Reference and Consciousness

OXFORD C O G N I T I V E S C I E N C E S E R I E S REFERENCE AND CONSCIOUSNESS

OXFORD C O G N I T I V E SCIENCE SERIES General Editors MARTIN DAVIES, JAMES HIGGINBOTHAM, PHILIP JOHNSON-LAIRD, CHRISTOPHER PEACOCKE, KIM PLUNKETT

Published in the series Concepts: Where Cognitive Science Went Wrong Jerry A. Fodor Context and Content Robert C. Stalnaker Mindreading Stephen Stich and Shaun Nichols Face and Mind: The Science of Face Perception Andy Young

REFERENCE AND CONSCIOUSNESS JOHN CAMPBELL

CLARENDON PRESS • OXFORD

This book has been printed digitally and produced in a standard specification in order to ensure its continuing availability

OXFORD UNIVERSITY PRESS Great Clarendon Street, Oxford 0X2 6DP 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 in Oxford New York Auckland Bangkok Buenos Aires Cape Town Chennai Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi Sao Paulo Shanghai Taipei Tokyo Toronto Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States ©John Campbell 2002 The moral rights of the author have been asserted Database right Oxford University Press (maker) Reprinted 2003 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, 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 book in any other binding or cover and you must impose this same condition on any acquirer ISBN 0-19-924380-8

PREFACE

In May 1995, Naomi Eilan and Michael Martin organized a one-day meeting in London on 'Attention and Consciousness' in which I was to be cosymposiast with Jon Driver. Driver's talk, presenting results about the connections between acts of perceptual attention in different sensory modalities, seemed to me fascinating but difficult to interpret in terms of its implications for the character of experience. I found, however, an immediate and intuitive connection between Driver's findings and a problem that had long interested me, namely the connections between references to objects made on the basis of perceptions of them in different sensory modalities, such as sight or touch or hearing. To refer to an object on the basis of seeing it seems to require attention to that object; to refer to an object on the basis of hearing it or touching it seems to require attention to that object. So in principle, I thought, it ought to be possible to use results on relations between acts of attention in different sensory modalities to illuminate the relations between, for example, demonstratives referring to seen objects and demonstratives referring to touched or heard objects. In pursuing this line of thought after the conference, though, it quickly became apparent to me that the fundamental problem is to articulate the relation between attention to an object and knowledge of the reference of a demonstrative referring to it. The issue of the relations among the sensory modalities is touched on only glancingly in this book, though I think there are ready extensions of the present approach to that topic. Rather, I have gone back to the original topic of the London conference, to find the relations now between three phenomena: attention, knowledge of reference, and our experience of the world. In finding my way around the basic conceptual links between these three notions—attention, knowledge of reference, and experience—I have been greatly helped by a long series of sceptical challenges from Philippa Foot. Christopher Peacocke provided a detailed set of comments on four chapters from an early draft. Quassim Cassam and Timothy Williamson helped me see the exact shape of many particular arguments in an area whose scope, though narrowly defined, is quite far-reaching. And most of all, I have tried in one way or another to address Michael Dummett's deeply thought-through set of challenges for any account of knowledge of reference.

vi

Preface

On vision science, I learned a lot from discussions with Charles Spence, Vincent Walsh, Kia Nobre, Peter McLeod, Alan Allport, Andrew Parker, and Stephen Palmer. In this book 1 appeal only to some simple, robust points about visual attention. But these people helped me find my way to them, and to see why even those findings might be problematic. In connecting the empirical with the philosophical I was also much assisted by a centre for such work in the U.K., the Consciousness and SelfConsciousness group based at Warwick University, in particular to Naomi Eilan, Christoph Hoerl, Teresa McCormack, and Johannes Roessler. At various points I have had sustained and helpful discussion of the approach with Paul Boghossian, Bill Brewer, David Charles, Tamar Gendler, Alison Gopnik, John O'Leary Hawthorne, Sydney Shoemaker, Charles Travis, and David Wiggins. Jeremy Butterfield forced a number of changes to my discussion of causation in Chapter 12; Michael Bacharach provided a generous stimulus to the remarks on common knowledge in Chapter 8. The basic forum in which I have discussed these ideas has been a set of graduate classes given in Oxford in most terms since 1995. I have had countless illuminating and searching comments from many participants; in particular, I would like to thank Stephen Butterfill, Imogen Dickie, Kostja New, Hanna Pickard, and Oliver Pooley. I have given talks from this material at a number of universities, and always learned from the discussions: my immediate memories of illumination include moments at Barcelona, Berkeley, Cornell, CREA (Paris), Edinburgh, Glasgow, Lisbon, London, NYU, Oxford, Palermo, Riverside, Salzburg, St Andrews, Stirling, UCLA, Vassar, and Warwick. Portions of this material have had formal commentaries from Julien Deonna, Jerome Dokic, Rae Langton, Brian Loar, Kirk Ludwig, and Michael Martin. Various early versions or alternative versions of this material have been published as: 'Molyneux's Question', in Enrique Villanueva (ed.), Perception: Philosophical Issues, 7 (Ridgeview: Atascadero, Calif., 1996), 301-18, with replies by Brian Loar and Kirk Ludwig. 'Sense, Reference and Selective Attention', Proceedings of the Aristotelian Society, Supplementary Volume, 71 (1997), 55-74, with a reply by Michael Martin. Attention and Frames of Reference in Spatial Reasoning', Mind and Language, 12 (1997), 265-77. 'Joint Attention and the First Person', in Anthony O'Hear (ed.), Current Issues in Philosophy of Mind: Royal Institute of Philosophy

Preface

vii

Annual Supplement, 43 (Cambridge: Cambridge University Press, 1998), 123-136. 'Sense and Consciousness', in Peter Sullivan and Johannes Brandl (eds.), New Essays on the Philosophy of Michael Dummett: Grazer Philosophische Studien, 55 (1998), 195-211. 'Immunity to Error through Misidentification and the Meaning of a Referring Term', Philosophical Topics, 26 (1999), 89-104. 'Memory Demonstratives', in Christoph Hoerl and Teresa McCormack (eds.), Time and Memory (Oxford: Oxford University Press, 2001), 169-86. 'Berkeley's Puzzle', in Tamar Szabo Gendler and John O'Leary Hawthorne (eds.), Imagination, Conceivability and Possibility (Oxford: Oxford University Press, 2002). I could not have begun on this work in earnest without the benefit of a British Academy Research Readership, which I held during 1995-7, which gave me the time to learn something about these matters. In the course of writing this book, I was given a renewed sense of purpose, on many occasions, by Allison Harvey; so my final thanks go to Michel Treisman, who introduced us.

This page intentionally left blank

CONTENTS

Introduction 1. Experiential Highlighting

1 7

2. What is Knowledge of Reference?

22

3. Space and Action

46

4. Sortals

61

5. Sense

84

6. The Relational View of Experience

114

7. The Explanatory Role of Consciousness

132

8. Joint Attention

157

9. Memory Demonstratives

177

10. The Anti-Realist Alternative

194

11. Indeterminacy and Inscrutability

216

12. Dispositional vs. Categorical

235

Bibliography

255

Index

265

This page intentionally left blank

Introduction It is experience of the world that puts us in a position to think about it. Without experience, we would not know what the world is like. Those who have experiences of which we know nothing may be able to think in ways of which we know nothing. Those who do not have our experience of the world will not be able to think of it as we do. These points show up even at the simplest levels. Someone who has never had experience of the colours will not be able to understand the concepts of the different colours. As Locke said, a scholarly and indeed brilliant individual born blind, on pursuing an investigation into the nature of the various colours, may eventually say of scarlet, "Tis like the sound of a trumpet!', but this will not reflect any knowledge of what scarlet is. Or again, if I have picked something up and I conceal it from you, I may make a series of remarks to you about 'this extraordinary object', but the most direct way for you to interpret my remarks is by experiencing the object itself; seeing the object would mean that you knew what I was referring to. This connection between reference and consciousness is one that has been lost sight of. The complexity of the issues surrounding the two topics individually has led to a fragmentation of effort in theorizing about them. Reference is one problem, consciousness is another. This means that it is easy to lose sight of ways in which the two problems illuminate each other. We understand reference better when we keep in view that it is at bottom a phenomenon of consciousness. We know better how to think about consciousness when we know that, whatever else is true of it, consciousness of objects is what provides knowledge of reference. This gives some discipline to theorizing about consciousness: of what we say about consciousness, we can ask whether it contributes to explaining how consciousness can be what provides knowledge of reference. There is a way of focusing discussion of this connection which appeals to the notion of attention. Reference and attention are generally taken to be different topics. Reference is one of the fundamental problems in philosophy, though the nature of reference is not often directly discussed by scientists: the general problem is how our thoughts and words connect to the things about which we think and talk. Attention, in contrast, is little

2

introduction

discusse d by philosophers, but one of the most intensively studied phenomena in cognitive science. To see why attention seems important in cognitive science, suppose for a moment that you think about a human being as an engineer might view the thing. There is a lot of machinery here, in the human being, which can be deployed now on this task, now on that. What makes the difference between the machinery being deployed on this object as opposed to that object, is a difference in what you are attending to. Attention is an element in the control of the whole system. So reference and attention seem individually to be important topics, in philosophy and psychology respectively, yet to have little to do with one another. Suppose, though, that you and I are sitting side by side looking at a cityscape, a panorama of buildings. If I am to think about any one of those buildings, if I am to formulate conjectures or questions about any of those buildings, if I am to be able to refer to any one of those buildings in my own thoughts, it is not enough that the building should simply be there, somewhere or other in my field of view. If it is simply there in my field of view, though unnoticed by me, I am not yet in a position to refer to it; I cannot yet think about it. If I am to think about it, 1 have to single out the building visually: I have to attend to it. And if I want to refer to that building, to make a remark about that building for your benefit, I have to draw your attention to it. That is what pointing is. Pointing is at once the most basic kind of reference to objects, and the single most useful way of drawing someone else's attention to an object. So reference and attention are not just different topics. When we think about demonstrative reference in particular—that is, reference made to a currently perceived object on the basis of current perception of it—it seems that reference to the object depends on attention to the object. So we should expect that philosophical problems about reference and psychological theorizing about attention should be capable of illuminating one another. It is attention as a phenomenon of consciousness that matters for knowledge of reference. If I am to understand a demonstrative referring to an object, it is not enough merely that the object be there somewhere in my visual field; I have to attend to it. But the attention that is needed here is, as it were, a matter of experiential highlighting of the object; it is not enough merely that there be some shifts in the architecture of my information-processing machinery, remote from consciousness. To understand how knowledge of reference depends on attention we will have to understand the relation between the experiential highlighting of an object and underlying shifts in the configuration of information-processing machinery. We can make a rough division between a description of what knowledge of reference is, and a description of what knowledge of reference does. When I say that knowledge of reference is provided by conscious attention

Introduction

3

to the object, this is a description of what knowledge of reference is, and I will amplify it by looking in detail at what conception of consciousness, and what conception of conscious attention in particular, we need to be appealing to here. But we also have to look at what knowledge of reference does—what the significance is of knowledge of reference. I think that, given a preliminary characterization of knowledge of reference as provided by conscious attention to the object, we can get at this by looking at the functional role of conscious attention, and, in particular, by articulating the relation between attention as a phenomenon of consciousness and attention as an information-processing phenomenon. The general line of my argument is this. There are many information-processing systems that run automatically, in parallel, and they are not, usually, directly affected by attentional shifts. For example, low-level vision or reflex movement may be of this type. But sometimes information-processing is carried out under more central control. For example, the motor processing involved in executing a voluntary action is not run automatically: it is responsive to the conscious intentions of the subject. Similarly, if you are, as it were, visually interrogating a scene, for example if you are a policeman looking for signs of trouble, the visual information-processing you are performing is in the service of your conscious objectives; once you believe you are off-duty your vigilance may relax and the information-processing is simply no longer performed. This implies that conscious attention to an object—the experiential highlighting of that aspect of your perception—means that this aspect of your experience has a different functional role to the functional role it had when it was not so highlighted. That aspect of your conscious experience is now connected to certain information-processing machinery in a way in which it was not previously so connected. It is one of the factors which cause that information-processing machinery to swing into play, and it also defines the objective of the information-processing machinery. So we can ask the question just how conscious experience of the object manages to identify the target of the information-processing machinery. Rather than seeing consciousness and information-processing as systems causally insulated from one another, we have to see them as causally explaining each other's movements, and we have to find the commensurability between the way in which conscious experience identifies the objects we think about, and the way in which the targets of informationprocessing are identified. Let me set this project in context. One of the defining problems of twentieth-century philosophy of language was the relation between knowledge of the reference of a term, and the pattern of use that you make of the term: the inputs and outputs to propositions involving the term, the ways in which you verify or act on the basis of propositions involving the term.

4

introduction

The difficulty is this. There is a common-sense picture of the relation between knowledge of reference and pattern of use. On the common-sense picture, your knowledge of reference controls the pattern of use that you make of the term. You use the term the way you do because you know what it stands for. In the later Wittgenstein and in Quine, the problem is that they think the common-sense picture cannot be sustained. There is only the pattern of use: there is no such thing as a knowledge of reference which controls the pattern of use, and to which the pattern of use is responsible. In later Wittgenstein, the form the resulting problem takes is that the pattern of use now seems arbitrary, since it is no longer thought of as controlled by knowledge of reference. This is the issue he confronts in his discussion of rule-following. In Quine, the form the problem takes is that when we have only the pattern of use to consider, we find that it seems to leave underdetermined the ascription of meaning to the terms of a language. This is Quine's problem of the indeterminacy of translation. In the ensuing discussion, amazingly, the common-sense picture—that you use the word the way you do because you know what it stands for—is all but lost sight of. I am suggesting that, by thinking of knowledge of reference as explained by conscious attention to the object, we can see how to reinstate the commonsense picture. Just to illustrate again the fundamental point about conscious attention, suppose that you are sitting in a lecture and your mind wanders off a little, the lecture fails to grip. So you look around idly at the other people in the audience, your gaze resting now on this person, now on that. In effect, you highlight now one aspect of your experience, now another. In effect, you put a yellow highlighter now over one or another part of your visual experience, as you wonder about this or that person. Now suppose that as you sit there, your neighbour whispers in your ear, 'Who's that man there?' To understand the remark you need to know who he means. So you need to single out the right person visually. It really is conscious attention that matters here. If, as you listened to your neighbour, the neural circuitry underpinning visual awareness blinked out of operation, leaving your visuomotor circuitry intact, it could happen that your visuomotor system, remote from consciousness, managed to lock on to the right person, so that you could, to your surprise, point to the right person. Perhaps in some sense your finger might be said to know who was being referred to. But you would not know who was being referred to until normal service was resumed and you achieved experience of the person. Of course, in an ordinary case it can happen that I am consciously attending to, for example, one building, while you are talking about another one in front of us. In such a case I might still understand exactly what you are saying in your use of the demonstrative 'that building', because I

Introduction

5

am relying on my having consciously attended to the referent a moment ago. I shall mostly ignore such complications in what follows. Or again, it can happen that I understand you perfectly, even though I am not 'giving you my full attention', as we would ordinarily say, perhaps because I am simultaneously worrying about my debts. But that is consistent with my understanding of your remark depending on my having singled out the object in experience, even if I subsequently made little of your remark. Or again, it can happen that I am consciously attending to the building to which you are referring, even though I do not realize that it is the building to which you are referring. Conscious attention to the building is not in itself an understanding of your remark: I have to make a link between that conscious attention and the demonstrative you use. In the first two chapters I will begin on the characterization of that link, and pursue it through Chapters 6 and 7. Finally, there are subtle and delicate cases in which an understanding of the demonstrative may involve some combination of descriptive knowledge and conscious attention, cases such as those in which I refer to the cast shadow of an object I can see, a gleam of light on a car roof, or a reflection in a puddle. Demonstrative reference in such cases may involve thinking of the referent as dependent on a substantial thing. But I will set aside the exact analyses of such cases in vhat follows, concentrating on the primal case of reference to the substantial objects themselves. Summing up, there are three levels we can look at. There is the level of conceptual thought about your surroundings. There is the level of conscious attention to your surroundings, which is more primitive than the level of conceptual thought, and which explains your capacity for conceptual thought by providing you with knowledge of reference. And there is the level of the underlying information-processing systems, such as highlevel visual processing or the visuomotor system, which conscious attention can cause to swing into play. It is the liaison between conscious attention and the underlying information-processing sub-systems which provides you with your capacity to use the term, to verify propositions involving a term, or to act on the basis of such propositions. The problem which led both the later Wittgenstein and Quine to abandon the common-sense picture on which knowledge of reference controls the pattern of use that you make of a term was what we might call the derivation problem. They held that there was no way in which the pattern of use of a term could be derived from knowledge of the reference of the term. So Wittgenstein spoke about the lack of any blueprint from which we could derive knowledge of how to use a term, and Quine spoke of the myth of meaning laid up in a museum from which a pattern of use could be derived. But by focusing on the liaisons between conscious attention and

6

Introduction

the underlying information-processing involved in verifying or acting on the basis of propositions, we can see how to solve the derivation problem, reinstating the common-sense picture of use as controlled by knowledge of reference. The basic parallel is between the way in which conscious attention to an object can control the information-processing involved in acting on that object, or verifying facts about that object, and the way in which knowledge of the reference of a demonstrative term means that you can act intentionally on the object, or verify demonstrative propositions about the object. This is what makes it possible to identify conscious attention to the object as what provides you with your knowledge of the reference of the demonstrative. But further to making remarks about its functional role, we have also to consider what conscious attention to an object intrinsically is, that it can constitute knowledge of reference. Conscious attention is not merely a subjective state. It is a relation between the subject and the thing experienced. So it is not exhausted by the liaisons between conscious attention and the underlying information-processing. Russell thought of acquaintance as a cognitive relation more primitive than thought about an object, which nonetheless, by reaching all the way to the object, made thought about the object possible. I will argue that this provides a model for the way in which we think of conscious attention to an object. It is a state more primitive than thought about the object, which nonetheless, by bringing the object itself into the subjective life of the thinker, makes it possible to think about that object.

1

Experiential Highlighting 1. Intuition and Explanation Suppose you say to me, 'What is that mountain over there?' To understand your question I have to know which mountain you are talking about. I can construct various descriptions that I might use to interpret your demonstrative, 'that mountain'. For example, there are 'the mountain she is looking at', 'the mountain she is asking me about', and so on. But ordinarily, I do not need to construct any such description. I have a more direct way of interpreting the demonstrative, 'that mountain'. I can interpret it simply by looking to see which thing you mean. Ordinarily, my knowledge of which thing you are talking about is provided by experience of the object. And it is not just having the mountain in my visual field that matters. I have to single the thing out visually, I have to see it as a figure against a background. Of course, grasp of a demonstrative like 'that mountain' does not mean that I have to sustain visual attention on the object continuously throughout the period during which I am thinking or talking about it. We can discuss 'that mountain' during a period in which my attention shifts around the scene. It remains that my understanding of the demonstrative depends on the act of visual attention. You might acknowledge that ordinarily we would use visual information to interpret the demonstrative, but question whether it has to be conscious. The idea of visual information that is not conscious is made vivid by cases of blindsight. A blindsight patient is one who has suffered damage to his primary visual cortex, as a result of which he has no awareness of objects in one half of his visual field. Nonetheless, when forced to guess about what is in the blind field, he may be reliably correct about, for example, the orientation, direction, and sort of the object in the blind field, perhaps to his own as well as our surprise (Weiskrantz 1986). So here we seem to have visual information in the absence of consciousness. Couldn't this subject use such visual information to achieve an understanding of a visual demonstrative? If the answer is that there is a problem because a blindsighted subject will not in fact have enough visual information, or the wrong sort of visual information, we could try supposing, as a thought

8

Experiential Highlighting

experiment, that we have a subject who has all the relevant visual information but is not yet conscious of the object in the blind field. (Cf. Block 1997 for a related notion of 'super-blindsight'.) Suppose that our blindseer can guess reliably at the properties of an object—say, a tree—in his blind field, and can act appropriately with respect to it: pointing in the right direction, avoiding walking into it, and so on. Why should we not say that he is able to understand the demonstrative 'that tree'? Of course, anyone has to agree that the blindsight patient may be able to construct descriptions which will allow him to give some interpretation or other of the demonstrative, 'that tree', descriptions such as 'the tree in my blind field'. The issue is whether the blindseer has the very same way of interpreting the demonstrative as the ordinary subject has. That is, the question is whether for the ordinary subject, consciousness of the object is not completely idle in an understanding of the demonstrative. In that case, the blindseer could have exactly the same understanding of the demonstrative as the ordinary subject has. One problem here is to explain what role experience could be playing in providing you with knowledge of the reference of the demonstrative. That is, even if it is compelling to common sense that experience is needed for an understanding of the demonstrative, we still need an explanation, a theoretical analysis, of why that should be so. What work could experience of the object be doing? Another, much simpler question is whether it can be made compelling to common sense that the blindseer does not have the same understanding of the demonstrative as is provided by experience of the object. Let me start with the simpler question. I think that the simplest way to grasp the common-sense difference between the blindseer and the ordinary subject is to consider an ordinary case in which you and I are sitting at a dinner table with a large number of people around and you make a remark to me about 'that woman'. There are a lot of people around; I can't yet visually single out which one you mean. So on anyone's account, I do not yet know which woman you are talking about. Suppose now that we add to the example. My visual experience remains as before: a sea of faces. I cannot consciously single out the person you mean. All I get consciously is the sea of faces. But now we add some of what the blindseer has. You refuse to give me any further clues as to which person you mean, but you say, Try to point to the woman I mean'. As first I protest that I can't do that, since I don't know who you're talking about, but I do try to point, and to my surprise you say I'm pointing right at the person you mean. Suppose now that my conscious experience remains a sea of faces, but we extend the reach of my reliable guessing so that it encompasses everything the blindseer can do. So I can make reliable guesses about what the person is eating, wearing, and so on, as well as

Experiential Highlighting

9

reaching and pointing appropriately. But so long as my conscious experience remains a sea of faces, there is an ordinary sense in which I do not know who you mean. The problem here does not have to do with whether I am reliable: we can suppose that I am quite reliable in my guesses and we establish this over a series of such cases. The point is rather that I do not know who you mean until I finally look at where my finger is pointing, or look to see who is wearing the clothes I described in my guesses. It is only when I have finally managed to single out the woman in my experience of the room, when it ceases to be a sea of faces and in my experience I focus on that person, that I would ordinarily be said to know who was being referred to. So it does seem to be compelling to common sense that conscious attention to the object is needed for an understanding of the demonstrative. If common sense is right about this, then the problem is to achieve a theoretical understanding of why that should be so. In some contexts—though not, I think, the present one—the importance of visual attention has to do with the sheer quantity of information you are receiving about the thing. You receive much more information about an object whose physical image is focused on the relatively small part of the retina called the fovea than you do about an object whose image falls on some more peripheral part of the retina, just because of the mass and type of nerve-endings packed in to the fovea. So sometimes attention matters because it can involve foveating the object, hence receiving more information about it. But this is not the explanation of the importance of visual attention for knowledge of reference. We have to resist the analysis that says: 'I know which mountain you are talking about only when I have foveated the mountain, because only then do I have enough visual information about it to know which thing you are talking about'. Sheer quantity of information is not to the point here. Since Helmholtz 1866/1925, psychologists have distinguished between overt and covert shifts of visual attention. An overt shift of attention involves movement of the head, eyes, or body, generally to bring the thing into focus on the fovea. But it is possible to shift visual attention covertly, without movement of the head, eyes, or body. You can shift visual attention around even though your eyes hold steady on a particular fixation point (cf. e.g. Posner 1988). If you say to me, 'Don't look now, but that man over there seems to recognize you', I may exactly maintain fixation on my wine-glass, but look at him, as we say, 'out of the corner of my eye'. Even though I will not get a mass of visual information in this way, I may still manage to single him out visually to a sufficient extent that I know to which man you are referring. ' In any case, an appeal to the mere quantity of information being received about an object could hardly be what explains why experience of the object plays a role in knowledge of the reference of a demonstrative,

10

Experiential Highlighting

since you could in principle receive quite a lot of information about the thing without experiencing it. So how are we to characterize the role of experience? This is the question that will occupy us through the whole book. But for preliminary orientation, it may be helpful to indicate the direction of reply. I think that what explains the role of conscious attention here is the combination of two points. First, there is what I will call the relational character of experience of the object. So long as we do not analyse experience of the object as a matter of the object causing some experiential effect in you, we can regard experience as making the object itself, the categorical thing, available to the subject. Experience of the object makes the object itself, rather merely some sensory effect of the object, available to the subject. Secondly, there is the functional role of conscious attention. Suppose that you have a complex scene before you—say, a number of people you have never met before. As you look over the throng, you attend now to one, now to another face in the crowd. What does this experiential highlighting of one rather than another person come to? It affects the functional role of your experience of that person. It means that you are in a position to keep track of that person deliberately over time, you are in a position to answer questions about that person on the basis of vision, and you are now able to act with respect to that person. In an ordinary situation, until you highlighted that person in experience, you could have done none of those things. Both of these points need some explanation and defence. We also need an account of the relation between these two points: I will argue that this is provided by the notion of a 'way' of attending to the object, which is both strongly relational, being individuated in terms of the object in question, and needed in an account of functional role. I will discuss this point in Chapter 5. In Chapter 6, I will set out the first point, the relational character of experience. I will pursue the ramifications of the relational character of experience through Chapters 7-12. In Chapters 1-4 I will be principally concerned with the functional role of conscious attention, to which I now turn.

2. Conscious Attention I have said that conscious attention to an object, the highlighting of your experience of that thing, affects the functional role of your experience of the object. Having once consciously focused on the object, you are now in a position to keep track of it deliberately, to answer questions about it, and

Experiential Highlighting

11

to act on it. But how exactly can it happen that your experience of the object has this functional role? Take, for example, your ability to answer questions about the object. Do you, as it were, somehow read off the answers from the intrinsic nature of your experience of the object? Or is some further process at work? Let me contrast two different types of explanation you might give of your verbal report: A. 'I say it's blue, because it looks blue.' B. 'I say it's blue, because the cell-firing in V4 is registering blue at that location.' In discussing these contrasting explanations, I do not mean to be appealing to anything idiosyncratic about colours and colour experience, as opposed for example to shapes and shape experience. I could equally well work with the contrast: Al. 'I say it's square, because it looks square.' B1. 'I say it's square, because the cell-firing in V3 is registering squareness at that location.' I use colour only because it is a good example of a visible characteristic of a demonstrated object. V4 and V3 are areas of the visual cortex thought to be involved in, respectively, the processing of colour and the processing of form information. Filling out an explanation of type B or Bl might be held to involve emphasizing neural areas upstream or downstream of V4 or V3; for present purposes this point is not critical. I think the first thing we have to say is that both types of explanation are, on the face of it, legitimate. Of course, in a particular case they could be mistaken. Perhaps the object doesn't actually look blue, but you made a mistake anyway. Or perhaps it was cell-firing in some other area that caused your verbal report. But the styles of explanation seem right. Explanations of type A (or Al) are most familiar to common sense; it seems, to common sense, absolutely compelling that such explanations are often correct. They are the kinds of explanation you might give when asked your reasons for making a particular report. In contrast, explanations in the style of B are not in general readily known to the person making a verbal report. But scientific accounts of vision depend very heavily on the legitimacy of explanations of this type. The picture of verbal report as caused by visual information-processing in various specialized pathways is fundamental to much of the experimental work on the visual pathways. I will give some examples in a moment. Information-processing explanations challenge the unity of the philosophers' notion of a representational state. Philosophers sometimes think of beliefs and desires as representational and suppose that it will be

12

Experiential Highlighting

possible to explain all other representational notions in terms of their relations to beliefs and desires. However, the representational states postulated in information-processing explanations are not themselves beliefs and desires, nor does it seem possible to explain their contents in terms of their relations to beliefs and desires. According to Chomsky 1976, we do have a unitary notion of 'representational state' which covers both ordinary beliefs and the states involved in, for example, analysing the grammatical structure of a heard sentence. The difference is only that some of these states—in particular, ordinary beliefs and desires—are accessible to consciousness, while others, such as those postulated by linguists, are not accessible to consciousness. But, according to Chomsky, the distinction between representational states accessible to consciousness and those not accessible to consciousness is of no particular theoretical interest. An alternative reaction is provided by Searle 1995. According to Searle, there is only one notion of representation and it is constitutively linked to consciousness; there are only the ordinary beliefs and desires, of which we can in principle become aware, and it does not really make sense to talk in terms of other kinds of representational state. On this view, informationprocessing explanations are problematic, for we can have intentionality only on the part of states which are in principle accessible to consciousness. The cautious view, though, is that we have (at least) two different types of representation (Davies 1995). On the one hand, there are the conceptual contents of ordinary beliefs and desires, to which consciousness may constitutively attach. On the other hand, there are the non-conceptual contents of information-processing states. The present point is about the relation between non-conceptual, information-processing content, and a third element: the content of conscious attention. I began with the remark that conscious attention is what explains our grasp of a particular type of conceptual content, namely, demonstrative content. We are now looking at the relation between conscious attention and information-processing content. If you are asked the question,'What colour is that block?', you must, to understand the question, consciously attend to the block. As we saw, it seems compelling that it is in virtue of your conscious attention to the block that you know which thing is in question. If you are to verify the proposition, That block is blue', it is not enough merely that there be appropriate cell-firing in visual area V4. After all, at any one moment, as you look at a complex scene, there will be many objects of various colours in your visual field. So there will be a lot of cells in V4 firing in response to the different colours in the scene. Your verbal response, 'That block is blue', has to be caused by just the right cell-firings; it has to be caused by exactly the cells which are firing as a consequence of that particular block

Experiential Highlighting

13

having the colour it does. So these particular cells have to be selected and used to affect the verbal response. What causes there to be that connection? I want to propose that what causes there to be that connection, between your verbal response and the cell-firing, is your conscious attention to the object. 3. The Causal Hypothesis Let me state this Causal Hypothesis formally: The Causal Hypothesis: When, on the basis of vision, you answer the question, 'Is that thing F?', what causes the selection of the relevant information to control your verbal response is your conscious attention to the thing referred to. What is the argument for this causal hypothesis? If you were not consciously attending to just that object, then just those cells would not be affecting your verbal response. Had you been consciously attending to a different object, a different set of cells would have been affecting your verbal response. And so long as you were to be looking at just that object, in just that context, it still would have been just those cells that affected your verbal response. These counterfactuals do not, in themselves, constitute the truth of the causal claim. But the counterfactuals are plainly true; otherwise, your verbal responses would not bear any relation to the objects of your conscious attention, and the visual verification of propositions about demonstrated objects would be simply impossible. Moreover, the causal thesis explains why those counterfactuals are true. There are two complementary lines of objection that you might have to this causal thesis. First, you might accept that there is a serious question: 'How does it come about that the firing of cells in V4 is connected to your verbal response?' But, you might say, the cause of there being this connection is not conscious attention to the relevant object. It is, rather, a further information-processing mechanism. To this objection we must immediately concede that there are illuminating information-processing models of the executive control of mental processes, and that the errors made in controlling which mental operations to perform can show that the causation of a verbal response by cell-firing in the visual cortex, under the guidance of conscious attention, is not a simple matter and may well be mediated by a number of small-scale mechanisms concerned with, for example, sequencing the various tasks the agent may be performing in roughly the same period (cf. Norman and Shallice 1986; Shallice 1988; Monsell 1996). But at the highest level of determining the objectives of the subject, there simply

14

Experiential Highlighting

is no alternative to appealing to the beliefs and intentions of the agent, and that includes the demonstrative beliefs and intentions of the agent. If we were blocked from appealing to the agent's intentions, we would simply have no idea where to begin in giving a model of control of the agent's mental operations. But what I have just been arguing is that an appeal to the agent's demonstrative intentions requires us to appeal to the agent's conscious attention to objects; we cannot acknowledge a role for intention, in the control of mental operations, without thereby acknowledging a role for conscious attention. We may have to appeal to the deepest aspects of an agent's personal life in explaining why his conscious attention has just the focus that it does, and we have no way of recasting this causal-explanatory work in information-processing terms. It can, of course, happen that people make verbal responses whose causation by visual processing is not under the control of conscious attention. That is exactly what happens in cases of blindsight, when the patient is reliably accurate in his verbal reports about the contents of the blind field. But the mere fact that such a thing occurs can hardly show that this is what always occurs. Conscious attention to objects can make a difference nonetheless. You might wonder whether it would not be possible for a sufficiently ingenious experimenter to establish that the onset time of selection of the relevant visual information to control verbal report is prior to conscious attention being given to the object. (That is, you might wonder whether the Causal Hypothesis is not vulnerable to a result in the style of Libet 1985, 1999.) But causation here is not a matter of onset time only; as Libet has emphasized, such a difference in onset time would not be enough to establish epiphenomenalism. Even such a point about onset time, were it to be tested, would be consistent with the view that conscious attention to the object is essential in sustaining and maintaining the selection of information for verbal report, in an ordinary case. Another way to put the point is to note that when you say to someone, 'What colour is that block?', it ordinarily matters whether or not they understand the question; in particular, whether they know which thing you are talking about. The question, for someone who proposes that all the causal-explanatory work in controlling their verbal response is achieved at the information-processing level, is whether we can give an account, in purely information-processing terms, of the distinction between knowing which thing is referred to by the demonstrative and not knowing which thing is referred to by the demonstrative. As we saw, common sense draws that distinction by asking whether the subject has consciously singled out the reference of the demonstrative. We have simply no way of getting the effect of that distinction in purely information-processing terms.

Experiential Highlighting

15

This line of objection to the causal thesis I am proposing—the thesis that it is conscious attention to the object that brings it about that your verbal response is controlled by the relevant cell-firing in V4—would fit with the view that conscious attention is epiphenomenal, that it is always causally idle. On this view, our ordinary demonstrative beliefs and intentions simply have no effect on our actions or on the world. But the line of objection would also fit with a less extreme position. On this view, conscious attention is not epiphenomenal, it does have a causal-explanatory role to play, but its causal-explanatory role is confined to the 'space of reasons'. On this view, conscious attention to just that object can be what brings it about that your verbal response to the question, 'What colour is that block?', is controlled by just the right aspect of your conscious experience. On this view, conscious attention to the block can be what brings it about that your saying, That block is blue', was caused by your experiencing just that block, rather than anything else, as blue. So the correctness of an explanation of type B can be underwritten by the focus of your conscious attention. But, you might argue, the correctness of an explanation of type A can never be underwritten by the focus of your conscious attention. Supposing that it can be so underwritten involves, on this view, an illegitimate crossing of explanatory levels, moving from one explanatory level— 'the space of reasons'—to another—the level of information-processing psychology. However, scientific accounts of vision are already heavily committed to level-crossing. For example, the appeal to information-processing to explain verbal reports is already a level-crossing in that sense: it appeals to information-processing to explain the allegedly 'space of reasons' phenomenon of verbal judgement. Moreover, once we accept the legitimacy of B-type explanations at all, it is hard to see how conscious attention could play a role in bringing it about that your verbal report is controlled by a particular aspect of your conscious experience, except by bringing it about that your verbal response is controlled by an appropriate aspect of the underlying information-processing. It is probably a mistake to suppose that there is any competition between type-A explanations and type-B explanations, even though there is this level-crossing. Suppose you consider the relation between a type-B explanation and the remark: I say it's square, because it is square. This remark is plainly consistent with a type-B explanation; we are being given descriptions of different aspects of a single causal process. The only problem with this explanatory remark is that it provides no clue as to just how the squareness of the object is affecting you; that is, through which sensory channel. So we might say:

16

Experiential Highlighting

I say it's square, because it's visibly square. And this also is plainly consistent with a type-B explanation. And suppose we now say, more cautiously: I say it's square, either because it's visibly square, or because it's just as if it's visibly square, This gives one way of interpreting a type-A explanation. On this interpretation, 'a looks F' can be glossed as: 'Either a is visibly F, or it's just as if a is visibly F'. So on this interpretation of type-A explanations they are plainly consistent with type-B explanations, describing different aspects of the same causal process.

4. The Role of Location What I am proposing, then, is that your knowledge of the reference of a demonstrative is provided by your conscious attention to the object referred to; and that your knowledge of reference thus supplied is what causes particular information-processing procedures to be used to verify propositions involving the demonstrative. Without yet going into the details—we will look at the details in Chapter 3—it seems straight off that the situation with respect to action on the basis of a demonstrative proposition will be similar. Conscious attention will be what causes the right information-processing systems to be selected to allow you to act on the basis of a demonstrative proposition. The pattern of use that you make of the demonstrative is explained by your knowledge of its reference. Incidentally, it is not just that conscious attention to the object will cause the right information-processing content to be selected. Conscious attention also defines the objective of the information-processing: it determines which object the information-processing has as its target. I want now to bring out how this approach lets us determine the sense of the demonstrative: that is, the way in which the object is given to you in consciously attending to it. Suppose for the moment that the causal hypothesis is correct, that it is your conscious attention to the object that brings it about that your prepositional judgement causally depends on a particular set of cell-firings, carrying a particular piece of information about the object. The next thing to understand is how there can be this causal-explanatory relation between the conscious identification of the object and the selection of particular information. How exactly does your identification of an object, at the level of your subjective life, bear on the selection of information for further pro-

Experiential Highlighting

17

cessing? That is, what is it about your identification of the object at the level of your subjective life that causes the selection of just the right underlying information to control your verbal reports? The very fact that this question demands an answer defines a sense in which there has to be a commensurability between content at the information-processing level and the content of consciousness. There is commensurability in the sense that there are causal-explanatory relations between the two types of content. One familiar way in which it is generally supposed that there is commensurability between the two types of content is that it is often supposed that a scientific account of vision could eventually explain why we have the kinds of visual experience that we do: for example, that the experience of redness may be causally explained in part by the fact that various cell-firings are registering the presence of redness (cf. Marr 1982). This does not presuppose that it is the very same type of content at both levels, but it does demand a commensurability between them. A parallel commensurability is presupposed between information-processing content and the conceptual content of verbal report, when we assume that what is happening at the information-processing level can explain our making the verbal reports that we do. I am not here concerned with this bottom-up commensurability, in which the content of conscious experience is explained in terms of information-processing content. The issue that concerns me is, as it were, a top-down commensurability: to explain in detail just how conscious attention to an object can identify the thing, so that, at the informationprocessing level, just the right information is selected to control your verbal reports. How in detail does conscious attention to an object serve to single out the right information to control verbal report and action? Suppose we consider a minimalist response to this question. The minimalist would say that, at the conscious level, the object is identified demonstratively, as 'that block', for example, and that at the information-processing level, the object is again identified demonstratively, as 'that block'. You might protest that these have to be quite different types of demonstrative, since the subjective 'that block' has to be grasped by the use of conscious attention, whereas there is no similar role for consciousness at the information-processing level. But we are already acknowledging that there must be commensurability between different types of content, and so we should allow that there could in principle be causal-explanatory relations between different types of demonstrative content. You might also protest, as against the minimalist, that we have as yet no account of how demonstratives are individuated, so we do not yet have any analysis of when we have commensurability between demonstratives at the two levels; but a minimalist may simply reject the demand for an account of individuation and take this

18

Experiential Highlighting

commensurability to be a primitive phenomenon. The most radical problem for minimalism is, rather, that we cannot assume that the relevant information-processing contents are to be specified using singular terms at all. The point here is that visual processing is carried out in specialized processing streams, and there is no such thing as the single stream in which all the information about any single object is carried (Zeki 1993). The way in which the object is identified at the level of visual attention has to be capable of singling out information from any of a range of specialized processing streams, to control verbal report. The evidence from physiology and from cognitive studies is that location is what is used in cross-referencing visual processing streams (Zeki 1993). In each specialized processing stream, determining colour, shape, or movement, for example, the processing generally carries information also about the locations of the colour, shape, or movement; so that if it matters whether a colour and shape discovered in different processing streams relate to the same object, a first approximation to the answer will be provided by asking whether the colour and shape are at the same place. If this is right, then when you consciously attend to the object, you will be able to single out which information streams are relevant to verification and action on the object so long as your experience identifies the object as the thing at a particular location. Of course, there is more to the singling out of the relevant information than merely location. If you are verifying or acting on the basis of propositions about a glass which is currently being clasped in a hand, the parts of the glass may be no closer to each other than they are to the parts of the hand. So location alone would not be enough to select the information relating just to the glass. There must also be some use of something like traditional Gestalt principles of grouping (cf. e.g. Palmer and Rock 1994; Prinzmetal 1995). As we shall see, it has been a traditional insight, shared by Moore 1962 and Evans 1982, and carefully expounded by David Kaplan 1989b from Michael Bennett, that the senses of demonstratives involve the locations of the objects seen. But the problem for the traditional insight has always been that articulating it seems to involve ascribing something like a descriptive sense to the demonstrative. It seems to involve supposing that the demonstrative 'that box' must mean something like 'the box at that place'. And this proposal always runs into the problem that you may in vision be having an illusion about where the thing is, and yet be demonstrating it successfully. The proposal I am making is that the sense of the demonstrative is indeed given by the seen location of the object, but the role of the experienced location of the object is not to provide a descriptive identification of it. It is, rather, to organize the information-processing

Experiential Highlighting

19

procedures that you use to verify, and to act on the basis of, judgements involving the demonstrative. It is entirely possible that the experienced location of the object should succeed in playing that role, so that you manage to have just the right information-processing procedures swinging into play to allow you to verify and to act with respect to the object, even though you are under an illusion about where it is. So we can individuate demonstrative senses by the seen locations of the objects, without supposing that location has to be providing a descriptive identification of the thing. And we have to keep in mind that there will be a role also for something like Gestalt organizing principles in individuating the senses of demonstratives. 5. The Empirical Evidence for the Role of Location The question I have been asking is, how does the selection of an object, at the level of conscious experience, affect the selection of information to control verbal report? There has to be something general-purpose about the selection of an object at the level of conscious attention. By singling out an object in experience, you enable yourself to answer any of a range of questions about its visible characteristics—not just its colour and shape, but its location, movement, orientation, and so on. So how is it that your singling out of an object in experience can have this general-purpose character? We could, in principle, use any of a wide range of strategies. For example, we could in principle use colour as a selection cue. Suppose that you are looking in the window of a shop selling jerseys and sweaters. The big display window is a mass of jerseys, all different shapes, colours, and sizes, from floor to ceiling; both the back of the display and the two side wings seeming to engulf you in jerseys. As you look, you focus now on one, now on another. Perhaps you spend some time comparing two. It is easy to imagine that you might select now one jersey, now another, to dwell on visually. What is not so easy to imagine is that you might simultaneously select all and only the red jerseys, from different parts of the display; that you might simultaneously highlight in experience all these different jerseys at their different places. It is not that there is a contradiction in supposing that someone might be able to do this. There is no a priori reason why we could not operate in that way. If we did, then at the level of conscious experience, we would simply highlight redness, and at the informationprocessing level, information from all the objects designated as red would be simultaneously selected. Certainly from a computational point of view we could construct a machine which, when given information about the

20

Experiential Highlighting

shape, size, colour, and so on of targets at various places, could simultaneously select the information from all the targets labelled as red, for further processing, in parallel, simultaneously, on all of it. So why can't we do that? The obvious answer is that though we could indeed construct a machine which worked in that way, that is not, as matter of engineering fact, the way in which we work. We visually select targets on the basis of location, and in general we can give focused attention to only one location at a time. The evidence that this is in fact how human vision works, selecting information from objects on the basis of location, comes from a variety of experiments. Suppose, for example, that you are shown a display of nine letters, three of them red, three of them green, and three of them brown. You are asked to name one of the red letters, and as many of the others as you can. If it were possible for you to focus simultaneously on all the red letters in the display, then the simplest strategy would be for you to do that: read out all the red letters, then name as many of the green and brown as you could. In fact, though, Tsal and Lavie 1988 found that this was not what subjects did. Rather, having first named one red letter, subjects then typically went on to name other letters spatially adjacent to the red letter. This implies that there are two steps in focusing onto the initial target red letter: (a) find a location at which there is redness; (b) report the letter found at that location. To complete the task, you then (c) go on to report the letters at adjacent locations. . Tsal and Lavie were building on a finding by Snyder 1972, who asked his subjects to name the single red letter in a display of otherwise variously coloured letters. Looking at the mistakes people made in trying to do this when under time pressure, he found that observers tended to report one of the two immediate neighbours of the red letter on about 35 per cent of the trials on which mistakes were made. If errors had been equally likely to involve all positions, this should have happened on only 18 per cent of the trials. Again, it looks as though the strategy being used is to find the location of the redness, then report which letter is at that location. On this analysis, it should happen that you can accurately report which letter is the red one only when you can also accurately report the location of the letter. Nissen 1985 confirmed that this is indeed so. In his review of the literature on this point, Pashler writes: In summary, location is special in this sense: when an observer detects a target defined on other dimensions ranging from simple to relatively complex, this provides information about the location of the stimulus being detected. A simple metaphor conveys the essence of this idea: when someone reads out information from a channel selective for a particular target, this channel is labelled with at least rough location information. (Pashler 1998: 98-9)

Experiential Highlighting

21

The force of the idea that location is a primary selection cue, as Pashler puts it (1998: 98), is illustrated by forced-choice tasks in which just one out of n possible targets is present, and your task is to say which one of the n targets is present on this occasion. In this task, you may also be asked about the location of the target. And if you get the location wrong, the probability that you are right about which target is present is close to 1/n; that is, you are at chance, you are just guessing (Shiffrin and Gardner 1972; Johnston and Pashler 1991). Suppose we accept that location is a primary selection cue. What are the implications of this for the Causal Hypothesis? What does it tell us about the way in which the experiential highlighting of an object can cause the selection of input information for further processing? What it all tends to show is that in experientially singling out an object, identification of the location of the object will enable the information relating to that location to be selected, for the control of verbal report. So if, at the level of experiential highlighting, you can identify the object by its location, this will enable the relevant information to be singled out to control your verbal reports about that object. And this is why location plays a special role in the way in which you identify the object at the level of conscious experience.

2 What is Knowledge of Reference? In this chapter, I want to set in broader philosophical context the observation that knowledge of the reference of a demonstrative is provided by conscious attention to the object. There is a connection between the notion of reference and the notion of truth. When you refer to an object and say something about it, whether or not what you say is true depends on whether what you have said is true of the object to which you have referred. So a term that stands for an object makes a contribution to determining whether a statement containing that term is true or false, by standing for that object. What exactly is the relation between knowledge of the reference of a singular term, such as a demonstrative, and your ability to verify, or find the implications of, propositions involving the demonstrative? Understanding this is fundamental to understanding the role of conscious attention. 1. Reference vs. Use

It often seems possible to formulate a hypothesis without knowing how to test it. A famous example in mathematics is the Goldbach Conjecture, that every even number is the sum of two primes. It certainly seems possible to understand this hypothesis, to know what it is for it to be true, without knowing how you would go about verifying or refuting it. In fact, it seems that to understand a proof of the conjecture you need to have mathematical concepts that go far beyond those you need simply in order to grasp the theorem itself. A similar point applies to hypotheses about the distant past. Take, for example, the view that among the Picts, succession to the kingship went through the line of the mother rather than the father, so that if the king died without children or siblings, his heir would have been the son of his mother's brother or sister, rather than the son of his father's brother or sister. This hypothesis is easy to formulate but hard to confirm or refute, and at the moment is still unresolved.

What is Knowledge of Reference?

23

Or again, a physicist may formulate a hypothesis without being able to confirm or refute it. You might conjecture that neutrinos spin clockwise when viewed from the direction in which they are going, for instance, without at all being able to verify it. Understanding the conjecture, knowing what it is for it to be true, is one thing, and knowing how to verify or refute it is another. There must presumably be some relation between the way in which you understand the hypothesis, and the way in which you set about verifying it. They are not simply unrelated. Your understanding of the hypothesis means that you know what it would be for the hypothesis to be true. And your knowledge of what it would be for the hypothesis to be true has to have some bearing on how you go about verifying or refuting it. A similar point applies to your knowledge of how to find the implications of a proposition—either its deductive implications for the truth of other propositions, or its significance for action. You could know what it is for the proposition to be true without yet having grasped its implications. But your knowledge of what it is for the proposition to be true must have some bearing on what you will take its implications to be. There are, I think, two types of relation we might expect to find here, between your knowledge of what it would be for the hypothesis to be true and your use of a particular way of verifying the hypothesis. First, your understanding of the hypothesis should be what causes your use of a particular method of verification or refutation. I will call this the Causal Link: The Causal Link: Knowledge of what it is for a hypothesis to be true is what causes your use of a particular way of verifying, or finding the implications of, that proposition. It is, though, not just that your use of the method of verification is caused by your understanding of the proposition. It ought also to be possible to view the use of a particular method of verification as being justified by your knowledge of what it would be for the proposition to be true. Your use of the method of verification has an objective, defined and grasped prior to the use of that method of verification. The objective is to find out whether the proposition is true. So it ought to be possible to view the method of verification as a way of meeting that objective. This is what I will call the Justification Thesis: The Justification Thesis: Knowledge of what it is for a hypothesis to be true is what justifies your use of a particular way of verifying (or finding the implications) of that proposition. Putting the Causal Link and the Justification Hypothesis together, we have what I will call the Classical View:

24

What is Knowledge of Reference? The Classical View: Knowledge of what it is for a proposition to be true is what causes, and justifies, your use of particular ways of verifying, and finding the implications of, that proposition.

Before looking at the case of demonstratives, it may be helpful to look at a much simpler example of the Classical View. Perhaps the simplest and most striking illustration of the Classical View is provided by the classical account of prepositional logical constants, terms such as 'and', 'or', and so on; I will look at this case in Chapter 5. But we can also apply the classical picture to our understanding and use of so-called descriptive names. An example of a descriptive name is the name 'Elmo', used as shorthand for 'the oldest tree in this garden', For there to be such a thing as Elmo, there must, of course, be exactly one tree in this garden which is older than all the others in the garden. There will be procedural rules for the use of the name 'Elmo'. Some of these rules—the introduction rules—will specify when a proposition containing the name 'Elmo' has been verified. Other procedural rules—the elimination rules- -will specify the implications of propositions containing the name 'Elmo'. The introduction rules for 'Elmo' will include: Exactly one tree in this garden is oldest

Any tree in this garden which is oldest is F Elmo is F

This tells you how to verify a proposition of the form, 'Elmo is F': establish both that there is such a thing as Elmo, and that anything meeting the condition for being Elmo is F. The elimination rules will include: Elmo is F exactly one tree in this garden is oldest Elmo is F any tree in this garden which is oldest is F Given that Elmo is F, you can move to the conclusion that there is such a thing as Elmo: the first elimination rule above spells out the implication of there being such a thing as Elmo. The second elimination rule spells it out that if Elmo is F, you can conclude that anything meeting the condition for being Elmo is F. These are the procedural rules. On the Classical View, what causes and justifies your use of these ways of verifying and finding the implications of propositions involving the term 'Elmo' will be your knowledge of the reference of the term:

What is Knowledge of Reference?

25

'Elmo' refers to the oldest tree in this garden. By denning what has to be the case for a proposition involving the name 'Elmo' to be true, this statement of reference defines the objective at which the introduction rule aims, and allows you to determine what we should take to be the implications of a statement using the name. In that sense, it justifies the use of those introduction and elimination rules above. We do not need to suppose that reflective reasoning is involved at all in the causation of your pattern of use of the term by your knowledge of the reference of the term. Your pattern of use may simply be systematically causally dependent on your knowledge of reference. If you varied the reference of the term, you would consequently vary which procedural rules you took to be correct. How does this Classical View apply to perceptual demonstratives, terms such as 'that man' or 'that whale', referring to currently perceived objects? As we have seen, there is a strong intuitive link between our understanding of perceptual demonstratives and consciousness of objects. The link is that you have to be conscious of the object in order to understand a statement made using the demonstrative. Suppose you and I are standing side by side on an observation platform high in the sky. I am gripping the railing tightly and staring at my hands. You make a series of remarks about a building visible in the distance. Since I am listening, I can formulate a number of descriptions: 'the building you are looking at', 'the building with the golddomed roof, and so on. But I do not understand your use of the demonstrative, 'that building', until I finally stop focusing on my hands and look. G. E. Moore put the point succinctly when he said: 'the prop, is not understood until the thing in question is seen.' (Moore 1962: 158). You do not understand the demonstrative except by being conscious of the object; or rather, there is a distinctive way of understanding the demonstrative that is provided by your consciousness of the object. So there is a theoretical role to be played by the notion of consciousness of the object. It has to explain the ability to understand the demonstrative. The intuitive link is not just between reference to a perceived object and consciousness of it, but more specifically between reference to the object and conscious attention to it. To refer to the perceived object, it is not enough that I have it in my field of view. As I look over the scene, it is not enough that the gold-domed building be there somewhere in my visual field. I must separate it visually, as figure from ground, I must visually discriminate it from its surroundings. I have to attend to it. Knowledge of the reference of the demonstrative is what causes and justifies the use of particular procedures to verify and find the implications of propositions containing the demonstrative. Conscious attention to the

26

What is Knowledge of Reference?

object, I will argue, is what causes and justifies the use of particular procedures for verifying and finding the implications of propositions containing the demonstrative. Hence, knowledge of the reference of the demonstrative is provided by conscious attention to the object.

2. Two Examples: Action and Verification If knowledge of the reference of a demonstrative is provided by conscious attention to the object, then we have to think of conscious attention to the object as what causes and justifies our use of patterns of verification and ways of finding the implications of demonstrative propositions. Just to illustrate how this view should work, I will give two simple examples. One relates to the output from a demonstrative proposition, looking at the link between knowledge of the demonstrative and action. The second example relates to the input to a demonstrative proposition, looking at the relation between knowledge of the demonstrative and verification of a demonstrative proposition. Briefly, the two examples I am about to describe are: (1) Catching a cricket ball. In this case, my conscious attention to the ball is what causes and justifies my use of particular procedures to catch that particular ball. (2) Finding whether one thing is enclosed by another. In this case, my conscious attention to an 'X' is what causes and justifies the use of a particular information-processing routine to verify the proposition that that particular 'X' is enclosed. In both these cases, computational processing is involved. I myself may not be aware of which computational procedures are being used; but it is my conscious attention to the object that defines the objective of the computation. It is in that sense that conscious attention justifies the use of the computational procedure: the use of the computational procedure can be justified or criticized by reference to the objective defined for the procedure by conscious attention. You might accept that there are information-processing systems in the human being, but suppose that they are insulated from the kinds of psychological phenomena familiar to common sense. You might believe that the dynamics of the two systems, information-processing and ordinary consciousness, are independent. But that would be a mistake. Which information-processing we perform is not somehow isolated from the explicit objectives that we have, the tasks that we want to carry out. Which

What is Knowledge of Reference?

27

information we process, and how, depends in part on what we are up to, what our objectives are. (1)1 turn to our first example, the fielder in a game of cricket trying to catch a falling ball. Suppose that you have to catch a ball. You can do this without being able to say how you do it. But to do it you have to be selecting information from the moving ball, and performing operations upon that information, if you are to catch the thing otherwise than by accident. According to McLeod and Dienes 1996, skilled fielders do this by finding a speed that will keep the acceleration of the tangent of the angle of elevation of gaze to the ball at 0. Even a skilled fielder, though, does not simultaneously perform this computation for every airborne object he sees. This information-processing is outside your conscious life; but it is being performed in the service of the goal you explicitly have, to catch 'that ball'— demonstratively identified. So then there is a question how this is done, how our information-processing capacities get harnessed to our conscious objectives. The key notion that we need, in explaining this connection, is attention. If you are to act intentionally with respect to the moving ball, you have to attend to it perceptually. And that conscious attention is what causes the selection of information from the ball, and selection of which operations to perform upon it. To sum up: if you are to act intentionally on an object, you must consciously attend to it, in the common-sense use of the term; but that act of attention must also cause the selection of suitable information for processing, and suitable processes to operate on it, if the information-processing of which you are capable really is to be harnessed to your objectives. The concept of conscious attention thus plays a role here in connecting our psychology, at the level described by common sense, with the information-processing described by psychologists. (2) My second example of the Classical View in operation relates not to action but to the verification of a demonstrative proposition. There are bottom-up processes involved in perception, which provide us with some basic information, and then there are top-down visual processes, selected according to the task in hand, applied to some portion of that basic information. So when you look at the scene before you, there are, as it were, many different visual questions you can ask. Is this object inside that one? Are the chairs arranged in a circle? Will you be able to cross the room by taking the direct route to the door? To answer these questions, you just have to look. But this kind of visual interrogation of the scene involves the exercise of visual skills, top-down processes that are selected by your decision to ask this question rather than that.

28

What is Knowledge of Reference?

Suppose you are shown a diagram with a figure containing a number of loops and an 'X', and you have to determine whether the 'X' is enclosed by the boundaries of the figure. In some cases this will be hard and in some cases it will be easy, but you are always exercising a visual skill. How are we to think of this visual skill? One proposal—from Ullman 1996—is that we think of the operation as involving a kind of 'colouring' process, so that from a given starting point, such as the 'X', there can be a spread of activation to neighbouring places, which stops when a discontinuity is encountered. If that process terminates, then the 'X' is enclosed. Otherwise, the 'X' is not enclosed. This kind of information-processing is not something of which we are conscious, but it can be elaborated to give an explanation of which cases we find easy and which we do not (Ullman 1984,1996). This processing involves the exercise of conscious attention. First, you have to select which place to begin from, you have to decide that the place of 'that "X"'—demonstratively identified—will be where you start. And secondly, you have to select which operation to apply to that place; you have to select the 'colouring' operation. There is no possibility of doing the whole thing without any selection; it would entirely defeat the purpose to try to apply the 'colouring' operation to all perceived places simultaneously, for example. Ullman suggests that this is a basic top-down attentional skill, one of a handful of basic routines that can be applied to the bottomup contents of perception in various combinations, for various purposes. These purposes are partly defined by the conscious attention of the subject. The structure of the view we have reached is that knowledge of the reference of a demonstrative is provided by conscious attention to the object. And conscious attention to the object is what causes and justifies the use of information-processing routines in acting on the basis of demonstrative thoughts about the thing, and it is what causes and justifies the use of information-processing routines in verifying propositions about that thing.

3. The Double Use of Feature Maps The Classical View demands that we keep knowledge of the reference of a term distinct from the use of ways of verifying that the object has particular properties. They have to be distinct, because the knowledge of reference has to cause and justify the use of methods of verification. It seems, though, particularly hard to keep knowledge of reference separate from method of verification in the case of demonstratives referring to perceived

What is Knowledge of Reference?

29

objects. The reason is that you can hardly have conscious experience of the object without already having, in some sense, verified its possession of various properties. It does not make sense to suppose that you could consciously attend to an object without being aware of it as having certain properties of shape, colour, and so on. You might therefore argue that the kind of high-level processing involved in finding whether one object is enclosed by another, which we discussed in the last section, is quite a special case. To verify that the 'X' is enclosed, after all, you will have to first identify the 'X'. And, you might argue, that will already involve you in having verified its possession of simple sensory properties, such as shape, size, and colour. (By a 'sensory property' I do not mean a 'secondary quality' in the traditional philosophical sense, but one of the properties used as primitive by the visual system.) So, you might argue, the Classical View cannot be sustained for the relation between knowledge of the reference of a demonstrative and verification that the object has particular sensory properties. Knowledge of reference in this case does not cause and justify the use of particular methods of verification. Rather, those verifications had to be performed already in order for there to be knowledge of reference. To deal with this objection there is a distinction we need. This is the distinction between (1) Using an object's possession of a property to single it out visually, and

(2) Verifying a proposition to the effect that the object has that property. Verifying a proposition involves the use of your conceptual skills, whereas visually singling out an object is a more primitive phenomenon. You could use the fact that an object is moving in a particular way to separate it visually from its background, without having formulated any proposition about its movement and indeed without having any concepts of motion at all. Similarly, it is possible to use colour as one of the properties by which you visually define the object—it is one of the properties in virtue of which you can be said to be seeing an object at all—even though you have not verified the proposition that the object has that property, and even though you do not have colour concepts. The obvious example here is an animal which may entirely lack concepts, but be able to use colour vision in separating objects from their backgrounds. Although we have to distinguish these two phenomena, visual singlingout and the verification of propositions, there is a connection between them. But before coming to the connection, let me give one more

30

What is Knowledge of Reference?

illustration of the distinction. For you might wonder whether this distinction can be drawn in the case of people who do have concepts of objects. If you are capable of grasping demonstratives referring to physical objects, can you be using colour vision to discriminate the objects around you, even though you do not have colour concepts and so cannot be said to be verifying propositions about the colours of those objects? Plainly you can. An example of this phenomenon is provided by children, who have the ability to refer to perceived objects, on the basis of colour vision, in place long before they have any grasp of colour concepts. Children have unusual difficulty in learning colour words, even the four basic terms, 'blue', 'green', 'yellow', and 'red'. This is not because of any delay in the maturation of colour vision. Even four-month-olds can discriminate, match, and recognize the four basic colours, so there is no reason to suppose that they lack experience of colour. And the difficulty is not because of a delay in learning which words are colour words. At an early stage, children will reply to the question, 'What colour is it?', by producing a colour word, but they may use the colour words randomly, or perseverate in using the same word for all colours. The situation seems to be that, as Bornstein puts it in his review of the developmental literature, 'early in life, sensory and linguistic colour knowledge seem to coexist, but a proper map connecting names and perceptions is late in developing' (Bornstein 1985: 78). At an early stage, then, children do not yet have colour concepts. But since they have colour vision, they can use the colour of an object as one of the characteristics that differentiates it from its background; they can use colour as an objectdefining characteristic without yet being able to verify propositions about the colour of the object. So there is certainly a distinction between visually singling out an object by use of its sensory properties, and verifying propositions to the effect that it has those sensory properties. But as I said, there is also a connection between the two phenomena: (1) visual singling-out and (2) the verification of demonstrative propositions ascribing sensory properties to perceived objects. To see this, let us put our distinction in the context of a fundamental topic in vision science: the way in which the visual system solves the socalled Binding Problem. As I said earlier, there is much converging evidence that different properties of an object, such as colour, shape, motion, size, or orientation are processed in different processing streams (Zeld 1993). This means that the visual system has the problem of reassembling individual objects, as it were, from the results of these specialized processing streams. A specific colour and shape, for example, have to be put together as the colour and shape of a single object, just when they are the colour and shape of a single object. We do not have perception of an individual object until this Binding Problem has been solved, and various

What is Knowledge of Reference?

31

simple sensory properties have been put together as properties of a single object. What strategy could be used to solve the Binding Problem? If the visual processing streams contain at least implicit information about the locations of the features being found—just where in the environment a particular colour or shape is, for example—then there is one effective strategy available. Features found at the same location could be put together as features of a single object. So if, in one stream, redness is found at a particular location, and, in another stream, squareness is found at that place, then the redness and squareness would be put together, so that you have perception of a red square. Of course, this could not be the whole story about how the Binding Problem is solved, because spatially overlapping objects, such as a hand holding a cup of tea, can be seen as plainly distinct, despite the fact that the parts of one object are as close to the parts of the other object as they are to each other. So there must be more than location at work. But it is nonetheless possible that location is a fundamental parameter in the way vision solves the Binding Problem. We can think of the various specialised processing streams as constructing maps showing the locations of particular features, and binding as being, in the first instance, a matter of features from the same location being put together as features of the same object. Anne Treisman's primordial Feature Integration Theory held that there is in the visual system a 'master map' of locations, which is scanned by a window (variable in size) of attention (see e.g. Treisman 1988). When the window selects a particular location on the master map, the features currently found at that location on all the various specialized feature maps are selected, put together to constitute an 'object token', and compared to stored representations so that the object can be categorized. The basic picture illustrating this approach is Figure 1 (p. 32). Spatial attention in Treisman's sense involves the singling out of a single location on the master map of locations, so that all features at the selected location can be bound together as features of a single thing. There is no very evident reason to think that spatial attention in this sense must be a phenomenon of consciousness. (For a striking example of spatial attention without awareness, see Kentridge, Heywood, and Weiskrantz 1999; for further analysis of the type of attention required for binding, see Briand and Klein 1987.) This kind of spatial attention is a precondition of consciousness of the object. The features must be bound for there to be experience of the thing. But the spatial attention itself may be a relatively low-level phenomenon. The kind of low-level exercise of attention that Treisman's model argues is required for binding, contrasts with the kind of exercise of conscious attention that I am arguing is required for knowledge of the reference of a demonstrative.

32

What is Knowledge of Reference?

Fig. 1. Framework proposed to account for the role of selective attention in feature integration. (From Treisman 1988. Copyright 1988 by Lawrence Erlbaum Associates.)

I have been distinguishing between (1) the use of properties of an object to single it out visually, and (2) the use of conscious attention to verify a proposition about an object. In our present terms, the point is that there is a double use for the feature maps, processing in parallel in early vision. They do have a role in determining which objects are there in the vicinity. It is only by finding which features are present and where, that it is possible for those features to be bound together to form object representations. The formation of an object representation is needed if there is to be experience of the object at all. Consequently, the binding of features established in early processing is needed for there to be any understanding of demonstra-

What is Knowledge of Reference?

33

tives referring to the objects in the vicinity. The second use of the feature maps is in verifying propositions about the observational properties of perceived objects; and this, as we have seen, is a separate use. It is one thing to verify a proposition to the effect that an object has a particular property, and another thing to use the object's possession of that property to single it out visually. Experience of an object is the upshot of low-level attention to a particular location. But once you have experience of the object, you can now attend to it consciously, keep track of it over time, find out more about it, and act intentionally on it. What happens in this second use of feature maps? Here we see the connection between the two phenomena. At this stage, the subject can be assumed to be capable of demonstrative reference to the object. What the subject has to do, to verify the proposition that the object has a certain colour, is to make that second use of a feature map in the right way. The right location has to be identified so that the verification of the judgement can be sensitive to which feature is represented as being at that location. This means that in demonstrative reference, the perceived location of the object is not simply a ladder that we throw away; it is not simply a feature of low-level processing which, although used in binding, is not itself registered in the way in which you experience the object. For the way in which you experience the object has to retain the capacity to single out the correct location, at the level of the feature map, when you attempt to verify the proposition. If your grasp of the demonstrative is to be capable of causing and justifying your use of feature maps to verify propositions about the observational properties of the object, then your grasp of the demonstrative must include information about the location of the thing. Hence, your experience of the object must include information about the location of the thing. This distinction between two uses of feature maps applies to the analysis of the kinds of experimental findings I discussed in Chapter 1, section 5. Consider, for example, Nissen's finding (Nissen 1985). Subjects were instructed to report the shape of, say, the red target. Nissen's finding was that subjects could give an accurate report of the shape of a target identified by its colour only if they could incidentally give correct reports of the location of the target. Her interpretation of the result is that subjects, in order to answer the question, consult the kind of feature map described by Treisman, and use the location of the target to identify the information relating to it in different processing streams. But then this is exactly an example of the kind of second use of feature maps that I am describing. We are not here dealing with the kind of low-level exercise of attention involved in the initial binding of features in ordinary perception, but a conscious attempt by the subject to find the red thing and then verify a

34

What is Knowledge of Reference?

proposition about its shape. (For more material relevant to the distinction between types of attention here, see Briand and Klein 1987 and Briand 1998). To sum up, it seems that we can sustain the Classical View, even when it is applied to the relation between knowledge of the reference of a demonstrative and verification of the most basic observable properties of the object. Even in this case, it can be your knowledge of the reference of the term, supplied by your conscious attention to the object, that causes and justifies the use of particular information-processing procedures—the second use of feature maps—to verify propositions about these observable properties of the thing. So we can continue to think of demonstrative reference on the Classical View. Your knowledge of the reference of the demonstrative is constituted by your conscious attention to the object itself. And this conscious attention to the object is what causes, and justifies, the use of particular information-processing procedures, reaching down to the feature maps, to verify demonstrative propositions about that object. (In this section, we have in effect recapitulated the argument of Chapter 1 from a slightly different perspective.)

4. Formally Describing the Use of a Demonstrative At this point we can be somewhat more explicit about the parallel between the Classical View as it applies to demonstratives and the Classical View as it applies to logical constants or descriptive names. We can try to formulate introduction and elimination rules for perceptual demonstratives to characterize their use, just as we use introduction and elimination rules to characterize the use of descriptive names. The empirical argument I have been giving is this. Suppose you are asked to verify a proposition about whether a demonstrated object in your visual field has a certain characteristic. Suppose, for example, that you are asked to verify whether that red object is round. The empirical argument is that in fact this involves you in first, determining the location of that red thing, then, second, determining whether there is roundness at that location. This suggests that the introduction rule for a perceptual demonstrative would typically be something like this: FEATURE MAP: Roundnesss at position p That figure is round However, the situation is actually more complicated than this. For the evidence suggests that visual attention can be allocated to objects rather than

What is Knowledge of Reference?

35

to locations. This is not inconsistent with the evidence that location is fundamental as a selection criterion, because objects are individuated partly by their locations. But the point is that location is not all that binds together an object. The kind of evidence that shows that visual attention can be allocated to objects rather than to locations is this. Suppose you have two figures, spatially overlapping and each continuously changing shape. People can track one of the figures, apparently to the exclusion of the other overlapping figure (Neisser and Becklen 1975). Or again, people are better—faster and more accurate—at judging two attributes of a single object than at judging two attributes of different objects (Duncan 1984). Again, suppose subjects are set the task of responding to a central object, while ignoring an object beside it. Interference from the irrelevant object can be increased when it and the target are grouped by Gestalt principles— for example, by moving in the same direction, or by having the same colour (Driver and Baylis 1989; Baylis and Driver 1992). So we ought to think in terms not of simple locations, but of 'visual objects', which will be individuated by some combination of (a) specific Gestalt principles, such as boundedness, sameness of colour, or common motion, and (b) location. So the kind of introduction rule we want is this: FEATURE MAP: Roundnesss at visual object X That figure is round Isn't 'visual object X'just a notational variant of 'that figure'? No, because at the input stage here we are still at a level of content more primitive than demonstrative content: individuation by Gestalt principles plus location is not yet conceptual demonstration of an object. We are talking only about a way of cross-referencing information in separate processing streams which does in fact relate to the same object, not about a level of representation at which we already have reference to objects (Prinzmetal 1995). What would an elimination rule for a perceptual demonstrative look like? As a first shot we could conjecture that the outputs of a perceptualdemonstrative judgement can be characterized in some such way as this: That cup contains water MOTOR SYSTEM: To reach water, move towards whatever is at position p, hand prehended thus and so On the classical approach I am recommending, the question of knowledge of reference comes up at the level at which we consider the justification that the subject has for the use of those procedures. This is where we have to appeal to the idea of conscious attention to the object. Which object you are consciously attending to is what causes you to use

36

What is Knowledge of Reference?

one information-processing procedure rather than another to verify the judgement. And it is the character of your conscious attention that causes you to use one information-processing procedure rather than another to find the implications of the judgement for action. Moreover, your conscious attention to the object does not merely cause your use of these information-processing procedures. It is also what you would have to appeal to in justifying the use of these information-processing procedures. It defines the point of these procedures, what the goal is of the computation. Suppose we go back to the descriptive name, 'Elmo', for which the introduction rule is: Exactly one tree in this garden is oldest

Any tree in this garden which is oldest is F Elmo is F

and the elimination rules are: Elmo is F exactly one tree in this garden is oldest Elmo is F any tree in this garden which is oldest is F And what causes and justifies your use of these ways of verifying and finding the implications of propositions involving the term 'Elmo' is your knowledge of the reference of the term: 'Elmo' refers to the oldest tree in this garden. What makes the relation between the reference axiom and the use of the term here seem so trivial is the easy commensurability between the reference axiom and the introduction and elimination rules. From your understanding of the phrase, 'the oldest tree in the garden', it is immediately evident why data to the effect that there is just one oldest tree in the garden, and that any oldest tree in the garden is F, for example, should be relevant to the verification of 'Elmo is F'. When we compare the case of the demonstrative, though, it is evident that we do not have the same immediate easy commensurability. Suppose we stick with the simplest introduction rule and the elimination rule already given: FEATURE MAP: Roundness at position p That cup is round

What is Knowledge of Reference?

37

That cup contains water MOTOR SYSTEM: To reach water, move towards whatever is at position p, hand prehended thus and so But conscious attention to the object has to validate the use of those introduction and elimination rules. In dealing with these specific proposed rules in particular, conscious attention to the object has to be of such a sort as to explain what the significance is of the location p. There has to be some commensurability between the way in which an object is identified at the level of feature maps and by the motor system, and the way in which the object is identified at the level of conscious attention. So what is the point of contact? I want to propose that it is found in what I will call the binding parameter. By the 'binding parameter', I mean the characteristic of the object that the visual system treats as distinctive of that object, and uses in binding together features as features of that thing. In Treisman's model, the binding parameter is a specific current location. A feature being perceived as currently at that location is what the system uses in binding it together with other features. Of course, in a more fully realistic model, the binding parameter will be more complex than simply location. It will include the aspects of Gestalt organization that are used in visually identifying that object. But location will evidently still be a component, just because we can see qualitatively identical objects simultaneously at different visual locations. For the moment, suppose we continue with the simplification that current location is the only binding parameter. Suppose now that conscious attention to the object—the highlighting of the object in visual experience—identifies the object as the object at a particular seen location. The identification of the target object as the thing at that location will then be enough for the visual system to know how to search the feature maps to find the shape, colour, or orientation of the thing. The binding parameter will then provide an address for the object, as it were—a way of identifying which object is in question, that can be used by the visual system when it has to verify or find how to act on the basis of propositions about that object. This way of putting it implies that at the level of conscious attention to the object, the very same frame of reference—the very same way of identifying locations—is being used as it is used by the initial feature maps in binding. But the commensurability between conscious attention and brain processing does not have to be achieved in just this simple way. Strictly speaking, all that is required is that there be a commensurability between the way in which locations are identified at the level of conscious attention and the way in which locations are identified at the information-processing

38

What is Knowledge of Reference?

level. All that is required is that conscious attention to the object should identify which thing is the target, in such a way that the informationprocessing sub-systems can, in one way or another, lock onto that thing in order to keep track of it over time, act effectively on the object, or verify propositions about it.

5. The Crick—Koch Hypothesis and an Analogy with Proper Names In thinking about the transactions between conscious attention to an object and the information-processing involved in verifying or acting on the basis of propositions about the object, it is sometimes helpful to consider an analogy. The analogy I have in mind is with the way in which gossip, or, more generally, information about individuals circulates in a community. All this information has to be sorted into bundles, so that each bundle contains information relating to just one individual; otherwise we would never know whether we were talking about the same or different people. The basic method we have for doing this is the use of proper names: we try to sort so that sameness of name is an indication that it is the same individual that is in question. That is not quite enough, though, since different people may have the same name. So we may add a tag. For example, if there are many different people named Williams, we might add tags like 'the milkman' or 'the mathematician'. Then the use of the name plus a tag gives us what we might call a bundling principle: a way of sorting together information as all relating to a single individual. Bundling principles will, then, play the same kind of role in organizing gossip as binding parameters play in the organization of visual information. It is worth remarking that bundling principles can do the work they do whether or not they are actually true of the object. It is perfectly coherent to suppose that Williams the mathematician might turn out not to be a mathematician at all, but a criminal who pretends to be engaged in mathematics. The tag, 'the mathematician', could still perfectly well play its role as a bundling principle. Similarly, Treisman could be right that location is a binding parameter, and you could be using the apparent location of an object perfectly well in sorting together all the visual information that relates to that object, whether or not the object is actually where it seems to be, rather than being reflected in a mirror or seen through a prism. So bundling principles or binding parameters can discharge their work without providing descriptive conditions which an object must meet in order to be the object to which you are referring. I think we can pursue this analogy in thinking about the point that there has to be a commensurability between the level of conscious attention and

What is Knowledge of Reference?

39

the level of visual information-processing. Suppose there are several different people called Reagan, and your informant Sally collects and provides you with information about each of them. So you stand to Sally somewhat as the level of conscious attention stands to the level of visual information-processing. Or rather—since the point of the analogy is not to encourage a homuncular picture of either conscious attention or visual information-processing, but to make it easier to think about the relations between different levels of content—the level of conscious attention stands to the level of information-processing somewhat as the content of your speech stands to the level of Sally's speech, in the analogy. Since there are several different people called Reagan, and Sally is collecting and providing you with information about each of them, Sally will be using tags like, 'the actor who was President of the U.S.A. in the early eighties' as bundling principles. Suppose that Sally gives you this information about all these various people, using the name 'Reagan'. You know that they all have the same name, so you are not in a position to use the name itself alone as a bundling principle. Now suppose that you want to have the following general capacities: (a) You want to be able to interrogate Sally for further information about any of the people about whom she is giving you information, and (b) You want to be able to instruct Sally to act on any one of the people about whom she is giving you information. To have these general capacities, you have to be able to identify the person in whom you are interested, for Sally's benefit. That is, your identification of the person about whom you wish to interrogate Sally has to be one that she can use to find the further information you have requested. Not just any way of uniquely identifying the thing you have in mind will do. Likewise, when you instruct Sally to act on a particular individual about whom she has given you information, you have to identify the individual in such a way that Sally can go about acting on that person; not just any way of uniquely identifying the thing will do. When you are interrogating Sally for further information about one of the Reagans, what you have to provide her with is a way of finding which bundle of information is relevant to your question. The simplest way would be to use the same tags that she uses. So you might say something like: 'How old is the Reagan who is an actor who was President in the early eighties?' But although that is the simplest procedure, it is not the only one that would work. You could identify the Reagan you mean as 'the one who was Governor of California', and that will work, even if it is not itself a tag

40

What is Knowledge of Reference?

that Sally uses as a tag, so long as she can work out that the Reagan who was Governor of California is the Reagan who was an actor that was President in the early eighties. So long as your identifier can be translated into a tag that Sally uses—and so long as Sally can make the translation— you will be in a position to interrogate her for further information about this person. A similar point applies, of course, to giving Sally instructions to act on one of the Reagans about whom she has given you information. She needs to have a way of identifying the target that she can use. So your way of saying which Reagan is in question has to be commensurable with the ways of identifying people that she uses in setting the parameters for her actions—in finding the address to which to write, for example. I am spelling out this analogy quite fully, partly because it may be helpful in reviewing the points that I have made so far. But I am also setting out the analogy because, as we shall see in a moment, it can be pursued to give a way of thinking about the bearing of the approach I am recommending on the Crick-Koch hypothesis (Crick and Koch 1990, 1992; see also Crick 1994). The Crick-Koch hypothesis is that the neural correlate of consciousness is to be found in the physiological mechanisms of binding. The Binding Problem was originally a physiological problem. Once it was observed that after the primary visual cortex, the visual system split into separate processing streams, it was apparent that neurons responding to different features of the same object may be anatomically separated. How, then, is it registered that these neurons are responding to features of the same object? The basic finding on which Crick and Koch build is that two neurons fire in synchrony when a single object, causes them to fire (Eckhorn et al. 1988; Gray et. al. 1989). This suggests that what registers the fact that a number of neurons are currently responding to features of one and the same object is that those neurons are currently firing in synchrony. There are many ways in which this line of thought can be developed. Crick and Koch argue that neurons in different regions are bound together by a mechanism that synchronizes the spikes of their firings in 40Hz oscillations; it is this synchronized firing that signals that they concern one and the same object. The Crick-Koch hypothesis is that this synchronized firing is the neural correlate of consciousness. This proposal is consistent with Treisman's cognitive model. The suggestion would be that the synchronized firing of cells in different processing streams is signalling the fact that those cells are registering the presence of various features all at the same location. Crick and Koch in fact suggest that the reason why binding is, as on Treisman's model, a phenomenon of attention, rather than being carried out automatically and in parallel for all detected features simultaneously, may be an engineering one: that the syn-

What is Knowledge of Reference?

41

chronized firing they describe may be a limited resource that cannot be applied to all perceived locations simultaneously. There is, indeed, a distinction between the process by which binding is carried out and the synchronized firing which signals that binding has been achieved. If Treisman's picture is correct, there must be some physiological realization of the process of singling out a location on the master map of locations, and finding which features, in the various processing streams, are at that location. At this point you might wonder whether the synchronized firing, if it exists, is not merely an epiphenomenon. Isn't the physiological basis of binding to be found in the physiological realization of the process of identifying a location on the master map and finding out which features are at that location? Whether the synchronized firing is merely an epiphenomenon depends on whether there are further physiological processes which respond to it. If synchronized firing is indeed epiphenomenal, then it is not obvious how it could be the neural basis of experience of objects; in that case, the physiological basis of the process of selecting all the features from a singled-out location would be a better candidate for being the neural correlate of experience of objects. But I set this point aside, to stay with the simple hypothesis put forward by Crick and Koch: that synchronized firing is the neural correlate of consciousness of objects. The immediate objection to this hypothesis is that the solution to the Binding Problem is a precondition of experience of the object, rather than being somehow the same as it. So the correlated firing that implements a solution to the Binding Problem may be necessary for consciousness of the object, without being itself the neural correlate of consciousness. The point I have been making, though, is that conscious attention to an object has to be able to identify that object for the benefit of informationprocessing systems. And the natural way for conscious attention to identify the object, for the benefit of the information-processing systems, is to use the parameter that was used in solving the Binding Problem for that object. In the simple Treisman model, the parameter is location. No doubt the simple Treisman model has to be made more complex, to allow for such basic possibilities as movement by the object, or some role for Gestalt grouping principles in binding together the properties of a single object. But whatever complex parameter we use in solving the Binding Problem, that will provide a kind of address for the object that is bound. And the way for conscious attention to identify the object, for the benefit of the information-processing systems, will be to use that complex parameter to identify the thing. Suppose we go back to the analogy, to the need for commensurability between the way in which you identify people and the way in which Sally identifies people. How are you to guarantee that you are using a way of

42

What is Knowledge of Reference?

identifying the object that Sally can use? One way would be just to operate in a hit-and-miss manner: use whatever identifiers occur to you. Some of them will be usable by Sally, and some will not. It's hard to believe that would be the right model for conscious attention. But how are you to have any guarantee that you are using the right identifiers, if you have not made a direct study of your informant—if you have not directly asked Sally which identifiers to use, or made an experimental study of your own visual system? One answer is that some hidden hand, such as evolution, has brought it about that you will be using suitable identifiers. Another, cruder way would be to use the very same physical tokens to issue requests and instructions to Sally as she uses in giving you information. So if she uses cards to give you information about a particular Reagan, just re-use those very same cards in giving her further questions or instructions. Using the same physical base twice over could give a guarantee of commensurability in the identifiers used. It would be quite a strong guarantee of commensurability. It should, I think, be testable which procedure is being used. If you are re-using the same identifiers, there is a kind of breakdown which should be impossible, one in which you simply use all the wrong codes in asking questions or issuing instructions and as a result get nowhere. If, however, there is merely a hidden hand guaranteeing that you are using suitable identifiers, it ought to be possible for the hidden hand to fail on some occasions, so that you use all the wrong identifiers and consequently get nowhere with your questions and instructions. The analogue for vision would be someone who consciously attends to the objects around him, and plans to act on those objects or to interrogate the scene for further information about them, but uses all the wrong identifiers at the level of conscious attention, so the information-processing systems cannot identify the target. It is hard to believe that this is possible, though there is always room for surprise about which pathological cases are possible; intuitively, the content of conscious attention to an object seems to be more tightly linked than that to the ways of identifying objects used in the underlying informationprocessing, so that they could not come apart. But the obvious way to secure that is to use the same physical vehicles to identify objects at the level of conscious attention as are used at the level of the underlying information-processing. Suppose you have a complex experienced visual scene before you. You consciously attend to one object, now to another. What does it come to, that you are engaging in this experiential highlighting of one rather than any other object in the scene? In effect, what I have been arguing is that it comes to this: when a particular seen object is given this experiential highlighting, when you are consciously attending to it, that affects the func-

What is Knowledge of Reference?

43

tional role of your experience of that object. Your experience of that object is now capable of controlling your information-processing, so that (a) you can use your information-processing capacities to verify propositions about that object, and (b) you can use your information-processing capacities to act on that object. For this to work, there must be some intrinsic aspects of your visual experience of the object, which can identify, for the benefit of subsequent information-processing, either to verify or to initiate action, which object is in question. But how exactly does your awareness of the object identify the target? What are these intrinsic aspects of the conscious experience, which identify for the information-processing system which object is in question? I am suggesting that a central aspect here is the experienced location of the object; or, more generally, the complex parameter used in binding. And in effect, what I have proposed is that we can find some support for the Crick-Koch hypothesis by asking how it is guaranteed that conscious attention is identifying the target in that way, using the complex parameter using in binding. The natural proposal is that conscious attention has as part of its neural correlate the very correlated firing that implements that solution to the Binding Problem. Of course, none of this implies that the only neural correlate of consciousness of an object would be this correlated firing. There is more to consciousness than that. You could have binding, and correlated firing implementing the binding, in the absence of consciousness. The thesis at its strongest is only that when you do have consciousness of an object—which will doubtless require the coordination of a mass of neural activity across many regions—the neural correlate of the fact that it is just that object which has been highlighted in conscious attention, is the correlated firing implementing the solution to the Binding Problem. This discussion of the neural correlate of conscious attention began from the need for there to be commensurability between the informationprocessing input that we use in verifying propositions about seen objects, and the demonstrative propositions themselves. I have been suggesting that one elegant way in which the effect could be achieved here is by the same neural circuitry serving as the vehicle of binding and as the vehicle of consciousness. But suppose for the moment we go back to the simplified introduction rule I formulated for demonstratives: FEATURE MAP: Roundness at position p That cup is round One simplification here is that location is being used as the sole binding parameter. Another point to remark is that a legitimate application of this

44

What is Knowledge of Reference?

rule does not require that the cup should actually be at position p, or even that the subject should believe that the cup is at position p. All that is required is that the cup should visually seem to be at position p. This is easy to see in practice. Suppose you are looking at the cup through a prism, or in a mirror. Then there is an ordinary sense in which the cup is not where it visually seems to be. But your visual system can still be using location as the binding parameter, and putting together all the information relating to that particular location in successful binding, even though the cup is not where it seems to be. And your judgements about the shape or colour of the cup, for example, can be perfectly sound even though location is being used as the binding parameter and even though the cup is riot where it seems to be. Your judgements about the shape or colour of the cup can be sound even though you do not believe that the cup is where it visually seems to be, perhaps because you suspect the presence of the prism or the mirror. There is a parallel with proper names: if the bundling principle for one collection of information circulating about an individual is 'Reagan, the actor who became President', you can use the fact that information is so tagged in forming beliefs about Reagan, even though you do not believe that he was in fact an actor; even though, say, you believe that the story of a Hollywood past is merely a romantic myth about the President. Does a similar point apply to action? I said that as a first shot, we could state the elimination rule for a perceptual demonstrative as follows: That cup contains water MOTOR SYSTEM: To reach water, move towards whatever is at position p, hand prehended thus and so But if the remarks I have just made about the introduction rule were correct, we should expect that, a parallel point would apply to the elimination rule. That is, we have to separate the role of location (a) as a binding parameter for action, allowing the putting together of all the information relevant to the action, and (b) as itself one of the parameters to be set in planning an action on an object. The point 1 have just been making is that success as a binding parameter—successful discharge of role (a)—does not require that the visual location of the object be correct, or even that the subject believe it to be correct. Presumably a similar point applies to action. But successful discharge of role (b)—setting the parameter for action—would require that the experienced location of the object be correct. At any rate, that would be so if we thought that the experienced location of the object is what we use in setting the parameters for action on the object. But as we shall see in the next chapter, that is not so. We do not, in general, use the experienced location of an object to determine how far

What is Knowledge of Reference?

45

away it is, in what direction, and so on. It remains true that location is a binding parameter for action, and that the experienced location of the object is what we use in identifying which thing is in question so that the motor system can swing into play appropriately when we act on the object. But let us, in the next chapter, look at these issues in more detail. We will reach a canonical statement of the introduction and elimination rules for demonstratives—a characterization of their use—in Chapter 5. What I have been proposing is that the notion of conscious attention to an object has an explanatory role to play: it has to explain how it is that we have knowledge of the reference of a demonstrative. This means that conscious attention to an object must be thought of as more primitive than thought about the object. It is a state more primitive than thought about an object, to which we can appeal in explaining how it is that we can think about the thing. As I said in the Introduction, this seems to be how Russell thought of acquaintance: acquaintance with an object is a state more primitive than prepositional thought about the object, which nonetheless explains how prepositional thought about the object is possible (Russell 1917). And by providing you with knowledge of the reference of the demonstrative, this conscious attention to the object will cause and justify the use of those introduction and elimination rules.

3 Space and Action So far I have talked about the experienced location of an object about which you are thinking demonstratively. I have said that the experienced location of the object matters because it is part of the way in which the object is identified, at the level of the subjective life. Location provides a way the object can be identified for the benefit of the neural informationprocessing systems involved in acting on the basis of propositions about that object, or verifying propositions about that thing. I have not, though, said anything about how locations themselves are identified.

1. Experience of Places How does experience of a place enable one to identify it? You might propose that place-identifiers such as 'there' identify places by their spatial relations to perceived objects. But a simple way of seeing that terms like 'there' do not work like this is to consider the possibility of illusions about the locations of perceived objects. If I am watching an object but subject to some illusion about its location—say, the desert heat is refracting the light so I am observing a scene which is actually taking place several miles away—then I may form the judgement, That palm tree is there', designating the place at which the tree seems to be. If 'there' referred merely to the place at which the palm tree is, there would be no possibility of error in that judgement of location: the term would automatically refer to the spot, several miles away, where the palm tree is. But we do have to keep open the possibility of error in the judgement: the judgement may be precisely an expression of my having been taken in by an illusion of location. There are other counterexamples to the idea that vision identifies places by their spatial relations to seen objects. In Wittgenstein's Philosophical Remarks (1975), he says that in visual space there is absolute position and absolute motion. If you watch two stars orbit one another in a pitch-dark night, you can see their movement even though their relative positions remain the same. In fact, a visual space in which there was only relative motion is not even imaginable. Suppose you were looking at a clock face

Space and Action

47

with twelve simple bars on it, no numerals. If your visual space had only relative motion, then if you were watching a single clock hand move between two bars on a clock face, you would be able to see the movement as it left the first bar, but the moment it reached the second bar everything would be exactly as it had been initially. We can't, says Wittgenstein, visualize that. In explaining how vision can identify places, if not by their relations to seen objects, it is natural to appeal to the idea of a frame of reference used in vision. The immediate problem here is that demonstratives such as 'there', which refer to places on the basis of your current experience of them, seem to be logically simple. These demonstratives do not involve us in giving some complex identification of a place by specifying coordinates along three axes. When he asserts that visual space is not merely relative, Wittgenstein tries putting the point by saying that it is as if we could always see, as well as the usual things, a set of axes which could be used to identify the locations of the things seen. But, as he says, even that is not correct, because if you could see a set of axes, those axes would already have an orientation in the visual field—as though they were located in relation to an unseen coordinate system used to give locations in visual space (Wittgenstein 1975: 254-5). The problem is to catch the sense in which this unseen coordinate frame can be used in identifying places. The place you are referring to, when you say 'there' and point, is the place to which you are consciously attending. So the question now is how attention is directed to one place rather than another. When, in experience, you highlight one place rather than another, how is it determined which place is in question? What happens is that the attentional spotlight ranges over the places identified by the frame of reference, and the reference of 'there' depends on where the attentional spotlight is focused. If we leave it there, though, we leave it seeming that the unseen frame of reference may be altogether remote from consciousness, perhaps a subpersonal control mechanism for conscious attention which does not itself affect the nature of the conscious life. How does the fact that we are using a frame of reference in our ordinary identifications of places show up in the subjective life, if not by the axes flaring up in the visual field? One way is through the fact that we can give expression to illusions of location, as in the example of the palm tree. Another is through the fact that the space of the visual field is absolute, in Wittgenstein's sense, in that we can perceive motion even of objects whose relative positions are unchanged. We can also consider propositions about the spatial relations between perceptually demonstrated places. Suppose you attend now to one place, now to another, forming the judgement, 'This place is to the left of that

48

Space and Action

place'. What is the epistemological status of this proposition? Is it merely a report of an empirical fact? It does not seem to be; if you understand the proposition at all then you are already in a position to know that it is true. Such a priori propositions, about the spatial relations between perceptually demonstrated places, articulate the frame of reference that we are using in our experiential identifications of places.

2. The Grounding Thesis Let's now return to the main topic, the role of conscious attention in our knowledge of the references of demonstratives referring to the objects we experience. These demonstratives can be exploited in intentional action. Suppose I see a coin on the ground in front of me. I might think,'I'll pick up that penny'. And I act on the thought. How exactly does my understanding of the demonstrative, 'that penny', contribute to my action? In particular, how does my experience of the location of the penny contribute to my action? One extreme position is epiphenomenalism. You might suggest that experience of the location of the penny is actually an epiphenomenon, so far as action goes. It plays no role in the explanation of your action. Of course, your action is spatially directed. But the spatial direction may be achieved entirely by sub-personal visuomotor systems. And, the epiphenomenalist says, there is no other role for your spatial experience to play here. To act on the thought, 'I'll pick up that penny', you do not need to formulate some proposition to the effect, 'That penny is there''. You do not need to refer demonstratively to the place where it is. Still, it might be argued that spatial experience could nonetheless be playing a role in the explanation of the spatial organization of your movements as you reach for the penny. When you act on an object, some aspects of your action are intentional and some are not. For example, when you pick up a glass, you intentionally picked up just that glass. But the exact configuration of your hand, as it moved towards the glass, would not ordinarily be intentional; it might not even occur to you that you were prehending your hand in any particular way at all. Nonetheless, the prehending of your hand might have been a causal consequence of your experience of the glass being of just the type it was. Similarly, even though you do not formulate propositions about the place of the object, still your experience of the location of the object may play a causal role in your action on the object. It may be that you are intentionally reaching for the glass just ahead and to the right. And even if you could not be described as moving in that way intentionally, it

Space and Action

49

could still be that your experience of the location of the glass was what was causing you to move in just that way. As we shall see in the next section, it does not seem likely that this characterization of the role of experience of location in the control of action can be sustained. But it is a very natural picture. It could be set out as a way of explaining two points, which any view does have to acknowledge: (1) Demonstrative identification of objects has a special role to play in action-explanation, and

(2) Demonstrative identification of an object may depend on experience of its location. On (1), suppose there are two boxes in front of me, and my task is to choose one of them by picking it up. The right box holds enough money to remove all my financial anxieties. The wrong box holds copies of my outstanding bills. This alone means that I have already a number of ways of identifying the right box. I can think of it as 'the box containing the money', or 'the box that does not contain the bills', or simply as 'the right box'. None of these ways of identifying the box is, though, sufficient for me to be able to reach out and pick it up intentionally. What would help, and ordinarily all that would help, is a demonstrative identification of the right box. So ordinarily, when I know 'The right box is that (seen) box', then I will be in a position to say, 'So I'll pick up that box', and now I can act intentionally on the thing. For many types of object-directed action, such as reaching towards or picking up the thing, having a demonstrative identification of the object will be enough for you to be able to perform that type of action on the object. And ordinarily there will be no other way of identifying the thing that would be sufficient to allow you to perform that type of action on the thing. It is in this sense that demonstratives seem to have a special role to play in action-explanation. Thesis (2) is a traditional point from the literature on demonstratives, though its explanation is disputed. In section 5 of this chapter, we will look at G. E. Moore's discussion of the point. For the moment, here is one natural thesis, which would describe just how demonstrative reference to an object, with its experienced location, contributes to your ability to act with respect to that object: Grounding: The meaning of a perceptual demonstrative is grounded in those aspects of perceptual experience that set the parameters for my action (how far I move, in what direction, and so on). A term like 'that penny' does not have the same meaning whenever it is

50

Space and Action

used. On any one occasion on which I use it, its meaning has to be filled out by the way in which I am perceiving the object. My current experience of the object gives me a kind of picture of the object—its shape and size, how far away it is, and so on. This picture of the object is the meaning that the term has on that occasion. And the way in which I am aware of the object— the picture of it that I consciously have—determines the coordinates of my action, it sets the details of the way in which I reach for the thing. The idea is that the perceptual information in which the meaning of a perceptual demonstrative is grounded gives a kind of descriptive meaning of the demonstrative, and this perceptual information controls the parameters of my action—how far I move, in what direction, how my hand is shaped, and so on. What explains the fact that perceptual demonstratives have this special link to action is that the meaning of the demonstrative is given by the pictorial content of the experience, which is what controls my movements. You might be encouraged in your adherence to Grounding by an uncritical reading of Alan Airport's conception of visual attention as 'selectionfor-action'. He describes it as follows: Any goal-directed action requires the specification of a unique set of (timevarying) parameters for its execution- -parameters that determine the outcome as this particular action rather than any other: as this particular vocative or manual gesture, this particular directional saccade, and so forth. Consider now what is required if these parameters are to be controlled by sensory (say, visual) information. Suppose that visual information has to guide manual reaching, for example, to grasp a stationary object or to catch a moving one. Clearly, many different objects may be present in the visual field, yet information specific to just one of these objects must uniquely determine the spatiotemporal coordinates of the end-point of the reach, the opening and closing of the hand, and so on. Information about the position, size and the like of the other objects in view, and also available, must not be allowed to interfere with (that is, produce crosstalk affecting) these parameters—though they may need to influence the trajectory of the reach in other ways. Consequently, some selective process is necessary to map just these aspects of the visual array, specific to the target object, selectively onto the control parameters of the action. (Allport 1989: 648)

To think that this amounts to an endorsement of Grounding, though, you would have to suppose that the visual information that is being used in setting the parameters for action must be part of the content of your experience of the object; and we shall see that this need not be so. To see already why there might be an alternative to Grounding, consider for a moment the example of acting intentionally on someone by ringing up that person. The analogue of 'setting the parameters of your action' here is knowing which number to ring. So you might argue that knowledge of the telephone

Space and Action

51

number is needed to execute this type of intentional action on a person. But suppose you have a telephone which stores Sam's number. In that case, all you need to do is press the button marked 'Sam' and the machine will set the parameters for your action. You do not need to know the parameters for your action yourself. Even if you had a mistaken belief about what Sam's number was, this procedure would still let you successfully ring Sam up. If we think this a better model for the role of demonstratives in actionexplanation, the conscious experience itself need not be what sets the parameters for your action, and experience of the object will have some quite different role in explaining the action. 3. Perception vs. Action Let me sketch the case against the idea that it is experience of the location of the object that causally controls the spatial organization of your action on the thing. The epiphenomenalist will, of course, welcome this case. But we will see that you can reject epiphenomenalism while also rejecting the Grounding Thesis. It is not that experience of the location of the object has no causal role to play; it is just that it does not play the causal role envisaged by the Grounding Thesis. First, there are a number of perceptual illusions that do not affect motor responses. The spatial content of your illusory experience indicates that the action should be organized in one way; but the visuomotor system is not taken in, and the spatial aspect of the action is directed correctly. One example is the so-called Roeloff effect, where a rectangular frame contains a stationary visual target. As the frame drifts to one side, the illusion experienced is that the frame is staying still while the target is moving in the other direction. So if the spatial content of the experience were controlling the action, an attempt to point to the target would follow the apparent movement. But in fact, when asked to point rapidly to the target, subjects are accurate in their pointing. The obvious analysis of the situation is that some relatively low-level visuomotor process, remote from consciousness, is setting the parameters for your action in pointing, rather than the spatial experience itself. The same point can be made by considering experiments in which a subject is asked to point towards a target which is moved during a saccadic eyemovement (that is, the rapid jump in eye-position that ordinarily occurs several times a minute) of the subject. The subject adjusts his pointing to keep track of the object, though there is no conscious awareness of any change in the position of the target (Bridgeman, Hendryn, and Stark 1975). Similarly, subjects asked to reach for a target, which was then moved

52

Space and Action

during a saccade, adjusted their reach to compensate for this movement, but were unaware of the change in position of the target (Goodale, Pelisson, and Prablanc 1986; Jeannerod 1997: 83). Again, it seems that the spatial content of the experience is not what is controlling the spatial organization of the movements of the subject. That is being accomplished by some relatively low-level visuomotor system. A related point is made by the so-called Ebbinghaus illusion, in which you present the subject with two target circles of the same size, one surrounded by a ring of circles smaller than it, the other surrounded by a ring of circles larger than it. The circle surrounded by a ring of smaller circles typically looks larger than the circle surrounded by a ring of larger circles. The illusion is preserved when the circles are not simply drawn on a page, but realized by thin solid discs such as poker chips. And the illusion can be manipulated so that two target circles of different sizes are made to appear the same size, by surrounding the smaller circle with a ring of smaller circles, and surrounding the larger circle with a ring of larger circles. Yet although the perceptual illusion is robust, people asked to pick up the discs set the parameters for their actions correctly: they scale their grip according to the actual size of the objects, so that the grip size is the same when the two circles are really of the same size, and different when the two circles are of different sizes (Goodale 1996; Aglioti, De Souza, and Goodale 1995). This situation is written large in the case of the patient DF studied by Goodale and Milner 1995. After carbon-monoxide-induced anoxia, DF was unable to recognize everyday objects or faces, and could not visually identify simple shapes. She could not even show how big objects were or in what orientations. So she did not seem to have experience of the spatial organization of her surroundings. She seemed to lack any basis for verbal reports of sameness or difference of shape, size, and orientation. Nonetheless, she had strikingly intact visuomotor capabilities. When she had to pick up an irregular object, she angled her fingers optimally for the grip, though she could not say which irregular objects were the same or different shapes. When she had to post a card through a slit, her movements were accurate: she oriented the card correctly and the size of her finger grip was correlated with object size. Similarly, Perenin and Rossetti 1996 looked at a blindsight patient, PJG, who was requested to post a card through a slot in the blind field, or to grasp blocks presented in the blind field. Despite the lack of awareness, the patient was accurate on those tasks: orientation of the hand posting the card was correlated with the orientation of the slot, and finger-grip size in reaching for the blocks was correlated with the size of the block. Here again, the parameters for action seem to be set by a low-level visuomotor system, remote from consciousness.

Space and Action

53

There are many cases which show the opposite dissociation, intact spatial experience and an impaired action system. In cases of optic ataxia, patients can accurately display the size of an object by holding up their fingers to show how big it is. Or they can show the orientation of a slot they can see by holding up a card in the same orientation. So they seem to have accurate experience of the spatial layout of their surroundings. But these patients cannot accurately grasp the object, and they cannot post a card through the slot in the right orientation. Their visuomotor capacities have been impaired (Goodale 1996). Our discussion so far points to the distinction between two visual pathways, sometimes referred to as the 'action' and 'perception' pathways. This distinction is partly motivated by the anatomical distinction between dorsal and ventral visual pathways, which were originally described as 'where' and 'what' pathways. One way of setting out the distinction is provided by Jeannerod 1997, from whom I take Figure 2 (p. 54). The suggestion is that the various processing streams can be divided into (a) the computation of the direction and distance of the object, given in body-centred coordinates, for activation of reach, together with (b) size, shape and so on being computed in allocentric terms, for activation of grasp. Both of these processing streams belong to the 'action' pathway. They are contrasted with the 'perception' pathway, in which (c) properties of the object such as size, shape, colour, and so on are bound together and the object categorized by the visual system. The 'action' pathway is a relatively low-level system, remote from consciousness, used in the fine control of motor movement. Once we have this distinction in place, the natural proposal is that intentional action on a demonstrated object must involve the 'perception' pathway. For there to be a demonstrative referring to the object, there must be an involvement of high-level semantic identification of the object involved, provided by processing in the 'perceptual' pathway. 4. The Binding Thesis Does this mean that the epiphenomenalist is right, and that there is no role for the experienced location of the object in the causation of movements towards or away from the object? What I have argued so far is that the epiphenomenalist is right that there is no especial reason to suppose that spatial experience is what causally controls the spatial aspects of intentional action on an object. But that is not to say that there is no role whatever for the experienced location of the object to play in the causation of your action on the object. Suppose you identify an object on the basis of

54

Space and Action

Fig. 2. Diagrammatic representation of central processes involved in object-orientated behaviour. Extrinsic attributes of an object (related to its spatial position) are processed in bodycentred coordinates, for activation of the reach. Its intrinsic attributes are processed in a different, parallel pathway for activation of the grasp, also pertaining to the dorsal visual pathway and to the posterior parietal areas. Shape analysis, using the same object primitives, is effected both in the dorsal pathway for visuomotor transformation and in the ventral pathway for perceptual identification. Only a few components of visual processing (for example, size cues, depth cues, etc.) are mentioned. Others are symbolized by the empty box. Semantic knowledge stored in the ventral pathway can improve visuomotor transformation using connections between the two pathways. (From Jeannerod 1994, 1997.)

your conscious experience of it, and choose to act on it: you think, 'I'll pick up that penny'. Suppose we agree with the epiphenomenalist that a lowlevel visuomotor system has to be swinging into play to set the parameters for your action in picking up the penny. There is still the problem that your visuomotor system has to be swinging into play so that you can reach the very object that you identified on the basis of experience. But how does the visuomotor system manage to connect with the right object? In terms of Jeannerod's picture, a parallel problem arises if you identify an object on the basis of semantic processing—if you think, for example, 'I'll have that penny'—and choose to act on it, there is the problem that the 'action' system has to swing into play to pick that very penny, the one that

Space and Action

55

has been identified on the basis of semantic processing. There is a form of binding problem here, which is: how do we make sure that the 'action' and 'perception' systems are dealing with one and the same external object? I think that the natural proposal here is the one that Jeannerod makes. The picture he has in mind is that there may be early processing areas shared by the two visual pathways in which the visual primitives and spatial localization are presented on the same map (Jeannerod 1997: 80). In effect, the suggestion is that it is location that is the binding principle. Jeannerod talks about 'the classical Aristotelian notion that what arises from the same point in the external world pertains to the same object' (80). So on this view, 'attentional mechanisms would play a role in binding different modes of representation into a single, higher-order one' (80). I said earlier that conscious attention to an object is what you use in interpreting a perceptual demonstrative referring to that object. Moreover, it seems in general possible to act on objects that you have demonstratively identified. Consequently, there is the problem of explaining what the relation is between conscious attention and the capacity for intentional action on the object. The present proposal is that conscious attention to the object will include some awareness of the location of the object, and that the target for processing by the visuomotor system can be identified as 'the object at that location'. For this to be successful, the experience of location would not have to be accurate. Moreover, even if the initial experience of location was accurate, there could be subsequent changes in the location of the object which were not picked up by the 'perception' system but which were compensated for by the 'action' pathway in setting the parameters for action on the thing. The role of conscious attention to the object is not directly to set the parameters for action on the thing, but to provide enough information about the thing to define a target for the visuomotor system. Once the target is defined, it is up to the visuomotor system to set the parameters for action on the object. The epiphenomenalist rightly opposed the idea that the role of conscious experience of the object in action was to set directly the parameters for action on the object. The epiphenomenalist concludes that there is, therefore, no role for the experienced location of the object in action on the object. I am now suggesting that that is a mistake. Rather, the role of experienced location is this: conscious attention is what defines the target of processing for the visuomotor system, and thereby ensures that the object you intend to act on is the very same as the object with which the visuomotor system becomes engaged. And the way in which conscious attention identifies the target for the visuomotor system is by the experienced location of the thing. This whole procedure may work even though your experience of the location of the object is not particularly accurate. If

56

Space and Action

you see a penny in a mirror without realizing that you are seeing it in a mirror, you may use its apparent location in verifying that it is brown, and prehending your hand correctly to pick it up, even though your experience of its location is not actually correct. There is an obvious analogy with the behaviour of a heat-seeking missile. Once the thing is launched, it sets the parameters for action on its target in its own way; but to have it reach the target you want, you have to have it roughly pointed in the right direction before it begins, so that it has actually locked on to the intended object. The role I am envisaging for conscious attention to the object, then, is to find the target of the information-processing, which need not itself be conscious, involved in acting on demonstrative propositions about the object. The reason why location matters in your experience of the object is that your conscious attention to the object has to be identifying the target for the benefit of these information-processing systems. And location is one of the ways in which the information-processing systems identity their target. Christopher Peacocke (1981) maintained that intentional action on an object requires that you have an intention to act on the object, as identified using a perceptual demonstrative. Nevertheless, the thesis, that demonstrative identification is essential for intentional action on an object, seems too strong as it stands. Telephoning someone is an intentional action that is intentional with respect to the person rung up, even though they might not be in when you ring. But ringing up Sam does not require that I currently be able to demonstrate Sam perceptually. The same point could be made about letter-writing, voting for a candidate in an election, or suing someone who has written something disagreeable about you. So you can act intentionally with respect to an object even though you cannot currently perceptually demonstrate the thing. You might point out that these counterexamples involve quite special actions on objects, rather than the reaching and grasping that Peacocke evidently had in mind. So you might say that perceptual-demonstrative intentions are essential if you are to perform a reaching or grasping action on an object. The kinds of actions in question will have to be delineated quite finely, though, for this strategy to help. You have only to think of someone with prosthetic limbs activated by spoken commands to see the possibility of someone whose reaching and grasping movements do not depend on demonstrative identification. But no doubt we can distinguish between ordinary reaching and grasping, and the reaching and grasping performed by a person with voice-activated prosthetic limbs. Even in the case of ordinary reaching for an object, though, it does not seem that the use of demonstratives is essential to action on an object. Suppose we have a blindsighted subject who has a hat present in his blind field. The subject is not in a position to identify the hat demonstratively;

Space and Action

57

lacking any visual experience of the hat, he cannot refer to it as 'that hat'. The most the subject can do is formulate the description, 'the hat (if there is one) in my blind field'. But suppose the experimenter says, 'There is a hat somewhere in your blind field; can you reach for it?'. Since the subject now has enough descriptive identification of the object to provide a rough identification of it for his visuomotor system, he is certainly in a position to reach for the hat; given the rough descriptive guide, his visuomotor system can do the rest. The most we can say, on behalf of the thesis that demonstratives are essential for intentional action, is that usually you will not, in the absence of experience of the object, have any knowledge of where the object is, in a form that you can use to identify the thing for the benefit of your visual system. If we ask why Peacocke maintained that demonstratives are necessary for intentional action on an object, one answer is proposed by him in Sense and Content: Consider an example in which we explain your walking up to a particular man by citing your intention to shake hands with that (perceptually presented) man. The direction in which you walk varies systematically with the position your experience represents him as occupying relative to yourself. If in intention you think of a man descriptively, as when you intend to shake hands with the richest man in London, then there is no saying on the basis of your then current psychological states in which direction you would or should walk. (Peacocke 1983: 159)

Peacocke's argument here is an appeal to the idea that the parameters of your action on an object are set directly by your current experience of the object; it is, in effect, an appeal to the Grounding Thesis, and we have seen, I think, that we have to reject that idea. The parameters for your action on an object are, in general, set by a visuomotor system which may be remote from consciousness. The role for the experienced location of the object in causing your action is simply to identify, for the benefit of the visuomotor system, which object is in question. Once we have grasped that this is what is happening, there is no very obvious reason why the target of your action could not, on occasion, be identified in some other way, for the benefit of the visuomotor system. Indeed, this seems to be exactly what happens in Hindsight.

5. G. E. Moore's Commonplace Book In his Commonplace Book, G. E. Moore is vividly aware of a special connection between a visual demonstrative and the experienced location of the object. He goes so far as to propose that we can gloss the meaning of

58

Space and Action

the demonstrative in terms of location. He says, 'Let us say "This thing" = "the thing that is here", and "that thing" = "the thing that is there'" (1962: 158). If we explain the meaning of the demonstrative in that way, then 'This thing is here' is equivalent to 'The thing that is here is here', which, as he says, is redundant, but not tautological, since it is not tautological that there is anything at all here. But. he goes on to remark a problem for the attempt to give such a gloss of the meaning of the demonstrative, in terms of a description identifying the object as the one at a particular place. If you ask, 'What's that thing there?', is the mention of a location serving to make explicit the way in which you are descriptively identifying the object? In the following passage, Moore argues that the identification of location is not serving as part of a descriptive identification of the object. This leaves him with the problem, unresolved in this passage, of explaining what the relation is between the visual demonstrative and the experienced location of the thing. What's that thing there? = (1) what's it called?, (2) what does it do, what's it for? I asked this at the Manse, and the answer was 'It's a "mixer", and is for beating up eggs, etc.'. Now this question is one which would only be asked by a person who was seeing or had recently been seeing the thing in question (a blind man might ask 'What's this thing here?', but not 'What's that thing there?', the seeing man says 'this thing here' when he's touching or nearly touching it), and he wouldn't understand the answer unless he was seeing it (the answer wouldn't be an answer to his question). Hence, 'there' for him does not merely mean, e.g. 'the nearest thing in that direction', but 'in the place where that thing is', where 'that thing' cannot = 'the thing that is there'. Can we say 'that thing' = 'the thing at which I am pointing' or 'the thing to which this finger points'? No, because the prop, in question is not understood until the thing in question is seen. 'There's a match-box there' 'That's a match-box': it answers 'Is there a matchbox anywhere about?'. The other doesn't. (Moore 1962: 158)

The proposal Moore is making in the second paragraph in this passage is that the object is not being identified by which place it is at; rather, the location in question is being identified as the location of the object. As he puts it, 'there' is being equated with 'the place where that thing is'. Consequently, we could not without circularity go on to identify which thing we were talking about by saying that it is 'there' (that is, at the place where 'that thing' is—which thing?). We have, though already seen some of the problems with this move— explaining the special connection between a visual demonstrative and the experienced location of the object by saying that the location is identified as the place at which the object is. The whole point of saying that experi-

Space and Action

59

enced location is a frame-of-reference phenomenon is that the locations are identified prior to identifying the objects at those locations. It is only because there is this prior identification of places that we can make sense of there being visual illusions about the locations of seen objects, or movements of seen objects which do not involve objects in changing their locations with respect to one another. In effect, Moore makes a point sympathetic to this in the last paragraph of the above passage, when he remarks that 'There's a match-box there' is an answer to the question, 'Is there a match-box anywhere about?'. The use of 'there' is to locate the matchbox in the domain of search. If it were identifying the location merely as 'the place occupied by that matchbox', the use of 'there' would add nothing to the remark, 'There's a match-box', which merely asserts the existence of a matchbox without saying that it is in the search domain. Hence, this use of 'there' must be exploiting some prior way of identifying places. Moore thus ends this passage with a problem unresolved. The problem is to explain the special connection between the visual demonstrative referring to an object and the experienced location of the object. There are two proposals he considers to explain the connection. First, he suggests that 'that thing' is equivalent in meaning to the description, 'the thing that is there (at that place)'. The problems with this proposal are (a) that it cannot explain why it should be essential to an understanding of the demonstrative that you should see the object itself; after all, you could understand the description without seeing the object itself, and (b) if the place in question is identified only as the location of the object, then you cannot say which object you have in mind by saying that it is the object at 'that place'; so some alternative account of how the place is identified would be needed. The second proposal he considers is that the place is identified merely as 'where that object is'. But as we saw, this would make errors of experienced location impossible, and leave unexplained the use of 'There's a matchbox there' as significantly different to There's a matchbox'. You might say that this shows only that the relevant descriptive condition is not 'the object at position X', but 'the object which looks as if it is at position X'. The point to notice about this is that there is something metarepresentational about this use of 'looks'. We might gloss 'the object which looks as if it is at position X' as 'the object which the visual system represents as being at position X'. And to determine which object that is, it is not enough to find which object actually is at position X; the experienced location is now merely a dummy in identifying which object is being referred to. If you tell me about Bill who has committed a murder, and I ask, 'Who is Bill?', it does not begin to answer the question to say, 'He is the person who I have just told you committed a murder'. When I do begin a

60

Space and Action

serious inquiry to identify who you were talking about, the idea of his having committed a murder may simply not be to the point: the fixing of reference may be much more fundamental than anything about the murder, it may relate entirely to his having been someone you have known for years, and so on. Nonetheless, Moore continues to insist that '"This is here" is certainly partly tautological, in some uses' (1962: 154). But if the discussion in this book has been right up to this point, that is not quite the right way to put it, because the experienced location of the object may be an illusion, so there is nothing tautological about the proposition which articulates that the object is at its experienced location. What we can say is that there is still a role for experienced location in fixing the meaning of the demonstrative, because it is the experienced location of the object that functions to control the verification of propositions about it and action on the basis of propositions about it.

4 Sortals 1. Styles of Conscious Attention Our theme is the role of conscious attention in providing knowledge of the reference of a demonstrative. Knowledge of reference, we have seen, can be viewed as causing and justifying the use of particular ways of verifying and finding the implications of propositions. So far, we have looked at the sense in which conscious attention to an object can be said to cause and justify the use of particular information-processing procedures to verify, or to act on the basis of, demonstrative propositions concerning that object. However, the things to which we can refer by means of demonstratives are of many different sorts. You can, for example, refer demonstratively to people, to valleys, and to clouds. These are different sorts of things, in the sense that there are striking differences in the ways in which their identities over time, and their boundaries at a time, are determined. In determining whether the person in the dock is identical to the person who committed the crime, for example, spatiotemporal continuity is perhaps one criterion being used; but the style of spatiotemporal continuity in question here is quite different to that relevant in finding whether this is the valley in which an infamous massacre was performed—we don't, for example, allow for the possibility of movement by valleys. And the identities of clouds have little of the complexity associated with the identities of persons. We register these differences between types of objects when we use what Locke (Essay, II. iii. 15) called a sortal concept. The idea of a sortal concept was introduced to the recent literature by P. F. Strawson in Individuals, as follows: A sortal universal supplies a principle for distinguishing and counting individual particulars which it collects. It presupposes no antecedent principle, or method, of individuating the particulars it collects. (Strawson 1959:168)

Examples of sortal concepts will include 'person', 'valley', 'cloud', and so on. The idea is that understanding such a term involves having a grasp, however provisional or inarticulate, of the criterion of identity relevant to a particular type of thing. In contrast, understanding terms such as 'red' or

62

Sortals

'square' does not involve any grasp of a criterion of identity; they can be applied indifferently to things of quite different sorts. As I said, I have been stressing the role of conscious attention in providing knowledge of the reference of a demonstrative. And different styles of conscious attention will be used in attending to different sorts of object. For example, if you are consciously attending to a person over a period of time, the way in which you keep track of that person will be quite different from the way in which you keep track of a valley to which you are attending over a period of time. In consciously attending to the person you will, for example, keep track of the movements of that person; but in the case of the valley you will make no such allowance for movement (certainly none for movement with respect to the flanking mountains). These differences in style of attention amount to differences in what I called the complex binding parameter used by the visual system in putting together the information true of the object. The binding parameter for a person will have to allow for the possibility of movement by the person; the binding parameter for a valley will not have to allow for any possibility of movement by the valley. As I said, the complex binding parameter in effect provides an address for the object, by which it can be identified, at the level of conscious attention, in a way that can be used in recruiting information from various processing streams to allow verification of propositions about the object, and action on the object. So the style of conscious attention to the object that is appropriate will depend on what sort of object is in question. The use of one style of conscious attention rather than another—that is, the use of one type of complex binding parameter rather than another— seems on the face of it to be a more primitive phenomenon than the ability to use sortal concepts to classify the objects to which you can attend. Animals other than humans plainly have a repertoire of binding strategies available to them: a cat keeping track of a mouse is plainly using different binding strategies than a cat keeping track of its home. It seems evident, also, that you could be using various styles of conscious attention in keeping track of various of the things around you in a new environment, for example, without yet having learned what sorts of things any of them are. Philosophers interested in the fact that we can make demonstrative reference to various quite different sorts of physical object have frequently proposed that our understanding of sortal concepts has a foundational role to play in making it possible for us to refer to objects of these various sorts. What I want to propose in this chapter, though, is that grasp of sortal concepts is a more sophisticated matter than is the mere capacity for demonstrative reference. What is important, in our capacity for demonstrative reference to different sorts of object, is our capacity to engage in different styles of conscious attention. And indeed, that capacity to engage

Sortals

63

in different styles of conscious attention is what plays the roles that have so often been assigned to our grasp of sortal concepts. In sections 3-5 of this chapter, I will try to define the roles that sortal concepts have been taken to play in our understanding of demonstratives. And I will argue that those roles are indeed best viewed as being taken by the more primitive phenomenon: capacities to engage in various styles of conscious attention. As a preliminary, though, I want to look first, in section 2, at the idea that our grasp of sortal concepts may underlie our capacity for various styles of conscious attention. We shall see that while this idea is engaging, it does not seem to be correct. The capacity for various styles of conscious attention is a more primitive phenomenon than grasp of sortal concepts. 2. What Justifies Binding? So far, I have talked about the binding strategies that the visual system uses to put together the various properties of a single object. The question I want to address now is: what causes and justifies our use of those strategies? Is there a role for sortal concepts in explaining how we come to use those strategies, or in explaining why those strategies are correct? To see just what question I have in mind here, and how a sortalist might answer it, it is illuminating to look first at the kinds of binding strategy used in hearing. Suppose that as you sit in a building you can hear the roar of a helicopter directly above the building; you cannot yet see the thing, though. Plainly you are in a position to use the auditory demonstrative, 'that helicopter'. There may be other sounds in the vicinity than the sound of the helicopter—people talking and so on. But your auditory system has put together the sounds produced by the helicopter. It has effectively solved the auditory analogue of the Binding Problem, for that object. How in detail the auditory system organizes sounds so that sounds produced by the same object are put together is not a topic I will address here (for discussion, see Bregman 1990). What I want to consider is a question that arises at the most general level: namely, what is the justification for the way in which the auditory system organizes the information it receives? And why does the auditory system organize sounds in the way it does? I think that in answering those questions you might suppose that there is a dependence of auditory organization on visual organization. First, when someone uses an auditory demonstrative such as 'that helicopter' to refer to a visible object, they must be able to understand an identity proposition of the form, 'That (heard) helicopter is identical to that (seen) helicopter'. You must be willing to allow that, in principle, you could be

64

Sortals

confronted visually with the helicopter you are hearing. Otherwise, your auditory demonstrative would not be referring to a helicopter at all, but to something quite different, namely, a sound. The organization of the auditory world is, for us, dependent on the organization of the visual world. It is because we know through vision of such things as helicopters that we can organize the auditory world as we do. You understand the auditory demonstrative 'that helicopter' only because you understand how an identity of the form, That (heard) helicopter is identical to that (seen) helicopter', could be true; there is no converse dependence of your grasp of the visual demonstrative, 'that (seen) helicopter', on the auditory demonstrative. So when your auditory system is organizing the auditory information from that helicopter into a single whole, it is using a principle of integration that depends on vision; it depends on the knowledge you have, on the basis of vision, of what such an object is. The integration that is done in hearing depends on the binding that is done in vision. Putting matters like this naturally raises a question about those who have been blind since birth. Surely such a person could still understand auditory demonstratives like 'that helicopter'. But: such a person does not have a potential to understand visual demonstratives referring to the same object. If hearing still has a dependent status for such a subject, it must be dependent on some other sense-modality, presumably touch. I do not want to attempt a serious discussion of this issue here, since it would take us very far afield. For the moment, I want to remark only that the concept formation of those born blind is obviously quite radically different to the concept formation of the sighted; and that those born blind are evidently aiming to live in a social world largely defined by the sighted, so there is evident pressure on them to conform their concepts as closely as possible to those of the sighted. The critical point for us is that the auditory demonstratives of the sighted may nonetheless depend on grasp of the connection between auditory and visual demonstratives. How is the connection made between auditory organization and visual binding? A simple, sortalist suggestion is this. You learn on the basis of vision what a helicopter is. You now grasp the sortal, 'helicopter'. In hearing, you aim to find whether there are such things as, for example, helicopters in your surroundings. So the organization of the auditory scene is driven by the objective of discerning things of the sorts with which you have been made familiar by vision. If this picture is right, then the organization of the auditory scene depends on our grasp of sortal concepts. So there is indeed a sense in which the conscious attention to the heard helicopter, which explains your grasp of the auditory demonstrative, 'that (heard) helicopter', has been focused by your grasp of such sortals as 'helicopter'.

Sortals

65

Even on its own terms, there are a number of ways in which this line of thought might be questioned. In particular, even if you accept that auditory integration is dependent on visual binding, you might question whether the link has to be mediated by sortal concepts. There is, after all, cross-modal binding, which seems to be a relatively low-level matter, in that it does not require the active involvement of sortal concepts. When you are looking right at something that is plainly making a noise, for example, the visual and auditory information is integrated in a way that does not depend on your conceptual thought about what is happening (cf. Ward et al. 1998 for a review discussion of cross-modal interaction). The purely auditory demonstrative may be, as it were, a limiting case of a cross-modal demonstrative, rather than being a separately motivated demonstrative dependent on visual demonstration. Whatever we say about that point in the argument, the point I want to stress for the present is that this argument at best establishes a role for sortal concepts in auditory binding to mediate the dependence of auditory binding on visual demonstratives. This line of argument does not establish a role for sortal concepts in causing or justifying the use of binding strategies in vision. Consider, for example, the use of location as a binding parameter. As I said earlier, it seems likely that the visual system uses location as a fundamental parameter in binding together different features as features of the same object: features at the same location are, as a first approximation, assigned to the same object. Someone who thinks there is a foundational place for sortal concepts in explaining how there can be conscious attention to objects may press the question: why is binding together features at the same location the right procedure for the visual system to use? What causes and justifies the use of this binding procedure? The null hypothesis is that there is nothing which justifies the use of one binding procedure rather than another. On this hypothesis, which binding procedures the visual system uses is more fundamental than either the question which objects there are in the environment, or the question which sortal concepts the subject employs. It is more fundamental than the question which objects there are in the environment, because, the argument runs, the very notion of the 'environment' is always relative to a particular type of creature; it depends on the ecological niche the creature inhabits. And, on this view, the way in which the perceptual system solves the Binding Problem is one of the things that determines what kind of world the creature inhabits. Moreover, on this view, which procedure the visual system uses to solve the Binding Problem is also more fundamental than the question which sortal concepts we use. For which sortal concepts we use depends partly on what our perceptual skills are; the most fundamental sortal concepts are observational concepts, which we can apply to

66

Sortals

objects on the basis of perception alone. But then the way in which our perceptual system solves the Binding Problem is one of the things which determines what kinds of sortal concepts we can have and use. So our way of solving the Binding Problem cannot be justified by reference to our use of one rather than another collection of sortal concepts. Rather, our grasp of sortal concepts simply has to work with whatever solution to the Binding Problem the visual system finds. In any area, there is always, I think, a presumption in favour of the null hypothesis; but it is a startling thought that the choice of a particular procedure to solve the Binding Problem may be quite unconstrained by anything other than the internal requirements of the visual system. We can see what the null hypothesis is saying by considering for a moment what would have to happen for the visual system to solve the Binding Problem in a radically non-standard way. Quine (1960) and Goodman (1951) used to talk about spatiotemporally scattered objects, which they claimed were just as real as anything else. So a typical spatiotemporally scattered object might comprise the top of one chair plus the base of a lightbulb. Non-standard binding would involve putting together the perceived properties of the top of the chair and the perceived properties of the base of the lightbulb as properties of a single, albeit spatiotemporally scattered, object. These features would be combined to give an 'object token', which could then be compared to stored representations to determine its characteristics. We could have a still more radically nonstandard form of binding. The properties of the top of the chair would all be bound, and the properties of the base of the lightbulb would all be bound, on the picture I just gave; the odd part is just putting the two together. But we could in principle have binding in which no two features from the same location were put together. The redness at this location, the squareness at that location, and the uprightness from a still further location could all be put together to give a single 'object token', to be compared to stored representations. The upshot would be a kind of collection of spatiotemporally scattered tropes. According to the null hypothesis, there is no justification to be given for the visual system operating in the ordinary way rather than in the ways I just described; there is no objective advantage in the standard approach. The visual system proceeds in whatever way it does; that is a primitive datum. The way in which the visual system proceeds will determine which object tokens are constructed, and that in turn will determine what stored object representations the subject has. This in turn will determine what sorts of objects are in the subject's environment; it is up to the subject to use one system of object representations rather than another to delineate which things in the surroundings he is thinking and talking about. It is

Sortals

67

natural to protest, as against the null hypothesis, that the way in which we actually bind objects has objective advantages over these bizarre alternatives. But the natural suspicion is that this protest is just a parochial, conservative reaction which elevates habit into a kind of transcendent superiority. What reason could be given for thinking that one way of binding features into objects is better than another? I think we can immediately say that it is not believable that there are no normative constraints on binding. There do appear occasionally individuals whose visual systems have problems with binding; see, for example, Friedman-Hill, Robertson, and Treisman 1995, or Robertson et al. 1997. These patients are quite seriously impaired. Though they may, for example, be above chance at saying which features are present in a scene displayed to them, they will be at chance when saying which combinations of features are present. Or, as in the case of the patient described by Humphreys and Riddoch (1987), they may be able to copy a drawing of a complex scene accurately, but be doing it without any identification of the objects involved. It is difficult to accept that since there are no norms of binding, these patients cannot be described as impaired. Rather, it seems that there must be normativity here, since these patients are so palpably impaired. However, you might have a view on which getting it right in your use of particular binding procedures is simply a matter of doing it the same way as everyone else. That is, you might have a view of binding which is like Chomsky's view of the innate systems he holds to be involved in language use. On this picture, there is such a thing as getting it right or wrong in your use of a particular grammar; but ultimately, rightness and wrongness here are just a question of whether you are in step with other people of the same species. Similarly, you might have, as it were, a 'community view' of binding procedures, on which rightness or wrongness is simply a matter of agreement or disagreement with others in your community. This view implies that the only problem with a non-standard binding strategy, such as putting together features at different locations, and the only problem with the impaired patients I just mentioned is social. Their only problem is that they bind differently to other people. But that seems entirely inadequate as an analysis of the problem. The problem with these subjects is that they cannot see the objects around them. The sortalist proposal, at this point, contains two elements. One is a causal hypothesis: that our visual systems use the binding strategies they do because we have the sortal concepts that we do. The other element in the sortalist proposal is normative: that what defines the objective of the binding strategies being used is the need to keep faith with the system of sortal concepts used by the subject. It has to be said, though, that neither of these ideas is compelling. Given the similarities between human vision and

68

Sortals

vision in other species, it seems somewhat unlikely that the use of particular binding strategies in humans evolved in response to our possession of particular sortal concepts. And it is difficult anyway to see how our possession of particular sortal concepts could have come first; our grasp of the concepts of particular sorts of observable objects depends on our abilities to perceive them, which in turn depends on having the relevant binding strategies in use already. Grasp of a system of sortal concepts thus seems to depend causally on possession of a relevant set of binding strategies, which makes it hard to see how there could also be a causal dependence in the other direction. The second element in the sortalist proposal is also difficult to accept. According to the second element, it is our system of sortal concepts that defines the objectives of the binding strategies that we use. Since the system of sortal concepts that we have presumably developed later than the binding strategies that we use, the sortal concepts are in effect providing a kind of after-the-fact justification for the use of those binding strategies, on this view. But if we have in fact developed a system of binding strategies in virtue of which we can see various sorts of objects, then we do not need the development of sortal concepts to provide a justification for the use of those strategies. The objective is simply to see the relevant objects, and binding strategies achieve that. To develop a system of sortal concepts consequently upon this achievement, and then maintain that the objectives of the binding strategies were after all being set by this system of sorta! concepts, adds nothing. We already have it in place that the objective of the binding strategies is to let us see the various sorts of objects around us, and there is no place for a further level of justification.

3. The Delineation Thesis Philosophers have, nonetheless, sometimes maintained very strong theses of sortal dependence. They have held that the ability to single out in experience a tree, or a mountain, or a lion depends on grasp of such sortal concepts as 'tree', 'mountain', or 'lion'. Their basic point is that there can be different things of different kinds in the same place at the same time. Suppose we consider a bend in the river. What object is there, at that point? There is, of course, the river, which continues on downstream. There is also a stage of a river: that is, the river as it is at this particular moment, for example. A 'river-stage' so defined cannot last longer than a moment, unlike the river. There is also a collection of water molecules at that point. This collection will have been diffused in the sea by tomorrow, unlike the

Sortals

69

river itself. There is also a stage of that collection of water molecules, at the same place: a momentary stage in the life of that collection of water molecules. Quine gave the basic argument here when he said: Pointing is of itself ambiguous as to the temporal spread of the indicated object. Even given that the indicated object is to be a process with considerable temporal spread, and hence a summation of momentary objects, still pointing does not tell us which summation of momentary objects is intended, beyond the fact that the momentary object at hand is to be in the desired summation. Pointing... could be interpreted either as referring to the river ... or as referring to the [collection of water molecules] . . ., or as referring to any one of an unlimited number of less natural summations... [SJuch ambiguity is commonly resolved by accompanying the pointing with such words as 'this river', thus appealing to a prior concept of a river as one distinctive type of time-consuming process. (Quine 1953a: 67)

The argument here is explicitly about pointing, which is a way of orienting attention—it is one of the control mechanisms of attention—rather than conscious attention itself. And you might argue—indeed, later in this chapter, I will argue—that Quine's argument should be read as an argument about the control of attention, rather than about what is involved in consciously singling out an object. But that is not the natural reading of the significance of Quine's comment. On the natural reading, Quine's comments imply that if you are to single out an object in experience, you have to be employing a relevant sortal concept, to delineate the boundaries of the object singled out. The idea is that that sortal concepts play, as it were, a top-down role in making it possible to pick out objects in experience. We could put the idea in terms of what I will call the Delineation Thesis: The Delineation Thesis: Conscious attention to an object has to be focused by the use of a sortal concept which delineates the boundaries of the object to which you are attending. It is, indeed, very often assumed that if you are to attend to Fs, you must already have the concept of an F (cf. e.g. Fodor 1998). But straight off, that is hard to believe. Of course, you cannot be intentionally attending to Fs without having the concept of an F; but for all that, you can be attending to an F without knowing that it is an F and without having that concept. According to Quine, the involvement of the sortal concept is needed for there to be a determinate answer to the question: 'To which object is the subject consciously attending?'. If we do not appeal to the subject's grasp of a sortal concept, how could we say what the difference is between attending to a river and attending to a collection of water molecules, for example? I think that we can answer this question by appealing to

70

Sortals

what I earlier called a complex binding parameter as focusing the subject's conscious attention. Since visual information-processing involves computation of the object's various characteristics in separate processing streams, the visual system has the problem of putting together all the information, in various processing streams, that relates to the same object. I have suggested that we can think of this as exploiting the location of the object together with something like the Gestalt organization of the characteristics found at that location. And, course, the visual system can bind together features over time, as when we keep track of a moving object, and across sensory modality, as when we assign heard speech to the person seen before us. This is a more primitive phenomenon, involving a more primitive level of information-processing content, than the application of sortal concepts. And I have argued that we should think of the complex binding parameter used as playing a role also in conscious attention to the object: it provides, in effect, a way of identifying the object which can be used to find information about that object in various processing streams, or to single out the object for the purpose of acting on the object. So the singling out of an object in experience need not involve the application of sortal concepts; only the mechanisms of binding. Whether you are consciously attending to a river or a mass of molecules, for example, will show up in how your visual system binds together information from the thing over time. If you have to keep moving downstream to keep track of the object of your attention, then you are attending to a collection of water molecules rather than a river. If, on the other hand, you are binding together information from any point in the course of the river, as all relating to a single object, then you are attending to the river itself. The distinction between consciously attending to a collection of water molecules, and consciously attending to a river, is not particularly hard to draw, even without appealing to any grasp of sortal concepts by the subject. It is, anyway, a fundamental point that demonstrative reference to an object can succeed even though you do not know what sort of thing you are dealing with. In a garden centre I exclaimed at the beauty of a particular plant, only to be told that it was plastic. There was no relevant ambiguity about which thing I was talking about. I had plainly referred to the plastic plant. But a plastic plant is not a plant, any more than a fake Rembrandt is a Rembrandt. You can succeed in singling out an object despite a mistaken belief about what sort of thing it is. A still more radical case is the case in which you succeed in singling out the object even though you have not the slightest idea what sort of thing it is. Suppose that by some chance, an ordinary teacup from today survives for a thousand years. And suppose that when it is discovered by our descendants, they and their circumstances have changed so dramatically that they have considerable conceptual diffi-

Sortals

71

culty in understanding what a teacup is. To grasp the concept they would need to have a lot of background filled in of which they know nothing as yet. But the discovered teacup prompts a lot of discussion. There are learned conferences speculating as to its nature. A strong body of opinion leans to the view that it 'probably had some religious significance'. It is kept in a glass case in a museum. Given the intense discussion it receives, it would be absurd to say that our descendants have not managed to 'single it out'. They can certainly make demonstrative reference to it; their experience of the object is sufficient to specify uniquely which thing is in question. In both these cases, the plant and the teacup, what is happening is that your visual system is managing to bind together the information from a single thing, and you are consequently able to attend consciously to it, even though you have not managed to apply the right sortal concept to it. Application of the correct sortal concept seems therefore to be a more sophisticated phenomenon than conscious attention to the object; the Delineation Thesis is simply false. The work that the Delineation Thesis allots to grasp of sortal concepts ought rather to be assigned to the principles used by the visual system in binding together the various characteristics of a single object. In particular, it is a capacity for binding, rather grasp of sortal concepts, that allows the subject to delineate the object in experience. Another way to put the point is in terms of the classical distinction between 'associative' and 'apperceptive' visual agnosias (Lissauer 1890; for discussion see Farah 2000). On the one hand, according to the classical distinction, there are 'associative' agnosias in which the patient sees the object, but does not recognize which object he is seeing—so the patient may be able to give quite a full description of the volumetric properties of, for example, a glove, but be at a loss to say what such a thing might be used for (cf. Humphreys and Riddoch 1987 on 'semantic access agnosia'). A patient of this type seems perfectly capable of using and understanding perceptual demonstratives. Such patients seem to be plain counterexamples to the Delineation Thesis; there is no question but that they have singled out the object in experience, though they are unable to give a semantic classification of it. The patient is even, in this kind of case, able to find what sort of thing it is that he is looking at. His understanding of the demonstrative means that he has the resources to understand a demonstration that the object is of this or that sort; it may be only the capacity for specifically visual recognition of the sort of the object that is impaired. What makes it so compelling that the patient is able to understand visual demonstratives is that his situation is, after all, not so very different from that of an ordinary subject given an unfamiliar view of an object of a familiar type, or a view of an object of a sort he has never seen before. In those cases, vision alone does not allow

72

Sortals

you to classify the object. But it is nonetheless compelling that in those cases you are still able to identify the object demonstratively, and to formulate hypotheses about that thing, or plans to act on it thus and so. Consider, in contrast, the so-called 'apperceptive' agnosics. An agnosic of this type may be able to copy complex figures, even though he insists that he has no idea what he is drawing. The patient will have no idea which objects there are to be found in the scene he has successfully copied, even though that is evident to an ordinary subject looking at the copy. Humphreys and Riddoch (1987) suggest that what is wrong here is, in effect, a problem with binding (cf. their discussion of 'integrative agnosia'). The patient is able to see the simple components of the scene and where they are; it is just that there is a difficulty with visual integration. When such an agnosic views a scene, is he in a position to understand demonstrative reference to the object before him? It seems evident that he is not. He cannot, on the basis of vision, verify propositions about the object he is seeing; he is in no position to act on the object, arid most fundamentally, his experience does not provide him with knowledge of which object is in question. It thus seems that for experience of an object to provide you with knowledge of the reference of a demonstrative referring to that object, there must be sufficient integration of the object in experience—the various features or parts of the object must provide experience of it as a coherent single thing—but this experience of the object as a coherent unity need not involve semantic classification of the object as an object of this or that sort. In Sameness and Substance, David Wiggins advances a form of the Delineation Thesis which seems designed to accommodate the possibility of singling out an object in experience even though you have made a mistake about what sort of thing it is. Wiggins says that Tor a thinker to single out or individuate a substance, there needs to be something . . . about his rapport with x or his relational state towards x and his practical sensibility in relation to x, which ... sufficiently approximates to this: the thinker's singling x out as x and as a thing of a kind f such that membership in f entails some correct answer to the question, "what is x?".' (Wiggins 2001: 7). The key point here is the idea that the subject has only to 'approximate towards' singling out the thing as a thing falling under the correct sortal concept f. So the subject can still single out the object so long as he is approximating towards correct sortal classification of it. But there are three problems with this. One is that it is not easy to see what is meant by 'approximating towards' in this context. When is one sortal misclassification a better approximation towards the right classification than another? Suppose that in the garden centre I had thought that the object was plastic, but not realized that it was a plant—perhaps I thought it was a gardening implement.

Sortals

73

Is that a better or worse approximation than when I thought it was a plant? There seems to be no way of answering this kind of question. The second problem is that it is simply a theorist's fiction to say that 'approximating towards the correct sortal classification' is what must be happening in the case of the teacup. It may be that as time goes on our descendants simply become less and less likely to reach a correct view about what sort of thing it is. There is nonetheless no indeterminacy as to which thing they are singling out. They are singling out the teacup. The third, most fundamental problem is this. If you are making a mistake in your classification of the object, how can that mistake be what allows you to single out the object? Surely you must be singling out the object despite your mistake, rather than because of it. How could success in a cognitive task—singling out—be constituted by a mistaken classification of an object? On the analysis I am proposing, the conscious singling out is done at a more primitive level than the one at which we have application of sortal concepts. Sortal concepts are applied to an object which you have already singled out, in consequence of your having succeeded in singling it out. And that is why singling out is consistent with incorrect, or no, sortal classification of the object. Sometimes what Wiggins says comes only to this: if you have singled out an object, then there must be a determinate answer to the question what sort of object it is. For example, he says: no substance has been singled out at all until something makes it determinate which entity has been singled out; and for this to be determinate, there must be something in the singling out that makes it determinate which principle is the principle of individuation for the entity and under what family of individuatively concordant sortal concepts it is to be subsumed. (Wiggins 2001:150-1)

This thesis has some independent plausibility; it is not, actually, a thesis about singling out, but a metaphysical thesis about the relation between existence and the possibility of sortal classification. It amounts to this: that every object can be subsumed under some type of sortal classification. If that is true, then of course singling out an object is singling out something that can be classified by a sortal. But then the point about singling out is merely a consequence of the more fundamental thesis, that everything admits of sortal classification. However, what Wiggins wants comes to more than this. Notice to begin with that in stating the metaphysical thesis, there would be no point in appealing to the notion 'approximating to'. The thesis is not that everything that exists approximates to being classified under some sortal or other. The metaphysical thesis is rather that everything that exists has some quite definite sortal classification. So the thesis about singling out comes to more than the metaphysical thesis: the idea is that singling out involves approximating towards the right sortal

74

Sortals

classification of the object. And Wiggins's remark was that to be singling out an object you have to be approximating towards singling it out as an object of a particular sort. On the face of it, this is a thesis about the work of the mind in singling out. It comes to more than merely the idea that each object is in fact of some definite sort. Suppose we accept the metaphysical thesis, that every object is of some definite sort. What is the argument for saying, further, that to single out an object you must approximate towards, or tend towards, singling it out as an object of some definite sort? It is not quite easy to pin down Wiggins's argument here, but here are some selective quotations which suggest a line of thought: Conceptualism [contends] that, even though horses, leaves, sun and stars are not inventions or artefacts, still, if such things as horses, leaves, sun and stars were to be singled out in experience at all ... then some scheme had to be fashioned or formed, in the back and forth process between recurrent traits in nature and would-be cognitive conceptions of these traits, that made it possible for them to be picked out. (Wiggins 2001: 150)

The role of the conceptual scheme deployed upon experience is then, on the interpretation I propose, specified in the passage which follows, and that I quoted above: there must be something in the singling out that makes it determinate which principle is the principle of individuation for the entity and under what family of individuatively concordant sortal concepts it is to be subsumed. (Wiggins 2001:151)

On this picture, the role of sortal concepts in singling out is to make it determinate which thing is being singled out. Without them, there would be no determinacy about what was being singled out. If you ask why it is not enough merely to experience a place to single out whatever it is that is at that place, the answer is: There will exist too indefinitely many items with too many distinct principles of identity and persistence which you might find in that place—the thing, the parcel of stuff that makes up the thing, and the mereological sum of all the components of that parcel, to name but three. (Wiggins 2001: 150 n. 13)

So this returns us to the argument from Quine with which I began this chapter. And exactly the same style of reply seems apt. First, it is a peculiarity of both Quine and Wiggins that they seem to find conscious attention to a particular place relatively unproblematic; there is no suggestion that attending to a place requires use of the sortal concept 'place'. Why attention to an object should not similarly be possible in the absence of a sortal concept is not explained. Moreover, it is perfectly possible to draw the distinctions Wiggins wants without appealing to sortal concepts.

Sortals

75

Conscious attention to the thing is manifestly quite different from conscious attention to the parcel of stuff that makes up the thing. Keeping track of the thing, even over quite a brief period of time, is manifestly different from keeping track of the parcel of stuff that makes up the thing; the difference will show up immediately the parcel of stuff is subjected to any fragmentation that does not affect the integrity of the thing, as when a small bit falls off. Someone attending to the thing will ignore the bit; someone attending to the parcel of stuff will either have to abandon the exercise or (depending on how we read 'parcel') may keep track of the totality of scattered stuff As for mereological sums, it is not obvious that they are visible at all, though their parts may be visible, so it is not obvious how you could be visually attending to such a thing. Notice also that Wiggins's argument above seems after all to demand that we have correct sortal classification of the object by the thinker. If we suppose that the mind does have to do the work of fixing the sort of the thing that is being singled out, it makes no sense to suppose that it might do the work by supplying an incorrect sortal classification. And in fact, if fixing the sort of the thing being singled out really is the task of the mind, it makes no sense to suppose that it could get it wrong; what else is there to determine what sort of thing is being singled out? If you suppose that something else—call it 'the world'—fixes the sort of the thing that is being singled out, then you can make sense of the idea that the mind could be making a mistake when it classifies the object. In that case there is no need for the mind to supply the sortal, in order to fix which sort of thing is being singled out. That task had been performed by 'the world'. But if you suppose that the world does not perform that task, then the talk of the mind doing its part by merely 'approximating towards' something or other, really makes no sense. Towards what is the mind supposed to be approximating? 4. Sortals as Orienting Attention There are two quite simple points which together can make it seem that there must be something right about the Delineation Thesis. One is that we usually do use sortal concepts in demonstrative constructions; very often, in discussion, you would not know which thing was in question unless your interlocutor used a sortal. The second point is that if someone uses an incorrect sortal classification in a demonstrative construction, then what they have said cannot be regarded as correct as it stands; once the facts are known, the statement has to be withdrawn and replaced by a use of the correct sortal classification. These points together can make it seem that

76

Sortals

(a) concepts are indeed essential to demonstrative reference, and (b) cases in which there seems to be singular reference but incorrect sortal classification are, after all, not cases in which the singular reference has been successful. But actually these points need to be independently explained, and they do not in the end offer any support to the Delineation Thesis. 1 will take them in turn. On the first point, there is a role for sortal concepts in demonstrative identification which is much less fundamental than the kind of role allotted to them by the Delineation Thesis, but which is commonplace and pervasive. When I commented on Quine's discussion of pointing, I remarked on the distinction between conscious attention to an object, and the control, or causation, of conscious attention. We can distinguish between the factors in virtue of which you can be said to be consciously attending to one particular object rather than anything else, and the factors which caused you to be attending to that object rather than anything else. For example, in the case of vision, it is arguable that you attend to a specific object in virtue, in part, of the fact that you are attending to the location the object is at. In contrast, I might orient your attention to the object by pointing, or by physically turning your head towards it while waggling the object in front of you. You can see the distinction between the two cases. Suppose that attending to the object is partly constituted by attending to the place, while the turning of your head is merely a cause of your attending to the object. Then you could have attended to the object, in the very same way, as a result of some other cause. For example, I could simply have pointed to the thing, or you might have become interested in it spontaneously. The upshot could still have been that you consciously attended to it in the very same way. As I said, we do not in ordinary communication confine ourselves to saying 'this' and 'that'; we very often do use phrases of the form, 'this F', where F is indeed a sortal term. Does this show that the Delineation Thesis is after all correct? I think it does not. We can distinguish between two ways in which a descriptive component can figure in demonstrative reference. Suppose you and I are strolling through the park and I point my umbrella and say something about 'that clock to the left of the fountain'. How does the descriptive component 'to the left of the fountain' relate to the demonstrative 'that clock' in this case? There are two possibilities. One is that it states a descriptive condition which has to be met by any object, if it is to be the reference of my term. So if there is no clock to the left of the fountain, I have not referred to anything. The other possibility is that the phrase, 'to the left of the fountain', is just a way of orienting your attention, exactly on a par with the flick of my umbrella, which aims to direct your attention onto an object I have already identified, prior to my use of the

Sortals

77

phrase or the umbrella. The identification of the object came first, and now I am trying, by hook or by crook, to direct your attention to that thing, and it does not really matter how I achieve that effect. So if the clock is not really to the left of the fountain, but I have still managed to direct your attention to it by means of my use of the phrase, then my term still refers and you understand what I have said. In these terms, we can say that one commonplace role for sortals in demonstrative identification is to orient attention to one object rather than another. The sortal really does play a role in the singling out of the object: it is what causes the orientation of attention to one thing rather than another. If I say to you, 'That is very old', you will very often have no idea which thing I am talking about until I supply the sortal, even if I point in roughly the right direction. You use the sortal to orient your attention onto the right object. But since the sortal here is functioning merely to orient your attention onto the right object, it can play its role successfully even if the object is not actually of that sort. That is why the notion of a 'plant' can be playing a role in referring to an object which is not actually a plant; it is 'good enough' in the sense that anyone looking in roughly the right direction will have their attention directed onto the right object by the sortal, even though the thing is not in fact a plant. This lets us see what was right about Wiggins' talk of 'approximating to' correct sortal classification. All we need here is that the approximation must be sufficiently close in this minimal sense: the use of the term must be successful in actually bringing it about that you attend to the right object. To say that the role of the sortal is merely to orient attention towards the right object, though, is also to say that the use of the sortal is dispensable. You could in principle have your attention oriented towards that object by some other cause. This is what happens in the case of the long-preserved teacup. Our descendants manage to orient their attention onto the thing without the use of any sortals at all. This place for the role of sortals does not give them the kind of constitutive role that the Delineation Thesis envisages in making it the case that you are consciously attending to one thing rather than another. It is merely an external causal factor in the act of conscious attention. But it does explain why typically you simply would not understand a demonstrative 'this' which was not, implicitly or explicitly, accompanied by a sortal. This point, about the role of sortals in demonstrative reference, should not be interpreted as a point about the semantic treatment of complex demonstratives in English. We can contrast simple demonstratives, such as 'this' and 'that', with complex demonstratives, such as 'this river' or 'that mountain'. Suppose that the contribution of the nominal 'F' in a phrase of

78

Sortals

the form 'that F' were merely to orient attention to the demonstrated object. Then the truth of a statement of the form, 'That F is G', would not require that the object referred to be F. But it does have to be acknowledged that when you use a demonstrative 'that F' to identify an object which is not in fact F, your statement cannot be regarded as correct (this was the second point we had to address). On the face of it, at any rate, if I point to a sundial and say, That clock is one hour slow', then what I say is not literally true; it implies that there is a clock which is one hour slow, and there may be no such clock. One analysis of the situation is that in addition to orienting attention, the nominal 'F' restricts what the demonstrative can refer to (cf. Kaplan 1989a, 1989b). On the view I am recommending, it would be surprising if ordinary English did work in this way. For demonstrative reference, on the view I am recommending, does not intrinsically require that you get it right about the sort of the object. But it would in principle be possible for English to impose an extrinsic requirement on reference: that when the simple demonstrative is coupled with a nominal, the nominal must apply to the object for the simple demonstrative to refer to it. It may be, though, that the nominal in a complex demonstrative contributes to the truth-conditions of a sentence containing it otherwise than by imposing a condition on the reference of the demonstrative. Strawson 1950 suggested that a sentence of the form, That F is G', says the same thing as a sentence of the form, That is the F which is G'. On this interpretation, the simple demonstrative refers without any conditions on reference being imposed by the nominal, though the nominal may indeed serve the pragmatic function of orienting attention. But the nominal does contribute to the truth-conditions of the sentence. The sentence cannot be true unless the nominal applies to the object to which the simple demonstrative refers. In a recent discussion, Lepore and Ludwig provide an analysis of the same general type, in which we have a self-standing use of a simple demonstrative and a quantificational analysis of the role of the nominal: The key to understanding demonstratives in complex demonstratives is to see the concatenation of a demonstrative with a nominal, as in That F', as itself a form of restricted quantification, namely as equivalent to '[The x: x is that and x is F]'. (Lepore and Ludwig 2000: 229)

On this analysis, the simple demonstrative refers without the nominal playing any role in fixing its reference. Such an analysis nonetheless acknowledges that a statement of the form, That F is G', cannot be true unless the object referred to is F; this is the second point with which we had to deal. And it provides no comfort to a proponent of the Delineation Thesis.

Sortals

79

5. Justification of a Way of Verifying Informative Identities So far I have been considering the Delineation Thesis, which assigns sortal concepts a 'top-down' role in singling out objects in experience. There is, however, another kind of role you might assign to sortal concepts—a 'bottom-up' role. Let me finally try to state what that putative role is. In considering ways in which conscious attention to an object can play a role in causing and justifying the use of particular procedures to verify propositions about the demonstrated object, I have so far concentrated on simple verification of the observable properties of the object, such as its colour or shape. There are also, though, propositions about the identity of the demonstrated object: propositions of the form, 'That F is (identical to). . .', where what fills the space occupied by the dots will be a singular term, such as another demonstrative, or a proper name, or a definite description. Some identity statements—that is, statements of the form' is identical to . . .'—are uninformative. For example, the statement that the Morning Star is identical to the Morning Star is simply an instance of the logical law that everything is identical to itself. In contrast, there are informative identity statements, such as, to use Frege's (1952) example, 'The Morning Star is identical to the Evening Star'. Such a statement cannot be verified simply by appealing to the law that everything is identical to itself. There are informative identity statements involving demonstratives. For example, I might say, 'That star is the Morning Star', pointing into the night sky. Again, establishing that proposition cannot be done simply by appeal to the laws of logic. Or if I return to my room after some time away, I might say, 'That table is the one I brought when I first came here'. Typically, establishing the truth of such an informative identity will involve the use of reasoning and inference, not just observation. Of course, there is the case in which you see the branches of a tree through one window, and other branches through another window. In this case, you might wonder, 'Is this tree (seen through one window) the same as that tree (seen through the other window)?'. And you might verify the informative identity proposition that they are identical, just by putting your head out of one of the windows. So here you might establish the identity just by observation. But often, observation alone will not establish an identity. If you want to establish that 'This man is the lost heir to the Smith millions', the reasoning through which you have to go may be quite demanding. Of course, different patterns of reasoning will be appropriate to different sorts of object. The kind of reasoning that you might use in establishing that 'this river' is identical to 'that river' will involve looking at the course of water through the local landscape, and checking the

80

Sortals

spatiotemporal continuity of the river-bed here with the river-bed there, and looking to see whether there are tributaries flowing into a principal river, to check that you are not identifying a tributary with the river into which it flows. This kind of reasoning is quite different from the kind of argument you would need to establish that this is the watch that was stolen from you, or again, from the kind of argument you would use to establish that this man is the heir to the Smith millions. And even if you are simply looking to see whether or not you have just one thing here, as in the case of the tree seen through the windows, or if you are simply exercising a recognitional capacity, as when you say 'That star is the Morning Star', different types of strategy will be appropriate to different sorts of objects when you are looking to see, or exercising your recognitional capacity. Grasp of these patterns of inference (or strategies of observation or recognition) is a procedural matter. It is a matter of which moves you make: which inferences you regard as valid, which procedures you use and regard as compelling. What justifies you in using such a procedure, or pattern of inference? It does not seem right to say that we can use whatever patterns of inference we like, because the correctness of a particular pattern of inference in a particular case after all depends on just which object you are dealing with. Nor does it seem right to say only that which pattern of inference to use depends only on what type of object you are in fact dealing with. To have the right to regard one rather than another pattern of inference as applicable in a particular case, we need more than simply that the object be of a suitable sort; you must in some sense have registered its relevant characteristics. At this point, it is a natural proposal that what justifies you in using one pattern of inference rather than another when verifying informative identities involving a demonstrated object is your knowledge of what sort of object is in question. This would be a 'bottom-up' use for sortal concepts. As against the Delineation Thesis, someone taking this line might accept that you can single out a particular object without your grasp of sortal concepts coming into play. Your conscious attention can be focused onto the object as a consequence, in part, of your visual system using a particular strategy to bind together information from that object. But once you have singled out the object in experience, you can go on to apply a sortal concept to it. And, you might say, it is your application of the sortal concept to the object that causes and justifies your use of a particular set of procedures to verify informative identities concerning the object. On this view, conscious attention to the object can cause and justify the use of a particular set of procedures to verify informative identities about the object only by causing and justifying your application of the right sortal concept to the object.

Sortals

81

On this picture, experience of the object matters only because it is instrumental in providing a sortal classification of the object, and the sortal classification amounts to an assurance that it is right to use the relevant pattern of inference in establishing informative identities using the demonstrative. You could in principle have had that information without experiencing the object. The trouble with this picture is that all that the sortal concept provides, when all the empirical work has been done and the subject's grasp of the concept made fully articulate, is at best an explicit statement of what the relevant pattern of inference is. It does not justify the use of the relevant pattern of inference, it simply says what it is. You might object that grasp of the sort of an object, when that grasp is made fully articulate and informed by all the relevant empirical discoveries, will involve spelling out just what the criterion of identity is for the object. So, for example, when you grasp that what you are referring to is a river rather than a star or a flower, spelling out the sort involves spelling out which pattern of reasoning is appropriate here. But this does not give a justification for the use of one pattern of inference rather than another; it is merely the explicit statement that one pattern of reasoning rather than another is appropriate. We might draw a comparison here with the justification of logical laws. Your grasp of the pattern of use of a classical logical constant is a procedural matter: it is a matter of which inferences you are prepared to engage in and regard as compelling. But a mere restatement of the laws themselves does not constitute a justification for your use of those laws; at best, it merely makes explicit the practice in which you are engaging anyway. We are not here being offered a justification for the use of the pattern of inference; we are simply accepting a piece of information to the effect that it is in order to use this pattern of inference, without having any understanding of why it is in order to do so. What supplies the understanding of why one pattern of inference rather than another is correct is rather, I suggest, your singling out of the object in experience. We should think of experience of objects as being what provides the justification for the patterns of inference that we ordinarily use in verifying informative identities. On this alternative picture, experience of the object is a common cause of both (a) your capacity to engage in the relevant pattern of inference, and (b) your capacity to provide a sortal classification of the object. And you might argue that experience of the object offers a common justification for (a) your use of the relevant pattern of inference, and (b) your application of the sortal concept. So we ought to be looking at the possibility that it is your experience of the demonstrated object that justifies your use of one pattern of inference rather than another in establishing identities involving that object. The capacity for sortal classification is merely an epiphenomenon. Different patterns of reasoning

82

Sortals

will be appropriate when establishing informative identities involving objects of different sorts. But so far, the sortal concept may play no role in justifying the use of one pattern of inference rather than another. The connection with the object that justifies the subject in using the relevant pattern of reasoning will also justify the subject in applying the relevant sortal concept. But this is a matter of there being a common cause for both (a) use of the right pattern of reasoning and (b) application of the sortal concept. This is not a picture on which the use of the sortal concept plays any role in justifying the use of a particular pattern of inference. At most, so far, you could regard the sortal concept as merely a tag indicating that one type of reasoning rather than another has to be used. In what sense does experience of, say, a dog justify the use of one type of inference rather than another to establish informative identities concerning that dog? If you have the thing in full view, you can keep track of it by keeping track of where it is, and you can put together different properties of the object as properties of the same object by using spatiotemporal continuity as your guide. Your visual system can use the spatiotemporal continuity of the object in keeping track of it. And your visual system can use the co-location of the properties you perceive in binding them together as properties of the same object. Of course there are many varieties of spatiotemporal continuity, appropriate variously to dogs, stars, and rivers. But the point is that the visual system can be using the appropriate type of spatiotemporal continuity in keeping track of the thing, whichever it is, from moment to moment, and in binding together the properties of the object as all properties of a single thing. There is no evident reason why there should be any involvement of the subject's grasp of sortal concepts in this; we are dealing with a more primitive phenomenon than sortal concepts, relating to the underlying visual system. Nonetheless, the type of procedure used by the visual system in organizing information as all relating to an individual object may provide the foundation for the patterns of reasoning we use in establishing informative identities involving the thing. The specific type of spatiotemporal continuity that is relevant is already in play in your capacity to see and keep track of the thing in the first place, so that you are in a position to understand a demonstrative referring to the object. We do have to contrast different types of experience of the thing. Suppose, for example, that late at night you are woken by the barking of a dog.' Is that dog injured?' you wonder, listening to the sound. You can certainly refer to the dog, on the basis of hearing alone, and form judgements or conjectures about it. But it does not seem that your auditory experience of the dog could be what directly causes and justifies your use of the general patterns of inference appropriate to the thing. For example, you might

Sortals

83

use patterns of inference which exploit spatiotemporal continuity as a criterion of identity for the thing, to establish that this dog is the one which your neighbour has just bought. But your auditory experience alone does not display the dog as a spatiotemporal continuant. As I said, it does seem to be very different in the case of a visual demonstrative, 'that dog', referring to a dog you currently have in view. Here we do seem to be directly confronted with the thing itself. And direct confrontation with the thing itself, as a spatiotemporal continuant for which particular strategies of keeping track are appropriate, does seem to be capable of providing a justification for your use of the demonstrative. This, I think, explains the dependence we remarked, in section 2 of this chapter, of the auditory demonstrative on the visual demonstrative. Of course, this distinction between ways of being presented with the object is simply invisible on a sortalist picture, on which the role of experience of the thing is merely to provide you with a sortal classification of it; for sortal classification of the object could be supplied just as well by hearing as by vision. But as I have said, we ought anyway to reject the sortalist picture. Knowledge of the reference of a demonstrative is provided by conscious attention to the relevant object. Our grasp of the identity conditions of an object over time, or the boundaries of the object at a time, is grounded not in grasp of sortal concepts, but in the style of conscious attention that we pay to the thing. And conscious attention to the object does not have to be focused by a grasp of sortal concepts; the various styles of conscious attention of which we are capable do not rely on our use of sortal concepts. Grasp of sortal concepts is a more sophisticated matter than the phenomena of reference and conscious attention.

5

Sense /. Informative Identities Frege (1952) introduces the notion of sense in terms of the informativeness of identities. The identity statement, 'The Morning Star is the Morning Star', is uninformative; the identity, 'The Morning Star is the Evening Star', is informative. The difference between the two identity statements has to be explained by a difference in the senses of the two singular terms. In these terms, the problem of the sense of a perceptual demonstrative arises because it is possible to have two demonstratives that refer to the same thing in different ways, so that the identity statement which uses the two demonstratives is informative. For example, suppose that you and I are rewiring the electrical supply to a house. We have a number of wires spilling out of a panel on the wall, and we also have a number of wires welling up from beneath the floorboards. We have to match up the wires from the wall with the wires from the floor. If I say, 'This wire [tugging at one from the wall] is the same as this wire [tugging again at that very same wire from the wall]', then 1 have simply expressed the law of identity and my statement is uninformative. If, however, I say, 'This wire [tugging at one from the wall] is the same as this wire [pointing at one from the floor]', then I have expressed an important truth; life or death may depend on it. In this case, the two demonstratives refer to the same thing. But there is evidently a difference between them. Frege calls this a difference in sense. And really that is all Frege gives us by way of a characterization of sense. Sense is that, sameness of which makes an identity uninformative, difference in which makes an identity informative. This emphasis on the notion of informativeness is unfortunate, because it has sometimes led commentators to think it telling to ask the question, 'informative to whom?' After all, you might say, if I already know that the Morning Star is the Evening Star, then it will not be informative to me to be told that the Morning Star is the Evening Star; whereas if I did not know that already, the remark will be informative. This can make it seem that Frege's criterion makes sameness or difference of sense depend on the idiosyncrasies of your prior state of knowledge. I think the right reaction

Sense

85

to this comment is to put Frege's criterion directly in terms of what is, and what is not, an instance of a logical law. To be 'uninformative', in this sense, is to be an instance of a logical law. In these terms, then, the point is that 'The Morning Star is the Morning Star' is an instance of the law of identity, whereas The Morning Star is the Evening Star' is not an instance of the law of identity. And no matter what my prior state of knowledge, even if I do already know that the Morning Star is the Evening Star, the remark that the Morning Star is the Evening Star will still be a substantial piece of information, in the sense that it is not merely an instance of the law of identity. We could, indeed, put the point in terms not of which truths are logical truths, but rather, which inferences are valid inferences. In these terms, the point is that the transition, 'The Morning Star is 5 million years old; the Morning Star is a planet; hence, some planet is 5 million years old', is valid as it stands, whereas the transition, 'The Morning Star is 5 million years old; the Evening Star is a planet; hence, some planet is 5 million years old', is not valid. This second argument is enthymematic, and has to be completed by filling in the missing premise, 'The Morning Star is the Evening Star'. Similarly, consider the inference, 'This wire [tugging at the one on the wall] has been disconnected; any wire that has been disconnected is safe to touch; hence, this wire [tugging at the one on the wall] is safe to touch'. This inference is evidently valid; contrast the transition, 'This wire [tugging at the one on the wall] has been disconnected; any wire that has been disconnected is safe to touch; hence this wire [pointing at one from the floor] is safe to touch'. This second inference is not valid; anyone who thinks otherwise is in for a shock. The second inference needs to have filled in the missing premise, 'This wire [tugging at the one on the wall] is this wire [pointing to the one from the floor]'. 2. Sense and Use The points I have made so far do not exhaust the notion of sense. They do not fully bring out the connection between sense and reference, on a Classical View. It does follow from the characterization I have given so far that sense must determine reference, in that sameness of sense guarantees sameness of reference. For suppose we had sameness of sense of two terms, but not sameness of reference. Then the identity statement connecting the two terms would be uninformative, or a logical truth, because of the sameness of sense, yet false, because of the difference in reference. And that hardly seems possible. Frege goes still further than this, though, in connecting sense and reference. He says that your knowledge of the reference

86

Sense

of the term is provided by your grasp of the sense of the term. And on the Classical View, knowledge of the reference of a term is what causes, and justifies, your use of particular ways of verifying, and finding the implications of, propositions involving the term. The natural question is now whether the way in which you have knowledge of the reference of the term will affect the use that you make of the term. That is, does it matter what sense you grasp the term as having, for which ways of using the term you will be caused and justified in employing? This approach applies to singular terms. Recall the descriptive name 'Elmo', for which the introduction and elimination rules were: Exactly one tree in this garden is oldest

Any tree in this garden which is oldest is F Elmo is F

and: Elmo is F exactly one tree in this garden is oldest Elmo is F any tree in this garden which is oldest is F On the Classical Model, use of these rules is caused and justified by your knowledge of the axiom: 'Elmo' refers to the oldest tree in this garden. Suppose now that there is exactly one palm tree in the garden. And suppose we have the descriptive name, 'Woody', for which the axiom is: 'Woody' refers to the palm tree in this garden. Then the rules for 'Woody' are: There is exactly one palm tree in this garden

Any palm tree in this garden is G

Woody is G

and: Woody is G There is exactly one palm tree in this garden And suppose, moreover, that:

Woody is G Any palm tree in this garden is G

Sense

87

Elmo is Woody —that the palm tree is the oldest tree in the garden. Then the identity, 'Elmo is Woody', is an informative identity in Frege's sense. The identity is not merely an instance of the law of identity. And though the two terms have the same reference, they do not have the same introduction and elimination rules. The rules for 'Elmo' and for 'Woody' are, as you can see, quite different. Again, we reach the conclusion that grasp of sense is what causes, and justifies, your use of a particular pattern of use of the term—a particular set of procedures for justifying and finding the implications of propositions involving the term. At this point we can see something of the constraints on an account of the sense of a demonstrative. The most immediate remark is that it should now be apparent how incomplete it would be to regard an account of demonstrative sense as having to do only with the classification of identity statements as uninformative or informative. Rather, having characterized the introduction and elimination rules for a demonstrative—the pattern of use of the demonstrative—we can say that grasp of the sense of the term has to be what causes, and justifies, your use of just these input and output rules. It justifies your use of these rules, in the sense that it defines the objective at which you are aiming in using these procedures, and allows us to display these procedures as ways of achieving that objective. And grasp of sense has to be the cause of your use of these procedures. It is not just a coincidence that you have the objective you do, and that suitable procedures swing into play for you to achieve it. To say that it would be incomplete to regard an account of sense as a way only of classifying identity statements as uninformative or informative is not to say that we can eliminate the notion of informativeness. It is still true that we can have different ways of referring demonstratively to one and the same object, as in the example of wires with which I began. It is just that these different ways of referring to the thing will cause and justify different procedures for verification and action. But when we have one procedure and when we have two will itself have to be understood in terms of informativeness. The different procedures will themselves ultimately be individuated in terms of informativeness. I have been arguing that what provides your knowledge of the reference of a demonstrative is your conscious attention to the object demonstrated. So what we need, in an account of sense here, is a characterization of ways of consciously attending to an object that can cause and justify the use of particular input-output patterns. So different ways of consciously attending to one and the same wire can validate different verification-action

88

Sense

patterns. This is exactly analogous to the point about 'Elmo' and 'Woody', that different ways of grasping one and the same reference can validate the use of quite different introduction and elimination rules. Knowledge of the reference of a demonstrative, provided by conscious attention to the object, causes and justifies the use of information-processing procedures to verify, and to find the implications for action of, propositions involving the demonstrative. So the issue of sense, as it applies to demonstratives, has to do with the relation between conscious experience and informationprocessing. What we have just found is that different ways of consciously attending to a single object can cause and justify the use of different informationprocessing procedures to verify, and to find the implications for action of, different demonstrative propositions involving reference to the same object. To characterize the sense of a demonstrative, then, we need to know how to characterize the content of conscious attention to the object, so that we can explain how one way of consciously attending to the thing can cause and justify: (1) use of one particular set of information-processing procedures to verify a proposition involving the demonstrative, and (2) use of one particular set of information-processing procedures to act on the basis of a proposition involving the demonstrative. What we have to do here, then, is to find a way of characterizing the content of your conscious attention to the object, so that we can see how the content of your conscious attention systematically causally affects which information-processing procedures swing into play in verification and in action. In Chapter 2,1 set out the relation between conscious attention to an object and the use of particular information-processing procedures to verify demonstrative propositions about the object. I described two types of verification. First, there was verification involving bringing to bear some high-level information-processing, such as the visual routines described by Ullman (1996). This is what you have to do to verify visually propositions to the effect that a particular object is enclosed by a boundary, for example. Secondly, there was verification that the object has a particular simple property, such as redness or roundness, where these properties are determined in establishing the initial feature maps used to find that there is an object there at all. In both cases, I said, it is your conscious attention to the object that causes, and justifies, the use of particular information-processing procedures to verify the proposition that the thing

Sense

89

has that property. But we can ask about the details of the causation here. Just what aspect of your conscious attention to the thing is operative in determining how the information-processing swings into play? Location is a key factor from the point of view of the informationprocessing mechanisms that have to swing into play. If a visual routine is to be brought to bear on an object, the first thing that has to be found is where the object is. If a feature map is to be consulted to find which property an object has, the first thing that has to be determined is the location of the object. In Chapter 3,1 argued for a parallel picture of the relation between conscious attention to an object and the use of particular informationprocessing procedures to act on the object. I said that conscious attention to the object again causes, and justifies, the use of these informationprocessing procedures. Which actions you perform must, in general, be determined by the direction of your conscious attention, if action is to be voluntary at all. And again, we can ask which aspects of your conscious attention to the object are causally operative in determining which information-processing procedures are to swing into play. Again, as we saw, the experienced location of the object is likely to play a key role, as that can identify the target of processing for the visuomotor system. The Gestalt organization of the visual field, when you are consciously attending to the object, must also play a role, since different objects can be close together, and there must be a way in which the subject can decide to verify visually a proposition about one rather than another of those objects, and it must be possible for the subject to decide to act on one rather than another of those objects. As I stressed in Chapter 3, it is not that the experience of the object directly sets the parameters for your action on it, but rather that this can identify the target for visuomotor processing. Similarly, the experienced Gestalt organization of the target can identify the target for processing to verify a proposition about it. The experienced location of the object may not involve the very same frame of reference for identifying locations as is used in cognitive processing. Nonetheless, it is the lead candidate for systematically affecting which location is selected at the level of cognitive processing, since it will have the same structure, a structure which can sustain systematic causal connections. Similarly, it is not quite immediate that the Gestalt organization of your experience of the object is just the same as the Gestalt organization used in cognitive processing; but the Gestalt organization of your experience is, again, the lead candidate when we seek a structure in the content of experience, so there could be systematic causal connections between it and the Gestalt organization used in cognitive processing. All this suggests that we can formulate the introduction and elimination rules for a visual demonstrative somewhat as follows. Suppose we have a

90

Sense

visual demonstrative, 'that wire', of whose reference you have knowledge by consciously attending to the object. The 'address' of that object will be given by the complex parameter used in binding together all the information related to that object. This parameter will typically, for a visual demonstrative, include the current location of the object, though it may involve some higher-order linkage between locations, in the case of an object of which you are keeping track as it moves over time. The parameter will also typically include the Gestalt organization of the object, though it may again involve some higher-order linkage between different styles of Gestalt organization, as the object changes over time (without changing out of recognition). Since conscious attention has to identify the object for the benefit of the information-processing system, there must be commensurability between the way in which the object is identified at the level of conscious attention and the way in which the object is identified in the information-processing subsystems. For present purposes, to give a reasonably uncomplicated exposition, I will suppose that commensurability is achieved in the simplest way, by having both conscious attention and the information-processing subsystems use the same complex parameter, p. Then we can formulate the introduction rule like this: FEATURE MAP: F-ness at p That wire is F The elimination rule will be: That wire is F MOTOR SYSTEM: To act with respect to an F wire, activate the visuomotor system to act with respect to the target identified by p. The cause, and justification, of your use of these introduction and elimination rules will be provided by the way in which you have knowledge of the reference of the demonstrative, which is supplied by your conscious attention to the object, this being focused by your experience of the object as at p.

3. Immunity to Error through Misidentification The sense of the demonstrative is given by conscious attention, focused by experienced location and Gestalt organization. At this point it may seem that what I am giving is a broadly descriptivist analysis of the meanings of demonstratives, on which the descriptive content of the demonstrative is

Sense

91

provided by the binding parameter. But that would be a mistake. The demonstrative does not have a descriptive content, not even one specified by the binding parameter. But the binding parameter is constitutive of the meaning of the demonstrative. We can see exactly how the meaning of the demonstrative is related to the binding parameter by looking at how this approach deals with a broadly logical point about demonstratives: namely, that judgements about the location of a demonstrated object may be mistaken, but they are nonetheless, in Sydney Shoemaker's (1984c) phrase, immune to error through misidentification. Suppose as I stand here I hear a voice, and I look round and think, 'That woman spoke'. There are different kinds of mistake I might make. It might be that no one spoke at all, that there is simply something wrong with my ears. In that case I made a mistake, but the mistake is not one of identification; it has to do with the predicate, whether there was any speaking. In contrast, it might happen that someone spoke all right, but it was not that woman. In that case, the error is an error of identification. I was right about the speaking, but wrong about who it was. We can put it like this: the evidential basis I have for saying, 'That woman spoke', is of such a kind that I could have a ground for doubt about whether it was that woman who spoke, which did not undermine my right to claim to know, on that same evidential basis, that someone spoke. In contrast, consider a judgement about the location of a demonstrated object: suppose we have a judgement like, 'That woman is there', in which I simply identify the place the person is at. A judgement of this kind can be mistaken. For example, if there is a mirror or a prism in between us, as could happen, then I might be quite wrong about where the person is. That is not a mistake of identification; it is a mistake about the predicate I am applying to the person. For there to be an error of identification, it would have to be that I make a judgement, That woman is there', and I could be rationally brought to doubt that it is that woman who is there, but still keep my right to think, 'Well, at any rate, someone is there'. And that seems flatly impossible. (Bear in mind that I am thinking of a case in which you are identifying the person only on the basis of your current perception, not as someone you recognize or whatever.) Intuitively, you are using the perceived location of the person to single out which person you have in mind. That perception of location might be illusory. But a mistake of identification would require you to be having a veridical perception of the location of some other thing, and mistakenly attributing that location to the person you are referring to. That is what seems blankly impossible. You are not simultaneously perceiving two different people to be at the same place. So you would have to be using the veridical perceived location of one thing to visually single out something

92

Sense

else, which is not at that location. And that does not seem to be a coherent possibility. So a mistake in thinking that the thing is at this particular place could not leave intact your claim to know that there is something or other at that place. The problem now is to say more explicitly just what the explanation is of this immunity to error through misidentification of judgements about the locations of demonstrated objects. There is an easy way to generate judgements which are immune to error through misidentification. Suppose we introduce a descriptive name, 'Frank', which has its reference fixed by 'the inventor of the postmark'. And suppose you have good reason to think that there was such a thing as the inventor of the postmark; it was not the product of a committee or an accidental artefact of the mailing system. Then, exploiting that background belief, and your understanding of the name, you form the judgement, 'Frank was a sole inventor of the postmark'. You are fallible about this. For there may after all have been no inventor of the postmark, despite your evidence to the contrary. But there is a kind of mistake you cannot have made. It cannot be that you are right about there having been a sole inventor of the postmark, but that you are just wrong about which person it was. It cannot be that you do know that there was a sole inventor of the postmark, but that you have made a mistake in thinking that it was Frank rather than somebody else. If you are right in thinking that there was a sole inventor of the postmark, then you cannot but be right in thinking that it was Frank who did it. To put it round the other way, if you are wrong about whether Frank invented the postmark, your mistake cannot be localized as an error of identification about who it was that invented the postmark. If you are wrong about whether Frank invented the postmark, you must also be wrong about whether there was a sole inventor of the postmark. This suggests a connection between immunity to error through misidentification, and the way in which the reference of a singular term is fixed. It suggests that the way in which it happens that there are judgements which are immune to error through misidentification, is that there are descriptive conditions on the reference of the singular term. So when the subject uses his grasp of the singular term to articulate a judgement in which the descriptive conditions are said to apply to the referent of the singular term, the result will be a judgement that cannot involve an error of identification. The point about the judgement, 'Sally spoke', is then that in an ordinary case, you are not using the fact of her speaking as a descriptive condition by which to fix the reference of 'Sally'; that is why your judgement 'Sally spoke' is subject to errors of identification. If we have a term whose reference is fixed by a descriptive condition, this will have the consequence that judgements applying that description to the

Sense

93

object referred to will be immune to error through misidentification. So you might propose that we can turn this procedure around. Whenever we find judgements that are immune to error through misidentification, we can conclude that we are dealing with a term whose reference is fixed by a descriptive condition, and find what the descriptive condition is by looking at the specific contents of the judgements which are immune to error through misidentification. As we saw, judgements about the locations of demonstrated objects seem to be immune to error through misidentification. Suppose you are looking at a number of technical instruments, lying on a workbench, each of which periodically emits its own characteristic noise. When you make a judgement about the sound being made by a particular instrument, there are two kinds of mistake you could make. Suppose you think, 'That meter is humming'. One kind of mistake is that there is no humming going on at all, there is just something wrong with your ears. Another possibility is that something is humming, but that thing is not the meter. So you have made a mistake of identification: you are right about whether something is humming, but you have just made a mistake about which thing it is. In contrast, it does not seem that judgements about the locations of these instruments could be similarly subject to errors of identification. A judgement of location can involve a mistake. For example, if there is an unsuspected mirror or prism in between you and the instrument, you might see the instrument perfectly well but make a mistake about where it is. That kind of mistake is not an error of identification. To be making an error of identification, you would have to be getting it right about whether something is at that location, only that thing is something other than the instrument you are identifying. This is what seems impossible. On the approach that works for descriptive names, this would suggest that the reference of the demonstrative, 'that instrument', is fixed by some such descriptive condition as 'the instrument at this location'. But we have already run into a complication. It does not seem that the descriptive model can be quite right here. For on the descriptive model, it should be a priori that 'That instrument is at that location (if any instrument is)'; just as it is a priori that 'Frank invented the postmark (if anyone did)'. But as I already remarked, it seems entirely possible that you should make a mistake about the location of a perceived object, because of mirrors or prisms or whatever. The point is only that it will not be a mistake of identification. What, then, is the explanation of immunity to error through misidentification for the case of demonstrative judgements about the location of an object? I will argue that the phenomenon here does have to be explained by appeal to the meaning of the demonstrative. But that is not because there is any descriptive element in the meaning of the demonstrative. Rather, it

94

Sense

has to do with the way in which the meaning of the demonstrative depends on the exercise of perceptual attention. On the analysis I have presented, it is your conscious attention to the object that provides your knowledge of the reference of the demonstrative. So your experience of the object has to provide enough information about it to identify the target of the information-processing involved in verifying and acting on propositions involving the demonstrative. What I have argued is that you use the experienced location of the object in identifying the target of this informationprocessing. The reason that experienced location is crucial here, I argued, is the role of location in solving the Binding Problem for the demonstrated object. Your use of a demonstrative, such as 'that person', depends on your having solved the Binding Problem. It is the fact that you are using the demonstrative 'that person' on the basis of your having bundled together all the information from location p as true of a single thing, that grounds your use of the demonstrative. It is because the solution to the Binding Problem gave a special place to location that judgements ascribing a location to the demonstrated object are immune to error through misidentification To see this, suppose we ask how it can happen that you might make an error of identification in using a demonstrative term. Suppose we stay with the case in which you make your judgement purely on the basis of vision, and that vision is using spatial location as its principle of binding. Then what has to happen for there to be an error of identification is that you do visually detect the presence of a certain attribute, so that you have the right to say, 'Something is F', even though there may be grounds for doubt as whether you are right in judging, 'That thing is F'. For you to have made such a mistake, you must have bound the attribute F-ness together, mistakenly, with a collection of attributes belonging to a different object than the object which is F. Since you are using spatial location as your principle of binding, what this means is that you have assigned F the wrong spatial location, in that it has been put together with a collection of attributes that do not belong to an object which is F. You are right in thinking that there is F-ness around, but you have mislocated it, which is how you have made your mistake of identification. But this can hardly happen with your ascription of location itself. Visual location is the principle which you are using to bind together a collection of features as features of a single object. So it does not make sense to suppose that you might have assigned the location to the wrong bundle of features. It is visual locations that individuate the bundles of features, so it does not make sense to suppose that the right location has crept into the wrong bundle. This shows a contrast between demonstratives and descriptions. An

Sense

95

attention-based account on which location is the principle of selection does not regard a demonstrative like 'that person' as equivalent to a description of the form 'the person at location x'. If you descriptively identify an object by means of its location, then there is no possibility of getting it wrong about where it is; a mistake about location can only mean that you have failed to identify anything at all. You can use attention to a location to bind together the features of a single object even though it is not where it seems to be. Suppose, for example, that you see an object—say, a person— through a prism without realizing that there is a prism there. You can use the apparent location of the features to select them all, binding them together as features of a single thing, even though the object is not at that place. When you use the demonstrative, 'that person', it does refer to the object. But your judgement, 'That person is at location x', will be mistaken. Still, the judgement of location is immune to error through misidentification. You cannot have a doubt about whether the demonstrated object really is at that location, which would leave intact your evidence that, at any rate, something is at that location. There are various models for the case of demonstratives; one we considered earlier is the case of ordinary proper names. Consider the way in which testimony uses names such as 'Ronald Reagan'. Here we have an individual who is, as it were, radiating information about himself into the community, information transmitted by testimony. Suppose we ask, 'What is it, for the ordinary hearer, that bundles all this information together, as all true of a single individual?' One answer would be that on each occasion on which the name is used, the hearer must assure himself by collateral information that it is the same thing that is being talked about. But this would evidently be false to the role that proper names have for us. It is rather the sameness of the name itself that assures us that it is the same thing that is being talked about. Of course, a single name may have different uses, so on occasion, a gloss has to be supplied to indicate which Ronald Reagan is in question. So, for example, you might gloss the name with, 'the former actor who was American President in the early eighties'. This problem of organizing information disseminated by testimony into usable clusters—clusters each of which relates to just one object—is, as we saw, in some ways analogous to the Binding Problem in vision. When is a judgement involving an ordinary proper name immune to error through misidentification? Suppose, for example, that you make the judgement, 'Ronald Reagan is the former actor who was American President in the early eighties'. Could your judgement involve a mistake of identification? Mistakes of identification certainly can in principle be made using ordinary proper names. For example, you might hear a news bulletin announcing that someone has been awarded a Nobel Prize in Economics, and

96

Sense

mistakenly think that it is Reagan who is being referred to. In that case, someone might provide you with grounds for doubt as to whether it was Reagan who is in question, without undermining your knowledge that someone has been awarded the Prize in Economics. Or again, at the start of the eighties, you might hear a broadcast announcing that Reagan has been elected President, and someone could challenge you at that point, accepting your evidence that a President had been elected but questioning your grounds for supposing it to be Reagan. Suppose, however, that we consider the position of an ordinarily well-informed speaker now, in the twenty-first century. Such a speaker uses some such gloss as, 'the actor who became American President in the early eighties', as his way of checking that the information he is getting concerns that Ronald Reagan; it provides his bundling principle, analogously to the use that the visual system makes of location. Suppose this speaker says, 'Reagan is the former actor who became American President in the early eighties'. This judgement is fallible. It takes a bit of an effort, but you can imagine being given strong evidence that Reagan was after all never elected President, or that he after all never was an actor. But can you envisage being given proof that though someone who was an actor was elected President in the early eighties, and though you do have knowledge of that existential proposition, still that person was not Reagan? You can imagine discovering that your whole picture of the early eighties is a complete hallucination, and that coincidentally there was a former actor who became President. But what is not conceivable is that you have knowledge that there was a former actor who became President in the early eighties, and that you are mistaken only in thinking that person was Reagan. The key point is to see that acknowledging this point does not commit you to thinking that 'Reagan' is a descriptive name, just because it is entirely possible that Reagan exists yet was neither an actor nor a President, though he was often referred to as such. The general point here is this. In the case of visual demonstratives, I have talked about binding principles, such as location. And in the case of proper names, I have talked about bundling principles, such as the gloss, 'the former actor who became President'. Such a binding or bundling principle does not deliver a definite description equivalent in meaning to the proper name. This comes out in a number of ways. Most dramatically, the binding or bundling principle is just a way of collecting together a cluster of information as all true of a single thing. For just that reason, it does not need to yield a definite description which actually is true of the object the name refers to. Nonetheless, when you articulate the judgement which ascribes the property used as your binding or bundling principle to the object referred to, the result will be a judgement which, though fallible, is immune to error through misidentification.

Sense

97

4. Trading on Identity I have been arguing that we have to think of knowledge of the reference of a demonstrative as provided by conscious attention to the object in question. For conscious singling-out of the object to be what explains your knowledge of reference, it cannot itself be a matter of demonstrative reference to the thing. Otherwise, conscious attention would presuppose what it explains. If visual experience of the object was simply one way among many of having a demonstrative thought about the object, there would be no saying how visual attention to the object could be what explained your knowledge of reference. There would be no primacy for perception in the understanding of a demonstrative; your understanding of the demonstrative could not be represented as depending on your perception of the object. To understand why perception of the object may be essential to your understanding of the demonstrative, we have to take visual attention to the object to be more primitive than demonstrative identification of the thing. Nonetheless, conscious attention to the object has a certain content, as we have just seen. We experience objects as having certain locations, and it is the fact that you experience the object as having a certain location that allows you to identify the object for the benefit of the underlying systems that have to swing into play to verify or act on the basis of propositions about the thing. Moreover, it is the fact that you are using the binding parameter for the object as the way of identifying it that explains the fact that judgements about the location of the object are immune to error through misidentification, though they can be simply mistaken. Suppose we return to the problem with which this chapter began, namely, to explain why some identities are informative, or, as I suggested we reformulate the question, to explain why some inferences involving identity are valid and others are not. The example I gave earlier of a valid inference involving trading on identity was this: This wire [tugging at the one on the wall] has been disconnected; Any wire that has been disconnected is safe to touch; Hence, this wire [tugging at the one on the wall] is safe to touch. This inference is evidently valid; contrast the transition: This wire [tugging at the one on the wall] has been disconnected; Any wire that has been disconnected is safe to touch; Hence, this wire [pointing at one from the floor] is safe to touch. This second inference is not valid; it needs to have filled in the missing

98

Sense

premise, 'This wire [tugging at the one on the wall] is this wire [pointing to the one from the floor]'. Some inferences involving a trading on identity are valid; some are not. In the case in which trading on identity is legitimate, your knowledge of the reference of the demonstrative should cause, and justify, your making the inferential step. So what, in general, can we say about knowledge of reference, in virtue of which it has that character? My argument has been that it is experience of the object that provides us with knowledge of reference. So can we say how it is that experience of the object causes, and justifies, trading on identity in some cases but not in others? We have seen that the way in which you experience the object can be seen as identifying the object by using the complex binding parameter by which the visual system put together the information from that object as all true of a single thing. So can we say that trading on identity is legitimate when the occurrences of the demonstrative are each being interpreted by an act of conscious attention for which the same complex binding parameter is being used? In the present context, that seems the simplest hypothesis. I think this hypothesis is basically correct, but it does need to be glossed a little. The problem emerges when we consider a case in which I am watching you manipulate, say, a matchbox. In line with the kinds of input move licensed by my grasp of the demonstrative, 'that matchbox', I might think, for example, 'that matchbox now contains my watch', having seen the watch in the thing. Unknown to me, though, you are actually manipulating two matchboxes, not one, even though it looks to me as though there is just one matchbox there. Then, as I see you filling with matches what looks like one and the same matchbox, I may think, 'That matchbox was empty a moment ago'. And I may be using the same complex binding parameter over the period. So, by our present hypotheses, I will be licensed in trading on identity to conclude that one and the same matchbox at one and the same time contained my watch and was empty. But something has gone wrong somewhere. I think that the basic point here is that there can be mistakes at the level of binding, and different degrees of error are possible. Suppose you are seated at a red traffic light, and for a split second it seems to you that the light has turned to green; but in fact this is an illusion: one of the traffic lights facing another set of traffic has turned green, and the mistake was to bind that greenness together with the rest of the perceptual information about the light facing you. In that case, it seems evident that you saw the light in front of you but just made a mistake about whether it had changed colour. (This is an error of identification.) In this case, there has been amistake at the level of binding, but not one that affects your ability to be think-

Sense

99

ing demonstratively about one particular object, one particular set of traffic lights. What concerns me now, though, are more radical cases, in which the mistakes in binding mean that there is no saying which object you are perceiving. Suppose, for example, that you are rapidly presented with, in succession, a pink X and a yellow T, but perceive a pink T (Treisman and Schmidt 1982). Did you perceive the T and make a mistake about its colour, or did you perceive the X and make a mistake about its shape? It seems quite obvious that there is no answer to such a question. This is a kind of limiting case of the matchbox example. If I say, 'That figure is pink' and 'That figure is a T', and trade on identity, I will get the conclusion that one and the same thing is both pink and a T, even though there is no pink T in the display. But where did I go wrong? It is not enough, to have a way of thinking demonstratively of an object, that you be consciously attending using a particular complex binding parameter. For there to be an object you are identifying, it must be that the bulk, the overwhelming majority, of the perceptual information that you are binding together does indeed all causally derive from just one object. Otherwise, if there are, for example, two different objects in play, as in the case of the matchboxes or the pink T, there is no saying which one you are identifying demonstratively. So perhaps we could put it like this: in order for there to be a way of identifying an object in place here at all, there must be just one object from which the relevant visual information causally derives. Given that this external condition is met, we can say that trading on identity will be legitimate if the uses of the demonstrative are interpreted by an act of conscious attention in which we have the very same complex binding parameter being used. There is, though, just one further line of thought we have to add here. It is not enough that there be a single object from which most of the visual information causally derives. It must be, further, that the subject does not believe there to be a number of objects from which the relevant information derives. Suppose we go back to the case of the matchboxes one more time. But this time, suppose that what is happening is that I am watching you at work with the matchbox, and there really is only one matchbox there, so things are just as they seem. But suppose that I suspect you of trickery, and believe there to be two matchboxes in play. Then it seems evident that, by my own lights, I cannot verify propositions about 'that matchbox', and trade on identity in inferences involving the demonstrative, in the usual way. For, by my own lights, the upshot would be that I was putting together information about different matchboxes as if it all related to a single matchbox. So for the trading on identity to be legitimate, it must be

100

Sense

not just that there is a single thing from which all the information causally derives, and that the information is all bound together by a single complex binding parameter. It must also be that I do not disbelieve that this is what is happening; that I do not believe that I am getting information about two different things and that my visual system is being fooled. We do not want to go so far as to demand that 1 should know that there is just one object here, as opposed to a pair of skilfully operated lookalikes; in most ordinary cases, most of us would not be able to rule out the hypothesis that there are many objects in play. Even if we believe there is only one thing in play, that is not because we can rule out a patiently constructed sceptical hypothesis on which there are two objects here being moved in some complex way that results in the illusion of a single object. But it is not, either, as though experience of the object is neutral on this point. We would ordinarily take it that our experience of the object is enough to allow us to interpret the demonstrative—to know which thing is being referred to— and, consequently, to recognize the correctness of inferences involving the demonstrative which trade on identity. This means that your ordinary experience of an object cannot be regarded as being neutral on the question whether there is one thing or two things in play. Just how we are, in consequence, to analyse the experience of the object involved in understanding the demonstrative is a matter to which I will return in Chapter 7. For the meantime, we can say that in an ordinary case, in which you experience the object—the object exists—and your visual information is causally deriving from just one thing, then you can trade on identity whenever you encounter a demonstrative which you interpret by an act of attention in which the very same complex binding parameter is used. These points, incidentally, all have their analogues in the case of proper names. Suppose we go back one last time to the name, 'Ronald Reagan'. You might think it inconceivable that an actor could ever have become President; you suspect that there are two or more different people here, the actor and the politician, for instance. In that case, you will mistakenly think that the information that is bundled together in gossip and testimony by the bundling principle, 'Reagan, the actor who became President in the early eighties', is actually a bundling together of information from at least two different individuals. So you will not be able to trade on identity in the usual way; indeed, you will not think that the name 'Reagan', as ordinarily used, serves to pick out just one rather than the other of those individuals (cf. Kripke 1979). What makes trading on identity legitimate in the ordinary case is that there is just one individual from whom derives the majority of the information bundled together by the tag, 'Reagan, the actor who became President in the early eighties'. Moreover, the ordinary speaker does not believe that there is more that one individual in play here. Whether

Sense

101

a mere lack of belief here is enough, or whether something more is required, is a question I will leave open for the moment. The next three sections are somewhat subsidiary to the main line of this book; the main line resumes at the start of Chapter 6. In section 6 of this chapter, I will look at the relation of the view I am developing here to the account of demonstratives in Kaplan's 1989a and 1989b. In particular, I will look at how the role of location in demonstrative identification can be accommodated within Kaplan's framework. In section 7,1 will contrast the role that we have given to location in demonstrative identification with the role found for it in Evans 1982. Finally, at various points I have mentioned an analogy between the Classical View of knowledge of the reference of a demonstrative, and the Classical View of our understanding of the propositional constants; I will now briefly set out the analogy explicitly. 5. Propositional Constants Perhaps the simplest and most striking illustration of the Classical View is provided by the classical account of propositional logical constants, terms such as 'and', 'or', and so on. The standard way of verifying a proposition of the form 'A&B', for example, is to derive it from A together with B:

A

B

A&B The standard way of finding the implications of a proposition of the form, A&B', is to derive A from it, and to derive B from it: A&B

A&B

A

B

Suppose you ask: what causes us to use the procedures? And is there any justification for our using them? The classical answer is that the cause, and justification, for your use of these procedures is your knowledge of the classical truth-table for '&', which shows how the truth or falsity of any sentence of the form 'A&B' is determined by the truth or falsity of the constituent sentences A, B: A B A&B T T T F FT F F

T F F F

102

Sense

The Classical View is that your grasp of the truth-table is what causes, and justifies, your use of the above ways of verifying, and finding the implications of, propositions of that form. Just to spell out what the truth-table says: whenever you have a complex proposition, such as 'Snow is white and grass is green', you can isolate the simpler statements from which it is built up—in this case, (a) 'Snow is white' and (b) 'Grass is green'. You can then ask what the relation is between the truth or falsity of the whole proposition, and the truth or falsity of the simple statements from which it is composed. The truth-table simply lists all the possible combinations of truth and falsity for the two simpler statements: that they are both true, that one or the other is false, and that they are both false. So there are four lines in all. Evidently, the whole statement, 'Snow is white and grass is green', is true if both constituent statements are themselves true. Otherwise—if either or both of the constituent statements are false rather than true—the whole statement is false. The truth-table simply sets this out. Whenever you have a statement of the form 'A&B', the whole thing is true if both constituents are true, and it is false otherwise. Whether ordinary speakers can in some sense be said to know truthtables for the ordinary logical constants used in everyday speech is a vexed question. (Cf, for example, Dummett 1991.) At the moment I am not aiming to force a decision on this difficult question. The point I am making is rather that the picture on which the use of a concept is caused and justified by your grasp of the concept is given an absolutely exact illustration by the view on which your use of a logical constant—the introduction and elimination rules you use for it—is caused and justified by your grasp of a truthtable for the logical constant. There is, indeed, a notorious crux over the threat of circularity at this point. How in detail is the causing and justifying of those patterns of use by your knowledge of the truth-table supposed to proceed? One easy answer is that the subject can by reflective reasoning establish that these introduction and elimination rules are the right ways to verify and find the implications of statements using a sign, '&', for which the above is the truth-table. You can establish that the rules will be truth-preserving; if the inputs are true, then, you can read off the truth-table, the outputs will be true. Moreover, the introduction rule demands as little as possible, consistently with its being truth-preserving; and the elimination rule allows you to extract as much as possible from the proposition, consistently with the elimination rule's being truth-preserving. The trouble is that the subject engaged in this kind of reflective derivation of the rules from the truthtable will have to be using deductive reasoning in the derivation. (Cf. Quine 1976.)

Sense

103

The sense in which the speaker grasps the justification of the practice need not, though, involve the use of reasoning at all. The point is rather this: which particular introduction and elimination rules the speaker employs must be systematically causally dependent on which truth-table the subject associates with the sign. So the use of inference rules in connection with a logical constant is systematically dependent on which truthtable you associate with the constant. Change one line of the truth-table the subject associates with the sign, and there is a corresponding change in the inference rules the subject uses in connection with the sign. Moreover, this grasp of the truth-table acquaints you with the validation of your practice. You might point out that even on a Classical View, there is a significant contrast between prepositional constants and demonstratives, in that there is a canonical way of identifying the semantic value of a prepositional constant, but no one canonical way of identifying the semantic value of a demonstrative. The canonical way of identifying the semantic value of a prepositional constant is by way of a truth-table, such as the table for '&' given above. In contrast, the referent of a demonstrative can be identified in many different ways, and there is no single way of identifying it that has canonical status. Some such point is certainly correct. But I will not here pause over just how the point is to be spelled out—just what notion of 'canonical' we need here to explain the point, or where exactly its significance lies. For our purposes, the important fact is that there are other, non-canonical ways in which the semantic value of a classical prepositional constant can be given, and the way in which the semantic value of a prepositional constant is given will determine the correct use of the propositional constant; so that the situation here really is, to that extent, parallel with the case of names such as 'Elmo' and 'Woody' which I discussed in section 2 above. The point is that two propositional connectives could have the same semantic value even though it is not a trivial matter that they do have the same semantic value. For example, consider the connectives'+' and '#', for which the tables are as follows: A B

A+B

AB

A#B

T T

T T

T F

T if 53 is prime, F otherwise F

T if 71 is prime, F otherwise F

FT

F

FT

F

F F

F

F F

F

T F

Since 53 and 71 are both prime numbers,'+' and '#' have the same semantic

104

Sense

value, and both have the same semantic value as '&'. They yield the same truth-values for the complex sentences which contain them, given the same truth-values for the constituent sentences. The two tables above provide different modes of presentation of the same semantic value. The reason it matters that these truth-functors have different modes of presentation of semantic value is that, as a result of the difference in mode of presentation, the introduction and elimination rules for the two truth-functors will be different: A

B

53 is prime

A

B

A+B

A#B

A+BA+B A

71 is prime

B

A+B 53 is prime

A#BA#B A

B

A#B 71 is prime

The difference in the modes of presentation of semantic value in the case of'+' and '#' is apparent in the fact that the two specifications of semantic value justify different sets of introduction and elimination rules. So the reason why we need the notion of a 'mode of presentation' of semantic value is to explain why it is correct to use particular sets of procedures in verifying or drawing the implications of propositions. The subject who grasps a particular mode of presentation of semantic value thereby grasps the justification for using a particular set of procedures for verifying or finding the implications of propositions. If this model is right, it confirms that we should think of grasp of sense—grasp of the mode of presentation of the semantic value of a term—as what causes our use of particular procedures to verify or find the implications of a proposition. The pattern of use you make of the term will be causally determined by the way in which you grasp its semantic value. So grasp of sense is what causes, and justifies, your use of a particular pattern of use of the term—a particular set of procedures for justifying and finding the implications of propositions involving the term.

6. Kaplan on Dthat and Demonstrations In 'Demonstratives' (1989b), David Kaplan gave an account of logical consequence for a language with demonstratives which aimed to characterize the difference between pairs of inferences such as those about the wires that I discussed above, in which trading on identity is or is not legitimate. Kaplan aimed to characterize the notion of formal validity for

Sense

105

a language with demonstratives. His analysis appeals to the notion of a 'demonstration'. On this account, trading on identity is legitimate when the two demonstrative terms are associated with the same demonstration; if the associated demonstrations are different, then the trading on identity is not legitimate. But what is a demonstration? Kaplan introduces the notion as follows. He says that some indexical terms require, in order to determine their referents, an associated demonstration: typically, though not invariably, a (visual) presentation of a local object discriminated by a pointing. These indexicals are the true demonstratives, and 'that' is their paradigm. Thedemonstrative(an expression) refers to that which the demonstration demonstrates.... [TJhe referent of a demonstrative depends on an associated demonstration.' (1989b: 491-2)

The problem with this informal introduction of the notion is that it does not distinguish between a demonstration as an instrument of communication, and a demonstration as my way of achieving a visual fix on the object. There is an oscillation in Kaplan's account of demonstratives between viewing 'demonstrations'—the (in general, non-linguistic) complements of a use of a demonstrative term such as 'this' or 'that'—as instruments of communication, and viewing a demonstration as the way in which the individual thinker achieves a perceptual fix on the object. When we are thinking of demonstration as an instrument of communication, it is simply the method by which, by hook or by crook, you get your audience to attend to the right object. One method would be to grasp your hearer's head and turn it firmly towards the object, meanwhile waggling the object itself. On that kind of understanding of the role of the demonstration, it really is a ladder you climb to secure understanding, but you throw it away at the conclusion. The trouble with this is that Kaplan is quite explicit about the fact that the demonstration has a further role: it has to do with the cognitive significance of the way in which the proposition is grasped, and in particular, with how the proposition affects your propensities to action; Kaplan approvingly quotes Perry on this (Kaplan 1989b: 532). So we need some account of just how the demonstration is to impact on action. Moreover, as I said, in his formal analysis of inferences involving demonstratives, Kaplan gives an account on which trading on identity in an inference involving demonstratives is correct just in case the two demonstratives on whose identity we are trading involve the same demonstration. And in giving the account of when trading on identity is legitimate, we are considering the propositions used in the argument as grasped by a single thinker at a time, and the 'instrument of communication' notion of a demonstration is not the relevant notion. Just to illustrate this point,

106

Sense

suppose you and I are watching a play and 1 notice that one of the actors seems about to faint. When I think, 'That man is about to collapse', I am visually discriminating him from his colleagues and the background. In this case, the visual discrimination, however exactly it is accomplished, is the demonstration, in that it determines the referent of the term. However, suppose I want to alert you to this situation, then my own visual discrimination is not quite what is needed to complete my use of the term, 'that man', when I say to you, 'That man is about to keel over'. If you say, 'Which one?', I have to say, 'The one second from the left on stage', The one playing King Lear', or perhaps even point. These communicative instruments succeed just when their upshot is that you visually discriminate the same man as I do. So there are two phenomena here that are covered by the notion of a 'demonstration': there is the communicative instrument, the means by which I focus your attention onto the right object, and there is the visual discrimination, the way in which I achieve my visual fix on the object. Suppose we consider a case in which you are sitting on my left, Bill is sitting on my right, and there are some actors on the stage. You whisper in my left ear, 'That actor (the one wearing a hat) is about to faint', while Bill whispers in my right ear, ' I met that actor (the one playing King Lear) once'. Can I immediately trade on identity, to conclude that, if what I have just been told is right, Bill has met someone who is about to faint? Notice first that the demonstrations associated with the two demonstratives, in the communicative sense of 'demonstration', are certainly different: you demonstrate him as 'the one wearing a hat', whereas Bill demonstrates him as 'the one playing King Lear'. However, the fact that the demonstrations are different does not of itself mean that the inference, as I understand it, cannot legitimately trade on identity. Kaplan suffers over the conflation of different notions of 'demonstration' at this point, since although he generally uses 'demonstration' in the communication-theoretic sense, he also says that the validity of the inference depends on whether we have the same demonstration here. But the demonstrations used in communication are merely instrumental to my achieving a visual fix on the person. And whether it would be legitimate for me to trade on identity depends on whether I achieved my visual fix on the object in the same way both times. This means that an account of how we understand demonstratives has to give some elucidation of the idea that there is such as thing as the way in which you achieve your visual fix on an object. Bearing in mind these problems about the interpretation of the notion of a 'demonstration', suppose we consider how the notion is to be used in giving an analysis of demonstratives. We could give the gist of Kaplan's analysis by saying that a use of a demonstrative like 'that yacht' is to be analysed by:

Sense

107

dthat + [demonstration] This puts a lot of weight onto the technical term 'dthat' which Kaplan introduces. He gives a canonical explanation of the notion in 'Afterthoughts'. Kaplan writes: The word 'dthat' was intended to be a surrogate for a true demonstrative, and the description which completes it was intended to be a surrogate for the completing demonstration. On this interpretation 'dthat' is a syntactically complete singular term that requires no syntactical completion by an operand. (A 'pointing', being extralinguistic, could hardly be a part of syntax.) The description completes the character of the associated occurrence of 'dthat', but makes no contribution to content. Like a whispered aside or a gesture, the description is thought of as offthe-record (i.e., off the content record). It determines and directs attention to what is being said, but the manner in which it does so is not strictly par? of what is asserted. The semantic role of the description is pre-propositional; it induces no complex, descriptive element into content. (Kaplan 1989a: 581) Ordinarily, then, 'dthat' is to be coupled with a description, which stands in for the demonstration used in conjunction with a demonstrative. In this framework, how could we make sense of the idea that the experienced location of an object has a special connection with the meaning of the visual demonstrative referring to it? In fact, Kaplan sets out a proposal of Michael Bennett's which addresses exactly this point: Michael Bennett has proposed that only places be demonstrata and that we require an explicit or implicit common noun phrase to accompany the demonstrative, so that: that [pointing at a person] becomes: dthat [the person who is there [pointing at a place]] (Kaplan 1989b: 527-8) This proposal of Bennett's is indeed the obvious way to accommodate, within Kaplan's framework, the insight that there is a special connection between the experienced location of an object and the visual demonstrative which refers to it. The trouble is that the demonstration, in Kaplan's account, is intended to fix the reference of the demonstrative. Consequently, if the descriptive phrase, 'the person who is there [pointing at a place]', is to fix the reference of the demonstrative, 'that person', then there is no possibility of referring to a person while being under an illusion as to the location of that person. But this rules out the possibility of demonstrative identification of objects seen in mirrors, through prisms, and so on; and these certainly are real possibilities. We could rule out the possibility of error here only by supposing that the place in question is

108

Sense

identified only as 'the place at which that person is located'; but then the spatial demonstration cannot fix the reference of the demonstrative. If we want, within Kaplan's framework, to accommodate the insight that there is a special connection between the experienced location of an object and the visual demonstrative which refers to it, then we have to keep Bennett's proposal but rework the explanation of 'dthat'. In effect, all that Kaplan tells us about the correct interpretation of 'dthat' in the passage quoted above is that when coupled with a demonstration which singles out an object uniquely, it yields a term whose contribution to any proposition is to identify the object on whose modal properties the modal status of the proposition depends. There is, though, more to say about the meaning of 'dthat'. We can ask how the prefixing of 'dthat' to a demonstration affects the cognitive role of propositions involving the term, and we can ask how the prefixing of 'dthat' to a demonstration affects the behaviour of propositions involving the term in inferences which trade on identity. The conclusions we reached already imply that the way to proceed here would be to: (a) keep the analysis of a demonstrative as: dthat + demonstration, but use the conception of a 'demonstration' as a way of achieving a visual fix on the object, rather than the conception of a demonstration as an instrument of communication; (b) interpret the prefixing of 'dthat' as implying that what follows is an indication of the binding principles governing your current attention to the object; this will determine the cognitive significance of the term (the way in which propositions involving it are to be verified, and the way in which to act on propositions involving the term); and (c) lay down that the validity of inferences trading on identity depends on whether the same binding principles are being used in connection with the demonstratives involved. Points (a)-(c) would, of course, be consistent with the point Kaplan was most anxious to urge in connection with 'dthat': namely, that only the object referred to, and not any of the descriptive or quasi-descriptive material involved in singling out that object, are relevant to the evaluation of the modal status of propositions involving 'dthat'. We might roughly sum up the gist of this approach by saying that, in addition to Kaplan's comments specifically on the evaluation of modal status, 'dthat' has been reinterpreted as a device signalling that what follows it is an articulation of the binding principles governing the act of attention that has been used to achieve a visual fix on the object referred to. This

Sense

109

reinterpretation is needed to articulate the cognitive role of demonstrative propositions and their role in inference. An implication of this is that not just any descriptive material can follow 'dthat', because your perceptual systems cannot use just anything to bind information as all relating to a single object. The special role that location has, as perhaps the single most pervasive binding principle, will then be reflected in Bennett's proposal that only identifications of objects by way of location can follow the use of 'dthat'. But we need not go as far as Bennett in the insistence that only location will do; we could allow that on occasion you may be able to identify an object perceptually without perceiving its location, and that in those cases the demonstrative you use will not depend on a demonstration identifying location. 7. Evans on Demonstratives I want finally to contrast the account I have given with Evans's (1982) remarks on demonstratives. Evans gives an important place to location in the meaning of a visual demonstratives. But he has a quite different explanation to the analysis I have been giving of why it is important. The contrast begins with the generality of his account. The varieties of reference he discusses are all in fact identifications of physical things. But he is trying to fit them to a mould designed equally for all types of singular reference, including reference to abstract objects. He says that 'in the case of a proposition of the form "a is F", knowledge of what it is for it to be true must be the result of two pieces of knowledge, one of which can be equated with an Idea of an object, and the other with an Idea of a property, or, more familiarly, a concept' (Evans 1982: 106). This point is the core of his conception of singular reference. Grasp of a singular way of thinking is to be explained, on this view, as requiring a capacity to think a certain range of thoughts. In effect, a 'functional' characterization is given. To grasp a singular idea a, one must be capable of grasping such thoughts as that a is F, that a is G, and so on, for each way of thinking of a property available to one. The task of a theory of singular reference is to explain what makes this possible. Evans's answer is that what makes this possible is that one's grasp of the idea a consists in what he calls discriminating knowledge of which thing a is. Evans's conception of discriminating knowledge is explained and defended by reference to his distinction between fundamental and nonfundamental singular ideas. This distinction is applicable equally to singular ideas of all types: to ideas of abstract as to ideas of concrete objects. A fundamental idea of an object identifies it as an object of a certain sort: as,

110

Sense

for example, a number, or a shape, or a mountain. It further identifies it as the possessor of the characteristics which ultimately differentiate it from all other things of the same sort. Thus a number is ultimately differentiated from all other numbers by its position in the number series; this is called the 'fundamental ground of difference' of the object. The fundamental ground of difference of a spatial object is taken to be its location at a time: its position then with respect to other objects. The most basic way in which one can have discriminating knowledge of an object is, on Evans's account, by grasping a fundamental idea of it; that is, by knowing what sort of thing it is, and those of its characteristics which ultimately differentiate it from all other things of that sort. As he says, 'Such an Idea constitutes, by definition, distinguishing knowledge of an object, since the object is differentiated from all others by this fact'; that is, the fact of its having the particular fundamental ground of difference that it does (1982:107). Yet why should we think that distinguishing knowledge in this sense is sufficient for singular reference? At this point Evans appeals to his doctrine about the way in which first-level concepts are understood. The doctrine is that in the first instance, possessing a concept of the property of being F is knowing what it is for a proposition of the form '6 is F' to be true, where '