acceptance procedure. Doing so is required, on the pragmatist account, for meaning by the sentence what we mean by it. But this standard of correctness is not absolute. Typically, the meaning-grounding acceptance procedure will produce different verdicts for different speakers, and for a given speaker at different times. Thus, for example, in the case of ‘killing one to save five is morally right’, some speakers will feel moral approval towards the action while other speakers won’t, and a speaker who used to feel moral approval towards the action might at some point cease to feel it, or vice versa. Hence, in this sense, it will be right for some speakers to accept the sentence and wrong for other speakers to do so. And it will be right for a speaker to accept the sentence at one time and wrong to do so at a different time. Notice that a speaker’s verdicts on whether it is correct or incorrect in this absolute sense to accept a sentence are necessarily linked to her own acceptance/ rejection of the sentence. Her verdicts are produced by ascribing absolute validity to her acceptances and rejections. The rules governing this kind of assessment make no room for accepting a sentence and treating its acceptance as incorrect.²⁸ We think that this absolute standard of correctness governs acceptance and rejection in some declarative discourses with pragmatist meaning grounds but has no application in other declarative discourses that can also be construed as having pragmatist meaning grounds. Declarative sentences that express what we regard as purely subjective preferences don’t have this feature. The sentence ‘night sailing ²⁸ This feature of the rules governing the absolute standard of correctness for acceptance and rejection is responsible, I think, for the phenomenon known as Moore’s paradox. See (Moore 1993).

154

 

is fun’ can be seen as having a pragmatist meaning ground, consisting in the fact that its acceptance is regulated by whether the speaker enjoys night sailing. This generates a speaker-relative notion of correct assertion—it is correct for a speaker to accept the sentence just in case she enjoys night sailing. However, we don’t recognize a second, absolute notion of correctness, common to all speakers at all times. If I enjoy night sailing, I will accept the sentence, but if you don’t enjoy night sailing there isn’t any sense in which it is wrong for you not to accept the sentence, or in which I was wrong in not accepting it a few years back, when I didn’t enjoy night sailing. The discourses on which we’ve focused, by contrast, do sustain this absolute standard of correctness. This feature is present in moral discourse on what I regard as the most natural understanding of how it operates. If I feel moral approval towards the killing of one to save five and consequently accept the sentence ‘killing one to save five is morally right’, there’s a clear sense in which I will treat its acceptance as correct at all times, and its rejection as incorrect at all times, for any speaker who at that time means by the sentence what I mean by it now. The semantic discourses we are focusing on also pass the test. The point is clear in the case of truth ascriptions. If you and I understand the sentence ‘the cat is on the mat’ in the same way, and I believe that the cat is on the mat but you don’t, then, in the speaker-relative sense of correct acceptance, not accepting ‘ “the cat is on the mat” is true’ is the right thing for you to do, but there is in addition an absolute notion of correctness attaching to truth ascriptions, under which I will regard your failure to accept ‘ “the cat is on the mat” is true’ as incorrect. Similarly, if Radical Interpretation recommends to me interpreting ‘Tisch’, as understood by Kurt, as meaning table, but it recommends to you not interpreting it in this way, it seems perfectly legitimate for me to say that in the absolute sense of correct acceptance it is wrong for you not to accept ‘ “Tisch”, as understood by Kurt, means table’. And if the meaning-grounding procedure for ascribing propositional-attitude predicates recommends to me ascribing to Peter the belief that that’s a horse but it recommends to you not to ascribe this belief to him, I can intelligibly say that it is wrong of you not to accept ‘Peter believes that’s a horse’. I want to propose that the pragmatist should construe the contrast between representational and non-representational discourse in terms of whether a discourse sustains this absolute standard of correctness. However, we need to be careful with exactly how this is done. One possibility would be to say that a discourse is representational just in case speakers take their acceptance and rejection of its sentences as subject to the absolute standard of correctness. But this position has serious disadvantages. Even if we assume that speakers are in agreement on this point, we think that the right account of the notion of representational discourse should make room for the possibility that speakers are wrong in their verdicts on which discourses are representational—that some of

  

155

their sentences fail to represent the world even though they mistakenly treat them as subject to an absolute standard of correctness, or vice versa. We could try to avoid this pitfall by making the representational character of a discourse depend, not on whether speakers take its sentences to be subject to an absolute standard of correctness, but on whether its sentences are as a matter of fact subject to an absolute standard of correctness. The problem with this proposal is that its explanans doesn’t seem much more tractable than its explanandum. The idea of treating acceptance of a sentence as subject to an absolute standard of correctness is perfectly intelligible, but what this proposal invokes is the idea of sentence acceptance actually being subject to an absolute standard of correctness, and it’s not clear how the pragmatist could construe this notion. One possibility would be to say that the absolute standard of correctness for the acceptance of a sentence is provided by the obtaining of the state of affairs that the sentence represents as obtaining. However, as I mentioned above, the pragmatist will not want to take things in this direction, as she will want to say that the existence of a state of affairs that a sentence represents as obtaining is a consequence of the sentence being representational, and not the other way round. I want to propose a different strategy for construing the notion of representational discourse in terms of speakers treating sentences as subject to an absolute standard of correctness. The proposal can be characterized as another application of the pragmatist approach, this time to the predicate ‘is representational’. On this proposal, the meaning ground of the predicate is not specified along the lines of the representationalist template. We don’t try to specify the state of affairs we represent as obtaining with ascriptions of the predicate to sentences—either in terms of whether speakers treat the sentences as subject to an absolute standard of correctness, or in terms of whether they are actually subject to an absolute standard of correctness, or in terms of other factors. I propose instead to provide a pragmatist specification of the meaning ground of ‘is representational’. On this approach, the predicate has the meaning it has by virtue of the fact that its ascription is regulated by whether you take the target sentence to be subject to an absolute standard of correctness. Regulating ascription of the predicate in this way is necessary and sufficient for meaning by the predicate what we mean by it. This approach immediately generates a speaker-relative notion of correctness for our acceptance or rejection of sentences of the form: R

‘Sentence S is representational’.

It will be correct for you to accept R just in case you take acceptance of S to be subject to an absolute standard of correctness, because regulating your acceptance of R in this way is required for meaning by R what we mean by it. However, this standard of correctness is not absolute. If two speakers disagree on whether acceptance of S is subject to an absolute standard of correctness, it will be right

156

 

for one to accept R and wrong for the other to do so. It’s not part of the view that a speaker taking acceptance of S to be subject to an absolute standard of correctness makes S, as understood by her, representational, or that if all or most speakers take acceptance of S to be subject to an absolute standard of correctness, that makes S, as understood by them, representational. As in other applications of the pragmatist template, this account of the meaning grounds of R doesn’t include a specification of the state of affairs that R represents as obtaining—of what has to be the case in order for things to be as R represents them as being. This treatment of the representational character of sentences will apply, in particular, to the representational character of sentences ascribing representational character, as: R*

‘ “Sentence S is representational” is representational’.

Once again, the pragmatist account won’t produce a specification of the state of affairs that this sentence represents as obtaining—what has to be the case in order for things to be as R* represents them as being. It will simply specify how your acceptance of R* needs to be regulated in order for the sentence to have the meaning it has—to ascribe representational character to R: you should accept it just in case you take acceptance of R to be subject to an absolute standard of correctness. According to the meaning-grounding ascription procedure for ‘is representational’, if you take R not to be subject to an absolute standard of correctness, you should reject R*. The resulting situation is perplexing, particularly when combined with taking S to be subject to an absolute standard of correctness and consequently accepting R. Rejecting R* appears to undermine acceptance of R. One would maintain that S is representational, but that this doesn’t amount to representing how things stand with S. There have been attempts to show that this kind of position is incoherent, but it’s not clear that the arguments on offer apply to the specific construal of discourse about representational character that I’ve presented here.²⁹ I see no reason to think that accepting R and rejecting R* involves a contradiction. On this point, I agree with the view defended by Robert Kraut: such claims as ‘S has truth conditions’ does not have truth conditions and ‘S is descriptive’ is not descriptive do not—pace recent arguments to the contrary—force the abandonment of the descriptive/expressive distinction or place it at risk. (Kraut 1993: 251)

²⁹ See (Boghossian 1990).

  

157

If we treat acceptance of the sentences of a discourse as subject to an absolute standard of correctness, we might want to include this fact in their meaning grounds. Consider, for example, the predicate ‘is morally right’. We regulate ascriptions of ‘is morally right’ by our feelings of moral approval. On the proposal we’ve considered so far, what makes the predicate have the meaning it has is the fact that its ascription is regulated in this way. Any predicate whose ascription is regulated in this way will have the same meaning as ‘is morally right’, as understood by us. But it is also a fact about our practice of ascribing ‘is morally right’ that we take these ascriptions to be subject to an absolute standard of correctness. On the current proposal, this is not part of the meaning ground of the predicate— someone who regulates her ascription of a predicate by her feelings of moral approval but doesn’t take these ascriptions to be subject to an absolute standard of correctness will still mean by her predicate what we mean by ‘is morally right’. What I’m proposing now is to change this—to say that the fact that we treat our ascriptions of the predicate as subject to an absolute standard of correctness is part of its meaning ground, and hence that someone who regulates her ascription of a predicate by her feelings of moral approval but doesn’t take these ascriptions to be subject to an absolute standard of correctness will not mean by her predicate what we mean by ‘is morally right’. I think this is a plausible move. The fact that we take our ascriptions of ‘is morally right’ as subject to an absolute standard of correctness is what makes it the case that ‘is morally right’ and ‘triggers moral approval’ have different meanings. The proposal also has an important dialectical advantage. I argued in Section 4.3 that once we have accepted that the sentences of a discourse have a nonrepresentational function as part of their meaning grounds, we can’t do full justice to the thought that these sentences perform the function of representing the world in the same way in which sentences with representationalist meaning grounds perform this function. The pragmatist position doesn’t face this problem, since pragmatist meaning grounds don’t include a non-representational function. However, one could complain that the problem hasn’t been fully solved. Pragmatist meaning grounds may not include a non-representational function, but they don’t include either the function of representing the world. Hence there’s still a contrast with representationalist meaning grounds, which have the representational function built into them. If having a non-representational function as part of its meaning ground is an obstacle to the representational status of a sentence, the thought goes, then not having the representational function as part of its meaning ground should be seen as having the same consequence. I’m suggesting that this objection can be rejected if we maintain that the meaning-grounding acceptance procedure for the target discourse includes the fact that we treat acceptance as subject to an absolute standard of correctness. Since treating acceptance as subject to an absolute standard of correctness is the meaning-grounding ascription procedure for ‘is representational’, this amounts to

158

 

building the representational character of the discourse into the meaning grounds of its sentences, thereby bringing the discourse in line with discourses with representationalist meaning grounds.

7.14 Wright’s Discipline: Cognitive Command In his book Truth and Objectivity (Wright 1992), Crispin Wright provided an illuminating development of the idea that the notion of truth aptitude should be construed in terms of features of the practices in which discourses are embedded—of what he calls the discipline of discourses (Wright 1992: 10). In this respect, the proposal sketched in the preceding section concerning representational character is closely related to Wright’s approach to truth aptitude. However, Wright proposes some conditions for truth aptitude that I don’t think should be included in the meaning-grounding ascription procedures for the predicate ‘is representational’. One such condition is what Wright calls cognitive command: It is a priori that differences of opinion formulated within the discourse, unless excusable as a result of vagueness in a disputed statement, or in the standards of acceptability, or variation in personal evidence thresholds, so to speak, will involve something which may properly be regarded as a cognitive shortcoming. (Wright 1992: 144)

As Max Kölbel has pointed out (Kölbel 2002: 22–8), there’s one reading of cognitive command under which the condition is satisfied by every discourse in which sentence acceptance is subject to an absolute standard of correctness. If I accept sentence S and take this acceptance to be subject to an absolute standard, then I will take everyone who fails to accept S, if they mean by S what I mean by it, as exhibiting a cognitive shortcoming—namely their failure to accept a sentence they should accept. On this reading, adding cognitive command to the meaning-grounding ascription procedures for ‘is representational’ would be a redundant move. There is, however, a different reading of the condition under which it is stronger than the existence of an absolute standard for acceptance of the sentences of a discourse. Under this reading, the cognitive shortcoming that has to be present in one party to every disagreement doesn’t include acceptance or rejection of the sentence in question. It has to concern the functioning of the cognitive mechanisms leading to acceptance/rejection—something has to be wrong with them over and above the fact that they produce the wrong verdict in this case.³⁰

³⁰ This is how Wright intends the condition to be understood. See (Wright 1992: 148–57).

  

159

When understood in this way, a discourse that is subject to an absolute standard of correctness for acceptance of its sentences could fail to exhibit cognitive command. I think it’s perfectly natural to treat moral discourse as satisfying this description. If I accept ‘killing one to save five is morally right’, I take acceptance of this sentence to be correct, and its rejection incorrect, by any speaker who means by the sentence what I mean by it. Nevertheless, I’m not in a position to rule out a priori that the sentence is rejected by another speaker as a result of the operation of cognitive mechanisms that are not defective in any way—over and above the fact that they produce the wrong result. If this is right, then including cognitive command in the meaning-grounding procedures for ‘is representational’ would bestow on the predicate a narrower meaning than I intend here.³¹ Someone who attaches to the predicate the meaning I’m trying to grasp in terms of ascription procedures should be prepared to ascribe the predicate to a sentence whenever they take its acceptance to be subject to an absolute standard of correctness, independently of whether they take the sentence to satisfy the cognitive-command condition. In later work, Wright has argued that the idea of an absolute standard of correctness that doesn’t satisfy cognitive command, in the strong sense, involves an incoherence. The problem is this. Suppose that acceptance of ‘killing one to save five is morally right’ doesn’t satisfy cognitive command. This circumstance opens the possibility of faultless disagreement concerning the sentence—cases in which a speaker accepts the sentence and another speaker rejects it even though neither of these attitudes involves any cognitive shortcoming. Suppose you accept it. If you take the discourse to be subject to an absolute standard of correctness, you will regard those who reject it as wrong relative to this standard—even those whose rejection involves no cognitive shortcoming. But according to Wright, this attitude cannot be maintained with intellectual integrity: ‘Apparently one of us has to be mistaken. But if one of us is mistaken, how can we tell who? Isn’t it just a conceit to think it has to be the other?’ (Wright 2006: 41).³² Now, it might be that the realization that an epistemic peer whose rejection of the sentence involves no cognitive shortcoming weakens your sense of moral approval towards the killing of one to save five, to the point that following the meaning-grounding acceptance procedure for the sentence dictates that you should now withdraw your acceptance. But it’s also possible that the discovery doesn’t have this effect, and your sense of moral approval towards the action ³¹ There may well be an important contrast between discourses that is tracked by the presence of cognitive command. I’m only claiming that cognitive command is not necessary for a discourse to perform the function of representing the world. ³² See also (Wright 2001). See (Davis 2015) for an insightful discussion of Wright’s argument. The view I’m defending here is close to the view that Davis calls divergentism. Standard strategies for dissolving the tension highlighted by Wright include relativist and contextualist treatments of the relevant discourses. For relativist treatments of other discourses, see (MacFarlane 2007; Egan 2006). For contextualism, see (Glanzberg 2007).

160

 

remains sufficiently strong to support acceptance of the sentence, according to its meaning-grounding acceptance procedure. According to Wright, adopting this attitude would be conceited, and intellectual integrity would force you to take one of two routes. The first option would be to rescind your acceptance of the sentence, in spite of your continued sense of moral approval towards the killing of one to save five. The second would be to abandon the idea that an absolute standard of correctness applies to acceptance of the sentence, over and above the subject-relative standard built into the meaning-grounding acceptance procedure, according to which your acceptance and your peer’s rejection are equally right. I want to argue that there is no reason for you to take either of these routes. Starting with the second, I’d like to call into question the assumption that if an absolute standard of correctness is in force in a discourse then faultless disagreement with respect to its sentences is impossible. We think that a discourse can be subject to an absolute standard of correctness even if faultless application of the acceptance procedures for its sentences doesn’t guarantee that we will only accept sentences that satisfy the absolute standard. We certainly don’t think that a question cannot count as factual unless we have a procedure for settling it which, when faultlessly applied, can’t fail to provide the right answer. And if faultless application of the acceptance procedure for a sentence doesn’t guarantee success, then faultless disagreement will always be a possibility. The first, route is also not mandatory. On the contrary, deciding on the acceptance of the sentence in line with your sense of moral approval towards the action is required for you to mean by the sentence what we mean by it. If the discovery of a faultlessly disagreeing peer doesn’t remove your sense of moral approval, then you wouldn’t be meaning by the sentence what we mean by it unless you accepted it, and if you take the discourse to be subject to an absolute standard, you should also regard the verdict of your faultlessly disagreeing peer as incorrect. Of course, you should be open to the possibility that you are wrong and she is right, but recognizing this possibility doesn’t prevent you from taking sides, by letting your verdict be dictated by your own sense of moral approval.

7.15 Ideal Conditions Another condition that I think should be left out of the meaning-grounding ascription procedures for ‘is representational’ is the existence of ‘ideal’ conditions under which application of the meaning-grounding acceptance procedure for the target sentence will be guaranteed to produce the right verdict under the absolute standard. As we saw in Section 4.1, this is a move that might be attractive to someone who wants to define the states of affairs represented by the target sentences in terms of the procedures employed for regulating their acceptance. But the pragmatist has no use for this move, as her proposed meaning grounds do

  

161

not require identifying the states of affairs represented by the sentences. Nevertheless, one might still maintain that unless we take these ideal conditions to exist, we won’t be ascribing to the sentence the task of representing the world. I see no reason for accepting this. I think a sentence can successfully discharge the task of representing the world even if, for any non-circularly specified set of conditions, it is a contingent matter whether the sentence’s acceptance procedure will produce the right verdict under those conditions. Likewise, I don’t think that in order for a sentence to represent the world, there have to be conditions under which the sentence’s acceptance procedure will produce a verdict that would not change upon further investigation. Wright has argued that in the circumstances that would enable the sentences of a discourse to have pragmatist meaning grounds, this condition is bound to be satisfied. He writes: Provided, as I have argued is presupposed by the very idea of assertoric content, there are generally acknowledged standards of proper and improper assertion within the discourse, there must be sense to be attached to the idea of a statement which under certain circumstances meets the standards of proper assertion, and then will or would continue to do so unless the considerations which led to its downfall were open to objection in some way. (Wright 1992: 48)

Pragmatist meaning grounds presuppose the existence of generally acknowledged standards of proper and improper acceptance—those that are given by the meaning-grounding acceptance procedures. However, in the cases we have discussed, there is every reason to expect that the relevant acceptance procedures will not generate the notion of guaranteed stable acceptance envisaged by Wright. There are no circumstances guaranteeing that either you and I would both feel moral approval towards the killing of one to save five or we would both feel moral disapproval towards the action. Likewise, there are no circumstances guaranteeing that my moral approval or disapproval would subsequently disappear only in ways that are ‘open to objection in some way’. However, on my view, this fact doesn’t in any way undermine the claim that the discourse has the function of representing the world or that it succeeds in discharging this function. The existence of the conditions envisaged by Wright should not be included in the meaning-grounding ascription procedures for the predicate ‘is representational’.

7.16 Conclusion In Chapter 5, I suggested that a pragmatist account of the meaning grounds of interpretations could be based on the characterization of our interpretative procedure provided by the Quine-Davidson notion of Radical Interpretation

162

 

(Translation). In the present chapter I have developed and modified this proposal. I have argued that the procedure that regulates our acceptance of compositional interpretations of compositional languages employs a doxastic criterion—we select interpretations on the basis of the belief ascriptions they produce as a result of the link between meaning and belief generated by the notion of sentence acceptance. I have argued that the charity criterion doesn’t provide an accurate characterization of our interpretative procedure and I have proposed to replace it with a combination of projection and familiarity, supplemented by belief ascriptions based on non-linguistic evidence. Then I’ve turned, more briefly, to truth ascriptions. I have presented a pragmatist account of their meaning grounds in terms of the connection between accepting one of your sentences and ascribing the truth predicate to it and, for the sentences of others, in terms of the connection between ascribing the truth predicate and interpreting the sentence as representing an obtaining state of affairs. Finally, I have considered the contrast between representational and nonrepresentational discourse. I have proposed to construe the contrast with another application of the pragmatist template—this time to sentences of the form ‘Sentence S is representational’. I have proposed to see the meaning of these sentences as grounded in an acceptance procedure focused on the existence of an absolute standard of correctness for acceptance of S. Chapters 6 and 7 have been devoted to developing pragmatist accounts of the meaning grounds of semantic discourses. In the next chapter I move from specific discourses to more general questions concerning the pragmatist approach. My target will be the assumption (RR) that a discourse with pragmatist meaning grounds cannot successfully perform the function of representing the world. I will consider what I regard as the main reason for endorsing this assumption and present a strategy for defusing this line of support.

8 Harmony and Abstraction Over the last three chapters I have developed an approach to semantic discourses according to which their sentences have pragmatist meaning grounds and they perform the function of representing the world in the same sense as sentences with representationalist meaning grounds do. Clearly, the proposal will be viable only if the two features it ascribes to the problematic discourses—non-representationalist meaning grounds and a representational function—are compatible with one another. In Section 4.3 I presented a line of reasoning according to which the deflationist account of truth ascriptions enables us to treat any discourse to whose sentences the truth predicate can be meaningfully ascribed as performing the function of representing the world. I then argued that the claim to representational status that a discourse would acquire in this way would be undermined by the inclusion in its meaning grounds of a non-representational function. The pragmatist accounts of the meaning grounds of the problematic discourses that I am defending don’t suffer from this problem. Their meaning grounds, on this approach, are given by the procedures employed for regulating the acceptance of their sentences. They don’t include ascribing to them a non-representational function. This obstacle to a genuine representational role doesn’t afflict every non-representationalist account of meaning grounds, but only those which, unlike my proposal, ascribe to the target sentences a non-representational function. There is, however, another line of reasoning that would show, if successful, that every non-representationalist account of meaning grounds, including mine, is incompatible with the claim that the target sentences perform the function of representing the world. My goal in the present chapter is to describe this problem and to propose a solution.

8.1 The Problem of Harmony The problem I want to focus on arises from the following thought: in order for a declarative sentence to succeed in representing things as being a certain way, there has to be a possible state of affairs that the sentence represents as obtaining. Notice that invoking this intuition doesn’t beg the question in favour of RR. The thought is simply that there has to be a state of affairs that the sentence represents as obtaining, not that its relation to this state of affairs has to be included in

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0008

164

 

the meaning ground of the sentence. This is in principle compatible with a representational sentence having a non-representationalist meaning ground, so long as this doesn’t preclude the existence of a state of affairs that the sentence represents as obtaining. However, there are reasons for thinking that this isn’t a possibility—that if the meaning ground of a sentence is non-representational, there can’t be a state of affairs that the sentence represents as obtaining. If this state of affairs exists, the relationship between the sentence and the state of affairs that plays this role will be a necessary condition for the sentence to have the meaning it has. If the role were occupied by a different state of affairs (or by no state of affairs), the sentence wouldn’t have the meaning it has.¹ This circumstance gives rise to a potential conflict between the existence of a state of affairs that the sentence represents as obtaining and ascribing to the sentence a non-representationalist (e.g. pragmatist) meaning ground. The conflict would come from the claim that the obtaining of the pragmatist meaning ground is a sufficient condition for the sentence to have the meaning it has. This claim is incompatible with the thought that the sentence’s relationship with the state of affairs it represents as obtaining is a necessary condition for the sentence to have the meaning it has, unless the pragmatist meaning ground is a sufficient condition for the sentence to represent the state of affairs it represents—unless, that is, it is impossible for the sentence to fail to represent that state of affairs so long as its pragmatist meaning ground obtains. It follows that if a sentence with a pragmatist meaning ground performs the function of representing the world, the obtaining of its meaning ground has to be sufficient for its relation to the state of affairs it represents as obtaining. I’m going to refer to this as the problem of harmony. How could this difficulty be overcome? What we need is that the fact that acceptance of the target sentence is regulated by its meaning-grounding acceptance procedure is a sufficient condition for the sentence to represent the state of affairs it represents. One way we might try to achieve this is to define the represented state of affairs in terms of the meaning-grounding acceptance procedures. However, this project faces the same difficulties that we raised in Section 4.1 for the project of using these definitions to provide representationalist meaning grounds for the problematic sentences. A second way in which we might try to achieve harmony is to propose that the state of affairs represented by a sentence should be singled out by its causal/ nomological links with the features of the use included in the pragmatist meaning ground. Take the case of ethical discourse as an example. Presumably, there are ¹ Once again, accepting this point doesn’t immediately enjoin a commitment to the assumption (RR) that only sentences with representationalist meaning grounds can perform the function of representing the world. Not every necessary condition for a fact has to be part of its ground, and specifically, not every necessary condition for a sentence to have the meaning it has has to be part of its meaning ground.

  

165

features of actions that bear causal/nomological relations to my feeling of moral approval. On the proposal under discussion, the property that we represent as instantiated by an action when we ascribe to it the predicate ‘is morally right’ would be singled out by these causal/nomological links. This approach faces difficulties that mirror the problems encountered by the definitional approach. Let R be the causal/nomological relation that a property has to bear to the feeling of moral approval in order to be the property that we represent as instantiated by an action when we ascribe ‘is morally right’ to it. The problem is that your feeling of moral approval might not be R-related to the same property as mine. Then, on this proposal, your ascriptions of ‘is morally right’ and mine would not represent the same states of affairs, and, a fortiori, would not have the same meaning, even if your acceptance of these sentences, as well as mine, is regulated by the feeling of moral approval.² A similar situation arises for trans-temporal or modal variation. Notice that this outcome doesn’t rule out the possibility that your acceptance and my rejection of ‘killing one to save five is morally right’ counts as a disagreement, since my feeling of moral approval might occasionally be triggered by an action that doesn’t instantiate the property to which it is R-related, or fail to be triggered by an action that instantiates this property. However, there would still be cases in which you and I regulate acceptance of the sentences in which we ascribe the predicate by our feeling of moral approval, but we are not disagreeing when you accept and I reject the ascription of the predicate to an action. I find this outcome implausible, but the main point for our immediate purposes is that it is incompatible with the pragmatist’s account of the meaning grounds of these sentences. The proposal offers no solution to the problem of harmony. As with the definitional strategy, one might try to solve the difficulty by treating the predicate as implicitly relational, referring to the relation that an action bears to a subject when the action instantiates the property that bears R to the subject’s feeling of moral approval. But once again harmony is not achieved in this way. Under this proposal, the sentence ‘killing one to save five is morally right’ represents different states of affairs as understood by you and as understood by me, and if your feeling of moral approval and mine bear R to different properties, one of these states of affairs might obtain while the other one fails to obtain. A third approach to the problem of harmony would be to maintain that the sentences from the problematic discourses have pragmatist meaning grounds and ² As we saw in Section 2.4, Terence Horgan and Mark Timmons, with their Moral Twin Earth thought experiments, develop a similar line of reasoning to undermine the view that moral terms refer to properties to which they bear a designated causal/nomological relation (Horgan and Timmons 1991). My goal here is more modest. I’m simply arguing that these views are incompatible with assigning pragmatist meaning grounds to the target terms, and hence that they offer no solution to the problem of harmony.

166

 

that each of them represents a state of affairs, while claiming that the state of affairs that plays this role in each case cannot be independently specified, but its relation with the sentence is necessitated by the employment of the meaninggrounding acceptance procedure. Thus, on this approach, for example, the state of affairs represented by the sentence ‘killing one to save five is morally right’ could not be specified without using this sentence, but the sentence could not fail to represent this state of affairs so long as its acceptance is regulated by the feeling of moral approval. This approach faces some important challenges. We are not given any reason to believe in the existence of the state of affairs represented by the sentence, independently of our conviction that the sentence represents the world. Likewise, no reason is given for thinking that the state of affairs, assuming it exists, couldn’t fail to be represented by the sentence so long as its acceptance is regulated by the feeling of moral approval towards the killing of one to save five. Rather than a solution to the problem of harmony, this approach offers no more than the bare assertion that things are as they would have to be for the problem to disappear. And even if we are prepared to accept these claims, the proposal would have to be supplemented with a plausible story about the epistemic status of speakers’ reliance on the procedures they employ for regulating their acceptance of the sentences of the discourse. I regard the problem of harmony as the main remaining obstacle to the treatment of the problematic discourses I’m defending. Ascribing to them pragmatist meaning grounds will be compatible with treating them as representing the world only if we can cogently maintain that there is a state of affairs that each of the target sentences represents as obtaining and that the connection between the sentence and the state of affairs is necessitated by the employment of the meaninggrounding acceptance procedure. My goal in the present chapter is to develop a strategy for achieving this.

8.2 An Idea of Quine’s The problem of harmony I’ve just presented with respect to the meaning grounds of sentences can also be formulated in terms of the meaning grounds of predicates, and this is the form in which I propose to discuss it. If a predicate is to be capable of representing the world, there has to be a property—the predicate’s referent—that we represent an object as instantiating when we ascribe the predicate to it. And if the meaning ground of the predicate is pragmatist, then the meaning-grounding ascription procedure for the predicate has to be sufficient for the predicate to refer to the property it refers to. What we need is a way of characterizing a predicate’s referent that satisfies this requirement.

  

167

The strategy I have in mind for defining the referent of a predicate can be introduced in terms of the account of the meaning of an expression that Quine outlines in the following passage: Just what the meaning of an expression is—what kind of object—is not yet clear; but it is clear that, given a notion of meaning, we can explain the notion of synonymity easily as the relation between expressions that have the same meaning. Conversely also, given the relation of synonymity, it would be easy to derive the notion of meaning in the following way: the meaning of an expression is the class of all the expressions synonymous with it. No doubt this second direction of construction is the more promising one. (Quine 1943: 120)³

I want to focus on Quine’s favoured option—to define the meaning of an expression as the class of all the expressions that are synonymous with it. My goal will be to explore an account of predicate reference that follows the same general approach as the account of the meaning of an expression that Quine proposes in this passage. My proposal will differ in substantial respects from Quine’s version, but before I go into these differences, I want to display the appeal that the approach might hold for the pragmatist, by considering the consequences of treating the predicate ‘is morally right’ as referring to the equivalence class of predicates that are synonymous with it. Notice that if the referent of ‘is morally right’ were the class of predicates synonymous with it, then in order to identify the referent of ‘is morally right’ it would suffice to specify a condition whose satisfaction by a predicate X is necessary and sufficient for X to be synonymous with ‘is morally right’, as ‘is morally right’ would refer to the class of predicates satisfying this condition. We saw in Chapter 1 that the meaning ground of a linguistic expression generates synonymy conditions for the expression. This suggests that if ‘is morally right’ refers to the equivalence class of predicates synonymous with it, then a nonrepresentationalist specification of its meaning ground might also be treated as an identification of its referent. This goes, in particular, for pragmatist meaning grounds. If ‘is morally right’ has a pragmatist meaning ground, the synonymy conditions that this generates will count as a non-circular identification of the equivalence class that ‘is morally right’ refers to. And, crucially, the resulting account will not be vulnerable to the problem of harmony. Now the obtaining of the pragmatist meaning ground of ‘is morally right’ is a sufficient condition for ‘is morally right’ to have the referent it has. Hence so long as the meaning ground of ‘is morally right’ obtains, the predicate can’t fail to have the referent it has. If ‘is morally right’ refers to the equivalence class of predicates that are synonymous

³ Similar views are defended in (Ayer 1936: 88) and (Carnap 1956: 152).

168

 

with it, a pragmatist account of its meaning ground is not vulnerable to the problem of harmony. I’m not going to propose that predicates refer to equivalence classes of synonymous predicates, but the proposal I’m going to articulate will hold the same promise for the pragmatist with respect to the problem of harmony.

8.3 Abstraction What Quine is proposing for meaning is a definition by abstraction. The method of definition by abstraction was widely used in mathematics since the end of the nineteenth century. It’s a method for defining a function from a domain of objects taken as given. The sentences in which definitions by abstraction are formulated are known as abstraction principles. If D is a given domain of objects, and f is a function with D as its domain (the abstraction operator), an abstraction principle has the form: A

For all x, y in D, f(x) = f(y) iff xRy

Notice that an instance of A is logically false unless R is an equivalence relation.⁴ In effect, an abstraction principle specifies necessary and sufficient conditions for two elements of D to have the same image under f. Definitions by abstraction were discussed by Frege in the Grundlagen (Frege 1980a). In one of his famous examples, D is the set of lines and f is the function pairing each line with its direction. The proposed abstraction principle can be formulated as follows (Frege 1980a: 74–5): For all lines x, y, the direction of x = the direction of y iff x is parallel to y. In the application that Frege focuses on, D is the set of concepts and f is the function pairing a concept F with the number of Fs. The proposed abstraction principle here is based on Hume’s principle (Frege 1980a: 73–4): For all concepts F, G, the number of Fs = the number of Gs iff there is a bijection between the extensions of F and G. If we take a function to be defined by an abstraction principle, one question that arises immediately is what the values of the function are. An abstraction principle tells us when two elements of the domain of the abstraction operator receive the ⁴ An instance of A is logically false also if R is inflationary. See (Fine 2002: 4). This is an important fact in other contexts, but it won’t play a role in our discussion.

  

169

same image, but it doesn’t seem to tell us anything else about the identity of the objects playing the role of images under the function—about the identity of directions or of numbers. Paolo Mancosu has introduced a useful distinction between two options one might take regarding the range of an abstraction operator. The contrast concerns whether its elements can receive an explicit definition independently of the abstraction principle. Mancosu refers to definitions by abstraction for which this question receives an affirmative answer as thin, and to those for which the question is answered in the negative as thick (Mancosu 2016: 24). As Mancosu points out, it is often unclear whether a definition by abstraction is meant to be understood as thin or thick, but there are some clear cases on both sides. Among those who proposed thin definitions by abstraction, there are two main traditions concerning the identity of the elements of the range of the abstraction operator. As Mancosu explains, in number theory the dominant approach was to treat as the image under f of an element a of D a specific representative of the equivalence class generated by R to which a belongs. However, in synthetic geometry and the theory of cardinals and ordinals, where it’s often not possible to pick a representative non-arbitrarily, the standard approach is to take the equivalence class itself as the image.⁵ The latter is what Quine is proposing for the function pairing each expression with its meaning—a thin definition by abstraction in which the image of an expression is the equivalence class generated by the synonymy relation to which the expression belongs. Other mathematicians clearly embraced thick definitions by abstraction. Thus, for example, Giuseppe Peano writes: Let u be an object; by abstraction one obtains a new object ϕu; one cannot form an equality ϕu = known expression since ϕu is an object whose nature is completely different from all those that have hitherto been considered. (Peano 1894: 45)⁶

And again, on the definition by abstraction of direction: The relation between two unbounded straight lines ‘a is parallel to b’ has the properties of equality. It has been transformed into ‘direction of a = direction of b’, or ‘point at infinity of a = point at infinity of b’. One cannot define an equality of the form: ‘point at infinity of a’ = ‘expression formed with the words of Euclid’s Elements’. (Peano 1894: 47)⁷

⁵ For the theory of cardinals and ordinals the approach was later abandoned, due to the discovery of its paradoxical consequences. ⁶ Cited in (Mancosu 2016). ⁷ Cited in (Mancosu 2016).

170

 

Notice that treating an abstraction principle as a thick definition by abstraction involves accepting that what the abstraction principle tells us about the identity of the values of the abstraction operator—i.e. which arguments they are paired with—can be regarded as a satisfactory specification of the identity of these items. The legitimacy of thick definitions by abstraction is hugely controversial. It is strongly rejected by Frege in the Grundlagen. And Russell’s verdict in The Principles of Mathematics is equally negative (Russell 1903: 113–16). The debate on their legitimacy regained momentum in philosophy in connection with the proposals advanced by neo-Fregean philosophers of mathematics.⁸ Here I don’t intend to take part in this debate. I propose instead to treat as an assumption that the debate has been settled in favour of the legitimacy of thick definitions by abstraction. I’m going to assume, specifically, that if an instance of A satisfies the conditions under which it can be used as a definition by abstraction of f in terms of R, then it can be legitimately used as a thick definition of f in terms of R, i.e. f will have been successfully defined even if the items in its range cannot be defined independently of A. I’m going to refer to this as the Peano assumption. Accepting thick definitions by abstraction enables us to formulate a variant of Quine’s proposal. On this position, the meaning of an expression will be defined by abstraction in terms of synonymy, as Quine proposes, but, contra Quine, it won’t be identified with an equivalence class of synonymous expressions. Meanings will be treated instead as sui generis entities whose identity is fully specified by the abstractionist link with synonymy. It follows from the Peano assumption that if Quine’s original proposal is legitimate, then the same goes for this thick version. My goal is to explore whether this general approach can be successfully applied to predicate reference. On this proposal, predicate referents would be introduced with a definition by abstraction of the function pairing predicates with their referents, but this definition would be treated as thick, thus avoiding the need to identify referents with equivalence classes of predicates, or with any other items that can be defined independently of the abstraction principle.

8.4 Two Readings of Abstraction Principles Accepting an abstraction principle as true doesn’t force one to accept the definition by abstraction that the principle could express. Take the following principle: N

For all persons x, y, x’s nationality = y’s nationality iff x and y are compatriots.

⁸ See (Wright 1983; Hale 1987; Hale and Wright 2001), and (MacBride 2003) for an excellent survey.

  

171

N is an instance of A. It is also clearly true. If we exclude from the range of the quantifier stateless persons and people with multiple nationalities, the relation . . . is a compatriot of . . . is an equivalence relation, and nationality is a function from persons to countries. But accepting all this doesn’t force us to treat N as a definition by abstraction of nationality. One could instead take the more plausible line of treating nationality as defined independently of N and treating N as a definition of the compatriot relation in terms of the nationality function. As this example illustrates, among those who accept an abstraction principle as true, there’s scope for disagreement between those who treat it as defining an equivalence relation in terms of a function (left-to-right) and those who treat it as defining, by abstraction, a function in terms of an equivalence relation (right-to-left). Consider now the abstraction principle that Quine is proposing to use to define meaning: M

For all expressions x, y, x has the same meaning as y iff x is synonymous with y.

The two approaches that Quine compares in the quoted passage correspond to the contrast between treating M as a left-to-right definition of synonymy in terms of meaning and treating it as a right-to-left definition of meaning in terms of synonymy. Taking the first option requires providing a definition of the function pairing each expression with the object playing the role of its meaning and then defining the synonymy relation in terms of which expressions receive the same image under this function. Taking the second option, favoured by Quine, requires providing a definition of the relation that each expression bears to its synonyms and then defining the meaning function by abstraction in terms of this relation. Hence, those, like Quine, who want to use M as a definition by abstraction of the meaning function in terms of the synonymy relation need to provide a definition of the synonymy relation independent of M, i.e. not in terms of whether expressions have the same image under the meaning function. Quine makes this point in the continuation of the passage: The relation of synonymity, in turn, calls for a definition or a criterion in psychological and linguistic terms. Such a definition, which up to the present has perhaps never even been sketched, would be a fundamental contribution at once to philology and philosophy. (Quine 1943: 120)

Now, Quine’s subsequent work contains a multipronged attack on the availability of the independent definition of synonymy the passage calls for. His arguments against the analytic-synthetic distinction (Quine 1980) and in support of the indeterminacy of translation (Quine 1960: ch. 2) would show, if successful, that it’s not possible to provide an independent definition of synonymy on which it

172

 

comes out as an equivalence relation, and hence as fit for sustaining a right-to-left reading of M. This outcome leads to a dead end for our attempts to define meaning. If (a) meaning can be defined only by abstraction with a right-to-left reading of M, but (b) it’s not possible to provide an independent definition of synonymy as an equivalence relation, it follows that meaning can’t be defined.⁹ This is not the place to assess Quine’s arguments, but I’m going to work on the assumption that they don’t succeed—that for at least some of our terms it is possible in principle to specify synonymy conditions. Notice that a pragmatist account of the meaning ground of a predicate entails that it’s possible to define independently of M the equivalence class of predicates that are synonymous with it. In what follows, I’m going to assume that there are no general reasons for thinking that individual predicates can’t receive this treatment.

8.5 Abstraction and Predicate Reference Our goal is to adapt to the notion of predicate reference the approach that Quine recommends for the meaning of an expression. Like Quine, we want to use an abstraction principle to define a function, from predicates to properties, in terms of a relation. The domain of the function will not include all predicates, but only those that can be successfully used for the task of representing the world. Call the predicates that satisfy this condition representational.¹⁰ The relation that will do the job for us is not quite synonymy. Synonymous predicates have the same referent, but the converse doesn’t hold in general. What we need is a modified version of synonymy for which the entailment also holds in the opposite direction. I’m going to use for the relation that satisfies this condition the label C-synonymy. We can now formulate the following abstraction principle: C

For all representational predicates P, Q, the referent of P = the referent of Q iff P and Q are C-synonymous.¹¹

My question will be whether C can be used to provide a plausible thick definition by abstraction of predicate reference.¹² There are circumstances under which an

⁹ I think that discussion of Quine’s ideas in this area pays insufficient attention to the role played by (a) in his overall line of reasoning. (Putnam 1994: 363) is an exception. ¹⁰ See Section 7.13, above. ¹¹ Another option worth exploring would be to stick to the standard notion of predicate synonymy and replace the biconditional in C with a conditional: For all representational predicates P, Q, the referent of P = the referent of Q if P and Q are synonymous. ¹² This use of the method of definition by abstraction differs from the use to which the method has been put by neo-Fregean philosophers of mathematics in two important respects. First, although the abstractionist definition of number has clear semantic consequences, the equivalence relation used in the abstraction principle, i.e. the existence of a bijection, is not a semantic relation. By contrast, in the

  

173

affirmative answer would be irresistible on our assumptions. Suppose we could provide necessary and sufficient conditions for C-synonymy independently of C. This would open the way to using C as a right-to-left definition by abstraction of predicate reference in terms of C-synonymy, perhaps adapting Quine’s idea by treating equivalence classes of C-synonymous predicates as referents. Now, according to the Peano assumption, it follows from this that it would be equally legitimate to use C as a thick definition of predicate reference, accepting that the referents of representational predicates cannot be defined independently of C. This is potentially good news for the pragmatist. Suppose that the pragmatist meaning ground of a predicate could be taken as defining, not synonymy conditions, but C-synonymy conditions for the predicate. Call global pragmatism the view that every representational predicate has a pragmatist meaning ground. If global pragmatism is true, it follows that we can provide a definition of Csynonymy independently of C. And, as we have seen, on the Peano assumption, this would force us to accept C as a legitimate thick definition by abstraction of predicate reference in terms of C-synonymy. Notice that this outcome wouldn’t be in principle undermined by the observation that the referents of some predicates can be defined independently of C. The global pragmatist could point at cases in mathematics in which thick definitions by abstraction exhibit this feature. Take for example the definition of rational numbers in terms of the following abstraction principle: For all integers x, y, w, z (y, z ≠ 0), rat(x/y) = rat(w/z) iff x * z = y * w. On the Peano assumption, we would be able to treat this principle as a legitimate thick definition of the rationals, where all we are told about the identity of a rational is which fractions represent it. And yet, we would be able to go on to identify some rationals, as defined in this way, but not all, with items already given. Thus, for all integers m, n (m ≠ 0), we would be able to identify rat(m*n/m) with the integer n, while the image of all other fractions would not be identifiable independently of the abstraction principle. For example, we would be able to identify the image of 6/3 under the rat function with the integer 2. The image of, say, 3/6, by contrast, would only be identified through the abstraction principle, in terms of which other fractions are paired with it. The global pragmatist could maintain that predicate reference follows this model. On this line, C would be taken as a thick definition by abstraction of application of the method that I’m proposing, the equivalence relation playing this role, synonymy, is a semantic relation. Second, neo-Fregeans have generally focused on applying the method to the reference of singular terms, although they have also considered extending it to predicate reference (Hale and Wright 2001: 432–3). In the application I’m proposing, by identifying the referent of the singular term ‘the referent of predicate P’ we would be ultimately identifying the referent of a predicate, P.

174

 

predicate reference, but some of the predicate referents defined in this way could be subsequently identified with items on our pre-existing catalogue of properties.

8.6 The Two-Way Approach However, the question whether global pragmatism would sustain a legitimate thick definition by abstraction of predicate reference is academic, as we know that global pragmatism is false: there are predicates for which C-synonymy cannot be defined independently of C. I take this to be a direct corollary of work on natural-kind terms by Hilary Putnam and Saul Kripke (Putnam 1975; Kripke 1980). The problem is that any characterization of our use of the predicate ‘is water’ that the pragmatist might try to employ as an account of its meaning ground will also be satisfied by our twins’ predicate ‘is twater’. However, ‘is water’ and ‘is twater’ are not C-synonymous, or even co-extensive. For predicates that exhibit this kind of behaviour, a pragmatist account of their meaning grounds is not an option. Global pragmatism is false. How should we react to this fact? One obvious possibility is to abandon the project of trying to use C to define predicate reference in terms of C-synonymy. If we take this line, we will accept that if predicate reference can be defined at all, it will have to be defined independently of C, by locating on our catalogue of properties the item that plays the role of referent for each predicate. This outcome would have devastating consequences for the approach I am defending. If we want to maintain that a representational predicate has a pragmatist meaning ground, avoiding the problem of harmony will force us to adopt one of the two strategies outlined in Section 8.1 for identifying its referent. If we think that these strategies face insurmountable difficulties, we will have to abandon the claim that the predicate succeeds in representing the world, and choose between ascribing to it a nonrepresentational role and treating it as having a representational role but failing to discharge it. As far as I can see, there’s only one possible alternative to this assessment. I don’t know if it is ultimately satisfactory, but it is certainly worth considering. So far we have conducted our discussion on the implicit assumption that we have to choose between using an abstraction principle such as C as a left-to-right definition of an equivalence relation in terms of a function for all its arguments and using it as a right-to-left definition by abstraction of a function in terms of an equivalence relation for all its arguments. On this assumption, the realization that it’s not possible to define C-synonymy for natural-kind predicates independently of C leaves us no option but to define predicate reference independently of C, and to take C as a left-to-right definition of C-synonymy in terms of reference for every representational predicate. However, if we abandon this assumption, a different approach becomes available. On this proposal, the direction in which C is used would be different for

  

175

different predicates. For some predicates, e.g. natural-kind predicates, their referents would be defined independently of C, and C would be treated as a leftto-right definition of C-synonymy in terms of reference. However, for other predicates, C-synonymy would be defined independently of C, and their referents would receive a (thick) definition by abstraction under a right-to-left reading of C. I’m going to label this as the two-way approach. If the two-way approach is a viable option, the position I’m advocating will be able to accommodate the behaviour of natural-kind predicates without abandoning the strategy for solving the problem of harmony offered by definitions by abstraction. If a representational predicate P has a pragmatist meaning ground, its referent will be identified by a right-to-left reading of C, as the item with which the reference function pairs those predicates that satisfy the meaning ground of P. And this way of defining the predicate’s referent will sustain a solution to the problem of harmony. Now a predicate that satisfies the meaning ground of P couldn’t fail to have the same referent as P. On this proposal, there would be a clear contrast between two kinds of representational predicates, depending on whether they have pragmatist or representationalist meaning grounds. However, this difference would not be taken to be reflected in the ontological status of their respective referents or of the predicate-referent link. The referents of predicates would be regarded as properties, whether they have been identified independently, as with predicates with representationalist meaning grounds, or through C, as with those with pragmatist meaning grounds. And in both cases predicates would represent the objects to which they are ascribed as instantiating the properties they refer to.¹³

8.7 Some Consequences of the Two-Way Approach Taking this line wouldn’t force us to deny some important differences between referents defined independently of C and those that are introduced by abstraction using C. One important difference concerns the causal efficacy of referents. Causally efficacious properties can be identified by their causal powers. Hence when a predicate has a causally efficacious referent it will be possible in principle to identify it independently of C. This might force us to accept that the referents that are identified by abstraction with a right-to-left reading of C won’t be causally ¹³ Øystein Linnebo (2018) draws a distinction between thick and thin objects, depending on whether their existence makes a substantial demand on the world, and argues that objects that have been defined by abstraction should be regarded as thin. This conclusion seems unavoidable if we agree with Linnebo that ‘the obtaining of an appropriate criterion of identity becomes sufficient for the existence of an object governed by this criterion’ (Linnebo 2018: 23). But on my proposal the existence of an appropriate criterion of identity for the referent of a predicate is not sufficient for the existence of this referent. The referent will not exist unless the predicate satisfies the conditions for counting as representational presented in Section 7.13.

176

 

efficacious.¹⁴ What the proposal is incompatible with is the idea that causal efficacy should be treated as a criterion of existence for properties. It follows from the proposal that the world contains causally inert properties—as they alone can serve as referents for predicates with pragmatist meaning grounds.¹⁵ Notice, by contrast, that this approach doesn’t force us to treat the referents introduced by a right-to-left reading of C as mind- or language-dependent. On the position I’m outlining, abstraction is seen not as bringing properties into existence, but as our tool for identifying independently existing properties.¹⁶ Another important difference between identifying the referent of a predicate independently of C and identifying it by abstraction with a right-to-left application of C is that in the former case, but not in the latter, we obtain an independent criterion for assessing the procedures that we use for deciding on the ascription of the predicate. When we establish that ‘is water’ refers to the property of containing mostly H₂O molecules, we acquire the means for performing an independent check on whether the transparent, tasteless, odourless liquids to which we have ascribed the predicate actually satisfy it. Defining a referent by abstraction doesn’t have this consequence. Establishing that the referent of ‘is morally right’ is the referent of those predicates that share the pragmatist meaning ground of ‘is morally right’ doesn’t provide me with an independent procedure for checking whether the predicate is actually satisfied by the actions that produce in me the feeling of moral approval. After the predicate’s referent has been identified in this way, we will still be reliant on our feelings of moral approval for deciding on the ascription of the predicate.¹⁷ Another important feature of the proposal follows from a general point about the sense in which a thick definition by abstraction enables us to identify the images under the abstraction operator of the objects in its domain. Suppose that we use Hume’s Principle as a thick definition by abstraction of the number of Fs. What the principle tells us about the identity of the image of F under this function is that it is identical with the image of every concept G such that there is a bijection between the extensions of F and G (i.e. such that F and G are ¹⁴ But see the discussion of the difference-making approach to the causal efficacy of propositional attitudes in Section 6.6, above. ¹⁵ The rejection of the causal criterion for property existence could be justified from a general hostility to substantive existence criteria, except, for predicate referents, those that follow from our criteria for when a predicate should be treated as representational (see Section 7.13, above). On the rejection of substantive existence criteria, see (Thomasson 2015: 115–21). ¹⁶ On this point, see (Fine 2002: 57). ¹⁷ This consequence of identifying a predicate’s reference by abstraction can be usefully characterized in terms of a distinction introduced by Paul Horwich between two senses in which the meaningconstituting property of a predicate can be said to determine its extension (Horwich 1995). In the first sense of determination, the meaning-constituting property singles out the set of objects that satisfy the predicate; in the second sense, extension determination means ‘merely that any two predicates with the same meaning-constituting property are co-extensional’ (Horwich 1995: 363). See also Horwich 1998b: 69–71). Pragmatist meaning-constituting properties (meaning grounds) determine extension in the second sense, but not in the first.

  

177

equinumerous). If we take the definition as thick, we will have to accept this as a satisfactory identification of the object playing this role. Notice, however, that the identification provided by Hume’s Principle doesn’t address a very natural motivation for wanting to know the number of Fs. If we want to know how many apples there are in the basket, our question won’t be adequately answered by the observation that the number of apples in the basket is the same as the number of Gs for every concept G that’s equinumerous with the apples in the basket. Our question will receive a satisfactory answer from Hume’s principle only if we independently know how many Gs there are for some concept G equinumerous with the apples in the basket. In general, we will be able to use Hume’s Principle to find out how many Fs there are only when there is a concept G equinumerous with F such that we know, independently of Hume’s principle, how many Gs there are.¹⁸ A similar situation obtains for the predicate referents identified with right-toleft applications of C. What C tells us about the identity of the referent of ‘is morally right’ is that it is identical with the referents of all other predicates whose ascription is regulated by the feeling of moral approval. But knowing this won’t reveal the identity of the referent of these predicates to someone who doesn’t independently understand one of them.¹⁹ C enables one to go from understanding of a predicate to knowledge of the identity of its referent. It doesn’t reveal the identity of the referent of a predicate to someone who doesn’t independently understand the predicate or one of its C-synonyms. In this sense, the referent identifications produced by right-to-left applications of C are parasitic on our understanding of the predicates whose referents are identified in this way. C reveals to me the identity of the referent of a predicate with a pragmatist meaning ground only if it is a predicate I understand or it is C-synonymous with a predicate I understand. For any other predicate with a pragmatist meaning ground, C will leave me in the dark about the identity of its referent. This is in sharp contrast with the situation for predicates whose referents are identified independently of C. If I understand ‘“is water” refers to the property of containing mostly H₂O molecules’, I know what referent is being ascribed to the predicate. On the other hand, understanding ‘“is morally right” refers to the same property as any predicate whose ascription is regulated by moral approval’ will not reveal the identity of the referent that’s being ascribed to the predicate to someone who doesn’t already understand some predicate with the relevant pragmatist meaning ground.²⁰

¹⁸ See in this connection my discussion of the role of what I call counting sequences in (Zalabardo 2000a). ¹⁹ When we lack a C-synonymous expression, we are in the same situation as an anthropologist who can give a complete description of the way the locals use a term of their language but can’t locate in her own repertoire a concept that would correspond to that pattern of use. ²⁰ This is in line with the approach that Hartry Field has labelled egocentric: ‘start from our own representations, which [ . . . ] are used as standards in attributing contents to others; and view all talk of content as some kind of projection from this’ (Field 2017: 523).

178

 

The two-way approach would provide the position I’m defending with a promising strategy for dealing with the problem of harmony. If successful, it would remove a major obstacle to treating representation as the primary function of predicates with pragmatist meaning grounds. The viability of the two-way approach can be challenged on two fronts. First, one might reject the general idea that an abstraction principle can provide a satisfactory definition of a function if the objects in its range cannot be identified independently of the principle. Second, one might accept in general the legitimacy of a thick definitions by abstraction, but object to the idea of using the approach for only some of the arguments of the abstraction operator. Nothing I’ve said here addresses the first kind of concern, as I’ve worked on the assumption that thick definitions by abstraction are in principle legitimate. Discharging this assumption would be a formidable task. I want to suggest that with respect to the second kind of concern the two-way approach stands on stronger ground. If we are prepared to accept that a thick definition by abstraction of a function can successfully identify the values of all its arguments, there is no obvious reason why we should oppose the possibility of achieving this for the values of some but not all of its arguments.²¹

8.8 Abstraction and the Problematic Discourses My original proposal was that the predicates and sentences of the problematic discourses have pragmatist meaning grounds but play the role of representing the world in exactly the same sense as expressions with representationalist meaning grounds. I now add to the picture I’m presenting the thought that the predicates of these discourses have referents, which are identified by abstraction principles based on their meaning-grounding ascription procedures. Likewise, their sentences represent states of affairs, which are similarly identified by abstraction principles based on their meaning-grounding acceptance procedures. Let’s consider briefly the picture that results for each of the problematic discourses. In the case of ethics, the predicate ‘is morally right’ will have a referent—a property that we represent an action as instantiating when we ascribe the predicate to it. This property will be sui generis, in that it will not be found in a preexisting catalogue of properties. However, its sui-generis character won’t render our access to the property playing this role problematic. Our belief that the referent of the predicate exists is grounded in our conviction that the predicate ²¹ Frege seems to be expressing his reservations about this move in the following passage: ‘It is not only about numbers that the relation of identity is found. From which it seems to follow that we ought not to define it specially for the case of numbers. We should expect the concept of identity to have been fixed first, and that then, from it together with the concept of Number, it must be possible to deduce when Numbers are identical with one another, without there being need for this purpose of a special definition of numerical identity as well’ (Frege 1980a: 74).

  

179

satisfies the criteria for representational character.²² We also have a way of deciding whether a predicate, as understood by a speaker, refers to this property. Our verdicts on this point are grounded in the abstraction principle with which the referent of the predicate is identified—a predicate, as understood by a speaker, will refer to this property just in case the speaker regulates its ascription by her feeling of moral approval. However, we shouldn’t see the abstraction principle as specifying the instantiation conditions of the property it singles out. The feeling of moral approval can’t do this, since it is prompted by different actions for different speakers at different times, and the definition by abstraction of the property doesn’t provide us with an alternative method for determining whether an action instantiates it. After the referent has been identified in this way, our feeling of moral approval remains our only procedure for determining which actions instantiate it. Notice that the phenomenon identified as the intrinsically motivating power of the predicate’s reference has a non-mysterious explanation on this account, on the natural assumption that the feeling of moral approval necessitates a motivation to perform or promote the actions that trigger it. When a speaker who ascribes a predicate to an action represents this action as instantiating the referent of ‘is morally right’, she couldn’t fail to be motivated to perform or promote the action. If she didn’t have the motivation, she wouldn’t feel moral approval, and if she applied the predicate without feeling moral approval, then the predicate, as understood by her, would not have the same referent as our predicate ‘is morally right’. We obtain a similar picture with truth ascriptions. On this account, the predicate ‘is true’ refers to a property, and by ascribing the predicate to a sentence we represent the sentence as instantiation this property. Once again, the property that plays this role is sui generis, in that we won’t be able to locate it in our preexisting catalogue of properties. This circumstance accounts for the impression that truth doesn’t have a hidden essence that a substantive theory of truth would have to reveal, but, on the approach I’m presenting, this circumstance shouldn’t be taken as making the property of truth somehow less substantial than the properties that were in our catalogue before the truth property was introduced by our abstraction principle.²³ Once again, when we identify the referent of the truth predicate by abstraction we don’t acquire a new procedure for determining whether a sentence instantiates its referent, whose verdicts would enable us to validate our procedures. After the referent of the truth predicate has been identified, it remains the case that the only procedure at our disposal for deciding whether the truth property is instantiated by a sentence is whatever procedure we use for regulating our acceptance of

²² See Section 7.13, above.

²³ On this point, see (Horwich 1998a: 38–9).

180

 

the sentence. This feature of truth is sometimes labelled as transparency (Kalderon 1997). For propositional-attitude ascriptions, once again, on this picture, there will be a property that, say, the predicate ‘believes that’s a horse’ refers to. The property that plays this role will be identified by abstraction in terms of the hybrid policy, as the property that we ascribe to a subject on the basis of whether the hybrid policy recommends ascribing to her the belief that that’s a horse. A speaker will represent a subject as instantiating this property with her ascriptions of a predicate just in case she uses this procedure for regulating its ascription. As before, when we identify the referent of the predicate in this way, we don’t acquire a new procedure for determining whether a subject satisfies the predicate, with respect to which we would be able to validate the results we obtain by applying the hybrid policy. In this case, unlike the previous ones, we face a problem concerning the thought that the properties that we represent subjects as instantiating when we ascribe propositional-attitudes to them appear to be causally efficacious, since we think of beliefs, desires, etc. as playing a causal role in the production of behaviour. As I suggested in Section 6.6, we might be able to achieve this on a difference-making account of causation. And if this is a viable approach, we might be able more generally to treat referents introduced by abstraction as causally efficacious. With meaning ascriptions we find a similar situation. There’s a property that ‘means table’ refers to. It’s the property that we represent a term, as understood by a speaker, as instantiating, when we ascribe the predicate to it. The property that plays this role will be identified by abstraction in terms of our interpretative procedures, as characterized in Chapter 7, as the property that we ascribe to a term, as understood by a speaker, on the basis of whether these procedures recommend interpreting the term as meaning table. As before, identifying the predicate’s referent by abstraction won’t remove our reliance on our interpretative procedures for deciding whether the property is instantiated by a term, as understood by a speaker.

8.9 Aufhebung? I have argued in this chapter that so long as we are prepared to accept thick definitions by abstraction as a legitimate method for defining a function and its values, we can use the pragmatist meaning ground of a predicate to single out the property it refers to. As a result, predicates with pragmatist meaning grounds, no less than those with representationalist meaning grounds, will be paired with properties by the reference function. Likewise, for sentences with pragmatist meaning grounds there will be states of affairs they represent as obtaining in the same way as for those with representationalist meaning grounds. The two kinds of predicate will be on a par with regard to their semantic relations to the

  

181

world. The only difference between the two kinds of predicate will be the character of our access to their referents. The referents of predicates with representationalist meaning grounds will be singled out by locating them in a pre-existing catalogue of properties, while the referents of predicates with pragmatist meaning grounds will be singled out by abstraction principles based on their meaning-grounding ascription procedures. In light of these parallels, we might want to question whether it still makes sense to treat both kinds of predicate as obtaining their meanings in fundamentally different ways. Once we’ve accepted that predicates with pragmatist meaning grounds refer to properties, wouldn’t we be able to say that they also obtain their meaning from their relation to the properties they refer to? I see no harm in saying this so long as it doesn’t distort our understanding of the situation. The main point we need to keep in mind is that this way of speaking shouldn’t be seen as legitimizing a left-to-right reading of the abstraction principles with which referents are identified. In these cases, sameness of referent cannot be seen as explaining why predicates with the same ascription procedures are C-synonymous. It’s still the case that having the same ascription procedures is what makes two predicates C-synonymous and what makes them have the same referent. The fact that pragmatist meaning grounds can be reformulated in this way can also be used to address the complaint that if the meaning ground of a predicate is pragmatist, its representational character is not built into its meaning ground.²⁴ I’m suggesting that the objection can be neutralized by pointing out that if a predicate has a pragmatist meaning ground, we can say that it has the meaning it has by virtue of the fact that it refers to the property singled out by the abstraction principle generated by its meaning-grounding ascription procedure. Then the function of representing the objects to which the predicate is ascribed as instantiating this property will be built into the meaning ground of the predicate, in the same way as for predicates with representationalist meaning grounds.

8.10 Conclusion In the present chapter I’ve addressed what I regard as the last remaining challenge to the view that predicates and sentences with pragmatist meaning grounds can successfully perform the function of representing the world. If a declarative sentence represents the world, there has to be a state of affairs that the sentence represents as obtaining. And if the sentence has a pragmatist meaning ground, its meaning-grounding acceptance procedure has to be sufficient for the sentence to

²⁴ See Section 7.13, above.

182

 

represent the state of affairs it represents. Likewise, if a predicate successfully performs the function of representing the world, there has to be a property that the predicate refers to, and if the predicate has a pragmatist meaning ground, its meaning-grounding ascription procedure has to be sufficient for the predicate to have the referent it has. I’ve proposed to meet this challenge by treating the referents of predicates with pragmatist meaning grounds as identified with abstraction principles based on the C-synonymy conditions generated by their meaning-grounding ascription procedures. I have suggested that this account of the referents of these predicates is compatible with the idea that the referents of other predicates should receive an independent specification in terms of which C-synonymy for them can be defined. I have considered some consequences of specifying the referent of a predicate in this way and I have outlined how the approach would work for the predicates of each of our target discourses. In the next chapter we widen our focus, from the provision of pragmatist meaning grounds for specific discourses to exploring the idea that pragmatist principles might be used to explain linguistic representation in general, including in those discourses to which the pragmatist approach is not directly applicable.

9 The Primacy of Practice In the preceding chapters I’ve developed the pragmatist strategy for specifying the meaning grounds of declarative sentences, and I have argued that sentences whose meaning grounds are specified along the lines of the pragmatist template can discharge the task of representing the world in the same sense in which sentences with representationalist meaning grounds achieve this. It follows from this that the RR assumption has to be rejected—a representationalist meaning ground is not a necessary condition for a sentence to discharge the task of representing the world. So far I’ve concentrated on the consequences of this situation for a range of discourses, including ethical discourse and semantic discourses, for which RR doesn’t seem to yield a satisfactory treatment. These are discourses that appear to discharge the task of representing the world even though the enterprise of finding representationalist meaning grounds for their sentences faces formidable difficulties. In the present chapter I’m going to discuss a more fundamental role that the pragmatist strategy can be called upon to play, over and above the vindication of the representational status of specific discourses. I’m going to argue that the representationalist strategy is not self-sufficient—it needs to be underpinned by an alternative strategy for specifying meaning grounds of declarative sentences, and the pragmatist strategy can play this role. The issue I’m going to discuss concerns the involvement of language in the specification of the meaning grounds of sentences and predicates. A specification of the meaning grounds of linguistic expressions, like any theory of anything, will be formulated in some language. This goes for pragmatist and representationalist meaning-ground specifications alike. However, I’m going to argue that in the case of representationalist specifications, the language in which they are expressed has to satisfy requirements that couldn’t be satisfied if the representationalist strategy were the only method at our disposal for specifying meaning grounds.

9.1 Theoretical Science Representationalism faces difficulties in a region of ostensibly representational discourse that we haven’t considered yet—sentences about unobservable entities postulated by scientific theories. Consider the sentence: G

‘Gluons have colour charge’.

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0009

184

 

G is a declarative sentence which appears to represent things as being a certain way, no less so than, say, the sentence ‘Giraffes have long necks’. On the standard analysis of universal sentences like these, they involve two predicates—‘is a gluon’ and ‘has colour charge’, or ‘is a giraffe’ and ‘has a long neck’. On the version of representationalism on which we have focused, what makes a sentence of this form have the meaning it has is the relation between each of the predicates and the property that plays the role of its referent. The sentence then represents every instance of the referent of the first predicate as instantiating the referent of the second—G, for example, will represent every instance of the referent of ‘is a gluon’ as instantiating the referent of ‘has colour charge’. The challenges faced by representationalism here have a different complexion from the other problematic discourses we’ve considered. In previous cases, the main challenge was to locate properties that could serve as referents in the catalogue of properties that the world contains provided by our metaphysics. However, for a naturalist philosopher, this problem doesn’t arise for terms in basic science. The naturalist delegates to science the task of compiling the catalogue of properties the world ultimately contains. The properties that the naturalist believes to exist are the properties that our best scientific theories postulate— i.e. the referents of the predicates that figure in these theories. Since ‘has colour charge’ is such a predicate, there’s no non-trivial task of locating its referent in the naturalist’s catalogue of properties the world contains. The problems faced by representationalism in this area are of a different kind. They concern our grasp of the identity of the referents of the relevant predicates. We’ve seen that it is perfectly possible for a predicate, as understood by a speaker, to refer to a property even though the speaker hasn’t identified the property that plays this role. This is the situation in which chemically illiterate English speakers find themselves with respect to the predicate ‘is water’. The predicate, as they understand it, refers to the property of containing mostly H₂O molecules, but they are not aware of this fact. However, this state of ignorance is purely contingent. More enlightened subjects can know the identity of the property the predicate refers to, and the chemically illiterate speakers can overcome this shortcoming themselves by learning some chemistry. In this case, the facts in which the representationalist locates the meaning ground of the predicate might not be known to the speaker, but they are in principle ascertainable by the philosopher with the right scientific knowledge. We can present in these terms the main idea of the difficulty I want to raise for a representationalist account of the meaning ground of ‘has colour charge’. The problem is that the situation in which chemically illiterate speakers find themselves with respect to the referent of ‘is water’ is analogous to the situation in which all of us find ourselves with respect to the referent of ‘has colour charge’, irrespective of the level of our scientific knowledge. And this shortcoming cannot be overcome. Our grasp of the referent of ‘has colour charge’ is necessarily of the same nature as the chemically illiterate speaker’s grasp of the referent of ‘is water’.

   

185

Hence, in this case, representationalism locates the meaning grounds of predicates in facts that are necessarily inaccessible to us. On the standard approach to the semantics of theoretical terms, their meaning resides in the role they play in the theory in which they figure (Duhem 1991; Poincaré 1952). This general approach produces an account of how theoretical terms are understood and an account of how they obtain their referents. Understanding of theoretical terms is construed as knowledge of the role they play in the theories in which they figure. To explain the reference of theoretical terms, the approach treats a theory as a description of the referents of the terms it introduces. If there’s a unique collection of items in the world that behave as the theory says the referents of its terms behave, then these items are the referents of the terms. If there are no items behaving in this way, then the terms introduced by the theory have no referents. If there are several collections of items exhibiting the requisite behaviour the situation is less clear, but we don’t need to concern ourselves with this case here. On this approach, the reference of ‘has colour charge’ follows the same template as the reference of ‘is water’. The referent of ‘is water’ is singled out as the unique property, if there is one, that satisfies a certain condition—something along the lines of being the underlying substance of the transparent, thirst-quenching liquid that fills our lakes. Likewise, the referent of ‘has colour charge’ is singled out as the unique property, if there is one, that satisfies a certain condition—behaving in the way that particle physics says the referent of the predicate behaves. Now notice the situation the illiterate speaker finds himself in with respect to the referent of ‘is water’: he knows the reference-fixing condition, but he doesn’t know which property satisfies this condition. I want to suggest that physicists know no more than this with respect to the referent of ‘has colour charge’—they know the reference-fixing condition, i.e. the theory in which the term figures, and this is all they know. And if knowing the reference-fixing condition in the case of ‘is water’ doesn’t suffice for knowing the identity of the referent, knowing the reference-fixing condition in the case of ‘has colour charge’ doesn’t suffice either for knowing the identity of the referent. Therefore, physicists are as ignorant of the identity of the referent of ‘has colour charge’ as chemically illiterate speakers with respect to the referent of ‘is water’. Furthermore, as we’ll see in the next section, there doesn’t seem to be anything they could do to overcome this deficit. But the representationalist specifies the meaning ground of ‘has colour charge’ in terms of the property it refers to. It follows that representationalism locates the meaning ground of the predicate in facts to which we necessarily have no access.

9.2 Lewisian Humility David Lewis has provided an argument for a claim that would have the same negative consequences for the prospects of representationalism in this area—the

186

 

claim that we cannot identify the fundamental properties. He refers to this claim as Humility. Let me start with a brief summary of the concepts that Lewis employs for formulating his argument. Assume, as Lewis does, that there is a true and complete final theory T. The vocabulary of T can be divided into theoretical terms that T implicitly defines (T-terms) and the rest (O-terms), which are interpreted independently of T. Lewis assumes, without loss of generality, that all the T-terms are names, and that no two of them name the same thing (Lewis 2009: 206). He also assumes that the O-terms suffice to express all possible observations (Lewis 2009: 206). Each of the fundamental properties that play an active role in the actual workings of nature will be the referent of a T-term. Lewis assumes that T can be formulated as a one-sentence axiom, the postulate of T.¹ This can be written as T(t₁, . . . , tn), where t₁, . . . , tn are the T-terms. Replacing T-terms by variables in the postulate of T we get the formula T(x₁, . . . , xn), in which only Oterms occur. Given the interpretation of the O-terms, an n-tuple may or may not satisfy this formula at a possible world. Call a tuple that satisfies this formula in actuality an actual realization of T and one that satisfies it in some possible world a possible realization of T. Since T is the true and complete theory and it implicitly defines the T-terms, it has a unique actual realization (Lewis 2009: 207). If we existentially quantify the free variables in T(x₁, . . . , xn), we obtain the Ramsey sentence of T: ∃x₁, . . . , xn T(x₁, . . . , xn). As Lewis observes, the Ramsey sentence of T ‘logically implies exactly the O-language sentences that are theorems of T’ (Lewis 2009: 207). Since we are assuming that the O-terms suffice to express all possible observations, it follows that any predictive success for T is equally a predictive success for the Ramsey sentence of T. Since the evidence for T consists in its record of predictive success, there is no way to gain evidence for T that is not equally evidence for the Ramsey sentence. (Lewis 2009: 207)

This, according to Lewis, has an important consequence, on the assumption (call it Variance) that T has multiple possible realizations: no possible observation can tell us which one is actual, because whichever one is actual, the Ramsey sentence will be true. (Lewis 2009: 207)

Therefore, if Variance is true, Humility follows. How does Lewis argue for Variance? As Lewis explains, ‘some T-terms might be names of other things [ . . . ] but for the most part the T-terms will name fundamental properties’ (Lewis 2009: 206).

¹ As he explains, this is not a substantive assumption (Lewis 2009: 206).

   

187

Let’s simplify matters by assuming that all T-terms name fundamental properties, and that the postulate of T says that they do. Properties fall into several categories: There are all-or-nothing monadic properties. There are all-or-nothing n-adic relations, at least for smallish n. There are properties that admit of degree, that is, magnitudes; more generally, there are scalar-valued, vector-valued, tensorvalued, . . . magnitudes. There are relational magnitudes. (Lewis 2009: 205)

We should expect the postulate of T to say which of these categories the property named by each T-term belongs to. It follows that for each position in an n-tuple of fundamental properties, there is a property category that the component of the tuple in that position has to belong to in order for the tuple to be a possible realization of T. If an n-tuple of fundamental properties satisfies this categorial condition, we’ll say that it is T-compliant. Lewis provides two arguments for Variance—the permutation argument and the replacement argument. The permutation argument, on which I’m going to focus, takes as its starting point the assumption that at least two properties in the actual realization of T fall in the same category.² As Lewis points out, ‘[a]n inconclusive reason to assume this is the expectation that current physics is not entirely wrong in its inventory of fundamental properties’ (Lewis 2009: 206). This assumption entails, and is equivalent to, the claim that there are multiple T-compliant permutations of the components of the actual realization of T. From this, Lewis derives Variance with the help of two metaphysical principles: combinatorialism and quidditism. Combinatorialism ‘tells us that possibility is preserved under permutations’ (Lewis 2009: 209). Quidditism tells us that if we permute same-category properties in the actual realization of T, ‘the permutation is indeed a different possibility’ (Lewis 2009: 209). If A is the unique actual realization of T, we can formulate in the following terms Lewis’s permutation-based argument for Humility: 1

There are multiple T-compliant permutations of the components of A. (Premise)

2

Every T-compliant permutation of the components of A is a possible realization of T. (Combinatorialism)

3

If a T-compliant permutation B of the components of A is a possible realization of T, then A and B are different possible realizations of T. (from quidditism)

4

VARIANCE: T has multiple possible realizations. (from 1, 2, and 3)

² The replacement argument avoids this assumption with a different argument for the existence of multiple T-compliant n-tuples of fundamental properties. The reasoning from this point is the same as in the permutation argument.

188

 

5

For every possible realization X of T, if X is the actual realization of T, the Ramsey sentence of T is true. (Premise)

6

If two hypotheses make the Ramsey sentence of T true, then no possible observation favours one over the other. (Premise)

7

For every possible realization X of T, no possible observation favours A over X as the unique actual realization of T. (from 5 and 6)

8

For some possible realization X of T, no possible observation favours A over X as the unique actual realization of T (from 4 and 7)

9

If for some possible realization X of T, my observations don’t favour A over X as the unique actual realization of T, then I don’t know the identity of the unique actual realization of T. (Premise)

10

HUMILITY: I can’t know the identity of the unique actual realization of T. (from 8 and 9)

Lewis presents this argument as sound, and is happy to accept its conclusion: ‘we are irremediably ignorant about the identities of the fundamental properties that figure in the actual realization of the true final theory’ (Lewis 2009: 214). He doesn’t see anything particularly devastating about admitting this kind of ignorance. ‘Why is Humility “ominous”?’, he asks, ‘Whoever promised me that I was capable in principle of knowing everything?’ (Lewis 2009: 211). I want to suggest that advocates of a representationalist account of the meaning grounds of terms referring to fundamental properties cannot take the same relaxed attitude towards Humility. According to representationalism, what makes one of these terms have the meaning it has is the fact that it refers to a fundamental property. Hence if, as Humility dictates, we can’t know the identity of fundamental properties, representationalism explains the meaning of these predicates in terms of facts to which we can have no access. Even if this consequence doesn’t render the view untenable, it certainly detracts from its appeal. It also opens the position to a familiar line of attack. If we can have no access to the items that the representationalist treats as the referents of theoretical terms, it’s hard to see how they could play an effective role in the explanation of linguistic phenomena.

9.3 Epistemological Challenges to Lewis’s Argument One standard strategy for resisting Lewis’s argument targets the epistemological principles that he employs for deriving Humility from Variance. This derivation invokes Premises 5, 6, and 9. 5 and 6 are fairly straightforward consequences of Lewis’s definitions. 9 is more substantial. It expresses a necessary condition for knowing the identity of the unique actual realization of T—my observations have to favour A over any other possible realization as the unique actual realization of T. Several strategies have been proposed for challenging this claim.

   

189

One option is to take issue with the epistemological approach that supports this kind of necessary condition for knowledge. 9 is an instance of the general principle that we don’t know p unless we can discriminate p from every alternative possibility. This general principle is a key ingredient in traditional sceptical reasoning. The sceptic uses it to advance from the observation that you cannot discriminate between the possibility that you have hands and the possibility that you are an envatted brain being fed impressions as of hands to the conclusion that you don’t know that you have hands. As Jonathan Schaffer has explained (Schaffer 2005), several extant antisceptical strategies seek to block the sceptical conclusion by rejecting the principle under discussion, arguing, to the contrary, that indiscriminable counterpossibilities are not always incompatible with knowledge. Dogmatist, anti-closure, and contextualist answers to scepticism offer different versions of this general strategy.³ Advocates of any of these positions will reject Premise 9, claiming that from the fact that my observations don’t favour A over some other possible realization as the actual realization of T it doesn’t follow that I don’t know the identity of the actual realization.⁴ A different strategy for resisting Premise 9 focuses not on conditions for knowledge, but on the notion of identification. According to Ann Whittle (Whittle 2006), identifying the actual realization of T does not require singling it out from other possible realizations. Hence the fact that we can’t obtain the evidence that would enable us to do this doesn’t force us to accept Humility. A third approach to resisting Premise 9 would exploit the fact that, as Lewis explains, my ignorance of the identity of the fundamental properties is ineffable ignorance. The ineffability of my ignorance arises from the limitations in the procedures at my disposal for referring to fundamental properties. As a result of these limitations, I can’t produce a sentence that identifies the referent of a theoretical term in a way that doesn’t make its truth value entirely obvious. Here’s Lewis’s illustration of the point: Which property occupies the seventeenth role? Is it the actual occupant of the first role? Is it the actual occupant of the second role? . . . Is it the actual occupant of the seventeenth role? . . . Is it the actual occupant of the nth role? No sooner do we ask our question this way than we seem to know the answer: it is the actual occupant of the seventeenth role. (Lewis 2009: 216)

³ See (Pryor 2000) for the dogmatist approach, (Nozick 1981) for the anti-closure approach, and (DeRose 1995) for the contextualist approach. For my approach to sceptical arguments based on indiscriminable counter-possibilities, see (Zalabardo 2012a: 146–50). ⁴ For some criticisms of this approach, see (Locke 2009). As Locke points out, one might try to use in this context Lewis’s own response to traditional scepticism, although Locke thinks that the strategy doesn’t work.

190

 

The conclusion Lewis draws from this circumstance is that our ignorance of the identity of the fundamental properties is ineffable. For every possible realization X of T, there is a contingent proposition true in all and only the worlds where X is the actual realization of T. What we don’t have, according to Lewis, is sentences that express these propositions ‘and do so in such a way that we can know which sentence expresses which proposition’ (Lewis 2009: 216). It’s unclear what role, if any, these considerations are supposed to play in Lewis’s argument for Humility. On my construal of Lewis’s reasoning, they don’t play any obvious role, but other authors have found in Lewis’s discussion a line of reasoning in which expressive limitations play a central role. Thus, according to Stephan Leuenberger (Leuenberger 2010), Lewis’s argument for Humility rests on the thought that propositions identifying the referents of the T-terms cannot be expressed in the O-language, but expressibility in the O-language is a necessary condition for knowledge.⁵ And Alexander Kelly (Kelly 2013) finds in Lewis an argument that derives the conclusion that we can’t know the proposition that A is the actual realization of T from the principle that we are not capable of grasping propositions of this kind. I’m not going to discuss these proposals or the general issue of the role that ineffability might play in an argument for Humility. What I want to highlight here is the possibility of using the ineffability of ignorance in a strategy for rejecting Humility. On this line of reasoning, ineffable ignorance is no real ignorance. Since there is no sentence identifying the fundamental properties whose truth value we are ignorant of, it follows that we are not ignorant of the identity of the fundamental properties.⁶ This line of reasoning would provide another route to the rejection of Premise 9. Our observations may not favour A over the other possible realizations of T, but this is no obstacle for knowing the identity of the actual realization of T. The tuple that does the job is identified by the sentence ‘T is realized by the tuple that actually realizes it’, and I know that this sentence is true.

9.4 Quidditism and Combinatorialism Another point at which Lewis’s argument is open to attack is his derivation of Variance. His reasoning here rests on two metaphysical principles: combinatorialism and quidditism. The involvement of combinatorialism in the argument is perfectly straightforward—it is used directly as a premise (Premise 2). The role of ⁵ David Yates refers to arguments for Humility based on these considerations as semantic (Yates 2018). ⁶ Shamik Dasgupta (Dasgupta 2015) has developed and rejected this line of reasoning, using an analogy between knowledge of the identity of the fundamental properties and knowledge of the positions of objects in Newtonian space.

   

191

quidditism is less clear. Quidditism is the view that one and the same property can be present at different worlds (Black 2000: 92). This principle does not figure as a premise in the argument. Its role is to provide support for one of the premises— Premise 3. But quidditism is not required for establishing Premise 3. This premise can be easily established without using quidditism, from other principles that Lewis accepts. The alternative derivation proceeds as follows, where A is the unique actual realization of T and B is a T-compliant permutation of the components of A: i.

The components of A are different from each other. (Lewis’s assumption)

ii. B is different from A. (from i) iii. T is uniquely realized in actuality. (Lewis’s assumption) iv. B doesn’t realize T in actuality. (from ii and iii) v.

Any possible world in which B realizes T is not identical with the actual world. (from iv)

And v is equivalent to Premise 3. This argument doesn’t rest in any way on quidditism. It doesn’t assume that the components of B in one world have to be numerically identical with the components of B in another world, rather than their numerically distinct images under the counterpart relation. The availability of this alternative argument means that rejecting quidditism would not deprive Lewis of Premise 3. Nevertheless, rejecting quidditism might still undermine the argument for Variance in a different way. It could be argued that if quidditism is rejected, combinatorialism loses all plausibility. If quidditism is false, talk of trans-world identity of properties will have to be understood as concerning not numerical identity, but a counterpart relation between numerically different properties inhabiting different worlds. It follows that a world in which B realizes T is a world in which T is realized by the tuple formed by the counterparts of the components of B. Now, whether a world satisfies this description depends on what properties in that world are paired with the components of B by the counterpart relation. I’m going to argue that it’s hard to see how the pairings could result in a tuple that realizes T. To see this, suppose that m is the property that plays the mass role in A and c is the property that plays the charge role in A. Suppose also that B is identical to A except that m and c have swapped positions. Now, a world in which B realizes T is a world in which the counterpart of m plays the charge role and the counterpart of c plays the mass role. But on no plausible account of the counterpart relation would it produce these pairings. We should expect instead that for any world w in which T is uniquely realized, the counterpart of m in w will be the property that plays the mass role in w, and the counterpart of c will be the property that plays the charge role in w. This would certainly be the outcome on the property counterpart relation that Lewis expects the anti-quidditist to

192

 

adopt: ‘two world-bound properties in two worlds are counterparts if the role of one in one world approximately matches the role of the other in the other’ (Lewis 2009: 211). On this plausible assumption about the counterpart relation, it follows that there is no possible world in which T is realized by B or, in general, by any permutation of the components of A. Hence, on this assumption, if quidditism is rejected, combinatorialism (Premise 2) will also have to go.⁷ I conclude that the rejection of quidditism would block the argument for Variance by undermining not Premise 3, but Premise 2.⁸ It might help to bring into focus my claim about the role of quidditism if we go back to how Lewis proposes to use it. According to Lewis, quidditism tells us that if we permute same-category properties in the actual realization of T ‘the permutation is indeed a different possibility’ (Lewis 2009: 209). I’ve argued that we can show, without invoking quidditism, that if the permutation is a possibility at all it is a different possibility. But if quidditism is false, the permutation won’t be a possibility, and combinatorialism will fail. I’ve argued that rejecting quidditism would undermine the argument for Variance by forcing us to reject combinatorialism. But anti-quidditism is not the only route to the rejection of combinatorialism. According to quidditism, the components of A themselves, not just their counterparts, might be present at non-actual possible worlds. This makes room for the possibility that at one of these worlds T will be realized by a given permutation of the components of A. But quidditism in no way entails that this will be the situation for any permutation of the components of A, let alone for every such permutation, as combinatorialism dictates. Quidditists might want to join anti-quidditists in rejecting combinatorialism. Hence a challenge to Lewis’s argument that targets combinatorialism directly will be of wider appeal than one that targets quidditism in the first instance.⁹

⁷ I don’t think my argument for this claim relies on a particularly stringent account of the counterpart relation for properties. On a permissive account of the relation (or in an undemanding context, on a contextualist account of the relation), there might be a world in which the counterpart of m plays the charge role—provided that there isn’t a better candidate at that world. What’s hard to see is how on any account of the counterpart relation, however lax, in a world that contains a property playing the mass role, m could have as its counterpart the property playing the charge role. See (Heller 1998) for an excellent discussion of a non-quidditist conception of properties. ⁸ In any case, for the purposes of my assessment of Lewis’s argument, what matters is not that rejecting quidditism would undermine Premise 2, but that this is the only way in which the move could block the argument. ⁹ Quidditism follows naturally from the Kripkean conception of possible worlds, according to which ‘ “[p]ossible worlds” are stipulated, not discovered by powerful telescopes’. On this conception, as Kripke points out, ‘[t]here is no reason why we cannot stipulate that, in talking about what would have happened to Nixon in a certain counterfactual situation, we are talking about what would have happened to him’ (Kripke 1980: 44). The quidditist would add, in the same spirit, that there is no reason why we cannot stipulate that, in talking about how mass would behave in a certain counterfactual situation, we are talking about how it would behave.

   

193

9.5 Ramseyan Structuralism The difficulties for representationalism that I’m considering in the case of theoretical science are based on the idea that we can know the referents of theoretical terms only through the role they occupy in the theories in which the terms figure. All we know is these roles, and this is not sufficient for knowing which properties occupy them. There’s a way in which one might try to address this difficulty that is akin in spirit to the strategies we considered in Section 4.1 for other problematic discourses. If all we know about the referents of theoretical terms is the roles they occupy, we would render their identity knowable by thinking of role occupancy as essential to their nature. This is the view known as structuralism. Structuralism was originally formulated in terms of causal roles (Shoemaker 1980, 1998; Hawthorne 2001). However, as Nora Berenstain has argued (Berenstain 2016), the focus on causal relations detracts from the appeal of the view, and a much more promising position results if we take the essence of properties to be constituted by all their nomological links, non-causal as well as causal. Structuralism seems to offer a promising strategy for resisting Lewis’s reasoning. It is generally seen as achieving this by challenging quidditism, but we’ve just seen that this wouldn’t be an effective approach. However, the view is more accurately characterized as challenging combinatorialism. Since the fundamental properties are components of the actual realization of T, the view can be formulated as concerning this tuple, as the view that realizing T is necessary and sufficient for its identity. We can easily provide a more precise formulation of the view. Start with sufficiency, the claim that realizing T is sufficient for the unique actual realization of T is the claim that the actual realization of T is its only possible realization. In other words, if A is the unique actual realization of T: S For every possible world w and every n-tuple X, if X realizes T at w, then X = A. Notice that S rules out the possibility that T is multiply realized in any possible world, but this seems to be a consequence of the claim that the nomological profile of properties is sufficient for their identity. The claim that realizing T is necessary for A is the claim that A couldn’t fail to realize T. One reading of this is as the converse of S: For every possible world w and every n-tuple X, if X doesn’t realize T at w, then X ≠ A, which is equivalent to: For every possible world w, A realizes T at w.

194

 

On this construal of the necessity claim, it has the obvious consequence that T is a necessary truth, but this consequence can be avoided by restricting the claim to worlds in which the components of A are instantiated: N

For every possible world w, such that the components of A are instantiated in w, A realizes T at w.

I’m going to refer to the conjunction of S and N as Ramseyan structuralism. Ramseyan structuralism poses a direct challenge to combinatorialism. According to combinatorialism, every T-compliant permutation of components of A is a possible realization of T. According to S, no permutation of components of A other than A itself is a possible realization of T. This blocks the transition from Premise 1 to Variance. There may be multiple T-compliant n-tuples of fundamental properties, but according to S exactly one of them is a possible realization of T—namely A itself, and Variance doesn’t hold.¹⁰ But Variance is an essential lemma in Lewis’s argument. Hence, if we adopt Ramseyan structuralism, Lewis’s argument for Humility can no longer be run. I’m going to argue, however, that this strategy is ultimately unsuccessful. Although it succeeds in blocking Lewis’s argument, it is powerless against a minor modification of his reasoning. The new argument for Humility bypasses Variance and the metaphysical principle that Lewis invokes to establish it. It runs Lewis’s epistemological reasoning not from Variance, but directly from the claim (Premise 1) that there are multiple T-compliant permutations of the components of A. This results in an argument with the following structure: 1

There are multiple T-compliant permutations of the components of A. (Premise)

5*

For every T-compliant permutation X of the components of A, if X is the actual realization of T, the Ramsey sentence of T is true. (Premise)

6

If two hypotheses make the Ramsey sentence of T true, then no possible observation favours one over the other. (Premise)

7*

For every T-compliant permutation X of the components of A, no possible observation favours A over X as the unique actual realization of T. (from 5* and 6)

8*

For some T-compliant permutation X of the components of A, no possible observation favours A over X as the unique actual realization of T. (from 1 and 7*)

¹⁰ Clearly, N is not involved in the argument against Variance. Hence the argument can be endorsed by positions according to which realizing T is sufficient but not necessary for the identity of A. On these positions, the components of A are instantiated in worlds in which A doesn’t realize T.

   

195

9*

If for some T-compliant permutation X of A, my observations don’t favour A over X as the unique actual realization of T, then I don’t know the identity of the unique actual realization of T. (Premise)

10

HUMILITY: I can’t know the identity of the unique actual realization of T. (from 8* and 9*)

The new argument is valid. If its premises are true, Humility follows. Resisting its conclusion requires rejecting at least one of its premises. Our options here are more restricted than in Lewis’s permutation argument. We are still assuming that 1 is true, and 5* (like 5) and 6 follow from Lewis’s definitions. The only candidate for rejection is 9*. There’s one clear route to the rejection of 9*. Premise 9 of the original argument is a logical consequence of 9*. If knowing the identity of the unique actual realization of T requires observations that rule out all other T-compliant permutations of A, as 9* dictates, then, a fortiori, it requires ruling out those T-compliant permutations of A that are possible realizations of A. Hence anyone who rejects 9 will also reject 9*. However, this route to rejecting 9* is of no interest for assessing the potential of Ramseyan structuralism for resisting Humility. If Premise 9 is false, then Lewis’s argument for Humility is unsound, and adopting Ramseyan structuralism is not needed for blocking it. What we are interested in is the possibility of rejecting 9* while accepting 9. Only if this is a coherent possibility will Ramseyan structuralism be efficacious in saving us from Humility. There’s room in principle for this approach, as the necessary condition for knowledge set by 9* is stronger than the one set by 9. 9 requires that my observations favour A over every other possible realization of T. 9* requires, in addition, that my observations favour A over those T-compliant permutations of the components of A that are not possible realizations of T. So long as you are prepared to accept that some T-compliant permutations of the components of A are not possible realizations of T, rejecting 9* is logically compatible with accepting 9. I’m going to argue, however, that this isn’t a coherent position. I’m going to make this case by highlighting the parallels between Ramseyan Structuralism and claims to the effect that certain properties of an object are essential to it. Consider, for example, the claim that the substance of which a material object was made is essential to it (Kripke 1980: 114–15). According to this position (call it substance essentialism), my desk couldn’t have been made of any substance other than the substance it was made of. Since it was made of wood, it couldn’t have been made of steel, plastic, etc. It was made of wood in every possible world in which it exists. I want to highlight two aspects of the contrast between substance essentialism and the opposite view (call it substance accidentalism), according to which my desk could have been made of other substances. Notice, first, that there’s a range of propositions that for the substance accidentalist represent metaphysical possibilities which according to the substance essentialist represent metaphysical

196

 

impossibilities. Thus, the proposition that my desk was made of steel represents a possible state of affairs for the substance accidentalist, but for the substance essentialist there is no possible world in which the proposition is true. In general, propositions representing my desk as made of a substance other than wood represent possible states of affairs for the substance accidentalist, but for the substance essentialist they represent metaphysical impossibilities. However, and this is the first point I want to highlight, the substance essentialist still has to recognize these as genuine, distinct propositions. They are needed, among other things, to provide the contents of propositional attitudes, since even if my desk could only have been made of one substance, it is perfectly possible for someone to believe of my desk that it was made of some other substance. The second point I want to highlight in this connection is that the substance essentialist can claim no epistemological advantage over her opponent. The claim that the substance an object was made of is essential to it doesn’t make it in any way easier to know what substance an object was made of. The conditions that a subject needs to satisfy in order to know that my desk was made of wood if substance essentialism is true are exactly the same as the conditions she would need to satisfy if substance essentialism were false. If knowing that it was made of wood would require evidence to rule out the hypothesis that it was made of steel if this hypothesis represented a possible state of affairs, it would still need this evidence if the hypothesis represented an impossible state of affairs. If it would require evidence to rule out the hypothesis that I’m a deskless envatted brain if this hypothesis represented a possible state of affairs, it would still need this evidence if the hypothesis represented an impossible state of affairs. Even knowledge that substance essentialism is true would be no help to the subject. The fact that my desk was made of the same substance in every possible world does not tell me which substance this is. There are important parallels between substance essentialism and Ramseyan structuralism. Both theses claim that the way things are in actuality in a certain respect is the way things are in every possible world. While the substance essentialist claims that my desk was made of the same substance in every possible world in which it exists, the Ramseyan structuralist claims that T is realized by the same tuple in every possible world in which it is realized. I want to argue that the two points I’ve highlighted concerning substance essentialism have equally plausible correlates concerning Ramseyan structuralism. According to combinatorialism, there are propositions representing T as realized by tuples other than the one that realizes T in actuality. They represent possible states of affairs—there are possible worlds in which they are true. According to the Ramseyan structuralist, these propositions represent metaphysical impossibilities—there are no possible worlds in which any of them is true. However, as with substance essentialism, this is no reason for denying the existence or distinctness of these propositions. If we think, as some authors have suggested (Kelly 2013; Dasgupta 2015), that these propositions are inexpressible

   

197

or ungraspable, we might question their suitability as the contents of propositional attitudes or the meanings of sentences. But this point applies independently of the metaphysical status of the states of affairs represented by these propositions. Treating them as metaphysical impossibilities doesn’t undermine the claim that they are genuine, distinct propositions. The second point is equally plausible. The Ramseyan structuralist can claim no epistemological advantage over her opponent. If she has views on the conditions that a subject would need to satisfy in order to know the identity of the actual realization of T if combinatorialism were true, her rejection of combinatorialism won’t abrogate these conditions. The same conditions need to be satisfied for knowing the identity of the actual realization of T whether or not there are possible worlds in which T is realized by other tuples. We now have what we need to establish the point that rejecting 9* is not an option for someone who accepts 9, i.e. the claim that knowing the identity of the unique actual realization of T requires evidence that favours A over every other possible realization of T as the unique actual realization. Combinatorialism says that every T-compliant permutation of the components of A is a possible realization of T. Hence 9 entails that if combinatorialism were true, knowing the identity of the unique actual realization of T would require evidence that favours A over every T-compliant permutation of its components as the unique actual realization of T. Hence 9 entails: E.

If every T-compliant permutation of the components of A were a possible realization of T, then knowing the identity of the unique actual realization of T would require evidence that favours A over every T-compliant permutation of its components as the unique actual realization of T.

Now we can bring to bear the point we made above: the same conditions need to be satisfied for knowing the identity of the actual realization of T whether or not there are possible worlds in which T is realized by other tuples. If knowledge of the identity of the actual realization of T required evidence that favours A over a T-compliant permutation B of the components of A if there were a possible world in which B realizes T, then knowledge of the identity of the actual realization of T would still require evidence that favours A over B if there were no possible worlds in which B realizes T. It follows from this point that E entails: Knowing the identity of the unique actual realization of T requires evidence that favours A over every T-compliant permutation of its components as the unique actual realization of T. But this is, of course, Premise 9* of the new argument. 9 entails 9*. Accepting the former and rejecting the latter is not a coherent position.

198

 

Summing up, Ramseyan structuralism sustains a strategy for resisting Humility only if the new argument is unsound. The new argument is unsound only if 9* is false. But 9* is false only if 9 is also false. And if 9 is false, Lewis’s original argument is unsound even if we accept its metaphysical premises. Ramseyan structuralism is of no help in avoiding Humility. The only viable strategy for resisting Lewis’s reasoning is rejecting Premise 9. For those who accept Premise 9, Humility is inescapable.

9.6 Other Options Theoretical science has also received construals along the lines of the two approaches considered in Chapter 2 for our original problematic discourses. One formerly influential option consists in interpreting scientific statements as representing not unobservable states of affairs, but the observable or phenomenal states of affairs that we regard as evidence in their support. Thus, for example, the sentence ‘an electron has just passed through the cloud chamber’ would represent the state of affairs of a white streak appearing in the cloud chamber. This position, known as verificationism, preserves the thought that the function of scientific statements is to represent the world. It only differs from standard realist construals in which aspects of the world scientific statements are taken to represent— they represent, according to the verificationist, the observable states of affairs that we treat as evidence for them. This is not the place to discuss the difficulties faced by verificationist construals of theoretical science.¹¹ I agree with the current consensus that the position is untenable. Theoretical science has also received non-representationalist construals, according to which its goal is not to represent things as being a certain way. This is the central idea of the position in philosophy of science known as instrumentalism.¹² According to instrumentalism, the function of scientific theories is not to represent how things stand in unobservable regions of the world. Its function is to serve as a tool for the prediction of observable states of affairs. The success of a scientific theory should be measured exclusively by the accuracy of the observable predictions it generates.¹³ Like the verificationist, the instrumentalist explains the meaning of scientific statements in terms of their observable consequences, but whereas the verificationist construes them as representing their observable consequences as obtaining, for the instrumentalist scientific statements

¹¹ See (Carnap 1967) for the most sophisticated implementation of the programme. Influential critics include (Quine 1980; Putnam 1963). ¹² The contrast between using a theory to predict observable states of affairs and using it to represent these states of affairs as obtaining is by no means straightforward. See (Nagel 1961). ¹³ See (Stanford 2016) for an overview.

   

199

don’t represent their observable consequences—their function is not to represent, but to generate successful observable predictions.¹⁴ Versions of instrumentalism enjoy some support in contemporary philosophy of science, and assessing their merits falls outside the scope of this book. I only want to point out here that the adoption of a non-representational construal of theoretical science has important consequences for our general account of the representational function of language. Consider first sentences involving natural-kind terms, like, e.g., ‘the liquid in this glass is water’ or ‘the trophy is made of gold’. Sentences like these are very high on the list of sentences that we think of as successfully representing things as being a certain way. However, in order to achieve this, the predicates that figure in them, ‘is water’, ‘is made of gold’, have to have properties playing the role of their referents. We think that the job of identifying the properties playing this role belongs to science, in sentences such as ‘ “is water” refers to the property of consisting mostly of H₂O molecules’ or ‘ “is made of gold” refers to the property of being made of the substance with atomic number 79’. Now, on a nonrepresentational construal of theoretical science, there are no properties that the predicates ‘consists mostly of H₂O molecules’ or ‘is made of the substance with atomic number 79’ refer to. Hence the sentences in which we seek to use them to identify the referents of ‘is water’ and ‘is made of gold’ achieve no such thing. This makes it hard to avoid the conclusion that sentences ascribing natural-kind predicates are no more in the business of representing the world than sentences involving the scientific terms with which we seek to specify their referents. Furthermore, on a prominent version of representationalism that we will consider later on, the referent of every representational predicate follows the model of natural-kind predicates—their referents have to be found in the catalogue of properties postulated by science—i.e. among the referents of the predicates of the vocabulary of science. If science is not in the business of representing the world and its predicates have no referents, then it is hard to see how this approach can avoid the conclusion that no sentence of any discourse succeeds in representing the world.

9.7 Pragmatist Meaning Grounds for Theoretical Science Assume that some version of Lewis’s reasoning succeeds in establishing that we can’t identify the referents of the predicates of the vocabulary of theoretical science. I have argued that this outcome would seriously detract from the appeal

¹⁴ Some versions of the instrumentalist approach hold on to the idea that scientific statements represent unobservable states of affairs, but maintain that the accuracy of these representations is not among their goals. See (van Fraassen 1980), and (Rosen 1994; Blackburn 2002) for critical discussion.

200

 

of a representationalist account of the meaning grounds of the sentences of theoretical science, since these would be specified in terms of items to which we can have no access. I want to argue that here, as in our original problematic discourses, the pragmatist strategy for specifying meaning grounds offers an appealing alternative. Applied to our simple example from Section 9.6, as a first approximation, the proposal would be that what makes the sentence ‘an electron has just passed through the cloud chamber’ have the meaning it has is the fact that its acceptance is regulated by whether a white streak has appeared in the cloud chamber. Needless to say, this application of the pragmatist template is unrealistically simplistic. There isn’t in general a one-to-one correlation between sentences about unobservable entities and states of affairs and the observable states of affairs on which we ground our decisions on the acceptance of those sentences. The links between these sentences and observation reverberate throughout the whole theory introducing the terms with which we represent the unobservable. A more realistic account would have to make reference to the relevant theory to specify the procedure for regulating the acceptance of the target sentences from which these sentences obtain their meaning. We achieve this if we say that what makes the sentence ‘an electron has just passed through the cloud chamber’ have the meaning it has is the fact that its acceptance is regulated by what our theory of elementary particles treats as observable consequences of the sentence. As we’ve seen in Chapter 8, adopting a pragmatist account of the meaning grounds of theoretical predicates is compatible with thinking of them as referring to properties, so long as we are prepared to treat these properties as identified with a right-to-left reading of an abstraction principle. For this we need to specify necessary and sufficient conditions for a predicate to be C-synonymous with a given theoretical predicate. I want to outline a basic strategy for achieving this by returning to Lewis’s framework, simplified in one major respect. We are going to focus on a scientific theory with a predicate, ‘ϕ’, as its only T-term. Let S be a speaker who accepts T. Our goal is to identify a feature of S’s acceptance procedure for sentences ascribing ‘ϕ’ whose presence in the acceptance procedure of a speaker S0 for her predicate ‘ϕ0 ’ can be treated as necessary and sufficient for ‘ϕ0 ’, as understood by S0 , to have the same referent as ‘ϕ’, as understood by S. Let T0 be a theory accepted by S0 with ‘ϕ0 ’ as the only T0 -term. When does ‘ϕ0 ’, as understood by S0 , have the same referent as ‘ϕ’, as understood by S? Our basic provisional answer to this question is that ‘ϕ’ and ‘ϕ0 ’ have the same referent when T and T0 are the same theory. To implement this idea, we need to provide a non-circular account of when T and T0 count as the same theory. Suppose that there is a bijection h between the extralogical vocabulary of T and the extralogical vocabulary of T0 such that h(‘ϕ’) = ‘ϕ0 ’ and every O-term t of T has the same referent as h(t). If s is a sentence of the language of T, let the O-translation of s into the language of T0 be the sentence that

   

201

results when we replace every term in T with its image under h. Now we can say that T and T0 are O-equivalent just in case, for every sentence s of the language of T, s is in T just in case the O-translation of s is in T0 . For our purposes, T and T0 will count as the same theory when they are O-equivalent. Intuitively, if T and T0 are O-equivalent, T will say about the referent of ‘ϕ’ what 0 T says about the referent of ‘ϕ0 ’. Hence we can express the fact that T and T0 are O-equivalent by saying that ‘ϕ’ and ‘ϕ0 ’ are O-synonymous. My provisional proposal is to take O-synonymy as C-synonymy for theoretical terms. This yields the following abstraction principle for theoretical predicates: CT

For all theoretical predicates P, Q, the referent of P = the referent of Q iff P and Q are O-synonymous.

CT would then enable us to identify the referent of any given theoretical predicate. In this way we would achieve what Lewis’s argument shows to be impossible on a representationalist account of the meaning grounds of theoretical terms. One problem with CT is that it is in conflict with an important aspect of our conception of the reference of theoretical terms: theory change doesn’t always change the referents of terms, even if the theories before and after the change are not O-equivalent. We also think that terms introduced independently by different, O-inequivalent theories can have the same referents. In order to accommodate this feature, we would need to replace O-synonymy in CT with the relation generated by our criteria of inter-theoretic co-reference, on the assumption that it is close enough to being an equivalence relation.¹⁵ Notice that this approach wouldn’t violate the non-circularity constraint. We don’t determine whether two theoretical predicates co-refer by ascertaining which property each of them refers to and checking whether they are the same. Our criteria for co-reference for theoretical terms operate on a direct comparison between the theories in which the terms figure.

9.8 A Pragmatist Progress I want to present the proposal that emerges from these ideas by reference to an influential account of linguistic representation, sometimes known as the

¹⁵ Our criteria for inter-theoretic co-reference are bound to be vague. Hence the relation of co-reference they define won’t be an equivalence relation. However, it could be close to being an equivalence relation in the following sense. Focusing on theories with a single T-term, if we take them to be arranged in a similarity space by our criteria of co-reference, we can turn the relation into an equivalence relation by removing from its domain the theories in the borderline areas. The smaller the area we need to remove to achieve this, the closer to being an equivalence relation the original relation will be.

202

 

Canberra Plan.¹⁶ I’m going to concentrate on the version put forward by Frank Jackson in (Jackson 1998). The problem Jackson is concerned with is very similar to the problem the present book is dealing with—discourses that appear to perform the function of representing the world but don’t satisfy conditions that seem to be necessary for achieving this. We can characterize Jackson’s construal of the problem as arising from two main assumptions. The first is a version of the idea that if a predicate is to perform the function of representing the world, there has to be a property it refers to—a property that we represent an object as instantiating when we ascribe the predicate to it. Here’s Jackson’s gloss on the assumption: I am simply supposing that predicates apply in virtue of how things are; if a predicate applies to one thing and not to another, this is because of something about how the two things are over and above the fact that the predicate applies to one and not the other. This supervenience of predication on nature is required for predicates to serve the purpose of saying how things are. Property talk, as we will be understanding it, is a way of talking of the nature on which predication supervenes: thus, ‘being F ’ or ‘the property of being F ’ picks out the nature in virtue of which ‘F ’ applies. (Jackson 1998: 15–16)

The second assumption that generates Jackson’s problem is physicalism—the view that the only properties the world contains are physical properties, which he characterizes as follows: physicalists can give an ostensive definition of what they mean by physical properties and relations by pointing to some exemplars of non-sentient objects—tables, chairs, mountains, and the like—and then say that by physical properties and relations, they mean the kinds of properties and relations needed to give a complete account of things like them. (Jackson 1998: 7)

Physical properties are then the properties postulated by the sciences that provide this complete account, or, equivalently, the referents of the predicates in the basic vocabulary of these sciences. It follows from these two assumptions, that if a predicate can do the job of representing the world, there’s got to be a physical property it refers to. Hence, if we want to vindicate the representational character of a discourse, we need to identify physical properties that its predicates can be plausibly said to refer to. The discourses for which this is a difficult task include, of course, the semantic discourses we’ve been concerned with:

¹⁶ For the label, see (Hawthorne and Price 2011).

   

203

Where, if anywhere, are the semantic properties of truth, content, and reference to be found in the non-semantic, physical or naturalistic account of the sentence? (Jackson 1998: 2)

They also include ethical discourse, the main focus of Jackson’s book. Jackson proposes a general strategy for locating the properties that can do the job in each case. His idea is that we should take as our starting point our folk theory involving a problematic predicate ‘P’, as revealed by our intuitions concerning which possible cases to describe using ‘P’. We would then use this theory to locate ‘P’s referent. If there is a physical property ∏ such that (a) if ‘P’ referred to ∏, our ‘P’-theory would approximate truth sufficiently and (b) every other physical property is such that if ‘P’ referred to it, our ‘P’-theory would be significantly further from the truth, we should conclude that ‘P’ refers to ∏. If there’s no physical property satisfying (a), we should conclude that ‘P’ has no referent, and hence that it isn’t representational. If there are physical properties satisfying (a) but none satisfying (a) and (b) the situation is more complicated, but we don’t need to concern ourselves with this problem here.¹⁷ The pragmatist strategy for specifying meaning grounds that I have developed in the preceding chapters sustains two modifications of this general picture. The first concerns the discourses for whose predicates physical referents can’t be easily found. According to the Canberra plan, if the search for suitable physical referents for the predicates of one of these discourses were unsuccessful, we would have to conclude that, appearances to the contrary, the discourse doesn’t manage to represent things as being a certain way. Then, as we saw in Chapter 4, we would have to choose between ascribing to the discourse a different, nonrepresentational function and maintaining that it has the function of representing the world but doesn’t manage to discharge it. The availability of the pragmatist strategy for specifying meaning grounds enables us to avoid this outcome. If we construe the meaning grounds of the problematic discourses along the lines of the pragmatist strategy, we can hold on to the idea that their sentences succeed in representing the world. We can agree with Jackson that this requires that the predicates of these discourses refer to properties, but this constraint can be satisfied with the properties we can identify by abstraction on the basis of our ascription procedures for the relevant predicates. Taking this line doesn’t force us to abandon physicalism, as the claim that the physical sciences provide a complete description of what the world is ultimately

¹⁷ Huw Price provides insightful critical discussion of the Canberra plan in (Price 2011a, 2011b). In his discussion of the semantics of the sentence ‘That’s cool’ in (Price 2011b), he characterizes his alternative to Jackson’s approach in terms that I find very congenial, except that in his view the alternative enjoins a commitment to functional pluralism.

204

 

like. Upholding physicalism simply requires maintaining that the scope of what the world is ultimately like is much narrower than that of what the world is objectively like. There is an objective difference between the 134 bus and the 43 bus, between a bull market and a bear market, between a Shakespearean sonnet and a Petrarchan sonnet, or between a democracy and a dictatorship, and this is so even though these objective differences can’t be realistically recovered from the physical account of what the world is ultimately like. Similarly, property talk should not be a problem in these cases. The property of being a 134 bus is the nature in terms of which the predicate ‘is a 134 bus’ applies, irrespective of whether it can be construed in terms of the properties postulated by the physical sciences. The idea that the world is ultimately physical could be given meaning in terms of a notion of supervenience, as the claim that states of the world that are identical in all physical respects are identical in all respects. So long as we accept this contrast between the narrower notion of what the world is ultimately like and the wider notion of what the world is objectively like, physicalism is perfectly compatible with treating predicates with pragmatist meaning grounds as referring to the properties we can identify by abstraction on the basis of ascription procedures. Then the physicalist can accept that there is a property of being true, of being morally right, of meaning table or of believing that’s a horse, even if none of these properties can be recovered from the account of reality provided by the physical sciences. The pragmatist strategy for specifying meaning grounds sustains a second, more radical modification of the picture presented by the Canberra plan. It consists in applying the strategy to the basic predicates of the physical sciences. As we have seen, here the problem for representationalism is not posed by the requirement that only physical properties can play the role of referents. If there are properties playing the role of referents with respect to these predicates, they are by definition physical. The problem in this case concerns our ability to identify the properties these predicates refer to, on a representationalist account of their meaning grounds. I have argued that the problem disappears if we adopt a pragmatist construal of the meaning grounds of these predicates. Then we can explain our ability to identify their referents in terms of the abstraction principles generated by our procedures for ascribing these predicates, or, more specifically, in terms of our criteria for co-reference for theoretical terms. This second modification of the Canberra plan has profound repercussions. According to the Canberra plan, our access to the properties that play the role of referents for our predicates is always ultimately based on our access to properties for which we can provide explicit definitions in terms of the language of the physical sciences. By endorsing pragmatist specifications of the meaning grounds of ethical and semantic discourses, we countenance a second method for gaining cognitive access to the properties that serve as referents of our predicates— abstractionist identifications based on the ascription procedures that figure in

   

205

the pragmatist meaning grounds of the relevant predicates. If the meaning grounds of the predicates of the physical sciences are specified in terms of the pragmatist method, our whole cognitive access to the properties that play the role of the referents of our predicates is ultimately explained in terms of abstraction. Abstraction becomes our primary means of access to the properties our predicates refer to. On this picture, explicit definitions of referents still play a role, including explicit definitions in terms of the language of the physical sciences, as in the case of natural-kind predicates. However, on the new picture, what these explicit definitions achieve is to pair the target predicates with properties to which we ultimately gain cognitive access through abstraction principles. On this picture, abstraction is ultimately our only method for gaining cognitive access to the referents of our predicates.

9.9 Pragmatist Meaning Grounds and Genuine Representation In Section 4.3 I argued that the quasi-realist approach doesn’t sustain a successful vindication of the representational character of the target discourses. The reason is that, by maintaining that the meaning grounds of the target sentences include a non-representational function, quasi-realism preserves a two-tier picture, of genuinely representational discourses with no function other than representing the world and discourses that we speak of as representational, although a non-representational function is essential to their meaning. I have argued that the pragmatist approach I’m defending doesn’t exhibit this shortcoming. Pragmatist meaning grounds, like representationalist meaning grounds, include no function other than representing the world—there’s no non-representational function that a sentence with a pragmatist meaning ground has to perform in order to have the meaning it has. Consequently, I’ve argued, there’s no sense in which the representational credentials of a sentence would be undermined by having a pragmatist meaning ground. One could argue, however, that my proposal faces a different version of the same worry. I argued in Chapter 8 that predicates with pragmatist meaning grounds, just as those with representationalist meaning grounds, refer to properties, so that a sentence that ascribes a predicate with a pragmatist meaning ground to an object will represent the object as instantiating the property the predicate refers to, in exactly the same way as a sentence ascribing a predicate with a representationalist meaning ground. However, it’s still the case that our access to the referents of predicates with pragmatist meaning grounds is different from out access to the referents of predicates with representationalist meaning grounds. Our access to the latter takes the form of explicit definitions, whereas the former can only be accessed through our grasp of abstraction principles. I have argued that this difference between two forms of cognitive access should not be seen as corresponding to an ontological difference between two kinds of

206

 

properties. On my proposal, the ontological status of a property is not affected by whether we can gain cognitive access to it with an explicit definition or we can access it only through an abstraction principle. However, this point could be called into question. It could be argued, to the contrary, that the only genuine properties are those for which explicit definitions can be provided, while properties that we can access only through abstraction principles have a more tenuous ontological status, with their existence dependent on the abstraction process through which we gain access to them. On this view, the form of representation achieved by a predicate with a representationalist meaning ground will still be superior to what can be achieved by a predicate with a pragmatist meaning ground. By ascribing the former to an object, we represent the object as instantiating a genuine objective property, whereas with the latter we can only represent objects as falling under the mental constructs that abstraction principles generate. If this complaint is legitimate, then the advantage I’ve claimed for my position over quasi-realism evaporates. Discourses with pragmatist meaning grounds will not be capable of the higher grade of representation that discourses with representationalist meaning grounds achieve. This complaint rests on the idea that a thick definition by abstraction should be seen as creating the items that the abstraction operator pairs with the elements of its domain, instead of granting us access to independently existing items. Then the abstraction principle associated with the meaning-grounding ascription procedure of a predicate will have to be seen as creating the property that is paired with the predicate as its referent, and this property will as a result not count as fully objective. This conception of what is achieved by thick definitions by abstraction is controversial. But if the pragmatist account of the meaning grounds of the terms of the basic sciences is accepted, we can bypass this controversy. In this situation, referents that can be identified with explicit definitions in terms of the language of the basic sciences can’t claim any ontological pre-eminence over those that we identify with abstraction principles. For those explicit definitions only take us back to predicates whose referents are identified by abstraction. In this scenario, which of the two methods of referent identification we employ for a given predicate cannot be treated as having any consequences for the ontological status of the referent, or for the possibility of using the predicate to produce objective representations of the world. There are no properties enjoying a higher ontological status than those that we identify through abstraction principles. And no linguistic representation of the world is more objective than what can be achieved with predicates with pragmatist meaning grounds. It also follows from this picture that the debate on whether the meaning grounds of a discourse should be specified along pragmatist or representationalist lines should not be seen as having any consequences for the level of objectivity that we attribute to the discourse. In particular, the aspiration to construe the discourse as fully objective should not be seen as recommending the representationalist

   

207

approach. Alternatively, we could maintain that questions concerning the level of objectivity of a specific discourse are really about nothing but which of the two strategies we should adopt for specifying the meaning grounds of its sentences.

9.10 A Regress Argument We have reached a picture on which our cognitive contact with the properties our predicates refer to is universally mediated by abstraction principles. Our route to this position was based on a Lewis-inspired argument to the effect that a representationalist account of the meaning grounds of the predicates of theoretical science would have undesirable consequences. In the present section I want to suggest that the same position can be reached through a much more direct route—one that doesn’t depend on the specific features of the language of theoretical science.¹⁸ This direct route is based on the reflection that explicit definitions of predicate referents are by their nature ultimately dependent on an alternative method for specifying predicate referents. An explicit definition of the referent of a predicate takes the following form: REF

‘ϕ’ refers to the property of ψ-ing.

In any instance of REF, two predicates are involved. One, the target predicate, is the predicate, ‘ϕ’, whose referent we are seeking to identify. The other, the identifying predicate, is the predicate, ‘ψ’, used for singling out the property that is to be assigned to the target predicate as its referent. I want to emphasise that I’m not claiming that what an instance of REF seeks to achieve is pairing a predicate, ‘ϕ’, with another predicate, ‘ψ’. Its goal is to pair a predicate, ‘ϕ’, with a property. But when this pairing is effected by an instance of REF, it involves our understanding of the identifying predicate, ‘ψ’. The property the target predicate, ‘ϕ’, is paired with, is the property that plays the role of referent with respect to the identifying predicate, as we understand it. The identifying predicate is a mere tool in the pairing of the target predicate with a property, but a necessary tool in the pairings effected by instances of REF. Not every instance of REF produces a successful identification of the referent of the target predicate. One obvious case of failure is the sentence: W

‘is water’ refers to the property of being water.

There is a sense in which W pairs the predicate ‘is water’ with a property as its referent. The singular term ‘the property of being water’, as we understand it, ¹⁸ I presented this argument in (Zalabardo 2019b).

208

 

refers to a property, and W, as we understand it, represents ‘is water’ as referring to that property. However, our understanding of W would enable us to pair the predicate with its referent only if we had already achieved this. Suppose that we haven’t. Imagine, for example, that we have a lay person’s understanding of the predicate ‘is water’ but not the knowledge of chemistry that is generally seen as required for grasping the identity of its referent. In this situation, W, as understood by us, will represent ‘is water’ as referring to the property of containing mostly H₂O molecules. However, we won’t be aware of this, and our understanding of W won’t enable us to gain cognitive access to the referent of ‘is water’. Notice that the real source of the problem with W is not that the identifying predicate is identical with the target predicate. We encounter the same difficulty in instances of REF in which the two predicates are different. Take, for example: AA

‘is Adam’s ale’ refers to the property of being water.

Once again, the sentence, as we understand it, represents the target predicate, ‘is Adam’s ale’, as referring to the property to which ‘is water’, as we understand it, refers—i.e. with the property of consisting mostly of H₂O molecules. However, due to our lack of chemical knowledge, we won’t be aware of the fact that this is the property with which AA pairs ‘is Adam’s ale’. Hence, our understanding of AA won’t succeed in giving us cognitive access to the referent of ‘is Adam’s ale’. Things are different, of course, if we know some chemistry, and have grasped the fact that ‘is water’, as we understand it, refers to the property of consisting mostly of H₂O molecules. In this situation, our understanding of AA will enable us to gain cognitive access to the pairing of ‘is Adam’s ale’ with the property of consisting mostly of H₂O molecules as its referent. These examples strongly suggest a necessary condition for understanding of an instance of REF to succeed in securing cognitive access to a pairing of the target predicate with a property as its referent: Understanding of an instance of REF can succeed in securing cognitive access to a pairing of the target predicate with a property as its referent only if the subject already has cognitive access to the pairing of the identifying predicate with the property to which it refers. I’m going to refer to this claim as the Basic Principle. The Basic Principle is all we need to undermine the idea that explicit definitions are our only means of cognitive access to predicate referents. A trivial regress argument shows that if understanding an instance of REF were the only method at our disposal for gaining cognitive access to a predicate-property pairing, cognitive access to these pairings would be impossible. Understanding of ‘ “ϕ” refers to the property of ψ-ing’ would provide you with cognitive access to the referent of ‘ϕ’

   

209

only if you already had cognitive access to the referent of ‘ψ’. This could be achieved in turn by your understanding of another instance of REF, with ‘ψ’ now as the target predicate. But if we had no other method for gaining cognitive access to the referent of a predicate, the ensuing chain could only extend to infinity, circle back upon itself or end with an instance of REF where you don’t have cognitive access to the referent of the identifying predicate. If cognitive access to predicate referents is possible, then there’s got to be some other way of achieving it. What method of referent grasp could play this role? One possibility is to postulate a faculty for direct grasp of properties, enabling us to bring a property to consciousness without the mediation of a predicate we understand. We could use this faculty to pair some predicates with their referents directly, without using an identifying predicate. These predicates could then form the base of a referent-access pyramid, serving as identifying predicates in instances of REF, whose target predicates could then be used as identifying predicates in further instances of REF. I don’t find this route very appealing. The idea of unmediated access to properties is deeply problematic, and in any case it’s only remotely plausible for a very restricted range of properties—mainly the sensory qualities of the empiricist tradition. In the first half of the twentieth century, some philosophers endorsed projects that would have made it possible to identify the referents of all representational predicates using identifying predicates referring to sensory qualities.¹⁹ If these projects had been successful, the idea that we enjoy unmediated access to sensory qualities would sustain a promising account of how we achieve referent grasp for our representational predicates. However, these hopes have been largely abandoned, and it is widely accepted that there’s only a very limited range of predicates whose referents can be identified using predicates that refer to sensory qualities. Hence, even if we accept that we have unmediated access to sensory qualities, the problem of how we gain cognitive access to the referents of our representational predicates will remain largely unsolved. The abstraction method for identifying the referents of predicates with pragmatist meaning grounds offers a more promising strategy for supplementing the explicit-definition method. On the resulting picture, our primary method for gaining cognitive access to predicate referents consists in the abstraction principles sustained by pragmatist specifications of the meaning grounds of predicates. Explicit definitions then enable us to extend the scope of our cognitive access to referents to predicates with representationalist meaning grounds. The picture outlined in Section 9.8 is an instance of this approach, in which the range of predicates to whose referents we gain access by abstraction comprises the predicates of the language of theoretical science and those other representational

¹⁹ The locus classicus of this project is (Carnap 1967). See also (Lewis 1929).

210

 

predicates whose referents cannot receive explicit definitions using identifying predicates drawn from the vocabulary of the physical sciences.

9.11 Tu quoque? According to the Basic Principle, your understanding of an instance of REF will give you cognitive access to the referent of the target predicate only if you antecedently have cognitive access to the referent of the identifying predicate. If this is accepted, saving our cognitive access to predicate referents from the threat of an infinite regress would seem to require accepting additional methods for achieving this which don’t rest on our cognitive access to the referents of other predicates. I claim that the abstractionist method I’ve outlined satisfies this requirement. It enables you to gain cognitive access to the referent of a predicate that doesn’t rely on your cognitive access to the referents of other predicates. This claim can be called into question. One might contend that the standards that the Basic Principle imposes on the identifying predicate of an instance of REF should also be applied to the predicates used in abstractionist referent identifications to specify conditions of co-reference. Consider as an example the following abstraction principle for the predicate ‘is morally right’: MR

For every speaker S and every predicate P of S’s language, P, as understood by S, has the same referent as ‘is morally right’ just in case S regulates her ascription of P on the basis of her feeling of moral approval.

According to the objection under consideration, understanding MR will enable us to gain cognitive access to the referent of ‘is morally right’ only if we already have cognitive access to the referent of ‘x regulates her ascription of y on the basis of her feeling of moral approval’. I want to argue that this objection is unsound. An instance of REF singles out the property that it pairs with the target predicate using a singular term formed with the identifying predicate and a nominalization device. This singular term has the same referent as the identifying predicate. If you understand the nominalization device, you will understand the resulting singular term if and only if you understand the identifying predicate, and you will have cognitive access to the referent of the singular term if and only if you have cognitive access to the referent of the identifying predicate. If you understand the identifying predicate (if you have the concept it expresses) you will understand the instance of REF. If, in addition, you have cognitive access to the referent of the identifying predicate, the instance of REF will enable you to gain cognitive access to the referent of the target predicate. However, if you understand the identifying predicate but have no cognitive access to its referent, you will understand the instance of REF, but this

   

211

won’t enable you to gain cognitive access to the referent of the target predicate. This is the intuitive idea behind the Basic Principle. The point I’m making concerning the singular term that figures in an instance of REF (‘the property of ψ-ing’) applies to other singular terms. You can understand the term ‘the person who was singing loudly on the street last night’ without having represented in thought the identity of the person the term refers to. If you are in this situation, then you will understand the sentence ‘the person who toppled the rubbish bins is the person who was singing loudly on the street last night’, but this understanding won’t enable you to represent in thought the identity of the person who toppled the bins. This is the phenomenon that I’m highlighting in the case of instances of REF. If you understand the predicate ‘is water’, you will understand AA, but if you don’t have cognitive access to the referent of ‘is water’, this understanding won’t enable you to gain cognitive access to the referent of ‘is Adam’s ale’. But things are different with abstraction principles. So long as I understand the relational predicate ‘x regulates her ascription of y on the basis of her feeling of moral approval’ (so long as I have the concept expressed by the predicate), I will be able to represent in thought the state of affairs of a speaker regulating her use of a predicate on the basis of her feeling of moral approval. Crucially, I’ll be able to represent in thought this state of affairs even if I don’t have cognitive access to the referent of the predicate, just as I can represent in thought the state of affairs of the liquid in the glass being water without having cognitive access to the referent of ‘is water’. Now, if I can represent in thought a speaker regulating her ascription of a predicate on the basis of her feeling of moral approval, I’ll be able to grasp the thought that if a speaker regulates her ascription of a predicate on the basis of her feeling of moral approval then the predicate, as understood by the speaker, bears some relation to ‘is morally right’. The thought expressed by MR is of this form, and grasping this thought is all that’s required on the abstractionist model for grasping the referent of ‘is morally right’. And, again, this can be achieved without grasping the referent of ‘x regulates her ascription of y on the basis of her feeling of moral approval’. Unlike instances of REF, abstraction principles enable us to progress from mere predicate understanding to cognitive access to predicate referents.

9.12 Conclusion In previous chapters, our discussion has been focused on declarative discourses that are widely seen as posing a problem for the representationalist model, on the grounds that we can’t seem to make room in our ontology for properties that could plausibly play the role of referents for their predicates. This circumstance

212

 

puts pressure on the RR assumption. If the requisite referents aren’t available, the sentences of these discourses can’t have representationalist meaning grounds, and then, according to RR, they can’t successfully discharge the task of representing the world. However, to many of us the sentences of the problematic discourses appear perfectly fit for representational duties. I have argued that the representational character of these discourses can be vindicated by abandoning the RR assumption and adopting a pragmatist account of the meaning grounds of their sentences. I have argued that pragmatist meaning grounds are perfectly compatible with the function of representing the world. On the version of the pragmatist approach that I’ve defended, representational predicates with pragmatist meaning grounds have referents, which we identify with the abstraction principles generated by our ascription procedures for the predicates. On the picture that emerges from the adoption of the pragmatist model for these discourses, the range of representational discourses includes both sentences with representational meaning grounds and sentences with pragmatist meaning grounds. In the present chapter I’ve discussed a more far-reaching application of the pragmatist model. I have argued that the application of the representationalist model to theoretical science is also problematic. The problem in this case is not lack of referents, since naturalism enjoins a commitment to the existence of the properties postulated by our best scientific theories. The problem here concerns our cognitive access to the properties that play the role of referents for the predicates introduced by scientific theories. Adapting David Lewis’s argument for Humility, I have contended that it’s not possible for us to gain cognitive access to the identity of the properties that play this role. I have proposed that the difficulty can be overcome by adopting a pragmatist account of the meaning grounds of these predicates. In the resulting picture, our cognitive grasp of the referents of theoretical scientific predicates consists in our understanding of the abstraction principles generated by our criteria of intertheoretical co-reference. This move has wide-ranging consequences for our understanding of the relationship between language and reality. On a standard version of the representationalist model, the referents of predicates are identified with the language of the natural sciences. Hence, if we adopt a pragmatist account of the meaning grounds of this discourse, it follows that our cognitive access to the referents of all predicates (with the only possible exception of predicates referring to sensory qualities) consists in our grasp of abstraction principles generated by the procedures that regulate our ascription of predicates. I have initially defended this outcome by considering problems faced by the application of the representationalist model to theoretical scientific discourse, but then I’ve argued that the same result can be reached on more general grounds, from the reflection that cognitive access to properties through explicit definitions is necessarily dependent on the availability of an alternative method for cognitive access to properties.

Epilogue The Meaning Grounds of Meaning-Ground Specifications

The enterprise of identifying the meaning grounds of sentences rests on the assumption that there are facts as to what makes it the case that a sentence has the meaning it has. The pragmatist, no less than the representationalist, is committed to this assumption. The dispute between pragmatists and representationalists with respect to a discourse doesn’t concern the existence of these facts, but their nature. According to the pragmatist, (it is a fact that) what makes the sentences of the target discourse have the meaning they have is features of the procedures we employ for regulating their acceptance. Suppose we had an independent account of the nature of meaning-ground facts—a specification of what makes it the case that an expression has the meaning ground it has—of the nature of the connection between an expression and its meaning-grounding fact. Then we could use these facts to provide a representationalist account of the meaning grounds of meaning-ground specifications, as, for example, the sentence: MR

‘What makes the sentence “killing one to save five is morally right” have the meaning it has is the fact that its acceptance is regulated by our feelings of moral approval towards the killing of one to save five’.

On this representationalist approach, MR would have the meaning it has as a result of its relation with the independently specified fact that the sentence represents as obtaining. If we took this line, meaning-ground specifications, representationalist and pragmatist alike, would have representationalist meaning grounds. My purpose in this brief epilogue is to explore the possibility of an alternative—a pragmatist account of the meaning grounds of meaning-ground specifications. On the pragmatist approach, meaning-ground specifications, such as MR, would have the meaning they have as a result of some features of the procedures by which their acceptance is regulated. Notice that this is perfectly compatible with a commitment to meaning-ground facts. This commitment would take the

Pragmatist Semantics: A Use-Based Approach to Linguistic Representation. José L. Zalabardo, Oxford University Press. © José L. Zalabardo 2023. DOI: 10.1093/oso/9780192874757.003.0010

214

 

form of regarding our acceptance of a meaning-ground ascription as subject to an absolute standard of correctness.¹ What are the acceptance procedures for meaning-ground specifications that can be treated as grounding their meaning? I want to argue that our acceptance of meaning-ground specifications is regulated by the same procedures that regulate our acceptance of interpretations. Let A be a sentence of our language. A specification of the meaning ground of A would take the form: MG1

What makes A have the meaning it has is the fact that P(A),

where P is a property of A, e.g. having its acceptance regulated by the speaker’s feeling of moral approval towards some action. Under our assumptions concerning meaning grounds, MG1 entails: MG2

A sentence B has the same meaning as A just in case P(B).

I’m going to concentrate on the meaning grounds of sentences of the form MG2, leaving to one side the additional content that sentences of the form MG1 might include. I want to argue that our acceptance of MG2 is regulated by our interpretative procedures. We should accept MG2 just in case we accept: MG3

Our interpretative procedures recommend interpreting a sentence B as representing the same state of affairs as A just in case P(B).

That is to say, we accept MG2, and MG1, only if the satisfaction of P by a sentence tracks whether interpreting the sentence as representing the same state of affairs as A maximally satisfies our interpretative criteria—on my account, the combination of projection and familiarity supplemented by non-verbal belief-ascription evidence (P+F+NVE). Applying this to our example, we can conclude that we should accept MR only if the following is the case: MRI

P+F+NVE recommends interpreting a sentence B as meaning that killing one to save five is morally right just in case acceptance of B is regulated by the speaker’s sense of moral approval towards the killing of one to save five.

It follows that the proposals I’ve made in this book concerning the meaning grounds of sentences in our target discourses should ultimately be assessed by

¹ See Section 7.13, above.



215

reference to P+F+NVE. I’m not going to undertake this task here, but I want to briefly illustrate the kind of issues on which the requisite validation will turn. Let’s focus on my proposal that what makes our predicate ‘is morally right’ have the meaning it has is the fact that its ascription is regulated by our feelings of moral approval. I’m claiming that this proposal would have to be validated by P+F +NVE—by whether regulation of a predicate by feelings of moral approval tracks whether P+F+NVE recommends interpreting the predicate as meaning is morally right. Let’s focus on projection. Interpreting a predicate ϕ, as understood by a speaker S, as meaning is morally right will satisfy the projection criterion to the extent that the actions to which S ascribes ϕ match the actions we would regard as morally right if we found ourselves in S’s epistemic situation—i.e. the actions towards which we would feel moral approval in that situation. Then my account of the meaning ground of ‘is morally right’ would be vindicated to the extent that the following condition is satisfied: Speakers ascribe a predicate to the actions to which we would feel moral approval in their situation just in case they regulate their ascription of the predicate by their own feeling of moral approval. The right-to-left direction of this condition requires that different subjects, if placed in the same epistemic situation, would feel moral approval towards the same actions. I think this condition is only satisfied imperfectly, since different subjects who find themselves in the same epistemic situation will not necessarily feel moral approval towards the same actions, on any reasonable construal of one’s epistemic situation. Notice, though, that my proposal might still be vindicated by P+F+NVE. If ascription of a predicate is regulated by the speaker’s feeling of moral approval, but the actions to which she feels moral approval differ from the actions to which we would feel moral approval in her epistemic situation, it might still be that interpreting her predicate as meaning is morally right has a familiarity advantage over rival interpretations that outweighs its less-than-perfect satisfaction of projection. The left-to-right direction requires that one’s sense of moral approval is the only effective way to pick out the actions to which we would feel moral approval in a given epistemic situation. For this condition to be satisfied, it would have to be the case that any other method for regulating ascription of a predicate is unlikely to replicate the verdicts of our sense of moral approval in a given epistemic situation. Specifically, subjects lacking a sense of moral approval would have to be incapable of classifying actions in a way that matches the classification that our sense of moral approval would produce in their epistemic situation.² ² On this point, see evidence suggesting that psychopaths have difficulties drawing the moral/ conventional distinction (Blair 1995; Blair et al. 1995). See (Nichols 2004: ch. 1) for discussion.

216

 

The issue of the validation of pragmatist meaning grounds by our interpretative practice has a special complexion in one particular case—the proposed pragmatist meaning grounds for interpretations. Consider the sentence: INT

‘What makes the sentence “A means that p” have the meaning it has is the fact that its acceptance is regulated by whether P+F+NVE recommends interpreting A as meaning that p.’

INT expresses our pragmatist account of the meaning grounds of interpretations. On a representationalist account of the meaning grounds of meaning-ground specifications, there’s nothing special about INT compared to other pragmatist meaning-ground specifications. On this view, INT has the meaning it has as a result of a semantic connection it bears to the (semantic) state of affairs it represents as obtaining. However, if we try to provide a pragmatist account of the meaning ground of INT, we face a situation that doesn’t arise with other pragmatist meaning-ground specifications. As with MR, INT will have to be validated by P+F+NVE. Hence, the P+F+NVE account of the meaning grounds of interpretations would have to be validated by P+F+NVE itself. Contrary to what it might seem, the double involvement of P+F+NVE doesn’t render the validation trivial or automatic. In order for INT to be validated, our interpretative procedures will have to have the following consequence: INTI

P+F+NVE recommends interpreting a sentence B as meaning that a sentence A means that p just in case acceptance of B is regulated by whether P+F+NVE recommends interpreting A as meaning that p.

Once again, I’m not going to address the question, whether our account of the meaning grounds of interpretations can be successfully validated in this way, but I want to give a general sense of the issues we will need to deal with when we undertake this task. On the proposal I developed in Chapter 7, the predicate ‘means that p’ has the meaning it has by virtue of the fact that its ascription to a sentence is regulated by whether P+F+NVE recommends interpreting the sentence as meaning that p. I’m claiming that this proposal would have to be validated by P+F+NVE—by whether the following two questions receive the same answer for every predicate ϕ: Is the ascription of ϕ to sentences regulated by whether P+F+NVE recommends interpreting them as meaning that p? Does P+F+NVE recommend interpreting ϕ as meaning means that p? Let’s focus on projection once again. Projection would recommend interpreting ϕ as meaning means that p to the extent that the sentences to which the speaker



217

ascribes ϕ coincide with the sentences to which we would ascribe ‘means that p’ if we found ourselves in the speaker’s epistemic situation—i.e. with the sentences that our application of P+F+NVE in that situation would recommend interpreting as meaning that p. It follows that my account of the meaning ground of ‘means that p’ would be vindicated by projection to the extent that the following condition is satisfied: Speakers ascribe a predicate to the sentences that our application of P+F+NVE in their situation would recommend interpreting as meaning that p just in case they regulate their ascription of the predicate by whether P+F+NVE recommends interpreting them as meaning that p. This requires that, in a given epistemic situation, P+F+NVE would recommend to different speakers the same verdict on whether a sentence means that p. Notice that satisfaction of this condition is not guaranteed. We saw in Chapter 7 that P+F+NVE could in principle support different interpretations when applied by different speakers, and on any natural construal of one’s epistemic situation, this will remain a possibility for speakers in the same epistemic situation. The support lent to INTI by the projection criterion will be inversely proportional to the extent of this variation. As with ‘is morally right’, when this support is less than perfect, the deficit might be compensated for by a familiarity advantage over rivals. I have sketched here a route map for a thoroughgoing pragmatism about meaning grounds. The starting point is the picture described in Chapter 9, in which the meaning grounds of all representational sentences are ultimately specified in terms of discourses with pragmatist meaning grounds. After this picture is accepted, we still face the question of how to specify the meaning grounds of meaning-ground specifications. I have suggested that the pragmatist template can also be applied here, and I have argued that this would require validation of our meaning-ground specifications in terms of our interpretative procedures. This includes validation by our interpretative procedures of our pragmatist specification of the meaning grounds of interpretations in terms of our interpretative procedures.

References Albritton, Rogers. 1959. ‘On Wittgenstein’s Use of the Term “Criterion” ’. Journal of Philosophy 56:845–57. Andreas, Holger, and Mario Günther. 2021. ‘Difference-Making Causation’. Journal of Philosophy 118:680–701. Apperly, Ian A., and Stephen A. Butterfill. 2009. ‘Do Humans Have Two Systems to Track Beliefs and Belief-Like States?’. Psychological Review 116 (4):953–70. Armstrong, David M. 1997. A World of States of Affairs. Cambridge: Cambridge University Press. Asay, Jamin. 2013. ‘Truthmaking, Metaethics, and Creeping Minimalism’. Philosophical Studies 163:213–32. Ayer, A. J. 1936. Language, Truth, and Logic. London: Victor Gollancz. Baillargeon, Renée, Rose M. Scott, and Zijing He. 2010. ‘False-Belief Understanding in Infants’. Trends in Cognitive Sciences 14 (3):110–18. Barnes, W. H. F. 1933. ‘A Suggestion about Value’. Analysis 1:45–6. Baron-Cohen, Simon, Alan M. Leslie, and Uta Frith. 1985. ‘Does the Autistic Child Have a “Theory of Mind”?’. Cognition 21 (1):37–46. Benacerraf, Paul. 1965. ‘What Numbers Could Not Be’. Philosophical Review 74 (1):47–73. Berenstain, Nora. 2016. ‘What a Structuralist Theory of Properties Could Not Be’. In The Metaphysics of Relations, ed. D. Yates and A. Marmodoro. Oxford: Oxford University Press. Berkeley, George. 1988. Principles of Human Knowledge and Three Dialogues. Oxford: Oxford University Press. Birch, Susan A. J., and Paul Bloom. 2004. ‘Understanding Children’s and Adults’ Limitations in Mental State Reasoning’. Trends in Cognitive Sciences 8 (6):255–60. Birch, Susan A. J., and Paul Bloom. 2007. ‘The Curse of Knowledge in Reasoning about False Beliefs’. Psychological Science 18 (5):382–6. Björklund, Fredrik, Gunnar Björnsson, John Eriksson, Ragnar Francén Olinder, and Caj Strandberg. 2012. ‘Recent Work on Motivational Internalism’. Analysis 72 (1):124–37. Black, Robert. 2000. ‘Against Quidditism’. Australasian Journal of Philosophy 78:87–104. Blackburn, Simon. 1980. ‘Truth, Realism, and the Regulation of Theory’. Midwest Studies in Philosophy 5:353–72. Blackburn, Simon. 1984. Spreading the Word. Groundings in the Philosophy of Language. Oxford: Clarendon Press. Blackburn, Simon. 1988. ‘Attitudes and Contents’. Ethics 98:501–17. Blackburn, Simon. 1993. ‘Circles, Finks, Smells and Biconditionals’. Philosophical Perspectives 7:259–79. Blackburn, Simon. 2002. ‘Realism: Deconstructing the Debate’. Ratio 15:111–33. Blackburn, Simon. 2006. ‘Antirealist Expressivism and Quasi-Realism’. In The Oxford Handbook of Ethical Theory, ed. D. Copp. Oxford: Oxford University Press. Blair, R. J. R. 1995. ‘A Cognitive Developmental Approach to Morality: Investigating the Psychopath’. Cognition 57:1–29.

220



Blair, R. J. R., L. Jones, F. Clark, and M. Smith. 1995. ‘Is the Psychopath “Morally Insane”?’. Personality and Individual Differences 19:741–52. Boghossian, Paul. 1989. ‘The Rule-Following Considerations’. Mind 98:507–49. Boghossian, Paul. 1990. ‘The Status of Content’. Philosophical Review 99:157–84. Boghossian, Paul. 1997. ‘What the Externalist Can Know a Priori’. Proceedings of the Aristotelian Society 97:161–75. Boghossian, Paul. 2003. ‘The Normativity of Content’. Philosophical Issues 13:31–45. Boyd, Richard. 1988. ‘How to Be a Moral Realist’. In Essays on Moral Realism, ed. G. SayreMcCord. Ithaca: Cornell University Press. Boyle, Matthew. 2009. ‘Two Kinds of Self-Knowledge’. Philosophy and Phenomenological Research 78:133–64. Brandom, Robert. 1983. ‘Asserting’. Noûs 17:637–50. Brandom, Robert. 1988. ‘Pragmatism, Phenomenalism, and Truth Talk’. Midwest Studies in Philosophy 12:75–93. Brandom, Robert. 1994. Making It Explicit. Cambridge, MA: Harvard University Press. Brandom, Robert. 2000. Articulating Reasons: An Introduction to Inferentialism. Cambridge, MA: Harvard University Press. Brink, David Owen. 1989. Moral Realism and the Foundations of Ethics. Cambridge: Cambridge University Press. Bronzo, Silver. 2011. ‘Context, Compositionality, and Nonsense in Wittgenstein’s Tractatus’. In Beyond the Tractatus Wars, ed. R. Read and M. A. Lavery. Routledge: New York. Burge, Tyler. 1979. ‘Individualism and the Mental’. In Midwest Studies in Philosophy 4:73–121. Byrne, Alex. 2011. ‘Transparency, Belief, Intention’. Aristotelian Society Supplementary Volume 85:201–21. Camerer, Colin, George Loewenstein, and Martin Weber. 1989. ‘The Curse of Knowledge in Economic Settings: An Experimental Analysis’. Journal of Political Economy 97 (5):1232–54. Carnap, Rudolf. 1937. Philosophy and Logical Syntax. London: Kegan Paul, Trench, Trubner & Co. Carnap, Rudolf. 1956. Meaning and Necessity. 2nd ed. Chicago: University of Chicago Press. Carnap, Rudolf. 1967. The Logical Structure of the World. Berkeley and Los Angeles: University of California Press. Carruthers, Peter. 2017. ‘Mindreading in Adults: Evaluating Two-Systems Views’. Synthese 194 (3):673–88. Carruthers, Peter, and Peter K. Smith, eds. 1996. Theories of Theories of Mind. Cambridge: Cambridge University Press. Churchland, Paul M. 1979. Scientific Realism and the Plasticity of Mind. Cambridge: Cambridge University Press. Churchland, Paul M. 1981. ‘Eliminative Materialism and the Propositional Attitudes’. The Journal of Philosophy 78:67–90. Cohen, L. Jonathan. 1992. An Essay on Belief and Acceptance. Oxford: Clarendon. Collins, John. 2002. ‘Truth or Meaning? A Question of Priority’. Philosophy and Phenomenological Research 65:497–536. Csibra, Gergely, Szilvia Bíró, Orsolya Koós, and György Gergely. 2003. ‘One-Year-Old Infants Use Teleological Representations of Actions Productively’. Cognitive Science 27 (1):111–33.



221

Csibra, Gergely, and György Gergely. 1998. ‘The Teleological Origins of Mentalistic Action Explanations: A Developmental Hypothesis’. Developmental Science 1 (2):255–9. Csibra, Gergely, and György Gergely. 2007. ‘ “Obsessed with Goals”: Functions and Mechanisms of Teleological Interpretation of Actions in Humans’. Acta Psychologica 124 (1):60–78. D’Arms, Justin, and Daniel Jacobson. 1994. ‘Expressivism, Morality, and the Emotions’. Ethics 104:739–63. Darwall, Stephen, Allan Gibbard, and Peter Railton. 1992. ‘Toward Fin De Siècle Ethics: Some Trends’. Philosophical Review 101:115–89. Dasgupta, Shamik. 2015. ‘Inexpressible Ignorance’. Philosophical Review 124:441–80. Davidson, Donald. 1963. ‘Actions, Reasons, and Causes’. The Journal of Philosophy 60 (23):685–700. Davidson, Donald. 1973. ‘Radical Interpretation’. Dialectica 27:313–28. Davidson, Donald. 1979. ‘The Inscrutability of Reference’. Southwestern Journal of Philosophy 10:7–19. Davidson, Donald. 1990. ‘The Structure and Content of Truth’. Journal of Philosophy 87:279–328. Davidson, Donald. 2001a. ‘A Coherence Theory of Truth and Knowledge’. In Subjective, Intersubjective, Objective. Oxford: Clarendon. Davidson, Donald. 2001b. ‘On Saying That’. In Inquiries into Truth and Interpretation. Oxford: Oxford University Press. Davidson, Donald. 2001c. ‘Theories of Meaning and Learnable Languages’. In Inquiries into Truth and Interpretation. Oxford: Oxford University Press. Davis, John K. 2015. ‘Faultless Disagreement, Cognitive Command, and Epistemic Peers’. Synthese 192:1–24. Dennett, Daniel C. 1987a. ‘Midterm Examination: Compare and Contrast’. In The Intentional Stance. Cambridge, MA: MIT Press. Dennett, Daniel C. 1987b. ‘True Believers’. In The Intentional Stance. Cambridge, MA: MIT Press. Dennett, Daniel C. 1988. ‘Science, Philosophy, and Interpretation’. Behavioral and Brain Sciences 11:535–46. DeRose, Keith. 1995. ‘Solving the Skeptical Problem’. Philosophical Review 104:1–52. Dreier, James. 2004. ‘Meta-Ethics and the Problem of Creeping Minimalism’. Philosophical Perspectives 18:23–44. Dreier, James. 2009. ‘Relativism (and Expressivism) and the Problem of Disagreement’. Philosophical Perspectives 23:79–110. Duhem, Pierre. 1991. The Aim and Structure of Physical Theory. Translated by P. P. Wiener. Princeton: Princeton University Press. Dummett, Michael. 1976. ‘What Is a Theory of Meaning? (II)’. In Truth and Meaning. Essays in Semantics, ed. G. Evans and J. McDowell. Oxford: Oxford University Press. Dummett, Michael. 1981. The Interpretation of Frege’s Philosophy. Cambridge, MA: Harvard University Press. Dummett, Michael. 2005. ‘The Justificationist’s Response to a Realist’. Mind 114:671–88. Egan, Andy. 2006. ‘Secondary Qualities and Self-Location’. Philosophy and Phenomenological Research 72:97–119. Epley, Nicholas, Boaz Keysar, Leaf Van Boven, and Thomas Gilovich. 2004. ‘Perspective Taking as Egocentric Anchoring and Adjustment’. Journal of Personality and Social Psychology 87 (3):327–39. Evans, Gareth. 1982. The Varieties of Reference. Oxford: Oxford University Press.

222



Fernández, Jordi. 2007. ‘Desire and Self-Knowledge’. Australasian Journal of Philosophy 85:517–36. Field, Hartry. 1972. ‘Tarski’s Theory of Truth’. Journal of Philosophy 69:347–75. Field, Hartry. 1975. ‘Conventionalism and Instrumentalism in Semantics’. Noûs 9:375–405. Field, Hartry. 1978. ‘Mental Representation’. Erkenntnis 13:9–61. Field, Hartry. 2017. ‘Egocentric Content’. Noûs 51 (3):521–46. Fine, Kit. 2001. ‘The Question of Realism’. Philosophers’ Imprint 1:1–30. Fine, Kit. 2002. The Limits of Abstraction. Oxford: Oxford University Press. Fine, Kit. 2012. ‘Guide to Ground’. In Metaphysical Grounding, ed. F. Correia and B. Schnieder: Cambridge University Press. Fodor, Jerry A. 1975. The Language of Thought. Harvard, MA: Harvard University Press. Fodor, Jerry A. 1987. Psychosemantics. The Problem of Meaning in the Philosophy of Mind. Cambridge, MA: MIT Press. Fodor, Jerry A. 1992a. ‘A Theory of Content, I: The Problem’. In A Theory of Content and Other Essays. Cambridge, MA: MIT Press. Fodor, Jerry A. 1992b. ‘A Theory of Content, II: The Theory’. In A Theory of Content and Other Essays. Cambridge, MA: MIT Press. Fodor, Jerry A., and Ernest Lepore. 1994. ‘Is Radical Interpretation Possible?’. Philosophical Perspectives 8:101–19. Foot, Philippa. 2002a. ‘Approval and Disapproval’. In Virtues and Vices: And Other Essays in Moral Philosophy. Oxford: Oxford University Press. Foot, Philippa. 2002b. ‘Moral Beliefs’. In Virtues and Vices: And Other Essays in Moral Philosophy. Oxford: Oxford University Press. Frankena, W. K. 1939. ‘The Naturalistic Fallacy’. Mind 48:464–77. Frege, Gottlob. 1980a. The Foundations of Arithmetic. Translated by J. L. Austin. Evanston, IL: Northwestern University Press. Frege, Gottlob. 1980b. ‘On Sense and Reference’. In Translations from the Philosophical Writings of Gottlob Frege, ed. M. Black and P. T. Geach. Oxford: Basil Blackwell. Geach, Peter T. 1960. ‘Ascriptivism’. Philosophical Review 69:221–5. Geach, Peter T. 1965. ‘Assertion’. Philosophical Review 74:449–65. Gergely, György, Zoltán Nádasdy, Gergely Csibra, and Szilvia Bíró. 1995. ‘Taking the Intentional Stance at 12 Months of Age’. Cognition 56 (2):165–93. Gert, Joshua. 2012. Normative Bedrock: Response-Dependence, Rationality, and Reasons. Oxford: Oxford University Press. Ghrear, Siba E., Susan A. J. Birch, and Daniel M. Bernstein. 2016. ‘Outcome Knowledge and False Belief ’. Frontiers in Psychology 7 (118). Gibbard, Allan. 1990. Wise Choices, Apt Feelings. Oxford: Clarendon. Gibbard, Allan. 2012. Meaning and Normativity. Oxford: Oxford University Press. Glanzberg, Michael. 2007. ‘Context, Content, and Relativism’. Philosophical Studies 136:1–29. Goldman, Alvin I. 1989. ‘Interpretation Psychologized’. Mind and Language 4:161–85. Grandy, Richard. 1973. ‘Reference, Meaning, and Belief ’. The Journal of Philosophy 70:439–52. Grice, H. P. 1957. ‘Meaning’. Philosophical Review 66: 377–88. Gross, Steven, Nicholas Tebben, and Michael Wiliams, eds. 2015. Meaning without Representation. Essays on Truth, Expression, Normativity, and Naturalism. Oxford: Oxford University Press. Hale, Bob. 1987. Abstract Objects. Oxford: Basil Blackwell.



223

Hale, Bob, and Crispin Wright. 2001. The Reason’s Proper Study: Essays towards a NeoFregean Philosophy of Mathematics. Oxford: Oxford University Press. Hare, R. M. 1952. The Language of Morals. Oxford: Clarendon. Harman, Gilbert. 1977. The Nature of Morality. An Introduction to Ethics. New York: Oxford University Press. Harman, Gilbert. 1988. ‘What Is the Intentional Stance?’. Behavioral and Brain Sciences 11:515. Hawthorne, John. 2001. ‘Causal Structuralism’. In Philosophical Perspectives, Vol. 15: Metaphysics, ed. J. Tomberlin. Atascadero, CA: Ridgeview. Hawthorne, John, and Huw Price. 2011. ‘How to Stand Up for Non-Cognitivists’. In Naturalism without Mirrors. Oxford: Oxford University Press. Heller, Mark. 1998. ‘Property Counterparts in Ersatz Worlds’. Journal of Philosophy 95:293–316. Hochberg, Herbert. 1967. ‘Nominalism, Platonism and “Being True Of” ’. Noûs 1:413–19. Horgan, Terence, and Mark Timmons. 1991. ‘New Wave Moral Realism Meets Moral Twin Earth’. Journal of Philosophical Research 16:447–65. Horgan, Terence, and Mark Timmons. 1992. ‘Troubles for New Wave Moral Semantics: The “Open Question Argument” Revived’. Philosophical Papers 21:153–75. Horgan, Terence, and Mark Timmons. 2006. ‘Cognitivist Expressivism’. In Metaethics after Moore, ed. T. Horgan and M. Timmons. Oxford: Oxford University Press. Horgan, Terence, and Mark Timmons. 2015. ‘Exploring Intuitions on Moral Twin Earth: A Reply to Sonderholm’. Theoria 81:355–75. Horwich, Paul. 1995. ‘Meaning, Use and Truth’. Mind 104:355–68. Horwich, Paul. 1998a. Truth. 2nd ed. Oxford: Oxford University Press. Horwich, Paul. 1998b. Meaning. Oxford: Oxford University Press. Hume, David. 1978. A Treatise of Human Nature. 2nd ed. Oxford: Clarendon. Hursthouse, Rosalind. 1991. ‘Arational Actions’. Journal of Philosophy 88:57–68. Hylton, Peter. 2007. Quine. New York: Routledge. Jackson, Frank. 1998. From Metaphysics to Ethics: A Defense of Conceptual Analysis. Oxford: Oxford University Press. Jackson, Frank, Graham Oppy, and Michael Smith. 1994. ‘Minimalism and Truth Aptness’. Mind 103:287–302. Jackson, Frank, and Philip Pettit. 1995. ‘Moral Functionalism and Moral Motivation’. Philosophical Quarterly 45:20–40. Johnston, Mark. 1993. ‘Objectivity Refigured: Pragmatism without Verificationism’. In Reality, Representation, and Projection, ed. J. Haldane and C. Wright. Oxford: Oxford University Press. Kalderon, Mark E. 1997. ‘The Transparency of Truth’. Mind 106:475–97. Kaplan, David. 1989a. ‘Afterthoughts’. In Themes from Kaplan, ed. J. Almog, J. Perry, and H. Wettstein. Oxford: Oxford University Press. Kaplan, David. 1989b. ‘Demonstratives: An Essay on the Semantics, Logic, Metaphysics and Epistemology of Demonstratives and Other Indexicals’. In Themes from Kaplan, ed. J. Almog, J. Perry, and H. Wettstein. Oxford: Oxford University Press. Kelly, Alexander. 2013. ‘Ramseyan Humility, Scepticism and Grasp’. Philosophical Studies 164:705–26. Köhler, Sebastian. forthcoming. ‘What Is (Neo-)Pragmatists’ Function?’. Australasian Journal of Philosophy. Kölbel, Max. 2002. Truth without Objectivity. London: Routledge. Kraut, Robert. 1993. ‘Robust Deflationism’. Philosophical Review 102 (2):247–63.

224



Kripke, Saul. 1980. Naming and Necessity. Oxford: Blackwell. Kripke, Saul. 1982. Wittgenstein on Rules and Private Language. Oxford: Blackwell. Laurence, Stephen, Eric Margolis, and Angus Dawson. 1999. ‘Moral Realism and Twin Earth’. Facta Philosophica 1:135–65. Leuenberger, Stephan. 2010. ‘Humility and Constraints on O-Language’. Philosophical Studies 149:327–54. Lewis, Clarence Irving. 1929. Mind and the World Order: Outline of a Theory of Knowledge. New York: Charles Scribners. Lewis, David. 1970. ‘General Semantics’. Synthese 22:18–67. Lewis, David. 1973. ‘Causation’. Journal of Philosophy 70:556–67. Lewis, David. 1974. ‘Radical Interpretation’. Synthese 23:331–44. Lewis, David. 1983. ‘New Work for a Theory of Universals’. Australasian Journal of Philosophy 61:343–77. Lewis, David. 1984. ‘Putnam’s Paradox’. Australasian Journal of Philosophy 62:221–36. Lewis, David. 1986. On the Plurality of Worlds. Oxford: Blackwell. Lewis, David. 2000. ‘Causation as Influence’. Journal of Philosophy 97:182–97. Lewis, David. 2009. ‘Ramseyan Humility’. In Conceptual Analysis and Philosophical Naturalism, ed. D. Braddon-Mitchell and R. Nola. Cambridge, MA: MIT Press. Linnebo, Øystein. 2018. Thin Objects. An Abstractionist Account. Oxford: Oxford University Press. Locke, Dustin. 2009. ‘A Partial Defence of Ramseyan Humility’. In Conceptual Analysis and Philosophical Naturalism, ed. D. Braddon-Mitchell and R. Nola. Cambridge, MA: MIT Press. Locke, John. 1975. An Essay Concerning Human Understanding. Oxford: Oxford University Press. Lycan, William G. 1971. ‘Noninductive Evidence: Recent Work on Wittgenstein’s “Criteria” ’. American Philosophical Quarterly 8:109–25. Lycan, William G. 1988. ‘Dennett’s Instrumentalism’. Behavioral and Brain Sciences 11:518–19. MacBride, Fraser. 2003. ‘Speaking with Shadows: A Study of Neo-Logicism’. British Journal for the Philosophy of Science 54:103–63. MacFarlane, John. 2007. ‘Relativism and Disagreement’. Philosophical Studies 132:17–31. Mackie, J. L. 1977. Ethics: Inventing Right and Wrong. Harmondsworth: Penguin. Mancosu, Paolo. 2016. Abstraction and Infinity. Oxford: Oxford University Press. Matthews, Robert J. 2007. The Measure of Mind: Propositional Attitudes and Their Attribution. Oxford: Oxford University Press. McCarty, David. 1991. ‘The Philosophy of Logical Wholism’. Synthese 87:51–123. McDowell, John. 1979. ‘Virtue and Reason’. The Monist 62:331–50. Miller, Alexander. 2013. Contemporary Metaethics: An Introduction. 2nd ed. Cambridge: Polity. Mitchell, P., E. J. Robinson, J. E. Isaacs, and R. M. Nye. 1996. ‘Contamination in Reasoning about False Belief: An Instance of Realist Bias in Adults but Not Children’. Cognition 59 (1):1–21. Moore, G. E. 1903. Principia Ethica. Cambridge: Cambridge University Press. Moore, G. E. 1993. ‘Moore’s Paradox’. In G. E. Moore: Selected Writings, ed. T. Baldwin. London: Routledge. Moran, Richard. 2001. Authority and Estrangement. Princeton: Princeton University Press. Nagel, Ernest. 1961. The Structure of Science: Problems in the Logic of Scientific Explanation. New York: Harcourt, Brace & World.



225

Nagel, Thomas. 1970. The Possibility of Altruism. Oxford: Clarendon Press. Nichols, Shaun. 2004. Sentimental Rules: On the Natural Foundations of Moral Judgement. Oxford: Oxford University Press. Nozick, Robert. 1981. Philosophical Explanations. Cambridge, MA: Harvard University Press. Onishi, Kristine H., and Renée Baillargeon. 2005. ‘Do 15-Month-Old Infants Understand False Beliefs?’. Science (New York) 308 (5719):255–8. Peacocke, Christopher. 1992. A Study of Concepts. Cambridge, MA: MIT Press. Peano, Giuseppe. 1894. Notations de Logique Mathématique. Introduction au Formulaire de Mathématique. Turin: Bocca. Perner, Josef, and Johannes Roessler. 2010. ‘Teleology and Causal Reasoning in Children’s Theory of Mind’. In Causing Human Action: New Perspectives on the Causal Theory of Action, ed. J. Aguilar and A. Buckareff. Cambridge, MA: MIT Press. Pettit, Philip. 1991. ‘Realism and Response-Dependence’. Mind 100:587–626. Plato. 1963. ‘The Sophist’. In The Collected Dialogues of Plato, ed. E. Hamilton and H. Cairns. Princeton: Princeton University Press. Poincaré, Henri. 1952. Science and Hypothesis. Translated by G. B. Halsted. New York: Dover Publications. Pollock, John. 1967. ‘Criteria and Our Knowledge of the Material World’. Philosophical Review 76:28–60. Premack, D., and G. Woodruff. 1978. ‘Does the Chimpanzee Have a Theory of Mind?’. Behavioral and Brain Sciences 1:515–26. Price, Huw. 2011a. ‘Immodesty without Mirrors: Makind Sense of Wittgenstein’s Linguistic Pluralism’. In Naturalism without Mirrors. Oxford: Oxford University Press. Price, Huw. 2011b. ‘Naturalism and the Fate of the M-Worlds’. In Naturalism without Mirrors. Oxford: Oxford University Press. Price, Huw. 2011c. ‘One Cheer for Representationalism?’. In Naturalism without Mirrors. Oxford: Oxford University Press. Pryor, James. 2000. ‘The Skeptic and the Dogmatist’. Noûs 34:517–49. Putnam, Hilary. 1963. ‘An Examination of Grünbaum’s Philosophy of Geometry’. In Philosophy of Science, the Delaware Seminar, Vol. 2: 1962–1963, ed. B. Baumrin. New York: John Wiley & Sons. Putnam, Hilary. 1975. ‘The Meaning of “Meaning” ’. In Mind, Language and Reality. Philosophical Papers, vol. 2. Cambridge: Cambridge University Press. Putnam, Hilary. 1978. Meaning and the Moral Sciences. London: Routledge & Kegan Paul. Putnam, Hilary. 1981. Reason, Truth and History. Cambridge: Cambridge University Press. Putnam, Hilary. 1994. ‘Model Theory and the “Factuality” of Semantics’. In Words and Life, ed. J. Conant. Cambridge, MA: Harvard University Press. Quine, Willard Van Orman. 1943. ‘Notes on Existence and Necessity’. Journal of Philosophy 40:113–27. Quine, Willard Van Orman. 1960. Word and Object. Cambridge, MA: MIT Press. Quine, Willard Van Orman. 1969. ‘Ontological Relativity’. In Ontological Relativity and Other Essays. New York: Columbia University Press. Quine, Willard Van Orman. 1980. ‘Two Dogmas of Empiricism’. In From a Logical Point of View. Cambridge, MA: Harvard University Press. Railton, Peter. 1986. ‘Moral Realism’. Philosophical Review 95:163–207. Ramsey, Frank. 1927. ‘Facts and Propositions’. Aristotelian Society Supplementary Volume 7:153–70.

226



Roessler, Johannes, and Josef Perner. 2013. ‘Teleology: Belief as Perspective’. In Understanding Other Minds: Perspectives from Developmental Social Neuroscience, ed. S. Baron Cohen, M. Lombardo, and H. Tager-Flusberg. 3rd ed. New York: Oxford University Press. Rorty, Richard. 1998. ‘Is Truth a Goal of Inquiry? Donald Davidson versus Crispin Wright’. In Truth and Progress. Philosophical Papers, vol. 3. Cambridge: Cambridge University Press. Rorty, Richard. 2000a. ‘Response to Barry Allen’. In Rorty and His Critics, ed. R. Brandom. Oxford: Blackwell. Rorty, Richard. 2000b. ‘Universality and Truth’. In Rorty and His Critics, ed. R. Brandom. Oxford: Blackwell. Rosen, Gideon. 1994. ‘What Is Constructive Empiricism?’. Philosophical Studies 74:143–78. Rubin, Michael. 2008. ‘Sound Intuitions on Moral Twin Earth’. Philosophical Studies 139:307–27. Rupert, Robert D. 2008. ‘Causal Theories of Mental Content’. Philosophy Compass 3:253–80. Russell, Bertrand. 1903. The Principles of Mathematics. Cambridge: Cambridge University Press. Russell, Bertrand. 1905. ‘On Denoting’. Mind 14:479–93. Russell, Bertrand. 1907. ‘On the Nature of Truth’. Proceedings of the Aristotelian Society 7:28–49. Russell, Bertrand. 1910. ‘On the Nature of Truth and Falsehood’. In Philosophical Essays. London: Longmans, Green. Russell, Bertrand. 1912. The Problems of Philosophy. London: Williams & Norgate. Russell, Bertrand. 1984. ‘Theory of Knowledge: The 1913 Manuscript’. In The Collected Papers of Bertrand Russell, vol. 7, ed. E. R. Eames and K. Blackwell. London: Allen & Unwin. Schaffer, Jonathan. 2005. ‘Quiddistic Knowledge’. Philosophical Studies 123:1–32. Schroeder, Mark. 2008. ‘What Is the Frege-Geach Problem?’. Philosophy Compass 3 (4):703–20. Scott, Rose M., and Renée Baillargeon. 2017. ‘Early False-Belief Understanding’. Trends in Cognitive Sciences 21 (4):237–49. Searle, John R. 1962. ‘Meaning and Speech Acts’. Philosophical Review 71:423–32. Sellars, Wilfrid. 1997. Empiricism and the Philosophy of Mind. Cambridge, MA: Harvard University Press. Shafer-Landau, Russ. 1998. ‘Moral Judgement and Moral Motivation’. Philosophical Quarterly 48:353–8. Shafer-Landau, Russ. 2003. Moral Realism: A Defence: Oxford University Press. Shah, Nishi. 2003. ‘How Truth Governs Belief ’. Philosophical Review 112:447–82. Shah, Nishi, and J. David Velleman. 2005. ‘Doxastic Deliberation’. Philosophical Review 114:497–534. Shaw, James R. forthcoming. ‘Wittgensteinean Notions of Uniformity and Kripkensteinian Skepticism’. In Kripke’s ‘Wittgenstein on Rules and Private Language’ at 40, ed. C. Verheggen. Cambridge: Cambridge University Press. Shoemaker, Sydney. 1980. ‘Causality and Properties’. In Time and Cause, ed. P. van Inwagen. Dordrecht: Springer. Shoemaker, Sydney. 1988. ‘On Knowing One’s Own Mind’. Philosophical Perspectives 2:183–209. Shoemaker, Sydney. 1998. ‘Causal and Metaphysical Necessity’. Pacific Philosophical Quarterly 79:59–77.



227

Simpson, Matthew. 2018. ‘Solving the Problem of Creeping Minimalism’. Canadian Journal of Philosophy 48:510–31. Skyrms, Brian. 1981. ‘Tractarian Nominalism (for Wilfrid Sellars)’. Philosophical Studies 40:199–206. Smith, Michael. 1994. The Moral Problem. Oxford: Blackwell. Soames, Scott. 1998. ‘Skepticism about Meaning: Indeterminacy, Normativity, and the Rule-Following Paradox’. Canadian Journal of Philosophy Supplementary Volume 23:211–49. Sonderholm, Jorn. 2013. ‘Unreliable Intuitions: A New Reply to the Moral Twin-Earth Argument’. Theoria 79:76–88. Stalnaker, Robert C. 1984. Inquiry. Cambridge, MA: MIT Press. Stalnaker, Robert C. 1997. ‘Reference and Necessity’. In A Companion to the Philosophy of Language, ed. B. Hale and C. Wright. Oxford: Blackwell. Stanford, P. Kyle 2016. ‘Instrumentalism: Global, Local, and Scientific’. In The Oxford Handbook of Philosophy of Science, ed. P. Humphreys. Oxford: Oxford University Press. Stevenson, Charles Leslie. 1937. ‘The Emotive Meaning of Ethical Terms’. Mind 46:14–31. Stevenson, Charles Leslie. 1944. Ethics and Language. New Haven: Yale University Press. Stich, Stephen P. 1981. ‘Dennett on Intentional Systems’. Philosophical Topics 12:39–62. Stich, Stephen P. 1983. From Folk Psychology to Cognitive Science: The Case against Belief. Cambridge, MA: MIT Press. Strawson, Peter F. 1949. ‘Truth’. Analysis 9:83–97. Tarski, Alfred. 1936. ‘The Concept of Truth in Formalized Languages’. In Logic, Semantics, Metamathematics. Oxford: Oxford University Press. Thomasson, Amie L. 2015. Ontology Made Easy. Oxford: Oxford University Press. Timmons, Mark. 1998. Morality without Foundations: A Defense of Ethical Contextualism. Oxford: Oxford University Press. Tresan, Jon. 2006. ‘De Dicto Internalist Cognitivism’. Noûs 40:143–65. van Fraassen, Bas C. 1980. The Scientific Image. Oxford: Oxford University Press. Wallace, John. 1977. ‘Only in the Context of a Sentence Do Words Have Any Meaning’. Midwest Studies in Philosophy 2:144–64. Wedgwood, Ralph. 2007. The Nature of Normativity. Oxford: Oxford University Press. Wellman, H. M. 1991. ‘From Desires to Beliefs: Acquisition of a Theory of Mind’. In Natural Theories of Mind: Evolution, Development and Simulation of Everyday Mindreading, ed. A. Whiten. Oxford: Basil Blackwell. Wellman, H. M., D. Cross, and J. Watson. 2001. ‘Meta-Analysis of Theory-of-Mind Development: The Truth about False Belief ’. Child Development 72 (3):655–84. Whittle, Ann. 2006. ‘On an Argument for Humility’. Philosophical Studies 130:461–97. Whyte, J. T. 1990. ‘Success Semantics’. Analysis 50 (3):149–57. Wiggins, David. 1998. ‘A Sensible Subjectivism?’. In Needs, Values, Truth. Oxford: Oxford University Press. Wikforss, Åsa Maria. 2001. ‘Semantic Normativity’. Philosophical Studies 102:203–26. Williams, J. Robert G. 2007. ‘Eligibility and Inscrutability’. Philosophical Review 116:361–99. Williams, Michael. 1999. ‘Meaning and Deflationary Truth’. The Journal of Philosophy 96:545–64. Williams, Michael. 2013. ‘How Pragmatists Can Be Local Expressivists’. In Expressivism, Pragmatism and Representationalism. Cambridge: Cambridge University Press. Wimmer, H., and J. Perner. 1983. ‘Beliefs about Beliefs: Representation and Constraining Function of Wrong Beliefs in Young Children’s Understanding of Deception’. Cognition 13:103–28.

228



Wittgenstein, Ludwig. 1974. Tractatus Logico-Philosophicus. Translated by D. F. Pears and B. McGuinness. 2nd ed. London: Routledge and Kegan Paul. Wittgenstein, Ludwig. 1981. Zettel. Translated by G. E. M. Anscombe. 2nd ed. Oxford: Basil Blackwell. Wittgenstein, Ludwig. 2009. Philosophical Investigations. Translated by G. E. M. Anscombe, P. M. S. Hacker, and Joachim Schulte. 4th ed. Oxford: Blackwell. Woodfield, Andrew. 1982. ‘On Specifying the Contents of Thoughts’. In Thought and Object: Essays on Intentionality, ed. A. Woodfield. Oxford: Clarendon Press. Woodward, James. 2003. Making Things Happen: A Theory of Causal Explanation. Oxford: Oxford University Press. Woodward, James. 2011. ‘Causal Perception and Causal Cognition’. In Perception, Causation, and Objectivity, ed. J. Roessler, H. Lerman, and N. Eilan. Oxford: Oxford University Press. Wright, Crispin. 1982. ‘Anti-Realist Semantics: The Role of Criteria’. Royal Institute of Philosophy Supplements 13:225–48. Wright, Crispin. 1983. Frege’s Conception of Numbers as Objects. Aberdeen: Aberdeen University Press. Wright, Crispin. 1984. ‘Second Thoughts about Criteria’. Synthese 58:383–405. Wright, Crispin. 1988. ‘Moral Values, Projection, and Secondary Qualities’. Aristotelian Society Supplementary Volume 62:1–26. Wright, Crispin. 1992. Truth and Objectivity. Cambridge, MA: Harvard University Press. Wright, Crispin. 2001. ‘On Being in a Quandary. Relativism Vagueness Logical Revisionism’. Mind 110:45–98. Wright, Crispin. 2006. ‘Intuitionism, Realism, Relativism and Rhubarb’. In Truth and Realism, ed. P. Greenough and M. Lynch. Oxford: Clarendon Press. Wright, Crispin. 2012. ‘Meaning and Assertibility: Some Reflections on Paolo Casalegno’s “the Problem of Non-Conclusiveness” ’. Dialectica 66:249–66. Yalcin, Seth. 2018. ‘Semantics as Model-Based Science’. In The Science of Meaning: Essays on the Metatheory of Natural Language Semantics, ed. D. Ball and B. Rabern. Oxford: Oxford University Press. Yates, David. 2018. ‘Three Arguments for Humility’. Philosophical Studies 175:461–81. Yolton, John W. 1996. Perception & Reality: A History from Descartes to Kant. Ithaca: Cornell University Press. Zalabardo, José L. 1995. ‘A Problem for Information Theoretic Semantics’. Synthese 105 (1):1–29. Zalabardo, José L. 1996. ‘Predicates, Properties and the Goal of a Theory of Reference’. Grazer Philosophische Studien 51:121–62. Zalabardo, José L. 1997. ‘Kripke’s Normativity Argument’. Canadian Journal of Philosophy 27:467–88. Zalabardo, José L. 2000a. Introduction to the Theory of Logic. Boulder, CO: Westview Press. Zalabardo, José L. 2000b. ‘Realism Detranscendentalized’. European Journal of Philosophy 8:63–88. Zalabardo, José L. 2001. ‘Towards a Nominalist Empiricism’. Proceedings of the Aristotelian Society 101:29–52. Zalabardo, José L. 2003. ‘Wittgenstein on Accord’. Pacific Philosophical Quarterly 84:311–29. Zalabardo, José L. 2009. ‘One Strand in the Rule-Following Considerations’. Synthese 171:509–19. Zalabardo, José L. 2012a. Scepticism and Reliable Belief. Oxford: Oxford University Press.



229

Zalabardo, José L. 2012b. ‘Semantic Normativity and Naturalism’. In Continuum Companion to Philosophy of Language, ed. M. García Carpintero and M. Kölbel. London: Continuum. Zalabardo, José L. 2015. Representation and Reality in Wittgenstein’s Tractatus. Oxford: Oxford University Press. Zalabardo, José L. 2016. ‘Empiricist Pragmatism’. Philosophical Issues 26:441–61. Zalabardo, José L. 2017. ‘Davidson, Russell and Wittgenstein on the Problem of Predication’. In Wittgenstein and Davidson on Thought, Language, and Action, ed. C. Verheggen. Cambridge: Cambridge University Press. Zalabardo, José L. 2018. ‘The Tractatus on Unity’. Australasian Philosophical Review 2:250–71. Zalabardo, José L. 2019a. ‘Belief, Desire and the Prediction of Behaviour’. Philosophical Issues 29:295–310. Zalabardo, José L. 2019b. ‘The Primacy of Practice’. Royal Institute of Philosophy Supplement 86:181–99. Zalabardo, José L. forthcoming-a. ‘Humility and Metaphysics’. Analytic Philosophy. Zalabardo, José L. forthcoming-b. ‘What Is Linguistic Interpretation?’. In Neopragmatism, ed. J. Gert. Oxford: Oxford University Press.

Index For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on occasion, appear on only one of those pages. absolute standard of correctness 153–8, 213–14 abstraction, definition by 168–81 thick 169–70, 172–4, 176–8, 206 abstraction principle 168–74, 178–81, 201, 204–7, 209–10, 212 left-to-right vs. right-to-left readings 171–2, 174–5, 200 abstraction operator 168–70, 176–8 acceptance 92–3, 135, 144–5, 152 Albritton, R. 89 Apperly, I. 119n.14 Asay, J. 80n.14 assertibility conditions 75–7, 83–6, 89 Ayer, A. J. 69–72, 167n.3 Baillargeon, R. 116, 120nn.15–16 Baron-Cohen, S. 115–16 Basic Principle 208–10 belief 45–9, 91–2, 94–5, 107–8, 135 Benacerraf, P. 35n.14 BD discourse 108, 110–11, 113–15, 122, 124–5, 128 Berenstain, N. 193 Berkeley, G. 50n.12 Birch, S. 120–1 Black, R. 190–1 Blackburn, S. 62n.3, 71n.7, 78, 104, 199n.14 Blair, R. J. R. 215n.2 Bloom, P. 120–1 Boghossian, P. 10n.15, 47n.10, 52n.16, 53n.18, 156n.29 Boyd, R. 22–3 Brandom, R. 101–4 Butterfill, S. 119n.14 Byrne, A. 127 Canberra Plan 201–5 Carnap, R. 70–1, 131n.3, 167n.3, 198n.11, 209n.19 Carruthers, P. 119n.13, 126n.25 causal theories of reference 142 causation 124–5, 175–6, 180 charity 114–15, 136–7, 139, 143–4

Churchland, P. 47, 69, 124 cognitive command 158–60 Cohen, J. 92n.11 Collins, J. 151 combinatorialism 187, 190–4, 196–7 compositionality 7, 132–4 conation 127 criterion 76, 84, 88–91 Csibra, G. 117–21, 120n.17, 126n.24 C-synonymy 172–5, 177, 181–2, 200–1 curse of knowledge 120–1 D’Arms, J. 71–2 Dasgupta, S. 190n.6, 196–7 Davidson, D. 64–7, 124–5, 130–2, 134n.9, 135–6, 141n.17, 143n.19, 148n.23, 151 Davis, J. 159n.32 Dawson, A. 30 Dennett, D. 64–5, 68, 74, 94, 94n.14, 107–12, 115, 129n.31, 143–4 desire 45–9, 94–5, 107–8 dispositions 52–3, 57–9 disquotational schema 150–1 doxastic procedure 135, 142–3 Dreier, J. 62n.2, 78n.10, 104–6 Duhem, P. 185 Dummett, M. 75–6, 84, 134 Egan, A. 62n.2, 159n.32 emotivism 70 endorsing 61–2 Evans, G. 126–7, 128n.30 ‘explanation’ explanation 105 expressivism 70–2, 93–4, 105 familiarity 140–1 faultless disagreement 159–60 Fernández, J. 123n.21 Field, H. 42n.3, 43n.8, 131n.3, 138n.13, 177n.20 Fine, K. 105, 168n.4, 176n.16 Fodor, J. 45–6, 47n.9, 53–4 Foot, P. 72n.8 Frege, G. 7–8, 131n.4, 168, 170, 178n.21

232



Frege-Geach problem 71, 98 Frith, U. 115–16 function 98–101

Kraut, R. 156 Kripke, S. 17, 51–3, 57, 74–7, 83–4, 142, 174, 192n.9, 195

Geach, P. 71 G-efficacy 112–13, 118 Gergely, G. 117–21, 120n.17, 126n.24 Gert, J. 119 Gibbard, A. 70–1, 78n.10 Glanzberg, M. 62n.2, 159n.32 G-optimal 112–13, 118 Grandy, R. 142n.18, 144n.20

Laurence, S. 30 Leslie, A. 115–16 Leuenberger, S. 190 Lewis, C. I. 209n.19 Lewis, D. 11n.17, 124n.22, 139, 145n.21, 185–92, 194, 201, 207 Linnebo, Ø. 175n.13 Locke, J. 50n.12 Lycan, W. 74, 84n.1, 110–12

Hale, B. 170n.8, 172n.12 Hare, R. M. 70–1 Harman, G. 17, 20, 26, 110n.5 Harmony 163–8, 175, 178 Hawthorne, J. 79n.13, 193, 202n.16 He, Z. 120n.16 Heller, M. 192n.7 Horgan, T. 22n.3, 28–32, 71n.7, 92n.11, 104 Horwich, P. 72–4, 79n.12, 92n.11, 93n.13, 151n.27, 176n.17, 179n.23 Hume, D. 53, 91–2 Hume’s Principle 168, 176–7 Humility 185–90, 194–5, 198 Hursthouse, R. 123 hybrid policy 119–23, 145 ideal conditions 62–3, 68, 160–1 indeterminacy 66–8, 114–15, 149 Information-Theoretic Semantics 46–9, 52–9, 63–4, 109 instrumentalism 74, 94, 198–9 Intentional Stance 64–5, 68, 74, 94–5, 97, 107–14 interpretation 49–50, 66–7, 130–2, 134, 150–1 interpretivism 63–8, 85–6, 94 intuitionism 33–5 ITS, see Information-Theoretic Semantics

MacBride, F. 170n.8 McDowell, J. 34n.13, 81n.15 MacFarlane, J. 62n.2, 159n.32 Mackie, J. 33n.9 Mancosu 169 Margolis, E. 30 Matthews, R. 124 meaning ground 2 Mentalese 45–8, 53–4, 58–9 Miller, A. 31n.8 mindreading 118n.12, 119n.13, 126 minimal intentional strategy 114–15, 117, 119–22, 127, 129 MI strategy, see minimal intentional strategy modal realism 5–6 Moore, G. E. 14–15, 33–4, 34n.12, 153n.28 Moorean premise 15–16, 27 moral approval 34–5, 60, 62n.2, 69–72, 84–8, 93–4, 96–8, 164–6, 176–9 Moral Twin-Earth 28–32 Moran, R. 127n.26 motivation 33–5, 60, 98, 179 Nagel, T. 81n.15 natural-kind terms 19–24, 28, 41, 45, 58, 174–5, 199, 204–5 naturalism 14, 17–20, 28–33, 42–4, 184 non-cognitivism 69–72, 93, 97–101 normativity 52–4, 57

Jackson, A. 120n.18 Jackson, F. 79n.13, 201–3 Jacobson, D. 71–2 Johnston, M. 97n.15 judgement 4, 6 justification 86–8, 90 justificationism 84–6, 91–2, 106

Onishi, K. 116, 120n.16 operational definition 19 Oppy, G. 79n.13 ostension 138

Kalderon, M. 179–80 Kaplan, D. 11n.16, 62n.2 Kelly, A. 190, 196–7 Köhler, S. 99–101 Kölbel, M. 158

Peacocke, C. 93n.12 Peano, G. 169 Peano assumption 170, 172–3 Perner, J. 120nn.16–17, 124n.22 Pettit, P. 97n.15



233

Poincaré, H. 185 Pollock, J. 84n.1, 86n.4 possible worlds 5–6 Premack, D. 118n.12 Price, H. 79n.13, 81n.16, 100n.17, 202n.16, 203n.17 projection 143–4 propositional attitudes 45–9, 94–5, 180 Putnam, H. 17, 19, 23–4, 30, 138n.13, 172n.9, 174

Smith, M. 33n.9, 79n.13 Smith, P. 126n.25 Soames, S. 55n.19 Sonderholm, J. 30–1 Stalnaker, R. 5n.7, 10–12, 109n.3, 124 Stich, S. 47, 69, 74, 144n.20 Strawson, P. 72 structuralism 193–8 subjectivism 61–2, 96 supervenience 35, 203–4

quasi-realism 78–81, 104, 198n.11, 205–6 quidditism 187, 190–3 Quine, W. V. O. 64–7, 130–2, 135–6, 138nn.13,14, 144n.20, 145n.21, 146n.22, 149–50, 167–73, 198n.11

Tarski, A. 42nn.4,6,7 teleological strategy 117–22 theoretical terms 185–6, 188, 193, 201, 204 Thomasson, A. 176n.15 Timmons, M. 22n.3, 28–32, 71n.7, 92n.11, 104 truth 37–45, 97, 150–1, 179–80 deflationism about 72–4, 78–81, 94 Tarskian definition of 41–5 truth aptitude 158 two-way approach 174–5

Radical Interpretation 64–8, 85–6, 95, 97, 130–2 Radical Translation, see Radical Interpretation Ramsey, F. 73n.9, 109n.3, 125n.23 rationality 110n.4 rejection 92–3 relativism 62n.2, 63, 67, 97 Representational Theory of Mind 45–9, 52–6, 58, 63–4, 109 representationalism 3–8 response dependence 97n.15 Roessler, J. 120nn.16–17, 124n.22 Rorty, R. 102–3, 104n.21 Rosen, G. 199n.14 RR assumption 8–10, 50, 59–60, 69, 77, 81, 97–8, 162–4, 183, 211–12 RTM, see Representational Theory of Mind Russell, B. 4–7, 10, 170 Schaffer, J. 189 Schroeder, M. 71n.7 Scott, R. 120nn.15–16 Searle, J. 71 Sellars, W. 101–2 Shafer-Landau, R. 33n.9, 34n.12, 81n.15 Shah, N. 127n.26 Shaw, J. 140n.15 Shoemaker, S. 127–8, 193 Simpson, M. 105

van Fraasen, B. 199n.14 variance 186–8, 190–2, 194 Velleman, D. 127n.26 verificationism 84–5, 93, 198 Wallace, J. 138n.13 Wedgwood, R. 25n.4 Wellman, H. 118n.12 Whittle, A. 189 Whyte, J. T. 109n.3 Wiggins, D. 72n.8 Williams, J. R. G. 139–40, 141n.16 Williams, M. 2n.4, 99, 101, 143n.19 Wittgenstein, L. 50–1, 58, 74–6, 84, 88–91, 106, 134n.9 Woodfield, A. 140 Woodruff, G. 118n.12 Woodward, J. 124n.22 Wright, C. 76, 84n.1, 86, 86n.4, 87n.5, 89, 91–2, 97n.15, 150, 158–61, 158n.30, 170n.8, 172n.12 Yalcin, S. 12n.18