The Phenomenal and the Representational 978-0198732556

Citation preview

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

The Phenomenal and the Representational

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

The Phenomenal and the Representational Jeff Speaks

1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

3

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 # Jeff Speaks 2015 The moral rights of the author have been asserted First Edition published in 2015 Impression: 1 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: 2014959432 ISBN 978–0–19–873255–6 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY 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.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

This book is dedicated to my dad, the first philosopher I ever knew.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

Preface As its title indicates, this book is about two kinds of properties of perceiving subjects: their phenomenal properties, and their representational properties. In particular, it focuses on three questions: What are phenomenal properties? What are representational properties? What is the relationship between phenomenal and representational properties? My answers to these questions are guided by two ideas, which have both been around for a long time. The first is that experience is transparent, in the sense that attention to one’s perceptual experiences is, or is intimately involved with, attention to the objects and properties those experiences present as in one’s environment. Though the label is due to Moore, versions of this idea can be found in earlier philosophers as well, and it has played a central role in recent work in the philosophy of perception. The second is that one of the roles of perceptual experience is to make objects and properties available to the perceiver for thought. The idea that perception must play this role is of course explicit in empiricists like Locke, and in recent work it has been emphasized by, among others, John McDowell and Mark Johnston. The term ‘availability’ is taken from Johnston’s paper, ‘The Obscure Object of Hallucination.’ These two ideas are central enough to the book that it used to be called ‘Transparency and Availability’—until an anonymous referee pointed out, reasonably, that no one would have any idea what a book with that title was about. If there’s anything original in this book, it’s the idea that, suitably formulated, these two ideas can go quite a long way in revealing the nature of the contents of experience, and the way in which those contents are related to phenomenal properties. Most of the book is concerned with tracing this argumentative path. The rest is concerned with some problems which arise from conclusions reached along the way. These problems are mainly metaphysical, and take us into questions both about the nature of propositions and the nature of phenomenal properties. As is true of any book like this, one can question the starting points, as well as some of the assumptions—often assumptions about whether certain sorts of experiences are possible or, about their veridicality conditions—made throughout. This book is the attempt to think through the consequences of certain assumptions which seem to me plausible starting points for thinking about

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

viii preface perception. It isn’t that I think that these assumptions are certain, or even, in some cases, immune from empirical refutation. They just seemed to me to be the best place to get started. . .

.

This book, while not unusually long, does contain unusually many chapters. The book came together as a result of my thinking about how to put together into a coherent picture various short arguments which seemed individually convincing to me; so this seemed to be the most natural way to present it. But, given the number of chapters, a kind of roadmap for the reader might be useful. Chapters 1 and 2 introduce the two sorts of properties mentioned in the title: phenomenal properties, and representational properties. These chapters emphasize a guiding theme of the book: that when theorizing about these properties, it is best to begin by focusing on the properties of experiencing subjects, rather than on the properties of experiences that those subjects undergo. Chapters 3–8 introduce a number of theses: various versions of intentionalism— roughly, the thesis that phenomenal properties supervene on representational ones—and the thesis that experience is, in a sense to be explained, transparent. They then argue from the latter thesis to a weak intentionalist thesis. Chapter 9 then extends this argument from the base case of visual experience to other types of phenomenal states. Chapters 10–13 ask whether it is possible for distinct states of different types— for example, a visual experience and a belief—to have the same content, but differ phenomenally. I argue that this is possible—a conclusion that rules out certain strong versions of intentionalism. This argument takes us (in Chapter 12) into questions about whether the contents of experience are best understood according to a Russellian or a Fregean view, and (in Chapter 13) into questions about whether a Fregean view of the contents of thought and language is wellsupported by Frege’s puzzle. The somewhat involved argument of Chapter 13 can be skipped without much loss by readers who are less interested in questions about the contents of thought and language than in questions about specifically perceptual representation (though, as I argue in these chapters, the former does have some consequences for the latter). The upshot of the argument to this stage is that phenomenal properties supervene on properties of bearing a certain representational relation (or relations) to Russellian contents. This raises the questions: what are these relations, and what sorts of things are Russellian contents? Chapters 14 and 15 offer initial answers to these questions, identifying the relevant contents with properties of a sort, and the relevant relation with a kind of specifically perceptual self-ascription.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

preface ix The defense of the sort of supervenience thesis just sketched raises the central question of whether phenomenal properties might be identified with representational properties. For reasons which become clear later, this question seems to me intertwined with questions about the scope of perceptual representation: questions about what sorts of things can be constituents of the contents of experience. Chapters 16–23 engage with the latter question. Chapters 16 and 17 introduce the idea that experiences make certain objects and properties available for thought, and attempt to refine this idea into precise principles which, in Chapter 18 and Chapters 20–23, are put to use arguing for specific views about the nature of perceptual contents. Chapter 19 engages with a problem in the metaphysics of propositions to which these views give rise. (Like Chapter 13, this chapter can be skipped without much loss by readers who wish to focus on issues about perception.) So far the argument of the book will have aimed to establish the claim that phenomenal properties supervene on certain relational representational properties. This raises a question which is crucial to attempts to give a physicalist reduction of phenomenal properties: do these relational representational properties all involve the same relation? Chapters 24–27 argue that they do not, and that certain sorts of attentional shifts force us to distinguish between at least two phenomenal relations. In Chapter 28, I turn at last to the question of whether phenomenal and representational properties can be identified. I suggest that there is strong reason to believe that they can be, and discuss two different ways in which such an identification might run—one which accepts the existence of distinct but indiscriminable phenomenal properties, and one which does not. In Chapters 29–33, I introduce the notion of a ‘phenomenally sneaky’ content as a way of arguing for the former option. In the concluding Chapters 34–37, I ask how the investigation of the relationship between the phenomenal and the representational carried out in Chapters 1–33 bears on the two main attempts to give a reductive physicalist account of phenomenal properties: functionalism and identity theory. I argue that the argument to this point presents some serious challenges for functionalist theories, but that the status of identity theories is less clear. And, on that somewhat inconclusive note, the book ends. Some readers of the manuscript suggested that I add a ‘summing up’ chapter at the end. Since I couldn’t think of a way to do this without just repeating what I’ve said here or what I go on to say, I’ve stuck with the admittedly abrupt ending to which you can look forward there. . . .

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

x preface I’ve developed many of the views defended in this book over a period of years, and owe an intellectual debt to more people than I can remember. Though my views were influenced by the writings of many philosophers about perception, I owe a special intellectual debt to the work of Alex Byrne, Mark Johnston, and Michael Tye. I owe something more to Scott Soames—though this book does not have much to do with the dissertation I wrote under his direction ten years ago, Scott provided me with a model of rigor and clarity in philosophy that I hope, to some degree, informs my work. I’m grateful for the comments and questions I’ve received after talks covering versions of parts of this material at Auburn, McGill, NYU, Oklahoma, Oxford, Rice, Rutgers, Western Michigan, and a Central APA session organized by Colin Klein. In addition to these groups, I owe thanks to my colleagues Robert Audi and Leopold Stubenberg, who gave me comments on the manuscript. I also owe special thanks to Antony Eagle, Daniel Immerman, Casey O’Callaghan, and Adam Pautz. Though he probably doesn’t know it, conversations with Antony had a big influence on the way I end up developing my positive view of the relationship between phenomenal and representational properties in the last third of the book. Daniel read through every page of a draft of the book in the spring of 2013; virtually every page of the book has benefited from his eye for unclarity and ability to spot confusion. He’s saved me from mistakes both silly and serious, and deserves much of the credit for what is good in the book. Casey has taught me more about perception than anyone else, and is a big part of the reason that I decided to work in the philosophy of perception in the first place. I benefited greatly from his comments on an earlier draft, and his influence will be obvious in many parts of the book. Adam’s characteristically incisive comments on an earlier draft showed me, among other things, the need to add Chapter 2 of the book, and to think through much more seriously than I had what it means to say that experience is representational. The book is, I hope, much better as a result of the changes made in response to his comments. Most importantly, I thank Elyse, Amelia, and Violet, the sources of more happiness than I could have imagined.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

Table of Contents Part I. Two Kinds of Properties of Subjects 1. Phenomenal Properties

3

2. Representational Properties

8

Part II. From Transparency to Intentionalism 3. Varieties of Intentionalism

21

4. Transparency

26

5. Two Impossible Scenarios

35

6. Phenomenism and Phenomenal Variance

41

7. The Interpersonal Constraint, the Time Constraint, and the Memory Constraint

44

8. Phenomenal Variance and Property Variance

52

9. Extending the Argument to Other Phenomenal States

55

Part III. Intermodal Intentionalism and Nonconceptual Content 10. Belief and Intermodal Intentionalism

59

11. Direct Arguments for Nonconceptualism

65

12. A Dilemma for Fregeanism about Sensed Contents

70

13. Frege’s Puzzle and the Content of Thought

76

Part IV. The Metaphysics of Representational Properties 14. Properties and the Nature of Russellian Contents 15. Properties and the Attitudes

99 108

Part V. Availability and the Scope of Perceptual Representation 16. Availability

121

17. Demonstratives and the Availability Requirement

126

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

xii table of contents 18. The Perceptual Representation of Objects

133

19. The Problem of Contingently Existing Contents

137

20. The Perceptual Representation of Natural Kinds

143

21. Independence, Fallibility, Anti-Circularity

147

22. Appearance Properties

155

23. Relativist Views of Sensible Qualities

166

Part VI. How Many Phenomenal Relations? 24. Phenomenal Relations

173

25. The Distinctions Between the Senses

177

26. Binding and Bodily Sensations

186

27. Shifts in Attention

189

Part VII. Phenomenal Identity and Indiscriminability 28. Identifying Phenomenal and Representational Properties

203

29. Phenomenal Content

208

30. Phenomenal Content and Phenomenal Continua

219

31. Perceptual Constancies and Phenomenal Match

226

32. Phenomenal Content and the Representation of Change

232

33. Indiscriminable Phenomenal Properties

235

Part VIII. The Reduction of Phenomenal Properties 34. Two Kinds of Theories of Phenomenal Properties

243

35. Functionalist Theories of Content

245

36. Functionalist Definitions of Phenomenal Relations

256

37. Identity Theories

263

Bibliography Index of Names Index of Topics

273 279 281

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART I

Two Kinds of Properties of Subjects

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

1 Phenomenal Properties It’s now standard to think about perception in terms of two sorts of properties: phenomenal properties, and representational properties. What are phenomenal properties? Most of us think that people and other animals are conscious, while tables are not—that is, there’s a certain property, being conscious, which we and other animals instantiate that tables don’t. There is something that it is like to be you or me; but nothing that it is like to be a table. But once we get this far, it’s hard to deny that this property of being conscious is a determinable—there are, after all, very many different ways of being conscious. Phenomenal properties, as I will use the term, are the determinates of this determinable. Phenomenal properties are the various ways of being conscious—much as the determinate colors are the various ways of being colored. What are phenomenal properties, so construed, properties of? Given this way of introducing the term, it should be uncontroversial that phenomenal properties are properties of subjects of experience—of things like you and me. A central aim of this book is to make some progress in understanding the nature of these properties of subjects. This focus on the properties of subjects runs counter to a tendency in some of the recent literature on perception to focus on the phenomenal character of experience. If we were to adopt this as our starting point, we would be focusing, not on a special class of properties of subjects of experience, but on a special class of properties of experiences themselves. Now, it may be that an investigation into the nature of phenomenal properties will end up investigating certain special properties of experiences; but it seems to me that it can’t be the right place for it to start. The initial contrast which gives us a grip on the notion of a phenomenal property is a contrast between the properties I have and the properties that tables have—not between the properties that my experiences have and the properties that tables have. When we talk about ‘what it is like,’ we are talking about properties of subjects, not of experiences; while there is something that it is like to be me right now, there is nothing that it is like to be one of my experiences.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

4 phenomenal properties Given this, why might study of phenomenal properties of subjects lead us to talk about certain properties of experiences? The following is a natural line of thought: subjects undergo various experiences; and which phenomenal properties those subjects instantiate depends on which experiences they undergo. Hence, one might think, we can explain what it is for a subject to instantiate certain phenomenal properties in terms of certain properties of the experiences that subject undergoes. ‘Phenomenal character’ can then be introduced as a term for those properties of experiences which fix the phenomenal properties of the subject of that experience. I think that many uses of ‘phenomenal character’ in the literature fit this way of thinking about it pretty well.1 But on this sort of use of ‘phenomenal character,’ it is not a name for a property on which we have an independent grip; it is, rather, a kind of theoretical term introduced partly by means of our grasp of the conceptually prior notion of the phenomenal property of a subject of experience. This fact about ‘phenomenal character’ leads to three dangers in taking the study of the phenomenal characters of experiences, rather than the phenomenal properties of subjects, as our starting point. The first is simply that it limits our theoretical options; it’s just not obvious that the phenomenal properties of subjects are best analyzed in terms of the properties of certain events, experiences, which involve those subjects. The second is that it can make the task of giving a theory of consciousness seem easier than it in fact is. If we take a theory of consciousness to be a theory of phenomenal character, we ignore the fact that a theory of phenomenal character does not, itself, provide a theory of the phenomenal properties of subjects. To get from a theory of phenomenal character to a theory of phenomenal properties, we need to explain what it is for a subject to have an experience which instantiates the relevant phenomenal character—that is, we need to explicate the relation between the subject and that phenomenal character which suffices for the subject to instantiate the phenomenal property corresponding to the phenomenal character. The last is that, as with other terms introduced in this way, there is a danger in theorizing about phenomenal character that one will lose sight of the theoretical role which gave the term its meaning in the first place. When that happens, we risk either falling into a merely verbal dispute about ‘phenomenal character’ or losing sight of the domain of facts—namely, the phenomenal properties of

1 For two cases in which the term is explicitly introduced in this way, see Byrne (2002) and Chalmers (2011), 104.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal properties 5 subjects, the facts about what it is like to be a conscious subject—which made this seem like a worthy topic of investigation in the first place. So while in what follows it will sometimes be convenient for me to express myself in ways which suggest that we are interested in properties of experiences, in every case the relevant claims can be reformulated as claims about the properties of the relevant subjects—and, in each case, that’s how the claims should be understood. I’ll return to instances of each of these three potential dangers of taking properties of experiences as our starting point in the chapters to follow. It is hard to avoid using the same words to describe phenomenal properties of subjects as one uses to describe the properties those subjects are representing as out in the world. But, of course, these are quite different sorts of properties. The property of being red is a property of (if anything) surfaces, which may or may not be conscious; the phenomenal property that I instantiate when I look at a red expanse is a property that only certain sorts of conscious beings can have. In order to keep these two sorts of properties separate, I’ll adopt the convention of using small caps to designate phenomenal properties. So, for example, RED is the phenomenal property that I experience when looking at a red expanse.2 It should be clear, I take it, that while there is no bar to a thing being both red and RED, being red is a quite different thing than being RED. So far I’ve been assuming that subjects really do instantiate phenomenal properties. Is this a reasonable assumption? It’s rare nowadays for philosophers to flatly deny that there is something that it’s like to have a certain sort of experience—and for good reason, since it seems blindingly obvious that there is. What’s a bit more common is for philosophers to deny that this is genuinely a matter of subjects instantiating certain phenomenal properties. For if ‘what it’s like’ is to be cashed out in terms of the instantiation of phenomenal properties, then it must make sense to ask whether distinct subjects, or subjects at two distinct times, are instantiating the same phenomenal property. And this is something which many philosophers have, for various reasons, wanted to deny.3 But there is a strong intuitive case to be made for the claim that we do instantiate phenomenal properties. For if we grant that there’s something that it’s like to be me right now, then surely we should also grant that we can compare what it’s like to be me right now with what it’s like to be me one second ago—after all, we regularly 2

I will, however, for ease of exposition sometimes be a bit loose in the way that I use terms like ‘RED,’ in the sense that I will sometimes use them as predicates which express a phenomenal property, and other times use them as abstract singular terms for those phenomenal properties. In each case, I think, it is pretty clear how the relevant points could be made without this kind of dual usage. 3 See, for example, Dummett (1975) and Stalnaker (1999).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

6 phenomenal properties make comparisons involving changes in ‘what it’s like’ over time, as when we attend to a throbbing pain. But if we grant this, then we’ve granted that we can compare what it’s like for me at t1 with what it’s like for me at t2. And if we can do this, then it must be possible for what it’s like for me at t1 to be identical to, or distinct from, what it’s like for me at t2. And phenomenal properties just are the properties which are singled out by phrases like ‘what it’s like for me at t1.’ That said, there are powerful arguments against the existence of phenomenal properties. One such argument comes from the apparent possibility of phenomenal continua—a series of n experiences in which the first is readily discriminable to the subject from the last, but in which each experience is indiscriminable from those which immediately precede and succeed it in the series. Given our informal glosses on the notion of a phenomenal property—as what it’s like for a subject at a time, or how things feel ‘from the inside’ for a subject at a time—the following two principles can seem pretty plausible: If what it is like to be A at t1 is indiscriminable from what it is like to be A at t2, then the phenomenal properties instantiated by A at t1 ¼ the phenomenal properties instantiated by A at t2. If what it is like to be A at t1 is discriminable from what it is like to be A at t2, then the phenomenal properties instantiated by A at t1 6¼ the phenomenal properties instantiated by A at t2. But if we accept these principles and the existence of phenomenal continua, then the claim that subjects of experience instantiate phenomenal properties entails a contradiction. For consider the first time t1 in a phenomenal continuum. By the description of a phenomenal continuum and the second principle above, it follows that the phenomenal property F1 instantiated at t1 6¼ the phenomenal property Fn instantiated at tn. But by the description of phenomenal continua and the first principle above, it follows that F1 ¼ F2, and F2 ¼ F3, and . . . Fn1¼ Fn. But then by the transitivity of identity it follows that F1 ¼ Fn—which contradicts our previous conclusion. This is indeed an argument which any believer in phenomenal properties must confront.4 But it seems to me that it is easier to reply to arguments like this in a 4

Indeed, it’s an argument which must be confronted by anyone who believes that there’s something it’s like to be the subject of a conscious experience. After all, our two principles about discriminability seem no less plausible when applied directly to ‘what it’s like to be A at t’ than they did when applied to ‘the phenomenal properties instantiated by A at t’—and identity is transitive, whether we are talking about identity claims involving properties or identity claims involving whatever the denier of phenomenal properties takes to be denoted by phrases like ‘what it’s like to be me right now.’

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal properties 7 principled way with a particular view of phenomenal properties in hand. Hence my strategy will not be to try to defuse skeptical arguments like this at the start, but rather—given the prima facie plausibility of the existence of phenomenal properties—to provisionally taken them on board at the outset and try, over the course of the book, to develop a theory of phenomenal properties which is suited to handle objections like this one. (I return to the problem of phenomenal continua in particular in Chapter 30.) With this introduction to the idea of phenomenal properties in hand, let’s turn to the second kind of property named in the title: representational properties.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

2 Representational Properties Many different kinds of things are said to instantiate representational properties; among them are sentences, utterances, mental states, and propositions. Our focus, though, is on the representational properties to which phenomenal properties stand in the closest relation. Since phenomenal properties are properties of conscious subjects, our interest is in the representational properties instantiated by those subjects. The paradigm examples of representational properties of subjects are their propositional attitudes; the properties ascribed by sentences like A believes that S or A supposes that S. These properties are, on standard views, relational properties: they are a matter of the subject of the ascription standing in a propositional attitude relation—like believing or supposing—to a proposition. As with phenomenal properties, the focus on properties of subjects is a departure from the usual way of talking about representational properties in the philosophy of perception. The most commonly used phrase for talking about these properties is ‘the content of experience’—which suggests that the relevant representational properties are not relational properties of subjects, but relational properties of experiences. But even granting that experiences instantiate representational properties of this sort, there is no conflict between the claim that experiences have these properties and the claim that subjects of experience, in virtue of having the experience, stand in relations to the relevant propositions as well. Indeed, it follows from the fact that I am having an experience which stands in some representational relation R to a proposition that I stand in the following relation to a proposition: having an experience which is R-related to that proposition. So we can focus on the representational properties of subjects without any loss of generality. And as with phenomenal properties, I think that something stronger can be said: I think that we have a firmer grip on the relational representational properties of subjects than on the alleged relational representational properties of experiences. We have a grip on what it means for a subject to stand in the belief relation or the supposition relation to a proposition; this helps us to get some

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

representational properties 9 initial grip on the thesis that subjects of experience, just in virtue of having that experience, stand in some distinctive relation to a proposition. By contrast, it’s often less than clear just what entities’ experiences are—and hence, partly for that reason, less than clear what it would take for one to stand in a relation of the right sort to a proposition. Now, to say that the representational properties of subjects are relations to propositions is not to explain what we mean by ‘representational property.’ After all, even if representational properties of subjects are relations to propositions, not all relations in which subjects stand to propositions are representational properties of those subjects. Coexistence is a relation in which I stand to every proposition that currently exists; that hardly means that I have a representational property involving each of these propositions. This makes it clear that the representational properties of subjects—their propositional attitudes—are a proper subset of their relations to propositions. This leads to the question: what does it take for a subject’s standing in a certain relation to a proposition to count as a genuine representational property—a genuine propositional attitude—of that subject? Or, in other terms: what does it take for a given relation between subjects and propositions to be a genuine representational relation—a genuine propositional attitude relation? Despite the ubiquity of talk about propositional attitude relations in contemporary philosophy, this question has received little attention.1 It is not an easy question to answer. But it is a pressing question in the present context because, without some account of the restricted class of relations to propositions we’re interested in, certain fundamental questions about the relationship between the phenomenal and the representational are trivialized. Consider, for example, the much-discussed question of whether phenomenal properties supervene on representational properties. If by ‘representational properties’ we simply mean ‘relations to propositions’ then this supervenience thesis is trivially true: at

1 A recent exception is Grzankowski (2014), who proposes that ‘for any attitude V, V is a propositional attitude just in case for any subject S and proposition p such that S stands in V to p, if p were true, then things would be as S V’s them to be.’ (5) This looks more like a test for whether a verb is a propositional attitude verb than whether a relation is a propositional attitude relation; if we try to turn it into a condition on relations, we end up quantifying into both subject and predicate position. And as a test for whether a given verb expresses a propositional attitude, it, while promising, is both insufficiently general and open to counterexample. It is insufficiently general because some propositional attitude relations aren’t expressed by sentences of the form ‘A V’s that S’—examples are the relations expressed in sentences like ‘It appears to A that S’ or ‘It seems to A that S.’ Other verbs, like ‘objects’ and ‘asks,’ seem to just be counterexamples to the test: from ‘A objects that S’ it does not follow that ‘There is a way that A objects the world to be.’ The latter, insofar as it makes sense, expresses something quite different than the former.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

10 representational properties some time t, I instantiate some phenomenal property F if and only if I, at t, stand in the relation corresponding to the open sentence x exists at the same time as y is true to the proposition that I am F. This is enough to establish the supervenience of phenomenal properties on the relations in which subjects stand to propositions— but it is obviously not enough to establish the supervenience of the phenomenal properties of subjects on their representational properties! Here is one way to make some progress on this issue. Intuitively, one mark of representational propositional attitude relations is that there is a tight connection between certain properties of the propositions to which a subject stands in those relations, and certain properties of the subject herself. So, for example, in the case of judgement, we know that, necessarily, for any subject that stands in the relation of judgement to a proposition p, if p is true, then that subject has the property of having correctly represented the world. Conversely, if p is not true, that subject has the property of having misrepresented the world. By contrast, the fact that I stand in the relation of coexistence to the proposition that 2 þ 2 ¼ 5 plus the fact that that proposition is false does not jointly entail that I have misrepresented the world. This is surely part of what makes judgement, but not coexistence, a representational relation. Let’s say that relations which are like judgement in this respect are truthsensitive relations. This notion can be defined as follows: R is a truth-sensitive relation iff necessarily, if A is R-related to p, then if p is not true, A is misrepresenting the world, and if p is true, then A is correctly representing the world. It seems plausible that—with a few qualifications2—truth-sensitivity is a sufficient condition for a relation to be representational. If standing in a certain 2 The qualifications involve relations which satisfy the above condition only because standing in those relations to a proposition entails standing in some other truth-sensitive relation to a modally equivalent proposition. To construct such a relation, begin with some truth-sensitive relation—say, judgement. Then we can define a relation using judgement as follows:

x judges that grass is green and y is the conjunction of some necessary truth with the proposition that grass is green. The relation corresponding to this open sentence comes out truth-sensitive, and it is a relation in which I presently stand to the infinitely many conjunctions of a necessary truth with the proposition that grass is green. But that hardly means that I have a representational property involving each of these conjunctions. The problem is that, while this is a genuine relation between subjects and propositions, it is a wholly derivative one, in the sense that I stand in this relation to the proposition that arithmetic is incomplete just in virtue of standing in the relation of judgement to the proposition that grass is

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

representational properties 11 relation to a false proposition is sufficient for a subject to misrepresent the world, that must be because standing in that relation to a proposition is a way of representing that proposition as being true. (As we’ll see in Chapter 15, we’ll also find reason to investigate some relational representational properties of subjects which are not relations to propositions. There I’ll return to the question of how the definition of truth-sensitivity might be generalized to handle relations of that sort.) This definition of truth-sensitive relations relies on a prior grasp of a kind of representational property. The idea behind this definition is that we can use our antecedent grip on the idea of a subject misrepresenting, or representing correctly—of a subject getting things right, or getting things wrong—to single out a class of representational relations. For this reason, the way that I’ve picked out the truth-sensitive relational properties which I want to talk about depends essentially on the fact that we are focusing on the representational properties of subjects, and not those of experiences. That is because it depends essentially on the idea that we have a pre-theoretic grasp of the notion of a subject of experience misrepresenting, or correctly representing, her environment. The idea that, in having certain experiences, a subject is correctly representing or misrepresenting her environment is not one which requires a great deal of philosophical indoctrination to accept. By contrast, if we focused on the properties of experiences, we would be relying on an antecedent grasp of the veridicality conditions of experiences—and it could reasonably be objected that ‘veridicality conditions of an experience’ is itself a philosophical term of art in need of explication.3 Truth-sensitivity is, even if sufficient, plainly not a necessary condition for a relation to be representational. One does not misrepresent when one intends that p be the case, even if p is not true—but intentions are representational mental states nonetheless. A sufficient condition will be enough for our purposes— though the importance of our present inability to give necessary and sufficient conditions for representational relations will resurface in Chapter 27. When we are interested in the relationship between phenomenal and representational properties, we aren’t just interested in the relationship between a subject’s phenomenal properties and any of the relations she bears to various propositions; we are, I think, interested in the relationship between her phenomenal properties and the truth-sensitive relations she bears to propositions. But of green. When I talk about truth-sensitive relations above, I’m focusing only on non-derivatively truth-sensitive relations. I ignore this complication in what follows. Thanks to Adam Pautz for helpful discussion here. 3

This is one difference between the present explication of perceptual representation and the ‘accuracy conception’ criticized in Pautz (2009b).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

12 representational properties course there is more than one type of truth-sensitive relation: belief, judgement, assertion, and conjecture are all truth-sensitive, but are all importantly different from each other. Just which truth-sensitive relation will we be focusing on in what follows?4 One might think that this is a question we could answer in the course of inquiry rather than at the outset. Consider, for example, the question of whether phenomenal properties supervene on representational properties. Couldn’t we just ask this question by asking whether there are any truth-sensitive properties of subjects which are such that any subjects alike with respect to that class of truth-sensitive properties are also alike with respect to their phenomenal properties? Why try to decide in advance what the best candidate for this class of truth-sensitive properties is? There are two problems with this way of proceeding. The first is strategic. I’ll be arguing in what follows that the phenomenal properties of subjects stand in a very close relation to certain of their truth-sensitive properties. But to mount an argument of this sort, we’ll need some assumptions about the relevant truthsensitive properties; and for these assumptions to be plausible, we will need more information about what the relevant truth-sensitive relations are. The second is more substantive. My concern, ultimately, will not just be with the question of whether the phenomenal supervenes on the representational, but with the question of whether phenomenal properties might be identical to representational properties. That means that we need to identify some type of truth-sensitive relational property which is plausibly not just necessarily correlated with phenomenal properties—but also a plausible candidate to give the identity of those properties. What kind of truth-sensitive property could this be? Consider a visual illusion—say, an experience of a surface as red which is in fact green. It seems quite plausible that, in undergoing this sort of illusion, the subject is misrepresenting the color of the relevant surface. It is plausible that if a subject is misrepresenting the world, he is doing so in virtue of standing in a truth-sensitive relation to some false proposition. Since the subject of our illusion is misrepresenting the color of the surface, it seems plausible that she is in some truthsensitive relation to a false proposition about the color of that surface. Could this truth-sensitive relation be judgement, or belief? For familiar reasons, it seems not, since one can have illusory experiences which one knows to be illusory—like the well-known Müller-Lyer illusion, in which two lines 4 My answer to this question closely resembles various discussions of perceptual representation in the literature; it is perhaps closest to Byrne (2009), especially }III and }VI.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

representational properties 13 which are the same in length appear to differ in length. A subject might be quite familiar with this sort of illusion and, when presented with an instance of it, be quite sure that the lines are of equal length, and hence be quite unwilling to judge, or form the belief, that the lines are of unequal length. Despite the fact that the subject is not making any false judgements or forming any false beliefs, it seems clear that the subject of this illusion is, at some level, misrepresenting her environment. But then in this sort of case, the relevant truth-sensitive relation in which the subject stands to a false proposition can’t be judgement or belief. One might reply that, even in the case of the Müller-Lyer illusion, we have some inclination to believe that the lines are of unequal length, or are such that, were certain counterfactual conditions to obtain, we would believe that the lines are of unequal length.5 The problem, though, is that being inclined to believe p (or being such that, were certain conditions to obtain, one would believe p) is not a truth-sensitive relation: being such that under certain conditions one would believe p, does not, even if p is not true, entail that one has misrepresented the world. Hence, even if it is true that one is, in the relevant sense, inclined to believe that the lines are unequal, that would not explain the initial datum that the subject of the Müller-Lyer illusion is, just by having that illusory experience, misrepresenting the length of the lines. This suggests that, in virtue of having a visual experience, one stands in some truth-sensitive relation to a proposition—and that this relation cannot be the relation of judgement or belief, or being inclined to judge or believe. By varying the example, we could argue for similar conclusions about auditory experience, tactile experience, and experiences of the other sense modalities. Hence this line of argument suggests that, in virtue of having a perceptual experience, the subject of that experience stands in some distinctive truth-sensitive relation to a content which is not reducible to judgement or belief. I will call these sensing relations, and will call the (monadic relational) property of standing in some sensing relation to a particular content a sensing property. So sensing properties will be relational properties of subjects of experience. For now, I will remain neutral on the question of whether we need to recognize various distinct sensing relations—perhaps corresponding to the different modalities of sense experience. That is a question to which I will return in Chapter 25. So, while it will be useful to talk about subjects visually sensing or auditorially sensing, I’m leaving open the question of whether, in the end, these are just two names for the same relation between subjects and contents.

5

See Armstrong (1968), 221–5.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

14 representational properties When we’re asking about the relationship between representational and phenomenal properties of subjects, we’ll be asking in the first instance about the relationship between phenomenal properties and sensing properties. Occasionally, usually to bring the discussion into contact with existing literature, I will use the more standard ‘content of experience’ language; that language can be translated into my preferred way of talking about the relevant representational properties of subjects via equivalences like the following: A senses that p iff A has an experience with the content that p A visually senses that p iff A has a visual experience with the content that p etc. Readers more comfortable with the standard ‘content of experience’ talk can use these to translate my usual way of talking into theirs. This way of setting our topic is not an attempt to give a ‘real definition’ of sensing relations. What I claim is only that I’ve succeeded in picking out a representational relation, or class of representational relations, about which we can now go on to theorize. And (as is the case generally) in so doing we do not have to limit ourselves to assumptions that follow from the description used to initially pick out the topic of our theorizing. So, for example, I might pick out the color red as the color of surfaces which cause me to instantiate certain phenomenal properties, but then go on to advance claims about redness (like the fact that surfaces which are red have certain reflectance properties) which don’t follow from the initial reference-fixing fact about my phenomenal properties. The same goes for theorizing about a planet, or theoretical particle—even if our initial ability to refer to that planet or particle utilized a description, our subsequent theorizing does not have to be simply a matter of unpacking that definition, and can include elements which are not derivable from it. Just so, in this case, I might notice that the propositions to which I stand in the visually sensing relation are closely connected to the phenomenal properties I instantiate while having the relevant visual experiences. That fact about that sensing relation is not derivable from the fact that it is truth-sensitive or from particular judgements about cases in which agents, in virtue of standing in that relation to a proposition, are misrepresenting the world. But that is just an instance of the general point that one’s theory of x needn’t simply be a way of unpacking the way in which one picked out x in the first place.6 6

Here’s another example, to which I will return later: once we notice that experience involves certain distinctive relations to propositions, we can ask questions about the relationship between the contents to which I am related by experience, and the contents to which I can stand in the relations of thought or belief. One might (I do) find it plausible that the fact that the relations to contents provided by experience explain our abilities to stand in other propositional attitude relations, like

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

representational properties 15 This elaborate way of picking out the relevant representational relations, and corresponding representational properties of subjects, is necessary because of the admittedly awkward fact that there is no English expression which perfectly expresses the sensing relations I’ve been trying to single out. If one is theorizing about belief, one doesn’t need to go to great lengths to explain just which representational relations one intends to talk about: one simply expresses that relation using the English word ‘believes.’ Unfortunately, nothing quite so easy is available in the case of the representational relations I’ve been trying to single out. The closest approximations are expressions like ‘appears that,’ ‘seems that,’ or ‘visually represents that.’ I’ll use these locutions occasionally throughout—but they are simply devices for picking out the sensing relations singled out above, rather than ordinary language expressions essentially involved in the explication of those relations.7 Not everyone, of course, will agree that I have succeeded in picking out a class of truth-sensitive representational properties about which we can go on to make various claims. This is because many philosophers doubt that there are any representational properties of subjects which are especially associated with experiences—or, to use the more standard way of describing this position, there are many philosophers who deny that experiences have contents. I think that the best response to this objection is to argue that the same considerations which count in favor of beliefs having content—that is, of believing being a representational relation—count in favor of the existence of representational sensing relations. Why do we think that beliefs have contents? Principally, I suggest, for two reasons. First, in having a belief, you can represent the world correctly or incorrectly. Thinking of belief as a truth-sensitive representational relation to a proposition explains this: you represent the world correctly iff the proposition to which you stand in the belief relation is true. Second, beliefs bear interesting relations to other sorts of states which seem to have contents, and these relations are most easily understood if we assign contents to beliefs as well. Consider, for example, assertion. If I assert that

thought and belief, to those contents. And one might then (as I do in Chapters 16–23) use this fact to argue for various theses about the nature of the propositions to which one is related by experience. But there is no requirement that these theses about the scope of perceptual representation be themselves derivable from the facts about the conditions under which subjects of experience represent correctly or incorrectly which were used to establish the existence of the relevant representational relations in the first place. 7 So, to use the terminology of Travis (2004), I’m not assuming that the relevant sensing relations are ‘looks-indexed.’

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

16 representational properties South Bend is boring in the winter, you might wonder why I’ve asserted this. A common, if partial, explanation of my so doing might be that I believe that South Bend is boring in the winter. But why is the latter an explanation of this assertion? Surely part of the story here involves some commonality between the assertion and the belief—and a very natural suggestion for this commonality is that belief and assertion are two different relations in which I stand to the very same content. But these reasons for thinking that belief is a representational relation seem to apply equally well to perceptual experience. After all, as the example of illusion discussed earlier suggests, one can represent the world correctly or incorrectly by having an experience. As in the case of belief, we can, if we think of experiences as involving a representational relation to a proposition, think of the subject as correctly representing iff the proposition to which the subject stands in the relevant truth-sensitive relation is true. And, like beliefs, perceptual experiences stand in interesting explanatory relations to other contentful states which are most easily understood if we also think of experiences as having contents. Just as I sometimes assert things because I believe them, so I sometimes believe things because I experience them. I might, for example, come to believe that a wall is red on the basis of a visual experience of a wall which, unbeknownst to me, is illuminated by red light. But how, exactly, does the experience explain the belief? As in the case of belief and assertion, it is quite natural to think of this explanation as involving some commonality between the experience and the belief—and, once we’re this far, a natural candidate for the commonality is that they are distinct relations to a single proposition—namely, the proposition that the wall is red. These are not knockdown arguments. Nothing I’ve said rules out the possibility that we might be able to construct alternative, non-content-involving treatments of the veridical/illusory distinction or of the relations between experience and other contentful mental states. But the parallels between experience and belief do show, I think, that our initial presumption should be in favor of the claim that there are distinctive representational properties of subjects associated with experience. If so, then the interesting question is whether there are any plausible arguments against this position. It’s a bit rare to come across an ‘in principle’ argument that perceptual experiences can’t have contents.8 More often, one comes across arguments which aim to show that there is no plausible treatment of the representational properties of certain specific sorts of experiences, or arguments that attributions 8

An exception is Travis (2004). For what seems to me to be a convincing reply, see Byrne (2009).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

representational properties 17 of representational properties to subjects of experience fail to earn their explanatory keep. As with arguments against the existence of phenomenal properties, it seems to me that the best way to respond to arguments of this sort is not to respond to them in a vacuum, but rather to respond to them in the context of a theory of the relevant representational properties. This is not to minimize the challenges which face any attempt to give a plausible account of sensing properties—as we’ll see later, there are many. But the only way to see whether these challenges can be met is by trying to construct a theory of those representational properties which meets them. So far, I don’t take myself to have accomplished very much. I’ve explained what I’m talking about when I talk about phenomenal and representational properties of subjects, and have given some preliminary reasons for thinking that there are such properties and that subjects instantiate them. But even if we are willing to take this on board, at least provisionally, this leaves pretty much open the questions of what the natures of the relevant phenomenal and representational properties are, and how they’re related to each other. Those are the questions which will occupy us for the rest of this book.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART II

From Transparency to Intentionalism

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

3 Varieties of Intentionalism Now that we have a rough fix on the representational and phenomenal properties we’re interested in, a question naturally arises: how, if at all, are these properties related? Much recent discussion in the philosophy of perception has focused on the thesis of intentionalism (sometimes called ‘representationalism’), which is the view that phenomenal and representational properties stand in a certain close relationship. It is, however, sometimes less than clear exactly what this close relationship is supposed to be. ‘Intentionalism’ has been used to stand for (among others) the theses that the representational explains the phenomenal, that it determines the phenomenal, and that it is identical to the phenomenal. This is confusing—not least because the last of these is inconsistent with the first. Rather than beginning with these more ambitious claims, I’ll begin with a supervenience claim: the claim that, necessarily, any subjects which differ in their phenomenal properties also differ in their representational properties. All of the theses just mentioned go beyond this bare supervenience claim; but if we want to know whether any of these theses are true, a good place to start is with the question of whether this supervenience claim is true. But there is not just one thesis of the supervenience of the phenomenal on the representational. As others have noted, these supervenience theses vary on the following two dimensions: • Local vs. global. Whether the supervenience thesis is supposed to hold for all mental states, or just for some proper subset of those states (for example, for perceptual experiences but not for bodily sensations). • Intramodal vs. intermodal. Whether the supervenience thesis is supposed to hold only for pairs of states of the same type (for example, of the same sense modality), or for arbitrary pairs of states. This more or less standard description of these dimensions of variance naturally accompanies the picture according to which we are interested in phenomenal and representational aspects of experiences. How can we state

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

22 varieties of intentionalism them in our preferred idiom, which focuses on the properties of experiencing subjects? In these terms, the global intentionalist thesis is an unrestricted thesis about the supervenience of phenomenal properties on representational properties: any two subjects who differ phenomenally in any way whatsoever must also differ in some representational property. The defender of a merely local intentionalist thesis, by contrast, carves out some special class of phenomenal properties for which the supervenience theses hold. The defender of any such merely local thesis, then, owes us some way of categorizing phenomenal properties. One might, for a start, try to distinguish visual phenomenal properties from auditory phenomenal properties, and distinguish perceptual phenomenal properties more generally from the phenomenal properties associated with bodily sensations. I think that these categories are clear enough for us to get a rough handle on which phenomenal properties are being referred to. ‘Visual phenomenal properties,’ for instance, is at least as clear as locutions one finds in the literature like ‘phenomenal character of a visual experience’—a property is a visual phenomenal property iff it is a phenomenal property a subject would instantiate in virtue of having a visual experience with a certain phenomenal character.1 In what follows, I’ll occasionally use phrases like ‘visual phenomenal properties’ or ‘color phenomenal properties’ in this way. I’ll be using them in a purely ‘reference-fixing’ way—just to draw the reader’s attention to a certain class of phenomenal properties of subjects. This sort of informal use of these terms is consistent with doubt about whether these distinctions between types of phenomenal properties can bear any serious theoretical weight, as they would have to if some local intentionalist thesis were true, and global intentionalism false. This is a point to which I’ll return later. In our terms, the intermodal/intramodal distinction is not about types of mental states and their properties, but about which relations between subjects and propositions have to be in the supervenience base for the facts about phenomenal properties. A good way to grasp the distinction is to begin by noting that there are various levels of detail at which one can describe a subject’s total representational state. One might, for instance, at a very coarse level simply list the propositions to which the subject stands in some propositional attitude relation or other. Or one might distinguish the propositions which the subject Not that I think that ‘phenomenal character of experience’ talk is more fundamental than ‘visual phenomenal properties’ talk—quite the opposite. But readers more accustomed to the latter can use this schema as a way of translating between the two. 1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

varieties of intentionalism 23 stands in truth-sensitive relations to from those to which the subject bears some non-truth-sensitive representational relation. Or one might introduce distinctions within the truth-sensitive attitudes—for instance, between the propositions which the subject believes or judges and the propositions to which the subject stands in some sensing relation. Or, cutting things more finely still, one might distinguish between various sensing relations, and list the propositions which the subject visually senses, auditorially senses, and so on. The issue between intramodal and intermodal intentionalisms is about how many distinctions between representational relations we need to make to provide a supervenience base for the phenomenal properties. Intuitively, this will tell us which modes of representing—which ways of being related to propositions—play a role in fixing a subject’s phenomenal properties. It should be clear that the global/local and the intermodal/intramodal distinctions are both matters of degree; and it should also be clear that they are orthogonal to each other. The global/local choice is about which phenomenal properties supervene on any representational properties. The intramodal/intermodal distinction is about, given some answer to the global/local question, just which representational properties are the minimal supervenience base for the chosen class of phenomenal properties. Simplifying away from the graded nature of the distinctions, this pair of distinctions generates four intentionalist supervenience theses, the entailment relations between which can be represented as follows:

Four kinds of intentionalism

Global intramodal intentionalism Local intramodal intentionalism

Global intermodal intentionalism Local intermodal intentionalism

Each of these four theses may be further subdivided according to the following threefold distinction: • Time-restricted intrapersonal vs. time-unrestricted intrapersonal vs. interpersonal. Whether the supervenience thesis is supposed to hold for pairs of

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

24 varieties of intentionalism phenomenal properties instantiated by arbitrary subjects, or only for pairs of phenomenal properties instantiated by a single subject, or only for pairs of phenomenal properties instantiated by a single subject within a certain restricted time interval. Again, there are obvious entailments within this dimension—every interpersonal intentionalism will entail the corresponding time-unrestricted intrapersonal intentionalism, and the latter will entail the corresponding time-restricted intrapersonal intentionalism—but this distinction is independent of the two distinctions mentioned above, giving us twelve intentionalist theses—three corresponding to each box in the above chart.2 As mentioned earlier, one of the central questions of the next few chapters will be: which is the strongest true thesis among these varieties of intentionalism? But first let’s ask: Why bother trying to figure out exactly which is the strongest true intentionalist supervenience thesis? Supervenience claims are not, in themselves, the most interesting claims in the world. But there are a few reasons for thinking that they are nonetheless of fundamental importance. First, as noted earlier, many of the most interesting claims in the philosophy of perception—like the claims that phenomenal and representational properties are identical, or that the two stand in certain determination relations—entail supervenience claims. So, if we can show that certain of these supervenience claims are false, this will show that, for example, identity claims which entail those supervenience theses are also false. Second, these supervenience claims are relevant to currently popular naturalizing programs in the philosophy of mind. A widely held view is that we can naturalize phenomenal properties by giving a naturalistically acceptable account of the facts about mental representation on which the phenomenal properties supervene; but to carry this out we need to know exactly which facts about mental representation (if any) form the minimal supervenience base for phenomenal properties, so we know exactly which facts about mental representation we need to give an account of. For example: do we have to give an account of what it is to bear some propositional attitude or other to a certain sort of content? Or also what it is to bear a certain attitude toward that content (that is, for the mental

2

We can also distinguish between intentionalist theses which are limited to the supervenience of the phenomenal on the representational and those which accompany this with the claim that the reverse direction of supervenience holds as well. This will become important later in Chapter 28 when we discuss the question of whether phenomenal properties can be identified with representational properties of any sort, but I’ll set it to the side for now.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

varieties of intentionalism 25 state to be of a certain type)? And if the latter, just which distinctions between attitudes must we be able to explain? Third, if it turns out that if there is some necessary connection between the phenomenal and the representational, this is something which we should want our view of the nature of perceptual experience to explain; we should want to know why, in the case of perceptual experience (and perhaps other mental states) this necessary connection obtains. But if we want to give an explanation of this necessary connection, it would be good first to know exactly what the necessary connection between the representational and the phenomenal is. My aim in the next few chapters will be to begin with an argument for the weakest intentionalist thesis: a time-restricted intrapersonal intramodal local intentionalism—and to argue from there for the truth of the corresponding interpersonal intentionalist thesis. We’ll then be in a position to ask whether a stronger intentionalist thesis—whether global, or intermodal, or both—is true.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

4 Transparency Many in recent years have thought that we can construct a good argument for one or another intentionalist thesis on the basis of the transparency of experience. I think so too. But, as many have noted, it’s less than obvious how the argument is supposed to go, because it’s less than obvious just what the thesis that experience is transparent is supposed to be. As is well known, recent discussion of this thesis is traceable to G. E. Moore’s report of his experience of a blue surface: . . . the moment we try to fix our attention upon consciousness and to see what, distinctly, it is, it seems to vanish: it seems as if we had before us a mere emptiness. When we try to introspect the sensation of blue, all we can see is the blue: the other element is as if it were diaphanous.1

Moore’s idea here is often summarized by saying that experience is transparent, because he seems to be saying that when we try to examine the features of an experience, we end up ‘looking through’ the experience and examining features of what the experience is an experience of.2 But how might we argue from this sort of claim to the truth of one or another intentionalist thesis? In this chapter I want to consider four ways in which we might try to construct an argument of this sort. The first three, which will be familiar from recent discussions of transparency, will all, though for different reasons and to different degrees, be found wanting. The last, which I’ll introduce at the end of this chapter and then develop over the next several chapters, is in my view successful. A natural way to formulate Moore’s transparency thesis is as a thesis about the scope of introspection. The idea is that when we introspect while having visual experience, what we find is always either some thing which is represented as in

1

Moore (1903), 450. The use of ‘transparency’ in this context is also due to Moore: ‘that which makes the sensation of blue a mental fact seems to escape us; it seems, if I may use a metaphor, to be transparent—we look through it and see nothing but the blue . . .’ (Moore (1903), 446). 2

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

transparency 27 one’s environment, or a quality which is represented as belonging to some aspect of the external environment. This thesis might be stated as follows: Positive Transparency When one introspects and one is instantiating a visual phenomenal property, nothing is available to introspection other than the objects represented as in one’s environment, and the properties they are represented as having. If this is right, our question is then how to use Positive Transparency in an argument for one or another intentionalist thesis. One very direct way to use Positive Transparency, which has recently been defended by Michael Tye, begins with the idea that ‘intuitively, we are aware of phenomenal character when we introspect.’3 But if we combine this with the thesis of Positive Transparency, which tells us that we are only aware in introspection of the properties we represent objects in our environment as having, it follows immediately that phenomenal characters must be properties we represent objects in our environment as having. Tye concludes that ‘The . . . phenomenal character of a perceptual experience consists in, and is no more than, the complex of qualities the experience represents. Thus, the phenomenal character of the experience of red just is red.’4 In Chapter 1 I suggested that taking ‘phenomenal character,’ rather than the phenomenal properties of subjects, as our starting point, could lead to confusion and verbal disputes. This seems to me to be an example of that. What, in the above passage, could Tye mean by ‘phenomenal character’? Earlier I introduced one meaning for ‘phenomenal character’: phenomenal characters are the properties of experiences which determine what it is like to be a subject of an experience with that phenomenal character. But this cannot be what Tye means by ‘phenomenal character’: on that usage, phenomenal characters are properties of experiences. And experiences aren’t the sorts of things that could be red. Could Tye instead be using ‘phenomenal character’ to pick out what we have been calling phenomenal properties—those properties of subjects that describe what it is like to be that subject? Hardly. While subjects of experience can, of course, be red, a subject’s phenomenal properties are not fixed by her color. Being red does not entail, and is certainly not identical to, being RED. So it seems that, when Tye says that the phenomenal character of the experience of red just is red, he is making a claim, neither about the phenomenal properties of subjects, nor about the properties of any experiences which determine the 3

Tye (2009), 119.

4

Tye (2009), 119.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

28 transparency phenomenal properties of the subjects of those experiences. So his thesis that ‘the phenomenal character of the experience of red just is red’ doesn’t itself tell us anything about phenomenal consciousness.5 To avoid a merely terminological dispute, let’s use ‘schmenomenal character’ as I suggested earlier we should use ‘phenomenal character’—as a name for the properties of experiences which determine what it is like to be a subject of that experience.6 And, as above, let’s use ‘phenomenal properties’ as a name for the corresponding properties of subjects—those properties of subjects which describe what it is like to be that subject. Then we can ask how Tye’s phenomenal characters are related to schmenomenal characters of experiences and to phenomenal properties of subjects. Now, though he does not use the term, Tye does have a view of schmenomenal characters. He holds that what it is for an experience to have a certain schmenomenal character is for that experience to represent a property like redness. He also has a view about phenomenal properties: what it is like to have an experience of red is for the subject of the experience to have the property of undergoing an experience which represents a property like redness.7 So, on Tye’s view, phenomenal properties are a matter of having an experience which represents phenomenal characters. These are intentionalist theses—but none of these claims follow from the conjunction of Positive Transparency with Tye’s thesis that phenomenal character is introspectable. Hence Tye’s discussion of transparency doesn’t succeed in establishing any interesting theses about the relationship between phenomenal properties of subjects and representational properties. Indeed, matters are worse than this, since Tye’s view of phenomenal properties undercuts the motivation for the key premise that ‘we are aware of phenomenal character when we introspect.’ The motivation for that premise is the idea that, in introspection, we are aware of what it’s like to have the experience. But, as discussed in Chapter 1, what it’s like to have an experience is a matter of the phenomenal properties of the subject of the experience—since the things that there’s something it is like to be are not experiences, but subjects of experience. Hence Tye’s argument relies on the view that, in introspection, we are aware of 5

Nor does it tell us anything about any representational properties—redness, after all, is not a representational property, and so, by Leibniz’s Law, if the phenomenal character of the experience of red is identical to it, that phenomenal character must not be a representational property either. Hence, despite the fact that Tye calls this thesis ‘strong property representationalism,’ it is hard to see it as any sort of intentionalist thesis at all, whether strong or weak, given that it says nothing about either phenomenal consciousness or any representational properties. 6 For a parallel move, see the discussion of ‘phenomenal character’ vs. ‘presentational character’ in Ch. 1 of Fish (2009). 7 See Tye (forthcoming).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

transparency 29 the phenomenal properties of subjects. But if we conjoin that view with Positive Transparency, we get the result that, in experience, we represent phenomenal properties as instantiated. And, if we combine that with Tye’s view of phenomenal properties, we get the result that, in experience, we represent the property of undergoing experiences which represent properties like redness as instantiated. But this is absurd.8 There is, however, a more promising way to use Positive Transparency to argue for intentionalism. That way begins with the observation that Positive Transparency seems to entail the following principle about introspectable differences between experiences: Transparency/Difference If there is an introspectable difference between two visual phenomenal properties of subjects, then there is a difference in the objects and properties those two subjects represent as in their environment. After all, if the only objects of introspection are the objects and properties presented as in the environment of the perceiver, how could there be an introspectable difference which was not a difference in one of the presented objects and properties? In addition to deriving the Transparency/Difference principle from Positive Transparency, we can also argue for it directly. Suppose that introspection reveals a difference in your experiences of a flat surface over time. Suppose (for simplicity) that the only introspectable aspect of your experience of the surface is its color. Can there be an introspectable difference between two experiences of the wall which is not a difference in the color that the wall seems to have? It seems not. And the line of thought here, one might think, generalizes; any change in your visual phenomenal properties will involve a change in the way your experience presents something as being, which just is a change in the properties you are sensing some object as having. The Transparency/Difference principle says that there is a necessary connection between introspectable difference and a difference in the objects and properties represented. Intentionalist theses assert a necessary connection between differences in phenomenal properties and differences in representational properties. To get from Transparency/Difference to intentionalism, then, we need there to be (i) a necessary connection between differences in phenomenal properties and introspectable differences and (ii) a necessary connection

8

For more discussion of these points, see Speaks (forthcoming-a and -b).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

30 transparency between differences in objects and properties represented and differences in representational properties. But, one might think, it is quite plausible that there are necessary connections of the sort described in (i) and (ii): In defense of (i): if the phenomenal properties of a subject are ‘what it’s like’ to be that subject, then it is plausible that any difference in phenomenal properties must be introspectable (even if not actually introspected); any difference between, for example, a subject at two times which was in principle not noticeable by the subject would not be a difference in what it is like for the subject at those times. In defense of (ii): if a subject represents certain objects and properties as in her environment, she stands in a representational relation to a proposition which represents those objects and properties as in her environment. If propositions p, q, represent distinct objects and properties as in the environment, then p6¼q. So, if two subjects differ with respect to the objects and properties they represent as in their environment, they stand in representational relations to distinct propositions. But then they differ in their representational properties. This gives us the following argument for intentionalism, which we might call the argument from introspectable difference:9 1 Necessarily, if two subjects differ in their visual phenomenal properties, there is an introspectable difference between them. 2 Necessarily, if there is an introspectable difference two visual phenomenal properties of subjects, then there is a difference in the objects and properties those subjects represent as in their environment. (Necessitation of Transparency/Difference) 3 Necessarily, if there is a difference in the objects and properties two subjects represent as in their environment, there is a difference in their representational properties. C Necessarily, if two experiences differ in their visual phenomenal properties, they differ in their representational properties. It is worth emphasizing the gap between the argument from introspectable difference and a related but distinct argument, which might be called the argument from focusing, which is suggested by many discussions of transparency in the literature. Moore’s thesis, according to the argument from focusing, is a thesis 9

For more on this argument, see Speaks (2009). For closely related arguments for intentionalism, see }3 of Byrne (2001), Tye (2002), and Kriegel (2009), 70.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

transparency 31 about what happens when we try to focus on the phenomenal properties corresponding to one portion of the visual field. It says that, when we do this, what we find is the property represented by this portion of the visual field. Given this, it seems plausible that any difference in the phenomenal properties corresponding to some bit of the visual field must entail a difference in the properties that bit of the visual field represents. But, further, it seems plausible that the phenomenal properties of the subject of an experience are fixed by the phenomenal properties associated with parts of that subject’s visual field (along with the way in which they are combined). And it seems plausible that the representational properties a subject instantiates in virtue of undergoing an experience are fixed by the representational properties of parts of the subject’s visual field (along with the way in which they are combined). But put these theses together, and one gets an intramodal intentionalist thesis: any two visual experiences whose subjects instantiate different phenomenal properties will also be experiences whose subjects instantiate different representational properties. There is much that could be questioned about this line of argument. (What is a visual field? Could it really have parts? And if phenomenal properties are properties of subjects, what does it mean for one to be ‘associated with’ part of a visual field?) But, without engaging these questions, we can show that something must be wrong with the argument. We can show this by, first, bringing out a way in which the supervenience thesis established by the argument from focusing is stronger than any of the supervenience theses discussed in the last chapter; and, second, by showing that this stronger supervenience thesis is false. The relevant distinction here is analogous to the distinction between global and strong supervenience theses. Ignoring some further distinctions which are not for our purposes relevant, the A-properties strongly supervene on the B-properties iff, if any two individuals in any worlds are alike with respect to the B-properties, then they must also be alike with respect to the A-properties. Global supervenience, by contrast, requires only that any two worlds which are the same with respect to the overall distribution of the B-properties are also the same with respect to their distribution of the A-properties. Now think of ‘portions of the visual field’ as analogous to individuals, and the subjects of the relevant experiences as analogous to worlds. We can then ask whether the intentionalist is claiming that any two portions of the visual field which represent the same property as instantiated must also be associated with the same phenomenal property—strong supervenience—or whether the claim is merely that any two subjects alike with respect to all of the objects and properties they sense as in their environment must also be alike with respect to their phenomenal properties—global supervenience.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

32 transparency The intentionalist theses considered in the last chapter were all analogous to global supervenience claims. But if the argument from focusing were convincing, it would establish strong supervenience. For that argument had us attending to one portion of our visual field, and, on the basis of that exercise, arriving at the supervenience of the phenomenal property associated with that portion of the visual field on the property represented by that portion of the visual field. But this strong supervenience claim is not true. Consider the perceptual constancies—like the phenomenon of color constancy, in which the color of an object is perceptually represented as constant through changes in illumination. If the color is perceptually represented as constant through such a series of experiences, then there must be some property—say, blue—represented by all of these experiences. And if this is true, then our strong supervenience thesis tells us that there must be some phenomenal property associated with a portion of the visual field which is common to all of these experiences. But, given color constancy, there just is no phenomenal property associated with a portion of the visual field common to all and only blue-representing experiences. The point of color constancy is that we get constant representation of blueness despite changes in the phenomenal properties associated with portions of the visual field. This is just where the ‘mosaic’ analogy for visual experience, and with it the argument from focusing, breaks down.10 It is a strength of the argument from introspectable difference, by contrast, that it does not entail the stronger supervenience thesis which seems to be falsified by color constancy. Indeed, given the above remarks about the plausibility of premises (1) and (3), it seems that Transparency/Difference provides the basis for a powerful argument in favor of intentionalism. But which variety of intentionalism is it an argument for? This is equivalent to the question: over which pairs of phenomenal properties and subjects do the premises of this argument quantify? The argument as stated focuses only on visual phenomenal properties. And if we restrict the phenomenal properties to consecutive phenomenal properties of a single subject in a single sense modality which differ in certain obvious ways, then, I think, each of the premises is extremely difficult to reject. But given this restriction, the conclusion of the argument is a claim which is weaker than even our weakest intentionalist thesis on the chart in Chapter 3 (a local, intramodal, 10

Here I am following part of the argument of Thompson (2006)—though we draw different morals from it. A parallel argument could be run involving color contrast, or common sensibles, rather than perceptual constancies. I discuss the latter case in Chapter 25.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

transparency 33 intrapersonal time-restricted intentionalism). A natural question is then whether, by relaxing this restriction, we might also be able to use this argument to defend a stronger intentionalist thesis. For instance, we might relax our restriction on the pairs of phenomenal properties over which the premises quantify, to allow not just consecutive phenomenal properties of a single subject, but (say) an arbitrary pair of visual phenomenal properties, whether or not they are properties of the same subject. The argument, thus modified, would then be an argument for an interpersonal intentionalism (albeit still a local, intramodal one). Wouldn’t the premises in this argument, thus modified, still be true? A reasonable case can be made that they would be; but things are not nearly so clear as with the version of the argument restricted to consecutive experiences of a single subject. One might now have a reservation about the first premise. Perhaps if two consecutive experiences of a single subject differ in phenomenal character, there is a clear sense in which there must be some introspectable difference between them. But if the two experiences are separated by a period of years, for example, are we so sure that a difference in phenomenal character must be introspectable? And if we are comparing the phenomenal properties of two different subjects, in which case there is no subject to do the introspecting, what does it even mean to say that a difference in phenomenal character between two experiences must be introspectable? And one might have a related doubt about the third premise: suppose that the contents of color experiences, for example, are not color properties but modes of presentation thereof, and suppose further that these modes of presentation do not rigidly designate the relevant color properties. Then two experiences separated by a sufficient interval, or had by different subjects, may differ in the circumstance of evaluation relevant to the determination of the color property designated by a single mode of presentation—which in turn could make possible a difference in color property represented without a difference in mode of presentation, and so without a difference in sensed content. These weaknesses in the argument from introspectable difference all stem from the universality of the premises. And in fact the universality of the Transparency/Difference principle itself might leave it open to objection. Instances of the principle involving certain paradigmatic introspectable differences—like the introspectable difference between ordinary visual experiences of red and green surfaces—do seem obviously true. But are we so sure that the principle holds for any introspectable difference? Consider, for example, pairs of visual experiences which differ only by a small change in the focus of the perceiver’s attention. There’s clearly an introspectable difference here—but it’s not so clear that there

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

34 transparency must be any difference in the objects and properties represented as in the perceiver’s environment.11 Though there’s much to be said for the argument from introspectable difference, we’ve seen that doubts can be raised about each of its premises—and it’s not easy to see how to argue directly for those premises. Hence my strategy will be to set Transparency/Difference and the argument from introspectable difference to the side, and begin with a much more limited assumption. The formulation of Moore’s claim which we’ll put to use doesn’t assume the impossibility of any introspectable difference between any phenomenal properties of any subjects without a difference in objects and properties represented, but rather assumes the impossibility of certain very dramatic phenomenal differences in consecutive experiences of a single subject without any corresponding difference in the perceptual representation of a certain class of properties. This is, I think, a kind of minimal interpretation of Moore’s claim. If, as Moore says, ‘[w]hen we try to introspect the sensation of blue, all we can see is the blue,’ one thing that this suggests is that certain combinations of phenomenal properties (the ‘sensations’ we introspect) and properties represented are impossible. So, for example, if when we introspect the sensation of blue, all we can see is the blue, it is hard to see how a series of visual experiences of the same shade of blue could yield wildly different introspectable sensations. And, conversely, it is hard to see how a constant introspectable sensation could be at one moment of blue, and at the next moment of red—this would indicate, contra Moore, that when we introspect the sensation of blue, we don’t ‘see’ the blue, but rather see something else. This is a negative rather than a positive approach to the transparency of experience. Rather than trying to establish a general necessary truth about experience which can be used directly in the derivation of one or more intentionalist thesis, we begin with certain more modest claims about the impossibility of certain sorts of experiences.

11

I return to cases of this sort, and shifts in attention more generally, in Chapter 27.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

5 Two Impossible Scenarios Here are two descriptions each of which, I claim, fails to describe a possible situation:1 Scenario A

Scenario B

Psychedelic phenomenology þ constant representation of color properties

Constant phenomenology þ psychedelic representation of color properties

A subject is looking intently at a well-lit surface which occupies the whole of the subject’s visual field. Over the course of a few seconds, she goes from being (as we would put it, were we to describe what it is like to be her) BRIGHT RED to being BRIGHT GREEN to being BRIGHT BLUE, and constantly repeats this pattern. And the subject’s memory is working normally—it’s working pretty much the way yours usually does when you have an experience lasting a few seconds. But, the whole time, she is visually sensing the wall to be a certain shade of red; it visually seems to her throughout that the wall is that shade of red; according to her experience, the wall is that shade of red throughout.

A subject is looking intently at a well-lit surface which occupies the whole of the subject’s visual field. The only thing notable about the series of phenomenal properties he instantiates is its monotony. For the duration of the experience he instantiates the phenomenal property CHARCOAL GREY. And the subject’s memory is working normally—it’s working pretty much the way yours usually does when you have an experience lasting a few seconds. Nonetheless, the subject is visually sensing the color of the wall to be rapidly changing from bright red, to bright green, to bright blue; it visually seems to him that the wall is changing from bright red, to bright green, to bright blue; according to his experience, the wall is changing from bright red, to bright green, to bright blue.

These resemble various examples of perceptual experiences which one can find in the recent philosophy of perception literature. See especially the discussion of dancing qualia in }7.5 of Chalmers (1996) and some of the examples in Levine (2001), Ch. 4. 1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

36 two impossible scenarios The claims that Scenarios A and B are impossible might seem too obvious to argue for. But, if this seems less than obvious, the best way to argue for it is to imagine being the subject of the experiences described above. Beginning with Scenario A, we can ask: can the subject notice the change in phenomenal properties, or not? It seems that this change in phenomenal properties must be noticeable; it is hard not to agree with Ned Block that ‘there is something in the nature of phenomenal character that precludes big fast unnoticeable changes.’2 So suppose that you are a subject who notices this rapidly and dramatically changing series of phenomenal properties, and suppose that you consider the question of how you are visually sensing the color of the surface to be. Would it be coherent for you to suppose that you are visually sensing the color of the surface as constant? Another way to put this question—recalling our introduction of sensing as a truth-sensitive relation—is: Would it be coherent for you to suppose that you are misrepresenting the world if it turns out that the surface before you is changing in color? It seems to me that the answer to these questions is, clearly, ‘No.’ Now, one might object that it is possible that psychedelic phenomenology could be combined with a judgement or belief that the color properties of the surface are constant; you could know, for example, that you just took a drug likely to produce psychedelic phenomenology. This is true but irrelevant, for the reasons given in Chapter 2: sensing is a representational relation which is distinct from the representational relations of judging and believing (and from being disposed or inclined to judge or believe). So the fact that you may well judge or believe the color of the wall to be constant does not change the fact that (when instantiating the phenomenal properties described in Scenario A) you would be visually sensing the color of the wall as changing. A similar thought experiment brings out the plausibility of the claim that Scenario B is impossible. Imagine yourself stuck instantiating CHARCOAL GREY, and noticing the constancy of your visual phenomenal properties. Would it be coherent for you to suppose that you are visually sensing the color of the surface to be rapidly changing—that, if the surface turns out to have been constant in color throughout, you will have been misrepresenting the surface before you? Again, it seems not. So it seems to me that the positive case in favor of these impossibility claims is quite strong. Are there any objections to these claims?

2

Block (1998), 669.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

two impossible scenarios 37 A first line of objection is based on judgements about certain other hypothetical scenarios. Here’s an example: Bob has contracted a rare disorder which systematically shifts his visual phenomenal properties when he is looking at things which are the color of ripe bananas. Whenever Bob looks at something which is this color, his visual phenomenal properties rapidly shift from BRIGHT RED to BRIGHT GREEN to BRIGHT BLUE, and constantly repeat this pattern. Despite this disorder, Bob lives a long and happy life, and gets used to the surprising appearance bananas (and other like objects) present to him. After a while, whenever Bob has an experience with this sort of psychedelic phenomenology, Bob immediately takes the object presented to be yellow—in something like the ‘second nature’ sort of way in which you take something which causes you to instantiate RED to be red. Surely, at this point, when Bob comes across a banana, and undergoes the psychedelic phenomenology just described, Bob now, after all these years, is visually sensing that the object is yellow. So goes the objection.3 I have to say that I feel no temptation at all to describe the case in this way. It seems to me quite obvious that in this sort of case, Bob is visually sensing the banana as rapidly changing in color, and not as constant in color—it’s just that Bob has adopted and internalized sensible policies of belief formation which adjust for the unusual tendencies of his visual system. You might be tempted to challenge this, by pointing out that these rapidly changing phenomenal properties are, after all, ‘the way in which Bob’s visual system responds to yellow things’—isn’t this enough for this rapidly changing phenomenal character to, for Bob, represent yellow? I don’t think so. Compare this to a case of color blindness. Certain sorts of color blind subjects, when looking at red objects, instantiate phenomenal properties characteristic of normally sighted subjects’ encounters with yellow-brown surfaces. These visual phenomenal properties are ‘the way in which their visual system responds to redness.’ But, for all this, we don’t say that these experiences represent the relevant object as red—we say, by contrast, that defects in their visual system prevent them from visually representing red objects as red. And the fact that a given color blind subject might be very good at correctly forming beliefs about the colors of red things wouldn’t change this. One way to press the point is to imagine a color blind subject who, at t1, ordinarily responds to both green and grey surfaces with GREY phenomenal properties. We might imagine that there is an innovative surgery designed to 3

For discussion of cases like this, see among other places, Biggs (2009).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

38 two impossible scenarios remedy the color blindness, but that this surgery proceeds in two steps. In the first step, at t2, the surgery makes the subject such that she responds to both green and grey surfaces by instantiating a GREEN phenomenal property. Then, at t3, in the second step, the surgery makes the subject such that (as with a non-color-blind subject) she responds to green surfaces with experiences having a GREEN phenomenal character, and to grey surfaces with experiences having a GREY phenomenal character. We can further imagine that the surgery is carried out while the subject is conscious, and looking at a surface whose left half is green and whose right half is grey. On the view in question, the subject’s GREY experiences at t1 will presumably represent surfaces as having one of the colors grey or green, but not as definitively having either—this, after all, is ‘how the subject’s visual system responds’ to green and to grey surfaces at t1. The subject’s GREEN experiences at t2 will, similarly, represent the surface as uniformly green-or-grey. At t3, the subject will instantiate a phenomenal property which is ‘half GREEN and half GREY’ and will represent the left half of the surface as green and the right half as grey. The subject will (presuming that her memory is functioning normally) report her experience by saying that her visual representation of the color of the surface changed dramatically from t1 to t2, and that, between t2 and t3, her visual representation of the left half of the surface was constant, whereas her visual representation of the right half changed back to the way it was at t1. But the proponent of the view that at t1 the subject’s instantiating GREY involves visually representing surfaces as grey-or-green must take the subject in question to be mistaken about each of these judgements about how she is visually representing the colors of the surface. She must, for instance, say that, despite the way it seems to her, the subject’s representation of the left half of the surface changed just as much between t2 and t3 as of the right half—the first from green-or-grey to green, and the second from green-or-grey to grey. This is very implausible. It’s also worth emphasizing that the example of the banana is, even if convincing, only a counterexample to the impossibility of Scenario A. It is very difficult to adapt the banana example to provide a convincing counterexample to the impossibility of Scenario B. And, for the purposes of the argument which follows, it will be enough if you’re convinced that one of Scenarios A and B is impossible. A different line of objection to the impossibility of scenarios A and B is that our intuition to this effect turns on a conflation between differences in representation and representation as different.4 Let’s focus again on Scenario A. One might defend its possibility by saying that while psychedelic phenomenology of the sort 4

Thanks to Timothy Williamson for suggesting this as a possible response to the argument.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

two impossible scenarios 39 described does entail that the subject represents the color of the surface as changing—that she represents it as different from one moment to the next— this does not entail that the color she represents it as being at one moment is distinct from the color that she represents it as being at the next moment. One can represent things as being different which one does not represent differently. The same line of thought could be extended to Scenario B, substituting sameness for difference throughout. I think that the problem with this line of thought is that it makes the experiences described in our two scenarios possible, but at the cost of making them possible representations with a (collectively) impossible content. In Scenario A, for example, it would entail that the subject of the experience at one moment (call it t1) visually represented the wall as red, at t2 visually represented it again as red, but through t1 and t2 visually represented it as varying in color. No wall could make this series of visual representations veridical. Analogously, in Scenario B the proposed strategy would entail that the subject of the experience at t1 visually represented the wall as red, at t2 visually represented it as green, but through t1 and t2 visually represented it as constant in color—again, we have a series of experiences whose contents are jointly impossible to satisfy. But it’s very hard to believe that experiences with the phenomenal characters described above could really have impossible propositions as their contents. Reflect again on Scenario B, and imagine yourself instantiating that monotonous, constant, phenomenal property. Could it really be the case that, in that experience—a matter of just about the most boring visual phenomenology possible—you were visually sensing the wall to be a way which no wall could possibly be? This seems to me very hard to believe. A last line of objection to the impossibility of Scenarios A and B is based on the phenomenon of color constancy, in which certain phenomenal properties associated with our visual experience of a surface change (with, for example, changes in illumination) without any change in our visual representation of the color of the relevant surface. If cases of this sort are possible, one might wonder, why not cases of psychedelic color phenomenology along with constant color representation? Again, it is worth noting that cases of color constancy are only even a prima facie challenge to the impossibility of Scenario A; they say nothing to the claim that Scenario B is impossible. But in the end it is doubtful whether color constancy is a serious challenge to the impossibility of Scenario A, either. Let’s suppose that, despite our stipulation that there are no changes in illumination, it is possible for a scenario like that described under the heading of psychedelic phenomenology þ constant content to be one in which the wall is

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

40 two impossible scenarios visually represented as having a constant color while changing in relative illumination. Still, on this interpretation, it is surely true that something is being represented as changing color; whether it is the wall, or the light under which it is being viewed, is irrelevant to what follows.5 And in fact we can make clear the irrelevance of illumination to the above cases simply by imagining that the subject is experiencing the above sort of psychedelic phenomenology while looking directly at a light source rather than at an opaque surface. Of course, to grant that Scenarios A and B are impossible is not to grant the truth of even the weakest intentionalist thesis on our chart in Chapter 3. To grant the impossibility of these scenarios is just to grant that certain dramatic changes in phenomenal properties in certain idealized situations is inconsistent with sameness of sensing properties. To get from there to an intentionalist thesis one has to argue—following the strategy pioneered by Alex Byrne in ‘Intentionalism Defended’—that the inconsistency remains even if we consider less dramatic phenomenal shifts and strip away the relevant idealizations. That, roughly, will be the task of the next few chapters. So in what follows I will assume that Scenarios A and B are impossible, and turn to what I take to be the more interesting question: are there any interesting theories of perception which imply the reverse?

5

This fits nicely with the view of color constancy defended in Hilbert (2005).

OUP CORRECTED PROOF – FINAL, 18/5/2015, SPi

6 Phenomenism and Phenomenal Variance I think that there are. Consider the negation of an interpersonal intentionalist thesis about experiences of color: Phenomenism: possibly, two subjects instantiate different color phenomenal properties, but visually sense the same contents. If we can show that phenomenists are committed to the possibility of our scenarios, this will amount to an argument that at least one local interpersonal intentionalist thesis is true. As it turns out, phenomenists also endorse a stronger claim: Super-phenomenism: possibly, two subjects instantiate arbitrarily different color phenomenal properties, but visually sense the same contents. Why should phenomenists be super-phenomenists? In part, because most phenomenists are motivated by the need to accommodate the possibility of spectrum inversion without misrepresentation. These are cases in which two subjects differ systematically in their visual phenomenal properties despite the fact that, intuitively, neither is misrepresenting the scene before them. According to one standard way of setting up the example, two people might differ—from birth, and systematically—in which things cause them to instantiate the phenomenal property BLUE and which cause them to instantiate YELLOW. Suppose that the first of the two is just like us, and that the second, for example, instantiates YELLOW when looking at the sky on a clear day. Surely, the defender of this line of thought holds, this is not a case of misrepresentation—after all, our inverted friend senses what we call ‘blue’ by instantiating YELLOW just as reliably as we sense that color by instantiating BLUE. But if this is the motivation for the view, then what’s needed is super-phenomenism. After all, in typical inversion scenarios, like the one just described, we get not just a little bit of variance in color phenomenal properties with no difference

OUP CORRECTED PROOF – FINAL, 18/5/2015, SPi

42 phenomenism and phenomenal variance in representational properties, but arbitrarily wide variance in color phenomenal properties with constant representational properties. And that’s what superphenomenism says to be possible. Perhaps some phenomenists are motivated less by cases of full spectrum inversion than by examples of ‘spectrum shift.’1 In these cases, phenomenal matching tests (and tests for identification of unique hues) seem to reveal relatively small—relative, that is, to the phenomenal differences exhibited by imagined spectrum inverted subjects—but systematic phenomenal differences between people of different ages, races, and sexes. But there seems to be no reason to think that people of different ages, races, and sexes can’t all be veridically sensing the colors of things. Hence (one might conclude) relatively small but systematic differences in color phenomenal properties must be compatible with sameness of representation of color properties. But if one finds spectrum shift cases convincing, one should also find more dramatic inversion cases convincing—for it seems that we can get to an inversionlike case by imagining a sufficiently long series of spectrum shift cases. And, in general, it’s hard to hold that small differences in phenomenal character are consistent with sameness of content but that large differences aren’t, since we can always get to the large differences by an accumulation of small differences. So it’s hard to be a phenomenist without being a super-phenomenist. But why should we think that the super-phenomenist is committed to the possibility of Scenario A or Scenario B? To see why, consider the following claim: Phenomenal variance: possibly, two subjects instantiate arbitrarily different color phenomenal properties, but visually sense the same color properties as instantiated. Super-phenomenists seem to be committed to phenomenal variance, because it’s hard to see why sameness of the content of an experience of color should entail a difference in the color property represented; presumably if we hold fixed the circumstance of evaluation, sameness of content should entail sameness of color property represented.2 So to show that phenomenism entails the possibility of Scenario A, it suffices to show that phenomenal variance entails the possibility of Scenario A. Scenario

1

See for discussion Block (1999). And phenomenal variance is directly entailed by the possibility of inversion scenarios as described, in which, for example, red/green spectrum inverts agree in their perceptual representation of the colors of things. 2

OUP CORRECTED PROOF – FINAL, 18/5/2015, SPi

phenomenism and phenomenal variance 43 A just is an instance of phenomenal variance in which the pair of phenomenal properties are instantiated consecutively by a single subject whose memory is working normally. So, phenomenal variance entails the possibility of Scenario A unless one of the following three principles is true:

The interpersonal constraint

Two subjects can instantiate arbitrarily different color phenomenal properties, but visually sense the same color properties as instantiated; but this is not possible for a single subject at different times.

The time constraint

A single subject can, at different times, instantiate arbitrarily different color phenomenal properties, but visually sense the same color properties as instantiated; but this is not possible for a single subject at times separated by less than some minimal interval t.

The memory constraint

A single subject can instantiate arbitrarily different color phenomenal properties, but visually sense the same color properties as instantiated, so long as those two experiences are not related by a certain memory relation M; but this is not possible for two experiences of a single subject which are connected by M.

So, to show that phenomenism entails the possibility of Scenario A—and hence to show that a local interpersonal intentionalist thesis about visual experience is true—it suffices to argue against the interpersonal constraint, the time constraint, and the memory constraint.3

3

I’m simplifying by ignoring the possibility that one might instead defend a combination of the time and memory constraints, which says that we can get variance in color phenomenology without variance in color property represented only if we have both a sufficient time interval and absence of the relevant memory relation M. The truth of some claim of this sort would be sufficient to block the derivation of the possibility of Scenario A. But the arguments I go on to offer against the time constraint and the memory constraint also work (if they work at all) against this sort of combination of the two.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

7 The Interpersonal Constraint, the Time Constraint, and the Memory Constraint I’ll consider these in turn.

The Interpersonal Constraint The interpersonal constraint is, I think, the least attractive of these three. We can argue against the interpersonal constraint as follows: Argument against the interpersonal constraint Consider two subjects, A and B, differing in their color phenomenal properties but not in the color properties they sense as instantiated. If we consider sufficiently long-lived and protean individuals, it will always be possible to imagine a single subject who is at one time relevantly just like A and at another time relevantly just like B.1 Calling this an argument may be a bit of a stretch. Really this is the flip side of the familiar phenomenist idea that if cases of interpersonal spectrum inversion without misrepresentation are possible, intrapersonal cases are too.2 The oddness of the interpersonal constraint is its combination of two claims which pull in different directions: first, the claim that there is no necessary, internal connection between phenomenal properties and represented color properties, so that different phenomenal properties may be associated in different individuals with the same represented color property; and, second, the claim that it is a necessary truth that, within the life of a single organism, the connections between phenomenal properties and represented color properties are immutable. This is one way of putting the intuition behind the ‘principle of recombination’ discussed in Byrne (2001), 216–17. 2 See, among many other places, Block (1990). 1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the interpersonal, time, and memory constraints 45 One way to bring out the oddness here is to recall the cases of spectrum shift discussed earlier. We’re just as disinclined, intuitively, to assert differences in the veridicality of color experiences between the experiences of individuals at different ages (intrapersonal shift) as between the experiences of people of different sexes and races (interpersonal shifts). But the proponent of the interpersonal constraint is forced to treat these cases differently. A natural thought is that one might defend the interpersonal constraint by defending a teleological theory of content, according to which the color property a certain phenomenal property represents for a creature is determined by the function of that state type—which, in turn, is fixed by the evolutionary history of the species of which the creature is a member. (Different teleological theories will give different views about exactly how the evolutionary history of a state type in a species fixes its content.) Since the evolutionary history of an organism can hardly change over the course of its life, it might seem that a teleological theory of content could explain why the color property represented by a given phenomenal state is unchangeable over the life of a creature—even though it may vary between distinct creatures, which might have different evolutionary histories. However, this sort of teleological theory of content does not in the end lend support to the interpersonal constraint, because it is possible that the function (in the relevant, biological sense) of a given state can change over the course of a creature’s life. And the proponent of a teleological theory of this sort had better hold that this is possible, since otherwise it is hard to see how teleosemantics could model the effects of learning on the contents of representational states.3 And, if this sort of change in function is possible, it is possible for the content of a state type—which is fixed by its biological function—to change as well. If this is possible for phenomenal state types—and there seems no reason why these would, of necessity, be an exception—then, if this sort of teleological theory of content is true, the interpersonal constraint is false.4

3 As proponents of teleological theories of content agree; see, for example, the discussion of nongenetic selection in Macdonald and Papineau (2006). 4 No reason, that is, short of there being some sort of identity between properties like having the phenomenal property RED and representational properties like the property of sensing the color red as instantiated. But such an identity claim would entail the falsity of phenomenal variance, and hence also the falsity of the interpersonal constraint. Dretske (1995), 15 does draw a distinction here between perception and belief; he says that the content of the former is ‘determined phylogenetically’ whereas the content of belief ‘is ontogenetically determined.’ But it is pretty clear from the context that this is put forth as a contingent fact about human beings rather than as a necessary truth about perception. He says, for example, that the functions of our sensory systems are fixed, if not immediately, at an early age; and qualifies the claim that we can’t change the representational character of an experience by saying ‘not easily, anyway.’

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

46 the interpersonal, time, and memory constraints

The Time Constraint As I said, I don’t think that many phenomenists will be tempted by the interpersonal constraint. So instead let’s think about whether the phenomenist might block the derivation of the possibility of Scenario A by endorsing the time constraint. It seems to me that this is a difficult thesis to endorse, for the following reason:5 Argument against the time constraint Consider a single subject at two times, t1 and t2, which are separated by the minimal time interval t. Because they are separated by t, it is possible that the subjects differ arbitrarily in color phenomenal properties, but sense the same color properties as instantiated; to fix ideas let us suppose that at the time of t1 the subject is such that when she is instantiating RED she is visually sensing the property red (and analogously for GREEN and green) whereas at t2 the subject is such that when she is instantiating RED she is visually sensing the property green (and analogously for GREEN and red). But presumably it is possible for the subject to instantiate a phenomenal property during t—let us suppose that it is RED. What color property is the subject visually sensing the color of the relevant surface to have? Since, by hypothesis, t is the minimal interval of time by which two experiences alike in color content but distinct in color phenomenology must be separated, she cannot be representing the color red, since it is separated from t2 by an interval less than t; and because it is also separated from t1 by an interval less than t, she cannot be representing the property green. And the subject can’t be visually sensing some third color property. For suppose that she was visually sensing the surface to be yellow. Presumably she was able to visually sense surfaces to be yellow at t1, and did so by instantiating YELLOW. We can then suppose that, at t1, the subject had an experience as of a half-yellow and half-red surface, and again generate a violation of the stipulation that t is the minimal interval of time by which two experiences alike in color content but distinct in color phenomenology must be separated. One reply open to the defender of the time constraint is to stipulate that there can be no experiences in the interval t; a subject can change in color phenomenal 5

For a different argument against the time constraint based on the claim that we come to know facts about phenomenal character by attending to the world as it appears to us, see the discussion of consecutive vs. non-consecutive experiences in Byrne (2001), 213–14.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the interpersonal, time, and memory constraints 47 properties but be alike in color property visually sensed only if separated by a sufficiently long experienceless interval. This view seems to me a bit ad hoc, in the sense that the proponent of an ‘experienceless interval’ view is going to end up treating very similar cases very differently. Suppose that we have a pair of subjects alike but for the fact that one underwent an experienceless interval of just long enough, and the other’s experienceless interval fell just short of the required length. Could this difference really be sufficient for the two to, from that point on, differ dramatically with respect to whether their visual representation of things as instantiating the property of redness is done via the phenomenal property RED or the phenomenal property GREEN? One might reply to this sort of worry by saying that what matters is not simply the passing of a minimal interval t; what matters is that the subject’s visual system, and its relation to the subject’s environment, undergo some change. Maybe, for example, what matters is that the functional roles (on some precisification of that notion) of the subject’s internal perceptual state change; on this sort of view, we would require an experienceless interval, but the duration of that interval could be designated (presumably, non-rigidly) in functional terms. Obviously, for this kind of view to work, it must be that the functional change in question entails the existence of some experienceless interval. The problem is that none of the functional properties usually suggested as candidates for the analysis of mental properties seem to have this entailment—if states can, for example, change what they reliably indicate, then there’s no reason why they can’t do this while the subject is enjoying an experience. (For that matter, there’s no reason why they can’t change from one moment to the next, rather than over a sufficiently long interval.) In addition to the worry about the arbitrariness of ‘experienceless interval’ views, there is a worry about how to describe certain transitional cases. Whatever the minimal experienceless interval is, there will be borderline cases—cases of subjects whose experienceless intervals are very close to it. Our choices about what to say about these borderline cases are familiar from other examples of vagueness. We can either say that there is a range of subjects for whom it is indeterminate whether or not the relationship between phenomenal properties and sensed content has switched, or we can say that there is a sharp cut-off point. But neither seems very attractive. We can’t say that borderline cases are ones in which the content of the subject’s experience after the interval is ‘indeterminate,’ since it is very implausible that it can be indeterminate whether a visual experience of any phenomenal character

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

48 the interpersonal, time, and memory constraints represents an object as red or green. Imagine a visual experience of something which is half-red and half-green—would the experience represent the object as indeterminately-red-or-green all over? One might try to solve this last problem by going supervaluationist, and saying that the two acceptable ‘precisifications’ of the experience’s content are: (i) red on left/green on right and (ii) red on right/green on left. On this sort of view it would come out true on every precisification, and hence true simpliciter—as it should— that the subject represents the left half of the object as having a different color than the right half. But there are other oddities. One is that, if we assume that it is impossible for a surface to be simultaneously red and green, it will follow that it is impossible for the viewed object to be such as to make the subject’s experience veridical (since, on the supervaluationist view in question, the experience will be veridical iff each of the two precisifications are true). But it does not seem that the subject of such a pedestrian experience could really be visually sensing the surface to be a way that no surface could be. So we’re forced to say that there must be a sharp cut-off point here—and, whatever the attractions of epistemic views of vagueness in other cases, it seems unattractive here. For one thing, it looks like the sharp cut-off point here would be (as in other cases of vagueness, on the epistemic view) undiscoverable. But then one could be in a position—absurdly—of being unsure whether, when one instantiates RED, one is visually sensing the same color as one used to sense when one instantiated GREEN. A slightly different way to defend the time constraint, without appealing to experienceless intervals, focuses on the fact that the constraint concerns arbitrarily large changes in color phenomenology with sameness of color property represented. Perhaps, one might think, such a transition is possible, but must be mediated by a series of much smaller changes. According to this sort of view, we can get from a situation in which RED involves the visual representation of red to one in which GREEN does—but only by sequentially instantiating a series of phenomenal properties which seamlessly connect RED and GREEN. Let’s call this a ‘baby steps’ theory. Let’s suppose that RED28 and RED29 are two consecutive phenomenal properties in this series. A first question is whether RED28 and RED29 are discriminable by the subject of the experience. If they are, then the baby steps theory does not really solve the problem that it was supposed to solve. For in that case we can imagine a series of experiences in which a subject rapidly fluctuates between instantiating RED28 and RED29 while constantly visually representing the surface before him as constant in color. This is not, perhaps, as spectacularly impossible as psychedelic

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the interpersonal, time, and memory constraints 49 phenomenology þ constant representation of color properties—but it is, still, pretty clearly impossible.6 So we have to presume that RED28 and RED29, together with the other phenomenal properties in the series, make up a phenomenal continuum: a series of experiences which are such that each experience in the series is indiscriminable from the next, but the first is discriminable from the last.7 This sort of baby steps theory faces many of the same problems as experienceless interval theories. In particular, it is hard to see why, if the facts about what representational state a given subject is in when she instantiates RED can vary over time from representations of redness to representations of greenness, that this must proceed via a very long series of distinct but indiscriminable phenomenal properties. Let t1 be a time at which S is such that her instantiating RED involves her visually sensing redness, and t2 a time at which her instantiating GREEN does so. There must be some properties F, G of S which are such that her being F at t1 and being G at t2 explains this switch. Are there any plausible candidates for F and G which are such that it’s impossible for a subject to be F at one time and G at another without, in between, taking the tour through the very many indiscriminable phenomenal properties separating RED from GREEN which we’re imagining our baby steps theory to require?

The Memory Constraint Finally, let’s consider the memory constraint. (Remember, this was the view that there is a certain memory relation M which is such that (i) experiences separated by a short interval when one’s memory is working normally are M-related and (ii) M-related experiences can’t differ arbitrarily in color phenomenology while agreeing in color property sensed, though this is possible for non-M-related experiences.) This constraint seems to be open to an argument much like our initial argument against the time constraint: Argument against the memory constraint Consider two experiences of a single subject, e1 and e2, which are not related by the relevant memory relation M. Because they are not M-related, it is possible for the subject to differ arbitrarily in his color phenomenal properties, but visually sense the same color properties as instantiated; let’s again suppose that 6

Remember that we can rule out cases of color constancy via the points made in Chapter 5. Some philosophers deny that phenomenal continua are possible; if they’re right, then of course this sort of baby steps theory is a nonstarter. I return to the question of whether phenomenal continua are possible in Chapter 30. 7

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

50 the interpersonal, time, and memory constraints at the time of e1 the subject is such that when RED he visually senses the property red (and analogously for GREEN and green), whereas at the time of e2 the subject is such that when RED he visually senses the property green (and analogously for GREEN and red). But memory relations (as Reid pointed out in his criticism of Locke’s theory of personal identity) are not transitive;8 in general, it is possible to remember some earlier experience, at which time you remembered some experience which you now can’t remember. So presumably it is possible that between e1 and e2 you have a perceptual experience, e*, which is M-related to both e1 and e2. But then we can argue in a way parallel to the argument against the time constraint that any assignment of phenomenal properties and sensing properties to the subject at the time of e* will contradict the memory constraint. However, one might also turn to discussions of personal identity for a solution to this problem.9 Even if ordinary memory relations aren’t transitive, we can use these relations to define a transitive relation. Using our initial memory relation M, we might define a new relation, indirect-M, as follows: e1 and e2 are indirect-M-related iff there is some series of experiences such that (i) e1 is the first and e2 is the last, and (ii) every experience in the series other than e1 is M-related to an earlier experience in the series. This has a certain similarity to the ‘experienceless interval’ way of defending the time constraint. In both cases, we have the view that the relation between phenomenal character and content can change for a subject between t1 and t2 only when that subject has an intervening quiet time—that he has no experiences at all between t1 and t2 (as in the ‘experienceless interval’ version of the time constraint) or that he has no experience which is M-related to both e1 and e2 (as on the present version of the memory constraint). Unsurprisingly, then, this way of defending the memory constraint is open to some of the same objections as views which rely on the existence of an experienceless interval. Here is one way to developing the argument from arbitrariness. Imagine two subjects each of whom have a chain of experiences stretching over a certain interval of time. The first subject’s experiences are all M-related, but for a single break in the chain; during the course of this series of experiences, it happens (as it might, given the memory constraint) that RED phenomenal properties shift from 8 See the example of the general, the officer, and the schoolboy in Essays on the Intellectual Powers of Man. 9 See, for example, the discussion of psychological connectedness vs. psychological continuity in Parfit (1984), 205–6.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the interpersonal, time, and memory constraints 51 first representing the color red to later representing the color green. The second subject differs from the first only in that her experiences have no break; each is M-related to an earlier experience. Could that really suffice for their RED experiences to differ in their content so wildly over the rest of their lives?10 But even setting aside worries of this sort, our modified memory constraint, which uses indirect-M rather than M, is open to serious objections. Let’s suppose that we have a subject who has two experiences, e1 and e2, which at the time of e2 are not M-related—and not indirect-M-related either. It seems extremely plausible that a subject’s phenomenal properties, and the facts about what that subject visually senses, can’t depend on later developments in the life of the subject. Let’s suppose that this subject satisfies the conditions (whatever they are) for variance in color phenomenology with no variance in color property represented; so let e1 occur at a time at which instantiating RED involves visually sensing red, and e2 occur at a time at which instantiating GREEN also involves visually sensing the color red. But now suppose that the subject later has another experience, e3, at which time the subject instantiates RED—and that e3 is M-related to both e1 and e2. e3 presumably represents some color property as instantiated— but which one? Not the color red, since e3 is M-related to e2; but also not any other color, since—given that at the time of e1 the subject is relevantly just like a normal, non-spectrum-inverted, human subject—we could vary the description of e1 to rule this out. So the proponent of the modified memory constraint must say that this sort of situation is not possible—once experiences e1 and e2 happen, this makes it metaphysically impossible for the subject to, later, have another color experience at which time she stands in the relevant memory relation to each of e1 and e2. But this is very hard to believe—why should this be impossible? One can, after all, ‘recover’ lost memories later in life.

10 We could also give a version of the argument from vagueness if the memory relation M admits of borderline cases, as it presumably will.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

8 Phenomenal Variance and Property Variance So far I’ve argued that phenomenism entails, falsely, the possibility of Scenario A—does it also entail the possibility of Scenario B? To answer this question we have to consider the relationship between phenomenal variance and the reverse claim, that we can get arbitrary differences in color property sensed while holding fixed color phenomenal properties: Property variance: Possibly, two subjects visually sense arbitrarily different color properties as instantiated, but have the same color phenomenal properties. A plausible argument can be made that phenomenal variance and property variance are equivalent:1 That phenomenal variance entails property variance: Imagine (as phenomenal variance tells us is possible) that A is instantiating RED while visually sensing redness as instantiated, and B is instantiating GREEN while also visually sensing redness as instantiated. Now let A go on to instantiate GREEN; presumably it is possible that A visually sense green while so doing (just imagine that A is a normal human subject). But this entails (given the facts about B’s experience) that a pair of subjects can be alike in phenomenal properties and differ arbitrarily with respect to which color property they visually sense as instantiated. And this just is property variance. That property variance entails phenomenal variance: Imagine (as property variance tells us is possible) that A instantiates the phenomenal property RED while visually sensing redness as instantiated, and B is also instantiating RED but is visually sensing greenness as instantiated. Now

1

See, among other places, the discussion of content inversion in Egan and John (unpublished).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal variance and property variance 53 let A go on to instantiate GREEN; presumably it is possible that in so doing she visually senses the property greenness. (Just imagine that A is a normal human subject.) Then B and A both visually sense the color property green as instantiated, despite differing widely in their phenomenal properties. And this just is phenomenal variance. So the phenomenist should be a super-phenomenist, the super-phenomenist must endorse phenomenal variance, and phenomenal variance is equivalent to property variance. So the phenomenist is committed to property variance. But property variance entails the possibility of Scenario B. The argument here is very much like the argument from phenomenal variance to the possibility of Scenario A. The proponent of property variance who wants to deny that Scenario B is possible, on the grounds that a subject can’t be alike with respect to her visual phenomenal properties while differing arbitrarily in which color properties she visually senses as instantiated, must either say that this is possible for distinct subjects but not for a single subject at different times, or that this is possible for a single subject at times separated by a sufficient interval of time t but not for a subject at times separated by less than t, or that this is possible for experiences of a single subject not connected by a certain memory relation but impossible for experiences which do stand in that memory relation. That is, he must accept the reverse interpersonal constraint, the reverse time constraint, or the reverse memory constraint:

The reverse interpersonal constraint

Two subjects can instantiate the same color phenomenal properties, but visually sense arbitrarily different color properties as instantiated; but this is not possible for a single subject at different times.

The reverse time constraint

A single subject can, at different times, instantiate the same color phenomenal properties, but visually sense arbitrarily different color properties as instantiated; but this is not possible for a single subject at times separated by less than some minimal interval t.

The reverse memory constraint

A single subject can instantiate the same color phenomenal properties, but visually sense arbitrarily different color properties as instantiated, so long as those two experiences are not related by a certain memory relation M; but this is not possible for two experiences of a single subject which are connected by M.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

54 phenomenal variance and property variance So, to show that phenomenism entails the possibility of Scenario B, it suffices to argue against the reverse interpersonal constraint, the reverse time constraint, and the reverse memory constraint. But the arguments against these three theses are exactly parallel to the arguments given against the interpersonal constraint, the time constraint, and the memory constraint earlier.2 We can sum up our two-pronged argument against phenomenism—and hence in favor of a local intramodal interpersonal intentionalism about visual phenomenal properties—as follows: Phenomenism

Super-phenomenism

Property variance

¬Reverse interpersonal constraint ¬Reverse time constraint ¬Reverse memory constraint

◊ Scenario B

Phenomenal variance

¬Interpersonal constraint ¬Time constraint ¬Memory constraint

◊ Scenario A

2 With the exception that the argument against the reverse interpersonal constraint is in one minor respect stronger than the argument for the original interpersonal constraint: teleological theories of content aren’t even an initial objection to the former, since the hypothesis that the contents of internal states are fixed by the evolutionary history of the relevant organism does not rule out the possibility that distinct internal states—which might be associated with distinct phenomenal properties—might represent the same color property.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

9 Extending the Argument to Other Phenomenal States The preceding chapters were an attempt to establish the truth of an interpersonal intentionalist thesis. The argument was only for the weakest such thesis—for the intramodal, local version of it restricted to the case of visual phenomenal properties. However, it seems that the line of argument pursued there generalizes readily to at least some other types of phenomenal properties, including those associated with perceptual experiences in other sense modalities, and at least some bodily sensations and moods. To show this, we can construct analogues of our examples of psychedelic phenomenology þ constant content and constant phenomenology þ psychedelic content for these state types. Consider the following cases: Hearing

Touch

Pain

A subject is listening to the radio. Over the course of a few seconds, her phenomenal properties change rapidly from the phenomenal property characteristic of your experience of listening to a motorcycle starting, to that characteristic of your experience of listening to a high-pitched whistle, to that of a quiet hum. But throughout, she is auditorially sensing the presence of a constant sound with the pitch, timbre, and loudness of a typical experience of white noise.

A subject is running his hand over the surface in front of him. Over the course of a few seconds, his phenomenal properties change rapidly from the phenomenal property characteristic of your experience of running your hand over silk, to that characteristic of your experience of running your hand over sandpaper, to that characteristic of your experience of running your hand over a furry stuffed animal. But, throughout, he is tactilely sensing the surface as perfectly smooth.

A subject is lying in bed. Over the course of a few seconds, her phenomenal properties change rapidly from the phenomenal property characteristic of your experiences of a throbbing toothache, to that characteristic of your experiences of a stubbed toe, to that characteristic of your experiences of a headache. But, throughout, she is sensing the constant presence of a cramp in her thigh.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

56 extending the argument to other phenomenal states Each of these could be modified in obvious ways to provide similarly absurd examples of constant phenomenology þ psychedelic representation of the relevant sensed properties: an experience with the constant phenomenology of listening to white noise but a content rapidly switching from representation of a high-pitched whistle to representation of a quiet hum, etc. Are these sorts of scenarios as plainly impossible as our original Scenarios A and B? It seems to me that they are. The only reason I can imagine for treating these cases differently would be the view that visual experiences do represent the world, whereas, for example, pains do not. The view that pains lack representational content faces well-known problems—including the problem of explaining the locatedness of pains, which figures in the example experience given above. But it’s by no means as obvious that pains involve truth-sensitive relations to contents as this is in the case of vision. Later, in Chapters 25 and 26—when we discuss the question of whether distinct sense modalities and types of bodily sensations are best understood as involving distinct representational relations or as involving aspects of a single integrated representational relation—we’ll find some reason for thinking that pain experiences do, like visual and other experiences, involve the subject standing in representational relations to contents. Till then, I’ll simply assume that they do—though the reader skeptical of this view can for now bracket this assumption without loss. The argument sketched in Chapters 4–8 in favor of intramodal intentionalism applies in the first instance only to visual experience, and hence establishes only a local intentionalist thesis. The point of this chapter has been to show that the form of argument exemplified there seems to generalize to establish intramodal interpersonal intentionalist theses about audition, touch, and pain. The interpersonal, time, and memory constraints are, after all, no more plausible when applied to these mental states than when applied to the case of vision. This raises the question of whether this form of argument generalizes to every type of phenomenal property—those associated with each sense modality, each type of bodily sensation, each mood, etc. If it does, then we have a recipe for establishing a global intentionalist thesis (albeit, so far, only an intramodal one). I don’t have any direct argument that it does—but it seems to me that consideration of cases counts in favor of the global thesis. No matter what sort of phenomenal state we consider, it always seems that we can construct cases of psychedelic phenomenology and constant representation of the relevant properties, and cases of constant phenomenology and psychedelic representation of the relevant properties—and for any type of phenomenal property, the cases seem, as in the case of our original Scenarios A and B, clearly impossible.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART III

Intermodal Intentionalism and Nonconceptual Content

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

10 Belief and Intermodal Intentionalism The last chapter presented an argument which pushes in the direction of a global rather than a local interpersonal intentionalism; let’s now turn to the question of whether the strongest intentionalist thesis is an intermodal or a merely intramodal thesis. As noted in Chapter 3, the distinction between intermodal and intramodal intentionalisms is one of degree. In our preferred idiom, talking about the representational properties of subjects rather than the representational properties of mental states, the distinction is framed in terms of distinct representational relations in which subjects stand to propositions rather than distinct mental state types. The intermodal intentionalist about a pair of representational relations R1 and R2 holds that, necessarily, for any proposition p and any pair of subjects which differ only in that one stands in R1 to p whereas the other stands in R2 to p, the two subjects instantiate the same phenomenal properties. The purest intermodal intentionalist holds this thesis for any pair of representational relations. Equivalently, the pure intermodal intentionalist holds that any subjects alike with respect to the class of propositions to which they stand in some representational relation or other—no matter which representational relations are borne to which propositions—will also be alike with respect to their phenomenal properties. The purest intermodal intentionalism seems to be open to a quick and decisive objection. Let’s suppose that you are visually sensing a homogenous red surface; let p be the proposition to which you stand in the visual sensing relation. Now imagine a subject representationally like you except for the fact that that subject believes p, and does not visually sense p. It seems clear that such a subject would differ phenomenally from you. But then it just follows that the purest intermodal intentionalism must be false. This argument can seem fatal to intentionalism if we begin by thinking that phenomenal and representational properties are, at root, properties of experiences and mental states. For on this way of looking at things, mental states are the things

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

60 belief and intermodal intentionalism which have or lack phenomenal character, and our question is what property of mental states their phenomenal character supervenes on. The foregoing argument seems to show that this can’t be the property of having a certain content, since beliefs and experiences can share a property of this sort while differing in their phenomenal character. But then isn’t this just to deny any sort of intentionalist thesis of the supervenience of phenomenal character on content? If one pursues this line of thought, then, if one wants to preserve some intentionalist thesis, one is forced to find some way of blocking the conclusion that experiences and beliefs can have the same contents. There are different ways of doing this, but one prominent strategy is to take phenomenal representational states to be self-representational. Perhaps my belief and my visual state might both represent the wall as red; the difference, this sort of theorist says, is that the visual state, but not the belief, also represents itself as so doing. It seems to me that this is an inadequate solution to a false problem. The solution is inadequate because it fails to do what it was designed to do—namely, block the possibility of beliefs and experiences which share a content. For it is very hard to see why we couldn’t have a belief or occurrent thought which not only represented the wall as red, but also represented itself as so doing. And even if this would be sufficient for the thought in question to have a phenomenal character—as some proponents of ‘cognitive phenomenology’ hold—it surely would not be sufficient for it to have the phenomenal character of a visual experience with the same content as that thought.1 Fortunately (for intentionalism) the problem is not well posed. Remember that the properties we’re ultimately interested in—phenomenal properties—are at root properties of subjects. Our question concerns the supervenience base for these properties; hence we must be looking for properties of subjects—not properties of experiences. So these properties will not be properties like ‘having a certain content,’ since subjects of experiences are not the sorts of things which ‘have a content.’ Rather, they will be properties like the property of standing in such-and-such representational relation to a content. Once we see this, there’s nothing to block the possibility that the supervenience base for phenomenal properties involves not just the contents to which these relations are borne, but 1 One might put extra conditions on the relevant sort of self-representation, as in Kriegel (2009). But, to focus on Kriegel’s theory for the moment, it’s not at all clear why occurrent thoughts couldn’t be, in his terms, non-derivatively, specifically, and essentially self-representing. Any mental singular selfrepresentational states satisfy the first two conditions; and there seems to be no reason to think that perceptual states, but not occurrent thoughts, have their contents essentially. This, of course, does not show that experiences are not self-representational—only that their being self-representational wouldn’t show that they can’t have a content in common with a thought or belief, and hence that their self-representational quality would not explain their distinctive phenomenal character.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

belief and intermodal intentionalism 61 the relations themselves. This is, in fact, just what the defender of intramodal intentionalism who denies the purest intermodal intentionalism says. This is, in a way, a completely obvious point. But it can be missed if we set up our question about phenomenal properties by thinking of them as, in the first instance, properties of experiences or mental states. Let’s return to our argument from the possibility that subjects can stand in sensing relations and the belief relation to the same content to the conclusion that the purest intermodal intentionalism is false. For reasons mentioned in Chapter 3, this is an argument one might wish to block, since every step away from intermodal intentionalism adds another piece to the supervenience base for phenomenal properties—and hence adds one more thing which needs to be explained away by a theorist seeking to naturalize phenomenal properties. One way to block this argument would be to claim that perceptual experiences and beliefs have different kinds of contents, so that no perceptual experience/ belief pair could share a content. This view has been a very popular one in recent philosophy—it is one of the views that is expressed by the thesis that perceptual experiences have ‘nonconceptual content.’ The idea is that there is a distinction between two categories of contents—the conceptual ones on the one side, and the nonconceptual ones on the other—and that (as a matter of necessity) one can stand in the belief relation only to the former, and can stand in the sensing relation only to the latter. We can state this thesis as follows: Nonconceptualism: Necessarily, for any content p, if a subject can stand in the belief relation to p, then no subject can stand in a sensing relation to p.2 ‘Conceptualism,’ as I’ll use that term, will just stand for the negation of Nonconceptualism—that is, the claim that there is some proposition which is possibly the content of both a belief and a perceptual experience. The label is, for reasons which will become clear, not entirely apt—one might believe the negation of Nonconceptualism without endorsing many of the diverse views which are presently associated with the term ‘conceptualism,’ largely because of the unfortunate diversity of the uses to which the term ‘concept’ is put. For that reason, I’d rather call myself an ‘Anti-Nonconceptualist’ than a ‘Conceptualist’—but the latter will do.

2 Why make this a modal claim? Because intermodal intentionalism is itself a modal claim, and it is presumably a necessary truth that no belief can have the phenomenal character of a perceptual state. Hence to save the purest intermodal intentionalism, we need the claim that it is impossible for beliefs and perceptual experiences to share a content.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

62 belief and intermodal intentionalism The literature on the topic of nonconceptual content usually frames the thesis in terms of the contents of experiences rather than the representational properties of subjects; in those terms, Nonconceptualism can be stated as follows: Necessarily, for any content p, if p can be the content of a belief, then it cannot be the content of a perceptual experience. In order to make contact with the existing literature, I’ll sometimes use this formulation, which we can for present purposes regard as equivalent to the ‘official’ formulation given earlier. It is natural to think that if the contents of perception and belief are necessarily disjoint in the way required by Nonconceptualism, this must be explained by the fact that perception and belief are relations to fundamentally different kinds of contents. What could such a difference in kind between contents be? One natural way of spelling this out would be in terms of the distinction between the following two conceptions of content: Russellianism: contents are structured entities the constituents of which are worldly items like objects, properties, and relations. If two mental states represent the same objects as instantiating the same properties and relations, then they have the same Russellian content. Fregeanism: contents are structured entities the constituents of which are ways of thinking about, or modes of presentation of, objects, properties, and relations. Two mental states can have the same Russellian content, and yet differ in their modes of presentation of that content. Given variation in current usage, it’s worth being clear on how I am using the term ‘Fregeanism.’ As I use the term, Fregeans are committed to the existence of entities—senses—which belong to an ontological category distinct from, and correspond many-one to, the objects and properties which they are modes of presentation of. Though this is a traditional use of the label, ‘Fregeanism’ is now sometimes used as a synonym for ‘anti-Millianism’—that is, the view that the contents of simple singular terms are not exhausted by their reference. One might, of course, be a Fregean in this second sense without being a Fregean in my preferred sense of the term—the Russell of ‘On Denoting’ was a Fregean in the second sense about ordinary proper names, but not, of course, a believer in Fregean senses. Though either use of the term might be useful in various contexts, for our purposes it will keep things clearer if we reserve ‘Fregeanism’ for the view stated above, on which it is not simply a synonym for ‘anti-Millianism.’ The most popular Nonconceptualist view assigns Russellian propositions as the contents of perceptual experiences, and Fregean propositions as the contents

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

belief and intermodal intentionalism 63 of thoughts, beliefs, and other like propositional attitudes—that is, that the contents to which we stand in sensing relations are Russellian, whereas the contents to which we stand in the representational relations of thought and belief are Fregean.3 Were this view correct, this would be a way to preserve the purest intermodal intentionalism, since in that case there could be no pair of subjects which differ just in that one bears a sensing relation to a content which the other believes. The importance of Nonconceptualism for the viability of this purest intermodal intentionalism should not be exaggerated. Even if Nonconceptualism were true, a pure intermodal intentionalism would still face some very difficult challenges. For example, Nonconceptualism wouldn’t help with the problem that the states of blindsighters seem as though they could have the same content as a normal visual experience without the usual phenomenal properties, nor would it help with the possibility of subpersonal but contentful psychological states which involve no phenomenal properties at all. For this reason, not too many contemporary philosophers would endorse the purest intermodal intentionalism, even if they sometimes express themselves using slogans—like ‘phenomenal character is representational content of a certain sort’—which can seem to entail the view.4 However, even if the purest intermodal intentionalism turns out not to be defensible for reasons independent of Nonconceptualism, questions about the truth of Nonconceptualism are still important, for two reasons. First, it matters not just whether modal differences are a part of the minimal supervenience base for facts about phenomenal properties; it also matters which those modal differences are. This matters because whatever modal differences are relevant to phenomenal properties must be explained by a theory of the nature of phenomenal properties. And it might be that the modal difference between occurrent judgement and perception is more difficult to give an account of than the difference between normal perceptual states and (for example) the states of blindsighters.

3 For a systematic development of a view like this—even if his scenario contents are not quite Russellian propositions—see Peacocke (1992). A variant assigns sets of possible worlds, or functions from worlds to extensions, as the contents of perceptual states. I’m skeptical of this view of the contents of perceptual experience for the familiar reason that it identifies apparently distinct but necessarily equivalent contents. But, for present purposes, the difference between this view of contents and the Russellian view doesn’t matter much. 4 Tye (2002), 141. See also Kriegel (2002), 180. Someone who has the resources to endorse the pure intermodal view—and is very aware of the sorts of problems discussed here—is Thau (2002). But his view comes (as he recognizes) with the high price of denying that we can name, or have beliefs involving, the properties represented in experience.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

64 belief and intermodal intentionalism Second, our aim in what follows isn’t just to understand the relationship between representational and phenomenal properties, but also, independently, to understand the nature of those representational properties. And to do this we need to understand the nature of the contents to which subjects stand in sensing relations. Discussion of Nonconceptualism is as good a way as any into the topic of the nature of the contents to which we stand in sensing and other representational relations, and this is a topic which will be a focus of the chapters to follow. In the next few chapters I’ll consider some arguments for Nonconceptualism. But before moving on to this, it is important to emphasize that this is a view which needs argument. Our default view, in the absence of argument to the contrary, should be that perceptual experiences and beliefs have the same sorts of contents. To see why, consider the analogous proposal that suppositions and beliefs have different kinds of content. This seems quite implausible, mainly because it seems that there can be something that at one time I suppose to be the case, and later come to believe to be the case. (This is reflected in ordinary speech—we might say that A believes what B is supposing.) The simplest view of this sort of transition— from supposing to believing—is that one really can suppose and believe exactly the same thing at different times—that one really can bear these two different attitudes to one and the same content. And this simple view entails that suppositions and beliefs have the same sorts of contents. But, as mentioned in Chapter 2, we seem to find just the same sorts of transitions between perceptual experiences and beliefs. On the face of it, it seems that your perceptual experience can represent the world as being a certain way, and that you can later come to believe that the world is just that way. The simplest view of this phenomenon is that a proposition which is at one time the content, or part of the content, of one’s perceptual experience can later be the content of a belief. But, plainly, if this is possible, then the contents of perceptual experiences are not different in kind from the contents of beliefs, and the contents to which we stand in sensing relations are not (in the sense presently under discussion) nonconceptual. So why would anyone endorse Nonconceptualism? There are, roughly speaking, two main strategies for defending this view. First, one might try to establish Nonconceptualism directly by presenting an argument for the conclusion that, whatever our views about the nature of the respective contents might be, the contents of perceptions and beliefs must be different sorts of things. Or, second, one might try a more indirect route, by separately defending views about the contents of perceptual experiences and beliefs which, together, entail that those contents are different sorts of things. Let’s consider these in turn.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

11 Direct Arguments for Nonconceptualism I think that attempts to argue directly for Nonconceptualism are usually really arguments for something else. The following informal arguments are a representative sample: [1] Consider an ordinary visual experience—say, the experience of looking at a grassy field. Now imagine trying to describe all of the information given to you by that visual experience. Surely, you wouldn’t be able to do this. To see this, imagine trying to describe the very slight but perceptible differences in color between two blades of grass—the distinctions made in experience are simply more fine-grained than those which you can make in words or in thought. [2] Our ability to have demonstrative thoughts about objects is explained by our perceptual representation of those objects. But then the contents of those perceptual representations can’t already be conceptual; if they were, they would presuppose rather than explain our ability to have demonstrative thoughts about objects. [3] It is implausible to think that all animals capable of perceptual representation possess concepts. Hence the contents of the experiences of these lower animals must be nonconceptual. But it is also implausible to think that the content of my visual experience of a colored surface must be different in kind from the content of the visual experiences of such an animal. So if the perceptual experiences of the lower animals are nonconceptual, so must be the perceptual experiences of human beings.1

1

These are informal restatements of arguments drawn from, respectively, Evans (1982), Heck (2000), and Peacocke (2001), 614. Evans states the intended conclusion of his argument as the claim that the move from perception to belief ‘takes the subject from one kind of state (with a content of a certain kind, namely non-conceptual content) to his being in another kind of . . . state (with a content of a different kind, namely, conceptual content)’ (227). Heck and Peacocke are also explicit that

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

66 direct arguments for nonconceptualism Consider first argument [1]. On the face of it, the phenomenon to which this argument points hardly shows that the information given in perception is of a different kind than the information about the world represented in a belief; it shows, at most, that there’s more of it in the case of perception. What this argument shows (if anything) is that certain propositions can be part of the content of a subject’s perceptual experience without that subject being in a position to entertain those propositions in thought or express them in language. The same can be said about [2]. Surely our ability to have thoughts and beliefs with certain contents would be easier, not harder, to explain in terms of perceptual experiences if the latter had just the same sorts of contents which one could later go on to think and believe. The alleged problem arises not with the idea that perceptual experiences and beliefs have the same sort of content, but rather with the idea that having a perceptual experience with a certain content presupposes or requires the ability to have thoughts involving that content. Hence the conclusion is not that perceptual experiences have some different sort of content, but rather that, in order for perceptual representation to do the explanatory work we want it to do, it must be possible for a subject to perceptually represent some content which, at the time of the perceptual experience, the subject is not able to represent in thought or belief. And [3] is more of the same. In this case, the key step in the argument is the first. From the fact that lower animals don’t possess concepts, it is inferred that the contents of their perceptual experiences must be nonconceptual. What does ‘possess a concept’ mean? Roughly, we can take ‘possesses a concept C’ to be equivalent to ‘is able to have thoughts or beliefs involving C.’ Then to say that lower animals possess no concepts is to say that these animals are not capable of thought or belief. But from this it hardly follows that the contents of their perceptual states are of a different kind than the contents of our beliefs; it only implies that they are able to perceptually represent contents which they, because incapable of any thought, can’t represent in thought. So arguments [1]–[3] have a lot in common. None really seems to be directed at Nonconceptualism, in the sense of the preceding chapter; and each seems to be directed instead at the quite different conclusion that, possibly, some subject of perception is capable of perceptually representing some contents which that subject is incapable of entertaining in thought. If, as is standard, we use ‘possessing a concept’ for, roughly, ‘able to have beliefs and thoughts involving that they’re trying to show that the contents of perceptual experiences are different in kind than the contents of beliefs and thoughts.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

direct arguments for nonconceptualism 67 concept,’ then one might express the conclusion of these arguments as the claim that one needn’t possess the concepts which figure in the contents of one’s perceptual states.2 Later, in Chapters 16 and 17, we’ll be interested in the question whether this claim about concept possession is true. But for now our question is whether the propositions which are the contents of one’s perceptual experiences are of a different kind than the propositions which can be entertained in thought, and it’s very hard to see how to get this conclusion from the arguments above. The arguments above (again, if they work) show that there is some proposition p and subject such that the subject is able to have perceptual experiences with p as content but not able to have beliefs with p as content. How is this supposed to show that, necessarily, for every proposition p and every subject S, it is not the case that S can have both perceptual experiences and beliefs with p as content? The apparent conflation of these two claims—that some subjects are able to perceptually represent some contents which they can’t entertain in thought, and that no subject can perceptually represent and entertain in thought any of the same contents—is a general feature of arguments for nonconceptual content. Very often we find the latter stated as the intended conclusion of an argument, but then, upon inspection, find an argument for the former.3 A natural thought, then, is that defenders of Nonconceptualism are best understood as making the tacit assumption that the former entails the latter—that if it is possible for a subject to perceptually represent some content which she can’t entertain in thought, then, in general, the contents of perception and thought must be different sorts of things. What we should ask, then, is why one might think that the truth of [A]

Possibly, a subject has a perceptual experience with content p but is not able to have thoughts or beliefs with this content (that is, does not ‘possess the concepts’ which would be used to specify the content of the perceptual experience)

shows that [B]

Perceptual experiences and thoughts have different kinds of content (from which it follows that no subject can have a perceptual experience and a thought which have the same content).

2 This view is often called ‘state nonconceptualism’; the term is due to Heck (2000), who was the first to clearly distinguish this thesis from the claim that perceptual states have a different sort of content than do thoughts and beliefs. 3 For a somewhat tiresome catalogue, see Speaks (2005).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

68 direct arguments for nonconceptualism One might argue that the truth of [A] is best explained by the truth of [B].4 We might ask: why are perceptual states such that one needn’t possess the concepts used to specify their content, whereas by contrast beliefs are such that one must possess all the concepts used to specify their content? Wouldn’t this be explained by beliefs, and not perceptual experiences, having a special, conceptual kind of content? Though this argument has some intuitive appeal, that appeal vanishes on inspection. Remember that ‘possessing a concept’ is just shorthand for ‘is able to have thoughts and beliefs involving a concept.’ So what is supposed to need explanation is that (i) one can have a perceptual experience with a certain content without being able to have a thought or belief with that content whereas (ii) one cannot have a thought or belief with a certain content without being able to have a thought or belief with that content. (ii) is just an instance of the triviality that no one does anything which they aren’t able to do, and so needs no explanation. (i), on the other hand, hardly cries out for explanation in terms of a distinction between two types of content. Why should the claim that some creatures possess mechanisms of perceptual representation which are more fine-grained than their mechanisms of belief formation, and hence able to represent more propositions than the latter, indicate anything more than just that? Inferring Nonconceptualism from (i) is like inferring from the fact that there are more channels on cable than on broadcast TV the conclusion that it is impossible for there to be a channel on both. One might try to save the argument from [A] to [B] by insisting that ‘possessing a concept’ is not simply shorthand for ‘able to have a thought or belief involving that concept.’ Perhaps, instead, to possess a concept is to satisfy some stronger condition—to have something like ‘full mastery’ or ‘full understanding’ of the concept, however these notions are to be explained. On that interpretation, we’d be arguing for Nonconceptualism on the basis of (i*)

one can have a perceptual experience with a certain content without fully grasping that content

and (ii*)

one cannot have a thought or belief with a certain content without fully grasping that content. 4

As in Bermudez (2007).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

direct arguments for nonconceptualism 69 One can of course decide to use ‘possess a concept’ in this way, and replacing (ii) with (ii*) does at least give us a non-trivial claim. But unfortunately (ii*) is false—this much is trivially entailed by our assumption that one can think thoughts with contents which one does not fully grasp. (Without this assumption, (i*) and (ii*) would be equivalent to (i) and (ii).) And this of course removes the point of contrast between perception and belief, and with it the very fact that was supposed to need explanation. When we look at direct arguments for the claim that perceptual experiences have nonconceptual content, inevitably what we find are arguments that there is some interesting difference between believing that p and having a perceptual experience with the content that p—but no argument that this is traceable to a difference in the contents rather than to a difference in the relevant representational relations. A more promising strategy for the defender of Nonconceptualism is the twopart strategy mentioned earlier, in which we try to determine the nature of the contents of perception and thought, respectively, and then use these conclusions to show that these are different sorts of contents. In the next chapter, I’ll do half of the work that the Nonconceptualist needs done by arguing that the contents of perceptual experiences are Russellian propositions. I’ll then turn to the question whether this view about the nature of perceptual content can be used to show that the contents of perceptual experiences are different from the contents of thoughts, and so, by establishing Nonconceptualism, defuse the present argument against the purest intermodal intentionalism.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

12 A Dilemma for Fregeanism about Sensed Contents Fregeans about sensed contents, as I’m using that label, hold that there can be differences in the propositions to which we stand in sensing relations which correspond to no difference in Russellian content—that is, which involve no difference in the objects and properties sensed as in the environment of the perceiver. The purpose of this chapter is to present a dilemma for the Fregean about sensed contents who is also an intentionalist. (Non-intentionalist Fregeans are open to the arguments of Chapters 4–8.) Let’s think for a second about the relationship between the modes of presentation which, according to the Fregean, represent color properties and those properties themselves. There are two options here: either the modes of presentation rigidly designate the relevant color properties, or they don’t. Suppose first that they don’t. Then it is possible for a pair of subjects to stand in sensing relations to the same Fregean content—involving the same modes of presentation—but which represent different color properties. The intentionalist Fregean must hold that those subjects instantiate the same phenomenal properties, since to do otherwise would be to say that phenomenal properties do not supervene on representational properties—which is just to endorse phenomenism, and reject intentionalism. So the Fregean who thinks that modes of presentation non-rigidly designate color properties must hold that when we have a pair of subjects alike with respect to the contents of their color experiences but which differ with respect to the color properties their experiences represent, those subjects are phenomenal duplicates. But then the Fregean must admit that phenomenal duplicates can vary with respect to the color properties they represent as instantiated in their environment. And (a familiar point from the foregoing) it’s hard to see how they can avoid claiming that phenomenal duplicates can vary arbitrarily with respect to the color properties they represent as instantiated, for two reasons:

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

a dilemma for fregeanism about sensed contents 71 • Fregean proponents of non-rigid senses typically endorse this view because they want to make room for the possibility of spectrum inversion without misrepresentation.1 But this requires not just some variance in the property represented, but wide variance. • As noted above, it’s hard to admit small variations in color property represented as consistent with phenomenal sameness without admitting arbitrary variation, because (i) it seems that all such variations are created equal, so that one is possible iff any other is, and (ii) from a long enough list of possible examples of small variations in represented property þ phenomenal sameness, we can derive the possibility of arbitrary variations in represented property þ phenomenal sameness. So it seems that the Fregean who is an intentionalist and who thinks that senses non-rigidly designate color properties must accept the thesis of property variance—and hence also the thesis of phenomenal variance. And this means that she must, if the argument of the preceding chapters works, admit the possibility of both Scenarios A and B. How about the Fregean who claims that the relevant modes of presentation rigidly designate color properties? A Fregean of this sort is not (as far as I can see) open to any argument like the preceding ones, and is not committed to the possibility of Scenarios A or B. Now, this sort of Fregean can’t use Fregean senses to do one of the main things they are used in the philosophy of perception to do: that is, explain the possibility of spectrum inversion without misrepresentation in a way consistent with intentionalism. (Because making sense of spectrum inversion in this way entails phenomenal variance, which entails property variance, which is inconsistent with the conjunction of intentionalism and the view that perceptual contents rigidly designate color properties.) But set that aside; there are other possible motivations for Fregeanism about sensed contents. For example, one might be motivated by the idea that the contents of beliefs are Fregean propositions, and the view that for perceptual experiences to provide reasons for belief, they must have the same sorts of contents as beliefs.2 But there are other problems for the intentionalist Fregean who makes use of rigidly designating senses. The Fregean, by definition, must claim that there are differences in the content of perceptual experience which don’t correspond to any difference in Russellian content—that is, which don’t correspond to any

1 2

See, for example, Chalmers (2006) and Thompson (2008). This is one reading of McDowell (1994).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

72 a dilemma for fregeanism about sensed contents difference in the objects and properties the perceiver represents as in her environment. Now consider a pair of subjects who differ with respect to the Fregean contents they sense but not with respect to the Russellian contents they sense. Must they instantiate the same phenomenal properties, or not? Suppose first that they don’t. Then there are a range of distinct phenomenal properties which subjects of experience might instantiate, even if those subjects sense just the same Russellian contents—and hence represent just the same objects and properties as in the environment of the perceiver. This does not immediately imply phenomenal variance, since it does not imply that arbitrary differences in phenomenal character are consistent with sameness in Russellian content, and hence sameness in color properties represented. But it does imply the following thesis: Fregean variance: Possibly, two subjects differ in their color phenomenal properties, but sense all the same objects and properties as instantiated. And one can argue from Fregean variance, using the falsity of the analogues of the interpersonal, time, and memory constraints, to the conclusion that consecutive experiences of a single subject whose memory is working normally can differ phenomenally despite agreeing with respect to their representation of all properties (not just the color properties) in the environment of the subject. This sort of scenario is not as spectacularly impossible as Scenario A—but it is still, I think, clearly impossible. Examine your own experiences. Can you imagine having a pair of consecutive visual experiences which differed phenomenally but in which the color of the represented object, and the lighting, and the background, and indeed every property of every object represented as in the environment, seemed just the same in the two experiences? It seems to me that—with one interesting exception, to be discussed in Chapter 27—we cannot.3 So let’s suppose instead that the Fregean who believes in rigidly designating senses requires that whenever a pair of subjects differ only by sensing different Fregean contents which correspond to a single Russellian proposition, those subjects instantiate the same phenomenal properties. So, according to this sort of Fregean, there are pairs of subjects which have the following three characteristics: (1) they instantiate the same phenomenal properties; (2) they sense just the same objects and properties as in the environment of the perceiver (that is, the propositions to which they bear sensing relations determine the same Russellian content); and

3 These are cases involving certain sorts of attentional shifts. I’ll explain in Chapter 27 why I think that the possibility of cases of this sort is no help to the proponent of Fregean variance.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

a dilemma for fregeanism about sensed contents 73 (3) despite these similarities, they stand in sensing relations to propositions which differ in their modes of presentation of the relevant objects and properties. I think that it is hard to get a handle on these supposed differences in Fregean sense. How could phenomenally identical representations of something be different modes of presentation of that thing? Let’s call this intuitive problem the ‘problem of vanishing senses.’ There are a couple of ways to make this more than an intuitive problem for the Fregean. One begins with a standard test for difference in Fregean sense between a pair of sentences. Speaking roughly, the test asks us to consider whether (holding fixed background beliefs) an attentive and reflective subject who understands both sentences could ever rationally take one to be true, and the other false. If this is possible, we conclude that the sentences differ in sense.4 There is a natural way to extend this criterion for difference in sense to a pair of experiences. We ask—holding fixed the beliefs and other relevant mental states of the subjects—whether it is possible for one subject to rationally take one of the experiences to be veridical, and yet rational for the other subject to take the other to be illusory.5 Does this test ever deliver the wanted result that subjects with characteristics (1) and (2) above differ in the contents they sense? It seems to me that it does not. If two subjects instantiate the same phenomenal properties, then (holding fixed background beliefs) it could never be rational for one, but not the other, to take herself to be veridically representing her environment. The Fregean might respond that this test for difference in content is only supposed to provide a sufficient, and not a necessary, condition for difference in content—hence the fact that subjects with qualities (1) and (2) fail this test does not entail that they sense the same contents, and so does not entail that no subjects satisfy all of (1)–(3).6 This is fair enough. But I think that this should still look a bit puzzling from the perspective of the Fregean. Why should subjects with qualities (1) and (2) always 4 Though, as we’ll see in Chapter 13, there are complications lurking below the surface of this sort of criterion for difference of Fregean sense. 5 Why stipulate that we hold fixed background beliefs? If we didn’t make this stipulation, it would trivialize the test for difference in sense, since presumably even if e1 and e2 are phenomenally identical and (by Fregean lights) have the same sense, it can be rational for A to regard e1 as veridical and B to regard e2 as illusory if, for example, B knows that she’s in an environment quite likely to produce illusory experiences. 6 While Frege himself seemed to want to provide necessary and sufficient conditions for difference of sense, many more recent defenders of Fregean views—like Evans (1982)—have been content to provide a sufficient condition for difference of sense.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

74 a dilemma for fregeanism about sensed contents fail the test for difference in sense, if some such pairs really do differ in sense? I think that the Fregean owes an answer to this question, and I’m not sure what the answer could be. Here’s a different way to try to bring out the oddity of subjects with qualities (1)–(3). We’re familiar from the philosophy of language and the philosophy of mind with pairs of sentences, and pairs of beliefs, which (according to Fregeans) differ in sense but not Russellian content. One would expect the differences in sense that can (according to the present view) exist between experiences alike with respect to phenomenal character and Russellian content to line up with these differences. But they don’t. If the differences in sense between experiences did line up with the differences between beliefs, then we should expect it to be the case that, typically, when we have a pair of subjects which satisfy (1)–(3), those subjects will be justified in holding different beliefs. (After all, the Fregean will typically claim that, in some cases, a Hesperus-belief will justify other Hesperus-beliefs without justifying any Phosphorus-beliefs, and analogously for other beliefs which differ in Fregean but not Russellian content.) Hence we should expect that, at least sometimes, there could be a pair of subjects alike but for the fact that their experiences at some time differed in sense (but not Russellian content), and who instantiated the same phenomenal properties, but who differed in which beliefs they were justified in forming. But I find this very hard to imagine. It seems to me that (again, holding fixed their background beliefs) a pair of subjects who are alike both with respect to their phenomenal properties and with respect to the objects and properties they sense as in their environment will always also be alike with respect to the beliefs it would be rational for them to form. Can we, for example, imagine such a pair of subjects, one of whom would be justified in forming the belief that Hesperus is over there but not the belief that Phosphorus is over there, and the reverse for the other? It seems to me not. Similarly, if the postulated differences in sense between experiences lined up with the differences in sense between sentences, then we should expect it to be the case that, typically, when we have a pair of experiences which satisfy (1)–(3), there should be some sentence which truly reports the content of one, but not the other. But again, I think that we just don’t find this. Consider again, the pair of subjects above, who are alike but for a difference in the sense of their current experience. Would it ever be correct to report the content of the experience of one, but not the other, with the words ‘Hesperus is over there’? It seems to me not. This adds a further layer of mystery to this variant of Fregeanism. Not only do we have to get a handle on how phenomenally identical experiences of a given

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

a dilemma for fregeanism about sensed contents 75 object or property could be distinct modes of presentation of that object or property—we also have to accept the claim that these distinct modes of presentation of that property cannot be differences in the sense of beliefs, or sentences, about that property. This means that the differences in sense but not Russellian content that we find between experiences are inaccessible to thought and to language. This is bad on its own—but particularly bad if one’s motivation for Fregeanism about the content of experience is Fregeanism about the content of thought together with the desire to maintain a tight link between experience and thought. We can think of the argument of this chapter as presenting the Fregean with the following dilemma: A dilemma for Fregeanism Do senses rigidly designate color properties?

No

Yes

Property variance (and hence also phenomenal variance, and the possibility of Scenarios A and B)

Are differences in Fregean content without differences in Russellian content compatible with a phenomenal difference?

Yes

No

Fregean variance

Problem of vanishing senses

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

13 Frege’s Puzzle and the Content of Thought The conclusion of the last chapter was that the contents of perceptual experiences are Russellian propositions.1 To turn this conclusion into an argument for Nonconceptualism, what we need is an argument for the claim that the contents of belief and thought are something else—for example, Fregean propositions. As emphasized earlier, this is a claim which very much stands in need of defense—and not just because the simplest explanation of our frequent and seamless transitions from perceptual experiences to beliefs is that they are different attitudes toward the same contents. From a purely metaphysical point of view, traditional Fregeanism is a very unattractive view, because it posits a new, sui generis category of entity—a sense—whose sole role is to function as the contents of linguistic expressions and mental states.2 Surely, if at all possible, we should avoid this route, and instead opt for a view like Russellianism, which builds contents out of entities—like objects, properties, and relations—in which we already have independent reason to believe. (Of course, this advantage is genuine only if the Russellian can solve the problem which exercised Russell—the problem of explaining how propositions are related to these objects, properties, and relations. More on this in Chapter 14.) This chapter asks whether there is any good argument in favor of a Fregean view of the contents of thought and of language. This is connected to our central topic—the nature of phenomenal properties and their relationship to representational properties—because of its connection to questions about intermodal vs. intramodal intentionalism. But because the focus in this chapter is on thought and language rather than on the representational properties most closely related to perceptual experience, the chapter is to some extent a digression, and readers who 1 Strictly, what follows is that they are no more fine-grained than Russellian propositions. For now I am ignoring the distinction between Russellian views of content and more coarse-grained views, like those which identify the relevant contents with sets of possible worlds. 2 Though this is not a fair criticism of all neo-Fregean views; it is, for example, no criticism of epistemic two-dimensionalism. See the second half of this chapter for discussion of this theory.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 77 are either antecedently disinclined toward Fregean views of thought and language, or who wish to focus on the representational properties involved in perceptual experience, can skip to Chapter 14 without much loss. The central arguments in favor of Fregean views of the content of thought and language are the various versions of Frege’s puzzle. The simplest version of the argument is that, whatever else we might think about names, our view of the meanings of names had better explain the distinction between trivial, uninformative identities like Hesperus is Hesperus and non-trivial, potentially informative identities like Hesperus is Phosphorus. But, given a plausible compositionality principle, we can’t explain this difference if we take the meanings of simple names to be the objects for which they stand. Hence, instead, we must take the meaning of a name to be a mode of presentation of that object—that is, a Fregean sense.3 Most responses to this sort of argument concede that, if Fregeanism were true, it would explain the difference in informativeness between sentences like the pair above. But, these objections say, it is not the best such explanation, for two sorts of reasons. First, there are other, equally plausible accounts of the data which don’t involve positing Fregean senses (for example, attempts to explain differences in informativeness in terms of facts about what sentences are typically used to pragmatically convey, or in terms of coordination relations between expression tokens).4 Second, the view that the contents of expressions are Fregean senses might lead to other problems which are far worse than the inability to explain the relevant differences in informativeness (like, for example, placing implausibly strong conditions on the truth conditions of attitude ascriptions, or leading to problems with the semantics of indexicals or variables, or entailing that names are non-rigid designators). It is, of course, a point of great controversy whether the relevant data can be explained, or explained away, without Fregean senses, and it is similarly controversial that Fregeanism leads to bad consequences like those just listed. While I think that the standard two-pronged objection to the argument from Frege’s

3 Of course, one might deny Millianism about names without being a Fregean—one could be a Russellian descriptivist. I set this view to the side here mainly because it is widely taken to have been discredited by Kripke’s arguments—this is fair game in the present context because eliminating this theoretical option helps rather than hinders the Fregean’s argument here. 4 See, among other places, Salmon (1986) and Fine (2007).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

78 frege’s puzzle and the content of thought puzzle to Fregeanism is a good one, here I won’t press that point. Rather than asking whether the explanation offered by Fregeanism of differences in informativeness is outweighed by other theoretical considerations, I want to ask whether Fregeanism really solves Frege’s puzzle at all. The basic idea behind the Fregean solution to Frege’s puzzle is that we explain the difference between our two identity sentences by, first, postulating the existence of distinct entities—the senses of the two names—and, second, claiming that differences in informativeness (or lack thereof) of sentences is to be explained by the difference between the senses assigned to the relevant names. Because senses are not elements of our ontology on which we have an independent grasp, but are entities postulated for, among other reasons, the explanatory purpose under discussion, it is reasonable to ask for a bit more information about sense—at least enough to enable us to derive the result that ‘Hesperus’ and ‘Phosphorus’ differ in sense. For without at least this, we won’t have a solution to Frege’s puzzle. There are, broadly speaking, two ways in which a Fregean might go about giving us enough information about senses to enable us to derive the result that ‘Hesperus’ and ‘Phosphorus,’ and like pairs of names, differ in sense. The first is to simply tell us what entities senses are, and in so doing give us enough information for us to be able to see which such entity is the sense of which linguistic expression. I’ll call one who takes this route an informative Fregean. Frege himself said notoriously little about what senses are; Frege was not an informative Fregean. But it isn’t as though Frege told us nothing about sense. Even if Frege did not give an account of what senses are which delivers the result that ‘Hesperus’ and ‘Phosphorus’ differ in sense, he did provide conditions for sameness and difference of sense which were designed to deliver this result. I’ll call a Fregean who explicates her theory by giving conditions of difference for sense without providing a constitutive theory of what senses are a minimal Fregean. I’ll begin by discussing some problems for the minimal Fregean’s solution to Frege’s puzzle; after that, I’ll turn to an instance of informative Fregeanism.

Minimal Fregeanism The usual way of stating the conditions under which a pair of names differ in sense is to, first, state the conditions under which a pair of sentences differ in sense and then, second, define sameness and difference of sense for subsentential expressions via difference in sense of sentences which differ only in the substitution of those expressions. Here I’ll assume that the second step of this method poses no serious problems, and focus on the first.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 79 As is well known, Frege suggested that we can give conditions on sameness and difference of sense for sentences as follows: Now two sentences A and B can stand in such a relation that anyone who recognises the content of A as true must thereby also recognise the content of B as true and, conversely, that anyone who accepts the content of B must straightaway accept that of A. (Equipollence)5

Other passages in ‘A brief survey of my logical doctrines’ and elsewhere suggest that Frege took the relation of equipollence in the above sense to be equivalent to the relation of having the same sense. One might then use this relation to provide conditions for sameness and difference of Fregean sense; here I’ll ask whether we might use Frege’s remarks to accomplish the less ambitious task of providing sufficient conditions for difference of sense. One might, at a first pass, formulate the wanted condition of difference for sense as follows: S1 and S2 differ in sense if some subject could be unsure whether S1 and S2 have the same truth value But, as others have recognized, this claim needs qualification—in particular, we need to put some constraints on the sorts of subjects who can be used to test the sameness of sense of a pair of sentences. We can’t, obviously, show that two German sentences differ in sense by pointing out that someone who doesn’t speak German might well be unsure whether they have the same truth-value; we need to at least require that the relevant subjects understand the sentences. And this isn’t the only qualification we need; surely a subject who understood two sentences with the same sense might be unsure about whether they have the same truth-value if that subject was drunk, or very tired, or confused, or distracted, or in any one of the myriad conditions which might cause us to make mistakes about even very simple matters. Later I’ll return to some questions about what, exactly, these conditions involve; but for now let’s set this question to the side, and simply abbreviate these conditions by saying that a subject who meets these conditions of understanding, rationality, and reflectiveness is ‘ideal.’ Then we might state our modified version of Frege’s equipollence criterion roughly as follows: S1 and S2 differ in sense if some ideal subject could be unsure whether S1 and S2 have the same truth value 5 Frege (1906/1991), 197. See Heijenoort (1977) for some skepticism as to whether Frege had an unequivocal view on this topic.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

80 frege’s puzzle and the content of thought The first question I want to focus on is one of interpretation: should we understand the values of ‘S1’ and ‘S2’ to be sentence tokens, or sentence types? It seems that however we answer this question, we run into problems. On the one hand, a plausible argument can be made that certain versions of Frege’s puzzle force us to go for the ‘sentence token’ interpretation. Consider, in particular, one of the instances of Frege’s puzzle discussed by Kripke in ‘A Puzzle About Belief.’ There Kripke discusses the example of Peter, who hears the name ‘Paderewski’ on two different occasions, once as a name for a famous pianist, and once as a name for a statesman. Peter, Kripke points out, may well wonder whether the two are the same, and hence may be unsure whether a token of ‘Paderewski is Paderewski’ is true. Such a use of ‘Paderewski is Paderewski’ is plainly relevantly like the stock example of ‘Hesperus is Phosphorus’ rather than the usual understanding of ‘Hesperus is Hesperus’—like the former, and unlike the latter, it is informative and non-trivial. But now consider Paul, who is under no such confusion. His tokens of ‘Paderewski is Paderewski’ are plainly relevantly like the stock example of ‘Hesperus is Hesperus.’ Hence if we want our Fregean theory to provide a general solution to Frege’s puzzle, that theory had better be able to assign a different sense to Peter’s tokens of this sentence than it assigns to Paul’s. But it is hard to see how our condition could do that, unless it is a condition on the difference in sense of sentence tokens, as opposed to types. This suggests that we should interpret the above principle as a principle about sentence tokens. But then it’s not clear that it even makes sense. The idea was supposed to be that if we want to figure out whether two sentence tokens that I utter have the same sense, we look at the attitudes which other speakers might take to those sentence tokens; we imagine, for example, our ‘Hesperus’ and ‘Phosphorus’ sentences being considered by someone prior to the discovery that the bright object visible in the evening ¼ the bright object visible in the morning. But these people aren’t considering the very sentence tokens that I consider; we don’t, for example, require that they be encountering the very same sound waves, or bits of ink on paper, that I am. Fortunately, this dilemma has a resolution. When we want to evaluate a pair of sentence tokens using our criterion for difference of sense, we do think about the attitudes that a subject could take toward some other pair of sentence tokens. But built into the test is the tacit assumption that when we do this, we can only look at a pair of sentence tokens which stands in some intimate relation Rd to the pair of sentence tokens we wish to evaluate for difference in sense. To give a clear statement of our principle for sameness and difference of

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 81 Fregean sense which incorporates this thought, it will be useful to adopt a few abbreviations: ?(s1, s2) iff an ideal subject considers sentence tokens s1, s2 and is unsure whether s1, s2 have the same truth value 〚s1〛 ¼ the sense of s1 Then we might state our principle of difference of sense as the claim that some instance of the following schema is true: Difference 〚s1 〛 6¼ 〚s2 〛 if ∃ s1 * ∃ s2 * ðRd ðs1 ,s1 *Þ & Rd ðs2 ,s2 *Þ & ?ðs1 *,s2 *ÞÞ The key questions about this principle concern the nature of the relation which should be supplied as the interpretation of the schematic letter Rd.6 In particular, we should ask whether this relation is, or is not, sufficient for sameness of sense. Intuitively, it seems clear that Rd should be sufficient for sameness of sense. For think about what the point of Rd is: the point of this relation is to tell us which sentence tokens we can look at to determine whether some pair of sentence tokens in which we are interested differ in sense. But it would be bizarre if we could derive conclusions about whether s1 and s2 differ in sense by looking at a pair of sentence tokens which don’t even have the same sense as s1 and s2. How could we ever expect to get information about whether s1 and s2 have the same sense by seeing whether an ideal subject could be unsure whether some pair of sentences which differ in meaning from s1 and s2 differ in truth value? This seems to me pretty conclusive; but in this case, we can do better than intuition, and can provide a kind of reductio of the thesis that Rd is not sufficient for sameness of sense. One way to run the reductio employs the plausible assumptions that Rd should be reflexive and transitive. The first needs little comment; the assumption of transitivity can be motivated as follows. Suppose that Rd is not transitive, and that we have three pairs of sentence tokens A, B, and C, such that A,B and B,C are Rd-related, but not A,C. Suppose now that the C-pair, but not the A- or B-pair, is ?-related. This would then entail that the B-pair differed in sense—but would, on the assumption that Rd is not transitive, be consistent with the A-pair having the same sense. But this is very odd. After all, given that A,B are Rd-related, if the B-pair were ?-related, this would be sufficient for the A-pair to differ in sense; surely, given this, the fact that the 6 As a shortcut in what follows, I’ll simply use ‘Rd’ as a term for the wanted relation, rather than as a name of the schematic letter.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

82 frege’s puzzle and the content of thought B-pair differs in sense should also be sufficient for the A-pair to differ in sense. But this would make Rd transitive. Given the reflexivity of Rd, and our supposition that Rd is not sufficient for sameness of sense, we know that at least one sentence token is Rd-related to a pair of sentence tokens which themselves differ in sense. Presumably at least sometimes this pair of sentence tokens will be such that they not only differ in sense, but can be shown to differ in sense by Difference; that is, presumably at least sometimes they will be, respectively, Rd-related to sentence tokens which are themselves ?-related. But this, plus the transitivity of Rd, will entail that our original sentence token differs in sense from itself, which is absurd.7 So it seems that Rd must be sufficient for sameness of sense. This is by itself no objection to Difference; but it does show that we can’t apply our criterion of difference for sense without being given a prior condition for the sameness of the sense of sentence tokens. The key question is then whether we can provide such a condition. Before going on to do this, though, let’s pause to consider the question of whether we really need to formulate our condition for difference of sense in terms of some relation between sentence tokens like Rd. First, one might resist the above argument that our criterion has to be formulated in terms of sentence tokens at all. Might we, perhaps, simply deny that the ‘Paderewski’ case is a genuine instance of Frege’s puzzle, and formulate our criterion in terms of sentence types? Then we’d be rid of the need to find the troublesome relation Rd between sentence tokens. In the end, though, the use of the ‘Paderewski’ example is dispensable. Given the existence of ambiguous expressions, the proponent of the ‘sentence type’ formulation of the criterion of sameness and difference owes us some non-orthographic explanation of what the relevant way of typing expressions is. But explaining the relevant sense of ‘is the same type as’ leads—by argument exactly parallel to the one just given—to the same problems as the attempt to specify relation Rd. Briefly: being of the same type (in whatever we are told is the relevant sense) must either be sufficient for sameness of sense, or not. Suppose for reductio that it is not. Then, given that every sentence token is of the same type (in the relevant sense) as itself, we can show that every sentence token differs in sense from itself. Let s* be some sentence token. Given that s* has the same sense as itself, it must be impossible for there to be an ideal subject who considers some sentence token s’ of the same type as s*, and some token s’’ of the same type as s*, and is unsure whether s’ and s’’ have the same truth-value. Given our supposition that 7 For a more detailed exposition of this line of argument, which weakens some of the assumptions employed here, see Speaks (2013).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 83 sameness of type is not sufficient for sameness of sense (and the assumption that every sentence token is of the same type as itself), it follows that s’ and s’’ might differ in sense. But then presumably it is possible for some ideal subject to be unsure whether sentence tokens which are, respectively, of the same type as s’ and s’’ have the same truth-value—which, given that s’ and s’’ are stipulated to be of the same type as s*, entails (given the transitivity of ‘is the same type as’ under the relevant interpretation) that s* differs in sense from itself. But this is absurd; so our supposition that sameness of type is not sufficient for sameness of sense must be rejected. Hence we end up with exactly the same problem. Whether we formulate our criterion in terms of sentence tokens or sentence types, we can get an interpretation of the criterion only if we are given a sufficient condition for sameness of sense of sentence tokens. This is hardly surprising, since in the present context there seems to be no important difference between specifying a way of grouping sentence tokens into types and specifying a relation between sentence tokens. A second way of resisting the move to Difference might be to question the need to introduce relation Rd in the first place. One might think that the key move in the argument, and the one most open to question, came when I said that when we ask whether an ideal subject could be unsure about whether a pair of sentence tokens have the same truth-value, we’re never really asking about an ideal’s subject’s consideration of those very sentence tokens—rather, we’re asking about an ideal’s subject’s consideration of some distinct sentence tokens which stand in some relevant relation R to the sentence tokens we wish to evaluate. But one might deny this by saying that when we apply principles like these in order to tell whether a pair of expressions differ in sense, what we typically do is consider some possible but non-actual scenario in which some ideal subject is considering a pair of sentence tokens which she understands, but is unsure whether they have the same truth-value. But, the objection continues, just as there is no problem in imagining a possible but non-actual scenario in which I exist, there is no problem in imagining a possible but non-actual scenario in which some actual expression token exists. So we can, after all, use ideal subjects in the way envisaged without introducing any relation between sentence tokens at all; all we have to do is consider non-actual scenarios involving the very expression tokens we wish to evaluate. This is perfectly correct as far as it goes. The problem, though, is that expression tokens—for example, particular inscriptions, or token sound waves—don’t have their contents essentially. Hence if we are allowed to consider, when applying our criterion for sameness and difference of Fregean sense, any

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

84 frege’s puzzle and the content of thought counterfactual scenario involving the relevant expression token, we will be able to trivially derive the result that every expression token differs in sense from itself. So we need some restriction on the counterfactual scenarios which we are permitted to consider. And then we face a problem already too familiar from the above: either this restriction is sufficient for sameness of sense, or it is not. If it is not, then we will be able (by argument parallel to the above) to derive the result that some expression token differs in sense from itself. And if it is, then again we reach the conclusion that we can interpret our account of sameness and difference of Fregean sense only if given an independent sufficient condition for sameness of Fregean sense. It is worth emphasizing that one cannot reply to this argument by saying that the Fregean does not intend principles like Difference to be (in some sense or other) reductive or explanatorily prior to the differences in informativeness which the Fregean ultimately wants senses to explain.8 The problem just raised is not a problem about these principles’ lack of explanatory priority; it’s a problem with—in the absence of a condition on sameness of sense—their lack of positive content. So let’s return to the question of how we might formulate such a condition for sameness of sense. An obvious way to provide such a condition is to consider the reverse direction of Difference, and try something like Sameness 〚s1 〛 ¼ 〚s2 〛 if ¬ ∃ s1 * ∃ s2 * ðRs ðs1 , s1 *Þ & Rs ðs2 , s2 *Þ & ?ðs1 *, s2 *ÞÞ We argued above that Rd would have to be sufficient for sameness of Fregean sense. If the same held for Rs, that would be bad news for Sameness, since that would imply that the principle could provide a sufficient condition for sameness of sense only with the help of a relation which did exactly that. We can’t just assume that what goes for Rd also goes for Rs, since the constraints on the latter are different than the constraints on the former. But in fact we can argue, in a similar if not exactly parallel way, that Rs must also be sufficient for sameness of sense. To do so we need only assume (i) that Rs is reflexive and (ii) that if some expression can be Rs-related to an expression from which it differs in sense, then this is true of any expression. Now select arbitrary sentence tokens s1 and s2 which we stipulate to have the same sense. If Rs is not sufficient for sameness of sense and (ii) is true, it follows that there will be a pair of expressions s1* and s2* which are such that Rs(s1,s1*) and Rs(s2,s2*) but s1

8

Independent considerations in favor of this view are given in Heck and May (2011).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 85 differs in sense from s1*. Given reflexivity, it follows that s1* can s2* differ in sense from each other. But then if Sameness is true it follows that there will be some ideal subject who understands both and doubts whether they have the same truth-value—and this means that Sameness will not deliver the verdict that our initial expressions s1 and s2 have the same sense. But nothing was assumed about these expressions other than that they have the same sense; so it follows from our supposition that Sameness will never succeed in telling us that a pair of sentences have the same sense. Note that this is not a reductio of the conjunction of Sameness with the supposition that Rs is not sufficient for sameness of sense. Quite the contrary; it shows that, if this supposition is true, then Sameness is vacuously true, by showing that the condition for sameness of sense which it offers is never satisfied. And a vacuously true condition is no good for our purposes; for, as we saw earlier, in order to be able to apply Difference, we need some condition for sameness of sense which sometimes actually tells us that a pair of sentence tokens have the same sense. And of course we can’t just accept the conclusion that Rs is sufficient for sameness of sense, since then Rs would do by itself just what we were hoping that Sameness would do with Rs’s assistance. So it looks like Sameness is either vacuous, or true only if we can, independently, come up with some sufficient condition for sameness of sense. Either way, it’s no use to the proponent of Difference. This problem seems a little bit surprising. A standard view on Fregean sense is that, even if we can’t give a perfectly clear account of what senses are, we can at least, via something like the conjunction of Difference and Sameness, provide a clear sufficient condition for difference of sense which is both plausible and will entail that (for example) coreferential names which can give rise to instances of Frege’s puzzle differ in sense. But if the foregoing is correct, this is a mistake. The best move for the defender of Difference seems to be to give up on Sameness, and try to come up with some other sort of condition on sameness of Fregean sense. But this turns out to be difficult to do. One idea—which of course is not so far from some of Frege’s own uses of the notion of sense—is to give a sufficient condition for sameness of sense in terms of some sort of reportability condition: Reportability Token sentence s1 (as used by A in C) has the same sense as token sentence s2 (as used by B in C*) if A could use s1, as he uses it in C, to truly report the claims that B makes with s2, as she uses it in C*.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

86 frege’s puzzle and the content of thought This fits nicely with a naive semantics for attitude reports, according to which an ascription ⌜A said that S⌝ is true iff the content of S in the context of the ascription is the proposition which the subject of the ascription said. Unfortunately, though, Reportability faces two very serious sorts of problems. The first is that it seems to lead to the result that just about any two tokens of a name differ in sense. Remember the example of Peter and ‘Paderewski’, discussed earlier. I suggested that given that Peter’s use of ‘Paderewski is Paderewski’ is relevantly like ‘Hesperus is Phosphorus’ rather than ‘Hesperus is Hesperus,’ any satisfactory Fregean treatment of Frege’s puzzle should assign different senses to the two tokens of ‘Paderewski.’ Let’s divide Peter’s uses of ‘Paderewski’ into pianist-tokenings and statesmantokenings. Suppose that Paul is an ordinary, un-confused user of ‘Paderewski,’ who knows that Paderewski is both a pianist and a statesman; we can further suppose, if it matters, that he doesn’t know that Peter is confused about the identity of the statesman and the musician. When Peter says, ‘Paderewski is my favorite pianist,’ it seems that Paul can truly report his speech by saying ‘Peter said that Paderewski is his favorite pianist’—and can do so even if he knows nothing about Peter’s ignorance, and does not intend to use ‘Paderewski’ in any special way. Hence some of Paul’s tokenings of ‘Paderewski,’ given Reportability, have the same sense as Peter’s pianist-tokenings. Further, when Peter says, ‘Paderewski is the very model of a corrupt politician,’ Paul can truly report his speech by saying ‘Peter said that Paderewski is the very model of a corrupt politician’—and, again, can do so without any special intentions or knowledge of Peter’s ignorance. Hence, it seems, some of Paul’s tokenings of ‘Paderewski,’ given Reportability, also have the same sense as Peter’s statesman-tokenings. But we already know (if the Fregean solution to Frege’s puzzle is to be general) that Peter’s pianist-tokenings of ‘Paderewski’ differ in sense from his statesmantokenings of ‘Paderewski,’ from which it follows, given the transitivity of identity, that some of Paul’s ‘Paderewski’ tokenings differ in sense from some of his other ‘Paderewski’ tokenings. Given that we can come up with Paderewski-type scenarios for virtually any name, it looks like this pattern of argument generalizes to the conclusion that distinct tokenings of a name for a single speaker will almost always differ in sense. This would certainly be a surprising result. But can the Fregean perhaps simply bite the bullet here, and accept the fact that every expression token differs in sense from every other expression token as a surprising consequence of her theory? Not without undermining the basic argument for Fregeanism. After all, on that view, both ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus’ would be identity

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 87 sentences involving name tokens with the same reference but distinct sense— which would leave the intuitive distinction between the sentences unexplained, and Frege’s puzzle unsolved. Fregeans could, of course, simply resist the intuitions about truth-values in the dialogue between Paul and Peter above. But doing this in a principled way would require some account of the conditions under which expression tokens have the same sense—which rules out the attempt to use intuitive judgements about who can correctly report what, in conjunction with Reportability, to provide these conditions. The second problem for Reportability is that it does not really escape the fundamental problem with Sameness. For the condition on sameness of meaning is stated in terms of what B could report using a given sentence token. But (to repeat a now-familiar line of reasoning) when we consider what B could report using a given sentence, we’re really asking about what B could report using a token of a given sentence type. And then we need to know what the relevant type is—or, equivalently, what relation a sentence token must stand in to the one we want to evaluate. This relation might either be sufficient for sameness of sense, or not. If the latter, our principle will be vacuously true and hence uninformative. If the former, then Reportability will provide a sufficient condition for sameness of sense only with the help of another principle which does exactly that.9 We could try instead to give conditions for sameness of sense using the intentions of the speaker: Intentions Token sentence s1 (as used by A in C) has the same sense as token sentence s2 (as used by B in C*) if A intends in C to use s1 with the same meaning as B used s2 within C*. The problem, though, is that my intention that my use of a term have the same sense as yours doesn’t guarantee that it will. One way to see this is that I can have multiple intentions in using an expression, and that these intentions might conflict—I might, for example, intend to use e with the same sense as A’s use of that expression and with B’s, even if, as it turns out, A and B are using the expression with different senses. (Just imagine that I am a committed Millian, and intend to use ‘Paderewski’ with the same sense as all of Peter’s tokenings of it.) As before, one could object that we might only consider B’s possible uses of that very sentence token. But (also as before) this delays rather than solves the problem. Given that sentence tokens don’t have their senses essentially, we’ll have to provide some restriction on the relevant class of possible tokenings, which will amount to giving a relation between token/world pairs which will face the dilemma just sketched. 9

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

88 frege’s puzzle and the content of thought At this stage, you might have the worry that I’ve been attacking a straw man. Can’t we just assume that our intuitive judgements about sameness of meaning are, by and large, correct, and use these judgements, along with something like Difference, to give us the desired result that ‘Hesperus’ and ‘Phosphorus,’ as we actually use them, differ in sense? Indeed, doesn’t our typical practice of evaluating semantic hypotheses always proceed via some assumptions about sameness of meaning across tokens? After all, we often evaluate some semantic hypothesis about an expression by seeing what it entails about various sentences in which that expression occurs—and if this methodology is to make any sense at all, it must presuppose that we can tell when two expression tokens have the same content. Surely we’re just being obstinate if we won’t let the Fregean help herself to that. This line of objection is not implausible. But I think that it is mistaken, for two reasons. First: it is true that we often make use of assumptions about sameness of semantic content across expression tokens when evaluating semantic hypotheses. But that is because the semantic hypotheses in question themselves entail claims about sameness of content across expression tokens. Consider, to take the simplest example, the Millian view that the meaning of a token of ‘that’ is the object for which the relevant token of ‘that’ stands. This tells us that, if this Millian view is true, any two tokens of ‘that’ which stand for the same object must have the same content: here we are relying in part on prior judgements about identities of objects to derive claims about sameness of content across expression tokens. Parallel remarks apply to any semantic hypothesis which tells us what the semantic content of a given expression is: any such hypothesis will provide us with some resources to make judgements about sameness of meaning across expression tokens. But of course this is exactly what the minimal Fregean who explicates his view via principles like Difference does not do: he does not tell us what the meaning of any expression is, but instead tries to tell us, without doing this, when expressions differ in meaning. So this sort of Fregean simply does not give us the resources to make the relevant judgements about sameness of meaning. Second, the Fregean is especially badly placed to suggest that we rely on our common sense judgements about sameness of content, for the Fregean is in the business of denying those judgements. Remember the case of Paul and Peter. Imagine these two talking about Paderewski’s involvement in politics and then, a few days later, talking about Paderewski’s musical abilities. There need be no confusion involved in either of these conversations; they could be, in any intuitive sense, perfect models of communication. By any common sense standard, Paul and Peter’s tokenings of ‘Paderewski’ in these conversations would be standard examples of tokens of name with the same meaning. But the Fregean—at least the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 89 Fregean who wants to give a general solution to Frege’s puzzle which applies to Peter’s use of ‘Paderewski is Paderewski’—is committed to denying at least one of these common sense judgements. Our problem is that Difference tells us nothing about difference of sense in the absence of a non-vacuous condition for sameness of sense which we don’t know how to provide. But this conclusion misses an obvious answer to our problem. For of course we do know how to provide a sufficient condition for sameness of sense: identity. Haven’t we, after all, been assuming throughout that every sentence token has the same sense as itself? Our problems have resulted from our trying to find not just a sufficient condition for sameness of sense, but something like a ‘suitably general’ condition for sameness of sense, which applies at least sometimes to distinct but synonymous expression tokens. This gives us a way to fill in Difference. For if identity is sufficient for sameness of sense, it might seem that we can simply plug in identity as the wanted relation Rd, giving us 〚s1 〛 6¼ 〚s2 〛 if ∃ s1 * ∃ s2 * ðs1 ¼ s1 * & s2 ¼ s2 * & ?ðs1 *, s2 *ÞÞ or, more simply, Restricted Difference 〚s1 〛 6¼ 〚s2 〛 if ?ðs1 , s2 Þ This principle applies only to sentence tokens which are such that some ideal subject reflectively considers them and is unsure about their truth-value. Just how restrictive this is depends upon how demanding our definition of ideal subjects is. It must be pretty demanding, if any principle like Difference is to hold. Given that principles like Restricted Difference are supposed to be not just true of Fregean sense but (in some sense or other) definitive of it, we can safely presume that they must be necessary if true. (It would be very strange if the Fregean were to say that the world could have been such that an ideal subject’s being unsure whether s1 and s2 have the same truth-value is consistent with s1 and s2 having the same sense.) So we know that h ððX is ideal & x is unsure whether s1 and s2 have the same truth-valueÞ  〚s1 〛 6¼ 〚s2 〛Þ which implies the following statement of the necessary conditions for being an ideal subject in the relevant sense: h ðx is ideal  ¬ ðX is unsure whether s1 and s2 have the same truth-value & 〚s1 〛 ¼ 〚s2 〛ÞÞ

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

90 frege’s puzzle and the content of thought So whatever else we might think about what it takes for a subject to be ideal, we know that being ideal in the relevant sense is some condition which is inconsistent with a certain sort of mistake: it is metaphysically impossible for an ideal subject to be uncertain as to whether a pair of same-sense sentences have the same truth-value. Have I ever, even in my best moments, been ideal in this sense? I don’t think so. Nor, I think, has anyone. No matter how well we understand a pair of sentences which have the same sense, and no matter how attentive, rational, reflective, etc. we are, it is surely always metaphysically possible that we have all of these properties and yet are not quite sure that those sentences have the same truth-value. If this is right, then Restricted Difference implies nothing about the difference in sense of any two sentence tokens actually produced—for there are no subjects around which are ideal in the relevant sense to be unsure whether they have the same truth-value and hence make them ?-related. One might respond that I’m taking the reliance on ideal subjects a bit too seriously here. We don’t, the objection goes, need to find some noncircular specification of a set of conditions which is such that satisfaction of those conditions is inconsistent with uncertainty in the face of sentences which share a sense. Rather, we can just think of ideal conditions in the following way: x is ideal with respect to a pair iff ð〚s1 〛 ¼ 〚s2 〛 iff x is sure that s1 is true iff s2 is trueÞ And this is of course a condition which (by Fregean lights) I’ve satisfied with respect to many pairs of sentence tokens at various times. I currently satisfy this condition with respect to the pair of tokens of ‘This chapter is getting too long’ that I have just uttered to myself, since they are in fact synonymous and I’m in fact sure that they are both true. The problem is that if we take this deflationary sort of view of ideal conditions, we empty Restricted Difference of any content. For suppose that I propose that the sense of a proper name is the object for which it stands. Surely this Millian thesis should be inconsistent with any criterion of difference for Fregean senses which deserves the name. But it is not inconsistent with the conjunction of Restricted Difference and the above take on ideal conditions. For to show that it is, we would have to find a subject ideal with respect to (say) a pair of tokens of ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus’ who was unsure whether they have the same truth-value. But suppose a candidate ideal subject, who is unsure whether these tokens have the same truth-value, is suggested. To see whether this subject really was ideal on the above definition, we’d have to first

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 91 know that the relevant sentence tokens differ in sense. But that is of course just the result that we wanted Restricted Difference to provide. So we’re stuck, it seems, with the result that Restricted Difference never directly implies anything about the sense of any actual sentence tokens. The question then arises as to whether, despite this fact, we might still use Restricted Difference to solve Frege’s puzzle. Any attempt to do this would, it seem, have to proceed in two stages: (a) argue that an ideal subject could be unsure whether coreferential tokens of ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus’ have the same truth-value, and (b) argue that on the basis of the fact that these tokens differ in sense for an ideal subject, it is reasonable to believe that at least some actual tokens of ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus’ differ in sense for we non-ideal subjects. One might worry about either step, but step (a) seems to me particularly suspect. The only grip we now have on the notion of an ideal subject is that of a subject in some condition which is such that, necessarily, no subject in that condition can be mistaken about whether a pair of sentences with the same sense differ in truth-value. But then our view about which subjects are ideal will be hostage to our views about sense—rather than (as we wanted) our views about sense being given by the facts about what ideal subjects could or would do. For suppose again that the Millian advances the thesis that the sense of a name is the object (if any) for which the name stands, and hence (given some standard sort of compositionality principle) holds that tokens of ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus’ have the same sense. Of course we non-ideal subjects can be unsure as to whether these sentences have the same truthvalue—but that is neither here nor there. What matters is whether an ideal subject could be similarly unsure. And it is far from clear whether this is possible. For suppose that the Millian’s claim is true. Then it just follows from the above necessary condition on ideal conditions that no genuinely ideal subject could be unsure whether tokens of ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus’ have the same truth-value. And this contradicts claim (a). One might object as follows: OK, but this just assumes that the Millian about names is correct. Suppose instead that we hold that the sense of a name is some mode of presentation of a reference. Surely it is possible for a subject to have as complete a grasp of two modes of presentation as is metaphysically possible, and still be unsure whether they present the same reference. So in the end we can at least be sure that—given Restricted Difference—as they are used by ideal subjects, ‘Hesperus’ and ‘Phosphorus’ differ in sense. And this surely gives

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

92 frege’s puzzle and the content of thought us some reason to believe that they differ in sense out of our non-ideal mouths as well. But it should be clear that this argument is going in exactly the wrong direction. We wanted to use our test for difference of sense to derive (at least) the result that typical tokens of ‘Hesperus’ and ‘Phosphorus’ differ in sense. What we’ve now seen is that to derive that result, we have to use it as a premise.10 Philosophers often ask whether, or claim that, the meanings of expressions are ‘individuated by Frege’s criterion of difference.’ Our conclusion so far seems to be that this claim is either trivially true or trivially false. If one simply means that one’s views about meaning should conform to the criterion, then it is trivially true—a claim that even the most diehard Millian can endorse. If, on the other hand, it means that Frege’s criterion of difference tells us what the meanings of expressions are, it is trivially false—since it is consistent with virtually any view of meaning.

Informative Fregeanism Let’s turn now to the informative Fregean who, unlike the minimal Fregean, need not rely on principles like Difference to explicate her view of sense, but instead simply tells us what entities the senses of expressions are. It is hard to see how to construct a general argument against informative Fregeanism, in part because the space of possible informative Fregean views is so large. So here I’ll focus on the

10

One might think that this shows that the Fregean should simply get rid of reference to what a subject who was rational, reflective, etc. would do, and focus on properties of actual subjects. One person who develops an account of Frege’s criterion along these lines is Schellenberg (2012). Her account is framed, not in terms of what an ideal subject would do, but in terms of what our actual rational commitments with respect to a pair of sentences are. Very roughly, and ignoring some important subtleties, the idea is that sentences differ in sense for a subject if that subject is not rationally committed to taking them to have the same truth-value. This does avoid the problems to do with ideal subjects; but it seems to me that those problems reemerge in another form. For consider a pair of subjects, who are, respectively, unsure whether the following pairs of sentences have the same truth-value: (1a) Hesperus is Phosphorus. (1b) Hesperus is Hesperus. (2a) Secretariat is a horse. (2b) Secretariat is a steed. Subjects can, as is well-known, be in uncertainty with respect to either pair of sentences while satisfying all ordinary standards for understanding the terms. (For an example which brings this out nicely, see Salmon (1990).) Such subjects might not be, in any ordinary sense, irrational. It seems to me that to explain the sense in which the second subject must be violating her rational commitments even though the first is not, we’ll have to use (as earlier) the premise that ‘steed’ and ‘horse’ have the same sense, whereas ‘Hesperus’ and ‘Phosphorus’ do not. But then we’d again be feeding into the account what we wanted to get out of it.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 93 question of whether what I take to be the most promising version of informative Fregeanism solves Frege’s puzzle. I think that analogues of the problems which this view encounters are likely also to be problems for other informative Fregean views, but that is not a point I will argue here. The version of informative Fregeanism I want to discuss is the version of twodimensionalist semantics which David Chalmers has defended in a series of important papers.11 Two-dimensionalist semantics has been put to a variety of interesting uses; our interest, though, is just whether it can provide a solution to Frege’s puzzle. For that reason, we can just focus on the treatment of names provided by epistemic two-dimensionalism. For our purposes, we can think of two-dimensionalist semantics as the view that a name has, in addition to the content a Russellian might assign to the name, a different sort of content which is more sensitive to the epistemic properties of the name. More specifically, we can define the epistemic intension of a name as follows. We begin with a space of scenarios, each of which can be thought of as a conjunction of sentences which is both epistemically possible—in the sense that its negation is not a priori—and epistemically complete—in the sense that for any sentence, the scenario epistemically necessitates either it or its negation. The epistemic intension of a name will be a function from scenarios to individuals. Roughly, a name n will have a reference in a scenario for a subject if there is some description ‘the F’ in the scenario such that, were the subject to learn that the scenario was actual, ⌜n ¼ the F⌝ would be a priori for that subject. How might the assignment of epistemic intensions to names help to solve Frege’s puzzle? The simplest way to explain the differences between identity sentences which lack cognitive significance and those that possess cognitive significance would be to say that an identity sentence lacks cognitive significance iff the two terms flanking the identity sign have the same epistemic intension. If we assume that there is a version of epistemic two-dimensionalism which delivers the result that a sentence is a priori iff it is true in every scenario and that ‘Hesperus is Phosphorus’ is a posteriori—neither a trivial assumption12—then this does explain the difference between ‘Hesperus is Hesperus’ and ‘Hesperus is Phosphorus.’ But (as Chalmers is aware) it seems to give the wrong result for identity sentences involving semantically complex expressions. Consider the following pair: 49 ¼ 49. 49 ¼ the positive square root of 2401.

11 12

See, among other places, Chalmers (2004), Chalmers (2006), and Chalmers (2012). For some doubts about the first, see Schroeter (2005) and Speaks (2010).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

94 frege’s puzzle and the content of thought These sentences seem to differ in cognitive significance in much the same way as our ‘Hesperus’ and ‘Phosphorus’ sentences—despite the fact that ‘49’ and ‘the positive square root of 2401’ have the same epistemic intension. (If they didn’t, then ‘49 ¼ the positive square root of 2401’ would be false in at least one scenario, and hence a posteriori.) A natural alternative is to explain cognitive significance by moving from epistemic intensions as defined above to more fine-grained structured epistemic intensions, thereby finding a difference between pairs of expressions, like ‘49’ and ‘the positive square root of 2401,’ which can be known a priori to be coreferential, on the grounds that the latter but not the former is semantically complex.13 One might then revise our account of cognitive significance to say that an identity sentence lacks cognitive significance iff the two terms flanking the identity sign have the same structured epistemic intension. This is a fair point about the above example, but I think does not in the end help very much. One worry is that the explanation of cognitive significance in terms of semantic complexity and simplicity seems to overgenerate. Consider the following sentence:14 49 ¼ the thing which is identical to 49. This seems to lack cognitive significance, and to be in that respect on par with ‘Hesperus is Hesperus’ and ‘49 ¼ 49.’ But on the proposal that we explain cognitive significance not by difference in epistemic intension but by difference in structured epistemic intension, the above sentence gets lumped in with ‘49 ¼ the positive square root of 2401.’ This seems to be a mistake. The notion of ‘cognitive significance’ is already notoriously difficult to pin down; if we are to claim that ‘49 ¼ the thing which is identical to 49’ has cognitive significance but ‘49 ¼ 49’ lacks it, it is far from clear that any intuitive distinction at all is being captured. But one might also wonder if the view is susceptible to counterexample in the way that simple Millian views appear to be: one might wonder, that is, whether two terms with the same structured epistemic intension can ever be used to formulate an informative identity sentence. To do this, we’d have to find a pair of such semantically simple expressions which are a priori equivalent. Such expressions are not easy to come by (a point which the epistemic twodimensionalist might reasonably take to count in favor of the plausibility of his view). But numerals may provide an example. There seem to be two plausible

13

See Chalmers (2012), 600. Thanks to Billy Dunaway for the example. Other variants include: ‘The number which is identical to 49 is 49,’ ‘That which is identical to 49 is 49,’ etc. 14

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

frege’s puzzle and the content of thought 95 treatments of numerals: either they are semantically simple expressions denoting numbers, or they are semantically complex descriptions uniquely satisfied by numbers. Either way there are problems. Suppose first that they are semantically simple. Then any two numerals for the same number have the same structured epistemic intension. But we can use such numerals to construct informative identities if we permit the use of multiple numeric systems within a single identity claim, like 48 ¼ xlviii. This looks a bit like identities combining terms from two languages, like London ¼ Londres. The latter case poses no problem for the epistemic two-dimensionalist, since she has room to distinguish (for at least some speakers) between the epistemic intensions of the two names for the city. Not so with numerals; given that the numeric identity is knowable a priori, the two names have the same epistemic intension, and hence also, on the supposition that the numerals are semantically simple, the same structured epistemic intension. So suppose instead that we think of numerals as semantically complex descriptions. Suppose, just to have an example, that ‘48’ has the structure of the description ‘four times ten plus eight.’ But then consider the identity sentence 48 ¼ four times ten plus eight. It looks like, on the present view of numerals, both of the expressions in this identity will have the same structured epistemic intension. But the identity is clearly informative. Of course, one might have another view about the structure of the description encoded by ‘48.’ But it seems likely that the present problem will re-emerge, since no matter what structure we choose we will be able to encode that structure using a descriptive phrase. And it seems like the description which results will always be able to be used to form an informative identity sentence.15 So it seems to me that the epistemic two-dimensionalist, like the minimal Fregean, fails to provide an adequate solution to Frege’s puzzle. Consideration of these cases shows, I think, that we should be a little less willing to grant the Fregean (as is usually done) the claim that Fregeanism, if true, would solve Frege’s puzzle. Hence we should be a bit less willing to grant that the central 15 Of course, one might treat some numerals as semantically simple—plausibly, 0–9 in the Arabic system—and others as complex. Then both of the above sorts of worries arise.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

96 frege’s puzzle and the content of thought argument given in favor of Fregean views of the content of thought and language is in the end an argument for Fregeanism at all. . . . Let’s take stock. The main conclusions argued for so far are (i) that intramodal interpersonal intentionalism is true, at least for perceptual experiences; (ii) that the contents of experience are Russellian; and—more tentatively—(iii) that the contents of belief and language are also Russellian. The argument for (iii) consists of the argument for (ii), the argument that the transitions between perception and thought should make the view that each are relations to the same kind of content the default view, and the argument of this chapter that the central argument for Fregeanism about thought and language—the argument from Frege’s puzzle—is a failure. (ii) and (iii)—plus the plausible assumption that it is metaphysically possible for a belief to represent just the same objects and properties as a perceptual experience—entail that (iv) the strongest intermodal intentionalism is false. While conclusion (i) provides some encouragement for the idea that we can identify phenomenal and representational properties, (iv) places some constraints on the form that this identity can take. To get clearer on the sort of property identity that might be in the offing, we’ll have to investigate two sorts of questions. First, Given that the contents of experience are Russellian propositions with objects and properties as constituents, just which objects and properties can be parts of the contents of experience? As we’ll see, our answer to this question is entwined with the answer to a second: What are Russellian propositions, and what does it mean for something to be a constituent of one?

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART IV

The Metaphysics of Representational Properties

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

14 Properties and the Nature of Russellian Contents One could give a very short answer to this question by simply taking propositions as a primitive category in one’s ontology, and taking the ‘constituent of ’ relation as primitive. But this seems to me less than satisfactory—and not just for general reasons of ontological parsimony. There’s something oddly anthropocentric about supplementing an ontology of objects, properties, facts, events, etc. with another category of entity whose sole roles are to be the contents of mental states and sentences, and the primary bearers of truth and falsity. Surely it would be better, all things being equal, to find a class of entity to play these roles among the types in which we already have independent reason to believe. One way to bring out the challenge of giving an account of the nature of propositions which have objects and properties as constituents is, as is well known, due to Russell himself. In the Principles of Mathematics (}54), he wrote Consider, for example, the proposition ‘A differs from B.’ The constituents of this proposition, if we analyse it, appear to be only A, difference, B. Yet these constituents, thus placed side by side, do not reconstitute the proposition. . . . A proposition, in fact, is essentially a unity, and when analysis has destroyed the unity, no enumeration of constituents will restore the proposition. The verb, when used as a verb, embodies the unity of the proposition, and is thus discriminable from the verb considered as a term, though I do not know how to give a clear account of the precise nature of the distinction.

The problem to which Russell is pointing here is often called ‘the problem of the unity of the proposition.’ One way to read Russell’s discussion here is as showing that propositions must be something over and above their constituents, since the list of A, difference, and B has the same constituents as the proposition that A differs from B (as does, for that matter, the distinct proposition that B differs from A). The puzzle, then, is to say what the extra ingredient, beyond these constituents, which is needed to make a proposition could be—and to say how this extra ingredient combines with the constituents of the proposition to form a single thing.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

100 properties and the nature of russellian contents If we begin with the thought that propositions must involve some ‘extra ingredient’ which binds the constituents of the proposition into an individual entity, it is very natural to take this extra ingredient to be some relation which holds between those constituents. On this sort of view, propositions are a matter of this relation holding between the relevant propositional constituents—which is to say that, on this view, propositions are a kind of fact. This is the sort of view which Russell himself held for a while. Notoriously, though, Russell identified propositions with facts which would suffice for those propositions to be true, which entails that there are no false propositions. (So, for example, he identified the proposition that A differs from B with the relation of difference really holding between A and B.)1 But one might adopt Russell’s idea that propositions are a kind of fact without opting for his views about which facts they are. This is a good way to view Jeffrey C. King’s theory of propositions.2 King’s theory is based in part on the idea that the ‘extra ingredient’ we’re looking for has something to do with the syntax of proposition-expressing sentences. This seems plausible especially when we think about another sort of example discussed by Russell, in which we transform a sentence which expresses a proposition into a string of words which expresses no proposition by substituting expressions of different syntactic categories which share the same content. For example, while on Russellian views ‘loves’ and ‘loving’ both have as their content the relation of loving, ‘John loves Jane’ expresses a proposition, while ‘John loving Jane’ does not. This is obviously due to the fact that strings with the syntactic form of the former—those consisting of a name, concatenated with a two-place predicate and another name—express propositions in English, whereas strings with the syntactic form of the latter—those consisting of a name, followed by an abstract singular term, followed by another name—do not. And this in turn suggests that the ‘extra ingredient’ in the proposition that John loves Jane, the thing which is in addition to Jane, John, and the relation of loving, is something which is in some way connected to the syntax of the proposition-expressing string. But how? A first point to make is that syntactic relations between words in a sentence, like the words themselves, make a semantic contribution to the proposition expressed by the sentence. King brings this point out nicely via the example of a possible language, Nenglish, which is like English but for the fact that concatenation of a name and a predicate expresses a proposition which is true iff the referent of the name does not instantiate the property expressed by the predicate. ‘Amelia talks’ would express a different proposition in Nenglish than it 1

Russell (1910).

2

See King (2007).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the nature of russellian contents 101 does in English—despite the facts that in both languages Amelia and the property of talking are the constituents of the proposition expressed by the sentence, and that in both languages the syntactic form of the sentence is that it is a name concatenated with a monadic predicate. King suggests that we think of the semantic significance (in English) of the relation between ‘Amelia’ and ‘talks’ in ‘Amelia talks’ as the following instantiation function from objects, properties, and worlds to truth values: the function which, given as argument an object o and property F, determines the truth value true at w iff o instantiates F at w. If our extra ingredient should have something to do with syntax, and syntactic relations make semantic contributions, this suggests that our extra ingredient should be some relation between the constituents of the proposition which is defined in part in terms of that semantic contribution—and hence, in the case of the proposition that Amelia talks, a relation defined in part in terms of the instantiation function. And that is exactly the sort of theory which King defends. Simplifying a bit, we can, in King’s view, describe the proposition expressed by ‘Amelia talks’ as follows: it is the fact of there being words x and y of some language such that x has Amelia as its content, y has the property of talking as its content, R(x,y), and R encodes the instantiation function.3 King’s focus was on giving a theory of the propositions expressed by sentences. But it must be said that King’s candidates for propositions do not look at all like natural candidates for the contents to which we stand in sensing relations. Whatever it is that I stand in a sensing relation to when I look at the blue sky, it is hard to believe that it is a fact about the existence of languages of a certain abstract sort. But this is the view to which we are led by King’s theory together with the plausible assumption that sensed contents are sometimes also the contents of thoughts and sentences.4 Plausible as the motivations which lead to King’s theory are, one might think that the problem just discussed arises from something strange about the way in which King treats syntactic relations. It seems plausible that King’s view goes wrong by identifying propositions with facts which partly concern linguistic

3

See, for example, King (2007), 37. See for refinements Ch. 4 of King, Soames, and Speaks (2014). In Ch. 10 of King, Soames, and Speaks (2014), King denies this assumption, and suggests that the contents of perceptual experience need a separate treatment (which could of course be analogous in various ways to King’s treatment of the contents of sentences). This seems to me a cost of King’s view, for reasons already discussed: it seems that at least sometimes my experience can represent the world as being some way and, as a result of that experience, I can form the belief that the world is just that way, and could then use a sentence to express that belief. The simplest view of these transitions is that there is, in at least some cases, a single content to which the subject is first related in experience, and then in belief, and then via a sentence which expresses that content. 4

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

102 properties and the nature of russellian contents expressions and syntactic relations, rather than facts which only involve the contents of those linguistic expressions and the semantic contributions of those syntactic relations. This seems natural in part because it amounts to the idea that we should think of syntactic relations in much the same way we usually think about expressions: as contributing their contents, but not themselves, to propositions expressed by sentences of which they are a part. A natural way to implement this idea in the present framework would be to simply identify our ‘extra ingredient’—the relation which holds between the constituents of the proposition—with the semantic contribution of the relevant syntactic relation. Such a view would share all of the advantages of King’s view, and would avoid the, in the present context, unnatural idea that propositions somehow concern linguistic expressions and syntactic relations. Attractive as this sort of view would be, it’s hard to see how to make it work so long as we are thinking of propositions as facts. For how, exactly, should we think about the semantic contributions of syntactic relations, on this sort of view? Exactly which fact, on this sort of view, would be the proposition expressed by ‘Amelia talks’? Here, it seems, we run into a genuine difficulty. Whereas King is able to explain quite clearly which relation holds between the constituents of the facts with which he identifies propositions, it is very hard to describe a relation which is contributed by the syntax of a sentence, and genuinely holds between the constituents of the proposition. It seems that concatenating a name with a predicate ‘says’ that the referent of the name instantiates the property expressed by the predicate. But to identify the semantic contribution of this bit of English syntax with instantiation would be to identify the proposition that Amelia talks with the fact that Amelia instantiates the property of talking, which would just repeat Russell’s mistake of identifying propositions with facts whose existence entails their truth. So, on the one hand, it is very plausible that syntactic relations make semantic contributions to propositions, and it would be extremely convenient if we could think of those semantic contributions as relations which held between the constituents of the proposition expressed by the relevant sentence; but, on the other hand, our inability to express these relations gives rise to the worry that this is just wishful thinking. But a solution to this problem is available if we think of propositions, not as facts, but as properties. Consider again the example of ‘Amelia talks.’ If we think of the semantic content of this sentence as a property, one natural view is that the property is the property of being such that Amelia talks. On this kind of view, what is contributed by the syntax of a simple predication—the semantic significance (in English) of this bit of syntax, in King’s terms—is something like the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the nature of russellian contents 103 three-place relation corresponding to the open sentence ‘__ is such that __ instantiates __’. In the case of the sentence ‘Amelia talks,’ the contents of the name and predicate fill in the second two slots to deliver the monadic property expressed by ‘__ is such that Amelia instantiates the property of talking.’ The view that propositions are a sort of property fits much of our talk about propositions rather naturally. We say that believing a proposition, for example, is taking the world to be a certain way. But if, as it seems, ‘ways things are’ are properties, this indicates that having a belief is taking a certain attitude toward a property. Parallel points might be made about mental states other than belief, and speech acts like assertion; we can hope, intend, and desire that the world be a certain way; and we can say, assert, and suggest that the world is a certain way. Again, if ways things can be are properties, this suggests that the objects of mental states and speech acts—that is, propositions—are properties. Supposing that propositions are monadic properties of this sort, what are these monadic properties properties of? They are properties of, if anything, everything. After all, given that Amelia does instantiate the property of talking, every thing instantiates the property of being such that Amelia talks. Hence the proposition that Amelia talks—that is, the property of being such that Amelia talks—is true iff it is instantiated. So, for example, it is sufficient for this proposition to be true that I, or my house, or a rock, instantiates the property of being such that Amelia talks. It is worth acknowledging that—given the account of truth just sketched—this view comes with some strings attached, because it places some strong constraints on our theory of properties. If propositions are properties which are true iff they are instantiated, then it seems clear that there must be uninstantiated properties—otherwise, we’d end up with Russell’s problem, as we’d have no account of the propositions expressed by false sentences. And, given that there are necessarily false propositions as well, parallel reasoning shows that we’ll also need to accept the existence of necessarily uninstantiated properties. And in general when we think about all of the propositions we need to make room for, it is pretty clear that we’ll need a very permissive theory of properties indeed.5 It might seem that this commitment to a very un-sparse view of properties is a serious cost of the view. But if we are already convinced Russellians, who think of the semantic contents of predicates as properties, this commitment is not much 5

A view of properties which I think would suit my purposes is outlined in van Inwagen (2004). But I think that the view of propositions I am developing would be consistent with various views of what properties are, so long as those views are not committed to any sort of principle of instantiation (or possible instantiation). The view I’m developing is, however, consistent with denying the existence of simple necessarily uninstantiated properties.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

104 properties and the nature of russellian contents of an extra burden. That’s because it is hard to see how Russellianism could be true without a fairly permissive theory of properties also being true. To see this, suppose that we adopted a sparse conception of properties according to which the only properties which exist are the fundamental properties mentioned in a completed physical theory, along with structural properties.6 Now consider a sentence which plausibly attributes a less-than-fundamental quality to something—say, ‘Bob is floundering.’ We need to supply something as the semantic content of ‘is floundering.’ Given the sparse view just sketched, it cannot be a simple property of floundering, there being no such property. So the content of the predicate must be some complex—presumably some conjunction and/or disjunction—of fundamental properties. Suppose, for purposes of argument, that this is the property of being ðF & GÞ or H or : : : for fundamental properties F, G, H . . . Let’s introduce simple predicates f, g, h . . . for these fundamental properties. Using ‘ºxłx’ as a singular term for the property which is expressed by ‘ł’, we can then ask whether ºx ðx is flounderingÞ ¼ λx ðx is ðf and gÞ or h or : : : Þ Suppose first that it is. Then (given Russellianism) the proposition expressed by ‘Bob is floundering’ will also be expressed by Bob is ð f and gÞ or h or : : : which, given plausible assumptions, entails that the two sentences should be substitutable salve veritate in the complements of belief ascriptions. But it’s very implausible that a complex belief ascription involving a large number of terms for fundamental properties could be true of an ordinary speaker who happens to think that Bob is floundering. So suppose instead that ºx ðx is flounderingÞ 6¼ ºx ðx is ðf and gÞ or h or : : : Þ Then the property of floundering must be distinct from the property referred to by the expression on the right side of the identity sign and, since the argument of the preceding paragraph seems to generalize, must be distinct from any property built up by operations like conjunction and disjunction out of fundamental properties. But this just contradicts the sort of sparse theory of properties under discussion, according to which predicates, like ‘is floundering,’ which don’t express 6

For one development of this sort of view, see Armstrong (1978).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the nature of russellian contents 105 fundamental properties must (if they express a property at all) express something built out of the fundamental properties. One might object, in the spirit of some recent work on fundamentality, that the proponent of a sparse view of properties needn’t deny that there is such a thing as the property of floundering; rather, she need only hold that this property is not fundamental, and is instead grounded in some other, more basic, properties. But this sort of sparse-ish view of properties is perfectly consistent with the present view of propositions. That view doesn’t require that propositions be among the fundamental entities; it’s enough that they exist. So perhaps the ontology of properties presupposed by this view of propositions is not, for the Russellian, a very serious extra cost. But even if this is true, and even if this view of propositions does make available a clear explanation of truth, there is one aspect of the traditional theory of propositions which it does not capture. This is the view that, as Scott Soames puts it, propositional attitudes are representational ‘because of their relations to inherently representational propositions.’7 Properties like the property of being such that Amelia talks are not inherently representational things; hence if propositions are properties of this sort, this aspect of the view of propositions common to Frege and (one version of the early) Russell must be rejected. This may seem like a cost; but there is also a benefit here. The idea that an entity can be intrinsically representational has seemed to many to be a puzzling one. If we can give an account of truth and propositional attitudes (about which more in the next chapter) without making use of entities of this sort, this is a good thing. Earlier I raised the question of what propositions are, and of what it means to say that something is a constituent of a proposition. I’ve defended an answer to the former question, but have so far ignored the second. For an object or property to be a constituent of a proposition is for that object or property to stand in some relation to that proposition. What we want to know is: what relation is this? It is not all that difficult to identify the relation which obtains between propositions and their constituents on the present view—though it is more difficult to do this in a metaphysically illuminating way. Consider the properties of being such that Amelia talks, of being such that Amelia walks, and of being such that Amelia sleeps. It seems very plausible that each of these properties stand in some particularly close relation to Amelia in which they stand to no other thing. Now consider the properties of being such that Amelia talks and of being such that Violet talks. It seems very plausible that these properties stand in 7

Soames (2012), 212.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

106 properties and the nature of russellian contents some particularly close relation to the property of talking in which they stand to no other property. Further, it seems that these two ‘particularly close relations’ have something in common, so that there is some relation R in which the property of being such that Amelia talks stands to both Amelia and the property of talking, but not to anything else. On the present view, R is the propositional constituency relation. One naturally wants to press further, and ask: what is this relation R? To answer this question in an informative way, we would need a theory of the nature of complex properties which, much as I would like to, I don’t know how to give. But I’m not sure that we need to be able to give a theory of this sort to make sense of talk about the constituents of propositions. Here I think it is crucial to distinguish between two uses of talk about propositional constituency, one classificatory and one explanatory. Corresponding to these two uses are two different theoretical roles. Classificatory uses of talk about propositional constituency use talk about propositional constituents to single out classes of propositions. On this use, the key elements of the theoretical role which the relation of propositional constituency is meant to play are given in principles like these: If p is a monadic predication, then, if o is a constituent of p, anyone who entertains p predicates some property of o. If p is a monadic predication, then, if F is a constituent of p, anyone who entertains p predicates F of something. We could obviously come up with similar principles for other types of propositions. If one is using ‘constituent of ’ in a merely classificatory way, then this talk is simply a shorthand way of singling out classes of propositions which make true generalizations like the above about the psychological states of subjects who have propositional attitudes involving those propositions. If we take principles like these to be definitive of the theoretical role of propositional constituency, then the claim that propositions have constituents is—while not trivial—pretty metaphysically unexciting. It is not trivial because it will be denied by certain very coarse-grained views of propositions. Consider, for example, the view that propositions are sets of worlds. The proposition that 2 is even seems to be a monadic predication with 2 as a constituent. But that won’t be the case if propositions are sets of worlds, since on that view there is only one necessary proposition, and to believe it one need not predicate any property of the number 2. But, even if it is non-trivial, the claim that propositions have constituents in the merely classificatory sense is metaphysically unexciting because it does not require that the relation of propositional constituency be a

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the nature of russellian contents 107 deep fact about the nature of propositions which is suited to explain central properties of propositions. If the classificatory use of ‘constituent of ’ talk is all we want, then I think that we need nothing more than the explanation of propositional constituency already given. But not all talk about propositional constituency is so theoretically noncommittal. One often sees the claim that some propositions are, for example, about certain objects because they have those objects as constituents. To go in for this sort of claim is to go beyond the classificatory use of ‘constituent of ’ talk to an explanatory use of that notion. To be able to make explanatory use of propositional constituency, we would have to provide some illuminating account of the ‘some close relation’ identified earlier. Otherwise it would be hard to see what the explanatory gain could be in moving from talk about aboutness to talk about the relation R, whatever it is, in which the property of being such that Amelia talks stands to Amelia and nothing else. This is the sort of theoretical pressure which makes attractive the idea that the relation of propositional constituency is, or is closely analogous to, parthood.8 While I’m not opposed in principle to some explanatory uses of the relation of propositional constituency, they won’t play any role in my argument. Hence when I talk about the constituents of propositions in what follows, I will be talking about the relation between propositions and objects, properties, and relations which satisfies the theoretical role associated with classificatory uses of ‘constituent’ talk. This, of course, does not answer the question of what sorts of things can be the constituents of the contents of perceptual experience. I’ll return to that question in Chapters 18–23. But first we need to address the prior question of how the present view that propositions are properties can be integrated into an account of sensing properties, and propositional attitudes more generally.

8

For some of the challenges and possibilities here, see Gilmore (forthcoming) and Keller (2012).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

15 Properties and the Attitudes If propositions are properties, and propositional attitudes are relations to propositions, then propositional attitudes must be relations to properties. These relations will presumably involve the property’s being instantiated. So, believing that grass is green will be believing-to-be-instantiated the property of being such that grass is green; imagining that grass is red will be imaginingto-be-instantiated the property of being such that grass is red; and so on. One might wonder: if someone believes that grass is green, what must they believe to instantiate the property of being such that grass is green? In the case of attitudes like belief and imagination, I think that there is no object which is such that it must be, for example, believed to be such that grass is green. Rather, it’s enough if the subject of the belief simply takes the property to be instantiated by something or other. The propositions which are the objects of such states are, after all, properties of everything or nothing. But the idea that belief and other propositional attitudes are relations to properties, and involve some attitude toward their instantiation, opens up the possibility that other representational properties of subjects are, similarly, relations to properties, but work a bit differently. Even if believing that grass is green does not require that the subject of the belief ascribe the property of being such that grass is green to any particular thing, other properties of subjects do have some requirement of this sort built into their definition. Consider, for example, the property of ascribing the property of being such that grass is green to Barack Obama. This is a genuine, if uninteresting, property of subjects—and it is a property which a subject possesses only if that subject takes the property of being such that grass is green to be instantiated by Obama, rather than some other thing. This is a representational property of subjects; and the relation corresponding to the open sentence x believes the property of being F to be instantiated by Barack Obama is a representational relation between subjects and properties. So the existence of representational relations and properties of subjects which require subjects to

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the attitudes 109 ascribe properties to particular things is not really up for debate; what is up for debate is whether there are theoretically interesting examples of such representational relations. A good candidate for such a representational relation is the one corresponding to the open sentence x believes the property of being F to be instantiated by x which expresses the relation of believing oneself to be some way. The reasons for thinking that this representational relation is a theoretically interesting one are familiar from discussions of first-personal mental states. As many have emphasized, it seems that we should distinguish between my representing a tiger as approaching me, and my representing a tiger as approaching an object o, where I am o—just imagine a case in which, unbeknownst to me, I’m looking at myself in a mirror, and seeing a tiger approach the person whose reflection I see. The question is how we can capture the distinction between these two representational properties of subjects. A first thought is that we can distinguish between these two representational properties by distinguishing between the two propositions which are their contents—namely, the propositions that a tiger is approaching me, and that a tiger is approaching o. But, as is wellknown, this is inconsistent with standard versions of the Russellian approach to content I’ve been defending, since these ‘two’ propositions involve the same relation holding between the same objects, and hence must be the same proposition. To put the point in linguistic form, it is hard for the Russellian to find a difference between the propositions expressed by A tiger is approaching n. A tiger is approaching me. where I am the referent of ‘n,’ just because it is hard to see what the content of ‘n’ and ‘me’ could be, if not their reference. There are well-known objections to descriptivist treatments of these terms, and Russellians can make no use of nondescriptive Fregean de re senses. This suggests that we should distinguish between the above representational properties of subjects not by the propositions they involve, but instead by the representational relation they involve. Perhaps the distinction is not between first-personal and non-first-personal propositions—for it is hard, from a Russellian point of view, to see what that distinction could be—but rather between first-personal and non-first-personal representational relations. Given our view of propositions as properties, it is natural to think that whereas the latter

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

110 properties and the attitudes involve taking a property to be instantiated by something or other, the former involve self-ascription of a property—that is, taking oneself to instantiate the property. This is a good first step. But this difference in attitude is not, by itself, sufficient to capture the distinction in question; we also need a difference in content. For the first-personal state described above can’t simply be believing oneself to instantiate the property corresponding to the open sentence x is such that a tiger is approaching o. where the subject is the value of ‘o.’ After all, one might well believe oneself to instantiate this property without being in the relevant first-personal state. I could see, in a mirror, a tiger approaching an object o, and come to believe that I, along with everything else in the universe, am such that a tiger is approaching o, without believing that a tiger is approaching me. Instead, we should take the content of the first-personal state to be the property corresponding to the open sentence x is such that a tiger is approaching x. Self-ascription of this property does seem to be sufficient for being in the relevant first-personal state. But this complicates our picture, for the property selfascribed by someone who is in such a state is not a proposition. It is, like a proposition, a monadic property; but it is not a property which is instantiated by everything or nothing. The property just mentioned is presumably at this moment instantiated by more than one thing, but fortunately not by me. This widens the gap between first-personal and non-first-personal representational properties; but the gap is not so wide that we should be baffled by the fact that a state of one type can give rise to a state of the other type. Indeed, the view promises to explain the asymmetry between first- and third-personal states, in that we seem to be always in a position to be in the relevant third-personal state on the basis of the corresponding first-personal state, but not—as the literature on the problem of the essential indexical emphasizes—the reverse. The former is explained by the fact that anyone—at least anyone capable of entertaining the relevant properties—who is in a position to see that some object o (whether o is oneself or not) is F is also in a position to see that anything else must be such that o is F. The fact that the opposite direction of inference is not always readily available is just an instance of the fact, familiar to Russellians, that one can believe the proposition expressed by a sentence

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the attitudes 111 n stands in relation R to m where both ‘n’ and ‘m’ are terms of direct reference for some object o, without being in any position to infer the truth of the proposition expressed by ∃x x stands in relation R to x. Just so, one might be in a position to know that I have the property expressed by x is such that a tiger is approaching y relative to an assignment of o to ‘y,’ where I am identical to o, without knowing that I am such that I have the property corresponding to x is such that a tiger is approaching x. So far I’ve been arguing we should recognize two categories of representational properties of subjects: those which involve taking a proposition—a property of everything or nothing—to be instantiated by something or other, and those which involve self-ascription of some monadic property which is, while closely related to one, not a proposition. But one might wonder at this stage why, if we recognize the existence of the second sort of representational property, we need to continue to believe in the first. Wouldn’t it be simpler to assimilate all mental states to the model proposed above for first-personal mental states? This view of the representational properties of subjects was, as is well-known, defended by Roderick Chisholm and David Lewis.1 While in one respect it is more parsimonious than my view—because it does not require a distinction between these two types of representational properties of subjects—I think that there are three reasons why we should not go this route. A first begins with the point that if we are going to deny that there is any distinction between the belief relation and the relation in which the subject stands to a property in first-personal belief-like states, then we’re committed to the claim that when someone has the intuitively first-personal belief that he is on fire, he has some belief whose content is a property like those corresponding to the open sentence x is on fire as on the view of Lewis and Chisholm, or x is such that x is on fire as suggested earlier. 1

See Chisholm (1981) and Lewis (1979).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

112 properties and the attitudes But this seems to violate seeming platitudes about the truth conditions of beliefs, like A’s belief is true iff what A believes is true. On this theory, what A believes—the content of A’s belief—is a property. But what does it take for the properties corresponding to the above open sentences to be true? For simplicity, let’s focus on the property of being on fire. We can’t say that this property is true iff it is instantiated, since (given the above platitude) this would entail that my belief that I am on fire can be made true by someone other than me being on fire. We can’t require for its truth that everyone instantiate the property, since the truth of my belief doesn’t require that everyone be in flames. And we can’t require that I instantiate the property, since that would make the truth of others’ first-personal beliefs that they are on fire hostage to my temperature rather than their own. We could, of course, get round this sort of problem by letting the property self-ascribed in the case of my first-personal belief that I am on fire be the property of being such that Jeff Speaks is on fire, since then we could say that what the subject believes is true iff the property is instantiated—but this, as noted above, would make the view unable to capture the distinctive firstpersonal nature of the target state. The best move here seems to be to abandon the platitude, and say that talk about the truth of a subject’s belief comes apart from talk about the truth of what the subject believes—but this, I think, is hard to accept.2 A second problem stems from the observation that, while in many cases nothing is lost by thinking of a propositional attitude with the content p as a belief that I am such that p is the case, in other cases, understanding the relevant mental state requires that we consider worlds where p is the case, but in which I am not such that p is the case, because I do not exist.3 The most striking case is perhaps the example of the desire that I not exist. On the Chisholm/Lewis theory, this is interpreted as the desire that I instantiate the property of nonexistence. But this seems wrong, since it is impossible that I have the property of nonexistence, and the desire that I not exist is not a desire for something impossible.4

2

One might try to soften the blow by saying that when one has a belief whose content is the property of being on fire, one typically also has a belief whose content is the property of being such that o is on fire, where o is the subject of the belief, and that this belief does satisfy the platitude. But the fact the platitude seems like a platitude because it seems to hold of all beliefs—not just some. 3 See Markie (1988) and Nolan (2006). 4 See King, Soames, and Speaks, Ch. 5 for discussion of some ways in which one might try to reply to this objection.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the attitudes 113 Finally, the foregoing objections aside, I think that the idea that all of my mental states involve attributions of properties to myself should strike us as odd. Imagine someone engaged in purely abstract reasoning—say, a mathematician. It’s very counterintuitive to think of the mathematician as trying to figure out what she is like—it seems like what she cares about is not which properties she has, but what the world is like, or what numbers are like. It should be possible for thinking to be less self-involving than this view makes it out to be. So let’s suppose that subjects have both kinds of representational properties: properties of taking propositions to be instantiated, and properties of selfascribing non-propositional monadic properties. This gives rise to the question: into which category do sensing properties fall? A reasonable case can be made that they should be placed in the second category. The first reason is that perceptual states are a typical (even if not the only) cause of first-personal representational states like those discussed earlier. When one senses an object as in a certain location and then, taking the experience at face value, ascribes that location to the object, the mental state of ascribing that location to the object is plausibly a first-personal mental state: one is ascribing to the object the property of being in a certain location relative to one’s own location and orientation. But, given that (as noted earlier) one can’t always infer firstpersonal states from the corresponding third-personal states, this fact about the origins of first-personal representational states suggests that they have their contents in common with perceptual experiences—which suggests that these, too, are selfascriptions (of a different sort, of course) of properties which are not propositions. A second reason for thinking of perceptual representation as first-personal comes from the possibility of pairs of experiences which are alike with respect to their third-personal content, but which clearly differ phenomenally.5 Here is one attempt to construct such a pair of experiences: Case 1: I am looking at a full-length mirror reflection of myself. I can see a stretch of ground between myself and the base of the mirror, and see that same stretch of ground reflected in the mirror. I see a reflection of a ping pong ball on that stretch of ground, ‘in the mirror.’ However, I do not represent a ping pong ball as on the ground ‘outside the mirror.’ Case 2: As with Case 1, except that I represent a ping pong ball as on the ground ‘outside the mirror,’ and not one ‘in the mirror.’ The ping pong ball is 5 Though I don’t think that he’d accept the present argument (since it depends on intentionalism), it was inspired by Christopher Peacocke’s offhand remark that ‘Examples of persons seen in mirrors suffice to make the point’ ((1992), 71) that a simple Russellian propositionalist treatment of the contents of experience won’t work.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

114 properties and the attitudes represented as the same relative distance from myself as the ball in Case 1 is represented from the image of myself. There is a clear difference in phenomenal properties between the subjects of Case 1 and Case 2. And there also seems clearly to be a difference in the way the subjects of those experiences visually sense their environments to be. But the problem is that, if we think of the sensing properties instantiated by these subjects as relations to propositions, it is very hard to see how to avoid the conclusion that they are relations to the same proposition. Consider for reductio the claim that the sensing properties instantiated in the two experiences are both relations to propositions. Let’s use ‘m’ as a name for the person represented as in the mirror, ‘s’ for the subject of the experience, and let the F-properties be the properties that the person in the mirror is visually sensed to have, and the G-properties be the properties that the ping pong ball is visually sensed to have. Then it seems that we would have the following assignments of propositions to the two cases: Case 1: x is such that m is d1 from the ball & m is F1 & the ball is G1 & s is d2 from the ball & . . . Case 2: x is such that m is d2 from the ball & m is F2 & the ball is G2 & s is d1 from the ball & . . . Since, in this case, m ¼ s, these will be the same proposition unless F16¼F2 or G16¼G2. The first possibility is a nonstarter, since there’s no reason to think that the two cases need involve any difference in the representation of the properties of the person seen in the mirror. But there’s also no obvious reason to think that it must be the case that G16¼G2; why think that there must be some property that the ball is represented as having in one experience but not the other? It might be represented as the same size, color, and shape in each experience, and by the construction of the case we know that it is represented as the same relative distance from m/s in the two cases. There are defensive moves that can be made here, though none seem very appealing to me. One might say that the contents of experiences of objects seen in mirrors don’t contain those objects as constituents; but this would face the problem of accounting for the facts about availability of de re thoughts to be discussed in Chapter 18. One might say that the features of the ball will be represented with more determinacy in Case 2, when the ball is nearer to me; but this seems to be at best a contingent feature of the case. One might say that the ball is represented as at a different absolute location in the two cases; but it seems implausible that, in addition to visually representing objects as some relative distance from ourselves

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the attitudes 115 (and from other objects in our environment) we also represent them as occupying an absolute location. By contrast, if we think of the representational properties involved in experience as self-ascriptions of non-propositional properties, the differences in content between the two experiences are not hard to find. The first experience will involve the self-ascription of the property x is such that x is d1 from the ball & m is F1 & the ball is G1 & s is d2 from the ball & . . . The second, by contrast, will involve the self-ascription of x is such that x is d2 from the ball & m is F2 & the ball is G2 & s is d1 from the ball & . . . These are distinct properties, since the self-ascription of the first involves selfascription of the property of being distance d1 from the ball, whereas the second involves self-ascription of the property of being the distinct distance d2 from the ball. This is as it should be. So there is some reason to think that sensing properties are not relations to propositions, but instead are self-ascriptions of non-propositional monadic properties. There are two further questions about this view of sensing properties which we need to address. The first is: how does this fit with the view of representational properties defended in Chapter 2? There I argued that perceptual experience involves a distinctive truth-sensitive relation, and that truth-sensitivity is sufficient for a relation to be representational. But truth-sensitivity was defined in terms of relations to propositions; can we make sense of the idea that a relation of self-ascription to a non-propositional monadic property is also truth-sensitive? Fortunately, the needed liberalization of the notion of truth-sensitivity is pretty obvious. If we recognize the existence of relations of self-ascription between subjects and non-propositional properties, such a relation R will be truthsensitive iff it satisfies the following condition: Necessarily, if A is R-related to F, then if A is not F, A is misrepresenting the world, and if A is F, then A is correctly representing the world. The veridical/non-veridical distinction is then reconstructed in the same way: a subject who stands in a sensing relation to a property F will be having a veridical experience iff she instantiates F. Suppose, then, that we take the distinctive relation to properties involved in perceptual experience to be a truth-sensitive kind of self-ascription. The second question about this view which we need to address concerns the objections given

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

116 properties and the attitudes above to the Chisholm/Lewis theory of belief. One might wonder whether these objections also apply to the present treatment of sensing properties. The first was that this view must deny the platitude that A’s belief is true iff what A believes is true This would be a problem for the present view if it were a platitude that A’s experience is true iff what A experiences is true But this is no platitude; we typically do not call experiences true or false, and still less do we say that ‘what A experiences’ is true or false. Indeed, depending on one’s faith in ordinary language about such matters, one might even take the fact that we don’t call experiences true or false to be some evidence for the view that experiences are, while closely linked to propositional attitudes, not quite relations to propositions.6 The second objection was based on the difficulty in making sense of states like the desire that I not exist. But this objection also seems not to arise in the case of perceptual experience, since it doesn’t seem possible that subjects perceptually represent their own nonexistence.7 The third objection was that the view of thought as self-ascription makes thought too self-involving. But again this seems not to be an objection to the view of experience as a kind of self-ascription, since experience just is self-involving—we are always representing things as standing in certain relations to us (as for example, a certain distance from us, or in a certain orientation relative to us). Even if the view is not undermined by the objections which make trouble for the Lewis/Chisholm treatment of belief, though, one might worry that the view that experiences are not quite relations to propositions after all somewhat undermines the intuitive case made for the view that experiences have contents. Wasn’t that view, after all, partly motivated by the transitions subjects often make between experience and belief, and the claim that this is best explained by supplying common objects to both types of states? There are two points to make here. The first is that experiences often, as noted earlier, give rise to other first-personal representational states; and this is easier to

6

To be sure, we also don’t say that intentions, desires, or fears are true—but that is plausibly because these states have a different ‘direction of fit’ than beliefs or experiences, and hence are not truth-sensitive relations. 7 Though there is a related problem for my sort of view, if it is extended to the case of desire. Just as I want to distinguish between belief and its first-personal counterpart, so, it seems, I should distinguish between desire and its first-personal analogue. But then I can’t treat the ‘desire’ that I not exist as a genuinely first-personal state—which seems ad hoc.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

properties and the attitudes 117 understand (at least, given our Russellian assumptions) on the present account than on one which assigns propositions as the contents of experience. The second is that, as noted earlier, there is no mystery in how first-personal states can give rise to third personal states; the needed transition is only one from self-ascribing the property of being such that one is F to taking to be instantiated the property of being such that I am F. Consider by analogy the standard view that propositions are both what sentences express (relative to contexts) and the contents of belief. One often motivates this view by noting that someone can believe what a sentence says, and asking how this is to be explained if not in terms of the fact that what the sentence says is the same sort of thing as the content of a belief. But it turns out that, given some plausible assumptions, many sentences might not quite semantically express propositions; perhaps instead they express fragments of propositions, which are then pragmatically enriched to deliver the assertion of full propositions.8 But proponents of this sort of view don’t thereby reject the intuitive datum with which they started, which show that there must be some close connection between the sorts of things which are the contents of sentences (relative to contexts) and the contents of beliefs; rather, they take it that the proposition fragments assigned to sentences by their semantics are sufficiently similar to propositions so as to make possible an explanation of the datum. I take the analogous view of the relationship between experience and belief. It’s worth noting one consequence of this view of propositions, as it applies to the case of perceptual experience, and that is that it brings about a happy unification of two views about experience which are often contrasted with each other. The first is the view that experiences are, or involve, representational properties; and the second is that experiences are, or involve, relations to properties or, as they are sometimes called, ‘sensible profiles.’9 This unification is happy because it offers responses to objections which have been thought to pose serious problems for each view. Those who think that experience involves a relation to a content sometimes meet the objection that the view that experience involves a relation to a proposition makes experience, in the veridical case, unacceptably distant from the scene perceived. Far better, goes the objection, if the experience were a relation to, not an abstract proposition, but something really there in the scene perceived, like the properties which are really instantiated by the objects perceived. But if the sensed contents are properties that, in the veridical case, are instantiated by the subject, the objection dissolves. 8

See, for example, Soames (2008).

9

As in Johnston (2004).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

118 properties and the attitudes On the other hand, property theorists sometimes meet the objection—from, among others, proponents of the idea that experiences are relations to propositions—that it is simply metaphysically bizarre to think that in hallucinations or illusions we are related to an uninstantiated property. But this is surely less shocking if it turns out that every false proposition is an uninstantiated property and hence that commonplace examples of false belief are, similarly, relations to uninstantiated properties.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART V

Availability and the Scope of Perceptual Representation

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

16 Availability So far our focus has been on the consequences of transparency for the connection between the phenomenal and the representational, and for the nature of the contents which we stand in sensing relations. To answer our questions about the scope of perceptual representation—about which objects and properties are constituents of the contents of the relevant experiences—let’s turn to the second thesis about experience mentioned in the Introduction: that one of its roles is to make new contents available for thought. Claims about the availability of contents are claims about the abilities of subjects. For a content to be available to a subject at t is for that subject to be able, at t, to have thoughts, beliefs, or other ‘cognitive’ mental states involving some propositions of which that content is a constituent. Why think that one of the roles of experience is to make contents available for thought? It’s quite plausible that, of all the objects in the world, some are now available to you for thought, and some are not. You are presumably in a position to have de re thoughts about yourself and other objects in your immediate vicinity; but you’re probably not presently able to have de re thoughts about the various socks currently in my sock drawer. There are various things that you might do to remedy this limitation on your cognitive abilities. Perhaps, for instance, you could do so by acquiring, from another competent speaker, a name for one of the socks. But one thing you could clearly do is have a visual experience of my sock drawer—that, at least, would put you in a position to have de re thoughts about some of the socks in the drawer. It’s plausible that having such an experience is only ever part of the explanation of the acquisition of the relevant ability—the subject must also be the sort of being that can have thoughts at all, and thoughts of the right type. But it seems pretty clear that it can be part of the explanation because, to return to the example above, you weren’t able to have de re thoughts about the items in my sock drawer prior to having the relevant experience.1 1 I don’t mean to imply that these are the only things you could do to acquire the relevant ability. The point that I am making here is, I think, independent of the question whether there is some sort

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

122 availability Given that we are thinking of experiences as involving sensing relations to contents, it seems very plausible, as Mark Johnston has argued, that there is some connection between the contents sensed and the contents which the experience makes available for thought.2 We can formulate the connection between sensed content and availability as a sufficient condition for two experiences to differ in content: Availability/Difference Necessarily, if two experiences differ in which thoughts they make available to their subjects (holding fixed the background beliefs and cognitive abilities of the subjects), then the subjects of the two experiences are sensing different contents.3 Given the above explanation of what it means for an experience to make a thought available to a subject, I think that Availability/Difference is quite hard to deny. For suppose that two subjects are sensing just the same content. Then they represent the world as being just the same way; they represent just the same objects as having just the same properties. How—holding fixed background beliefs and other mental states—could one make available thoughts about an object or property which the other did not? It’s worth emphasizing the difference between Availability/Difference and the following stronger principle: Necessarily, if two experiences differ in that one makes available a content F to its subject which the other does not (holding fixed the background beliefs and

of ‘acquaintance requirement’ on de re thought about objects. For skepticism about such a requirement, see among other places Hawthorne and Manley (2012). 2

See Johnston (2004). Johnston frames the topic in terms of the objects, rather than the contents, of experience. 3 To see the importance of the parenthetical qualification, imagine a frog’s visual experience of a particular fly, and my experience of the same fly; and suppose that frogs are incapable of having de re thoughts about flies. Surely this difference between my cognitive capacities and that of a frog does not show that frogs can’t have visual experiences which represent particular flies as having certain properties, just as mine do. There is one sort of exception (due to Daniel Immerman) to Availability/Difference. It seems that (i) distinct token experiences can share a content; but it also seems that (ii) having a token experience puts one in a position to have de re thoughts about that token experience. But these two ideas are jointly inconsistent with Availability/Difference, since the pair of experiences whose existence is guaranteed by (i) would, by (ii), be an example of a pair of experiences with the same content which differed in which de re thoughts they made available. However, this seems to me to be an isolated sort of counterexample, which does not do much to threaten the core intuition behind Availability/Difference. Given that we are able to have de re thoughts about token experiences, any principle which permits the comparison of the contents of distinct token experiences would run into this problem. To avoid it, we could simply append ‘other than de re thoughts about that token experience’ to the above formulation.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

availability 123 cognitive abilities of the subjects), then the first subject stands in a sensing relation to a content of which F is a constituent. The intuitive case for Availability/Difference did not depend on this stronger principle—and that’s a good thing, for this stronger principle is likely false. There seems nothing to block the possibility that acquiring the ability to have thoughts involving one content might (in conjunction with background mental states, perhaps) give one the ability to have thoughts involving some other content. But if this is in general possible, then there is nothing to block there being a pair of experiences which differ with respect to some content F, and hence also differ in that the first but not the second makes F available for thought to some subject—but also, in virtue of those differences, further differ in the fact that the first but not the second makes some distinct content G available for thought. The stronger principle above would license us to infer that G is part of the content of the first experience—Availability/Difference, fortunately, does not. It’s important at this stage to consider a question about how this discussion of Availability/Difference fits into the method of this book. In Chapter 2, I noted that subjects stand in a variety of representational relations to contents, and singled out a class of these—sensing relations—which fix the conditions under which subjects misrepresent, and correctly represent, the world in experience. I then argued in Chapters 3–12 that we stand in these relations to Russellian contents, and that certain of our phenomenal properties supervene on the facts about which Russellian contents we stand in sensing relations to. When I advance claims about availability here and in what follows, I’m making claims about the relations between those very same sensing relations and the abilities to have certain thoughts. But, given the multiplicity of the representational relations in which we stand to contents, one might wonder how we know that a principle like Transparency/Difference holds for these sensing relations, rather than for some other representational relation in which subjects stand to contents. This is a fair question, given that none of the above principles follow immediately from the connections between sensing and the conditions under which subjects misrepresent their environments (by which we picked out sensing relations in the first place) or between sensing and phenomenal properties (which we argued for earlier). I think that the best answer to this question is to note that there is a tight connection between the phenomenal properties associated with an experience and the contents made available by that experience. A given color experience might make available thoughts about a particular shade. A change in the phenomenal

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

124 availability properties of the subject of that experience will make available thoughts about a different shade. This connection between phenomenal properties and contents made available for thought, moreover, seems to hold necessarily. This suggests that, whatever the availability-explaining representational properties subjects instantiate in experience, they must be representational properties which are necessarily connected to phenomenal properties. Sensing properties, we have seen, are also necessarily connected to phenomenal properties. So, while nothing strictly rules out the possibility that subjects stand in some other sort of representational relations to contents which also stands in a necessary connection to phenomenal properties, the most economical view seems to be that the representational properties which explain the facts about availability are the sensing properties which have been our topic so far. Let’s return to the discussion of Availability/Difference. That is a principle which allows us to draw conclusions about differences in sensed content across subjects from facts about which thoughts are available to those subjects. But it is very natural to think that if some principle of this sort, which involves comparisons between subjects, is true, then some more basic principle which governs the connection between sensed content and availability for thought of particular subjects must also be true. A natural and simple suggestion is something like If an object or property is part of the content sensed by a subject at t, then that object or property is available for thought to that subject at t. But this seems excessively strong, since it rules out the possibility of creatures who instantiate sensing properties, but are not able to have thoughts about the world—and, prima facie, it is hard to see why such creatures should be impossible. We could qualify the principle so that it only applied to creatures who were able to have thoughts. But it is not obvious that this qualification would go far enough. Just as it is plausibly possible that there are creatures able to perceptually represent the world but not have thoughts about it, so it seems possible that there are creatures with quite refined perceptual systems which have the ability to think, but whose cognitive abilities are quite rudimentary. Such creatures would be counterexamples to the above principle, even if qualified. In the end, for our purposes, it will be best to simply bypass these sorts of worries. What will matter for our purposes is whether normal adult human beings are the sorts of creatures just described: whether our cognitive abilities are, relative to our ability to perceptually represent the world, rudimentary. Let’s state the thesis that they are not as follows:

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

availability 125 The Availability Requirement If an object or property is a constituent of the content sensed by a normal adult human subject at t, then that object or property is available for thought to that subject at t.4 This principle, if true, allows us to derive negative claims to the effect that a certain object or property is not perceptually represented by a subject at t from the corresponding negative judgements about that subject’s abilities to have certain thoughts involving those objects and properties. This principle does not entail Availability/Difference—but if it were false, then this would cast some doubt on Availability/Difference. For suppose that the Availability Requirement were false, and that there could be ‘hidden’ contents of experience, which were not available to the thought of the subject in question. In that case there seems nothing to block there being a pair of experiences with the same content, but where part of the content is hidden in the case of one experience, but not the other. And that would falsify Availability/Difference. Why think that the Availability Requirement is true? One simple argument for it is given by reflection on examples. Consider your present visual experience of the shapes and colors of objects around you. Can’t you consider whether those objects really have those colors and those shapes? If so, then the color and shape properties presented in experience are available to you for thought. If this seems to be true of representations of color and shape properties, then one might think that this generalizes to all perceptual representation of objects and properties. For any way your perceptual experience represents the world as being, can’t you wonder whether it really is that way? Even if a prima facie case can be made in favor of the Availability Requirement, the claim that every part of the content of your experiences are available to you for thought (at least at the time of the experiences) sounds a lot like the claim that the contents of experiences are concepts, and that for a concept to be part of the content of an experience, you must possess that concept. But this claim has been the target of a number of challenging and influential arguments in the recent literature about nonconceptual content; in the next chapter, I discuss some of these. 4

It might be worth adding a word about why the restriction to times is included in the statement of the Availability Requirement. Suppose that I’m at the store buying paint for my house, and am looking at a 5 x 5 array of color squares. While having the experience, I am able to have thoughts about each of the 25 colors; I can simply attend to it and then think about, for example, whether that particular color is one that would look good on the staircase. But this ability might well be fleeting; I might individually consider each of the 25 colors in this way, but, 10 minutes later, after inspecting a few more 5 x 5 arrays, have lost the ability to have de re thoughts about the second color from the left in the fourth row of the first array. An example of this sort is discussed—though a different moral is drawn—in }3 of Kelly (2001).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

17 Demonstratives and the Availability Requirement The intuitive case in favor of the Availability Requirement was that for any way the objects in your environment are presented as being, you can always judge that, or wonder whether, they are that way. This suggests that, as John McDowell emphasized it in response to Evans’ classic defense of nonconceptual content in The Varieties of Reference, one fundamental way in which we are able to have thoughts involving the properties we sense as in our environment is via demonstration. For demonstrative thought to play the role in the defense of the Availability Requirement that I’d like it to, the following thesis must be true: Demonstrative Availability If a property is a constituent of the content sensed by a normal adult human subject at t, then the subject is able, at t, to demonstrate that property. The point made above—that for every way our experience presents an object as being, we can judge that, or wonder whether, it is that way—amounts to a way of pointing out the prima facie plausibility of Demonstrative Availability. Nonetheless, it’s a thesis about which many have been very skeptical. Why? The main arguments against the Availability Requirement (and hence also against the stronger thesis of Demonstrative Availability) are versions of arguments [1] and [2] (ostensibly) for Nonconceptualism discussed in Chapter 11. It’s worth emphasizing, to belabor a point already belabored in Chapter 11, that the sense of ‘nonconceptual content’ that is relevant to the Availability Requirement is not the same sense that is relevant to Nonconceptualism, as defined earlier. Since using ‘nonconceptual content’ in both of these two ways is obviously a bad idea, I won’t do it, and instead will stick to the language used to state our theses about Availability, which are framed as theses about the relation between sensed contents and the abilities of subjects to have certain thoughts.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

demonstratives and the availability requirement 127 Argument [1] was the argument from fineness of grain. Now, at first glance, the appeal to demonstratives seems (as McDowell noted) ready made to answer this argument. Even if I lack general color terms for each of the distinctive shades of green represented as exemplified by the blades of grass in my field of vision, I can surely think about these colors by using demonstratives like ‘that color’ or ‘that shade.’ However, one might worry that this response to the argument from fineness of grain introduces too many distinctions.1 Suppose you and I are both examining a blade of grass, and that each of us is representing the blade of grass as the same color. I could refer to the color represented by my experience as ‘that color’ while you do the same with ‘that shade.’ But surely these complex demonstratives differ in content; hence it can’t be that both capture the contents of our respective color experiences, these being the same. But, the argument goes, there’s no nonarbitrary way to choose between these demonstratives; so if one captures the contents of the relevant color experiences, both do. Since both can’t, neither does, and Demonstrative Availability is false. There are two ways to respond to this argument. The simplest is to say that the relevant demonstratives are ones which directly refer to the color presented in the relevant experience. I think that it is plausible that we could use either ‘that shade’ or ‘that color’—or both—in this way. We can also, more simply, just use ‘that’— so long as this bare demonstrative is accompanied by attention to the relevant color, and the intention to demonstrate it. But there is also a less committal option. Perhaps complex demonstratives are not devices of direct reference, but instead are, as Jeffrey King and others have argued, devices of quantification.2 On (one version of) this sort of view, a sentence like ‘That color is F’ will express the proposition expressed by [the x: color(x) & Gx] Fx for some contextually determined property G. In some cases, the contextually determined property will be the property of being identical to the intended referent of the demonstrative phrase; so, in some cases, the contextually determined property will be the property of being identical to some perceptually represented color.3 In this case, no less than in the direct reference case discussed earlier, the color property which is a constituent of the sensed content will also be 1

See Peacocke (1998). See King (2001). Many of the points made in this chapter would also carry over to a view like that of Richard (1993), according to which complex demonstratives are articulated terms. 3 See the discussion of perceptual vs. descriptive intentions in Chapter 2 of King (2001) for one plausible way of working this out. 2

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

128 demonstratives and the availability requirement a constituent of the subject’s demonstrative thought—despite the fact that on this view of demonstratives, unlike the direct reference theory, the relevant property will not be the semantic content of a demonstrative phrase used by the subject. Demonstrative Availability is, as stated, intentionally neutral between these two different treatments of demonstratives. For a subject to be able to demonstrate a property might be for the subject to be able to use a demonstrative device of direct reference to refer to that property; but it might, equally, be for the subject to be able to use a complex demonstrative device of quantification whose contextually determined restrictor includes that property as a constituent. Either is enough to satisfy the Availability Requirement. Let’s turn now to argument [2] from Chapter 11, which, like [1], is also best viewed as an argument, not against Nonconceptualism, but against the Availability Requirement. The most popular form of this argument is best introduced by an example. Suppose that I am looking at the color of my shirt sleeve, and say ‘That is that color,’ while demonstrating the sleeve. It seems that what I say is (presuming that my shirt sleeve exists, and has a color) linguistically guaranteed to be true: in this sort of context, ‘that color’ just automatically refers to the color, whatever it is, of the object demonstrated. Let’s suppose that this color is blue34. But now suppose, as is surely possible, that I’m in fact having an illusory experience of the color of the sleeve. Then my experience is representing the color of the sleeve as different than it actually is—as some color other than blue34—which means that I have not, after all, succeeded in demonstrating the color my experience represents the sleeve as having. Rather, the objection goes, demonstratives always pick out the property really instantiated by the relevant object in the world. But then I am not able to demonstrate the properties represented as in my environment in illusory experiences, and Demonstrative Availability is false. While the description of the above case is quite plausible, to conclude from it that demonstratives never pick out the property represented by the relevant experience in cases of illusion is to generalize much too quickly. In fact, this should have been clear from the initial argument in favor of Demonstrative Availability: if I can intelligibly wonder in some cases whether the sleeve is that color, then I must be able to use demonstratives to demonstrate properties represented as being in my environment, without committing myself to the claim that they are actually instantiated by objects in that environment. The obvious thought is that we just have here two ways of using demonstratives, which are, plausibly, distinguished by the intentions of the speaker—in the default case perhaps I intend to demonstrate the real color of the sleeve, but

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

demonstratives and the availability requirement 129 I can also, if I wish, easily single out for demonstrative thought the color my experience presents the sleeve as having.4 Innocent as this thought seems, many have thought that it is inconsistent with Conceptualism, which I endorsed in Chapter 11. That is the intended conclusion of the following very influential argument from Richard Heck: Surely I can form a concept of the color my desk appears to me to have. Is that not the demonstrative concept [the Conceptualist] needs? Yes and no. Yes, in the sense that I am inclined to believe that we can have demonstrative concepts of this kind. But no, because [the Conceptualist] cannot respond to my argument in this way. . . . To allow that my concept that color might not denote the color my desk actually has, but the one it appears to me to have, is obviously to deny that its reference is fixed by a ‘sample.’ And that claim is not one [the Conceptualist] can easily abandon, for what alternative conception of how the concept’s reference is fixed is available to him? The most obvious alternative would be to say that its reference is fixed by the content of my perceptual experience. But to say that would be to appeal to a level of perceptual representation [the Conceptualist] does not want: If the content of my perceptual experience is to fix the content of my demonstrative concept of the color experience presents to me, my concept of that color cannot also be part of the content of that experience. If it were, the content of the demonstrative concept would be fixed by the content of that same concept.5

Heck is surely correct that the reference of ‘that color,’ uttered with the intention to refer to the color my desk appears to have, is fixed by the property I am visually representing the desk as having. But the idea that this is somehow in conflict with Conceptualism seems to me to rest on a conflation of two uses of the term ‘concept.’ Heck’s argument turns on the following three claims, which jointly entail the falsity of Conceptualism: (a) Some demonstrative concepts must have their reference fixed by the contents of the subject’s experience. (b) If Conceptualism is true, then demonstrative concepts are parts of the contents of experiences. (c) If a demonstrative concept is part of the contents of a subject’s experiences, then the reference of that very concept is not fixed by the contents of the subject’s experience. 4

See Bengson, Grube, and Korman (2011). Heck (2000), 495–6. In the above I replace references to John McDowell with the generic label ‘the Conceptualist’ for conformity with the terminology used so far. Heck’s argument, I take it, is not an ad hominem directed at McDowell, but rather an argument aimed at anyone who (like me) tries to combine McDowell’s endorsement of Conceptualism with the view that we can form demonstrative concepts of the colors things visually appear to have, but don’t really have. Arguments related to Heck’s against the Availability Requirement are presented in Tye (2006) and in Roskies (2008, 2010). I discuss these in Speaks (forthcoming-c). 5

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

130 demonstratives and the availability requirement I agree that there is an interpretation of ‘concept’ that makes (a) true, and am willing to grant for purposes of argument that there is an interpretation that makes (b) true. What I doubt is that there is any interpretation of ‘concept’ that makes (a) and (b) both true. As is well-known, ‘concept’ is used in two quite different ways in the philosophy of mind: it is used to stand both for mental representations—the mental analogue of linguistic expressions, a kind of internal bearer of content—and for the contents of those mental representations. The problem is that neither interpretation makes both (a) and (b) true. For suppose we interpret ‘concept’ in Heck’s argument as meaning ‘mental representation.’ Then (a) looks reasonably plausible. But (b) looks extremely implausible. Saying that mental representations are parts of the contents of experience is a bit like saying that linguistic expressions are parts of the propositions expressed by the sentences in which they figure. In general, this simply won’t be the case; the proposition expressed by ‘Grass is green’ is about grass and greenness, and not about ‘grass’ and ‘green.’ Certainly it is no part of Conceptualism to deny this triviality. So suppose we interpret ‘concept’ as meaning ‘content of a mental representation.’ Then, on the simplest Russellian view of the content of experience, the relevant concept will just be a color property—hence, plausibly, will just be the reference of the relevant demonstrative representation, and not something which, as (a) says, must have its reference fixed.6 To be sure, Heck had in mind a Conceptualist view which took a Fregean, rather than a Russellian, view of the contents of experience and thought. But the same problem for Heck’s argument emerges if we take this view of content. Suppose that, with the Fregean, we take the contents of mental representations to be modes of presentation of objects and properties. Modes of presentation just are not the sorts of things which ‘need their reference fixed.’ As understood by Frege, modes of presentation at least determine conditions of satisfaction, and thereby determine a reference—which is the thing, whatever it is, that meets the relevant conditions of satisfaction. We don’t need to give modes of presentation conditions of satisfaction—they have these conditions of satisfaction essentially. And we don’t need to determine what meets those conditions of satisfaction— that is fixed by what the world (or, more generally, the circumstance of 6

On the more complex, but still Russellian, view of demonstratives as devices of quantification, the content of the demonstrative will not be a constituent of the content of the relevant experience— even though the same color property will be a constituent of both the content of the demonstrative and of the relevant experience. This is why earlier I was only willing to grant for purposes of argument that there is an interpretation of ‘demonstrative concept’ that makes (b) true.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

demonstratives and the availability requirement 131 evaluation) is like. So we don’t, contra (a), need to explain how the reference of a mode of presentation is fixed.7 Since there is no interpretation of ‘concept’ on which demonstrative concepts are both things which need to have their reference fixed by the relevant experience, and are things which are parts of the contents of that experience, the claim that we can demonstrate sensed but uninstantiated properties does not entail Nonconceptualism. So the Conceptualist need not adopt any special views to escape Heck’s argument; that argument fails for reasons independent of any particular view of sensing properties.8 Here I part company with fellow Conceptualists, like Bengson, Grube, and Korman, who take Heck’s argument to be one of the most serious challenges to their position. Partly in response to this challenge, they posit nonconceptual states of property awareness that are independent of and irreducible to experiential representational states. They take themselves to be forced to posit such states because, as they put it, The content of E is meant to be that o is thus. The concept thus obviously cannot fix its own reference, nor is there any other constituent of the content of E that can fix its reference.9

7

I don’t think that this depends on any contentious theory of modes of presentation. We could, to pick an example of a less traditional Fregean view, adopt the neo-Fregean semantics of epistemic two-dimensionalism developed in Chalmers (2006) and discussed in Chapter 13, and think of modes of presentation as epistemic intensions. As with Fregean senses, once we have an epistemic intension of some expression in hand, we don’t need to do anything else to ‘fix its reference’—its reference is fixed by the idealized a priori connections that hold between the canonical description of the actual world and the application of that expression. 8 But even if it is true that uses of demonstratives to pick out perceptually represented colors which are not properties of the ‘sample’ pose no particular problem for the conjunction of Conceptualism with the Availability Requirement, related cases can be used to present a puzzle for Availability/Difference. Suppose that I am having an illusory experience of a brown desk as green. Even if I can use ‘that color’ to pick out the shade I represent the desk as instantiating, it seems that I can also use ‘that color’ to pick out the color the desk actually has—to use ‘that color,’ that is, in a way which makes ‘That desk is that color’ come out true. Now consider a different experience which is like the first one with respect to content, but is veridical. It seems like ‘that color’—when used in this second, sample-directed way—will differ in content between the two experiences. In the first, it will pick out a shade of brown; in the second, it will pick out a shade of green. Doesn’t this mean that the two experiences, despite having the same content, make different contents available for thought? I’m inclined to think that it does not, because I’m inclined to think that ‘that color’— used in a way in which it is intended to pick out the color of the table, whether or not that is the color perceptually represented to the subject—is best treated as a complex device of quantification whose content does not include the relevant color property as a constituent of the restrictor. But my views here are tentative, and I’m not sure what the right thing to say about this sort of case is. 9 Bengson, Grube, and Korman (2011), 175.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

132 demonstratives and the availability requirement Here, as in Heck’s argument, we have concepts as things which both are constituents of the contents of experience and things which need to have their reference fixed. Modes of presentation satisfy the first condition but not the second, and mental representations satisfy the second condition but not the first. Nothing, as far as I can see, satisfies both. In general—contra influential arguments to the contrary—there is no inconsistency in conjoining Demonstrative Availability with Conceptualism, and holding that we acquire demonstrative concepts of, for example, certain colors from experience. If ‘acquisition of demonstrative concepts’ is just acquisition of an ability to have thoughts involving the contents of the relevant demonstratives, then this is just a matter of a subject standing in a sensing relation to some content in which some color property figures, and this experience giving the subject the ability to have thoughts, judgements, etc. whose contents also involve that property. The situation is in fact just the opposite of what the present line of objection suggests: Conceptualism makes it much easier to see how demonstrative phrases and thoughts could inherit their contents from sensing properties. If the two are relations to the same sorts of things, then this is just a matter of a subject’s coming to stand in one relation to a content explaining her having the ability to stand in another relation to that same content.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

18 The Perceptual Representation of Objects In Chapter 12, I argued that the contents of perceptual experiences can be no more fine-grained than Russellian contents. But this leaves open the question of exactly which Russellian contents we can stand in sensing relations to. Do they, for example, ever contain objects as constituents, or are they always purely general propositions whose constituents include only properties and relations? And what sorts of properties and relations can they contain? The Availability principles defended in Chapters 16 and 17 give us the resources to make some progress on these questions. Let’s address the first of these questions first. Consider a pair of experiences of a pair of subjects, A and B. The two experiences are indiscriminable, and each is a visual experience of an unlabeled white golf ball against a green background. We can argue, using Availability/Difference, that A and B differ with respect to the contents they visually sense. Let’s call the golf balls involved in the experiences, respectively, ball-A and ball-B. We can presume that, prior to their experiences, neither A nor B were able to have de re thoughts about either ball. But things change with their visual experiences. During and after his experience of ball-A, A is able to have de re thoughts about ball-A. He can, for instance, judge of that ball that it is dimpled. Of course, he’s in no position to do this of ball-B; he couldn’t have de re thoughts about ball-B before his experience of ball-A, and it is hard to see how his experience of ball-A could have helped. And since the situations are symmetrical, B seems to be in just the opposite position with respect to the two golf balls. But then it follows from Availability/Difference that the two subjects are visually sensing different contents: holding fixed the relevant facts about the two subjects, the two experiences nonetheless make available different de re thoughts to them.1 1 As earlier, this use of facts about which thoughts are made available by experiences to determine their contents runs parallel to Johnston’s discussion of the objects of experience in Johnston (2004).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

134 the perceptual representation of objects How do the sensed contents differ? It’s hard to see how there could be any difference in the properties of the golf ball, the background, or the relations between the two sensed by the two subjects. So it is plausible that if the two subjects differ in the contents they are visually sensing, they must differ in something other than the properties they represent as in their environments. But, once we get this far, it is hard to deny that the golf balls themselves are constituents of the contents of the relevant experiences. After all, how else could these contents differ, other than one having ball-A as a constituent and the other having ball-B as a constituent? The most plausible way to resist this conclusion seems to be to reject Availability/Difference. And the example of the golf balls might seem to suggest a way to do this: one might think that which thoughts a given experience makes available for thought to a subject depends not just on the contents sensed by the subject of the experience, but also on the causes of the experience. So perhaps, to use our current example, the experiences of our two golf balls don’t differ in content; the crucial difference between them, which explains the differences in the availability of our two de re thoughts, is that in the first case ball-A is among the causes of the experience, whereas in the case of the second experience, ball-B is among the causes. The obvious challenge for a view of this sort is to explain exactly which causal relation is such that o’s bearing that causal relation to an experience is sufficient for that experience to make available de re thoughts about o. It obviously is not enough for o to be one among the causes of the experience. If Bob is the one who put ball-A there, then he’s among the causes of the first experience, but that plainly need not put A in a position to have de re thoughts about Bob. Similarly for an individual molecule which is a constituent of golf ball A. So our question is: what special causal relation is such that an object’s standing in that relation to an experience is sufficient to give the subject of that experience the ability to have de re thoughts about that object? The problems to which attempts to meet this challenge lead are familiar from discussions of the causal theory of perception. One might suggest that we should require not just that o is among the causes of the relevant experience, but also that some of o’s properties are represented in that experience. Since neither Bob nor the molecules which compose ball-A are white, round, and dimpled, this looks promising. But of course another golf ball—ball-C—could have been among the causes of the experience. (Maybe it bumped ball-A into place.) But this would not put A in a position to have de re thoughts about ball-C. One might, instead, suggest that we require the golf ball to cause the experience ‘in the normal way’ that objects cause experiences of them—whatever that is. But

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the perceptual representation of objects 135 this is too strong, since we might have abnormal, but genuinely veridical, experiences of objects which put us in a position to have thoughts about those objects.2 A different, but related, suggestion would be that o must be among the immediate causes of the experience, as ball-A is, rather than merely helping to bring about the experience, as ball-C does. The distinction, though, is pretty obscure—is a golf ball seen through a pane of glass an immediate cause of the experience of the ball? Moreover, lots of things can be among the immediate causes of an experience and yet not be thereby available to the thought of the subject of the experience—like photons. Instead we might require not just that o’s properties are represented, but that they are represented as properties of o. This would do the trick—but at the cost of re-introducing objects as parts of the contents of experience, which was the thing that the detour through causes of experiences was meant to avoid. Maybe there’s some way around these problems. But they make me think that Availability/Difference is pretty secure, and hence that the present argument in favor of objects being parts of the contents of experience is a difficult one to resist.3 . . . However, the view that we bear sensing relations to Russellian propositions with objects as constituents leads to at least three apparent problems. The first is that it opens the door to an immediate argument for externalism about sensing properties, which to many has seemed to be in conflict with the conjunction of the intentionalist theses already defended and a plausiblesounding internalist thesis about phenomenal properties. The argument for externalism, given the foregoing, is pretty hard to resist: consider again two experiences, which differ only in that one is of an object x, and the other is of an indiscriminable object y. The first puts the subject of perception in a position to have de re thoughts about x, but not y; the second puts the subject of the perception in a position to have de re thoughts about y, but not x. So, by the Availability/

2

See Noë (2003). The causal theorist might reply with a style of argument which descriptivists often use against causal theorists of reference: they might just say that the proponent of the idea that objects are parts of the contents of experience owes some account of the facts in virtue of which one experience includes o1 as part of its content whereas the other includes o2. Call these the ‘constitutive facts.’ Given any account of the constitutive facts, the causal theorist can simply piggyback on this account and explain the difference in availability in terms of the constitutive facts, without thinking of the object as part of the content of the relevant experience. But there are two problems with this. The first is that there simply may be no non-content-involving list of constitutive facts which does the job. The second is that if constitutive facts can be provided, they will presumably be derived from some more general reductive analysis of representation—but then they will support, rather than undercut, the claim that the relevant objects are parts of the contents of the experiences. 3

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

136 the perceptual representation of objects Difference Principle, the two subjects are instantiating distinct sensing properties. But, given that x and y are indiscriminable, it seems clear that this difference in sensing properties need not show up as a difference in the intrinsic properties of the subjects of the two experiences.4 The second stems from the fact that, in the example of the golf balls, A and B seem to instantiate indiscriminable phenomenal properties while sensing different contents. But, if we assume that indiscriminable phenomenal properties are identical, this example suggests that, even if phenomenal properties supervene on certain representational properties, the latter do not supervene on the former—the experiences of A and B seem to provide a counterexample to any such supervenience thesis. And this raises the worry that the truth of our intentionalist supervenience theses cannot be explained by the identity of the relevant representational and phenomenal properties. I’ll return to these worries in Chapter 28. Both will be easier to address once we have a fuller picture of the constituents of sensed contents (Chapters 20–23) and of sensing relations (Chapters 24–27). A third worry is more immediately pressing: this is the worry that Availability/Difference must be false, on the grounds that a thesis which it entails—that contents can have contingently existing objects, like golf balls, as constituents—leads immediately to absurdity.

4 It is worth noting that this argument for externalism does not rely on Russellianism, Millianism, or any contentious views about content. It relies only on the Availability/Difference Principle.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

19 The Problem of Contingently Existing Contents The problems here stem from the fact that propositions can’t exist unless their constituents do; and this, plus the conclusion of the preceding chapter and the assumption that some things exist only contingently, entails that some propositions exist only contingently. One might point out, correctly, that the claim that propositions can’t exist without their actual constituents doesn’t obviously follow from anything I’ve said about propositional constituency—why assume, for example, that propositions have their constituents essentially? A plausible argument for this claim has been given by Timothy Williamson.1 The argument is that singular propositions, like the proposition that a particular golf ball is white, are essentially about the objects which are their constituents. It’s just hard to see how that proposition could exist, and still be that proposition, without being about that golf ball. But it also seems quite plausible that nothing can stand in a relation, like that expressed by ‘is about,’ without existing; and from these two claims it follows that it is a necessary truth that if the proposition that that golf ball exists, the golf ball does too. Why might this be a worrisome consequence of the views defended so far? The central problem can be summed up by presenting the following reductio, due to Alvin Plantinga, of the view that propositions can’t exist unless their constituents do.2 Following Plantinga, I’ll call that view ‘Existentialism’:

1

See Williamson (2002). See Plantinga (1983). I’ve modified the argument by (following David (2009)) combining two of Plantinga’s premises into premise (7). Related arguments are also discussed in Prior (1969) and in Williamson (2002). 2

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

138 the problem of contingently existing contents 1 Necessarily, if the proposition that Socrates does not exist is true, then the proposition that Socrates does not exist exists. 2 Necessarily, if the proposition that Socrates does not exist exists, then Socrates exists. 3 Necessarily, if the proposition that Socrates does not exist is true, then Socrates exists. 4 Necessarily, if the proposition that Socrates does not exist is true, then Socrates does not exist. 5 Necessarily, if the proposition that Socrates does not exist is true, then Socrates exists and Socrates does not exist. 6 Possibly, Socrates does not exist. 7 If possibly Socrates does not exist, the proposition that Socrates does not exist is possibly true. 8 The proposition that Socrates does not exist is possibly true.

Serious Actualism Existentialism

6,7

C Possibly, Socrates exists and Socrates does not exist.

5,8

1,2

3,4 Contingency

The argument is a reductio—but of which premise? Plantinga took the argument to be a reductio of (2), Existentialism; and it is not hard to see why. Premises (4) and (7) look trivial on a first reading, and, of the three named premises, Existentialism looks, at first glance, like the easiest to give up. For many philosophers—including me—the theses that not everything exists necessarily and that if a property is instantiated then there is something which instantiates it are among their core metaphysical convictions. One might think that, in the present context, we could just dodge this problem. After all, the contents to which we stand in sensing relations are not propositions, and hence are not the sorts of things which are true or false at worlds. So why worry about problems which essentially involve those topics? The answer is implicit in the discussion of different sorts of attitudes toward contents in Chapter 15. Even if sensing properties involve a certain sort of selfascription of a property which is not a proposition, they give rise to propositional attitudes, and in particular to beliefs. The process by which this occurs involves ‘filling in’ unfilled argument places in the property with individuals, not the subtraction of individuals from the content. So it’s pretty hard to avoid the view that if a particular golf ball can be a constituent of a sensed content, it can also be a constituent of a proposition which could be the content of a belief, or the semantic content of a sentence. Indeed, the problem is particularly pressing in the present context. That’s because the usual ways of avoiding Existentialism altogether let the proposition that Socrates does not exist have as a constituent, not Socrates himself, but, instead, either some Fregean sense which is a mode of presentation of Socrates, or

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the problem of contingently existing contents 139 some description involving a haecceity of Socrates. But Chapter 12 has already given us an argument against the former view, and, whatever their uses elsewhere, the use of haecceities in the philosophy of perception seems singularly inapt. When I look at a golf ball, I just don’t perceptually represent a haecceity of that golf ball as instantiated.3 In my view, the right response to this reductio is to reject premise (7): a proposition can be true at a possible world without being such that, were that world actual, it would be true. I’ll defend this response in two steps. First, I’ll present a dilemma for a proponent of the above reductio which is designed to show that, if (7) is true, then (6)—the thesis that some things exist only contingently—must be false. Either way—whether (7) is false, or (6) is—the reductio is blocked. Second, I’ll explain what it means for a proposition to be true at a world in such a way that it does not follow from a proposition’s being true at w that, were w actual, that proposition would be true. The dilemma begins with the following formula:4 ðEÞ 9x

 x does not exist:

If (E) is false, then Contingency (premise 6 in the argument) is false, and the argument against Existentialism is defused. So we can assume that this formula is true. If it is true, then the open sentence x does not exist must be true at some world relative to some assignment of a value to ‘x.’ That is, the proposition expressed by this open sentence relative to some assignment must be possible. If premise (7) is true, it follows that it is also possibly true; that is, it follows that there is some world w such that, were w actual, then the proposition expressed by x does not exist relative to some assignment of a value to the variable would be true. But there could be no such world, since it is impossible that something be the value of a variable without existing. Hence either (6) or (7) must be false.

3

This might not seem so obvious if we think of haecceities not as exotic primitive individual essences, but simply as world-indexed versions of more familiar properties. But haecceities so construed are no help in the present context, since the problem just is that there are no differences in the representation of such ‘more familiar properties’ between the two golf ball experiences described in the previous chapter. 4 The idea behind this dilemma came from a talk by Timothy Williamson, though he shouldn’t be held responsible for the formulation of the argument which follows.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

140 the problem of contingently existing contents Another way to put much the same point is that it seems that the proponent of (7) must endorse the following formulation of the Converse Barcan Formula: 9x

 Fx !  9x Fx

For, if the converse Barcan formula were false, it would seem that there could be some proposition—one expressed by ‘Fx’ relative to an assignment of a value to the variable—which was true with respect to some world without it being the case that any world is such that, were that world actual, this proposition would be true—and to admit this possibility is just to admit the possibility of a gap between a proposition’s being true at a world and that proposition’s being such that, were the world actual, that proposition would have the property of being true. But if the Converse Barcan Formula is true, then the dilemma above re-emerges. For (E) plus the converse Barcan formula implies that

 9x x does not exist: which certainly seems false. So (E) must (given (7)) be false. And in this case, again, premise (6) of Plantinga’s reductio is false, and the argument against Existentialism is blocked. This argument, if successful, blocks the reductio of Existentialism; but it does not tell us whether it is (6) or (7) that we should reject. Since (6) seems to me quite plausible, I suggest that, absent other arguments, the best course of action is to hold (6) and reject (7). But taking this line requires us to give some account of what it means for a proposition to be true at a world. Proponents of (7), after all, typically think that (7) is true because it is entailed by the following simple and attractive theory of truth at a world: p is true at w iff were w actual, p would be true If we are to reject (7), we need to explain what truth at a world could be, if it is not this—and do so in such a way as to make sense of the claim that the proposition that Socrates does not exist could be true at w even if, were w actual, that proposition would not exist. Here’s one way to do that, in steps: (i) Let’s begin by assuming the view of propositions as properties sketched earlier, according to which the proposition that Amelia talks is the property of being such that Amelia talks. For such a property to be true is for it to be instantiated. (ii) We associate with each such property another property, which we can call that proposition’s truth condition. Our aim is that, for any

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the problem of contingently existing contents 141

(iii)

(iv)

(v)

(vi)

(vii)

proposition p, we want p’s truth condition F to be such that a possible world w instantiates F iff p is true at w. It should be uncontroversial that propositions have truth conditions, in this sense. Note that since we are assuming actualism, we are assuming that possible worlds (like everything else) actually exist and actually have properties. When we talk about worlds instantiating truth conditions, we are talking about how possible worlds actually are—not about how they would be under certain other conditions. Then what we need to reject (7) is a view of truth conditions which is such that the truth condition for the proposition that Socrates does not exist is such that, if it is actually instantiated by w, then, were w actual, the proposition that Socrates does not exist would not exist. We then let the truth condition for the proposition that Socrates does not exist be the following property: the property of being such that, were w actual, Socrates would not exist. The Existentialist and anti-Existentialist alike should agree that this is a truth condition for the proposition that Socrates does not exist; that is, that a world w instantiates this property iff the proposition that Socrates does not exist is true at that world. But from the fact that this property is a truth condition for the proposition that Socrates does not exist, plus Existentialism, it follows that (7) is false. If a world instantiates the property of being such that, were it actual, Socrates would not exist, then were it actual, well, Socrates would not exist. And from this plus Existentialism it follows that the proposition that Socrates does not exist wouldn’t either. And this is enough to falsify premise (7).

Proponents of Plantinga’s reductio often respond to views which reject (7) with incomprehension. Often, such views rely on a distinction between truth in a world and truth at a world; and many have not been convinced that this distinction has been sufficiently explained by its proponents. But the above sketch of an account of truth at a world relies on no such distinction.5 It does rely on a distinction (at step (iv)) between the properties which a possible world instantiates and the properties which would exist if that world were actual; but it can hardly be denied (at least by anyone willing to talk about properties and worlds) that this distinction makes sense. 5 This is true even if the distinction can be reconstructed in the above terms. A proposition p will be true at w iff w instantiates p’s truth condition. It will be true in w iff were w actual, p would have the property of being true—that is, if w instantiates p’s truth condition and, were w actual, p would exist.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

142 the problem of contingently existing contents Where should the anti-Existentialist object? They might say that on their view, this assignment of a truth condition to the proposition that Socrates does not exist does not falsify (7). This is correct but irrelevant; the point is not to provide an argument, independent of Existentialist assumptions, for the claim that (7) is false; it’s just to provide a principled way to resist a reductio of the view. A better objection is that I have not given a general theory of truth conditions, but have just provided a truth condition for the single proposition that Socrates does not exist. This is true. And it’s no trivial matter to do better, for reasons familiar from discussions of deflationary theories of truth; we can’t simply generalize the theory by saying that for any proposition p, p’s truth condition is the property of worlds corresponding to the open sentence were w actual, p since we’d be illegitimately using ‘p’ once in object and once in sentence position. Better to generalize by types of propositions, in something like the following way: If p is an attribution of a monadic property F to o, then the truth condition for p is the following property of worlds: the property of being such that, were the world actual, o would instantiate F. If p is the negation of a proposition q, then the truth condition for p is the following property of worlds: the property of not instantiating q’s truth condition.6 There are of course non-trivial problems providing a fully general and satisfactory account of this type; but it seems to me quite plausible that it should be able to be done. Opponents of Existentialism will likely see this view of truth at a world as excessively complex. Wouldn’t it be easier if we weren’t Existentialists, and simply said that, just as a proposition is true iff it is instantiated, it is true at w iff were w actual, it would be instantiated? I agree; it would be easier. But things aren’t always easy, and I think that reflection on the case of perceptual experience shows that it is difficult to see how any of the usual anti-Existentialist views of singular contents could be correct. If the foregoing is right, there is no incoherence in thinking of contingently existing objects as among the constituents of sensed contents, even if this entails that those contents exist only contingently. Let’s now turn to the question of which properties can be among the constituents of those contents. 6 This isn’t the only possible treatment of negation in this framework; for brief discussion of some other options, see Speaks (2012).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

20 The Perceptual Representation of Natural Kinds Availability/Difference shows us that objects can be among the constituents of sensed contents; does it show us anything about which properties can so figure? One topic of recent interest has been the question of whether perceptual experiences can represent natural kind properties, like the properties of being water or being a pine tree. Susanna Siegel has done more than anyone to defend the view that they can; and her principal argument for this conclusion has employed what she calls ‘the method of phenomenal contrast.’1 The idea behind this method is that we can mount a plausible argument for the claim that some type of property is a constituent of sensed contents if we can come up with a pair of experiences which differ phenomenally from each other, where this difference is best explained by a difference in the representation of some property of the type in question. Siegel suggests as an example two experiences of some pine trees, before and after acquiring an ability to recognize pine trees. After acquisition of this ability, the pine trees just seem to ‘stand out’; they become visually salient to you in a way that they weren’t before. This is clearly a phenomenal difference; and, if this sort of phenomenal difference must be explained by a difference in the representation of some property or other, the hypothesis that it is best explained by the fact that the second experience but not the first represents the property of being a pine tree seems quite plausible. Though I think that arguments of this sort have considerable force, I think that the inference from a phenomenal difference to a difference in properties represented can be resisted. This is not because there can be differences in content which entail no differences in which objects and properties are represented—as Fregeans would say—but because the phenomenal differences might also be

1

See Siegel (2006, 2010).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

144 the perceptual representation of natural kinds explained by a difference in the representational relation rather than the content represented.2 The issue here is related to the distinction between intermodal and intramodal intentionalisms, and the question of which differences in relations to contents can make a difference to a subject’s phenomenal properties. In Chapter 27, we’ll see that a reasonable case can be made for the claim that some phenomenal differences are explained by attentional shifts which involve no difference in which objects and properties are sensed as in the subject’s environment. If this is right, then we might worry that the phenomenal contrasts to which Siegel draws attention are best explained by attentional shifts of this sort. Perhaps both experiences represent the shape properties of pine trees, but certain such shape properties become the objects of the subject’s attention after acquisition of the recognitional disposition. (Maybe acquisition of a recognitional disposition of this sort just is acquisition of an ability to attend to certain perceptually represented properties, or to similarity relations between such properties.) However, my aim here is less to evaluate the merits of Siegel’s case for the conclusion that kind properties are represented in experience than to suggest a different sort of argument for the same conclusion. The argument is quite similar to the argument of Chapter 18; indeed, it’s exactly parallel but for the fact that we substitute indiscernible natural kinds for indiscernible objects. Consider, for example, the pair of subjects imagined in Putnam’s famous ‘twin earth’ thought experiment, one of whom is looking at an actual sample of water, and the other of which is in a waterless environment in which a chemically distinct colorless odorless liquid—twater—fills the lakes and rivers. It seems clear that the actual subject is, during and after his experience, in a position to have de re thoughts of the kind water. During the experience he can demonstrate the kind by using a demonstrative phrase like ‘that stuff ’ or a name like ‘water’; and he might entertain various hypotheses about that kind. He might, for example, come to believe of the kind that samples of it are composed primarily of H2O molecules. The situation of our twin-earthling is quite similar, except that his de re thoughts are about the kind twater rather than the kind water. He might also come to believe of twater that samples of it are composed primarily of H2O molecules—but his belief, unlike that of his earthling counterpart, would be false— and presumably necessarily so. Hence the two thoughts are not the same.

2 This would be an objection to Premise (1) of the argument of Siegel (2006). She considers various objections of this sort to her argument, and convincingly argues against each; I don’t think, however, that those objections rule out the sort of attention-based explanation just given. Of course, that explanation might fail for other reasons.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the perceptual representation of natural kinds 145 And of course neither need be able to entertain the thoughts of the other. Our earthling might well come to wonder whether there are other natural kinds which are perceptually indiscriminable from water, but this general thought would not be a way to have de re thoughts about twater, as opposed to the (presumably many) possible kinds which are possibly perceptually indiscriminable from water. Hence, by Availability/Difference, it follows that our two subjects stand in sensing relations to different contents. But—as in the example of the golf balls—it is hard to see what the relevant difference in content could be, if not the fact that water is a constituent of the first whereas the distinct kind twater is a constituent of the second. Given that the two experiences could be indiscriminable from the point of view of the subject—water and twater, after all, look just the same, and each could be viewing the relevant liquids in just the same conditions—it is hard to see what properties, other than these distinct natural kinds, one subject but not the other might be sensing as instantiated. Despite the close parallels between the above argument and the argument of Chapter 18, the view that experiences represent kind properties is not nearly so popular as the view that they can have singular contents with objects as constituents. I’m not sure why this is, since it’s hard to imagine an argument against one which would not be equally good as an argument against the other. One objection against the idea that we perceptually represent kinds begins with the observation that, just as there can be distinct but indiscriminable natural kinds, objects of one kind can look quite different from one another. To get such an example, we need look no further than the example of water, which looks quite different at low temperatures (as ice) than it does at high temperatures (as steam). If we perceptually represent the kind water when presented with it in liquid form, then presumably we do the same when presented with it in solid and gaseous form. But now imagine having a visual experience of some ice, followed by a visual experience of some steam. If we do perceptually represent kind properties, then each experience will represent the kind water as before us. Hence one would think that it would be evident, on the basis of our perceptual experiences, that both the ice and the steam are of the same kind. But obviously nothing of the sort is evident just on the basis of the perceptual experiences—children are surprised to learn that, for example, snow is water.3 3

Alex Byrne (2009) discusses a different example of the same type, but for a slightly different purpose: his aim is to block Siegel’s ‘phenomenal contrast’ argument for the perceptual representations of kinds, rather than to show directly that kinds are not perceptually represented. However, one of his objections seems to rest on the assumption that if one perceptually represents one object as of kind K and also represents another as of kind K, one will be in a position to infer the conclusion

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

146 the perceptual representation of natural kinds If this argument works, it also shows that we don’t perceptually represent objects. After all, we can imagine re-presenting ball-A, now cleverly disguised, to the subject discussed in the preceding chapter. It need not be ‘perceptually evident’ to the subject that she is being re-presented with the same golf ball. Fortunately, this style of argument is not convincing, since it assumes that we have a kind of ability to invariably identify commonalities in the contents of our mental states that we just don’t have. To see this, move from the case of perception to the case of belief. It is presumably uncontroversial that kind properties can be parts of the contents of beliefs. But now consider beliefs formed about the kinds present after the ice-experience and the steam-experience. These will involve the same kind; but this may well not be evident to the subject of the belief. And if this is possible for belief, why not think that it is possible for sensing relations as well?

that both objects are of kind K. This inference, for the reasons given here, does not seem to be in general valid.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

21 Independence, Fallibility, Anti-Circularity So far I’ve used Availability to argue for liberal views about the contents of experience—namely, that they can include objects and natural kinds as constituents. Can this principle also be used to argue for the conservative conclusion that some class of contents don’t figure in the contents of experience? I think so—but only with the help of a few other principles about sensing properties, which I’ll introduce and defend in this chapter. This chapter, and the next two, are in one sense a digression: with one minor exception, none of the arguments in Chapters 24 and following depend on the conclusions argued for here.1 I include them for two reasons. The first reason is that I think that some of the principles defended earlier can, with the addition of some plausible principles defended in this chapter, make some progress in providing a principled basis for deciding between various views about the properties represented in experience. There has been a fair amount of discussion of the question of which properties get represented in experience; but there has been less discussion of the question of which general principles might help us to decide these questions.2 The second reason is that the discussion helps to tie up some loose ends about the right things to say about the cases of spectrum inversion and spectrum shift which played a prominent role in the arguments of Chapters 4–8. The first two principles I want to introduce in this chapter are based on the thought that experience is, even if reliable, neither incorrigible nor omniscient with respect to any class of properties it represents as instantiated:

1 The exception comes in the discussion of the representational properties of subjects who differ only with respect to the focus of their attention in Chapter 27. 2 There are of course exceptions. One prominent example—Siegel’s method of phenomenal contrast—was already mentioned in the last chapter.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

148 independence, fallibility, anti-circularity Independence If a subject can sense F as instantiated, then F can be instantiated without being sensed to be so. Fallibility If a subject can sense objects x, y, . . . as instantiating a property or relation F, then a subject can sense this even if x, y, . . . do not instantiate F. These principles are based on the idea that what we represent in experience are aspects of one’s environment which are there whether or not one is sensing them; hence, for any candidate sensed content, we should be able to imagine experiences with that content in the absence of the sort of state of affairs which would make that experience veridical, and that state of affairs in the absence of that content being sensed. The letter, if not the spirit, of these principles is threatened by certain claims which are necessary if true. If there are properties which are both perceptually represented and necessarily instantiated by every environment, these will be counterexamples to Fallibility; and if there are properties which can’t be instantiated, but can be perceptually represented (like, on some eliminativist views, colors) these will be counterexamples to Independence. I’m ignoring this complication here; to get around it we could add ‘if it is possible for x, y, . . . to fail to instantiate F’ to the consequent of Fallibility and ‘if F can be instantiated at all’ to the consequent of Independence. I think that Fallibility and Independence are independently plausible; this is reinforced by the fact that when we consider examples of properties which, on pretty much any view, are perceptually represented—like length, shape, and texture properties—they all pretty clearly satisfy Independence and Fallibility. This is at least some evidence that the principles are general truths about the properties we sense to be instantiated. Of course, certain sorts of idealists, who think that no properties satisfy the consequents of Independence and Fallibility, won’t be especially impressed by this argument from examples. I’m happy to set these sorts of views to the side, and to concede to their proponents that there’s no easy way to demonstrate that any properties satisfy Independence and Fallibility. My target is really the view that some perceptually represented properties satisfy Fallibility and Independence, but not all. In this chapter, I want to consider two arguments against this view. The first makes use of a certain sort of idealist argument, due to Berkeley, to defend Independence. The second introduces and argues for a third constraint on sensed contents—Anti-Circularity—and shows how that principle can provide some support for both Fallibility and Independence.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

independence, fallibility, anti-circularity 149 The first argument, adapted from Berkeley’s Dialogues, can be stated using the following condition on classes of properties: Necessarily, 8X (x has an A-property iff x has a B-property) Let’s call any classes of properties which satisfy this schema linked. Of course, to say that two classes of properties are linked is not to say that having a particular A-property entails having a particular B-property, or vice versa—it’s just to say that having some A-property or other entails having some B-property or other, and vice versa. Berkeley famously presented an argument that color and shape properties are linked: ‘Can you even separate the ideas of extension and motion, from the ideas of all those qualities which they who make the distinction term secondary? . . . [T]ry if you can [to] frame the idea of any figure, abstracted from all particularities of size, or even from other sensible qualities. . . . Since . . . it is impossible even for the mind to disunite the ideas of extension and motion from all other sensible qualities, doth it not follow, that where the one exist, there necessarily the other exist likewise?’3

One way to read Berkeley here is as pointing out that we cannot visually conceive of an object with shape but not color, and as inferring from this that it is impossible that an object have a shape but not a color. But if this is right, then color and shape properties are linked. And this form of argument seems to have substantial generality. Just as it seems to show that color and shape are linked, reflection on touch seems to show that texture and temperature properties are linked, and reflection on audition seems to show that volume and pitch properties are linked.4 An obvious criticism of this argument is that the inference is invalid; we can’t, in general, infer from our inability to conceive of a situation that the situation is impossible. Some situations are presumably too complicated for us to grasp, but are nonetheless possible. And perhaps synaesthetes who perceptually represent combinations of qualities which we do not would arrive, by Berkeley’s method, at views about which properties are linked which would diverge from ours. So, one might think, even if Berkeley is right that we can’t imagine an experience of an object with a shape but no color, we can’t infer from this that there couldn’t be such an object. But it’s not clear that this reply to Berkeley’s argument is successful. It seems plausible that it is a necessary truth that an object can’t be (at the same time) red 3

Berkeley (1713/1998), First Dialogue, Part III. Though admittedly these categories of properties must be construed rather broadly—broadly enough to allow that one feels the texture of the air when one feels its temperature. 4

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

150 independence, fallibility, anti-circularity and green all over, and that we can know that this is a necessary truth. But how do we know this? Presumably by something like the following method: we try to conceive of an object which is both red and green all over and, in failing to do so, we somehow see that it is impossible for an object to satisfy this condition. The background thought here seems to be that if it were possible for something to be both red and green all over, then this would be the sort of thing of which we could conceive; but we can’t. But this seems to be just the same sort of inference involved in Berkeley’s argument: we’re inferring the impossibility of an object satisfying a certain condition from a certain sort of failure to conceive of an object satisfying that condition. So there is the worry that, if we take Berkeley’s argument to show something about our imaginative capacities rather than shape and color themselves, we will be forced to say that we don’t, after all, have knowledge of color incompatibilities, but only of our inability to imagine certain color combinations. If we want to preserve our knowledge of color incompatibilities—which, admittedly, not everyone does—it’s not easy to see why Berkeley’s argument for the conclusion that color and shape properties are linked is not as good as our argument for the claim that nothing can be red and green all over. So let’s suppose that the A-properties are linked to the B-properties, and suppose we know that every A-property satisfies Independence. Does it follow that the B-properties do too? We know that if the A-properties satisfy Independence, then any A-property can be instantiated without being perceptually represented. But if this is true, then presumably any A-property can be instantiated by an object which is such that none of that object’s properties are perceptually represented. (It would be weird if a perceptually represented property F could only be instantiated by o if o were represented as having some property other than F.) But, since we know that the A- and B-properties are linked, it then follows that at least some of the B-properties must also satisfy Independence. This imposes a strong constraint on any view to the effect that some property which can be the constituent of a sensed content fails to satisfy Independence: it must not be linked with any property which does satisfy Independence. But, if Berkeley’s style of argument is in general an effective one, plausible candidates for such properties will be hard to come by. Let’s turn to the second argument. This begins with a third constraint on sensed contents: Anti-Circularity If a subject can sense F as instantiated, then, if F is a complex property, F cannot be one of its own constituents.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

independence, fallibility, anti-circularity 151 Note that this principle does not say that no property can be one of its own constituents. It seems to me quite plausible that this more general principle is true; and if it is true, it obviously entails Anti-Circularity.5 But we’ll only need to employ the weaker principle, restricted to the contents to which we can stand in sensing relations. Why think that Anti-Circularity is true? Suppose that, for some properties F, G, it is proposed that F ¼ the property of standing in a relation R to the property of being such that something is G Suppose further that G¼H. It then seems to follow that F ¼ the property of standing in a relation R to the property of being such that something is H But just this form of reasoning can be used to argue in favor of Anti-Circularity. Suppose for reductio that Anti-Circularity is false, and that for some perceptually represented property F, F ¼ the property of standing in a relation R to the property of being such that something is F By reasoning analogous to that just employed, it follows from this property identity that F ¼ the property of standing in a relation R to the property of being such that something has the property of standing in a relation R to the property of being such that something is F and by another application of the same reasoning, that F ¼ the property of standing in a relation R to the property of being such that something has the property of standing in a relation R to the property of being such that something has the property of standing in a relation R to the property of being such that something is F and so on ad infinitum. And this seems to show that no property like F could be perceptually represented by beings with anything like our cognitive limitations

5

Johnston (2001), }6 derives something like Anti-Circularity from the more general claim.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

152 independence, fallibility, anti-circularity because, if it could be, any perceptual experience which represented F as instantiated would have an indefinitely complex content.6 (It’s worth noting that this way of defending Anti-Circularity depends essentially upon the view that the contents of experience are Russellian, and hence have as their constituents properties rather than modes of presentation thereof. There’s no such immediate route from the claims that a content contains a mode of presentation of a property and that the property is indefinitely complex to the conclusion that the content itself is indefinitely complex; to the extent that we can make sense of talk about modes of presentation, there’s no obvious reason why complex properties couldn’t have simple modes of presentation.) As mentioned earlier, if we accept Anti-Circularity, this provides us with an argument in favor of Fallibility and Independence. Suppose that Fallibility were false, and that it is a necessary truth that, if a property is represented as instantiated by o, then it is. This surprising fact would cry out for explanation; and the obvious explanation is that what it is for an object to be F is for that object to be perceptually represented as being F. But this would violate Anti-Circularity. Hence a plausible argument leads from the falsity of Fallibility to the falsity of Anti-Circularity—and so also from the truth of the latter to the truth of the former. Parallel remarks apply to Independence. . . . Before moving on, I want to consider one worry one might have about Fallibility, and that is that that principle runs counter to the claim, defended earlier, that the contents to which we stand in sensing relations can have objects among their constituents. Fallibility is, as stated, just a principle about representation of properties; but if we perceptually represent objects as well as properties, we might reasonably expect some analogous principle to hold of representation of objects in experience. But it might seem that no such principle does hold of our perceptual representation of objects. What, after all, makes it the case that subject A senses ball-A, rather than ball-B, as white, round, etc.? Just that ball-A happened to be the ball in front of him. But if this explanation is general, then it might seem that it is impossible to perceptually represent o as thus and so

6 Note that this argument for Anti-Circularity depends only on the classificatory use of ‘constituent’ talk discussed in Chapter 14, since it only makes the relevant impossibility claim about contents to which we can stand in representational relations. The argument depends on the impossibility of our representing certain contents, not on, for example, principles of mereology— the latter sort of argument would obviously rely on a more metaphysically heavy-duty notion of propositional constituency.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

independence, fallibility, anti-circularity 153 without o being before one; and that rules out the sort of misrepresentation which seems as though it should be possible. In reply to this, one might point out that certain sorts of hallucinations are plausibly examples in which one stands in a sensing relation to a content with an object as a constituent, despite the fact that the object is not in one’s environment. After the death of a loved one I might have a hallucinatory experience which represents that person as sitting in his chair, even though there’s no one there— this is enough to show that one can misrepresent by visually sensing the presence of an object in one’s environment that is in fact absent. But there is one respect in which these kinds of hallucinations seem very unlike perceptual misrepresentation of colors and shapes: they seem to depend on prior acquaintance with the object misrepresented as present. This point—emphasized by Mark Johnston in ‘The obscure object of hallucination’—brings out a genuine difference between perceptual representation of objects and of colors. We might put this by saying that colors are, while objects and natural kinds are not, ‘hallucination-available’: one can gain the ability to have de re thoughts about a certain shade in a hallucination, but one can’t gain the corresponding ability to have de re thoughts about objects from a hallucination; rather, the presence of the object in the content of a hallucination always depends on the prior ability to have de re thoughts about the object. I’ll return to the significance of this point for a certain sort of identity claim involving representational and phenomenal properties later, in Chapter 29. Hallucination of a dead loved one thus might seem like a rather special kind of misrepresentation, which involves the thoughts and beliefs of the subject in a more direct way than most sorts of misrepresentations involving sensing properties. One might still wonder whether we can have misrepresentation of objects in a setting which is more similar to standard cases of sensing objects as having properties which they in fact lack. We might try to construct such cases using examples of object constancy. Suppose that I am watching ball-A roll across the ground, and that, thanks to a clever experimental set-up, ball-B is substituted for ball-A in such a way that the experience appears seamless to me—it is, from my point of view, indiscriminable from an experience of a single golf ball rolling at a more or less constant rate across some grass. (Maybe I’ve been given a pill which causes a temporary hallucination while the balls are switched.) It seems plausible that this is an illusion of some sort—I seem to visually represent the ball as the same throughout the series of experiences, and it isn’t the same ball throughout. So this might well seem to be a case in which I mistakenly perceptually represent ball-B as ball-A—that is, a case in which the content I visually sense includes the false identity claim

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

154 independence, fallibility, anti-circularity Ball-B = Ball-A I think that this is fairly plausible; it’s not obvious what other false content could be part of the experience, and it does, as mentioned earlier, seem like there’s some sort of misrepresentation going on here. One might of course say that this misrepresentation is just misrepresentation at the level of judgement rather than at the level of sensing properties. But, against this, the sense that this experience involves some sort of false representation persists even if one is aware of the experimental set-up, and hence has no inclination to judge that the same ball is present throughout. This is at least some indication that, among the sensed contents of the latter portion of the imagined experience is the false identity claim that Ball-B=Ball-A. However, one might worry that, from the point of view of the preceding chapters, there is something puzzling about this view, and that is that this sort of case of misrepresentation seems to be a counterexample to Demonstrative Availability. On the view in question, the content of the experience at the end of the series in which Ball-B is really before me has both Ball-A and Ball-B as constituents. But it’s hard to see how I could use a demonstrative, like ‘that ball’ to refer to the ball which I represent as present but is not really present—in this case, Ball-A. It is hard to summon special uses of demonstratives, as in the color case, which refer to what is represented as in the environment rather than what really is in the environment; the demonstratives seem invariably to refer to the golf ball that’s really there.7 One might then wonder whether these are counterexamples to the Availability Requirement, which Demonstrative Availability was used to defend. But they are not, for the reason mentioned earlier, which is that for objects misperception requires prior acquaintance, which is sufficient to secure availability for thought. Hence even if these are counterexamples to Demonstrative Availability, they are counterexamples which prove the rule—the Availability Requirement—of which Demonstrative Availability is (on this view) a special case.8

7

On the other hand, hallucinations do seem to permit these special uses of demonstratives. It is a bit puzzling that the present sort of case should be different—unless we don’t perceptually represent identities and such cases are impossible. 8 They are, for this reason, not counterexamples to a weakened principle about demonstrative thought, which is that if an object or property is available for thought in virtue of being part of the content of some perceptual experience, then that object of property is available for demonstration. Thanks to Casey O’Callaghan here.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

22 Appearance Properties Let’s apply the principles from the preceding chapters to a particular thesis about perception: the thesis that the contents of experience include appearance properties. Appearance properties are best introduced by thinking about alleged cases of spectrum inversion without misrepresentation. So far we’ve argued against the ideas that there can be variation in visual phenomenal properties which is not also variation in the content visually sensed (that is, phenomenism), and that there can be variation in the content visually sensed which is not also variation in the objects and properties represented by the experience (that is, Fregeanism). But these two ideas together entail that stock examples of spectrum inverted (or spectrum shifted) subjects must differ with respect to their representation of the properties of some object in their environment. And doesn’t this entail, implausibly, that at least one of these subjects must be misperceiving? This last step follows only with the help of two further assumptions: that the relevant difference in the representation of properties is a difference in the representation of color properties; and that standard examples of ‘monochromatic’ surfaces really do have only one color. Someone who denies the first assumption is, in my terms, an appearance property-ist: Appearance property-ism: in addition to color properties, color experiences represent appearance properties; apparent cases of spectrum inversion without misrepresentation are cases in which spectrum inverted subjects perceptually represent the same color properties as instantiated, but differ with respect to their perceptual representation of the appearance properties.1

1 The term, and the view, is due to Sydney Shoemaker. See among other places Shoemaker (1994, 2000). There is, it should be said, a close affinity between appearance property-ism and some views of perception in the recent literature which are called ‘Fregean’; see the discussion in Chapter 10 of two different uses of ‘Fregean.’ Compare, for example, the Fregean view of Chalmers (2006)—on which the contents of color experience are conditions like ‘it must be the property that normally causes phenomenally red experiences (in normal conditions for the perceiver)’—and the classical

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

156 appearance properties The foregoing gives us the materials for three arguments against appearanceproperty-ist views.

Appearance property-ism, phenomenal variance, and Scenarios A and B A first argument is based on the impossibility of Scenarios A and B. Since appearance properties are introduced explicitly to solve spectrum inversion and shift problems, the appearance property-ist must endorse the possibility that arbitrary differences in color phenomenal properties are, while incompatible with sameness of representation of appearance properties, compatible with sameness of representation of color properties. But this just is the thesis of phenomenal variance, which, by the argument of Chapters 4–8, entails the possibility of Scenario A, and (via property variance) the possibility of Scenario B. One response on behalf of the appearance property-ist would be to defend phenomenal variance by, for example, defending the interpersonal constraint, the time constraint, or the memory constraint. But a different sort of response is to say that the argument just given is a bit too easy. After all, an appearance property-ist might say, how can we be so sure that Scenarios A and B really are impossible, once we have appearance properties on the table? Why, for example, are we so sure that psychedelic phenomenology really is inconsistent with sameness of representation of color properties, as opposed to sameness of representation of appearance properties? It might well seem to an appearance property-ist that any argument against her view which simply assumes the impossibility of Scenarios A and B is, in some objectionable sense, begging the question. I can see why it would seem this way. But I have to say that this line of response strikes me as unconvincing. If I add to the description of Scenario A that the subject is also visually sensing some other properties of the wall as rapidly changing, that just doesn’t make it any easier for me to believe that she is visually sensing the color of the wall as constant. And, likewise, the fact that a subject is instantiating the phenomenal property CHARCOAL GREY and visually sensing another property of the wall as constant while visually sensing the color of the wall as rapidly changing seems just as clearly impossible as Scenario B did in the first place.

appearance property-ist view of Shoemaker (2000) that such visual experiences represent ‘a disposition to produce such experiences in creatures with visual systems of one or more sorts’ (467).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

appearance properties 157 A less table-thumping reply is to say that this derivation of the possibilities of Scenarios A and B from appearance property-ism is a way of substantiating a persistent, if somewhat hard to pin down, objection to these views: the charge that these views make colors, in some sense or other, impossible to see, or hidden behind a veil of appearance properties.2 Appearance property-ists might plausibly reply by, first, pointing out that this complaint is as it stands purely metaphorical and, second, saying that there is a perfectly good sense in which we do see colors—appearance properties just provide the ways in which we see them. It seems to me that the present argument against appearance property-ism provides one way of spelling out the intuitive objection. If appearance propertyism really does entail that Scenario B is possible—and hence that it is possible for a subject to, constantly and monotonously, instantiate GREY while visually sensing the color of the relevant object as first red, then blue, then green—then, it seems to me, there is a good sense in which the appearance property-ist does think of the colors as invisible. Indeed, there is a quite specific sense in which color properties turn out to be invisible on this view: it is impossible, on this view, for two subjects who differ representationally only in the way that they visually sense the colors of things to be to differ in their phenomenal properties.3 Colors, in that sense, just don’t show up in the phenomenology. This is a consequence that we should want to avoid. So far I’ve been talking about appearance properties, as they’re usually talked about, in the context of color perception. But we could introduce analogous properties corresponding to the perceptual experience of any sensible quality. Consider, for example, tactile representations of texture. We might get to wondering whether there could be certain sorts of texture-inversion without misrepresentation, and hence introduce properties—texture appearance properties—to serve as the contents of tactile experience, and explain the phenomenal differences between texture-inverts. The above argument rules out not just the appearance properties people have in mind when they’re thinking about color perception, but all appearance properties of this sort—for just the reason that it seems plausible that, as discussed in Chapter 9, we can come up with equally impossible parallels of Scenarios A and B for auditory, tactile, and other modes of sensory representation.

2 3

This objection is pressed in Tye (2000b). So on this view, colors are, in the terminology of Chapter 29, ‘phenomenally silent.’

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

158 appearance properties

Fallibility, Independence, and Availability A second argument against appearance property-ism is based on the constraints on perceptual representation developed in the preceding chapters. The best way to present the argument is to begin by asking: What, exactly, are appearance properties? There are several candidates. The first, and simplest, view of appearance properties identifies them with occurrent appearance properties like x causes S to instantiate RED at t relative to an assignment of the subject of the experience to ‘S’ and the time of the experience to ‘t.’ But this violates Independence—since nothing could have this property without being experienced as such—and a plausible argument can be given that it violates Fallibility as well: Suppose for reductio that it did not. Then I could perceptually represent an object o as instantiating an occurrent appearance property without it doing so. Given the definition of appearance properties, we know that this will involve me instantiating the phenomenal property RED. There are two possibilities: either o is a cause of my experience, or it is not. If it is, then, contra our supposition, o does instantiate the relevant occurrent perceiver-specific appearance property, and my experience is veridical. But if it is not, then my experience doesn’t represent o at all, and hence, contra our supposition, does not represent o as instantiating the relevant occurrent perceiver-specific appearance property. This argument is not airtight, because it does not rule out the possibility of a hallucination of o, which represents o rather than some other object in virtue of a prior acquaintance with o. But it at least shows that misrepresentation of appearance properties construed as occurrent would have to be a pretty special thing. This pushes us toward a dispositional treatment of appearance properties; and then the question is how, exactly, the relevant dispositions should be understood.4 A maximally undemanding construal of the relevant dispositions would identify appearance properties with purely general dispositional appearance properties like the one corresponding to the open sentence x is disposed to cause some subject to instantiate RED in some circumstances

4

See Shoemaker (2000), 466–7 for a brief discussion of some of the options here.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

appearance properties 159 But this delivers some odd results. Presumably, by appropriately varying the perceiver or by varying the conditions of observation, we can show that every object will instantiate this purely general dispositional appearance property. But it seems plausible that the content sensed when I instantiate the phenomenal property RED at a time should be incompatible with the content sensed when I instantiate the phenomenal property GREEN at that time. This is so even if we want the property represented by my instantiating RED to be compatible with the properties represented by an invert’s instantiating GREEN, or even by my later instantiations of this phenomenal property. There are two reasons for this. First, it seems that when I look at a green thing and a red thing side by side, my experience must have a content which is such that, if the experience is veridical, there is some way the first object is that the second object is not. Indeed, if we think about the series of experiences described by Scenario A, it seems quite plausible that for each of those experiences to be veridical, the surface perceived must be changing in some way over the course of the experiences. But on the purely general dispositional view of appearance properties, the object will instantiate all of the relevant appearance properties throughout. Hence the veridicality conditions entailed by the identification of appearance properties with purely general dispositional properties are implausibly weak. Second, if something can simultaneously instantiate both the appearance property associated with RED and the appearance property associated with GREEN, it would seem puzzling that I couldn’t simultaneously visually represent the object as having both of these properties. After all, it is hard to think of another case in which an object can be both F and G, and F and G are properties which I can represent in my perceptual experience, but I can’t perceptually represent an object as both F and G. But purely general dispositional appearance properties would be an exception to this general rule, since I can’t simultaneously visually represent an object as being disposed to cause a subject to instantiate RED and GREEN in some circumstances.5 A better proposal, perhaps, would be to appeal to more restricted dispositional appearance properties: properties which specify the relevant type of perceiver, and relevant type of circumstance, rather than existentially generalizing over types of perceivers and circumstance. On this view, they would be properties like that corresponding to the open sentence

5

This follows the argument in Egan (2006), }4.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

160 appearance properties x is disposed to cause subjects of type S to instantiate RED in circumstances of type C 6 given some specification of the relevant type of perceiver and circumstance. This is, I think, the most promising view of appearance properties. The central question is what the values of ‘C’ and ‘S’ are—of, that is, exactly how restricted the appearance properties which are constituents of the contents to which we stand in sensing relations are. There are three related problems with answering this question in a satisfactory way. The first is a dilemma about what fixes these values on a given occasion; the second is a dilemma about how fine-grained the relevant types should be; and the third, and most fundamental, is the problem that any plausible way of responding to the first two problems leads to conflict with the Availability Requirement. The first dilemma is whether the subject must always be of the represented type S and whether her circumstance must always be of type C. If so, this seems to lead to conflict with Fallibility, since then any RED experience will attribute to the object of the experience the property of being disposed to cause RED experiences in the subject of the experience in the circumstances she is actually in. And of course this is a property that the object of the experience, in virtue of having caused that experience, will have. But the other possibility—that occasionally the subject and her circumstance are not of types S and C—is puzzling. For then it is hard to see what could fix the types of perceiver and circumstance which get into the content of an arbitrary visual experience. A second dilemma concerns how demanding or permissive C and S should be. On the one hand, they must be demanding enough that a subject spectrum shifted or inverted relative to me must not fall under the same subject-type as me—lest appearance properties fail to do the job for which they were introduced. Given the ubiquity of actual cases of spectrum shift—between subjects of different ages, sexes, and races—this will mean that most actual subjects will fall under different types.7 On the other hand, we face some pressure not to slice the types that finely. For consider me now and myself two years from now, at which time my visual system will have changed enough so that these two time slices of myself will be spectrum shifted with respect to each other. It will follow that these two time slices of myself will never have visual experiences with the same content: there will be

6

See Shoemaker (2000).

7

See Block (1999).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

appearance properties 161 distinct types of subjects, S1 and S2, such that the first’s color experiences will all have content involving type S1 and the second’s color experiences will all have content involving type S2. This result is a bit surprising; but it is also inconsistent with the conjunction of the claims (i) that these two time-slices of myself are capable of instantiating the same phenomenal property, and (ii) that phenomenal properties are identical to representational properties. The first is independently plausible— though one could deny it if one is prepared to accept the possibility of distinct but genuinely indiscriminable phenomenal properties. The second is, one might think, strongly suggested by the supervenience of the phenomenal on the representational. We’ve already seen that there is good reason to believe that some intentionalist supervenience theses are true: for any phenomenal property P, there is some representational property R such that, necessarily, for any subject x, if Rx, then Px. We haven’t, yet, identified the representational properties which will give us the strongest true intentionalist supervenience claim; but, even in advance of doing that, we can ask why this relationship between the phenomenal and the representational holds: why is it impossible to change a subject’s phenomenal properties without also changing her representational properties? An appealingly simple answer to this question is: phenomenal properties just are a certain sort of representational property. But this, plus the present view of appearance properties, rules out the possibility that these distinct time slices of me might ever instantiate the same phenomenal properties. These two issues about how to define the values of ‘S’ and ‘C’ show that it is no easy task to figure out what the types of perceivers and circumstances which figure in the contents of visual experiences could be. That fact makes it all the more surprising to be told that we’ve all been perceptually representing these types in our visual experiences all along. Consider, for example, the fact that spectrum shifted subjects will have to represent different subject-types with their experiences. One would think that the question of whether two subjects are spectrum shifted with respect to each other is a highly empirical one, detectable only by hue matching tests and the like. But if the present version of appearance property-ism is correct, this should be detectable simply by the two subjects comparing the perceiver-types which are among the constituents of the contents of their respective color experiences. Indeed, this is not just surprising, but in conflict with the Availability Requirement. That principle requires that if, as the view of appearance properties as properties like restricted dispositional appearance properties says, we represent these types of circumstances and perceivers in our color experience, then these

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

162 appearance properties properties of circumstances and perceivers must be available to the thought of ordinary adult human subjects who have just had the relevant sort of color experience. The problem is that most normal color observers are in no position to specify, or have thoughts about, types of observers and circumstances which meet the constraints just outlined. The conflict is even more stark if we endorse Demonstrative Availability, as the relevant types are simply not there to be demonstrated.8 This is a problem which can be avoided by the appearance property-ist who is also a Fregean, for they can make room for the idea that while appearance properties are restricted dispositional properties, our representation of those properties involves only modes of presentation of them, which themselves make no reference to the relevant type and circumstance. But, given the Russellian view defended in Chapter 12, it is hard to see how this could work. The Russellian would have to say something like this: properties

8

This may be the sort of worry Egan (2006), 508 has in mind when he objects that types of perceivers are the wrong sorts of things to show up in perceptual content. One might think that we could get around this conflict with the Availability Requirement by adopting Egan’s ingenious view of appearance properties as ‘centering features’ which replace reference to types of perceivers and circumstances with indexical reference to the perceiver and (perhaps) circumstance of the relevant experience. This view would fit neatly with the view of sensing properties as involving a kind of selfascription of a property defended in Chapter 15, and is I think an improvement on the more standard appearance property-ist views described earlier with respect to the objections discussed in this section (though it’s on par with more traditional varieties of appearance property-ism when it comes to the other two main objections of this chapter). Centering features have as their extension a set of pairs of objects and subjects—where we can think of the latter as including a specified circumstance and time. Suppose then that we identify an appearance property characteristic of our experiences of red things with a centering feature which has as its extension the set of object/ subject pairs which satisfy x is disposed to cause y to be RED Suppose that I instantiate RED. It seems to follow that I will be veridically representing the relevant centering feature—if an object is causing me to instantiate RED in certain circumstances, then it must be disposed to cause me to instantiate RED in those circumstances. But then centering features will entail the falsity of a principle quite like Fallibility. (It can’t be exactly like Fallibility, since that is a principle about properties, and centering features are not properties.) One might get around this argument by not building in everything about the circumstance of the subject into the value of ‘y’, and instead including some information about the relevant conditions of manifestation in the specification of the appearance property itself. Then we’d be identifying the appearance property with a centering feature which has as its extension the set of object/subject pairs which satisfy something like x is disposed, in conditions C, to cause y to be RED This makes room for misrepresentation of appearance properties, since something could cause me to instantiate RED without being disposed to do so in C. But this leads to the sort of conflict with the Availability Requirement characteristic of restricted dispositional conceptions of appearance properties, since typical subjects needn’t be able to have thoughts about whatever the relevant conditions of manifestation turn out to be.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

appearance properties 163 like being disposed to cause subjects of type S to instantiate RED in circumstances of type C are not strictly speaking constituents of visually sensed contents; rather, these complex properties are part of the analysis of the constituents of visually sensed contents. But this simply does not tell us what properties (in addition to color properties) are parts of the contents of the relevant experiences. We might present this dilemma for the appearance property-ist as follows: A dilemma for appearance property-ism What are appearance properties?

Occurrent properties

Dispositional properties

¬Independence ¬Fallibility (almost)

Purely general or restricted dispositional properties?

Purely general

too weak veridicality conditions + unexplained incompatibilities of experience-types

Restricted

¬Availability Requirement

Anti-Circularity A third, and last, argument against appearance property-ism is based on Anti-Circularity. By themselves, appearance property-ist views don’t violate this constraint. But, with the addition of the claim, mentioned earlier, that the phenomenal not just supervenes on the representational, but is identical to it, we can generate a conflict. Let’s use ‘appearance-red’ as a generic name for the appearance property we represent as instantiated in typical experiences of red things. In this case, then there will be some true property identity roughly of the form RED ¼ the property of bearing the sensing relation to a content of which appearance-red is a constituent

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

164 appearance properties But it’s hard to put this together with any of our accounts of appearance properties. Suppose that for simplicity we pick occurrent perceiver-specific appearance properties. Recall that on that view, appearance-red = the property of causing me now to instantiate RED Put these two together, and we get the following property identity as a result: appearance-red = the property of causing me now to bear the sensing relation to a content of which appearance-red is a constituent which, given that appearance properties are supposed to be the sorts of properties which can be perceptually represented, violates Anti-Circularity. One might resist the above inference on the grounds that it rests on a substitutivity principle which can be rejected.9 From the fact that a subject has an experience with a content which can be specified using a predicate f, and the property expressed by f is identical to the property expressed by some predicate g, does it really follow that the contents of the subject’s experience can also be correctly specified by replacing f with g? One might think not—just as one might think that truth of a belief ascription which involves the name ‘Hesperus’ doesn’t entail the truth of the ascription which replaces ‘Hesperus’ with ‘Phosphorus.’ And if this sort of substitution principle can be rejected, then the first property identity above can be accepted and the second rejected. This is an appealing move for the appearance property-ist—but, again, not one which can be sustained if the positions defended earlier are correct. Chapter 12 defended a Russellian view of sensed contents, according to which the constituents of these contents are objects, properties, and relations. If F is a constituent of a Russellian proposition, and G=F, then G must be a constituent of that proposition too. It’s hard to see how one could deny that without denying Leibniz’s Law. This problem can’t, of course, be blocked by switching from the occurrent conception of appearance properties to any of the other candidate views of appearance properties discussed here. It can be blocked simply by denying that there is any true identity between representational and phenomenal properties.10 See Cohen (2009) }6.4. Cohen is defending a color relativist position rather than appearance property-ism—but, as we’ll see in the next chapter (and as Cohen acknowledges) the views are similar in many respects. 10 It can also be blocked by, as Kriegel ((2009), Ch. 3) suggests, identifying appearance properties with dispositions to cause, not certain phenomenal properties, but certain brain states. One problem here is that, as Kriegel recognizes, one must then either deny the possibility of multiple realizability or take the relevant dispositions to extremely gerrymandered ones, which build in the fact about the brain states which various sorts of creatures would come to be in on the basis of various sorts of experiences. This sort of proposal would also come into stark conflict with Availability, since 9

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

appearance properties 165 I’ll return to this topic later; for now it’s worth noting that, given Anti-Circularity, appearance property-ism at least sharply constrains our view of the relationship between phenomenal and representational properties.

subjects of experience aren’t typically in a position to have thoughts involving the neural property which underwrites their experience—let alone the various other neural properties which possibly underwrite phenomenally identical experiences.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

23 Relativist Views of Sensible Qualities Some properties, like natural kinds and appearance properties, are controversial candidates for inclusion in the contents to which we stand in sensing relations. Others, like colors, aren’t (at least among people who think that experiences have contents). But even if it is uncontroversial that experiences sometimes represent objects as colored, there’s still substantial controversy about what it is for an object to be colored. Just as some of the constraints developed earlier can be used to argue against the view that experiences represent appearance properties, so they can be used to argue against certain views of the nature of sensible qualities, like colors. Earlier I mentioned that the Russellian intentionalist might block the conclusion that at least one of a pair of spectrum inverted subjects must be misperceiving by denying either (i) that the relevant difference in the representation of properties is a difference in the representation of color properties or (ii) that standard examples of apparently monochromatic surfaces really do have only one color. If the appearance property-ist denies (i), the color relativist denies (ii). While this way of thinking about the views can make them seem quite different, they also have important affinities; and that’s because the most plausible relativist construal of colors identifies colors with the one or another of the candidates for appearance properties discussed earlier. This does not mean that the color relativist is open to exactly the same arguments as the appearance property-ist; the color relativist’s view need not conflict with the impossibilities of Scenarios A and B, and hence avoids the first argument of the previous chapter. But it seems that if the second and third arguments of that chapter work against the appearance property-ist, they are just as effective against the color relativist. Indeed, in one respect the color relativist is worst off, because the conflict of color relativism with the Availability Requirement is more stark than in the case of appearance property-ism. Whereas the appearance property-ist can avoid

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

relativist views of sensible qualities 167 conflict with the Availability Requirement by adopting some view of appearance properties other than the restricted dispositional view, it seems as though the color relativist must appeal to these restricted dispositional properties to make sense of the phenomenon of color constancy. If we want to say, as seems plausible, that cases of color constancy involve some property perceptually represented as invariant and some other property perceptually represented as changing, then, even if fine-grained perceiver-specific occurrent properties are good candidates for the latter, it seems that we’ll need to appeal to more coarse-grained properties to play the former role. It’s hard to see what these could be, if not non-perceiver-specific but restricted dispositional properties.1 . . . It might seem like the conclusion of the preceding line of argument leads to an unacceptable conclusion. I identified four ways in which we might make room for the possibility of spectrum inverted subjects, neither of whom is misrepresenting the properties of the relevant surface: phenomenism, Fregeanism, appearance property-ism, and color relativism. But we’ve now found reason to reject all four. Does that mean that we have to deny that inversion (or spectrum shift) without misrepresentation is possible? Yes. But I don’t think that this result is as devastating as is commonly supposed, because I think that any way of arguing that neither of our spectrum inverted subjects is misrepresenting makes use of premises which are far from obvious. Let’s consider three such arguments. One standard line of argument holds that denying the possibility of inversion without misrepresentation involves unfairly privileging one of the subjects over the other. This is not quite right, at least in the case of color, since we might be color eliminativists and take both subjects to be misrepresenting. But even if we do say that one is right and one is wrong, it’s not obvious that this is objectionable. This conclusion wouldn’t, after all, entail that one of the subjects was irrational or blameworthy for her color judgements. If we think that objects really do have colors, then it seems to me no more odd to say that one but not the other of a pair of spectrum inverted subjects gets the color right than it is to say that my perceptions are veridical whereas the perceptions of someone in a Cartesian skeptical scenario indistinguishable from my own are not. One might challenge this view by asking what, in the inversion case, grounds the fact that one subject is right and the other wrong. But if we are presuming the falsity of 1

See, for example, the discussion of ‘coarse-grained colors’ in Cohen (2008).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

168 relativist views of sensible qualities color eliminativism, it’s hard to see what’s wrong with answering, ‘The color of the object.’2 This response to spectrum inversion cases can be turned into a challenge to those who think that it is implausible to deny that spectrum inverted subjects could each be correctly visually sensing the color of an object. It is quite plausible that it is possible for a subject to be in a Cartesian skeptical scenario indiscriminable from my own, and that such a subject would have false beliefs about his situation, but would not be in any way blameworthy or irrational for holding those beliefs. I say the same thing about someone spectrum inverted with respect to me. What property does the spectrum invert have, and the Cartesian subject lack, which makes this view plausible in the latter case but not in the former? A second, more challenging line of argument in favor of the possibility of spectrum inversion without misrepresentation is based on the possibility that the spectrum inverted subjects could share a language.3 For surely then, one might think, ‘red’ means the same thing out of each of their mouths. But if each is a competent user of the language, then each can surely report her beliefs using sentences of it, so that when each sincerely asserts some sentence of the form ‘n is red’ they must have a belief with the same content. But there’s no reason to believe that either should be unable to form perceptual beliefs in the usual way; so presumably, in the standard case, when each forms a perceptual belief which they would sincerely express via some sentence of the form ‘n is red,’ they are visually sensing the same content (or at least the same color content). But this entails that spectrum inverted subjects could each be correctly representing the color of the relevant object, despite the differences in their phenomenal properties. This is a formidable argument. But I think that its weakest link is the first: the assumption that each of the spectrum inverted subjects must use ‘red’ with the same content. Rejecting this assumption is not, obviously, a move that one could make if one were attracted to the Wittgensteinian idea that when it comes to meaning ‘nothing is hidden’4—but, setting that aside, there seems to me nothing especially implausible in holding that, for an individual language user, terms for properties, like colors, which figure in the perceptual experiences of that language user, get their contents from the contents of the experiences which dispose that language user to predicate the term of something.5 If this view is right, then 2

For similar sentiments, see Byrne and Hilbert (2007). This line of argument is forcefully presented in Thau (2002). 4 Wittgenstein (1953), }435. 5 The ‘of that language user’ qualification is important—given that redness is never among the contents of the experiences of someone blind from birth, ‘red’ would not be a term of the relevant class for that language user. 3

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

relativist views of sensible qualities 169 someone spectrum inverted with respect to me will simply mean something different by ‘red’ than I do—even if in certain situations this difference in meaning will be undetectable. A last challenge focuses, not on spectrum inversion, but on spectrum shift. Here we can’t escape with convicting merely possible subjects of misrepresentation; given widespread actual spectrum shift, the line of thought above generalizes to the conclusion that almost all of us are, almost always, misrepresenting the colors of things. But even this conclusion is not all that surprising. In Chapter 11, I criticized the argument from fineness of grain for Nonconceptualism. But one thing that argument has right is that sensed contents typically represent the world in a much more determinate way than do the contents of our beliefs and thoughts. With the greater determinacy of sense representation come advantages and disadvantages. Among the former is an increase in the number of distinctions in color we can make on the basis of visual experience. But among the latter is a greater chance of misrepresentation. Take some belief you have about the color of some object which you encounter reasonably often but which is not in your presence right now—say, your coat. Suppose that you were forced to make your belief about the color of your coat much, much more determinate. It is highly likely, if you are at all like me, that you would in this process turn a true (but relatively indeterminate) belief about the color of your coat into a false (but much more specific) one. If we were restricted to forming such specific beliefs about the colors of things, it would be quite unsurprising if most of our color beliefs turned out to be false. But this provides something of a model for sensing properties. We cannot but sense that objects have quite determinate colors; given this fact, is it really that surprising if it turns out that we misrepresent the colors of most things most of the time?

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART VI

How Many Phenomenal Relations?

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

24 Phenomenal Relations On the view being developed, when one has a perceptual experience one stands in a sensing relation to a content. Our main focus so far has been on the nature of the contents: we’ve defended the view that these are Russellian; that they sometimes have, among other things, objects and natural kinds as constituents; and that they are a kind of property. Let’s now turn our attention to the sensing relations in which, in perceptual experience, we stand to those contents. In Chapter 15, I argued that sensing relations involve a certain kind of selfascription. But that view does not answer another question: the question of how many sensing relations there are. At first, this question seems very easily answered: there are quite a lot of sensing relations. After all, we ordinarily distinguish visual representation from auditory representation and the other senses, and distinguish all of them from, for example, pain states and itch states and the states associated with other bodily sensations. And surely the relation of sensing by a visual experience is distinct from the relation sensing by an auditory experience. But once we see how easily sensing relations are multiplied, the question of how many sensing relations there are quickly begins to look like a merely verbal question. Given any sensing relation R, we can find more simply by putting extra conditions on R. If the relation of sensing by a visual experience is distinct from the relation sensing by an auditory experience, so sensing by a pleasant visual experience is distinct from the relation sensing by a horrible visual experience. There are, after all, many contents to which I have stood in the first relation but not the second and, unfortunately, vice versa. The reasonable view is surely that we can simply decide, for certain purposes, to distinguish between these sensing relations, and for other purposes, count them as of the same type. This after all, seems like the right view about debates about how many senses there are—how many senses we ought to distinguish seems to depend, in part, on the purposes of the categorization.1 1

See Macpherson (2011) for different axes on which the senses might be individuated.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

174 phenomenal relations Our purpose in investigating sensing properties and sensing relations is the task of understanding the nature of phenomenal properties. Keeping this purpose in mind will help us to turn our question about the number of sensing relations from verbal to substantive. A first step is to define the following condition on relations between subjects and contents: R is a phenomenal relation iff possibly, two subjects are representational duplicates but for differences in the contents to which they are R-related and those subjects are instantiating distinct phenomenal properties.2 Belief is a good example of a relation to a content which does not seem to satisfy this condition; two subjects who differ only in what they believe will not instantiate different phenomenal properties. So belief is not a phenomenal relation.3 By contrast, the relation to contents in which I stand when I have typical visual experiences does seem to satisfy this condition. It seems that there could be a subject who was a mental duplicate of me but for the fact that that subject is visually sensing the wall before him as blue rather than green; such a subject would instantiate different phenomenal properties than me. Supposing that our aim is to find a minimal supervenience base for phenomenal properties, what we want to know is how few phenomenal relations we need to recognize in order to find a supervenience base for the phenomenal. This gives us the following sufficient condition for distinctness of phenomenal relations:

2 There are some complications involving this definition of phenomenal relations if there are necessary connections between certain mental states. For example, visually representing will trivially fail to satisfy this definition if, for example, it is impossible to visually represent x as blue without believing that you are having a visual experience, since then it will be impossible for there to be a pair of subjects who only differ mentally in the contents of their visual experiences. We could get around this by changing the definition to: ‘representational duplicates but for . . . (and any other mental differences which are entailed by this difference in the propositions to which they are R-related).’ But this definition would mistakenly count factive mental states which are about phenomenal states—for example, knowing that I am visually representing such and such—as themselves phenomenal. We could get around this by requiring that the states in question be non-factive; but this seems ad hoc. As a fallback, we could simply add this condition and claim only that the condition which results is sufficient for being a phenomenal relation; that would be enough for what follows. 3 Here I’m assuming that a global intramodal intentionalism is true. If it weren’t, then by this definition it would follow trivially that every relation to a content, including belief, would be a phenomenal relation. For suppose that intentionalism was false of, say, pains. Then two subjects could differ phenomenally with respect to their sensations of pain without any difference in the contents of their pain sensations. The two subjects might also be representational duplicates but for a difference in the contents of one of their beliefs. Then they would be a pair of subjects which differed phenomenally but were representational duplicates but for a difference in the contents of one of their beliefs, which would make belief a phenomenal relation. Parallel arguments could be constructed for any mental state type.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal relations 175 If R, R* are phenomenal relations, then R6¼R* if possibly, there are a pair of subjects, A and B, which are such that: (a) the only representational difference between A and B is that there are one or more contents to which A is R-related and which B is R*-related, and (b) the phenomenal property instantiated by A 6¼ the phenomenal property instantiated by B For suppose that relations R1, R2 meet this condition. Then, if we did not include the distinction between R1 and R2 in our list of the subvening representational properties of subjects, we will have failed to provide a supervenience base for the phenomenal—in that case there would be a pair of subjects with different phenomenal properties but no corresponding difference in their representational properties. I will go on to ask how many phenomenal relations there are. But this should be understood as shorthand for the question: how many phenomenal relations do we need to distinguish in order to provide a supervenience base for the phenomenal? Because it is not a verbal question when a pair of subjects differ in their phenomenal properties, it is not a verbal question which list of phenomenal relations is the minimal list required to provide a supervenience base for the facts about the phenomenal properties of subjects.4 This question about the number of phenomenal relations can be thought of as breaking into two. First, how many sensing relations are there? Second, are there any phenomenal relations which are not sensing relations? Sensing relations, recall, were defined in part in terms of the notion of a truth-sensitive relation to contents; but the notion of truth-sensitivity plays no role in the definition of a phenomenal relation, and there is nothing to stop there being some phenomenal relation which is representational but not truth-sensitive. More on this possibility in Chapter 27. The relevance of our question about the number of phenomenal relations to attempts to analyze or naturalize phenomenal properties should be clear. I argued (in Chapters 10–13) that the purest intermodal intentionalism is false, because we can stand in sensing relations to the same contents as we can stand in belief relations, and that the distinction between sensing and believing a content can involve a phenomenal difference. In the present terms, this was to claim that subjects can stand in phenomenal and non-phenomenal propositional attitude relations to the same content. From this it follows that phenomenal properties 4 We can make a similar point if we wish to identify phenomenal and representational properties. If phenomenal properties are identical to relations to certain contents, it can’t be a mere matter of convention whether we count relations as the same or different—any more than it can be a mere matter of convention which phenomenal property I’m instantiating now.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

176 phenomenal relations don’t supervene simply on facts about what the contents of a subject’s mental states are; rather, they supervene only on these facts plus facts about the phenomenal relations in which subjects stand to those contents. This has obvious consequences for any attempt to identify phenomenal and representational properties; the relevant identity will not be of the form phenomenal property F ¼ the property of standing in some propositional attitude relation or other to p but rather of the form phenomenal property F ¼ the property of being R-related to p for some phenomenal relation R. But even if the example of belief shows that one can’t explain phenomenal properties in terms of content alone, one might still hope that one can explain phenomenal properties given a satisfactory account of content plus an account of a single phenomenal relation. This is no trivial task, but it certainly seems easier than packaging an account of mental representation with independent accounts of each of several distinct sensory modes of being related to a content. In the next chapter, we’ll investigate this question by thinking about whether the distinctions between the senses force us to recognize distinct phenomenal sensing relations.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

25 The Distinctions Between the Senses Let’s consider first the question of whether the usual candidates for the senses— vision, hearing, smell, taste, and touch—each involve their own phenomenal relation. If they did, it would be bad news for the theorist who wanted to identify phenomenal and representational properties so that she could go on to give a naturalistically acceptable account of the latter (and hence also of the former). Such a theorist would owe some account of the distinctions between the relations of visually sensing, auditorially sensing, and the others; and this, from the point of view of intentionalism, is not easy to do. One might try to explain the distinctions between the senses in terms of the properties represented by saying that, for example, any sensing relation whose contents include color properties is an instance of visual sensing. But this doesn’t seem to help the cause of this theorist, who says that phenomenal properties supervene on content plus sense modality but not on content alone—one can’t combine this with the view that sense modality supervenes on content, since then it would follow from the supervenience of phenomenal character on content plus modality that phenomenal character supervenes on content alone. Alternatively, one might try to analyze the distinctions between visually representing, auditorially representing, et al. in terms of the phenomenal properties distinctive of the various sense modalities; but this is objectionably circular from the point of view of an intramodal intentionalist who tries to explain phenomenal properties in terms of content þ the phenomenal relation distinctive of that sense modality. One might try to explain the distinction in terms of the sense organs involved. But the fact that they involve the eye does not seem to be a necessary feature of visual experiences; we can imagine a blind person being fitted with a device which allowed her to enjoy visual experiences, even if the device bypassed the eye.1 1

See the discussion of prosthetic vision in Macpherson (2011).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

178 the distinctions between the senses Bad news though it may be, it might seem as though there is a pretty difficult to resist argument in favor of the view that each sense involves its own phenomenal relation. Given the sufficient condition for distinctness of phenomenal relations stated in the last chapter, we know that, if all of the senses involve the same phenomenal relation to a content, any difference in phenomenal character between an experience in one modality and an experience in another modality must also correspond to some difference in content between the two experiences. But one might think that any view of this sort runs into an obvious problem with common sensibles: qualities which are represented by more than one sense modality. Consider a visual experience of a square and a tactile experience of a square. Both cases are instances of sensing a certain property: squareness. So, one might think, if phenomenal character supervenes on content, there must be some phenomenal commonality between the two experiences. But, if we think about the two experiences, it is very hard to see what this phenomenal commonality could be—there just seems to be nothing in common between the way that shapes look, and the way that they feel.2 There is a standard reply to this objection, and that is just to note that, as it stands, it is not an objection to any intentionalist thesis at all.3 As in other cases, the appearance of a conflict here comes from the focus on properties of experience, rather than properties of subjects of experience. The supervenience theses in which we are interested are theses about the supervenience of a subject’s phenomenal properties on her representational properties. But the example of tactile and visual experiences of shape is not even an attempt to provide a counterexample to any thesis of this sort, since, while it is clear that someone touching a square will differ in her phenomenal properties from someone seeing a square, there is no argument that the contents to which the subject stands in sensing relations will be the same, as opposed to merely overlapping. And, pretty obviously, they won’t be the same, since, for example the visual experience will represent the shape as having some color, and the tactile experience won’t. (This is analogous to the point made in Chapter 4 about ‘strong supervenience’ vs. ‘global supervenience’ versions of intentionalism; common sensibles are only even a prima facie problem for ‘strong supervenience’ versions of intentionalism, which we’ve already found reason to reject.)

2 Not everyone agrees; see John (2005), 176. John thinks that there is a phenomenal similarity between these experiences, but that it is hard to notice because it is overshadowed by the many phenomenal differences between those experiences. It’s difficult to know how to argue this point. 3 See, for example, Tye (1995), 156–7.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the distinctions between the senses 179 In our terms, this is just to say that cases involving common sensibles do not meet the sufficient condition for distinctness of phenomenal relations stated in the last chapter, because they aren’t cases in which two subjects are representational duplicates but for the fact that one bears a visual and the other a tactile relation to the same content. What would it take for an example of this sort to meet that sufficient condition? Focusing on the example of vision and touch, what we would need, it seems, is a case in which a visual experience and tactile experience had exactly the same content, and in which there was a difference in the subjects’ phenomenal properties. This sort of case really would force us (so long as we hold to intentionalism) to posit distinct phenomenal relations for vision and touch. But it’s at least not easy to see how the distinctions between senses could give us a case of a this sort—even in cases of common sensibles, there always seems to be some difference in content between experiences in one modality and experiences in another. The foregoing shows that we lack an argument (at least of this form) for the claim that each sense is associated with its own phenomenal relation. But there are also facts about the relations between the senses which provide strong evidence that we should not think of the senses as each involving a fundamentally different sensing relation. These problems can be brought out by considering how the proponent of the view that each sense does involve a fundamentally independent sensing relation should think of the content of a total perceptual experience, which involves input from more than one sense. On this sort of view, sensing is a disjunctive relation: one senses a content iff one visually senses it or auditorially senses it or . . . . On this picture, the total content of one’s perceptual state is simply the conjunction of the propositions which are visually represented, those which are auditorially represented, and so on. This is a view of the relationship between the senses which Casey O’Callaghan has aptly dubbed the ‘composite snapshot’ conception of experience. However, as O’Callaghan has argued, this view of the relationship between the senses seems inadequate to handle the phenomenon of intermodal binding.4 Cases of perceptual binding are cases in which a perceptual experience represents several properties as properties of a single object. Intramodal examples are familiar. When I look at a tomato, I don’t just see that redness and roundness are both instantiated in my environment, but rather that both are instantiated by the same thing—and, crucially, no view of the contents of my visual experience can 4 See O’Callaghan (2008). This is also related to the argument of Tye (2007), though Tye focuses on ‘phenomenal unity’ rather than binding.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

180 the distinctions between the senses neglect the fact that, in such an experience, I do more than visually represent that there is something red and that there is something round; I also visually represent that there is something which is both red and round. As is often the case, we can argue for this conclusion using the veridicality conditions of such an experience.5 Consider a visual experience which, intuitively, represents something as both red and round. We would consider this experience to be illusory if had in a situation in which there is red object which is not round and a round object which is not red in the environment of the subject. But then it follows that there must be more to the content of my visual experience than that there is something red and that there is something round, since both of these propositions are true. A very natural suggestion is that the false content— the content in virtue of which the experience counts as illusory—is the content that something is both round and red. But it is very plausible that there are also cases of intermodal binding. Sometimes we represent a sound as coming from some object which we also visually perceive to have a certain color, or feel a surface as cold that we also visually represent as blue. Think about a case in which I hear a barking sound as coming from a black and white dog. My visual experience represents there being a black and white dog in such and such location relative to me, and my auditory experience represents there being a barking sound in my vicinity. But there must be more to the content of my experience than this, since I perceptually represent the white-andblackness as a property of the source of the barking—and this content cannot be visually represented (since you can’t visually represent a barking noise) and cannot be auditorially represented (since you can’t auditorially represent colors). This indicates that there is at least one proposition—that something is both black and white and barking—which is sensed, but is neither visually sensed nor auditorially sensed. This is a problem for the composite snapshot conception of perceptual representation, since it indicates that there is more to the content of my perceptual experience than the conjunction of the propositions that I visually represent, auditorially represent, etc. As above, we can defend this view of the content of the barking dog experience by considering the veridicality conditions of the experience. If it turns out that there is something barking in my vicinity, but not the dog, we would count the experience as illusory, which suggests that there must be some false sensed

5 Here and in the following chapter, I simplify a bit by treating the objects of sensing relations as propositions rather than the non-propositional monadic properties of Chapter 15, and hence as true or false simpliciter rather than true of the subject of the relevant experience. The relevant points could easily be translated into the latter terms.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the distinctions between the senses 181 content. But the relevant visually sensed content—that there’s a black and white dog there—is true, as is the relevant auditorially sensed content—that there is a barking sound in my vicinity. A very natural candidate for the false content is then the ‘intermodal’ content: the one that predicates barking and black-andwhiteness of the same thing. And if there is such a thing as sensing such contents, the composite snapshot picture is false, since such contents can’t be represented by propositional attitudes associated with any of the five individual senses. One might try to block this argument by pointing out that we don’t just auditorially sense that there is a barking noise somewhere or other in the vicinity, but rather that that noise is located in a particular relative location. Couldn’t the relevant false sensed content be a false representation of the location of the sound’s origin? This is a plausible view of the contents of auditory experience, but doesn’t affect the present argument very much. Presumably even if we represent sounds as located, we don’t represent them as located at absolutely precise points; rather, we represent sounds as located in regions of space. But now just imagine that the genuine source of the sound—perhaps a very small microphone attached to the dog’s collar—is within the region at which the sound is represented as located. We would still take the experience to be illusory, despite the fact that the auditorially sensed content—that there is a barking noise there, or coming from there—is true. Hence to find a false proposition we again need to go to the cross-modally represented proposition. The best reply on behalf of the composite snapshot view is to emphasize that the contents of perceptual experiences are not general propositions that something or other is white and black and that something or other is barking, but rather singular propositions which predicate barking, and black and whiteness, of some particular dog d. (We’ve already seen, in Chapter 18, that perceptual experiences do sometimes have singular propositions of this sort as their content.) This version of the composite snapshot picture has the resources, it would seem, to handle the datum about illusions just mentioned. After all, if the subject just described has a visual experience which has the singular content that d is white and black and an auditory experience which has the content that d is barking, then, if the white and black thing is not the barking thing, either the visually represented or the auditorially represented singular proposition will be false—which is the result the proponent of the composite snapshot conception should want. However, there are three reasons why this sort of reply seems inadequate. The first begins with the observation that capturing the veridicality conditions of an experience is necessary but not sufficient for capturing its content. To see this, consider by way of analogy the belief states of two subjects. The first might

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

182 the distinctions between the senses believe of some object o that it is F, and also believe of o that it is G; the second might believe both these things, but also do something else: she might integrate the two beliefs, thus forming the belief of o that it is both F and G. This is a genuine difference in the beliefs of our two subjects; in general, there is no true closure principle to the effect that if a subject believes two propositions, he also believes their conjunction. However, the difference between their beliefs is not accompanied by a difference in the truth conditions of the belief sets of the two subjects; one’s beliefs are true iff the other’s are. There is surely room for an analogous difference in perceptual content without a difference in veridicality conditions. One can’t, however, demonstrate the existence of such a difference using illusions, which play upon our intuitions about misrepresentation; so how can we argue that the picture provided by the composite snapshot theorist who relies on singular propositions is still missing something? As noted earlier, we have a grip on the distinction between, on the one hand, believing of some object that it is black and white and believing of the same object that it is barking, and, on the other hand, believing of some object that it is both black and white and barking. While believing and sensing aren’t the same thing, we can sometimes use the former as a rough test for the latter, by asking: what beliefs would I form if I simply took my perceptual experience at face value? In the case just described, one of the beliefs I would thus form would, it seems, be the belief which predicates of the dog the conjunctive property of both being black and white and barking. This is some indication that this conjunctive proposition is also part of the content of the experience. Contrast this with a case in which I visually represent that d is black and white and auditorially represent that d is barking, but do not perceptually represent d as both barking and black and white. (Suppose that, unbeknownst to me, I’m seeing d’s reflection in a mirror, and therefore visually represent him as to my right, but that I (correctly) auditorially represent the bark as coming from the left.) If I took this experience at face value, I would not form the conjunctive belief which predicates black and whiteness and barking of d. That is an argument that even if the modified composite snapshot conception can get the veridicality conditions of the experience right, it’s still not giving an adequate account of the contents of the experience. However, second, there’s also reason to doubt that it can get the veridicality conditions right, because it is hard to see how it can get the right singular propositions to be the contents of the relevant visual and tactile experiences. Consider the illusory experience discussed earlier, in which the barking noise is in fact not produced by the dog, but instead by a small microphone hidden inside the dog’s collar. The result that the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the distinctions between the senses 183 composite snapshot theorist should want is that the content of the subject’s auditory experience attributes barking to the dog. But why should this be the content of the subject’s auditory experience? Why not think that, instead, the subject correctly represents the source of the sound as the dog’s collar, rather than its mouth? After all, the auditory experience might well be produced by a causal chain of the sort characteristic of veridical auditory experiences. Third, and last: we might reasonably doubt whether singular propositions can do all of the work for which they were introduced, because we can give plausible examples of ‘intermodal’ perceptual representations which are not cases of perceptual binding, and make no essential use of singular propositions. One way to do this is by adapting some examples from Susanna Siegel’s discussion of the perceptual representation of causation. Here’s one of Siegel’s examples: Suppose you are playing catch indoors. A throw falls short and the ball lands in a potted plant, with its momentum absorbed all at once by the soil. You see it land, and just after that, the lights go out. The ball’s landing in the plant does not cause the lights to go out, and by hypothesis you do not believe that it does. Nevertheless it may seem to you that the ball’s landing somehow caused the lights to go out. This is the first case. In the second case, you likewise see the ball land and the lights go out. But this case is unlike the first: you do not have any feeling that the ball’s landing caused the lights to go out . . . It seems plain that there can be a phenomenal difference between two such experiences.6

Siegel argues, plausibly, that corresponding to this phenomenal difference is a difference in sensed content: in the first experience, the ball landing is perceptually represented as the cause of the lights going out and, in the second case, this is not represented. If we grant that the cases do involve a difference in the phenomenal properties of the subjects, it can be turned into a case which forces us to recognize the existence of sensed relations between items which cannot each be the object of any one sense. That is because, even if Siegel’s example concerns a pair of visual experiences, there is, as others have noticed, no reason why examples of this sort have to.7 Just replace the example of the lights going out with a loud booming noise—then, if Siegel’s argument is sound, it follows that we perceptually represent a ball landing as the cause of a sound. But this can’t be visually represented— since we can’t visually represent loud booming noises—and it can’t be auditorially represented, since we can’t auditorially represent the noiseless path of a ball through the air. So we have a perceptual experience which represents a causal relation as obtaining between the instantiations of an exclusively visible quality

6

Siegel (2009), 526.

7

See Nudds (2001), 218–19 and O’Callaghan (forthcoming).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

184 the distinctions between the senses and an exclusively audible one; hence that perceptual experience can’t belong to any of the five sense modalities. The cases of cross-modal representation we’ve discussed have the following consequence: there are propositions which are the contents of experiences but are not the contents of any of the propositional attitudes which we imagined to be associated with each of the five senses. But one might think that this shows less that the composite snapshot conception of perceptual representation is fundamentally flawed than that it is simply incomplete. Perhaps the composite snapshot theorist is right that corresponding to each sense is an independent representational system; but perhaps, in addition to these, there is a sort of sixth, composite modality of perceptual experience whose job it is to represent the ‘intermodal’ propositions described earlier. Let’s call the propositional attitude associated with this extra composite modality of sense perception ‘C-representation.’ As we’ll see, closer attention to the contents of this supposed extra phenomenal relation make this modified version of the composite snapshot conception seem less than attractive. To see this, think about a case of intermodal binding like those discussed earlier, in which a subject visually senses o as F, auditorially senses that o is G, and, intuitively, also senses that o is both F and G. In this case, the modified composite snapshot theorist will say that, in addition to what the subject visually senses and auditorially senses, she also C-represents the proposition that o is both F and G. Now, ordinary sorts of perceptual representation, just like belief and other truth-sensitive propositional attitudes, clearly distribute over conjunction. If I visually sense an apple as both round and red, it follows that I visually sense the apple as red, and that I sense it as round. And it is really not easy to see how a type of sensing relation could fail to distribute over conjunction; surely, no matter what sort of perceptual representation we’re talking about, if I perceptually represent, in any sense modality, an object as having both of two properties, I must also, in that sense modality, represent it as having each of those properties. So if, in a case of intermodal binding, I C-represent o as both F and G, I must also C-represent o as F and C-represent o as G. But then it looks like Crepresentation is swallowing up the other species of perceptual representation: in normal cases of intermodal binding, C-representation will redouble the representational efforts of the visual system, the auditory system, etc. But at this point those other systems of perceptual representation look redundant—why should we believe in visual representation, etc., if the work that they are supposed to do is already done by C-representation? The natural thought at this point is that C-representation is not an addition to visually sensing, auditorially sensing,

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

the distinctions between the senses 185 et al., but rather that the latter are simply abstractions from the former. If this is right, then the best view of the various senses is as each involving a single common sensing relation. Here’s another way to bring out the oddness of thinking about C-representation as a sensing relation on a par with sensing relations associated with the other five senses. Suppose for reductio that one can C-represent that something is red without visually representing that it is red. What phenomenal properties would the subject of such an experience instantiate? She couldn’t instantiate RED, since that would seem to entail, contra our supposition, that the subject visually senses redness as instantiated. The subject had better instantiate some phenomenal property or other, if C-representation is to be a phenomenal relation at all. But what phenomenal property could it be? The best conclusion is that C-representation without visual sensing is impossible, and that it is a necessary truth that if a subject is C-representing that o is red, that subject is also visually sensing o as red. But then it looks increasingly hard to see what the real distinction between Crepresenting that o is red and visually sensing that o is red is supposed to be. They are representational properties with the same content and are associated with the same phenomenal properties; indeed it’s hard to think of any property which the one has but the other does not. The most reasonable conclusion is that there is no real distinction here, and that what we have is just one phenomenal relation which we can call by different names. Remember that the distinction between intermodal and intramodal intentionalisms is one of degree: we’ve already seen that there is good reason to think that the purest intermodal intentionalism is false, since there is good reason to think that it is possible for subjects to believe and sense the same contents. What the argument of this chapter shows is that the purest intramodal intentionalism must also be false. That on which the phenomenal character of perceptual experience minimally supervenes can’t be content plus the particular propositional attitude associated with one of the five senses, since there aren’t distinct propositional attitudes associated with each of the five senses. Instead, when it comes to perceptual representation, there’s just one basic phenomenal relation. Visually sensing is to auditorially sensing as traveling by foot is to traveling by car; the first two are different ways of sensing, just as the latter two are different ways of traveling. But, just as one would never try to analyze traveling as a disjunction of the various ways in which one could travel, so we should not analyze sensing as a disjunction of the various ways in which one could sense.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

26 Binding and Bodily Sensations So cases of intermodal binding push us, with respect to the distinctions between the five sense modalities, away from an intramodal intentionalism and toward an intermodal intentionalism. But these sorts of cases can also be used to show more than this, since we can generate cases of intermodal binding not just between different modalities of perceptual experience, but also between perceptual experiences and simultaneous bodily sensations.1 Imagine that you look down at your thigh as you feel a sharp, stabbing pain there; you see what seems to be a small but sharp knife protruding from just that part of your thigh where you feel the pain. In such a case, it may well perceptually seem to you that the knife—the object with just that color and shape—is the source of the pain you feel in your thigh. As a test for this claim, we can, as before, turn to an illusion designed to mimic the case just described. Imagine that the grey object which seems to be protruding from your thigh is in fact a plastic toy knife with the end cut off which is pressed against your thigh, and that the stabbing pain was caused by a toxin placed in the cup of coffee you’re drinking. In this sort of case, it again seems that the subject is misrepresenting his environment, and it is quite plausible that this misrepresentation is a matter of misrepresentation involving his sensing properties rather than, for example, a false judgement. As usual, we can provide evidence for this classification by pointing out that illusions of this sort could persist despite the subject’s knowing that the relevant experience is not veridical, and hence having no temptation to make the relevant judgement.2 Just as the cases of intermodal perceptual binding discussed in the previous chapter can be used to argue that the senses are not associated with distinct propositional attitudes, so cases of binding involving bodily sensations can be 1

Tye (2003) also emphasizes the unity between perceptual experiences and concurrent bodily sensations, though his emphasis is on phenomenal unity rather binding (which he calls ‘object unity’). But the kind of view I argue for here is in some ways very similar to Tye’s. 2 This would be similar in some ways to the rubber hand illusion; see Botvinick and Cohen (1998).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

binding and bodily sensations 187 used to show that we should not think of perceptual experience, on the one hand, and various bodily sensations, on the other, as associated with distinct propositional attitudes. For, if we did, we would need some distinct propositional attitude—the analogue of C-representation—to have as its contents propositions like the proposition that that grey thing is the source of my pain, which can neither be the contents merely of visual experience (since you can’t see pain) nor of my pain sensations (since you can’t feel greyness). But then we would run into just the same problems making sense of the relationship between this new sensing relation, on the one hand, and perceptual experience and bodily sensations on the other, as we did making sense of the relationship between C-representation and the individual senses. We now have two reasons to reject the intramodal intentionalist thesis that the supervenience base for the facts about phenomenal character must distinguish between the sensing properties which involve a perceptual sensing relation and those which involve a sensing relation characteristic of one or another type of bodily sensation. The first reason is just that we have no example of a bodily sensation and a perceptual experience which would force us, via the sufficient condition for distinctness of phenomenal relations in Chapter 24, to distinguish between the phenomenal relations involved in bodily sensations and in experience. The second is that this sort of merely intramodal intentionalism presupposes that perceptually representing that such-and-such and having a bodily sensation with the content that such-and-such are fundamentally distinct sensing relations; but examples of binding show that there is good reason to reject this view.3 Just as we should reject the thesis that perceptual representation is an amalgam of more fundamental propositional attitudes associated with each of the senses, so we should reject the thesis that sensing properties in general are an amalgam of perceptual representation and representation by various bodily sensations. Instead, we should take these distinctions, like the distinctions between the senses, to be abstractions from a more fundamental propositional attitude. When I introduced sensing relations in Chapter 2, I was agnostic about how many sensing relations we need recognize to find a supervenience base for the

3

It also provides a further argument against a local intentionalism which says that the supervenience of phenomenal content on content holds for perceptual experiences but fails for bodily sensations, and against the view that bodily sensations, unlike perceptual experiences, lack a representational content (as in McGinn (1988)). These views only seem to make sense if we can draw a clear line between perceptual experiences and bodily sensations, and cases of binding like the example of the knife make it hard to see how we could do that. An interesting further question is whether these sorts of binding examples might be extended in some way to emotions and moods.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

188 binding and bodily sensations phenomenal. I take the last two chapters to make a strong case for the conclusion that we need recognize only one. In what follows, then, I’ll set the agnosticism of Chapter 2 aside and, rather than referring generically to ‘sensing relations’ will simply refer to the one phenomenal relation of sensing in which, I think, subjects of various sorts of perceptual experiences and various sorts of bodily sensations all stand to contents.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

27 Shifts in Attention So far the following picture has emerged: every state with an associated phenomenal property is associated with a single truth-sensitive representational relation, which I’ve called ‘sensing.’ Necessarily, if any two subjects are sensing the same content, then the two subjects instantiate the same phenomenal properties. This is a global intentionalist claim, and one which is only one step removed from a purely intermodal intentionalism, in that it claims that we need recognize only one phenomenal relation, and hence that just one relation to contents—the attitude of sensing—is a part of the supervenience base for phenomenal properties. I’ve already sketched the attractions of this sort of view. However, a reasonable case can be made that this simple view is too simple, and that we need to recognize the existence of multiple phenomenal relations. There are a few different ways in which one might try to argue for this conclusion. One would be to start from the thought that there is such a thing as cognitive phenomenology, and hence something that it is like to make a judgement or entertain a thought. Since (as argued in preceding chapters) it is very plausible that one could judge the same content as another subject senses, but very implausible that what it is like to judge a content is identical to what it is like to sense that content, the existence of cognitive phenomenology would force us to recognize judgement (or whatever phenomenal relation is claimed to be distinctive of cognitive phenomenology) as a phenomenal relation distinct from sensing. A different line of argument would focus, not on cognitive phenomenology, but on the phenomenal properties associated with imagination. Even if the contents which we imagine are typically far less complex and less determinate than the contents we sense, it seems possible for there to be subjects capable of imagining contents as complex and determinate as the contents we sense; and it is not implausible to think that, even in this case, the phenomenal properties of the subjects carrying out the relevant imaginative acts would differ from those of subjects sensing those same contents.1 This would force us to recognize the existence of imagination as a phenomenal relation distinct from sensing. 1

Thanks to Brian Cutter for helpful discussion here.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

190 shifts in attention In this chapter, I’ll pursue a third route to the existence of multiple phenomenal relations. I’ll argue that certain kinds of attentional shifts show that attention must be thought of as a phenomenal relation distinct from sensing. Consider, for example, a visual experience of the following lines on a white sheet of paper large enough to fill the perceiver’s visual field:

Compare two visual experiences of these lines: in the first, the perceiver’s attention is focused on the intersection of the second vertical line from the left with the horizontal line; the second differs only in that the perceiver shifts his attention to the point of intersection to the right, between the horizontal line and the third vertical line from the left. It is undeniable that the two subjects differ in their phenomenal properties; what it is like to be the first is plainly discriminable from what it is like to be the second. But do the two subjects sense different contents? A natural first thought is that if there is a difference in sensed contents, this must be a difference in the representation of the location, shape, or color of the figure or one of its parts; given that the background of the figure is an uninterrupted stretch of solid white, the change in focus between the two points of intersection does not bring with it a change in the representation of anything on the periphery of the perceiver’s visual field. But there does not seem to be any such difference in the representation of the properties of this figure. The figure does not seem to move relative to the subject when one shifts one’s attention from one point of intersection to another. (To make the case clearer, we can imagine that this shift in attention does not involve any eye movement; it is possible, even if a bit unnatural, to shift one’s attention from one point in the visual field to another without foveation.) And given the simplicity of the figure, it does not seem plausible to claim that one experience represents a given portion of the lines with more detail or determinacy. Nor, if the points of intersection are close enough together, is there any clear difference in the determinacy with respect to which the locations of the points of intersection are represented.2 2

One might object that this last point rests on implausible assumptions about the size of the area a typical visual experience represents with clarity. But we can concede that we only visually represent color and shape properties with a high level of determinacy within a surprisingly small area of the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

shifts in attention 191 Moreover, we can put some pressure on the intentionalist by remembering that intentionalism is a modal claim. Therefore, what the intentionalist must say about these cases is not just that, in the case of human visual experience, there happens to be a difference in sensing properties, but also that, for any possible pair of experiences which involve this sort of attentional shift, there is such a representational difference. This seems to me implausible. Even if, in the case of human beings, attentional shifts always involve some difference in the determinacy of the representation of (for example) relative location, this seems to be a contingent fact about the workings of the human visual system. Surely there could be a creature for whom attentional shifts were possible without this sort of representational difference; and the mere possibility of such creatures is enough to make trouble for the ambitious intentionalist thesis under discussion. But the intentionalist who concedes that there is no difference in the representation of the location, shape, or color of the represented figure or its parts is not quite out of options. The intentionalist might reply by finding some overlooked class of properties with respect to which the two subjects do, contrary to initial impressions, differ in sensed content. This is a time-honored intentionalist strategy for responding to potential counterexamples to the supervenience of phenomenal character on content. Often, the plausibility of such counterexamples rests on an impoverished view of the sorts of properties represented in experience, and the intentionalist can offer a plausible treatment of the problematic cases by calling attention to a difference in the representation of the overlooked properties to correspond to the relevant phenomenal difference.3 But in this case, what sensed properties could we be overlooking? The intentionalist might point out that, as emphasized in Chapter 15, it is plausible that, in perceptual experience, we often represent egocentric properties, like relative distance and relative orientation. In such cases, the subject is not just representing properties of the objects in her environment; she’s representing those objects as standing in certain relations to her. This suggests a way of handling the sort of attentional shift described earlier: the intentionalist might be tempted to say that our two experiences differ with respect to which points of intersection are represented as prominent to the perceiver. After all, if we can make use of perceptual representation of egocentric locations and orientations, why not also representation of egocentric relations of perceptual prominence?

visual field (one estimate is an area of one degree of visual angle). The example can be adapted so that the shape in question is smaller than that area, but still large enough for the subject to be able to shift attention from one point of intersection to another. 3

For an excellent example of this strategy in action, see Tye (2000), Ch. 4.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

192 shifts in attention However, there is an important disanalogy between the representation of egocentric relations of distance and orientation and the representation of egocentric relations of perceptual prominence. The latter, but not the former, involves representation of mental properties of the subject of the experience. After all, to say that one point of intersection on the horizontal line is more prominent than the other is just to say that one, but not the other, point of intersection on that line is attended to by the perceiver. So, to say that in one experience the perceiver senses a given point of intersection as prominent just is to say that in one experience the perceiver senses herself, or her own experience, as attending to that point of intersection. Why is this bad? First, there is a worry about trivializing intentionalism. If intentionalism is to be a substantive thesis, it surely can’t be legitimate to simply slide without further argument from the premise that a subject instantiates a certain phenomenal property to the conclusion that the subject senses herself to be instantiating some mental property corresponding to the phenomenal property for which we were trying to find a representational correlate. This is the sort of built-in response to alleged counterexamples which should make us suspicious. In each case, we should require that the intentionalist should make plausible the idea that the representational property adverted to is a genuine representational property of the experience. In the present case, the idea that we not only attend to aspects of the represented scene but also, just in virtue of so doing, sense ourselves as so attending, has no initial plausibility. Second, we can provide positive arguments that perceptual prominence is not genuinely sensed using the principles of Fallibility, Independence, and AntiCircularity, discussed in Chapter 21. First, perceptual representation of oneself as attending to an object seems to violate Fallibility. What would it be like to have a perceptual experience which represented a feature of the scene as perceptually prominent without that feature being perceptually prominent? This question seems impossible to answer, because it seems clear that if it seems to me that something is perceptually prominent, then, just in virtue of that, it is. We simply don’t have a grip on what perceptual representation of perceptual prominence—as something over and above perceptual prominence itself—could be. The same problems result if we think of perceptual experiences not as representing facts about what the subject is attending to, but as representing facts about what that subject is perceptually representing. Suppose (for reductio) that an experience represented an experience as representing an object as purple but that the experience did not, in fact, represent the object as purple. Would the subject’s phenomenal properties include those characteristic of visually sensing

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

shifts in attention 193 things as purple, or not? If so, then it looks like, contra our supposition, the subject’s visual experience would be representing the relevant object as purple. But if not, then it seems that (again contra our supposition) the subject would not be visually sensing the experience as representing the object as purple. (In virtue of what would it be representing the experience as representing the object as purple, rather than as some other color?) The view under discussion also violates Independence. Suppose for reductio that the proponent of the view under discussion holds that there could be a shift in the subject’s attention which the subject did not sense as such. Then we would have a difference in phenomenal properties corresponding to the shift in attention, because it is hard to see how two experiences could differ in the focus on the subject’s attention without differing phenomenally. (Just try to imagine the two experiences of the line-intersections described earlier, but without any difference in phenomenal character.) But then the phenomenal difference corresponding to this attentional shift would correspond to no difference in sensing properties, which contradicts the intentionalist thesis we’re considering. Hence the intentionalist must admit that there can be no unrepresented shifts in attention, and Independence fails. The view would also violate Anti-Circularity—at least if we’re inclined to identify phenomenal and representational properties. For then the intentionalist would be identifying the phenomenal property corresponding to attention to a point with the perceptual representation of that very property. It’s hard to see what other sort of representational difference the intentionalist might appeal to. So, if we agree that we cannot appeal to perceptual representation of perceptual prominence, then it seems that we have here a genuine example of a subject whose phenomenal properties change over a period of time during which her sensing properties remain constant. One can see in hindsight why defenders of intentionalism4 have overlooked the possibility of this kind of counterexample; in fact, we need only recall the argument from introspectable difference for intentionalism discussed in Chapter 4. Intentionalists are often motivated by the transparency of experience, glossed as the view that the only things available to introspection on perceptual experience are the objects and properties that experience represents as in the environment of the perceiver. From here it seems but a short step to the conclusion that any introspectable difference between experiences—that is, any difference in phenomenology, or phenomenal character—must correspond to some difference in the objects and properties presented as in the perceiver’s 4

Like Speaks (2009).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

194 shifts in attention environment, and so also to a difference in content. What this step misses—what the inference from Positive Transparency to Transparency/Difference misses—is that a difference in phenomenal properties can be generated not just by a change in what is introspected but also by a change in where one’s introspective gaze— that is, one’s attention—is focused. Here we face a dilemma. On the one hand, attentional shifts seem to provide a counterexample to even local intramodal intrapersonal time-restricted intentionalist theses (after all, we’ve been talking about the visual experiences of a single subject, and we’ve relied on no universal claims about phenomenal states in general). On the other hand, we’ve already seen that these intentionalist theses are quite difficult to deny without absurdity. The dilemma, though, is a false one. We can accept the intentionalist theses argued for earlier while accepting that we have a failure of the supervenience of phenomenal properties on sensing properties if we accept attention as a genuine phenomenal relation, distinct from sensing. We might, in other words, just take the argument of this chapter to show, not that intentionalism is false, but that sensing and attending meet the sufficient condition for distinctness of phenomenal relations stated in Chapter 24. This is another place where it’s important to remember that we’re ultimately interested in properties of subjects rather than properties of experiences or mental states. If we don’t keep this in mind, these sorts of attentional shifts look extremely puzzling, since there’s a clear sense in which attention does not add anything to the total representational state of the subject. The subject already, after all, visually represents both points of intersection. If we are focused on the properties of the subject’s experience, it can seem puzzling that just ‘doubling up’ the same content—by attending to a point one already represents—could make a difference to one’s phenomenal properties. But when we keep in mind that phenomenal properties are properties of subjects and hence are to be identified, not with contents, but with relations to those contents, it becomes intelligible how shifts in attention can change one’s phenomenal properties. It is a matter of a change in the content to which one is standing in the phenomenal relation of attention without any change in the content one senses.5 5 This is a good place to tie up a loose end from Chapter 12. One of the versions of Fregeanism about sensed contents considered there held that pairs of experiences which differ in Fregean but not Russellian contents might differ in their associated phenomenal properties. I argued against this thesis (Fregean variance) that we cannot imagine pairs of discriminable phenomenal properties which involve no difference at all in the objects and properties sensed as in one’s environment. But the Fregean might use the material of this chapter to fashion a reply, and say that the attentional shifts described here generate differences in sense (rather than being a second phenomenal relation, as I’ve suggested). The problem which this view faces is basically the problem of vanishing senses

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

shifts in attention 195 Of course, introducing an extra phenomenal relation only aids the intentionalist cause if an intramodal intentionalist thesis is true of attention. But it seems fairly clear that it is. It seems as though any instance of a subject’s consecutive phenomenal properties varying due to an attentional shift will be one in which the contents to which the subject stands in the attention relation also change. So, for example, the change in phenomenal properties in the example discussed at the beginning of this chapter corresponds to a change in the subject’s attention being directed first at one point of intersection, and then at the other. Were this not the case, we could have a kind of psychedelic phenomenology with constant attentional focus—which will be as clearly impossible as our example of psychedelic phenomenology þ constant representation of color properties above. One can then argue, using the arguments against the interpersonal constraint, time constraint, and memory constraint from Chapter 7, to an intramodal intentionalist thesis about attention.6 So far this looks like a reasonably neat way of handling these cases within a generally intentionalist view of phenomenal properties. But there is a way in which I have been making things seem a bit tidier than they in fact are. That is because, if we make the move that I am suggesting—taking attending to be a phenomenal relation distinct from sensing—it is less than perfectly clear whether this is a vindication of intentionalism, or a way of abandoning it. Intentionalism is the view that a subject’s phenomenal properties supervene on her representational properties. But even if we hold that a subject’s phenomenal properties supervene on the facts about what she senses and what she attends to, it is not clear that this is a version of intentionalism, because it is not clear that properties like attending to a particular object or property are really representational properties. As noted in Chapter 2, not all relations in which subjects stand to contents are representational relations. There I made the case that we can give a sufficient condition for a relation to be representational: a relation to a proposition is representational if it is truth-sensitive. That definition was then generalized in Chapter 15 to make room for representational truth-sensitive relations to discussed in Chapter 12, since the relevant differences in sense would not on this view correspond to any differences in sense between beliefs or sentences. One can’t, for example, easily imagine a difference in the beliefs which one of our experiences of looking at the intersecting lines would justify and the other not, or an example of a sentence which would express the content of one but not the other experience. 6 I’m being intentionally non-committal about how, exactly, we should understand the contents of attentional states. One plausible view is that their contents are properties much like (though of course in typical cases less complex than) the properties with which I identify the contents of sensing.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

196 shifts in attention monadic properties which are not propositions. But there is no way to generalize the definition in any principled way which will deliver the verdict that attention relations are truth-sensitive. One simply can’t misrepresent the world by virtue of attending to one perceptually represented property rather than another. So, to decide whether attention relations are really representational relations, it looks like we will need the thing which we could not provide in Chapter 2: necessary as well as sufficient conditions for a representation to be representational.7 It may be that, in the end, the question of whether attention relations are representational is something of a verbal question; the bounds of the category of the representational might not be fixed enough to deliver a determinate verdict. In that case, the question of whether intentionalism is true is also, to that extent, a verbal question. Attention relations clearly involve a kind of directedness toward objects and properties; they have that (admittedly inchoate) description in common with paradigm representational relations. On the other hand, they don’t seem to determine veridicality conditions, truth conditions, or any sort of satisfaction condition—and to that extent seem very unlike paradigm representational relations. Perhaps the best thing to say is that whether or not the boundaries of the representational include attention is, in the end, not the most important thing. Whether or not they do, we can still use attending, alongside sensing, in an attempt to explain the nature of phenomenal properties. Whether that theory is in the end best classified as a kind of intentionalism, or not, seems like a secondary question. What matters more is whether otherwise plausible analyses of representational properties can be extended to handle the attentional properties of subjects. In what follows I’ll continue to describe the view I’m developing as a kind of intentionalism. But that label should be understood to come with the qualification that some of the relations to objects and properties which appear in the supervenience base for phenomenal properties aren’t obviously genuinely representational relations. . . . I’d like to close this chapter by bringing out a consequence of a view which recognizes attending, alongside sensing, as a phenomenal relation: it provides a natural and unified treatment of two cases which would otherwise be quite puzzling for intentionalism.

7 Or at least sufficient conditions which pick out a different class of representational relations than the truth-sensitive ones.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

shifts in attention 197 The first of these is Mach’s example of seeing a box as a square, and then as a diamond. There is a clear difference in phenomenology between the two experiences, but no obvious difference in representational content. The natural first line of response for the intentionalist is to say that some properties of the shape are represented by the experience in which the subject sees it as a square, but not in the experience in which the subject sees it as a diamond, and vice versa; for example, one might say that when the figure is seen as a square one’s visual experience represents a symmetry about the bisectors of the sides of the shape, whereas when the figure is seen as a diamond one’s visual experience represents a symmetry about the bisectors of the angles of the shape.8 However, as Fiona Macpherson has pointed out,9 it is possible to see a box as a square while visually representing the symmetry in the bisectors of the angles of the shape, as is shown by the example of a perceptual experience of the following figure:

This figure can be seen as a square or a diamond, despite the fact that, in both cases, the symmetry about the bisectors of the angles is perceptually represented (thanks to the dotted lines). I suggest that the intentionalist should respond to this case in the same way as she should respond to the example of the intersecting lines: she should say that the difference in phenomenal properties between seeing the box as a square and seeing it as a diamond is not explained by any difference in the sensing properties of subjects, but rather by a difference in the contents to which the subject is standing in the attention relation. When the box is seen as a square, one is attending to symmetries involving the sides, whereas when it is seen as a diamond one is attending to symmetries involving the angles. Against this suggestion, one might argue that one can see the box as a square even while attending to the symmetry in the bisectors of the angles.10 But this seems not to be the case. If we attend to the angle bisector symmetry by attending 9 See Peacocke (1992). See }7 of Macpherson (2006). Macpherson (2006), for example, says that ‘it seems perfectly possible to see a square as a square while focusing intently on its angle bisector symmetry’ (103). 8

10

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

198 shifts in attention to the intersecting dotted lines, this shift in attention generates the Gestalt shift to seeing the figure as a diamond. This is strong evidence that the difference in phenomenology between seeing the box as a square and seeing it as a diamond is, like the difference between the two experiences of the intersecting horizontal and vertical lines above, due to a shift in attention. The second proposed counterexample to intentionalism is due to Bernard Nickel, who asks us to consider two perceptual experiences of a 3 x 3 grid of squares like the following:

which differ only in which groups of squares appear as prominent.11 In one such experience, the corner and center squares appear prominent, and in the other the remaining four ‘side’ squares appear as prominent. There is, as Nickel says, a clear difference in phenomenal properties here, and Nickel argues convincingly against a number of different attempts to find a difference in sensed content. Intuitively, though, it seems that we have the same phenomenon here as in the case of the intersecting lines and the case of the square/diamond: the relevant difference in phenomenal properties is explained by a shift in attention from one group of boxes to the other. So, as before, if we can think of attentional states as involving a phenomenal relation to contents other than sensing, this sort of example need not worry the intentionalist who is willing to make the distinction between sensing and attentional states part of the supervenience base for facts about phenomenal properties. One has to be careful here; we could always respond to counterexamples to intentionalism by simply multiplying phenomenal relations without end. Is appeal to the distinctness of sensing and attending cheating in this way? We don’t want the appeal to the phenomenology of attention to be an unprincipled way for the intentionalist to simply relabel any proposed counterexample as a mere shift in the phenomenology of attention without any change in sensing properties.

11

Nickel (2006), 284.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

shifts in attention 199 I think that there are two responses to this worry, one intuitive and one a bit more theoretically loaded. The intuitive point is just that each of the three cases really do seem to essentially involve attention. It is difficult even to describe the initial example of the intersecting lines without describing it as a shift in attention or focus. The same is true of the two experiences of the box pictured earlier with dotted lines marking bisectors of the shape’s angles; to generate the two different experiences, one directs a subject ‘to change his focus of attention’ from the dotted lines to the symmetry of the sides, and back. Analogous remarks apply to the 3 x 3 grid. It is also telling that in each case it is natural to describe the two experiences by talking about what aspects of the relevant figure are ‘prominent.’ By contrast, it would not be natural to extend this model to other putative counterexamples to intentionalism.12 One clue that this is the correct response to these cases is the isolated nature of the relevant phenomenal differences. There is a sense in which these kinds of shifts in attention are not part of specifically visual phenomenology at all: similar cases can easily be generated for any of the other sense modalities, or for bodily sensations. (Imagine listening to a duet in an otherwise silent environment while shifting your focus of attention from one voice to the other, or shifting your attention between your toothache and the itch in your foot.) This suggests that it is not unreasonable for the intentionalist to respond to these cases not by trying to find some differences in contents sensed, but instead in terms of the subject’s attentional properties. The second line of response involves the voluntariness of at least some attentional shifts. Typically, changes in, for example, visual phenomenal properties cannot be varied about at will; one cannot go from being RED to being GREEN simply by deciding to do so. But in each of the cases above, we can effect the relevant phenomenal change without bringing about any such ‘external’ change in the scene perceived; this makes it more plausible that these changes are due to attentional shifts since, in at least many cases, we can shift our attention from one element of a represented scene to another at will.13 So there is a reason to believe that a principled line can be drawn between perceptual phenomenology and the phenomenology of attention.14 If so, perhaps 12

For example, the phenomenal differences between visual experiences of trees of the same height located at different distances from the perceiving subject—discussed in Peacocke (1992)— could hardly be described as just a shift in attention or focus. 13 This is not to say that there is no involuntary aspect to these phenomenal changes; as Macpherson (2006), }4 argues, there is. The point is just that the relevant phenomenal shifts can, in the standard case, be initiated at will without any change in stimulus from the environment. 14 That being said, the idea that sensing and attending are distinct phenomenal relations does give rise to a puzzle which was pointed out to me by Casey O’Callaghan. One would think that, if sensing and attending were genuinely distinct relations to things, we might sometimes be unsure

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

200 shifts in attention attentional shifts are not counterexamples to intentionalism as such; but they are counterexamples to an intentionalism which places only representational properties involving the single phenomenal relation of sensing in the supervenience base for phenomenal properties.

whether some object of our attention is or is not identical to some perceptually represented object. But we never seem to get ‘Frege problems’ of this sort.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART VII

Phenomenal Identity and Indiscriminability

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

28 Identifying Phenomenal and Representational Properties My aim so far has been to arrive at views on two topics: the nature of representational properties involving phenomenal relations, and the modal relationship between these representational properties and phenomenal properties. On the first topic, my aim has been to defend the view that the contents to which we stand in phenomenal relations are Russellian contents, which turn out to be a certain sort of complex property, and that phenomenal relations involve a kind of self-ascription of those properties. On the second topic, we’ve arrived at the view that the strongest true intentionalist thesis is an intramodal, interpersonal intentionalism, which consists of the following two supervenience claims (as earlier, I use ‘sensing’ as a general term for the single propositional attitude associated with perceptual experience and bodily sensation): Necessarily, for any two subjects, if both subjects are sensing the same content, then the phenomenal properties associated with those acts of sensing are the same. Necessarily, for any two subjects, if both subjects are attending to the same objects and properties, then the phenomenal properties associated with those attentional states are the same. and, more tentatively, the further claim: Necessarily, for any two subjects, if both subjects are sensing that p and both subjects are attending to the same objects & properties, the phenomenal properties of the two subjects are the same (what it is like to be one subject is just what it is like to be the other subject). This claim is more tentative for two reasons. First, while the argument of Chapter 9 provides some support for a global intentionalism, and the arguments of Chapters 25 and 26 suggest that all perceptual experiences and bodily

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

204 identifying phenomenal and representational properties sensations involve a single representational relation of sensing, I haven’t explicitly considered sources of phenomenal difference which at least seem to be distinct both from perceptual modalities and from bodily sensations—for example, moods. It’s not totally clear that the style of argument for intentionalism of Chapters 4–8 has application there. Second, and relatedly, while I have argued that sensing and attending should be regarded as distinct phenomenal relations, and hence that there are at least two phenomenal relations, I have given no argument that there are not more than two phenomenal relations. As mentioned in Chapter 27, one might argue that phenomenal properties associated with occurrent thought or imagination force us, via the sufficient condition for distinctness of phenomenal relations in Chapter 24, to add thinking or imagining to our catalogue of phenomenal relations. I’m setting this view to the side here, not out of skepticism about these alleged phenomenal properties, but because if there were more than two phenomenal relations that would make the arguments in the chapters which follow easier rather than harder to make. Indeed, for the purposes of the next few chapters we can also set aside the complications raised by attention in the last chapter—though the importance of the existence of multiple phenomenal relations will resurface in Chapter 36. These claims give us a supervenience base for phenomenal properties. The rest of this book will be concerned with the question: Given these supervenience theses, and given the theses about sensed content advanced above, what could phenomenal properties be? One might just say that RED is what it is, and not another thing. This is certainly true; but it would be disappointing if this is all we could say about the nature of phenomenal properties. A more informative answer to our question—which has, in different forms, been defended by many philosophers in the last few decades—is suggested by our supervenience claims. If it is impossible for there to be a change in phenomenal properties without a corresponding change in representational properties, it would be nice to know why this is so. And a plausible explanation of why this is so is that phenomenal properties just are representational properties: what it is for a subject at a time to instantiate a certain phenomenal property just is for that subject to be representing the world in a certain way. In addition to providing a neat explanation of our supervenience claims, this sort of ‘identity claim’ explanation of the truth of intentionalism has also been thought by many to be a key step in the program of naturalizing the mind. If phenomenal properties just are certain kinds of representational properties, then, if we can provide a theory of what it is for a subject to have a representational property of the right kind—a task which has seemed to many to be easier than providing a theory of phenomenal properties—we will have explained the nature

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

identifying phenomenal and representational properties 205 of the phenomenal itself, and therefore gone a long way toward solving the traditional mind-body problem. But if this sort of identity claim promises explanatory gains, it also carries with it an important commitment. Because identities are symmetric and necessary if true, if some identity claim of this sort is true, it must be the case not just that phenomenal properties supervene on representational properties, but also that the relevant representational properties supervene on the phenomenal properties. But is any such thesis true? We’ve already seen one reason to doubt that it is. In Chapters 18 and 20, I argued that the contents of (veridical) perceptual experiences are Russellian contents which sometimes have objects and natural kinds among their constituents. But there can be perceptually indiscriminable objects and perceptually indiscriminable natural kinds. Consider two such objects—like our example of two new golf balls. A subject who senses one, it seems, needn’t differ in phenomenal properties from a subject who senses the other—after all, an experience of one would be indiscriminable from an experience of the other (so long, of course, as we hold fixed the lighting and other features of the experience). But if sensed contents can have objects as constituents, the two subjects will differ in their sensing properties, which seems to give us an example of a pair of subjects which differ in their sensing properties but not their phenomenal properties. And from this it follows that the representational does not supervene on the phenomenal, which rules out any identification of the representational with the phenomenal. Setting aside for now the complications introduced by attention, there are, broadly speaking, three different replies to this problem. The first two begin with the thought that, because indiscriminable phenomenal properties are always identical, the example of the golf balls shows that it is possible for a pair of subjects with the same phenomenal properties to differ in their sensing properties. One might then have two different responses to this situation. The first is quite simple: Distinctness Phenomenal properties are distinct from any representational property. The second tries to reconcile the claim that the golf ball experiences generate the same phenomenal properties with the claim that phenomenal properties are identical to representational ones: The Phenomenal Content Thesis Subjects of indiscriminable phenomenal properties who differ in their sensing properties still have some part of the content of their sensing properties

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

206 identifying phenomenal and representational properties in common, and that we can use this ‘part of its content’—intuitively, the part of content which does supervene on phenomenal properties—to formulate a true identity claim between phenomenal and representational properties. A third response is to deny that the example of the golf balls shows that subjects can have the same phenomenal properties but differ in their sensing properties, by denying the assumption that indiscriminable phenomenal properties must be identical: Simple Identity Phenomenal properties are identical to sensing properties; hence subjects are alike with respect to their phenomenal properties iff they are alike with respect to their sensing properties. The fact that two subjects’ sensing properties can differ only in which external particular they represent shows that there can be distinct but indiscriminable phenomenal properties. I have little to say about Distinctness, other than to say that it seems to me that it should be a position of last resort. If there really is a necessary connection between phenomenal and representational properties of the sort we have been discussing—such that it is metaphysically impossible that two subjects differ in their phenomenal properties without also differing in the relevant representational properties—this would be utterly mysterious if the two sets of properties were genuinely distinct. Why, in that case, couldn’t we vary the former properties without varying the latter? This is of course just to echo the standard rhetoric used to press the point that supervenience theses, like those sketched at the beginning of this chapter, need some explanation. While this seems very plausible to me, I confess that I would not know how to argue against someone who simply denied this explanatory claim.1 In what follows, my strategy will be, in the next two chapters, to consider some problems that arise for the Phenomenal Content

1

One might also wonder how generally this line can be pushed. To mention a topic treated in the closing chapters of this book, I’m attracted to a view on which phenomenal and representational properties are identical, but cannot be identified with any physical property of subjects. But despite this, I’m somewhat inclined to think that possession of at least some phenomenal properties supervenes on the intrinsic physical properties of subjects. But this combination of views is in obvious tension with the idea that supervenience theses need explanation in terms of identity claims.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

identifying phenomenal and representational properties 207 Thesis and Simple Identity. I think that at this stage a reasonable attitude would be to prefer both of them to Distinctness, if either is otherwise plausible. One can think of the chapters to follow as an exploration of the relative merits of these two different ways of identifying phenomenal and representational properties.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

29 Phenomenal Content A good place to begin our discussion of phenomenal content is with its motivations. Why would we want to single out some proper part of the content of experience as its phenomenal content? One important motivation is adherence to something like the following principle: Distinctness/Discriminability Necessarily, for any phenomenal properties F, G such that F6¼G, if A instantiates F and B instantiates G, then what it’s like to be A is discriminable from what it’s like to be B1 If one did not think of phenomenal properties as satisfying Distinctness/Discriminability, then there would be no reason to look for a special, restricted sort of content which experiences of our indistinguishable golf balls have in common: one could say (as the proponent of Simple Identity does) that phenomenal properties are identical to the sorts of representational properties discussed earlier, which include perceptually indiscriminable external objects, like our golf balls, as constituents.2 This is objectionable only if we think that any pair of subjects such that what it is like to be one subject is indiscriminable from what

1 Distinctness/Discriminability, so understood, is related to the view that phenomenal properties are, in Williamson’s sense, luminous: that is, properties of subjects which are such that instantiating that property is sufficient to be in a position to know that one is instantiating that property. However, there are differences: phenomenal properties might be luminous but still fail to satisfy the above condition if, as seems plausible, one can be in a position to know that x is F and know that y is F without being in a position to know whether x and y are both F. This distinction is important in the present context because one might defend the claim that one can always know which phenomenal property one is instantiating by using a demonstrative of the sort discussed in Chapter 17 to form the belief that I am instantiating this property. But one can use the same demonstrative expression on distinct occasions to refer to the same thing without knowing whether one has referred to the same thing on those occasions. 2 Unless, that is, one is attached to an internalist thesis about phenomenal properties. More on this in Chapter 37.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content 209 it is like to be the other subject must be instantiating the same phenomenal property. To make sense of Distinctness/Discriminability, and hence the notion of phenomenal content, we need to say a bit about what ‘discriminable’ means here. Following Timothy Williamson, I think of discriminating between a and b as ‘activat[ing] the knowledge that a and b are distinct.’3 Corresponding to different kinds, and sources, of knowledge are differing kinds of discrimination, and—given that discriminability is the ability to discriminate—different kinds of discriminability. What kind of discriminability is relevant to the above principle? A paradigm example of phenomenal properties which are supposed to be indiscriminable in the relevant sense are the ones corresponding to the two golf ball experiences described earlier; it’s useful to think about that example in order to rule out some kinds of discrimination which are clearly not relevant. Suppose that the subject of the two golf ball experiences were to reason as follows: I know on the basis of testimony that the experiences were of different golf balls; and I know on the basis of philosophical argument that this entails that they differ in content. But I also know that phenomenal properties are identical to representational properties; hence the phenomenal properties I instantiated during those two experiences must be distinct. I am happy to concede that—if the premises of this informal argument are true— this line of argument shows that what it was like to be that subject during one golf ball experience is discriminable, in one sense of discriminable, from what it was like to be that subject during the other experience. But this is plainly not the sense of ‘discriminable’ relevant to proponents of phenomenal content. So then what is that sense? The basic idea is something like this: a pair of phenomenal properties is discriminable in the relevant sense iff a subject could come to know that those phenomenal properties were distinct solely on the basis of introspection and memory. That is: For any phenomenal properties F and G, F is discriminable from G iff possibly, a subject instantiates F at t1 and instantiates G at t2 and at t2 is able to know that F6¼G solely on the basis of (i) introspection of her experience at t2 and (ii) memory of her introspection of her experience at t1.4 3

Williamson (1990), 7. This would (trivially) lead to incorrect results if there were pairs of phenomenal properties which were are not possibly instantiated by the same individual, or if there were phenomenal 4

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

210 phenomenal content It is worth noting that, given the fallibility of memory and introspection, we can’t simply infer from the fact that a subject on one occasion is not able to tell the difference between what it is like to be her at that time and at some earlier time that the two experiences were genuinely indiscriminable. Failure of discrimination on one occasion does not entail indiscriminability.5 This is a point to which I will return later. The above remarks about discriminability put us in a position to state a bit more clearly the Phenomenal Content Thesis stated informally in the previous chapter. In doing so, we need to remember that our interest in the question of whether phenomenal and representational properties could be identical further constrains the sort of ‘phenomenal content’ thesis we are interested in. For, in looking for a kind of representational property that supervenes on phenomenal properties, we have to remember that the wanted identity thesis requires us to preserve supervenenience in the other direction as well. This gives us the following formulation: Phenomenal Content Thesis There is a class of representational properties R such that, (i) necessarily, A and B are alike with respect to their R-properties iff what it is like to be A is indiscriminable from what it is like to be B and (ii) every phenomenal property is identical to some R-property. The Phenomenal Content Thesis is equivalent to the conjunction of three claims: the claim that phenomenal properties are identical to some representational properties; Distinctness/Discriminability; and the (uncontroversial) claim that discriminable phenomenal properties are distinct. For purposes of thinking about this thesis, it will be harmless to focus on clause (i) of the definition—(ii) is needed only to distinguish the Phenomenal Content Thesis from the (presumably less than attractive) view that, despite the fact that phenomenal properties supervene on the R-properties and the R-properties supervene on the representational properties, phenomenal properties are distinct from R-properties.6 properties which were for some reason impossible to introspect or remember. We won’t need any examples of that sort, so I will set this kind of worry to the side. 5 This point is emphasized by Williamson (1990). See the discussion of presentation-sensitivity in }4.2. 6 It’s important to be clear that the Phenomenal Content Thesis is a regimentation of the term ‘phenomenal content’ which fits some, but not all, uses of that term in the literature. The point of this regimentation is to clearly distinguish between the two strategies for identifying representational and phenomenal properties discussed at the conclusion of the previous chapter. Given that we are reserving the term ‘phenomenal content’ for representational properties which satisfy the Phenomenal Content Thesis, it is clear that some popular arguments for the existence of

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content 211 Before moving on to consider how the proponent of the Phenomenal Content Thesis should think about some cases which provide challenges for the view, it is worth adding a note on the dialectical situation. In the end, I am inclined to deny Distinctness/Distinguishability. So I will not be too worried by objections about whether the above-defined notion of discriminability is clear enough for us to be able to do any real work with it; if it isn’t, then so much the worse for the Phenomenal Content Thesis. That said, I do think that we have a grip on the notion of discriminability at play here, and I think that there are powerful intuitions which support Distinctness/Distinguishability. When people say things like ‘Our notion of phenomenal character seems essentially tied to our notions of appearing the same as, or being perceptually indistinguishable from . . . this ought to strike one as a conceptual truth.’7

I don’t agree—but I do see where they are coming from, and I think that it’s an important question whether this sort of view of phenomenal properties can, in conjunction with the constellation of other views defended so far, be sustained. As noted earlier, one case which seems challenging for the proponent of the Phenomenal Content Thesis is our example of the golf ball experiences. We have an argument that the subjects of those experiences instantiate different sensing properties; but, given Distinctness/Discriminability, their phenomenal properties must be identical. Hence the Phenomenal Content Thesis requires us to find some representational properties which the two subjects have in common which are also rich enough to provide a supervenience base for phenomenal properties. What could these be? A natural first thought is to set aside the singular propositions involving the distinct golf balls, and focus instead on the corresponding existentially quantified propositions. On this sort of view, we can identify phenomenal properties with representational properties like the property of sensing that there is something or other which is white, round, etc. This seems to give us what we want, since the

phenomenal content won’t, in our sense, establish the existence of phenomenal content. An important example is the plausible argument, defended by, among others, Siewert (1998), that there is a kind of content which is graspable just in virtue of attention to the phenomenal character of one’s experience. This sort of argument may show that there are representational properties which supervene on phenomenal properties; but it does not show that those representational properties are substantial enough to support the reverse direction of supervenience, and hence does not show that the Phenomenal Content thesis is true. This is of course no objection to Siewert’s argument; just a reminder of the significance of being clear about just what phenomenal content must be in order to secure the wanted identity of representational and phenomenal properties while holding fixed Distinctness/Discriminability. 7

Deutsch (2005), 10.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

212 phenomenal content experiences of these two golf balls do seem to be alike in content so long as we ignore their singular component. This sort of maneuver suggests a general strategy for friends of phenomenal content: to obtain the phenomenal content of an experience, replace every singular content with the corresponding existentially quantified general content. This strategy, however, has a weakness: it commits us to the claim that we stand in sensing relations to these sorts of existential contents. We already know from the argument of Chapter 18 that we sometimes sense a particular object o to have a property F. The present proposal would require that, whenever we do this, we also sense the content that something is F. But it is very hard to see why this should be so. It would be explained by the sensing relation being closed under entailment, so that anyone who senses p also, for any q such that p entails q, also senses q—but this sort of closure principle is, for reasons familiar from the discussion of belief, wildly implausible.8 Given that sensing seems (as noted in Chapter 25) to distribute over conjunction, this sort of closure principle would entail that anyone who stands in the sensing relation to any content stands in that relation to every necessary truth. But surely this is not right. And absent the truth of some such closure principle, it’s hard to see why it would not be possible to stand in the sensing relation to a singular content without standing in that relation to the corresponding existential content. One might reply to this argument by conceding this point about sensing, and defining a different relation between subjects and contents, sensing*, as follows: A subject senses* that something is F iff for some object o, the subject senses that o is F This is surely a perfectly good definition of a relation between subjects and existential contents. Given this definition, the phenomenal content theorist who wants to make use of existentially quantified contents might then just identify phenomenal properties with sensing* properties. The problem with this move is that sensing* does not seem to be a truthsensitive representational relation. For suppose that I have a hallucination of a white golf ball, and hence sense* that there is something which is white, round, etc. Now suppose that there is something which is white, round, etc., but that this thing existed on the other side of the world 100 years ago. This would make the content to which I stand in the sensing* relation true, but would not make it the

8

See Soames (1987).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content 213 case that I am correctly representing the world. Hence sensing* is not a truthsensitive relation.9 The proponent of sensing* might reply that this objection underestimates the complexity of the contents to which we stand in the sensing, and hence also sensing*, relations. For surely when I have the golf ball experience, I don’t just represent tenselessly the existence of some white, round thing in some place—I represent the existence of such a thing at some particular time t and location l. Hence, so the reply continues, the presence of a white, round, etc. thing 100 years ago on the other side of the world would not make my sensed* content true. But introducing particular times and locations into sensed* contents reintroduces the problem which the sensing* relation was supposed to solve in the first place.10 Just as there can be perceptually indiscriminable golf balls, there can be indiscriminable times and locations; hence, if sensed* contents include particular times and locations, we will have subjects who differ in their sensing* properties but instantiate indiscriminable phenomenal properties. This, in the presence of Distinctness/Discriminability, entails that phenomenal properties can’t be identified with sensing* properties. In the end, though, this needn’t worry the proponent of phenomenal content, because she is not forced into the view that we stand in a representational relation to these sorts of existential contents. This is because there is no reason for the proponent of phenomenal content to identify the representational properties mentioned in the Phenomenal Content Thesis with relations to contents. To say why, it will be useful for the moment to abstract from the details of the account of perceptual contents given in Chapter 15, and pretend that the contents of perceptual experiences are propositions rather than non-propositional monadic properties. Now consider again our experiences of the golf balls. Simplifying (and ignoring for now questions about the visual representation of size and location), let’s say that these have as contents the propositions expressed by b1 is white & round & of such-and-such size & . . . b2 is white & round & of such-and-such size & . . . We need to identify some representational property in common between these phenomenally identical experiences. But to do this we needn’t say that each subject senses the further content

9 There is room in logical space for the view that sensing* is representational, but not truthsensitive. But that would be a bit surprising, given the close connection between sensing and sensing* and the fact that sensing is truth-sensitive. 10 This point is made in Schroeder and Caplan (2007).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

214 phenomenal content ∃x (x is white & round & of such-and-such size & . . . ) Instead, we can just point out that both experiences have as their content some value of the propositional function corresponding to the open sentence x is white & round & of such-and-such size & . . . This propositional function is not, of course, a proposition; but this shouldn’t matter. The point of phenomenal content is to find some representational property of subjects with which phenomenal properties can be identified—not some content with which phenomenal character, thought of as fundamentally a property of experiences, might be identified. That representational property of subjects might be: the property of being such that there is some value of the relevant propositional function such that the subject stands in the relation of sensing to it.11 There’s no need to, in addition, require that these subjects bear the attitude of sensing to the proposition that this propositional function is true of something. Having presented the view in terms of propositions and propositional functions, it’s easily modified to fit the view that the contents of experience are selfascribed non-propositional monadic properties. The view of the phenomenal content theorist would then be that phenomenal properties are identical to the property of sensing some value of the function from objects to the relevant class of non-propositional monadic properties—some value, we might say, of the relevant ‘content function.’12 The golf ball example was apparently problematic for the Phenomenal Content Thesis because it illustrates the fact that subjects can instantiate indiscriminable phenomenal properties while differing in which objects they sense to have certain properties. But it isn’t just that this can sometimes happen with sensing of objects; rather, it seems to be a general feature of sensing that objects, like individual golf balls, are in the following sense ‘phenomenally silent’:

11 A related view is developed, but rejected, in Tye (2009), }5.5. Tye’s main objection to the view depends on the argument from transparency criticized in the first few pages of Chapter 4. 12 One subtlety which I will mention but then set to the side stems from the much-discussed fact that hallucinations can be indiscriminable from veridical experiences and illusions. This requires us to think of ‘values of content functions’ broadly enough that a gappy content, in which no object is supplied in place of the deleted constituent, counts as a value of the relevant function. Otherwise we would be able to find no representational property in common between a subject who is sensing that o is red and a subject having an indiscriminable hallucination—and this would be enough to falsify Distinctness/Discriminability. This is an aspect of the phenomenal content theorist’s picture which would have to be worked out in more detail; here I’m going to just suppose that some strategy along these lines will prove satisfactory.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content 215 A class C of constituents of sensed contents is phenomenally silent iff for any x, y ∈ C, for any two subjects, if the only representational difference between those subjects is that one senses a content with x as a constituent and the other senses a content with y as a constituent, then the two subjects instantiate indiscriminable phenomenal properties. This definition suggests a strategy for the defender of the Phenomenal Content Thesis, which we can call the strategy of constituent deletion. This strategy says that, whenever confronted with a subject who stands in the sensing relation to a content which contains some phenomenally silent constituent c, the proponent of the Phenomenal Content Thesis should identify the relevant phenomenal property, not with the property of standing in the relation of sensing to a content, but rather with the property of, for some value of the relevant content function, standing in the sensing relation to it. This appears to be a perfectly general way of dealing with phenomenally silent contents, which would otherwise provide counterexamples to the conjunction of Distinctness/Discriminability with the identity of representational and phenomenal properties. It is plausible that the phenomenal content theorist should also employ the strategy of constituent deletion when it comes to the perceptual representation of natural kind properties, because kind properties, like individual objects, seem to be phenomenally silent. It is difficult to imagine a pair of subjects who differ representationally only with respect to substitution of distinct kinds in sensed contents who nonetheless instantiate discriminable phenomenal properties. And so the proponent of the Phenomenal Content Thesis can handle these experiences in just the same way as the pair of golf ball experiences described earlier: she can say that the representational property with which the relevant phenomenal property should be identified is the property of standing in the relation of sensing to some value of the content function corresponding to the open sentence obtained by replacing a term for the relevant natural kind with a free variable. Let’s say that any constituent of a content which is not phenomenally silent is ‘phenomenally loud.’ Then the present suggestion is, in effect, that phenomenal properties are identical to properties of sensing some value of content functions which themselves have only phenomenally loud constituents. There’s a lot to be said for this view. For one thing, it allows us to preserve the identity of representational and phenomenal properties and a plausible externalism about sensing properties while holding on to the thesis, which many take to be quite plausible, that a subject’s phenomenal properties supervene on her intrinsic properties. Many have thought that these theses are jointly inconsistent; the key to finding a way to consistently hold all three is to (i) as above, obtain the relevant

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

216 phenomenal content representational properties via constituent deletion of phenomenally silent contents and (ii) hold that contents are, if externalist, then phenomenally silent.13 And a strong case can be made that this distinction between types of sensed constituents—externalist þ phenomenally silent on the one hand, and internalist þ phenomenally ‘loud’ on the other—is exhaustive. In fact, this claim gets some support from some observations about hallucination—already mentioned in Chapter 21—which are due to Mark Johnston. Johnston pointed out the following asymmetry between veridical and hallucinatory experience: there are some contents with which one can become originally acquainted in either hallucinatory or veridical experience, and others with which one can become originally acquainted only in veridical experience. In the terminology introduced earlier, some contents are hallucination-available whereas others are not. Why should this fact matter for the Phenomenal Content Thesis? For our purposes, two aspects of hallucinatory experience are especially salient: In hallucinations, the standard external accompaniments to having a perceptual experience involving a given phenomenal property are absent. For any given veridical experience, there is a corresponding indiscriminable hallucination. The first aspect of hallucination should lead us to expect that, if there really is a distinction between externalist and internalist sensed contents, internalist but not externalist contents should be hallucination-available. The second aspect of hallucination should lead us to expect that, if there really is a distinction between phenomenally silent and phenomenally loud contents, that all phenomenally loud contents should be hallucination-available. And this, strikingly, is exactly what we find. Sensible qualities like colors are the paradigm case of hallucination-available contents. Johnston illustrates the point with the example of super-saturated red: ‘the paradigm red—the reddest of the reds—can only be presented in delusive experience. One can come to know what ‘supersaturated’ red is like only by afterimaging it. While one is afterimaging it, one could compare how much more saturated it is than the reds exhibited by the reddest of the standard Munsell color chips, there before one on the table. Likewise, a painter might discover in hallucination a strange, alluring color, which he then produces samples of by mixing paints in a novel way. Here we have all the signs of de re knowledge of quality.’14

13

Here a content c is externalist iff there could be a pair of intrinsic duplicates who are such that one senses a content of which c is a constituent, and the other does not. 14 Johnston (2004), 130.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content 217 Here we have a hallucinatory experience which contains a given determinate color property—and, as a result of that experience, we are put in a position to have thoughts involving that property which we were, prior to the hallucination, not able to have. But, as Johnston notices, there is a sharp contrast here with hallucinations of objects and natural kinds: ‘I can hallucinate my mother talking to me on the phone, but I could not do this unless I already had an independent way of making singular reference to my mother. If I had been abandoned to the monks at birth and knew nothing of my mother or of mothers in general, then I could not hallucinate my mother talking to me. Even if I hallucinated a woman who happened to look just like my mother, there would be nothing that would make that hallucination of my mother, as opposed to my aunt, or any other woman who appeared like her.’15

It is not difficult to make a parallel case for natural kinds. This strikes me as a strong argument in favor of the distinctions on which the present version of the Phenomenal Content Thesis relies. Given that hallucinations can be had in the absence of the usual connections to the external world, if there are externalist contents, we should expect these to be hallucination-unavailable. We have seen that we can give strong arguments that objects and natural kinds are externalist, and these are exactly the contents which turn out to be hallucination-unavailable. The fact that hallucinations can be phenomenally indiscriminable from veridical experiences leads us to expect phenomenally loud contents to be hallucination-available, and phenomenally silent contents to be hallucination-unavailable. Colors and shapes are paradigm cases of phenomenally loud contents, and these turn out to be hallucinationavailable, whereas external particulars turn out to be both phenomenally silent and hallucination-unavailable. One wonders whether it could really be an accident that when we look at these three distinctions between possible contents of perception— externalist vs. internalist phenomenally silent vs. phenomenally loud hallucination-unavailable vs. hallucination-available they all just happen to draw the line in just the same place. So far, then, smooth sailing for the Phenomenal Content Thesis. How might one go about arguing against this thesis? The basic idea behind any such argument has to be that there can be a pair of indiscriminable phenomenal

15

Johnston (2004), 129.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

218 phenomenal content properties corresponding to sensing properties which differ with respect to the representation of members of a class C of properties, despite the fact that members of C are not, in the above sense, phenomenally silent. Let’s call properties of this sort phenomenally sneaky. More formally: A class C of properties is phenomenally sneaky iff (i) there can be a pair of subjects who instantiate indiscriminable phenomenal properties and who differ with respect to which C-properties they sense, and (ii) there can be a pair of subjects who instantiate discriminable phenomenal properties and are alike representationally except with respect to which C-properties they sense.16 If the C-properties satisfy (i), then representation of C-properties cannot be a part of the representational properties mentioned in the Phenomenal Content Thesis, since, if they were, a pair of phenomenal properties of the sort guaranteed by (i) would falsify Distinctness/Discriminability, and hence the Phenomenal Content Thesis. But if the C-properties satisfy (ii), representation of the C-properties must be a part of those representational properties, since otherwise a pair of phenomenal properties of the sort guaranteed by (ii) would be both discriminable and identical to the same representational property—which again would falsify the Phenomenal Content thesis. That is why phenomenally sneaky contents can’t be handled by the strategy of constituent deletion. So to argue against the Phenomenal Content Thesis, it suffices to argue that some properties are phenomenally sneaky. The next three chapters consider three attempts to do just that.

16

This is slightly oversimplified: strictly, we need also to add the condition

(iii)

no sub-class of the C-properties is phenomenally silent.

For without this, we could trivially come up with a class of phenomenally sneaky properties by taking the union of some class of phenomenally silent and some class of phenomenally loud properties. Since I won’t be considering any gerrymandered classes of this sort, I’ll stick with the simpler version in the text going forward.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

30 Phenomenal Content and Phenomenal Continua The first of these comes from much-discussed examples of phenomenal continua. Imagine, to take the standard example, that we have a series of color patches such that each color patch in the series is perceptually indiscriminable from the next, but that the first is discriminable from the last. If this relation holds between the color patches as experienced by some subject, then it seems that an analogous relation could hold between the phenomenal properties of that subject: each phenomenal property in the series is indiscriminable (in the sense given to this term in the last chapter) from the next, but the first is discriminable from the last. Let’s call any such series of phenomenal properties a phenomenal continuum. If there are phenomenal continua, then colors are phenomenally sneaky. For, given the sort of intentionalist thesis argued for earlier, the first and last experiences in the series must involve the sensing of distinct color properties. But sameness in the representation of color properties is transitive: if a subject at t1 represents the same color property to be instantiated as at t2, and analogously for t2 and t3, it follows that the subject represents the same color property to be instantiated at t1 and at t3. Hence it follows that if the first and last experiences in the series differ in their representation of color properties, so must at least two consecutive experiences in the series. But by the description of phenomenal continua these consecutively instantiated phenomenal properties will be indiscriminable from each other; hence color properties satisfy clause (i) of the definition of phenomenally sneaky properties. But they plainly also satisfy clause (ii), since subjects could be alike representationally but for their sensing of the color of one or more objects, and yet instantiate discriminable phenomenal properties. Hence if there are phenomenal continua, then colors are phenomenally sneaky, and the Phenomenal Content Thesis is false.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

220 phenomenal content and phenomenal continua One might try to block the argument that color properties satisfy clause (i) by handling determinate colors in something like the way I suggested the proponent of the Phenomenal Content Thesis should handle phenomenally silent contents. Perhaps we should say that only relatively less determinate colors can be part of the properties representation of which is identified with phenomenal properties. One could say this even if color experience is of more determinate colors, since we could identify phenomenal properties with the property of sensing some color in a specified range. For reasons familiar from the discussion of propositional functions and the strategy of constituent deletion, this is consistent with denying that color experiences ever simply represent surfaces as falling with in this range, and instead always represent them as having some particular color in the range. But this won’t work. On pain of falsifying intentionalism, any two determinate colors sensings of which give rise to discriminable phenomenal properties must be assigned to distinct ranges. Let these two colors be the ones represented by the first and last experiences in the phenomenal continuum described earlier. Then if we consider the colors represented by the intervening experiences, there must be two ‘adjacent’ colors which are mapped onto distinct relatively less determinate color properties; which is enough to show that our relatively less determinate color properties will also satisfy clause (i) of the definition of phenomenal sneakiness. Though running the argument from the possibility of a phenomenal continuum to the falsity of the Phenomenal Content Thesis via the phenomenal sneakiness of colors will be convenient for reasons which will become clear later, one could also establish the same result by arguing that phenomenal continua show the falsity of Distinctness/Discriminability, which is entailed by the Phenomenal Content Thesis. For the first experience in the series must involve a different phenomenal property than the last. And if the subject instantiates a different phenomenal property at the beginning of the series than at the end, then there must be a change in phenomenal properties at some point in the series, which means that there must be a pair of consecutive moments in the series at which the subject changes from one phenomenal property to another. But since what it’s like to be the subject at each moment in the series is indiscriminable from what it is like to be that subject at the next moment, this entails the falsity of Distinctness/Discriminability. It is tempting to try to escape this second version of the argument by saying that it shows not that Distinctness/Discriminability is false, but just that ‘sameness of phenomenal character is intransitive.’ But it is hard to see how this could be a satisfactory stopping point. Talk about the ‘phenomenal character of experience’ is just a way of talking about the phenomenal properties of subjects. And if

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content and phenomenal continua 221 there are such things as phenomenal properties, sameness of these properties just can’t be intransitive. Indeed, it’s a bit hard to see what this would mean, if not the incoherent claim that there are times t1, t2, t3 such that A is alike in being F at both t1 and t2, alike in being F at both t2 and t3, but not F at both t1 and t3. Hence, if it’s to be coherent, denying the transitivity of ‘sameness of phenomenal character’ must be a way of denying that there are such things as phenomenal properties of subjects. But this, for the reasons mentioned in Chapter 1, is hard to swallow. Surely if I think about what it’s like to be me right now, I’m thinking about a way that I am right now—and that’s just what it is for something to be a property of mine.1 Anyway, it does not matter much which way we run the argument. Either way, the conclusion is that if there are phenomenal continua, then the Phenomenal Content Thesis is false. So: are there phenomenal continua? Or, more to the point: could there be phenomenal continua? One might try to argue that there could be, on the grounds that our abilities to discriminate between colors must have limits. Surely, the argument goes, there must be some amount of difference d between colors which is such that our visual system cannot ‘register’ differences of less than d. But then it might seem that to construct a phenomenal continuum, all we need to do is construct a series of color patches, each of which differs from the next by less than d, but which is long enough that the first differs from the last by well more than d.2 Wouldn’t a series of experiences of these color patches be a phenomenal continuum? Not obviously. What does it mean to say that our visual system cannot register differences of less than d? First, it might mean that we can never have a pair of experiences which veridically represent as distinct colors which differ by less than d. While this seems like a plausible claim about our capacities of visual representation, there’s no obvious way to derive from it the possibility of a phenomenal continuum. On this interpretation, after all, we couldn’t have a series of veridical experiences of the color patches just described, since we can only represent differences in color greater than d. And then what’s the argument that experiences which represent differences greater than d would not always be discriminable?

This seems to run counter to Dummett’s claim (in ‘Wang’s Paradox’) that ‘there are no phenomenal qualities, as these have been traditionally understood’ (323). But part of what Dummett means by ‘as these have been traditionally understood’ is something like ‘understood as obeying Distinctness/Discriminability.’ Later I’ll return to the question of whether, as Dummett seems to think, thinking of phenomenal properties as not meeting this constraint is a kind of objectionable changing of the topic. 2 For arguments of this sort, see Dummett (1975) and Deutsch (2005). 1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

222 phenomenal content and phenomenal continua On the other hand, one might take the claim that our visual system can’t register differences less than d to mean that, while we can represent differences of color of less than d, it is never the case that a pair of experiences which differ in the representation of color in such small ways are ever discriminable. But this is less an argument for the existence of a phenomenal continuum than a simple assertion that phenomenal continua are possible.3 What’s wrong with the alternative hypothesis that our color experience is ‘quantized’ in the sense that, while there are differences in color too small to be perceptually represented by us, any experiences which do differ in their representation of color will also be discriminable from each other? I think that the best argument for the possibility of phenomenal continua is the simpler one that most of us, at one time or other, have had an experience which is enough like a phenomenal continuum for us to be fairly sure that such a series of experiences is possible. Consider the experience of watching the color of some patch of the sky while the sun is setting over the ocean. Let’s consider these experiences second-by-second—letting each second correspond to a distinct color patch in the example above. It seems pretty clear that the phenomenal property that one instantiates at each second is indiscriminable with respect to color from the phenomenal property that one instantiates at the next second. (If this does not seem clear, just sub-divide the experience into smaller segments.) But then this series of experiences is very much like a phenomenal continuum. Now, this is not quite a phenomenal continuum, because there are noticeable differences between consecutive experiences in the series—the sound of the waves, the feel of the wind, all typically change quickly enough that they might differ from moment to moment. But it is similar enough to a phenomenal continuum that it at least strongly supports the possibility of phenomenal continua. All we have to do is imagine a very still day, on which one is wearing excellent noise-canceling headphones, etc. To deny that this is would be a phenomenal continuum it seems that one must say that, if one attended to one’s experience more closely, one could, no matter how small the time intervals in question, distinguish between what it is like to be oneself at one moment and at the next. But this seems very implausible. This simple ‘argument from experience’ is, however, open to an objection. No one should say (for reasons discussed in Chapter 29) that undiscriminated phenomenal properties must thereby be identical; what the proponent of the Phenomenal Content Thesis needs is rather the claim that indiscriminable phenomenal properties (in the sense of indiscriminable described earlier) be identical. 3

Essentially this point is made in Fara (2001), }3.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content and phenomenal continua 223 And how are we supposed to get from the fact that our subject on the beach is not able, in her situation, to make the relevant discriminations, to the conclusion that no one could? It seems to me, though, that this inference is quite a plausible one. Our beachgoer is, after all, in a pretty good position to make the relevant discriminations. The experiences in question are consecutive; and to imagine the case, we need not imagine her to suffer any deficiencies in the operation of her memory or the thoroughness of her introspection. Suppose that we try to block this inference by saying that there is a possible subject who does discriminate between the relevant phenomenal properties.4 Call this subject ‘the super-introspector.’ Note that when we are imagining the superintrospector, we are not imagining someone with better vision than the beachgoer, who is therefore capable of distinguishing more colors in the sky; that sort of subject would instantiate different phenomenal properties than the beachgoer, and hence could not be used to show that the phenomenal properties instantiated by the beachgoer are discriminable. We are supposed to be imagining someone who represents exactly the same colors in the sky as does the beachgoer, and instantiates just the same phenomenal properties, but is just better at distinguishing between those phenomenal properties. It is very unclear whether a super-introspector, so described, is possible. For suppose that she is. Then we can ask whether the super-introspector can herself undergo a series of experiences like the beachgoer’s: that is, a series of experiences the first of which the super-introspector discriminates from the last, but every consecutive pair of which the super-introspector cannot, despite being quite attentive, discriminate. It seems that this must be possible; it is very hard to imagine what it would be like to be a subject who could not undergo such a series of experiences.5 But that leaves us with two options. Either it is possible for there to be someone who is yet better at introspecting than the super-introspector—a super-duper-introspector—or there is not. If there is not, then the series of experiences just described is a phenomenal continuum, and the possibility of the super-introspector doesn’t in the end challenge the argument for the existence of phenomenal continua. If there is, then we can ask the same questions about 4

Thanks to Daniel Immerman for helpful discussion here. At least, it is hard to imagine being a subject able to make the perceptual discriminations we can make who was not able to undergo a phenomenal continuum. One can imagine a subject with very coarse-grained perceptual discrimination abilities like this—say, one which can only represent things as pure white or pure black. It does seem that phenomenal continua would be impossible for such a creature. 5

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

224 phenomenal content and phenomenal continua the super-duper-introspector as we asked about the super-introspector. The same process could be repeated indefinitely. Hence to use the gap between undiscriminated and indiscriminable phenomenal properties to block the possibility of phenomenal continua, we need to posit an infinite series of possible subjects, who—despite being alike with respect to the colors they visually represent—are such that each individual in the series can discriminate between phenomenal properties which the previous individual cannot. I find this difficult to believe. So examples like the beachgoer provide, I think, a good enough argument that we should believe in the possibility of phenomenal continua unless we’re presented with a compelling argument for their impossibility. Are there any such arguments? In answering this question, it is worth keeping in mind that different writers attach different senses to ‘phenomenal continuum’ and that, in some cases, the apparent disagreement between my defense of the possibility of phenomenal continua and others’ defense of their impossibility may be merely verbal. As is well known, phenomenal continua can be used to construct a kind of sorites argument; and different interpretations of ‘phenomenal continuum’ correspond to different inductive premises in the corresponding sorites arguments. So, for example, when Delia Graff Fara argues against the possibility of phenomenal continua, she has in mind a series of color patches experience of which would force us to reject the transitivity of ‘looks the same as.’ By contrast, I’m concerned to show that there is, or at least could be, a series of experiences which would force us to reject the transitivity of ‘is indiscriminable from.’ It’s not obvious that these are quite the same. Fara, for example, is willing to allow that there may be differences in how things look which are such that ‘the change in appearance is too slight, or too slow, for us to notice it.’6 But if the differences between a pair of experiences are too slight for me to notice them, then it looks like the relevant pair of phenomenal properties will be, in my sense, indiscriminable, despite the fact that the pair of experiences differ with respect to how things look to the subject of the experiences. A difference in the colors objects look to have (in Fara’s phenomenal sense of ‘looks’) entails a difference in the colors they are visually represented to have. But this entails the falsity of Distinctness/ Discriminability, and makes quite plausible the existence of a phenomenal continuum in my sense. Since, as Fara points out, the possibility of such a pair of experiences is consistent with the impossibility of phenomenal continua in her 6

Fara (2001), 927. Note that Fara does not assert that such pairs of experiences are possible; she just thinks that it’s an open possibility—which is consistent with the impossibility of phenomenal continua in her sense—that they are.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content and phenomenal continua 225 sense, it seems plausible that the disagreement between the conclusion of this chapter and her denial of the possibility of phenomenal continua is merely verbal. This difference gives us the resources to block someone who might try to use Fara’s central argument against the possibility of phenomenal continua in her sense against the possibility of phenomenal continua in mine. That argument is that phenomenal continua would, if possible, falsify platitudes like ‘If any two patches look the same in respect of colour, then if one looks red so does the other’ (907). While I agree that platitudes like this look hard to deny, I think that it’s not too implausible to respond to Fara’s argument by distinguishing the following two interpretations of ‘looks the same’: (i) ‘is visually represented as having the same color’; (ii) ‘is indiscriminable from.’ We can grant the truth of Fara’s platitude under interpretation (i), while still denying it under (ii). So it seems to me that the case for phenomenal continua, and against the Phenomenal Content Thesis, is a reasonably strong one. And of course this line of argument against the Phenomenal Content Thesis is a general one. If phenomenal continua are possible with color representation, they’re also possible with perceptual representation of size, shape, location, etc. Hence if phenomenal continua are possible, all of these properties are phenomenally sneaky.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

31 Perceptual Constancies and Phenomenal Match A quite different way of arguing for the existence of phenomenally sneaky properties begins with examination of perceptual constancies. Cases of perceptual constancy are typically presented as cases in which a subject’s visual system represents some quality of a perceived object as constant despite certain phenomenal changes over the course of the subject’s experience. So, for example, cases of color constancy are cases in which the color of an object is represented as constant despite changes in illumination; cases of size constancy are cases in which the size of an object is represented as constant despite changes in relative distance (and hence changes in the size of the relevant retinal image); cases of shape constancy are cases in which the shape of an object is represented as constant despite changes in relative orientation (and hence changes in the shape of the relevant retinal image). It is natural to think of these as cases in which a subject’s phenomenal properties are a function of the sensing of two different represented qualities. Consider, for example, a normal visual experience of a receding object. There is an obvious phenomenal difference between an experience of the object when it is relatively near and when it is relatively far; and this corresponds to a difference in the representation of the object’s relative distance while the object’s size is represented as constant. The subject’s phenomenal property is a function from the pair of the constant quality (size) and the variable quality (relative distance) which are constituents of the content which the subject senses. Good evidence for this is that any difference in ‘size phenomenology’ brings with it a difference in the representation of either represented size, represented relative distance, or both. This seems to be a general feature of cases of perceptual constancy. In each case, the phenomenal properties instantiated seem to be function f of some quality c represented as constant and some other quality v represented as

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

perceptual constancies and phenomenal match 227 variable.1 The point of putting things in this way is to bring out a consequence of the Phenomenal Content Thesis, which is that it is impossible that f ðc1 ,v1 Þ is indiscriminable from f ðc2 ,v2 Þ for c1 6¼ c2 and v1 6¼ v2. Call cases meeting this description cases of phenomenal match. Let’s take size constancy as an example. If cases of phenomenal match for size constancy were possible, then subjects could instantiate indiscriminable phenomenal properties while sensing contents which differ with respect to the size and relative distance of some object. The problem for the phenomenal content theorist is that we can’t pursue the same strategy of constituent deletion with these sorts of qualities as we pursued with respect to the perceptual representation of phenomenally silent contents, like particular objects and natural kinds. In the latter case, we could ignore representational differences which did not register phenomenally by employing the strategy of constituent deletion. But if we pursued the same strategy with respect to, for example, size properties, the remaining content functions would obviously be too sparse to be a supervenience base for phenomenal properties. And that is of course just to say that, if there really are cases of phenomenal match involving size constancy, size and relative distance properties are phenomenally sneaky. Of course, this is only a problem for the Phenomenal Content Thesis if there is reason to believe that cases of phenomenal match are possible. But a case can be made that they are. Here’s a possible example involving shape constancy: Subject 1: An ordinary coin is suspended in the air in front of me and rotated slightly so that, as we might say, it ‘looks elliptical.’ Though we might describe the case this way, it seems to be a standard case of shape constancy, so it is plausible that I am sensing the shape of the coin as constant (circular) and its orientation as varying as it is rotated. Subject 2: I get an elliptical piece of metal done up in just the right way to ‘look like’ a rotated coin. I can hold the piece of metal in my hand and see its shape clearly. It is then suspended in the air in front of me so that my phenomenal properties are indiscriminable from those in the experience above. Presumably when the piece of metal is in my hand I am able to visually sense it correctly as elliptical; and there

For a discussion of constancy which fits this characterization well, see Hilbert (2005). It is controversial, though, that this characterization is correct. For alternative views, see Thompson (2006) and Cohen (2008). 1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

228 perceptual constancies and phenomenal match seems to be no reason to think that I should sense its shape as changing while it moved to the position in which it is suspended in the air in front of me. So, plausibly, at the end I am still sensing the shape of the coin as elliptical. It is a substantive claim that cases of the foregoing sort are possible.2 But if they are, then it seems that we can have pairs of indiscriminable phenomenal properties of subjects who differ with respect to the shape and relative orientation properties they sense objects to have. And that, for the reasons already sketched, entails the phenomenal sneakiness of size and relative orientation, and the falsity of the Phenomenal Content Thesis. It is not hard to see how analogous cases might be constructed for other perceptual constancies. For example, we can imagine differently sized but otherwise indiscernible objects moved different distances from the subject (against a featureless background) to create a pair of experiences with the same phenomenal character. If the initial experiences of the objects’ sizes are veridical—and there’s no reason to doubt that they could be—then the initial experiences of their sizes will represent them as of distinct sizes. If the objects exhibit size constancy in the usual way, as the objects are moved further away from the subject, the size properties sensed will remain constant. Hence in this situation (again, if it is possible), we will have subjects instantiating indiscriminable phenomenal properties who differ with respect to the size and relative distance properties they sense objects to have— which entails the phenomenal sneakiness of size and relative distance. We might also imagine objects of different lightness viewed through changes in level of illumination, or objects of different hue viewed through changes in the color of the light under which they are presented. Let’s consider how the phenomenal content theorist might respond to these examples. Consider again the two ‘quarter’ experiences given earlier. What the proponent of the Phenomenal Content Thesis needs is some representational property which is such that (i) the two subjects each have this property and (ii) any two subjects which have this property have the same phenomenal properties. One might then suggest the following disjunctive property: the property of either sensing a circular shape at (say) a 30 degree relative orientation or sensing an elliptical shape at a 0 degree relative orientation. This seems to satisfy (i) and (ii). But this is just a stopgap, since once we see how examples like the ‘quarter’ 2

Because it might be that, when we set up conditions, including the background, in such a way that we can generate indiscriminable phenomenal properties, we get failures of constancy: perhaps in those conditions, it will not seem to the subject that the rotated object is maintaining a constant shape. I don’t see any reason to think that this would be the case, but this is not the sort of thing reliably decided from the armchair.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

perceptual constancies and phenomenal match 229 experiences can be generated, we can also see that, if these cases are possible, then indefinitely many other such examples are also possible. And that means that we won’t be able to rest with the disjunctive property mentioned above; instead we’ll need a disjunctive property with indefinitely many disjuncts to handle the indefinitely many shape/orientation combinations which can be combined to generate a phenomenal property indiscriminable from those associated with the quarter experiences. This is, on its own, counterintuitive; but it is especially puzzling when we remember that the proponent of the Phenomenal Content Thesis is committed to the identity of representational and phenomenal properties. So the relevant question is: could phenomenal properties be identical to indefinitely long disjunctive properties like the one sketched here? This seems to conflict with the fact that I can grasp, and self-predicate the phenomenal properties I am now instantiating; for from this it would follow that I can grasp and self-predicate this indefinitely complex disjunctive property. But I don’t think that this is the sort of thing I can do. There are two ways in which one might try to save the spirit of the ‘disjunctive properties’ proposal while avoiding this problem. This first is to adopt a very coarse-grained view of the contents of perceptual experience. If we think of the contents of perceptual experiences as, for example, sets of possible worlds, then we can avoid the use of disjunctive representational properties, since there is a set of circumstances in which experiences with the phenomenal character of the ‘quarter’ experience are veridical. It is the set of worlds in which the quarter is circular and at a 30 degree orientation or . . . (fill in the relevant disjunction here). And specifying the relevant set of worlds via disjunction needn’t make the content itself disjunctive; on this conception of content, there’s no such thing as a disjunctive content, since contents are simply unstructured collections of circumstances. But this very coarse-grained view of the contents to which we stand in sensing relations is difficult to maintain. One reason is that it entails that sensing relations are closed under necessary consequence; and this, for reasons already mentioned in Chapter 29, looks implausible. A second is that it seems unsuitable for the purposes of the proponent of the Phenomenal Content Thesis, because, by minimizing differences between contents, it also makes it harder for phenomenal properties to supervene on sensing properties. And of course if we want to defend a claim of identity between representational and phenomenal properties, we need both supervenience claims to be true. One way to see the problems here is by thinking about subjects standing in sensing relations to impossible contents, possible examples of which include the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

230 perceptual constancies and phenomenal match waterfall illusion and imaginary (but presumably possible) perceptual experiences with phenomenal characters resembling that of the experience of looking at an Escher drawing.3 Consider any pair of such experiences. They will have just the same content, on the coarse-grained conception of content, since both will be false at every world—but of course the subjects of those experiences might differ widely in their phenomenal properties. Hence phenomenal properties can’t be identified with phenomenal relations to such coarse-grained contents, since these coarsegrained views of content would threaten to make all but the phenomenally silent contents phenomenally sneaky. The second way of modifying the ‘disjunctive properties’ solution to the problem posed by phenomenal match is in effect a more sophisticated version of the strategy of constituent deletion. If we pursued the simpler version of that strategy, we would end up identifying the phenomenal property instantiated in our quarter experiences with some property like the property of, for some shape property S and orientation property O, sensing some value of the propositional function f(S,O) where for simplicity we can suppose that the propositional function includes all of the facts about the sensed content other than the represented shape and orientation of the coin. This won’t do, since there will obviously be shape and orientation properties which, when supplied as arguments to f, will yield sensing properties which entail phenomenal properties which are discriminable from those exemplified by the subject of the quarter experiences. (Just let S be, for example, the property of being cube-shaped.) But perhaps this problem can be solved just by restricting the quantification over shape and orientation properties to those shape/orientation pairs which, when given as argument to f, will deliver as value a content such that any subject sensing that content will instantiate a phenomenal property indiscriminable from the one instantiated by a subject of one of the quarter experiences. Let R be a relation between shape and orientation properties which holds only when this is the case. Then we might suggest identifying the relevant phenomenal property with the following representational property: the property of, for some shape property S and orientation property O such that R(S,O), sensing some value of the propositional function f(S,O) This delivers basically the result at the which the ‘disjunctive property’ response was aiming without requiring that subjects stand in the sensing relation to 3

For discussion of the former case, see Crane (1988).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

perceptual constancies and phenomenal match 231 disjunctive contents—much as the original use of the strategy of constituent deletion secured basically the result at which the ‘existential contents’ theory was aiming without requiring that subjects stand in the sensing relation to existential contents. Perhaps something along these lines could work. But I worry that, as stated, it has the advantages of theft over honest toil. As presented here, the relation R was defined partly in terms of the indiscriminability of phenomenal properties. But if we were allowed to help ourselves to facts about phenomenal properties in the definitions of the relevant representational properties, it would be no difficult task to formulate a theory which would make the Phenomenal Content Thesis true: For any phenomenal property F, F ¼ the property of instantiating some representational property R which is such that, necessarily, for all x, Rx ! Fx This would be to take the strategy of constituent deletion to its breaking point: one would be, in effect, ‘deleting’ the whole representational property. This is surely not a thesis of the identity of representational and phenomenal properties worth the name. But it seems to me that the same can be said for the theory which is formulated using relation R. One could improve upon that theory by specifying relation R in some way which did not make use of facts about the indiscriminability of, or indeed any other properties of, phenomenal properties. It is a substantial claim that this could be done. But suppose that it could. The specification of the relevant relation between shape and orientation properties will presumably be quite complex. But this will then lead to a problem parallel to the original problem for the ‘disjunctive properties’ response: our ability unproblematically to grasp and self-predicate phenomenal properties, along with the fact that none of us are at present in any position to grasp the wanted relation R, will entail the falsity of the relevant identity claim.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

32 Phenomenal Content and the Representation of Change The problems for the Phenomenal Content Thesis which result from perceptual constancy seem to stem from the fact that the content sensed by a subject at t can vary depending upon the phenomenal properties of the subject in the time immediately leading up to t. It seems essential to the ‘quarter’ experience, for example, that one inspect the quarter (and the elliptical fake quarter) before they are moved into the relevant orientations. We can also use this feature of the content of experience to construct an argument against the Phenomenal Content Thesis using an example of the representation of motion: Subject 1: Suppose I am playing a carnival game in which I have to shoot a duck which moves slowly from left to right. But the game is broken, so that the duck moves very, very slowly (though visibly). In fact, it moves slowly enough that were I to quickly open and shut my eyes, it would be impossible for me to tell that the duck was moving at all. Now imagine some interval t1 in the series where the duck is moving which is short enough that it is indiscriminable from the open-and-shut eyes case. Subject 2: Next, imagine another series of experiences the next day, when the game is really broken, and I am staring at a stationary duck, and select an interval t2 of the same duration as t1. If the foregoing is correct, it seems that (if the Phenomenal Content Thesis is true) the subjects of t1 and t2 will instantiate the same phenomenal properties. After all, it was stipulated that the duck was moving slowly enough that, had I opened and shut my eyes at the beginning and end of t1, I would not have been able to tell that the duck was moving; so it seems as though the two phenomenal properties will be indiscriminable. But it seems that these two subjects will sense different contents. t1 is part of an interval during which I am sensing the duck as in constant motion; hence during t1 I am presumably representing the duck as in motion. There’s no reason to

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

phenomenal content and the representation of change 233 think that the subject of t2 is misrepresenting the world, so presumably the subject of t2 is sensing the duck as stationary. But then it follows, given that the two subjects instantiate indiscriminable phenomenal properties, that motion is phenomenally sneaky. One might think that this is just another example of the difficulty posed by cases of phenomenal continua, discussed in Chapter 30. (Discussions of phenomenal continua have often involved discussion of the case of motion; Dummett, for example, discusses perception of the movement of the minute hand on a clock.) But, while related, the cases are a bit different. The present case is not one in which two experiences each represent different sorts of movement despite being indiscriminable; it’s a case in which we seem to have a pair of indiscriminable experiences, one of which represents motion and the other of which does not. This, like the examples involving perceptual constancy, is a challenge to the existence of phenomenal content. And as in the case of perceptual constancy, the defender of phenomenal content can reply in one of two ways: by denying that the two subjects really do instantiate indiscriminable phenomenal properties, or by denying that they differ in their sensing properties. It seems to me difficult to defend the first of these views, for much the same reason as it seems to me difficult to defend the view that there could not be a phenomenal continuum. The pair of experiences just described strikes most people as clearly possible, because most people have had experiences relevantly similar to it. This strikes me as strong evidence that a pair of experiences like this is possible. So let’s consider instead the possibility of denying that the two subjects differ in their sensing properties. One initially plausible way to do this is to say that, although the full experience of which t1 is a part represents the duck as moving, t1 itself does not. But the problem with this is that the full experience can be divided without remainder into t1-sized chunks. If we think of the full experience as a series of such chunks, we’ll then have to say that no experience in the series represented the duck as moving, but that the whole series of experiences together did. But this has some odd consequences. Presumably if the experience at t1 did not represent the duck as moving, it represented the duck as stationary. (This is anyway needed to get us to phenomenal content, since it should be uncontroversial that the experience at t2 represents the duck as stationary.) But that means that (again by parity), during the full experience of the moving duck, the subject is constantly representing the duck as stationary. How can this be reconciled with the fact that the subject of the experience is representing the duck as in constant motion, and the obvious fact that the subject could be accurately representing the scene before her?

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

234 phenomenal content and the representation of change There is also a more radical response available to the defender of phenomenal content, and that’s just to deny that it makes sense to talk about the phenomenal character, or content, of part of an extended visual experience.1 But this is, again, hard to believe—and even hard to understand when we translate this into talk about the properties of subjects. There surely was something that it was like to be the subject of t1, and this is sufficient for the subject at t1 to instantiate a phenomenal property. And we can certainly imagine asking, of that subject: At what point did he sense the duck as being at that location? This question should have an answer; but any such answer will attribute a sensing property to a subject at a particular time. . . . Reflection on these three sorts of experiences over time—phenomenal continua, perceptual constancies, and the perceptual representation of motion—suggests that there is no such thing as phenomenal content, because there is no class of representational properties which is both such that phenomenal properties supervene on those representational properties and is such that differences in those representational properties entail discriminable phenomenal properties.

Siewert (1998) seems to flirt with this position when discussing experiences over time: ‘typically, then phenomenal character of our visual experience during a given time . . . will be identifiable only relative to the phenomenal character of experience lying outside that time’ (225). Another radical reply would be to deny that we ever sense temporal properties and relations. See Chuard (2011) for a defense. 1

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

33 Indiscriminable Phenomenal Properties If phenomenal sneakiness is possible, then the Phenomenal Content Thesis is false—which means that, if phenomenal properties are identical to representational properties, Distinctness/Discriminability must be false. On this sort of view, while discriminability is still sufficient for difference of phenomenal properties, its absence is not sufficient for identity of phenomenal properties. To many people, this view will seem like a somewhat crazy last resort. After all, think about the way that the notion of a phenomenal property is usually introduced. We say things like: A and B are instantiating the same phenomenal property iff things ‘seem the same from the inside’ to A as to B—but it certainly seems plausible that if what it is like to be me is indiscriminable from what it is like to be you, then things ‘seem the same from the inside’ to each of us. Indeed, one might think that any view of phenomenal properties which rejects Distinctness/Discriminability is just changing the subject: one might think that what we’re interested in when we’re discussing the nature of phenomenal properties just is some property which obeys Distinctness/Discriminability. A proponent of this view might say: Let’s grant you your use of the term ‘phenomenal property.’ Perhaps, in your sense, there can be distinct but indiscriminable phenomenal properties; but I’m interested in a class of properties which can’t be distinct without being indiscriminable. Let’s call these ‘schmenomenal properties.’ Even if you can give a satisfactory treatment of phenomenal properties, this doesn’t explain the nature of the schmenomenal properties—and this just leaves the core of the problem of phenomenal consciousness unsolved. The problem with this line of reply is that one can’t simply stipulate the existence of a class of properties in this way. In effect, the strategy for defining schmenomenal properties is to begin with a relation R—here, indiscriminability—and to say that the schmenomenal properties are that class C of properties which are

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

236 indiscriminable phenomenal properties such that, necessarily, x and y have their C-properties in common iff they are R-related. But, for any such relation, it’s a substantive claim that such a class of properties exists. And if there are cases of phenomenal continua, indiscriminability will not be transitive, and there will be no schmenomenal properties. However, there’s a related, more basic concern which need not involve an attempt to stipulate the existence of schmenomenal properties. And that’s just that one might reasonably wonder what it could mean to say that there are distinct but indiscriminable phenomenal properties. What could it mean for A’s situation to be indiscriminable from B’s while what it is like to be A is distinct from what it is like to be B? In response, the main thing to say is that property identities can give us principled reasons to distinguish between properties which we’d otherwise have had no basis to distinguish. The present idea is that we should identify phenomenal properties with certain representational properties—properties of sensing certain contents, and attending to certain objects and properties. Given that we understand the distinction between the representational properties associated with experiences of looking at distinct but perceptually indistinguishable golf balls, this identity claim should help us understand what it means to distinguish between indiscriminable phenomenal properties. We needn’t be committed to the idea that we would have reasons which are independent of this property identity to claim that indiscriminable phenomenal properties could be distinct. An analogy might help. Consider the identity between heat and mean molecular motion. Given that there are differences in mean molecular motion which are not even in principle detectable by human observers, it follows from this theoretical identity that there are differences in the heat of substances which are not even in principle detectable by human observers. But would these indiscriminable distinctions among heat properties have been as readily acceptable prior to the theoretical identity in question being accepted? There would, I think, have been some temptation to say that the idea that surfaces could differ in heat in ways which can’t be felt by any observer is incoherent. But given the identity of heat with mean molecular motion, this temptation goes away. I suggest that we should take a parallel view of the identity of phenomenal and representational properties and the rejection of Distinctness/Discriminability. We can see that there must be distinctions between indiscriminable phenomenal properties once we recognize (i) that phenomenal properties are identical to representational properties and (ii) that there can be distinctions in representational properties which are not, even in principle, discriminable by human subjects.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

indiscriminable phenomenal properties 237 I think that this is enough to make the rejection of Distinctness/Discriminability intelligible. Despite the fact that the rejection of this principle is not very popular, actual arguments in favor of Distinctness/Discriminability are not easy to find; more often, one finds the assertion that something like this principle is ‘a conceptual truth’ or something of that sort. This sort of claim is not without plausibility; but we should always be open to the possibility that what seem to be conceptual truths can be falsified by theoretical identities of the sort we’re now considering. One person who does give an explicit argument against the rejection of Distinctness/Discriminability is Michael Tye. Tye considers the possibility of someone denying that a pair of indiscriminable experiences have the same phenomenal character, and points out that any subject of such a pair of experiences will not be conscious of any phenomenal difference between them. He then says that ‘The question now becomes why they are not conscious of any such difference. Surely the most straightforward and natural answer is that in some cases there is no difference: the phenomenal character is the same.’1

This is an argument of the same family as an argument often brought against ‘disjunctivist’ views of experience, according to which veridical experiences and hallucinations are not members of a common psychological kind. Against such a view, it’s pointed out that hallucinations can be indiscriminable from veridical experiences, and then argued that the disjunctivist can provide no account of what such a hallucination/veridical experience pair have in common. One might take Tye to be pointing out that, in effect, anyone who denies Distinctness/ Discriminability faces a parallel objection. Whether this is right depends upon how we’re thinking of the objection to the disjunctivist. One way of thinking about the argument is as simply pressing the point that it is baffling how mental events of fundamentally different kinds could be indiscriminable. This sort of intuitive argument is no problem for the present sort of view. Even if we take it that distinct phenomenal properties can be indiscriminable, there’s no claim that properties have nothing in common. They will both be representational properties of the same type: both will be a matter of 1

Tye (2009), 560. Just before this passage, Tye presents another argument against Distinctness/ Discriminability, which seems to go something like this: if we say that two experiences can be indiscriminable but not phenomenally the same, then we need some analysis of ‘indiscriminable’ (since we can’t understand it as meaning ‘has the same phenomenal character’). But, Tye thinks, there is no plausible analysis to fill in here. I’ve already in effect answered this argument with the explanation of ‘indiscriminable’ given in Chapter 29: roughly, two phenomenal properties are indiscriminable iff one cannot know that they are distinct on the basis of introspection and memory.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

238 indiscriminable phenomenal properties subjects standing in the relation of sensing to some property, and in the sorts of cases of indiscriminable but distinct phenomenal properties discussed earlier, the properties to which this relation is borne will have a lot in common. A different way of thinking about the argument is as relying on the premise that, in general, when x and y are indiscriminable, this fact should get some explanation solely in terms of the nature of x and y. But this premise can be safely rejected; for at least some sorts of indiscriminability, the fact that x and y are indiscriminable in that way will be explained, not in terms of x and y alone, but also in terms of the nature of the relevant sort of indiscriminability. An example helps to make the point. Suppose that we have a balance scale used for comparing the weight of collections of grains of sand, each of which has the same weight as every other, with a display on the front which shows the letter ‘L’ if the sand on the left side of the balance weighs more than the sand on the right side, ‘R’ if the sand on the right weighs more, and ‘?’ if the scale does not detect either side as having more weight than the other. Suppose further that the scale, though very reliable, cannot detect the difference in weight between any two collections of sand which differ in number by one grain. This scale enables us to define a kind of discriminability; x and y will be scalediscriminable iff one can know on the basis of what the scale says that x and y are of different weights. Relative to this sort of discriminability, grains of sand will be sneaky in something like the way that colors turn out to be phenomenally sneaky when we reflect on phenomenal continua. But it would be bizarre to claim that we should expect an explanation of the fact that a 17-grain lump is scaleindiscriminable from an 18-grain lump in terms of identity of weight properties. So why should we expect the fact that certain distinct phenomenal properties are (in the sense explained earlier) indiscriminable to get an explanation in terms of identity of phenomenal properties? This way of thinking about the example of the scale is based on the sort of indiscriminability we get from phenomenal continua; we could modify the example to better fit cases of phenomenal match. Suppose that there are two sorts of grains of sand: type A grains weigh 13 micrograms, and type B grains weigh 5 micrograms. Then imagine modifying the earlier example in two ways. First, the scale is a digital rather than a balance scale; second the scale displays ‘A’ if the mixture contains more type A than B grains, ‘B’ if more B than A, and ‘?’ if the scale can’t tell. For some mixtures, it is easy to see how this will be possible. For example, if the scale registers a weight of 41 micrograms, it will be able to tell that there are two type A grains and three type B grains, and hence will display a ‘B’. But for other mixtures, this won’t be possible; if the scale registers a weight of 65 micrograms, this might be because there are five type A grains on the scale

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

indiscriminable phenomenal properties 239 or because there are thirteen type B grains on the scale, and so the scale will display a ‘?’. So a pair of mixtures, one with thirteen B grains and one with five A grains, will be indiscriminable by this scale. And, as before, we should not expect that the fact that these two ‘mixtures’ are indiscriminable in respect of their composition by A and B type grains to be explicable solely in terms of the identity of their composition by A and B grains. But note that while in neither case will we get an explanation of the relevant sort of indiscriminability just in terms of the identity of the properties which are (or are not) discriminated, in both cases we will be able to give an account of discriminability partly in terms of these properties. In the example of the first scale, two lumps will be discriminable iff they differ in number by more than one grain of sand. Can we do the same thing for the sort of discriminability of phenomenal properties we have in mind? I don’t see why not. We’ve encountered three sorts of cases in which we might be inclined to say that we have distinct but indiscriminable phenomenal properties: sensing properties which differ only with respect to phenomenally silent contents; sensing properties which differ only with respect to small differences in representation of some phenomenally sneaky property (as in phenomenal continua, or the cases of the experience of change); cases of phenomenal match, in which two pairs of sensed properties ‘cancel each other out.’ This is a slight simplification, since there’s no reason that two experiences couldn’t differ both in phenomenally silent contents and in some small difference in some phenomenally sneaky property; such a pair of experiences would be indiscriminable but would not fit into any of the three categories just defined. But setting this to the side, we can say that a pair of phenomenal properties is indiscriminable iff they fall into one of these three categories. Turning this into a more precise account of indiscriminability would then be just a matter of making precise our definitions of the three classes of cases just described; and this, while highly non-trivial, does not seem impossible in principle. Summing up: the most plausible version of the idea that indiscriminable experiences present a problem for this sort of view is the idea that we should be able to get some sort of explanation of why an arbitrary pair of experiences is, or is not, discriminable in terms of phenomenal properties. We can either require that such an explanation be given in terms of identity of phenomenal properties or not. If we do, then the demand for explanation should be rejected, since (as the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

240 indiscriminable phenomenal properties examples of the scales show) it is not generally true that indiscriminability in respect of some property-type should be explained in terms of identity of the relevant properties. If we don’t, then there’s no reason to think that the demand for explanation can’t be met. . . . On the basis of the arguments for phenomenal sneakiness, I’m tentatively inclined to reject Distinctness/Discriminability, and accept Simple Identity. In the remainder of the book, I want to think about the constraints placed by this view—along with some of the theses which led us to it—on the two main attempts to give a theory of phenomenal properties in terms of the physical properties of subjects.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

PART VIII

The Reduction of Phenomenal Properties

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

34 Two Kinds of Theories of Phenomenal Properties There are, broadly speaking, two main approaches to the construction of a physicalist theory of phenomenal properties. The first is a type-type identity theory. Let’s say that the internal states of an organism are the intrinsic physical properties of that organism. Then an identity theory of phenomenal properties identifies phenomenal properties with internal states. The second sort of theory is a functionalist theory of phenomenal properties. As I’ll use the term, a functionalist theory of phenomenal properties is a theory according to which what it is for a subject to instantiate some phenomenal property F is for the subject to be in some internal state s such that for some accidental relational property R, R(s).1 Hence we’ll want the functionalist to provide some mapping from phenomenal properties to relational properties of internal states such that ☐ 8x ðFx iff ∃ y ðy is a state of x & RyÞÞ Given such a theory, we might then go on to state a corresponding property identity: F ¼ the property of being in some internal state y such that Ry This is a less demanding definition of ‘functionalism’ than is often given; often the view is defined by conjoining the definition just given with the claim that the relevant relational property of internal states should be a functional role. In what follows, it will be useful to have the broader class of theories in mind. The 1

The point of this restriction to non-essential relational properties is to avoid making identity theories entail trivially functionalism. Without it, we could state any identity theory which identified a phenomenal property F with some intrinsic physical property I as the functionalist theory that for a subject to instantiate F is for that subject to be in some internal state y such that y has the relational property of being identical to I.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

244 two kinds of theories of phenomenal properties functionalist theories we’ll be interested in will be reductive theories, in the fairly weak sense that they will not define the relevant relational properties in terms of the phenomenal or representational properties of subjects. We could also describe functionalist theories without using the standard ‘internal state’ talk, and speak only about properties and relations. In those terms a functionalist theory of phenomenal properties will be a theory that identifies phenomenal properties with the property of having some intrinsic property which instantiates some accidental relational property, to be specified by the theory. The relative advantages and disadvantages of these theories regarding issues like mental causation and the alleged multiple realizability of mental states are well-worn topics, and I won’t discuss them much here. Instead, I’ll focus on the narrower question of whether the line of argument traced to this point might shed light on the plausibility of these two approaches to understanding the nature of phenomenal properties. One way to think about the constraints the present approach to phenomenal properties places on these theories is to remember the form of the proposed identity between representational and phenomenal properties. The idea is that a phenomenal property F is identical to the property of standing in certain specific phenomenal relations (sensing and attending) to particular contents. Given the truth of some such identity claim, we should expect any theory of phenomenal properties to give necessary and sufficient conditions for a subject to stand in one or both of our phenomenal relations of sensing and attending to a content. We can separate out this explanatory target into two components: A theory of content: an explanation of what it is for a mental state to have a certain content. A theory of phenomenal relations: an explanation of what it is for a subject to bear a certain phenomenal relation like sensing (rather than some nonphenomenal relation like belief, or some distinct phenomenal relation like attending) to a content. A theory of phenomenal properties need not be presented as consisting of these two parts; but it’s convenient to think about functionalist theories of phenomenal properties as divisible in this way. In the next two chapters, I’ll ask how well— given the results of the preceding chapters—functionalist theories can accomplish these two tasks. I’ll return to identity theory in the closing chapter.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

35 Functionalist Theories of Content It is very plausible that there is a stable dependence relation between the physical internal states of subjects and the phenomenal properties of those subjects. In this chapter, I’ll adopt the convention of referring to the internal state which is causally sufficient for a subject S to instantiate a given phenomenal property— say, RED—as ‘REDS.’ It is worth emphasizing what this terminological convention does not assume. It does not assume that the phenomenal properties of subjects supervene on their intrinsic physical properties. Nor does it assume that the physical state causally sufficient for a given subject to instantiate RED or any other phenomenal property must be constant over time, or that for distinct subjects S and S*, REDS ¼ REDS*. Hence there is no assumption that ‘REDS’ is either modally or temporally rigid. All that is required is the very plausible assumption that there are causally sufficient conditions for a subject to be in a certain phenomenal state at a time which are specifiable in terms of the intrinsic physical properties of the subject at that time.1 This way of talking makes the simplifying assumption of uniqueness. This assumption of course is false; if there is one internal state which is causally sufficient for a subject to instantiate RED, there will be many which include it and hence also causally suffice for this. This won’t matter for the argument to follow, and will make it easier to state the sort of constraint on functionalist theories I’m interested in. We can think of the job of a functionalist theory of content as that of specifying a relation R between internal states-types and the properties those internal states represent (for a given subject) as instantiated. Given this set-up, we can state at

1 Hence I’m relying on Weak rather than Strong C-Dependence, in the terminology of Byrne and Tye (2006).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

246 functionalist theories of content least one constraint on functionalist theories of content which follows from the foregoing: [C] Necessarily, there is no subject S such that REDS and GREENS both bear R to redness over the same interval. Suppose for reductio that [C] is false. Then take the subject whose possibility is guaranteed by [C]’s falsity, and imagine that subject being first in state REDS, and then, immediately after, being in GREENS. This would be a case of psychedelic phenomenology þ constant representation of color properties; but, as argued earlier, such cases are impossible. My aim in this chapter is not to show that no functionalist theory can meet [C]. What I do think, though, is that [C] rules out many of the most plausible functionalist theories of content, and hence both substantially reduces the plausibility of functionalism about phenomenal properties and sharply constrains the range of functionalist theories of content which we should consider. Intuitively, to show that a functionalist theory fails to meet [C], one has to do two things: (i) show that distinct mental representations can bear R to just the same property, and (ii) show that such co-reference is possible when tokenings of those mental representations are, respectively, causally sufficient for the subject to instantiate RED and GREEN. With this in mind, let’s consider a few different ways in which a functionalist might try to specify the relevant representation relation R.

Indication Theories Indication theories of content say that the content of a perceptual state is determined by what would cause the subject to be in that state, were the subject in certain ideal or optimal conditions.2 Do theories of this sort fail to meet [C]? There certainly seems to be no impossibility in the claim that a given state of affairs could, in conditions optimal for the subject, sometimes cause one internal state and sometimes another. The question, though, is whether this is possible where those internal states are respectively individually causally sufficient for the subject to instantiate RED and GREEN. What we would need to show, in other words, is that some modal claim like the following is true: Possibly, there is a subject S for whom color red. 2

REDS

and

GREENS

both indicate the

A theory of this general sort is defended in, among many other places, Stalnaker (1984).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist theories of content 247 If this modal claim is true, indication theories entail that the subject whose possibility it guarantees would, when alternating between the two states REDS and GREENS, be an instance of psychedelic phenomenology þ constant representation of color properties. Hence, if scenario A really is impossible, this claim would entail the falsity of indication theories of content. The claim that there could be a subject S for whom REDS and GREENS both indicate redness is not obvious. But a case can be made for it. Imagine first that our subject is an evolved creature, and that we think of optimal conditions for such creatures as ‘involving the various components of the visual system operating as they were designed to do in the sort of external environment in which they were designed to operate.’3 It is well known that the process of evolution often issues in creatures who are not ‘put together’ in the way one might expect a creature designed ex nihilo for the relevant environment to be put together. This sort of observation makes it very plausible that, even if it seems odd a priori, it is possible for a creature E to have evolved to operate in an environment in which both REDE and GREENE were caused by red surfaces. Perhaps there was a surprising absence of green things in the environment, and red things were very important to the creature’s survival—so important that it was evolutionarily useful to have redundancy in one’s red-indicating capacities. It is important to be clear about what sort of redundancy is being imagined here. It is not just redundancy in the sense of there being multiple internal states which indicate a given property—the possibility of that kind of redundancy should be uncontroversial. It is, rather, a specifically phenomenal redundancy— the existence of multiple internal states which indicate the same property but are individually causally sufficient for distinct phenomenal properties. Once this is made clear, it is also clear that the indication theorist can reply to the argument by denying that this sort of phenomenal redundancy really is possible. On this sort of view, our description of E, despite initially seeming coherent, in fact fails to describe a possible creature. Now, on the one hand, this is a bit difficult to accept, because it really seems as though a creature could evolve the sort of phenomenal redundancy described above—but, on the other hand, functionalists about the phenomenal are already committed to blocking this sort of conceivability-possibility inference in other cases. The novel problems posed by [C] for functionalist theories only emerge when we turn to the phenomenal properties of non-evolved creatures. These problems can be brought out by extending an argument due to Ned Block.4 Familiar spectrum inversion scenarios feature inverting lenses which 3

Tye (2000), 138.

4

See especially Block (1998).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

248 functionalist theories of content would correlate our ordinary experiences of yellow things with BLUE phenomenal properties and our ordinary experiences of blue things with YELLOW phenomenal properties. But we can imagine more sophisticated inverting lenses, which do two things. First, they give our experiences of blue things a GREY phenomenal character. Second, when confronted with the light reflected from a normal yellow surface, they randomly vary between two settings. When they assume setting A, they do what conventional inverting lenses do, and cause me to instantiate BLUE; but when they are in setting B, they just ‘let the light through,’ causing me to instantiate YELLOW. Now imagine (adapting an example of Michael Tye’s) a Swamp-duplicate of me who materializes on earth with the sophisticated lenses as part of its visual system. Our question is: what is true of Swamp-me when the lenses are in, respectively, setting A and setting B? There appear to be three options: (1) Swamp-me is just like me when I have the lenses in; when the lenses are in setting A, he instantiates BLUE and is in a state which indicates blue, and when the lenses are in setting B, he instantiates YELLOW and is in a state which indicates yellow. (2) Swamp-me differs from me both in his phenomenal properties and in what his internal states indicate; when the lenses are in setting A, he instantiates YELLOW and is in a state which indicates yellow, and when the lenses are in setting B, he also instantiates YELLOW and is in a state which indicates yellow. (3) Swamp-me is like me with respect to his phenomenal properties, but his internal states indicate different things than mine; when the lenses are in setting A, he instantiates BLUE and is in a state which indicates yellow, and when the lenses are in setting B, he instantiates YELLOW and is in a state which indicates yellow. Tye opts for (1); and it is easy to see why this is an attractive option.5 It seems quite plausible that Swamp-me will instantiate the same phenomenal properties as I do when I have the lenses on, and the simplest picture would be one on which he also represented the same color properties as I do. So, if the indication theory of content is true, his internal states had better indicate the same properties as mine. The problem with this is that it’s hard to see how conditions optimal for Swamp-me could be anything other than the conditions in which it materialized. But Swamp-me materialized with the lenses in place. So it seems as though the internal state Swamp-me is in when the lenses are in setting A—which only occurs when he is confronting a yellow surface—will indicate yellow, not blue. 5

See Tye (2000), 137–9.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist theories of content 249 The same goes for the state he is in when the lenses are in setting B. This suggests that option (1) can’t be the correct description of the case. In order to block this result, one would have to argue that, despite the fact that Swamp-me materialized with the sophisticated inverting lenses in place, the conditions which are optimal for Swamp-me are ones in which the lenses are removed. This is just what Tye argues about a similar case, on the grounds that the presence of the lenses ‘interferes with the production of accurate color experiences.’6 But it is not easy to see what the principled basis for saying this could be. A color experience is accurate iff the color it represents the surface as being is the color the surface has; hence if we build in facts about what is and is not an accurate color experience, we are building in facts about the contents of those experiences. But this can’t be fair game for any theorist interested in giving a non-circular account of the facts in virtue of which experiences have the contents they do. A different way to press this point would be to point out that a creature might well evolve into a duplicate of Swamp-me—that is, a duplicate of me with the sophisticated inverting lenses in from birth. It should be uncontroversial that the state that that creature is in when the lenses are in setting A indicates yellow. So what is the principled reason for having the conditions which are optimal for Swamp-me be the ones which are optimal for me, rather than the conditions which are optimal for Swamp-me’s evolved duplicate? This line of argument suggests that Swamp-me’s internal states, unlike mine, indicate yellow whether the lenses are in setting A or setting B. Could it be, as option (2) has it, that Swamp-me’s phenomenal properties also differ from mine, so that Swamp-me instantiates YELLOW even when the lenses are in setting A? This is a difficult bullet to bite—which is presumably why Tye goes for option (1). It is just hard to believe that, when Swamp-me materializes beside me as I’m wearing my lenses, that our phenomenal properties could be radically different. And, as Tye recognizes, further problems result when we consider a duplicate of me which materializes without the inverting lenses. Presumably that being’s phenomenal properties will be the same as mine without the lenses in and, if we both insert the lenses, our phenomenal properties will stay in sync. But, as Tye says, the suggestion that ‘what it is like [for the Swamp-duplicates] will vary, depending upon whether the lenses were in place from the start or inserted by others, seems absurd.’7

6 Tye (2000), 138–9. Note that Tye is discussing the more standard case of inverting lenses, rather than the present one with the lenses which switch between settings A and B. I take it that this difference is irrelevant to the question about what conditions are optimal for Swamp-me. 7 Tye (2000), 138.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

250 functionalist theories of content This leaves us with option (3). But this entails the possibility of Scenario A. For on this view, BLUEswamp-me¼the lenses being in setting A and YELLOWswamp-me¼the lenses being in setting B. But then it follows that both BLUEswamp-me and YELLOWswamp-me indicate yellow. Given that the indication theorist identifies relation R with indication, this falsifies our constraint [C]. To generate an instance of psychedelic phenomenology þ constant representation of color properties, just imagine a situation in which Swamp-me is looking at a yellow surface, and the lenses are rapidly switching from setting A to setting B. Swamp-me’s phenomenal properties will be rapidly changing from BLUE to YELLOW and back again; but he will be constantly representing the color of the surface before him as yellow.

Co-variational Theories The central problem pressed here against indication theories was that such theories leave open the possibility of a kind of ‘co-reference’—distinct phenomenal states indicating the same color—and that this sort of co-reference, conjoined with an indication theory of content, entails the possibility of psychedelic phenomenology þ constant representation of color properties. Intuitively, then, it seems that what we want to do is modify our functionalist theory so that it does not permit this sort of co-reference. There are a few different ways in which we might do this. One easy way is to require, not just that REDS be caused by nothing other than redness in conditions optimal for S, but also that REDS be invariably caused by redness in those conditions. That is, one might think, all we need to do is to switch from an indication theory to a co-variational theory. Consider again the (allegedly) possible situation in which both REDS and GREENS indicate a single color—say, redness. This would then be a case, if we add the suggested stipulation to the indication theory, in which neither REDS nor GREENS would co-vary with a color in optimal conditions, and hence would be a case in which neither represents a color. Hence our co-variational theory would not entail that psychedelic phenomenology is consistent with representation of a single color property throughout. But this is hardly an advance, for our co-variational theory would still be consistent with a kind of psychedelic phenomenology þ constant representation of color properties—namely, a constant representation of no color properties at all. And that seems just as clearly impossible as the original scenario A. Analogous points will apply to the co-variationist’s treatment of the case of non-evolved creatures and the inverting lenses.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist theories of content 251 Here’s one way to bring out the weirdness. Imagine, what is surely possible, that some other states, which we can call ORANGES and YELLOWS, do co-vary with colors in optimal conditions—with orange and yellow, respectively. But in this kind of case, if the co-variational theory is to be believed, if the subject is looking at a screen with colors being projected upon it, and her phenomenal properties switch from YELLOW to ORANGE to RED, what has happened is that the screen first visually seemed yellow to the subject, then visually seemed orange to the subject, and then ceased to seem to have any color at all. This is hard to believe. Surely this sort of change in phenomenal properties can’t be a switch from representing the relevant surface as having a color to simply failing to represent it as colored at all.8

Internalist Functionalism Let’s consider another, more drastic way in which the functionalist might respond to this problem: she might endorse a purely internalist functionalist theory—for example, a conceptual role semantics which is, in Harman’s sense, ‘solipsistic.’9 If a functionalist theory has the consequence that any intrinsic duplicates are alike with respect to the representational properties of their internal states, then such a theory will not, obviously, permit the permutation of the representational properties of visual states via the permutation of the relationship between those states and instantiations of the various colors in the world, and hence won’t be open to an argument like the ones just given. The problem is that it is not easy to see how to construct such a theory, since even more ‘internalist’ versions of functionalism typically appeal to versions of one of the theories sketched earlier in giving their account of the content of perceptual experience. A good example here is Brian Loar, in Mind and Meaning, who says that ‘Perceptual input conditions are needed to secure uniqueness of systematic role because the internal constraints on beliefs do not individuate them.’10 Loar—despite the internalist flavor of much of his theory—goes on to give a kind of indication theory for observational beliefs, which can be criticized 8 The discussion of zombie replicas in Tye ((1995), 194–5) suggests that he would deny the possibility of this situation, on the grounds that the creature would simply not instantiate any phenomenal properties at all when in the state which, in optimal conditions, is sometimes caused by red surfaces and sometimes by green. But this runs into problems similar to those discussed in connection with indication theories. Imagine lenses which cause both blue and yellow surfaces to cause me to instantiate BLUE, and imagine a non-evolved creature which materializes beside me, and is an intrinsic duplicate of me with the lenses in. Tye seems forced to say that that creature would lack any phenomenal properties when looking at a blue or yellow surface. And that, just like option (2) in our previous example, is tough to swallow. 9 10 See Harman (1987). Loar (1981), 65.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

252 functionalist theories of content on much the same grounds given for criticizing indication theories of perceptual content.

Limited Holism A better response to the problems we’ve been discussing is to incorporate a kind of holism into our functionalist theory of content. So far, we’ve been thinking of R as a relation which is such that, intuitively, we can see what a given state REDS is R-related to just by looking at REDS and at what causes, and would cause, S to be in this state in various conditions. But we could instead define R as a relation between the subject’s collection of ‘color phenomenal states’ and the color properties; maybe such a phenomenal state represents a color iff the ‘best oneto-one pairing’ of all the phenomenal states with the colors assigns that phenomenal state to that color.11 The main virtue of this sort of theory in the present context is that it promises to block the sorts of arguments given earlier, since it guarantees a one-to-one pairing between internal states (of the relevant sort) and colors, thus (unlike indication theories) making it impossible for there to be a subject S for whom REDS and GREENS both represent redness and (unlike co-variational theories) making it impossible for REDS to represent a color while, say, ORANGES represents no color at all. The key question for this sort of view is how ‘best one-to-one pairing’ should be defined. In particular, we should ask whether the best one-to-one pairing of internal states with the colors they represent is something which can change during the life of the organism, or not. Suppose first that it cannot. Then we will run into problems with non-evolved creatures which are parallel to those discussed in connection with Block’s example of the swampman and the inverting lenses given earlier. Suppose that without the lenses, blue surfaces typically cause me to instantiate the phenomenal property BLUE, whereas yellow surfaces typically cause me to instantiate YELLOW. Once I put the lenses on, blue surfaces typically cause me to instantiate the phenomenal property YELLOW and yellow surfaces typically cause me to instantiate BLUE. Putting on the inverting lenses changes, not just my phenomenal properties, but also the colors I sense things to have. So, if I am looking at a surface which is blue, and put on the lenses, my phenomenal properties change It is worth noting that a careful statement of this view would not define R in terms of ‘color phenomenal states,’ since what it is for something to be a phenomenal state is part of what we want our theory to explain. 11

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist theories of content 253 from BLUE to YELLOW, and, whereas I previously represented the surface as blue, I now represent it as yellow. (Were this not the case, we could generate an instance of psychedelic phenomenology þ constant representation of color properties by imagining a case in which the lenses are taken in and out several times while the subject is looking at a surface; changing the surface while changing the lenses could generate a case of constant phenomenology þ psychedelic representation of color properties.) We can then consider two non-evolved creatures which materialize to be, respectively, duplicates of me without the spectrum-inverting lenses, and me with the spectrum-inverting lenses. Call these creatures swamp-me and swampinverted-me. It is hard to deny that swamp-me and swamp-inverted-me are, respectively, phenomenal duplicates of me without the shifting lenses and me with the inverting lenses. For suppose that we were to deny this. Then, if we imagine swamp-me having the lenses added immediately after having materialized—turning him into an intrinsic duplicate of swamp-inverted-me— we would have to deny that, after the operation, swamp-me had become a phenomenal duplicate of swamp-inverted-me.12 But this seems absurd, for just the reason Tye gives: it seems absurd that two non-evolved creatures could differ systematically in their phenomenal properties just in virtue of one having materialized with the lenses in, and the other having the lenses added just moments after having materialized. But this case can then be used to give a reductio of the view that the ‘best one-toone pairing’ cannot change over time. For consider which is the best one-to-one pairing of phenomenal states to colors for swamp-me and swamp-inverted-me at the moment at which they materialize. Presumably the best pairing for swamp-me will map BLUE to blue, and will map YELLOW to yellow—these colors are, after all, the typical causes of these phenomenal properties for me when I am not wearing the inverting lenses. And presumably the best pairing for swamp-inverted-me will map YELLOW to blue, and will map BLUE to yellow—these colors are, after all, the typical causes of these phenomenal properties for me when I am wearing the inverting lenses. Given the supposition that the best one-to-one pairing for a subject can’t change, this entails that, in the case where swamp-me has the lenses added moments after he materializes, he differs from swamp-inverted-me in surprising ways. If we present the two with a surface which is blue, the two creatures—which are now intrinsic duplicates—will each instantiate the phenomenal property YELLOW, 12 Or we could deny that the addition of the lenses has any effect at all on swamp-me’s phenomenal properties. But that seems even worse.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

254 functionalist theories of content but will differ in the color property they sense the surface to have. Swamp-me will be representing the surface as yellow, whereas swamp-inverted-me will be representing it as blue. There are two separate worries here. The first is that this result conflicts with the intentionalist thesis argued for in Chapters 4–8. But the second is just that this result seems absurd, for the same reason that it seems absurd that a swampbeing’s phenomenal properties could vary dramatically depending upon whether it materialized with inverting lenses in place, or had those lenses inserted moments after coming to be. It seems just as absurd that the contents of that being’s experiences could vary dramatically for the remainder of its existence due to this sort of seemingly inconsequential difference in its origins. So let’s suppose instead that the proponent of a holist functionalist theory says that the best pairing of states with represented colors can vary over time. This avoids the problems just discussed, with the differences between the swampbeings who differ only slightly in when the inverting lenses are inserted. But, on plausible assumptions, this sort of theory is open to a diachronic version of the argument brought against indication theories of content: they entail the possibility of psychedelic phenomenology plus constant representation of color properties. To have a particular case in mind, let’s stick with swamp-me, who materializes as a duplicate of me without the inverting lenses, and then has the lenses permanently inserted sometime very soon after he materializes. Let t be the time at which the lenses are inserted. Then t seems like the most plausible candidate for the time at which the ‘best pairing’ of internal states to represented colors changes. Suppose that swamp-me is looking at a blue surface which he senses to be blue, and instantiates BLUE in the time leading up to t. After t, he will instantiate YELLOW. Since the best pairing has changed, he will still be sensing the color of the surface to be blue. Given that there appears to be nothing in the case which would prevent swamp-me’s memory from working normally, this will be a case of psychedelic phenomenology þ constant representation of color properties. This is not an argument that this theory falsifies our constraint [C]; there is after all no time at which both BLUEswamp-me and YELLOWswamp-me stand in the representation relation R to the same color. But we do have consecutive moments t1 and t2 such that the color which BLUEswamp-me represents at t1 ¼ the color which YELLOWswamp-me represents at t2. And that is all we need to generate an instance of Scenario A.13 13 One might say that there is no one moment when the best pairing switches; perhaps the best pairing can only shift over a sufficiently long interval of time. But this runs into the problems,

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist theories of content 255 So, while the kind of limited holist theory of content under discussion appears to be the best option for the functionalist, it faces a dilemma. Either the best pairing of states with represented colors can change, or it cannot. In the latter case, it is hard to see how to avoid saying implausible things about the phenomenal and representational properties of non-evolved creatures. In the former case, it is hard to see how to avoid the view entailing the possibility of psychedelic phenomenology þ constant representation of color properties.

discussed in Chapter 7, with the denial of the time constraint. A further sort of problem for limited holist theories of this sort arises if they permit ties between distinct mappings from internal states to represented colors. The obvious way to handle these cases would be to treat them as cases of representational indeterminacy. But this too can lead to problems, since it is implausible that a RED and GREEN state could both represent a surface as ‘red or green’—and, if it could, we could use this possibility to construct an instance of scenario A. See the discussion of borderline cases in Chapter 7.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

36 Functionalist Definitions of Phenomenal Relations As noted earlier, it is quite plausible that we can stand in phenomenal and nonphenomenal relations to the same contents. For this reason, even a successful functionalist theory of content would not be an adequate theory of phenomenal properties, for we’d still need a functionalist account of what it is for a subject to be in a phenomenal rather than a non-phenomenal relation to the relevant content. And, if the argument of Chapter 27 is correct, even this would not be quite enough, since we also need our theory to distinguish between distinct phenomenal relations. In this chapter I’ll discuss in turn three topics: (i) whether a standard sort of functionalist theory can provide a theory of phenomenal vs. non-phenomenal relations; (ii) whether the same sort of theory can distinguish between distinct phenomenal relations; and (iii) whether a certain sort of higher-order theory might fare better with either task. Beginning with (i), it will be useful to focus attention on two sorts of nonphenomenal relations in which subjects can stand to contents: non-phenomenal sensory states (like subpersonal processing states, and the states of blindsighters), and cognitive states, like judgements and beliefs. The proponent of our reductive program owes an account of the distinction between these representational relations and the sensing relation. The most well-developed functionalist strategy for handling this problem is due to Michael Tye, who suggested the following two part solution:1 (a) Phenomenal states are poised, in the sense that they ‘stand ready and available to make a direct impact on beliefs and/or desires.’2 (b) The contents of phenomenal states are nonconceptual.

1 2

Note, though, that Tye’s view has since changed. See Tye (2009) and Tye (forthcoming). Tye (2000), 62. See also Tye (1995), Ch. 5.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist definitions of phenomenal relations 257 The ‘poisedness’ condition (a) is intended to handle the cases of subpersonal representations and blindsighters, but it is less plausible that it will help to distinguish phenomenal states from judgements and beliefs, since these are typically poised to have a direct impact on desires and (other) beliefs. This is where (b)—the nonconceptual condition—is supposed to come in. Tye explains what he means by ‘nonconceptual’ as follows: ‘[T]o say that a mental content is nonconceptual is to say that its subject need not possess any of the concepts that we, as theorists, exercise when we state the correctness conditions for that content.’3

This raises a question which we’ve already encountered: what does it mean to ‘possess a concept’? As noted earlier, a natural idea is that to possess a concept is to be able to have thoughts involving that content. But in the present context this is more than a natural idea: this is really what possessing a concept must mean if the notion is to play the required role. If possessing a concept required some stronger condition, then judgements would be nonconceptual, which, on Tye’s theory, would make them phenomenal states. If it were a substantially weaker condition, then it would be hard to see why, on that weakened definition, perceptual states would qualify as nonconceptual. Indeed, even without weakening the conditions on concept possession, a strong case can be made (as argued in Chapters 16 and 17) that—at least in the case of normal adult human beings—perceptual representation of a content requires the ability to have thoughts involving that content. Tye’s claim about sensed contents, after all, immediately entails the negation of the Availability Requirement, for which we have at least a reasonably plausible argument. But let’s set that sort of argument to the side; even if the Availability Requirement were false, there are more basic problems with the use of condition (b) to explain the distinction between phenomenal and non-phenomenal states— problems that, I think, point to general problems with functionalist accounts of phenomenal relations, whether or not they take exactly the form of Tye’s theory. Assuming that concept possession is glossed in terms of the ability to have certain thoughts, a first pass at the intended interpretation of ‘nonconceptual’ is: a mental content is nonconceptual iff its subject need not be able to have thoughts or beliefs with that content

3

Tye (2000), 62. See also Tye (2006), 507.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

258 functionalist definitions of phenomenal relations But this is not (as Tye and others have noticed) a condition on types of contents at all, but rather a condition on types of mental state—after all, contents themselves don’t ‘have subjects’—mental states do. So better to say: a mental state type M is nonconceptual iff its subject need not be able to have thoughts or beliefs with its content This is progress. But we still need to do some more unpacking, because we need to make the role of the modal locution—‘need not’—clear. The intended interpretation seems to be something like this: 8M M is nonconceptual iff 8s ½ðs is in MÞ !  ðs is in a state of type M with content p & S is not able to have beliefs or thoughts with content pÞ This could be restated as a principle about relations to contents rather than mental state types; since the differences don’t matter for present purposes, I’ll stick for now with Tye’s preferred formulation in terms of properties of mental states. It’s worth emphasizing that the modal aspect of this definition of nonconceptual content is not dispensable. If we were to remove the possibility operator from the above, this would lead to an immediate problem with the theory. For even if we suppose for the sake of argument that the Availability Requirement is false and that I sometimes perceptually represent the world in ways which I can’t also represent in thought, this is clearly not true of all of my perceptual experiences—it is not, for example, true of my perceptual experience when I’m looking at an otherwise featureless solid white expanse. But this fact does not rob my experience of its phenomenal character—as it would, were the ‘unmodal’ version of the above correct. But the use of this modal operator in the account of what it is for a state to be nonconceptual is, given Tye’s explanatory aims, puzzling. This is because the above explanation of what ‘nonconceptual content’ means builds in facts about sameness of mental state type across possible worlds; it explains what it means for a state to be nonconceptual in terms of what is possible when a subject is in a state of that type. But this raises circularity worries—worries which are even more fundamental than the worries that one might have in this context about relying on facts about what it is for a state to be a thought or belief. After all, the whole point of this use of the notion of nonconceptual content is to provide a naturalistically acceptable account of a type of mental state—namely, the phenomenal ones. But we can no more do this in terms of facts about what is possible for people in states of that type than we can explain facts about personal identity in terms of facts about what properties it is possible for that person to have. One might surmount this worry by appending to Tye’s account an independent theory of sameness of mental state types—but the worry is that it

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist definitions of phenomenal relations 259 is hard to see how to do this without just giving a theory of the relevant state types, which is what Tye’s theory was supposed to be doing. This seems to me to be a very general problem for functionalist theories of phenomenal relations (and indeed any representational relations). Let’s suppose, with the functionalist, that what it is to stand in a phenomenal relation to a proposition is for the subject to be in some internal state which (a) is a phenomenal state and (b) has the relevant proposition as its content. It is hard to see how we could analyze the property of being a phenomenal state without appealing to facts about the possible properties of creatures who are in that state. But it is very hard to see how this could mean anything other than ‘creatures who are in states of the same type.’ And it is hard to see how we could spell out ‘of the same type’ without the very analysis of ‘is a phenomenal state’ which we were hoping that our functionalist theory would provide. .

. .

Let’s turn now to question (ii) mentioned at the start of this chapter. The problem of distinguishing phenomenal from non-phenomenal relations is hard enough. But we know from the foregoing that this is only half the problem of reducing the attitudes: we also need an account of what distinguishes distinct phenomenal relations, like attention and perceptual representation. After all, we need an account of the phenomenal difference between pairs of experiences which differ with respect to the location of the subject’s attention—and, as the examples of Chapter 27 show, we can’t always give such an account in terms of a difference in the contents to which the subject stands in phenomenal relations, since in some cases there will be no such difference. Obviously neither of the strategies just discussed—appeals to nonconceptual content or poisedness—will work here, since typically the contents of attentional states are poised to directly affect belief and desire, and it seems plausible that they will be nonconceptual by Tye’s criterion iff the contents of perceptual states are. The functionalist can hardly take the distinction between sensing and attending as primitive. So one naturally looks here for some functional difference—some connection which attending to x has to other mental states which merely sensing x does not. But if we think about the sorts of covert attentional shifts discussed in Chapter 27, it is not easy—for me at least—to see what this functional difference could consist in. No matter which point of intersection my attention is focused on, I’m able to demonstratively refer to both, immediately form beliefs about both, undertake actions involving both, etc. What functional difference could explain the phenomenal difference between subjects who differ only slightly in the focus of their attention? .

. .

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

260 functionalist definitions of phenomenal relations While some of these criticisms of functionalist theories of phenomenal relations generalize to various other sorts of functionalist approaches, they don’t apply to theories which might try to explain phenomenal relations in terms of the higherorder attitudes of the subject in question. Although higher-order theories are often thought of as competitors to functionalist theories, they fit the very broad characterization of functionalist theories given in Chapter 34, since they explicate the properties of mental states which are their targets in terms of accidental relational properties of those mental states—namely, the relational property of being the object of a certain sort of higher-order mental state. Higher-order theories of consciousness have been developed with great subtlety over the last few decades, and there’s no chance here of doing justice to the wide variety of theories of this general sort. But—if we think of such theories as part of an attempt to give an account of phenomenal relations—the foregoing gives us the resources to formulate a dilemma for these theories. The idea behind a higher-order theory of phenomenal relations will presumably be that whether or not a subject bears a phenomenal relation to content is a matter of whether the subject has an appropriate higher-order attitude toward her own relation to that content (that is, toward her own mental state). We can then ask the proponent of a higher-order explanation of phenomenal relations about the content of the relevant higher-order state. The higher-order state could represent the content of the first-order state, the content together with the relation to which the subject stands in that content, or these things together with the fact that that relation is a phenomenal relation. These three options for the higher-order theorist can be illustrated by thinking of a subject who senses some content p. Then the higher-order state might have any of the following contents: (i) that the subject bears some attitude relation to p (ii) that the subject bears the attitude of sensing to p (iii) that the subject bears the attitude of sensing to p and that sensing is a phenomenal relation The worry is that none of these will do the job. To fix ideas, it will be useful to introduce two imaginary subjects: A is, among various other mental states, sensing that p Setting aside possible higher-order representational differences, B is a mental duplicate of A, but for the fact that she is not sensing that p, and is instead thinking that p

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

functionalist definitions of phenomenal relations 261 We should agree that A and B will be instantiating distinct phenomenal properties. What we want is an explanation of this difference. We know, given the description of A and B, that we will not get it from our theory of content; so we must get it from our theory of phenomenal relations. The question is whether the higher-order theorist can provide the missing piece. Let’s now consider our three ways of construing the contents of the relevant higher-order states. The problem with (i) is obvious: if we add to the story about both A and B the fact that each represents herself as bearing some attitude relation to the proposition p, we will have simply failed to find a difference between A and B, and hence failed to find a higher-order difference which could explain the phenomenal difference between them. The problems with (ii) are more subtle. Let’s add to the story about A that she represents herself as sensing p, and add to the story about B that she represents herself as thinking p. We’ve thus succeeded in finding some higher-order difference between A and B; what is less clear is that we’ve succeeded in explaining the phenomenal difference between A and B. For why should representing oneself as thinking vs. representing oneself as sensing make such a big difference? Surely not just the fact that one is representing oneself as standing in distinct relations to contents; no such massive phenomenal difference is to be found between two subjects, one of whom represents herself as judging p and the other of whom represents herself as supposing p. Hence, it seems, the phenomenal difference between A and B must, on view (ii), be explained in terms of some important difference between sensing and thinking themselves. But this is just the difference of which we wanted our theory of phenomenal relations to give us an account. Here is a different way to put the point. A theory of phenomenal relations should be such that, given a specification of all of the subject’s mental states, the theory enables us to determine which of them are phenomenal states and which are not. But now suppose that we have our pair of subjects, one of whom is sensing that p and thinks that she is sensing that p, and one of whom is thinking that p and thinks that she is thinking that p. What could our theory of phenomenal relations say which would enable us to determine that the first subject’s state of sensing that p is a phenomenal state, whereas the second subject’s state of thinking that p is not? The theory can’t say, ‘A state is phenomenal if the subject has a higher-order representation of herself as standing in a particular relation to its content,’ since both subjects satisfy that description. The theory could say, ‘A state is phenomenal if the subject has a higher-order representation of herself as standing in a particular relation to a content, and that relation is the sensing relation.’ But then we’ve simply built into our theory the thing that we wanted it to explain; the fact that sensing is a phenomenal relation is taken as primitive,

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

262 functionalist definitions of phenomenal relations rather than explained in terms of some higher-order cognition of the relevant mental state. One might think that option (iii) for the higher-order theorist solves this problem; perhaps the key is that the subject represents sensing as a phenomenal relation, but does not represent thinking as a phenomenal relation. But this faces two problems. First, it entails the wildly implausible result that any creature which instantiates phenomenal properties—like, presumably, dogs—takes itself to stand in phenomenal relations. (I don’t even think that I took myself to stand in phenomenal relations until a few years ago, when I began thinking of phenomenal properties in these terms.) Second, it neglects the existence of distinct phenomenal relations. For we can also imagine a subject C who is a mental duplicate of A except for possible higherorder differences, and except for the fact that whereas A is attending to q, C is attending to a distinct content r. Given that (as discussed in Chapter 27) the content of a subject’s attentional state is always part of the content of which she senses, A and C might (accurately) each represent themselves as sensing q and r and as bearing phenomenal relations to q and r. The only difference is that A will (we can suppose) represent herself as attending to q whereas C will represent herself as attending to r. But then we can ask the same question we asked of option (ii): given that both A and C represent themselves as standing in a phenomenal relation to the relevant contents, what explains the difference between the phenomenal properties of A and C? This difference must, it seems, be explained in terms of the difference between the phenomenal relations represented—between, that is, sensing and attending themselves. And, as before, this is just the difference that we wanted our theory of phenomenal relations to give us an account of.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

37 Identity Theories Let’s turn to the second attempt to give a physicalist account of phenomenal properties: identity theories, which identify phenomenal properties with intrinsic physical properties of subjects. As before, my aim here is not to give anything like a full evaluation of the prospects of identity theory. Instead, I want to ask a much more limited question: does the foregoing discussion of phenomenal properties provide—as we saw that it did in the case of functionalism—any reason to doubt an identity theory for phenomenal properties? So I won’t engage with much-discussed topics in this area, like the multiple realizability argument against identity theories, except insofar as the sort of view of phenomenal properties I’m discussing sheds new light on those issues. Though the connection between the phenomenal and the representational doesn’t bear as directly on identity theory as on functionalist theories, I think that there are four points of contact between them worth discussing: the consequences of the view that sensed contents include as their constituents contingently existing external things; the relationship between the falsity of Distinctiness/Discriminability and the multiple realizability argument against identity theories; a potential problem for identity theories posed by cases of phenomenal match; and a question about the reducibility of phenomenal relations. I’ll discuss these in turn.

Phenomenal Properties and Sensing Relations to Contingent Things It seems that identity theory is inconsistent with Simple Identity. For that view identifies phenomenal properties with representational properties which involve relations to contingently existing objects, like golf balls. But if we conjoin this with the claim that phenomenal properties are identical to intrinsic physical properties (like neural properties) then—given the serious actualist thesis that only existing things can stand in relations—we have the conclusion that the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

264 identity theories relevant intrinsic properties are such that their instantiation entails the existence of some particular golf ball. But that’s not a property that any neural state, or any other intrinsic physical property of a perceiving subject, could have.1 Another way to run basically the same argument, but without the serious actualist assumption, is to say that if phenomenal properties are identical to representational properties which involve relations to things like individual golf balls, then phenomenal properties must be externalist properties (in the sense that they fail to supervene on the intrinsic properties of their bearers). It is, after all, very hard to see how we could explain what makes one golf ball rather than another a constituent of the content of a given sensing relation other than some connection between the particular golf ball and the subject of the experience; and such connections could vary without any change in the intrinsic physical properties of the subject of the experience. But identity theories identify phenomenal properties with intrinsic neural properties; hence identity theories are false. Those antecedently skeptical of externalism about phenomenal properties might find this last challenge to identity theories less than persuasive; but some lessons from the foregoing call the credentials of this skepticism into question. For one thing, we might recall the example of the property identity involving heat. The idea that phenomenal properties are externalist is less surprising once we see that a reasonable case can be made that they are identical to representational properties, and that the relevant representational properties—those which involve the sensing relation—are externalist. Second, this externalist thesis, given the falsity of Discriminability/Distinctness, leaves open the possibility that the phenomenal properties of intrinsic duplicates are always indiscriminable; and perhaps the intuition that intrinsic duplicates must instantiate indiscriminable phenomenal properties is the intuition that defenders of phenomenal internalism are really after. The best move for the identity theorist in response to the problem posed by Simple Identity is to borrow a page from the proponent of the Phenomenal Content Thesis, and pursue the strategy of constituent deletion. On this view, while subjects of experience do sometimes instantiate representational properties which essentially involve relations to contingently existing objects, these are not the representational properties with which we should identify phenomenal properties. For anyone who instantiates an object-involving referential property like that corresponding to the open sentence

1 This is of the same form as the ‘separation arguments’ discussed by Pautz (ms.)—though there the focus is on relations to properties, rather than contingent particulars.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

identity theories 265 x senses that o is F relative to some assignment of a value to ‘o’ also instantiates the non-objectinvolving property corresponding to There is some value C of the content function that y is F such that x senses C Properties like this, for all we have said, might supervene on subjects’ intrinsic physical properties; so there is no bar to the identity theorist identifying physical properties with them. If a parallel move is made in the case of natural kinds, then we lose our case for externalism about phenomenal properties, and the argument against internalist identity theories goes away. This shows that the list of views about phenomenal properties given earlier— Distinctness, the Phenomenal Content Thesis, and Simple Identity—is not exhaustive. That assumed that, given that we bear sensing relations to contents with contingent things as constituents, the only reason for identifying phenomenal properties with a subset of the sensing properties not including these externalist properties was allegiance to Distinctness/Discriminability. But that is not true; one might also be motivated by the view that phenomenal properties are intrinsic physical properties of subjects, and hence that they cannot be identical to any relations involving external things. The result is that, even if the identity theorist can’t endorse Simple Identity, she can endorse an identity thesis about the phenomenal and the representational which is almost as simple—it differs from Simple Identity only in employing constituent deletion to remove contents with objects and natural kinds as constituents from the relevant identities. This is a view which has many of the attractions of the Phenomenal Content Thesis—including the considerations given in favor of that thesis at the end of Chapter 29—but without the baggage of Distinctness/Discriminability. One might wonder, though, whether we can mount an argument against identity theories parallel to the one given against the Phenomenal Content Thesis. To do so we would need to identify externalist contents which are, like colors, phenomenally sneaky rather than phenomenally silent. If externalist contents—like objects or natural kinds—were such that representation of those properties could sometimes make a difference to the phenomenal properties of the subject of the experience, one might think that this would entail that intrinsic duplicates could sometimes differ in their phenomenal properties. And this would of course be bad news for the identity theorist. To mount such an argument, what we would need is an example of a pair of discriminable experiences which differ only with respect to the representation of

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

266 identity theories objects or kind properties. And one might think that we already have one on hand. In Chapter 20, I discussed Siegel’s example of two experiences of some pine trees, one before and one after the acquisition of an ability to recognize pine trees. As Siegel says, it seems quite plausible that these experiences will be discriminable, and hence differ phenomenally—the pine trees might well seem to ‘stand out’ from the other trees more after acquisition of the recognitional disposition than before. If that is right, does this pair of experiences provide an example of the sort we need? Not quite. If we accept Siegel’s view of the contents of the two experiences, they differ in that one represents some of the trees as belonging to some kind K, whereas the other experience does not. So, in the simplest case, if the content of the first experience is some content p, the content of the second will be p & K(tree 1) & K(tree 2) & . . . adding conjuncts for each tree represented as being of the relevant kind. If K is the kind pine tree, then, given that there are kinds distinct but indistinguishable from pine trees, it is very hard to deny that the property of sensing the above conjunctive content is an externalist one. It is just hard to see how, absent some relation to pine trees, one could visually sense the trees to be of this kind, rather than some distinct but indistinguishable kind; and it is hard to see how that relation to pine trees could supervene on one’s intrinsic properties. But even if the two experiences do differ phenomenally and even if the representation of K does not supervene on one’s intrinsic properties and phenomenal properties are identical to representational properties, this does not show that phenomenal properties are not intrinsic. That is because the identity theorist can respond to this case by constituent deletion: by identifying the phenomenal property instantiated by the subject of the second experience with the representational property corresponding to the following open sentence: ∃k x senses that p & k(tree 1) & k(tree 2) & . . . This is the property of, for some kind k, sensing both that p and that various trees fall under k. There are two complementary reasons why this move is available, despite the phenomenal difference between representing the trees as pine trees and not representing them as belonging to any kind at all. The first is that even if Siegel is correct that one who senses this content will differ phenomenally from the subject of the first experience—who simply senses p—there is as yet no reason to believe that someone who senses any of the contents corresponding to the above open sentence will differ phenomenally from someone who represents the

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

identity theories 267 trees as falling under kind K in particular. The second is that even if the property of sensing that various trees are pine trees is an externalist one, it is far from obvious that the property of sensing that various trees fall under some common kind or other is.2 For a case to be genuinely problematic for the identity theorist, we need more than a pair of discriminable experiences which differ in that one represents some objects as falling under some kind and the other does not. We need, instead, a pair of kinds K, K* such that (i) there could be a pair of discriminable experiences which differ only in that whereas one represents some objects as falling under K, the other represents objects as falling under K*, and (ii) a pair of intrinsic duplicates could be such that one perceptually represents some objects as falling under K whereas the other represents some objects as falling under K*. But it is not easy to come up with such a pair.3

Distinctness/Discriminability and Multiple Realizability In Chapters 30–32 I described some types of experiences which were designed to show that certain properties are phenomenally sneaky. If these examples are convincing, they show that Distinctness/Discriminability is false. This is a line of argument which one might resist in various ways; what I now want to ask is how, if this argument is correct, its conclusion might bear on identity theories. In one way, as I’ll argue now, it is a point in favor of identity theories, since it provides some help in resisting one of the central arguments against identity theories of phenomenal properties. As is well known, one of the central problems for identity theories of phenomenal properties is the fact that they are inconsistent with the possibility of As Siegel is aware: see Siegel (2010), }4.3. The best case that can be made, I think, focuses on suggestive cases of associative agnosia. In some of these cases, subjects are unable to recognize the kinds to which things belong on the basis of visual experience, despite unimpaired visual representations of the shapes, colors, and other ‘lowlevel’ properties of things. It is plausible that the phenomenal properties of agnosiacs differ from those of people without agnosia. In most cases, agnosiacs are relevantly like Siegel’s pine tree viewer before acquisition of the recognitional capacity: they simply fail to represent kind properties which we might. In some cases, though, they appear to misrepresent the kind to which a thing belongs. If there is a phenomenal difference between the agnosiac’s misrepresentation of the thing as falling under kind K and the veridical representation of the thing as falling under K*, we might seem to have a case of the wanted sort. But in the best examples I know of like this—for example, a subject misrepresenting a can-opener as a key—condition (ii) is not obviously met. Externalist intuitions are not nearly as robust in the case of artefact kinds as in the case of natural kinds; and indeed it seems unlikely that intrinsic duplicates could differ in that one visually represents things as can-openers while the other visually represents them as keys. For an illuminating discussion of these cases, see Bayne (2009). 2 3

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

268 identity theories multiple realizability. Any argument from multiple realizability against an identity theory of some mental property M depends upon the provision of a pair of organisms at worlds and times which are such that (i) they both instantiate M and (ii) the physical property F which realizes M in one organism 6¼ the physical property G which realizes M in the other organism. A common way to challenge arguments of this sort is to bring out a tension between (i) and (ii). To the extent that the physical realizers differ, one might have doubts about whether the mental properties are really the same; to the extent that the mental properties are really the same, one might have doubts about whether it is really impossible to find a level of description at which the physical realizers are the same. In response to any alleged example of multiple realizability, the identity theorist has a choice about whether to reject premise (i) or (ii); and the right response can vary depending on the case.4 Hence one way of responding to an alleged instance of multiple realizability is to reject (i)—the thesis of the sameness of mental type. This can seem hard to do in the case of phenomenal properties if something like Distinctness/Discriminability is true, since it can seem plausible that physically quite different organisms could nonetheless be in indiscriminable states.5 Hence the fact that we have some independent reason to reject Distinctness/Discriminability looks like good news for the identity theorist inclined to take this path in responding to the multiple realizability argument. Of course, this is a bit of a double-edged sword, for if phenomenal properties are identical to representational properties, then the identity theorist who blocks an instance of the multiple realizability argument by distinguishing between indiscriminable phenomenal properties can do so only if she can also provide a plausible difference in the relevant representational properties of the two organisms. Whether this can always be done is an open question.

Identity Theories and Perceptual Constancies In another way, though, the falsity of Distinctness/Discriminability might be used in an argument against identity theories. Or, more specifically, one of the examples used against this thesis might also prove problematic for identity theories. These are the cases of phenomenal match discussed in Chapter 31. 4

See, for example, Bechtel and Mundale (1999) and Couch (2004). Shapiro (2000) argues, in effect, that (i) and (ii) are such that any reason to think that one is true in a given case will also be a reason to think that the other is false. 5 Not that this assumption can’t be questioned; as Zangwill (1992) points out, plausible arguments for it are not all that easy to come by.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

identity theories 269 Recall the (allegedly) indiscriminable experiences of the rotated quarter and the elliptical coin. I suggested that, given that both could occur at the end of a series of experiences which exhibit a standard sort of shape constancy, the two indiscriminable experiences could have different contents. Hence the identity theorist who accepts the identity of phenomenal and representational properties must say that the subjects of these two experiences are exhibiting distinct phenomenal properties P1 and P2, which then must be identical to distinct intrinsic physical properties I1 and I2. We’ll then have the following view: I1 ¼ P1 ¼ the property of sensing a round silver object rotated relative to me I2 ¼ P2 ¼ the property of sensing an elliptical silver object perpendicular to my line of sight Consider the series of experiences which, in the example described above, terminates in the subject instantiating P1. The subject first looks at a coin which is perpendicular to her field of vision, which she correctly represents as round. The coin is then rotated. If this is a standard case of shape constancy, the subject will represent the shape of the coin as constant, but its relative angle as changing. At the end of this interval, the subject is instantiating P1; but during the interval the subject instantiated, in succession, a series of many phenomenal properties P0 to Pn. Call this ‘Series 1.’ If the present version of identity theory is true, the subject will also, during Series 1, instantiate a series of intrinsic physical properties I0 to In which are, respectively, identical to the phenomenal properties instantiated by the subject during that interval. Now consider a different subject, who differs from the subject just described in only one way: immediately after instantiating the series of intrinsic properties I0 to In, the subject then, rather than instantiating I1, instantiates I2. Call this Series 2. Given the truth of identity theory, the subject of the Series 2 experiences will instantiate the same series of phenomenal properties as the subject of Series 1, except for the fact that the series of experiences will terminate in P2 rather than P1. But we know that P2 and P1 are indiscriminable. So it seems plausible that what it is like to be the subject of Series 1 will be indiscriminable from what it is like to be the subject of Series 2. Despite this, the representational properties of the subjects will, if identity theory is true, differ in a significant way. As mentioned earlier, the subject of Series 1 will be representing the shape of the coin as constant, and its orientation as gradually changing, throughout the series. So will the subject of Series 2— except that, suddenly, at the end, the subject will represent the shape of the coin as changing from round to elliptical, and its orientation as also changing instantaneously from tilted to perpendicular. And the subjects will exhibit this difference

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

270 identity theories despite the fact that the phenomenal properties instantiated by the two subjects will be, throughout the entire series, indiscriminable. This seems to me impossible, for much the same reason that cases of constant phenomenology þ psychedelic color representation seem impossible. It is hard to see how an apparently smoothly changing series of phenomenal properties could coexist with a radically discontinuous series of sensory representations. So, if identity theory implies the possibility of Series 2, this seems to me to be a serious problem for that theory. The identity theorist has some not-implausible replies available. First, she might reject the possibility of phenomenal match; as mentioned in Chapter 31, the assumption that there are indiscriminable phenomenal properties like P1 and P2 is a substantive one. Second, she might say that it is metaphysically impossible to instantiate I2 just after instantiating I0 to In, and hence that, for that reason, Series 2 is impossible. For this to work, the intrinsic properties with which the identity theorist identifies phenomenal properties would have to be historical properties, in the sense that they entail things about the past states of the subject. This is an available position; but it may be a slightly uncomfortable one for the identity theorist to take, since the same intuitions which militate in favor of the supervenience of phenomenal on intrinsic properties also militate in favor of the supervenience of phenomenal properties at a time on the subject’s intrinsic properties at that time. Indeed, partly because this position denies that phenomenal properties supervene on the intrinsic states of a subject at a time, it also leads to problems related to those posed by non-evolved creatures for functionalist theories of content. If the foregoing is correct, it is impossible for a non-evolved creature to materialize and instantiate I1 or I2—since these are properties which entail facts about the history of things which instantiate them. But it certainly seems possible for a nonevolved creature to materialize and instantiate a phenomenal property, P3, which is indiscriminable from P1 and P2. Since the creature can’t be in I1 or I2, we know that P36¼ P1 and P36¼ P2. If phenomenal properties are identical to representational properties, the creature must be representing some content. But this is puzzling, for two reasons. First, given that the content can’t be either of the contents associated with P1 and P2 (or, by parity, any content much like those), it is hard to see what it could be. Second, if identity theory is true, P3 must be identical to some intrinsic property of the creature. Call this property I3. I3 is apparently not a historical property, since it can be instantiated by a non-evolved creature. But then it seems that we can imagine a new series of experiences, Series 3, in which a subject instantiates I0 through In, and then instantiates I3. This series will exhibit the same combination of phenomenal

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

identity theories 271 continuity and representational discontinuity as Series 2, and hence seems impossible. But, given that I3 is by hypothesis not a historical property, it is hard to see how the identity theorist can avoid holding that Series 3 is possible.

The Reducibility of Phenomenal Relations The last point of contact between the identity theorist and the foregoing account of phenomenal properties begins with a point which is due to Hartry Field and has been emphasized in important recent work by Adam Pautz.6 Just as there are mental properties, there are mental relations. Phenomenal properties are examples of the former; phenomenal relations, like sensing and attending, are examples of the latter. Reductive theories often focus on phenomenal properties; but if a wholesale reduction of the mental to the physical is what is wanted, then it seems as though mental relations, no less than mental properties, need to be accounted for. Functionalist views of the sort discussed in Chapters 35 and 36 are at least of the right form to answer this challenge. Take, for example, the relation corresponding to the open sentence: x senses that something is F An indication theorist, to take one example, might simply identify this relation with the relation x is an internal state which has such-and-such functional role and indicates that something is F Of course, as we saw in Chapters 35 and 36, there are problems with functionalist theories of this sort. But we can, on a functionalist view, at least see how we might begin to supply a physical relation with which the above sensing relation between subjects and properties might be identified. But the same, as Pautz argues, cannot be said for the identity theorist. Suppose that the identity theorist proposes the following identities, where (as above) the ‘I’ properties are intrinsic properties and the ‘P’ properties are phenomenal properties: I1 ¼ P1 ¼ sensing that x is red21 I2 ¼ P2 ¼ sensing that x is red22 I3 ¼ P3 ¼ sensing that x is red23 ...

6

See Field (1978) and Pautz (forthcoming).

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

272 identity theories This gives us, for each monadic phenomenal and representational property, some physical property to which it is identical. But that still leaves us the representational relation of sensing. To what physical relation could this be identical? This is a tough question for the identity theorist to answer. Suppose for simplicity that sensing is a relation between subjects and properties. The problem is that, given that properties like I1 are intrinsic properties, they could be instantiated by subjects—like brains in vats—which bear no interesting physical relation (counterfactual or otherwise) to surfaces which instantiate the property they represent. The only option for the identity theorist who wants to reduce mental relations as well as mental monadic properties is to identify the sensing relation with the massively disjunctive relation corresponding to the following open sentence: (x instantiates I1 & y ¼ red21) ∨ (x instantiates I2 & y ¼ red22) ∨ (x instantiates I3 & y ¼ red23) ∨ . . . where we will need a disjunct, not just for every phenomenal property which can be instantiated by a human being, but for every phenomenal property which is possibly instantiated by anything. Perhaps a disjunctive analysis of this sort would be extensionally adequate; but it is hard to believe that it tells us what the sensing relation is. One worry here, which Pautz emphasizes, is that it is hard to see how we could manage to refer to this long disjunctive property, given that we have no acquaintance with many of the physical and phenomenal properties that will appear on the list. This worry is more pressing if we adopt the Russellian view of content defended here, according to which the semantic content of a two-place predicate is a dyadic relation— for, on that sort of view, it is hard to see how to avoid the absurd conclusion that ‘senses’ is synonymous with the massively long disjunctive predicate which expresses the property described earlier.7 This is no immediate objection to an identity-theoretic view of monadic phenomenal and representational properties. But it does suggest that the truth of that sort of theory—if the phenomenal is identical with the representational— would have as a surprising (and presumably unwelcome) consequence the irreducibility of phenomenal relations.

7

For more detailed argument that the identity theorist can provide no plausible physical relation with which the sensing relation can be identified, see Pautz (forthcoming), }5.

OUP CORRECTED PROOF – FINAL, 10/6/2015, SPi

Bibliography David M. Armstrong (1968). A Materialist Theory of the Mind. Routledge. David M. Armstrong (1978). Universals and Scientific Realism: A Theory of Universals Vol. II. Cambridge University Press. Tim Bayne (2009). Perception and the Reach of Phenomenal Content. Philosophical Quarterly 59 (236):385–404. William P. Bechtel, and Jennifer Mundale (1999). Multiple Realizability Revisited: Linking Cognitive and Neural States. Philosophy of Science 66 (2):175–207. John Bengson, Enrico Grube, and Daniel Z. Korman (2011). A New Framework for Conceptualism. Noûs 45 (1):167–89. George Berkeley (1713/1998). Three Dialogues Between Hylas and Philonous. Oxford University Press. José Luis Bermúdez (2007). What is at Stake in the Debate on Nonconceptual Content? Philosophical Perspectives 21 (1):55–72. Stephen Biggs (2009). The Scrambler: An Argument Against Representationalism. Canadian Journal of Philosophy 39 (2):215–36. Ned Block (1990). Inverted Earth. Philosophical Perspectives 4:53–79. Ned Block (1998). Is Experiencing Just Representing? Philosophy and Phenomenological Research 58 (3):663–70. Ned Block (1999). Sexism, Ageism, Racism, and the Nature of Consciousness. Philosophical Topics 26 (1):39–70. Matthew Botvinick and Jonathan Cohen (1998). Rubber Hands ‘Feel’ Touch that Eyes See. Nature 391, 756. Alex Byrne (2001). Intentionalism Defended. Philosophical Review 110 (2):199–240. Alex Byrne (2002). ‘Don’t PANIC: Tye’s Intentionalist Theory of Consciousness.’ In A Field Guide to the Philosophy of Mind (winter 2002/3). Available online at . Alex Byrne and Michael Tye (2006). Qualia ain’t in the Head. Noûs 40 (2):241–55. Alex Byrne and David R. Hilbert (2007). Truest Blue. Analysis 67 (1):87–92. Alex Byrne (2009). Experience and Content. Philosophical Quarterly 59 (236):429–51. David J. Chalmers (1996). The Conscious Mind: In Search of a Fundamental Theory. Oxford University Press. David J. Chalmers (2004). Epistemic Two-dimensional Semantics. Philosophical Studies 118 (12):153–226. David J. Chalmers (2006). Perception and the Fall from Eden. In Tamar S. Gendler and John Hawthorne (eds.), Perceptual Experience. Oxford University Press. David J. Chalmers (2011). The Character of Consciousness. Oxford University Press. David J. Chalmers (2012). Propositions and Attitude Ascriptions: A Fregean Account. Noûs 45 (4):595–639.

OUP CORRECTED PROOF – FINAL, 10/6/2015, SPi

274 bibliography Roderick Chisholm (1981). The First Person: An Essay on Reference and Intentionality. University of Minnesota Press. Philippe Chuard (2011). Temporal Experiences and Their Parts. Philosophers’ Imprint 11 (11):1–28. Jonathan Cohen (2008). Colour Constancy as Counterfactual. Australasian Journal of Philosophy 86 (1):61–92. Jonathan Cohen (2009). The Red and the Real: An Essay on Color Ontology. Oxford University Press. Mark B. Couch (2004). Discussion: A Defense of Bechtel and Mundale. Philosophy of Science 71 (2):198–204. Tim Crane (1988). The Waterfall Illusion. Analysis 48 (June):142–7. Marian David (2009). Defending Existentialism? In M. Reicher (ed.), States of Affairs. Ontos. Max Deutsch (2005). Intentionalism and Intransitivity. Synthese 144 (1):1–22. Fred Dretske (1995). Naturalizing the Mind. M.I.T. Press. Michael Dummett (1975). Wang’s Paradox. Synthese 30 (3–4):201–32. Andy Egan (2006). Appearance Properties? Noûs 40 (3):495–521. Andy Egan and James John, unpublished. A Puzzle about Perception. Gareth Evans (1981). Understanding Demonstratives. In Herman Parret (ed.), Meaning and Understanding. Clarendon Press. Gareth Evans (1982). The Varieties of Reference. Oxford University Press. Delia Graff Fara (2001). Phenomenal Continua and the Sorites. Mind 110 (440):905–35. Originally published under the name ‘Delia Graff.’ Hartry Field (1978). Mental Representation. Erkenntnis 13 (July):9–61. Kit Fine (2007). Semantic Relationism. Blackwell. William Fish (2009). Perception, Hallucination, and Illusion. Oxford University Press. Gottlob Frege (1906/1991). A Brief Survey of My Logical Doctrines. Reprinted in his Posthumous Writings, ed. P. Long (Wiley-Blackwell, 1991). Alex Grzankowski (2014). Attitudes Towards Objects. Noûs 49 (1):1–15. Gilbert Harman (1987). (Nonsolipsistic) Conceptual Role Semantics. In Ernest LePore (ed.), New Directions in Semantics. Academic Press. Gilbert Harman (1990). The Intrinsic Quality of Experience. Philosophical Perspectives 4:31–52. John Hawthorne and David Manley (2012). The Reference Book. Oxford University Press. Richard Heck (2000). Nonconceptual Content and the ‘Space of Reasons.’ Philosophical Review 109 (4):483–523. J. Van Heijenoort (1977). Sense in Frege. Journal of Philosophical Logic 6 (1):93–102. David R. Hilbert (2005). Color Constancy and the Complexity of Color. Philosophical Topics 33 (1):141–58. James John (2005). Representationism, Phenomenism, and the Intuitive View. Philosophical Topics 33 (1):159–84. Mark Johnston (2004). The Obscure Object of Hallucination. Philosophical Studies 120 (1–3):113–83. Mark Johnston (2006). Better than Mere Knowledge? The Function of Sensory Awareness. In Tamara Szabo Gendler and John Hawthorne (eds.), Perceptual Experience. Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/6/2015, SPi

bibliography

275

David Kaplan (1977). Demonstratives. In Joseph Almog, John Perry, and Howard Wettstein (eds.), Themes From Kaplan. Oxford University Press. Lorraine Keller (2012). Whence Structured Propositions? Dissertation, University of Notre Dame. Sean D. Kelly (2001). Demonstrative Concepts and Experience. Philosophical Review 110 (3):397–420. Jeffrey C. King (2001). Complex Demonstratives: A Quantificational Account. MIT Press. Jeffrey C. King (2007). The Nature and Structure of Content. Oxford University Press. Jeffrey C. King, Scott Soames, and Jeff Speaks (2014). New Thinking About Propositions. Oxford University Press. Uriah Kriegel (2002). Phenomenal Content. Erkenntnis 57 (2):175–98. Uriah Kriegel (2009). Subjective Consciousness: A Self-Representational Theory. Oxford University Press. Saul A. Kripke (1979). A Puzzle about Belief. In A. Margalit (ed.), Meaning and Use. Reidel. Saul A. Kripke (1980). Naming and Necessity. Harvard University Press. Joseph Levine (2001). Purple Haze: The Puzzle of Consciousness. Oxford University Press. David Lewis (1979). Attitudes de dicto and de se. Philosophical Review 88 (4):513–43. Brian Loar (1981). Mind and Meaning. Cambridge University Press. Graham Macdonald and David Papineau (2006). Introduction: Prospects and Problems for Teleosemantics. In Graham Macdonald and David Papineau (eds.), Teleosemantics: New Philosophical Essays. Oxford University Press. Fiona Macpherson (2006). Ambiguous Figures and the Content of Experience. Noûs 40 (1):82–117. Fiona Macpherson (2011). Individuating the Senses. In Fiona Macpherson (ed.), The Senses: Classical and Contemporary Readings. Oxford University Press. Peter J. Markie (1988). Multiple Propositions and ‘De Se’ Attitudes. Philosophy and Phenomenological Research 48 (4):573–600. John McDowell (1977). On the Sense and Reference of a Proper Name. Mind 86 (342):159–85. John McDowell (1994). Mind and World. Harvard University Press. Colin McGinn (1988). Consciousness and Content. Proceedings of the British Academy 74:219–39. G. E. Moore (1903). The Refutation of Idealism. Mind 12 (48):433–53. Bernhard Nickel (2006). Against Intentionalism. Philosophical Studies 136 (3):279–304. Alva Noë (2003). Causation and Perception: The Puzzle Unravelled. Analysis 63 (2):93–100. Daniel Nolan (2006). Selfless Desires. Philosophy and Phenomenological Research 73 (3):665–79. Matthew Nudds (2001). Experiencing the Production of Sounds. European Journal of Philosophy 9 (2):210–29. Casey O’Callaghan (2008). Seeing What you Hear: Cross-modal Illusions and Perception. Philosophical Issues 18 (1):316–38.

OUP CORRECTED PROOF – FINAL, 10/6/2015, SPi

276 bibliography Casey O’Callaghan (forthcoming). Not All Perceptual Experience is Modality Specific. In Mohan Matthen, Dustin Stokes, and Stephen Biggs (eds.), Perception and Its Modalities. Oxford University Press. Derek A. Parfit (1984). Reasons and Persons. Oxford University Press. Adam Pautz (2009). What Are the Contents of Experiences? Philosophical Quarterly 59 (236):483–507. Adam Pautz (2010). Why Explain Visual Experience in Terms of Content? In Bence Nanay (ed.), Perceiving the World. Oxford University Press. Adam Pautz (ms.). Why Consciousness Cannot Just be in the Head: A New Argument against Biological Theories. Adam Pautz (forthcoming). How Can Brains In Vats Experience Space? A Puzzle for Internalists. Forthcoming in A. Pautz and D. Stoljar (eds.) Themes from Block. MIT Press. Christopher Peacocke (1992). A Study of Concepts. M.I.T. Press. Christopher Peacocke (1998). Nonconceptual Content Defended. Philosophy and Phenomenological Research 63 (2):381–8. Christopher Peacocke (2001). Phenomenology and Nonconceptual Content. Philosophy and Phenomenological Research 62 (3):609–15. Alvin Plantinga (1983). On Existentialism. Philosophical Studies 44 (1):1–20. A. N. Prior (1969). The Possibly-True and the Possible. Mind 78 (312):481–92. Thomas Reid (1785/2002). Essays on the Intellectual Powers of Man. Pennsylvania State University Press. Mark Richard (1993). Articulated Terms. Philosophical Perspectives 7:207–30. Adina L. Roskies (2008). A New Argument for Nonconceptual Content. Philosophy and Phenomenological Research 76 (3):633–59. Adina L. Roskies (2010). ‘That’ Response Doesn’t Work: Against a Demonstrative Defense of Conceptualism. Noûs 44 (1):112–34. Bertrand Russell (1903). Principles of Mathematics. Routledge. Bertrand Russell (1910). On the Nature of Truth and Falsehood. In Bertrand Russell (ed.), Philosophical Essays. Longmans, Green. Nathan Salmon (1986). Frege’s Puzzle. Ridgeview. Nathan Salmon (1990). A Millian Heir Rejects the Wages of Sinn. In C. A. Anderson and J. Owens (eds.), Propositional Attitudes: The Role of Content in Logic, Language, and Mind. CSLI Publications. Susanna Schellenberg (2012). Sameness of Fregean Sense. Synthese 189 (1):163–75. Timothy Schroeder and Ben Caplan (2007). On the Content of Experience. Philosophy and Phenomenological Research 75 (3):590–611. Laura Schroeter (2005). Considering Empty Worlds as Actual. Australasian Journal of Philosophy 83 (3):331–47. Lawrence A. Shapiro (2000). Multiple Realizations. Journal of Philosophy 97 (12):635–54. Sydney Shoemaker (1994). Phenomenal Character. Noûs 28 (1):21–38. Sydney Shoemaker (2000). Phenomenal Character Revisited. Philosophy and Phenomenological Research 60 (2):465–7. Susanna Siegel (2006). Which Properties are Represented in Perception? In Tamar S. Gendler and John Hawthorne (eds.), Perceptual Experience. Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/6/2015, SPi

bibliography

277

Susanna Siegel (2009). The Visual Experience of Causation. Philosophical Quarterly 59 (236):519–40. Susanna Siegel (2010). The Contents of Visual Experience. Oxford University Press. Charles Siewert (1998). The Significance of Consciousness. Princeton University Press. Scott Soames (1987). Direct Reference, Propositional Attitudes, and Semantic Content. Philosophical Topics 15 (1):47–87. Scott Soames (2008). Drawing the Line Between Meaning and Implicature—and Relating Both to Assertion. Noûs 42 (3):440–65. Scott Soames (2009). Philosophical Essays: Natural Language: What It Means and How We Use It. Princeton University Press. Scott Soames (2012). Propositions. In Gillian Russell and Delia Fara (eds.), The Routledge Companion to the Philosophy of Language. Routledge. Jeff Speaks (2005). Is There a Problem About Nonconceptual Content? Philosophical Review 114 (3):359–98. Jeff Speaks (2009). Transparency, Intentionalism, and the Nature of Perceptual Content. Philosophy and Phenomenological Research 79 (3):539–73. Jeff Speaks (2010). Attention and Intentionalism. Philosophical Quarterly 60 (239):325–42. Jeff Speaks (2012). On Possibly Nonexistent Propositions. Philosophy and Phenomenological Research 85 (3):528–62. Jeff Speaks (2013). Individuating Fregean Sense. Canadian Journal of Philosophy 43:5–6, 634–54. Jeff Speaks (forthcoming-a). What are Debates about Qualia Really About? Philosophical Studies. Jeff Speaks (forthcoming-b). Is Phenomenal Character Out There in the World? Philosophy and Phenomenological Research. Jeff Speaks (forthcoming-c). Content and the Explanatory Role of Experience. In James Genone, Rachel Goodman, and Nick Kroll (eds.), Singular Thought and Mental Files. Oxford University Press. Robert Stalnaker (1984). Inquiry. Cambridge University Press. Robert Stalnaker (1999). Comparing Qualia Across Persons. Philosophical Topics 26 (1/2):385–406. Michael Thau (2002). Consciousness and Cognition. Oxford University Press. Brad J. Thompson (2006). Color Constancy and Russellian Representationalism. Australasian Journal of Philosophy 84 (1):75–94. Brad J. Thompson (2008). Representationalism and the Conceivability of Inverted Spectra. Synthese 160 (2):203–13. Charles S. Travis (2004). The Silence of the Senses. Mind 113 (449):57–94. Michael Tye (1995). Ten Problems of Consciousness: A Representational Theory of the Phenomenal Mind. MIT Press. Michael Tye (2000). Consciousness, Color, and Content. M.I.T. Press. Michael Tye (2000b). Shoemaker’s The First-Person Perspective and Other Essays. Philosophy and Phenomenological Research 60 (2):461–4. Michael Tye (2002). Representationalism and the Transparency of Experience. Noûs 36 (1):137–51. Michael Tye (2003). Consciousness and Persons: Unity and Identity. M.I.T. Press.

OUP CORRECTED PROOF – FINAL, 10/6/2015, SPi

278 bibliography Michael Tye (2006). Nonconceptual Content, Richness, and Fineness of Grain. In Tamar S. Gendler and John Hawthorne (eds.), Perceptual Experience. Oxford University Press. Michael Tye (2007). The Problem of Common Sensibles. In Ralph Schumacher (ed.), Perception and Status of Secondary Qualities. Kluwer. Michael Tye (2009). Consciousness Revisited: Materialism Without Phenomenal Concepts. M.I.T. Press. Michael Tye (forthcoming). Speaks on Strong Property Representationalism. Philosophical Studies. Peter van Inwagen (2004). A Theory of Properties. In Dean W. Zimmerman (ed.), Oxford Studies in Metaphysics, Volume 1. Clarendon Press. Timothy Williamson (1990). Identity and Discrimination. Blackwell. Timothy Williamson (2002). Necessary Existents. In A. O’Hear (ed.), Logic, Thought, and Language. Cambridge University Press. Ludwig Wittgenstein (1953). Philosophical Investigations. Macmillan. Nick Zangwill (1992). Variable Realization: Not Proven. Philosophical Quarterly 42 (167):214–19.

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

Index of Names Armstrong, David 13, 114 Audi, Robert x Bayne, Tim 267 Bechtel, William 268 Bengson, John 129, 131 Berkeley, George 148–50 Bermúdez, José Luis 68 Biggs, Stephen 37 Block, Ned 36, 42, 44, 160, 247 Botvinick, Matthew 186 Byrne, Alex x, 4, 12, 16, 30, 40, 44, 46, 145, 168, 245 Caplan, Ben 213 Chalmers, David 4, 35, 71, 93, 94, 131, 155 Chisholm, Roderick 111–12, 116 Chuard, Phillippe 234 Cohen, Jonathan 164, 187, 186, 227 Couch, Mark 268 Crane, Tim 230 Cutter, Brian 189 David, Marian 137 Deutsch, Max 211, 221 Dretske, Fred 45 Dummett, Michael 5, 221, 233 Dunaway, Billy 94 Eagle, Antony x Egan, Andy 52, 159, 162 Evans, Gareth 65, 73, 126 Fara, Delia Graff 222, 224–5 Field, Hartry 271 Fine, Kit 77 Fish, William 28 Frege, Gottlob 73, 78–9, 105 Gilmore, Cody 107 Grube, Enrico 129, 131 Grzankowski, Alex 9 Harman, Gilbert 251 Hawthorne, John 122 Heck, Richard 65, 67, 84, 129–32 Hilbert, David 40, 168, 227 Immerman, Daniel x, 122, 223

John, James 52, 178 Johnston, Mark vii, x, 117, 122, 133, 150, 153, 216–17 Keller, Lorraine 107 Kelly, Sean 125 King, Jeffrey C. 100–2, 112, 127 Klein, Colin 10 Korman, Dan 129, 131 Kriegel, Uriah 30, 60, 63, 164 Levine, Joseph 35 Lewis, David 111–12, 116 Loar, Brian 251 Macdonald, Graham 45 Macpherson, Fiona 173, 177, 197–9 Manley, David 122 Markie, Peter 112 May, Robert 84 McDowell, John vii, 71, 126–9 McGinn, Colin 187 Moore, G. E. vii, 26, 30, 34 Mundale, Jennifer 268 Nickel, Bernard 198 Noe, Alva 135 Nolan, Daniel 112 Nudds, Matthew 183 O’Callaghan, Casey x, 154, 179, 183, 199 Papineau, David 45 Parfit, Derek 50 Pautz, Adam x, 11, 264, 271–2 Peacocke, Christopher 63, 65, 113, 126, 197, 199 Plantinga, Alvin 137–41 Prior, Arthur 137 Reid, Thomas 50 Richard, Mark 127 Roskies, Adina 129 Russell, Bertrand 62, 76, 99–100, 105 Salmon, Nathan 77, 92 Schellenberg, Susanna 92 Schroeder, Timothy 213 Schroeter, Laura 93 Shapiro, Lawrence 268

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

280

INDEX OF NAMES

Shoemaker, Sydney 155–60 Siegel, Susanna 143–7, 183, 266–7 Siewert, Charles 211, 234 Soames, Scott x, 101, 105, 112, 117, 212 Stalnaker, Robert 5, 246 Stubenberg, Leopold x Thau, Michael 63, 168 Thompson, Brad 32, 71, 227 Travis, Charles 15–16

Tye, Michael x, 27–30, 63, 129, 157, 178–9, 186, 191, 214, 237, 245, 247–53, 256–9 Van Heijenoort, J. 79 Van Inwagen, Peter 103 Williamson, Timothy 38, 137, 139, 208–10 Wittgenstein, Ludwig 168 Zangwill, Nick 268

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

Index of Topics appearance properties 155–65 attention and higher-order theories of consciousness 262 as a distinct phenomenal relation 194–200 as representational relations 195–6 functionalist theories of 259 shifts in 33–4, 72, 144, 190–200 availability vii, 114, 118–35, 143–6, 153–4, 158–63, 166–7, 257–8 and appearance properties 158–63 and relativism about colors 166–7 Availability/Difference 122–4 Availability Requirement 125 demonstratives and 126–32 of de re thoughts 133–5, 153–4 of thoughts about natural kinds 143–6

and externalism about sensed contents 263–7 and multiple realizability 267–8 and phenomenal match 268–71 definition of 243–4 of phenomenal relations 271–2 illusion 16, 118, 128, 153, 181–2, 186, 214 Müller-Lyer 12–13 waterfall 229–30 intentionalism and the identity of phenomenal and representational properties 203–7 intermodal vs. intramodal 21–3, 59–64, 76, 96, 144, 173–201 local vs. global 21–3, 55–6, 187 transparency and 26–54 varieties of 21–5

color as hallucination-available 153, 216–17 constancy 32, 166–7, 226, 228 demonstrative concepts of 127–32 relativist views of 166–7 common sensibles 178–9 content as a kind of property 99–118 Fregean views of 62–3, 70–97, 109, 130–2, 138, 143, 162, 167, 194 non-propositional 108–18 possible worlds views of 63, 76, 229–30 Russellian views of 62–3, 96–118, 130, 135–6, 152, 162–3, 205, 272 two-dimensionalist views of 93–5, 131 see also propositions, representational properties, sensed contents Converse Barcan Formula 140

multiple realizability 164, 267–8

demonstratives 65–6, 126–32, 144, 154, 162, 208 Frege’s puzzle 76–96, 199–200 functionalism about phenomenal relations 256–62 about sensed contents 245–55 definition of 243–4 higher-order theories of consciousness 260–2 idealism 148 identity theory

nonconceptualism and intermodal intentionalism 62–4 arguments for 65–9, 129–32 definition of 61 distinguished from a view about concept possession 66–9, 126 pain 55–6, 174, 186–8 perceptual binding 179–88 perceptual constancy involving color 32, 166–7, 226, 228 involving objects 153–4 involving shape 227–31, 269–71 involving size 227 phenomenal continua 6–7, 49, 219–25, 233, 236, 238–9 phenomenal properties as identical to intrinsic physical properties 263–72 as identical to representational properties 203–6, 235–41 as properties of subjects rather than experiences 3–7 associated with attention 189–200 functionalist theories of 245–62 in relation to phenomenal character 4–5 indiscriminable but distinct 208–40 see also intentionalism, phenomenal relations

OUP CORRECTED PROOF – FINAL, 16/5/2015, SPi

282

INDEX OF TOPICS

phenomenal relations and attention 190–200 and cognitive phenomenology 189 and imagination 189 and the distinction between perception and bodily sensations 186–8 and the distinction between the senses 177–85 condition for distinctness of 175 definition of 174 functionalist theories of 256–62 identity theory and 271–2 phenomenal sneakiness and perceptual constancies 226–31 and phenomenal continua 219–25 and the representation of change 232–4 definition of 218 phenomenism 41–3, 52–4, 155, 167 propositional attitudes as a proper sub-class of relations to propositions 8–9 as relations to properties 99–119 distinguished from first-person representational relations 109–11 associated with experience 12–17 see also phenomenal relations, truth-sensitive relations propositions as a kind of property 99–107 as having constituents 105–7 see also content, representational properties, sensed contents

representational properties as properties of subjects 8–9 first-personal 108–18 how to define 9–11 see also content, propositional attitudes, sensed contents, truth-sensitive relations sensed contents and the representation of change 232–4 as contingently existing 137–42 as distinct from propositions 108–18 as including natural kinds 143–6 as including objects 133–6 as non-propositional 108–18 as Russellian 70–5 as self-representational 60–1, 191–3 spectrum inversion 42–5, 71, 155–69, 247–54 spectrum shift 42, 45, 169 Swampman 248–54 transparency of experience vii–viii, 26–34, 121, 123, 193–4, 214 negative view of 34 positive view of 27 truth-sensitive relations associated with experience 12–17 definition of 10–12 distribute over conjunction 184 to properties other than propositions 115–16