Bermúdez, J.L., Thinking Without Words PDF

Thinking without Words
PHILOSOPHY OF MIND SERIES
Series Editor
David J. Chalmers, University of Arizona
Self Expressions
Minds, Morals, and the Meaning of Life
Owen Flanagan
The Conscious Mind

In Search of a Fundamental Theory
David J. Chalmers
Deconstructing the Mind

Stephen P. Stich
The Human Animal

Personal Identity Without Psychology
Eric Olson
Minds and Bodies

Philosophers and Their Ideas
Colin McGinn
What's Within?
Nativism Reconsidered
Fiona Cowie
Purple Haze
The Puzzle of Consciousness
Joseph Levine
Consciousness and Cognition

A Unified Account
Michael Thau
Thinking
without
Words
Jose Luis Bermudez
OXPORD
UNIVERSITY PRESS
2003
OXPORD
UNIVERSITY PRESS
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
Copyright © 2003 by Oxford University Press, Inc.

Published by Oxford University Press, Inc.
198 Madison Avenue, New York, New York, 10016
www.oup.com
Oxford is a registered trademark of Oxford University Press

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,
electronic, mechanical, photocopying, recording, or otherwise,
without the prior permission of Oxford University Press.
Library of Congress Cataloging-in-Publication Data

Bermudez, Jose Luis.
Thinking without words / Jose Luis Bermudez.
p. cm.—(Philosophy of mind series)
Includes bibliographical references and index.
ISBN 0-19-515969-1
1. Philosophy of mind. I. Title, n. Series.
BD418.3 .B47 2003
128'.3—dc21 2002029277
1 3 5 7 9 8 6 4 2
Printed in the United States of America
on acid-free paper
Acknowledgments
This book was begun during a two-year period of leave from teaching and admini-
stration that started in September 1998. The first year was spent at the Centre de
Recherche en Epistemologie Appliquee (CREA) at the Ecole Polytechnique in
Paris. I am extremely grateful to the Ecole Polytechnique for appointing me as
Charge de Recherche for the 1998-99 academic year. During this period I benefited
from a European Research Fellowship jointly funded by the Royal Society of
Edinburgh and the Caledonian Research Foundation. The second year was made
possible by a semester's sabbatical from the University of Stirling and a matching
period of leave under the Arts and Humanities Research Board's Research Leave
scheme. I owe a considerable debt to my colleagues in the Philosophy Department
at Stirling for allowing me to benefit from these opportunities.
A very preliminary version of the material in this book was presented in Paris in
spring 1999 in a course of lectures entitled "La pensee sans le langage," offered as
part of the DBA en Sciences Cognitives run by the Ecole Polytechnique, the Ecole
des Hautes Etudes en Sciences Sociales, and the Universite de Paris IV. My
thoughts were sharpened by the questions of the students who attended the course.
A slightly later version was delivered in Spanish under the title "El pensamiento
sin el lenguaje" at the University of Barcelona in April 2000.1 am very grateful to
Manuel Garcia Carpintero for extending the invitation and to Carolina Arrowsmith
for helping me with translating the lectures into Spanish. In August 2000 I deliv-
ered the same set of lectures at the Universidad Nacional in Bogota, Colombia. In
Bogota each lecture was followed by a commentary and a period of probing and
fruitful discussion with students and faculty. A version of chapters 7 and 8 was pre-
sented on a second visit to the Universidad Nacional in November 2001. Thanks
are due to Juan Jose Botero and his colleagues and students.
Material from the book has been presented at various colloquia and conferences.
The minimalist conception of nonlinguistic thought discussed in chapter 3 was ini-
tially put forward in my contribution to a symposium entitled "Thought without
Language," organized by John Campbell at the 1998 annual conference of the Eu-
ropean Society for Philosophy and Psychology in Lisbon. Some of the ideas in
chapters 4 and 5 were originally developed for a conference entitled "Animal
Minds," at Oxford University, organized by Susan Hurley in April 1998. Later ver-
sions were delivered at CREA in June 1998 and at the Belief Ascription Confer-
ence in San Marino organized by Marina Sbisa in December 2000. Very early
vi Acknowledgments
versions of the discussion of the language of thought hypothesis were delivered at

the University of Aix-en-Provence in January 1999 and the University of Geneva
in March 1999. The distinction between three levels of rationality that occupies
most of chapter 6 was first worked out during a fortnight's stay at the Universidao
Nacional Autonoma de Mexico (UNAM) Centra de Neurobiologia in Queretaro,
Mexico, in November 1998. My thanks to Jose Luis Diaz for his hospitality and
encouragement. The basic idea of chapters 8 and 9, that intentional ascent requires
the capacity for semantic ascent, was arrived at in Paris during various meetings of
the Action, Perception, Intentionalite, Conscience (APIC) seminar—in response to
the refusal of participants in the seminar to believe the arguments I offered in The
Paradox of Self-Consciousness for the dependence of conceptual thought on lan-
guage. Fran£ois Recanati, Elisabeth Pacherie, Joelle Proust, Jerome Dokic, and
Pierre Jacob were particularly helpful.
The first version of the manuscript was greatly improved by comments from
four referees who have waived their right to anonymity. Donald Griffin, Colin
Allen, David Braddon-Mitchell, and Chris Gauker made numerous probing criti-
cisms and helpful suggestion. I have also greatly benefited from written comments
from Peter Sullivan and Jonathan Lowe.
The final reworking of the manuscript was carried out at Simon Fraser Univer-
sity in Vancouver, where I was Visiting Professor for the 2002 spring semester. I
am very grateful to the faculty and students who attended my Thinking Without
Words course and helped me toward better formulations of many of my claims and
arguments. My final revisions of the manuscript were greatly helped by extensive
comments from David Chalmers, who has been an exemplary series editor.
Preface
Can creatures who do not have a language think? On the one hand, many types of
nonlinguistic creatures behave in ways that seem to require treating the creatures
in question as thinkers. The evidence is not simply anecdotal. Much of the most
exciting and influential recent research in developmental psychology, cognitive
archeology, and cognitive ethology explicitly assumes that the capacity for thought
is not in any way tied to language possession. On the other hand, we do not really
have any way of attributing thoughts to nonlinguistic creatures other than by crude
analogy with the attribution of thoughts to language-using creatures. We have no
theoretical framework for understanding the content and nature of nonlinguistic
thought or the mechanisms of reasoning and reflection of which nonlinguistic crea-
tures might be capable. And in the absence of such a theoretical framework, the
practices of explanation within which the attribution of thoughts to nonlinguistic
creatures seems so necessary remain without a secure foundation.
In this book I develop a framework for thinking about the thoughts of nonlin-
guistic creatures that acknowledges the differences between thinking without
words and thinking with words while nonetheless enabling us to attribute to non-
linguistic creatures thoughts with many of the characteristic features of language-
based thoughts. As will emerge, the thoughts of nonlinguistic creatures can have
compositional structure—they have distinguishable components that can feature in
further thoughts. They possess determinate contents, and it is often possible, de-
spite the claims of many philosophers, to identify and spell out with reasonable ac-
curacy the precise way that nonlinguistic creatures are thinking about their envi-
ronment. Yet there are limits to the range of thoughts that nonlinguistic creatures
can entertain because, as I argue in chapters 8 and 9, certain types of thinking (all
those involving intentional ascent, or thinking about thoughts) require a linguistic
vehicle.
Theoretical discussions of nonlinguistic thought tend either to deny that there
can be any such thing or to speculate about the vehicles of nonlinguistic thought.
Each of these general approaches is associated with a particular conception of the
relation between thought and language. One very influential philosophical concep-
tion of the nature of thought, derived from the writings of Frege, is based on the
principle that the study of thoughts can only proceed via the study of the sentences
that express them. Theorists impressed by this conception of the nature of thought
will find the notion of thought without language deeply problematic. Many of the
problems that appear insuperable from the perspective of the classical philosophi-
viii Preface
cal approach seem less threatening if we shift paradigm and adopt the dominant
understanding of the nature of thought within cognitive science, construing all
thoughts (those of language-using creatures as well as those of nonlinguistic crea-
tures) as relations to sentences in an internal language of thought. Yet, whatever
the merits of the language of thought hypothesis as an account of the mechanics of
cognition, it cannot provide a full account of nonlinguistic thinking. In particular,
it does not provide us with an appropriate epistemology for the thoughts of nonlin-
guistic creatures.
In this book I develop a different approach to thought without language. Unlike
the two approaches just outlined, the account I offer of thinking without words em-
phasizes the epistemological and explanatory dimensions of the problem. The best
way of approaching the problem of thought without language, I suggest, is through
providing an epistemological basis for the practice of attributing thoughts to non-
linguistic creatures and for the psychological explanations within which those at-
tributions take place. Accordingly, the book contains considerable detailed discus-
sion of empirical material from developmental psychology, cognitive psychology,
and studies of animal behavior, both experimental and ethological. The theory that
emerges is philosophical, in the sense that it is based on a particular philosophical
conception of the nature of thought and thinking, but that philosophical view is de-
veloped and refined through detailed consideration of the practical implications of
attributing thoughts to non-language-using creatures.
The basic problems that set the framework for this book are set out in chapter 1,
where I outline the different types of question posed by the forms of psychological
explanations of the behavior of nonlinguistic creatures given in various parts of the
cognitive and behavioral sciences. Chapter 2 explores the differing responses
to these questions given by the two approaches to the nature of thought outlined
earlier, and shows how neither can provide a fully satisfying account of thinking
without words. Chapter 3 considers a deflationary or minimalist construal of the
nature of nonlinguistic thought that might be deployed to finesse the apparent need
to attribute thoughts to creatures that are not language-users. The aim of the mini-
malist proposal is to show that thinking behavior in nonlinguistic creatures can be
understood in nonpropositional and perceptual terms, rather than through the attri-
bution of propositional attitudes such as beliefs and desires. In opposition to this I
suggest that there are important types of nonlinguistic thought that cannot be ac-
commodated in the manner proposed by the minimalist. The requirements of psy-
chological explanation often demand that we attribute to nonlinguistic creatures
thoughts that are structured, represent the world in a highly determinate way, and
reflect the particular mode of presentation under which the creature apprehends its
environment. Chapters 4 and 5 are devoted to showing how we might go about at-
tributing such determinate and structured thoughts to nonlinguistic creatures. In
chapter 41 explain how a theorist might fix an ontology in a way that will allow the
theorist to determine what objects a particular non-language-using creature is ca-
pable of thinking about—or, in other words, that will elucidate how the creature
"carves up" its world into bounded individuals. Chapter 5 explores how a seman-
Preface ix
tics can be provided for nonlinguistic thoughts in a way that both does justice
to philosophical constraints on acceptable theories of content and provides the
ethologist or developmental psychologist with a workable method of assigning
contents to the beliefs and desires of nonlinguistic creatures.
A theory of nonlinguistic thought is incomplete without an account of nonlin-
guistic reasoning and the norms of rationality by which such reasoning is gov-
erned. In chapter 6 I show how an account of nonlinguistic rationality emerges
when we pose the question: What could count as evidence that a nonlinguistic
creature is behaving rationally? There are several different forms of evidence that
can come into play here. At the most sophisticated level, a creature is behaving ra-
tionally when it is sensitive to the consequences of different courses of action, but
there are types of rationality that do not involve such consequence-sensitivity. Dif-
ferent forms of rationality are appropriate to different types of explanation, and
I draw a distinction between level 1 rationality and level 2 rationality that maps
onto the distinction between explanations of the type proposed by the minimalist
and explanations that make use of belief-desire psychology. Reasoning and ration-
ality are, of course, correlative notions, and in chapter 7 I pursue the question of
the forms of inference available at the nonlinguistic level. I offer an account of
protoinference that respects the differences between linguistic and nonlinguistic
thought (and in particular the impossibility of explaining nonlinguistic reasoning
in formal terms), while at the same time offering analogues at the nonlinguistic
level of some basic forms of inference.
In chapters 8 and 9 I address in more general terms the scope and limits of non-
linguistic thought. I argue that there are certain types of thinking for which a lin-
guistic vehicle is essential—and by this I mean a public language rather than a
private language of thought. Roughly speaking, a linguistic vehicle is required for
all types of thinking that involve intentional ascent, or what is sometimes called
metarepresentation. I argue in chapter 8 that intentional ascent requires semantic
ascent, on the grounds that intentional ascent requires the ability "to hold a thought
in mind" in a way that can only be done if the thought is linguistically vehicled. In
chapter 9 this argument is developed and the limits of nonlinguistic thought plotted
in the context of a range of different types of thought. These language-dependent
cognitive abilities range from second-order reflection on one's own beliefs and de-
sires and the capacity to attribute thoughts to others to the ability to entertain
tensed thoughts and to deploy logical concepts. Many of these language-dependent
cognitive abilities, however, have primitive analogues that do not involve inten-
tional ascent and hence are available at the nonlinguistic level.
A final word. In previous work I have mapped the distinction between linguistic
and nonlinguistic cognition onto a distinction between conceptual and nonconcep-
tual content. In The Paradox of Self-Consciousnesss I proposed that genuine con~
cept mastery involves an ability not simply to make judgments involving those
concepts but also to justify those judgments and to reflect on the grounds for them.
Since these are paradigmatically language-dependent activities, it follows that con-
cept mastery requires the possession of a language. This line of thought still seems
x Preface
to me to be fundamentally correct, and the discussion of the differences between

linguistic and nonlinguistic cognition in chapters 8 and 9 provides tools for devel-
oping it further. However, I have deliberately written this book in a manner that
does not involve or require a particular interpretation of the distinction between
conceptual and nonconceptual content. It seems to me that a clear view of the dis-
tinction between conceptual and nonconceptual content is more likely to emerge
from a clear understanding of the differences between linguistic and nonlinguistic
cognition than vice versa.
Contents
Note to the Reader xv
1 The Problem of Thinking without Words 3
1.1 Morgan's canon and psychological explanations

of behavior 6
1.2 The parameters of the problem 10
2 Two Approaches to the Nature of Thought 13
2.1 The Fregean approach: Thoughts as the senses

of sentences 14
2.2 The language of thought hypothesis 22
2.3 The language of thought and the epistemology

of thought without language 27
2.4 The road ahead 31
3 Minimalist Approaches to Nonlinguistic Thought 34
3.1 Thinking-how and thinking-that: An intuitive

contrast 36
3.2 Minimalist approaches to nonlinguistic thought 38
3.3 The minimalist construal of intentional action

and explanation 46
3.4 The argument from planning 51

xii Contents
3.5 The argument from intelligent action 56
3.6 The argument from psychological explanation 60
3.7 Taking stock 62
4 Ascribing Thoughts to Nonlinguistic Creatures:

Toward an Ontology 64
4.1 Success semantics: Problems and prospects 65
4.2 Success semantics and radical interpretation 69
4.3 Two types of reification 72
4.4 Dishabituation and higher-order principles 78
4.5 From object-properties to ontology 82
5 Ascribing Thoughts to Nonlinguistic Creatures:

Modes of Presentationn8888
5.1 Individuals, modes of presentation and

the sortal problem 898
5.2 Structured thoughts and psychological explanation 95
5.3 Success semantics and nonlinguistic thought:

An overview 104
6 Rationality without Language 109
6.1 Theoretical and practical rationality 109
6.2 Level 0 rationality 116
6.5 Nonlinguistic rationality and psychological

explanation 128
Contents xiii
7 Practical Reasoning and Protologic 133
7.1 The decision-theoretic model of practical reasoning 133
7.2 Inference and proto-logic 140
8 Language and Thinking about Thoughts 150
8.1 Language as a cognitive tool 151
8.2 Intentional ascent and semantic ascent 158
9 The Limits of Thinking without Words 165
9.1 Belief revision and second-order cognitive dynamics 166
9.2 Higher-order desires 170
9.3 Intentional ascent and understanding other minds 171
9.4 Embedding and intentional ascent 178
9.5 Two types of general thought 183
9.6 From domain-specificity to domain-generality 185
Afterword 189
Notes 195
References 205
Index 219
This page intentionally left blank
Note to the Reader
The problems with which this book is concerned are of broad interdisciplinary in-
terest, and I have written the book with an interdisciplinary audience in mind. I
hope that the general line of argument will be of interest to anyone with a theoreti-
cal, experimental, or practical interest in nonlinguistic creatures and how they rep-
resent and think about the world. The book does not presuppose a philosophical
background, but there are certain sections that will perhaps be most directly of im-
mediate concern to philosophers. Readers without a philosophical background
could omit sections 2.1,2.2, and 2.3 on an initial reading, moving straight from the
end of chapter 1 to the final section of chapter 2. In chapter 3, section 3.2 could ini-
tially be omitted.
Thinking without Words
1
The Problem of Thinking
without Words
The recent "cognitive turn" in the behavioral and cognitive sciences has drastically
expanded the domain of the cognitive. High-level cognitive abilities are being
identified and studied in an ever-increasing number of species and at ever-earlier
stages of human development. The contemporary behavioral sciences have almost
completely abandoned a longstanding tenet in the study of cognition, namely, that
language and thought go hand in hand, and hence that the study of thought can
only proceed via the study of language. Until recently, even those who held that
thought could in principle exist without language had little idea how to study
thought except through the language by which it is expressed. But current practice
in the study of animal behavior, in the study of prelinguistic infants, and in the
speculations of cognitive archaeologists about the evolutionary prehistory of
Homo sapiens, has left these assumptions far behind.
Our understanding of the early stages of human development has undergone a
sea change. Many developmental psychologists have come to speak of prelinguis-
tic infants as little scientists, possessing, testing, and refining theories about the na-
ture of the physical world (Gopnik and Meltzoff 1997). Complex experiments are
regularly set up to identify the predictions that infants as young as 3 months make
about the structure of physical objects and their dynamic and kinematic properties;
about the trajectories that objects take through space-time; and about what will
happen when objects interact (Baillargeon 1995, Spelke 1990). For example, when
they are 3 months old infants are sensitive to the solidity of objects. They show
surprise when one object appears in a place that it could only reach by passing
through another object. It is tempting to conclude, and many developmental psy-
chologists have concluded, that these infants have classified something as an ob-
ject and have correlative expectations about how that thing will behave on the
basis of that classification. There are important differences, of course, between
how different developmental psychologists interpret these expectations and ac-
4 Thinking without Words
companying inferences. Some see them as involving the operation of a primitive

theory (a naive physics) and correlatively as involving something approaching in-
ference. Others see them as primarily perceptual. All are agreed, however, in re-
jecting the traditional view of the prelinguistic infant as operating in a more or less
unstructured perceptual universe with a small range of sensorimotor schemas en-
abling primitive types of behavior.
The study of animal behavior has been no less drastically transformed (Allen
and Bekoff 1997, Dawkins 1986, Griffin 2001, Hauser 2000). The new discipline
of cognitive ethology is essentially the study of the mental states of animals and
of how those mental states manifest themselves in behavior. Unlike traditional ap-
proaches that have remained in the laboratory and attempted to account for animal
performance on complicated but artificial tasks in terms of various forms of asso-
ciationist learning, cognitive ethologists are prepared to study animals in the wild
as they deal with the practical problems that arise in foraging, finding mates, con-
structing shelters, and raising their young. Cognitive ethologists, unlike the older
generation of comparative psychologists, have little time for the project of trying
to explain how an animal behaves in terms of nonrepresentational stimulus-
response mechanisms or the fixed behavior patterns known as innate releasing
mechanisms. They start from the assumption that animals have certain desires and
certain beliefs about how the world is organized and act on the basis of those be-
liefs to try to ensure the satisfaction of their desires. Then they look at a species'
natural behaviors, interpreting them as sophisticated strategies for pursuing the de-
sires the members of that species seem to have. A good illustration of the degree of
cognitive sophistication now attributed to animals are the various behaviors that
have been analyzed as conscious attempts to manipulate conspecifics and members
of other species. Deception behaviors have been identified at all levels of the phy-
logenetic ladder, from the much-studied examples in higher primates such as
chimpanzees (Byrne 1995) to the broken-wing display of the plover (Ristau 1991),
the false alarm calls of the great tit (M011er 1988), and the role of sentinel birds in
mixed flocks (Munn 1986).
The study of human prehistory has taken on a new face. Cognitive archaeolo-
gists are finding evidence of thinking behaviors long before even the earliest
plausible dates for the emergence of language (Corballis and Lea 1999, Mellars
and Gibson 1996, Mithen 1990, 1996). Influential current accounts of the mind of
prehistoric man identify evolutionary stages of high-level but manifestly nonlin-
guistic cognition. Stephen Mithen has argued, for example, that the early prehis-
tory of the human mind was characterized by highly specialized cognitive mod-
ules, bodies of knowledge dedicated to specific aspects of the natural and social
world (Mithen 1996). These "multiple intelligences" permitted complex intellec-
tual skills and inferences. The social intelligence of the early human mind ini-
tially emerged from the constraints of social living, but then rapidly in its turn
expanded the possibilities and parameters of communal existence. The natural
history intelligence was tied up with what appear to have been the complex hunt-
ing and foraging strategies of the omnivorous early hominids. On Mithen's view
Thinking without Words 5
the emergence of language proper did not make cognition possible. Rather, it per-
mitted the integration of previously separated domain-specific modules. Merlin
Donald has offered a rather different, but no less cognitivist, conception of the
life of the prelinguistic hominids (Donald 1991). Prelinguistic hominids were ca-
pable of representing the world intentionally, of learning complex motor skills by
imitation, and of constructing novel motor routines from a recursively structured
motor vocabulary of basic movements. The integration of these individual skills
into a social environment, with group mimetic acts, social coordination, and sim-
ple forms of teaching, facilitated the emergence of complex tool making, patterns
of hunting that varied according to the season, primitive rituals, and a highly
ramified social structure. For Donald, as indeed for Mithen and many other stu-
dents of human prehistory, these sophisticated forms of instrumental and social
cognition are a precondition for the emergence of language, not a consequence of
that emergence.
These three areas of cognitive ethology, developmental psychology, and cogni-
tive archaeology are becoming ever more closely integrated. It is usual to find cog-
nitive archaeologists appealing to developmental work on infant grasp of naive
physics or naive biology to support their claims about the domain-specific special
intelligences of early man, just as they draw on the various studies of social cogni-
tion in nonhuman primates in developing theories about the social life of our earli-
est hominid ancestors. Research into animal cognition is now employing the disha-
bituation paradigm, which has proved so fruitful in the study of infant cognition
(Hauser 1998, Munakata et al. 2001). Similarly, in developmental psychology ar-
guments from evolution are playing an increasingly prominent role in the identifi-
cation and explanation of the different levels of cognitive development, as indeed
are arguments from the theory of animal learning (Gallistel et al. 1991). These in-
terconnected bodies of knowledge and research share the common assumption that
the domain of the cognitive far outstrips the domain of the linguistic.
This assumption that language is not necessary for thought will be the subject of
this book. Relatively little work has been done on elucidating the types of thinking
that are being attributed to different types of non-language-using creatures. The
consequences of the basic assumption have been far more deeply explored than its
theoretical background. This, of course, is how science proceeds. Niceties of con-
ceptual framework are not at the fore when there is a major paradigm shift. But the
new paradigms in ethology, developmental psychology, and the study of hominid
prehistory are sufficiently well established for the more theoretical questions now
to demand attention. In this book I will work toward providing a framework for
understanding nonlinguistic thought, drawing on experimental work but trying to
integrate it with the researches of philosophers on the nature of thought in a way
that does justice to the more important insights that have emerged from very differ-
ent approaches to the study of thought, language, and the relation between them.
The remainder of this chapter is devoted to clarifying the precise challenges posed
by the practice of explaining the behavior of nonlinguistic creatures in psychologi-
cal terms.
1.1 Morgan's Canon and Psychological Explanations of Behavior

Experimental psychologists, particularly those concerned with comparative psy-
chology (the psychological study of nonhuman animals) frequently appeal to a
methodological principle originally proposed by Lloyd Morgan in 1894. The first
formulation of Morgan's canon, as it is known, came in Morgan's book An Intro-
duction to Comparative Psychology, where he writes that
In no case may we interpret an action as the outcome of the exercise of a higher psy-
chical faculty, if it can be interpreted as the outcome of the exercise of one which
stands lower in the psychological scale. (Morgan 1894, 53)
Morgan himself did not provide a worked-out theory of what he terms the psycho-
logical scale, but it seems clear from his later formulations of the canon that he in-
terpreted it in an evolutionary and developmental manner. In the revised edition of
the Introduction he explicitly formulates the canon in evolutionary terms.
In no case is an animal activity to be interpreted in terms of higher psychological
processes, if it can be fairly interpreted in terms of processes which stand lower in the
scale of psychological evolution and development. (Morgan 1903, 59)
Morgan's canon has frequently been taken to motivate an extreme distrust of

attributing psychological states to nonhuman creatures, and is often cited as an ob-
jection to the whole enterprise of cognitive ethology. Clearly, given that it is ex-
plicitly formulated in evolutionary and developmental terms, it can hardly be ap-
plied without begging the question as a methodological principle for the study
of prelinguistic infants and early hominids (since in both these cases the issue at
stake is precisely when the "higher psychological processes" emerge in the course
of psychological evolution and development). Nonetheless, Morgan's canon is fre-
quently held to impose requirements on explanation that are incompatible with the
attribution of thoughts to prelinguistic infants and early hominids. Consider, for
example, the following exposition of Morgan's canon.
In Morgan's case, the principle amounted to this. Where there is a pattern of animal
behavior which must be explained, both as to form and to origin, and in the simplest,
but at the same time most adequate way, the experimenter should appeal to factors
observable in the situation in which the animal has been placed, in the behavior itself,
and in the machinery by which the behavior is made possible. It is not incumbent on
him to pass over these factors in order to appeal to a verbal construct, to a mind, or to
any other kind of mental factor which lies outside of, behind, or within the behavior-
situation. (Griffith 1943, 322)
The injunction here to seek an explanation in purely behavioral terms is just as

applicable to developmental psychology and cognitive archeology as it is to
ethology.
There are important questions about whether the appeals that have been made to
Morgan's canon do justice to Morgan's own approach to comparative psychology.
It is clear, for example, that Morgan was not simply advocating that one always
seek the simplest explanation. Explaining a behavior in terms of beliefs and desires
is frequently much simpler than explaining it as the result of a complex process of

instrumental conditioning. But, at a minimum, Morgan was suggesting that we
should not trust psychological explanations of behavior unless we are convinced
that those explanations are indispensable—that is to say, unless we are convinced
that the behavior in question cannot be explained in nonpsychological terms. This
is not, I suspect, a methodological principle with which very many of the authors
cited in the preceding section will disagree.
The roots of the cognitive explosion in the study of nonlinguistic creatures lie in
an increasingly widespread sense that the existing tools for the nonpsychological
study of behavior have produced patently inadequate pictures of infant and animal
cognition and human evolution. Psychological explanations are appealed to be-
cause they seem required to make sense of the complex behaviors that can be both
observed and experimentally demonstrated. And of course they are appealed to
very selectively. Few advocates of extending psychological explanation to the
realm of the nonlinguistic think that we should completely set aside our nonpsy-
chological models of behavior. Few would deny, for example, that the cognitive
ethologist should be ready to characterize many types of behavior in terms of in-
nate releasing mechanisms—namely, fixed patterns of behavior that are more com-
plex than reflexes, often involving a chained sequences of movements rather than a
simple reaction, and that yet seem to be instinctive (Tinbergen 1951). Nor is it
being proposed that the experimental study of animal behavior should completely
abandon explanations formulated in terms of processes of conditioning, whether
classical or instrumental.1 It seems clear that much of infant behavior is best un-
derstood in terms of the sensorimotor schemas so carefully studied and docu-
mented by Piaget. And many cognitive archeologists have devoted considerable
time to learning the techniques of stone-knapping and Neolithic tool construction
with a view to investigating just how cognitively sophisticated they are.
Psychological explanations of nonlinguistic behavior are hard won—and, of
course, only provisional. It may well be that a nonpsychological explanation will
emerge for behavior that we currently think requires a psychological explanation.
It is hard to see how any such possibility could be ruled out in principle. What
seems highly unlikely, however, is that satisfactory explanations in wholly nonpsy-
chological terms will be found for all the behaviors for which psychological expla-
nations are canvassed. It is difficult to imagine that researchers into animal behav-
ior, infant development, and human evolution will return to the methodological
precepts of behaviorism. Even though they might, in any given situation and in
complete conformity with Morgan's canon, always investigate whether a nonpsy-
chological explanation might be forthcoming, it is (I think) inconceivable that psy-
chological explanations will be banished from the study of nonhuman creatures.
The shift to the cognitive has now cast too many roots and the independence of
thought from language is now too deeply engrained for that to be possible.
If Morgan's canon is to be applied effectively, however, we require both a
specification of the distinctive features of psychological explanation and some sort
of understanding in operational terms of how behavior requiring such explanation
is to be identified. The issues that this raises will recur throughout this book, but
in the remainder of this section I will sketch out some of the principal respects in
which psychological explanations are to be distinguished from nonpsychological
or mechanistic explanations.2
The first distinguishing characteristic is that psychological explanations are te-
leological. That is to say, they explain an organism's behavior either in terms of the
purposes and desires that the behavior is intended to satisfy or, more minimally,
simply in terms of those that it does satisfy. On either construal, however, the be-
havior will come out as goal-directed in a way that mechanistically explicable be-
havior is not. It may be the case, for example, that a particular innate releasing
mechanism brings about the satisfaction of a desire. Consider, for example, the
pecking response in herring gull chicks (Tinbergen 1973). Newly hatched herring
gulls are particularly sensitive to the sensory input correlated with the length,
movement, and coloration of the adult herring gull's bill, and when they encounter
such input they respond by pecking vigorously at whatever it is that presents the
appropriate input (usually, of course, the adult's bill tip). The adult herring gull
responds by feeding the chick. Yet the chick's pecking is not a goal-directed re-
sponse, even though it can be described as bringing about the satisfaction of a
desire. It is not a goal-directed behavior because it is an invariant response to the
appropriate stimuli.
This brings us to the second characteristic of psychological explanations,
namely, the fact that the behavior they seek to explain cannot be explained and pre-
dicted as a function of invariant responses to detected stimuli. Innate releasing
mechanisms, such as the herring gull pecking response, have the following charac-
teristics (Lea 1984).
• They are triggered by specific stimuli.
• They always take the same form.
• They occur in all members of the relevant species.
• Their occurrence is largely independent of the individual creature's history.
• Once launched they cannot be varied.
• They have only one function.
This means, of course, that if one can identify a member of a given species and has
some understanding of the innate releasing mechanisms characteristic of members
of that species at the appropriate stage of development, then one will be able
straightforwardly to predict what the creature will do when it registers stimuli of
the appropriate type. Registering the relevant stimulus causes the appropriate re-
sponse, and this can be fully understood, explained, and predicted without any ap-
peal to an intermediary between stimulus and response. Similar input-output links
can be seen in the case of sensorimotor schemas and various types of conditioned
behavior. Psychological explanations of behavior only become necessary when no
such input-output links can be identified. The essence of a psychological explana-
tion is that it explains behavior in terms of how the creature in question represents
its environment, rather than simply in terms of the stimuli that it detects. Psycho-
logical explanations involve appealing to representational states that function as

intermediaries between sensory input and behavioral output.
How does this work? How can the appeal to psychological states allow us
to make sense of behaviors that cannot be understood as invariant responses to
stimuli? When we appeal to representational states to explain behavior we rarely, if
ever, appeal to single states operating in isolation. The issue here is one of cogni-
tive integration. The behavior of organisms that are suitably flexible and plastic in
their responses to the environment tends to be the result of complex interactions
between internal states. Organisms respond flexibly and plastically to their envi-
ronments in virtue of the fact that their representational states respond flexibly and
plastically to each other, most obviously through the influence of stored represen-
tations on present representations. The possibility of learning and adaptation de-
pends on past representations contributing to the determination of present re-
sponses, and hence interacting with them. Moreover, representing a particular
feature of the environment does not have effects on behavior just when a creature
is confronted by that very feature. It can also be effective when the creature is con-
fronted by something that is relevantly similar to the represented feature, or when
the feature is absent and there is nothing relevantly similar in the environment (as
when a creature determines to leave its shelter because it cannot detect either a
predator or anything predator-like in the vicinity).
The role of representations in explaining behavior is closely tied to the possi-
bility of misrepresentation. Tropistic and classically conditioned behavior can be
explained without reference to representational perceptual states because the re-
sponse is invariant once the creature in question has registered the relevant stimuli.
That is to say, the behavior can be explained by the way things are in the immedi-
ate environment, together with the assumption that the creature is appropriately
sensitive to the relevant stimuli. The need to appeal to how things are taken to be
comes in only when the law-governed correlation between stimulus and response
breaks down. This can come about either when the response occurs in the absence
of the stimulus, or when the stimulus occurs and is registered without the response
following. In the first of these cases one might say that there is a representation of
the stimulus and that this is what generates the response, and in the second case
one might say that even though the stimulus is there it is not represented in a way
that would bring about the appropriate response. Both of these are different types
of misrepresentation, and the existence or possibility of misrepresentation is a use-
ful index of the need to advert to psychological explanations.
Jerry Fodor has suggested how the possibility of misrepresentation might be ex-
plained by the lack of a lawlike correlation between stimulus and response (Fodor
1986, 14-16). Explaining why there should be an invariant connection between
stimulus and response requires postulating the existence of transducers (devices
that transfer sensory stimuli into physical output that is in a suitable form to inter-
act with the motor system). We need transducers to explain how an organism can
pick up stimuli at all, and when there is an invariant connection between stimulus
and response we need to postulate the existence of transducers that are sensitive to
specific properties—what are often known as dedicatedtrransducers. transducers. T

and operation of these dedicated transducers, hooked up in the appropriate ways to
the motor system, is all that we need to explain the invariant behavior in question.
Now, in the case of intentional behavior that does not involve a lawlike connection
between stimulus and response, it is clear that we are not dealing with dedicated
transducers. Creatures that are behaving intentionally are of course sensitive to
particular properties of the stimulus, but these are not properties that can be de-
tected by dedicated transducers. So how are they detected? In some sense their
presence needs to be "inferred from" the presence of properties that can be picked
up by transducers. The need for such a process of inference, however, brings with
it the possibility of error. Hence the possibility of misrepresentation.
I will show later on in the book how these characteristics of psychological ex-
planations and the behavior that they are proposed to explain can be applied in
practice. For the moment I need simply note the following three characteristics of
psychological explanations.
1. They serve to explain behavior in situations where the connections be-
tween sensory input and behavioral output cannot be plotted in a lawlike
manner.
2. They rely on the cognitive integration of different psychological states.
3. They appeal to psychological states that admit of misrepresentation.
That such explanations are, as a matter of fact, applied to nonlinguistic creatures is
beyond doubt. That they are applied to nonlinguistic creatures as a matter of neces-
sity is highly plausible. As remarked earlier, there are no good reasons for thinking
that the practice of psychological explanation should be confined to language-using
creatures. I have so far, however, said little about precisely how these explanations
should be understood. As I will show in the next section, these very general com-
ments about the role of psychological explanations in explaining the behavior of
nonlinguistic creatures leave open a range of important questions about the type of
explanation at stake and the nature of the representational states that they attribute.
1.2 The Parameters of the Problem

A theorist confronting the wide range of psychological explanations applied to
nonlinguistic creatures, and convinced, in the light of Morgan's canon and the gen-
eral considerations brought out in the previous section, that these psychological
explanations are indispensable, will find it natural to ask just how we are to under-
stand the thoughts and types of thinking that are now being attributed to creatures
not previously taken as candidates for thinking at all. Questions such as the follow-
ing spring to mind.
• Can nonlinguistic creatures properly be described as having beliefs about
the way the world is (or indeed, knowledge of how the world is)? Can they
properly be described as having desires, or hopes or fears? How does the
psychological vocabulary appropriate for characterizing nonlinguistic crea-
tures relate to that which we apply to language-using creatures? What sorts
of things might nonlinguistic thoughts be? How can animals and prelin-
guistic infants grasp thoughts in a way that will influence behavior?
• Do the thoughts of nonlinguistic creatures represent the world in a determi-
nate way (do they have determinate contents)0! Can we characterize those
thoughts in the way we characterize the thoughts of language-using crea-
tures, by using "that—" clauses that specify their content? Are there distinc-
tive ways that nonlinguistic creatures classify objects in thought? Do the
thoughts of nonlinguistic creatures have any sort of determinate structure?
Should they be assimilated to sentences, or are they in fact closer to percep-
tual representations and mental images?
• If it is the case that nonlinguistic creatures do have beliefs and desires with
determinate contents, how are we to work out what those contents are? As
observers of animal or infant behavior we of course only have behavior to
go on. How are we to attribute thoughts on the basis of behavior in a way
that will explain behavior?
• Does the attribution of thoughts to nonlinguistic creatures explain their be-
havior in anything like the way that the attribution of thoughts explains the
behavior of language-using creatures? Can we say that a nonlinguistic crea-
ture behaves in a certain way because it has beliefs and desires that render
that particular course of action a rational one to adopt? What sort of transi-
tions between thoughts can nonlinguistic creatures make? Are they properly
described as making inferences?
These questions, all of which will be tackled in this book, fall into four different
groups. Questions in the first group are broadly metaphysical. They all concern the
nature and mechanics of nonlinguistic thought. Questions about whether nonlin-
guistic creatures can properly be described as having beliefs and desires fall under
this heading, as does the issue of how such beliefs and desires (if they exist) might
have effects on behavior. The second group of questions concerns the semantics of
nonlinguistic thought. These are questions about how we should understand the
content of nonlinguistic thought and about the different types of thinking available
to language-less creatures. Under this heading fall, for example, the issue of
whether nonlinguistic thoughts are structured and the question of whether any
sorts of thoughts are in principle unavailable to nonlinguistic creatures. The third
group of questions is largely epistemological. Even if all the metaphysical ques-
tions are answered satisfactorily, we will still need some account of how we can
come to attribute thoughts to nonlinguistic creatures. We need to have a reliable
method for working backward from behavior to the thoughts generating that be-
havior. Without this we will be unable to deploy our theory of nonlinguistic
thought in explaining behavior. The fourth and last group of questions has to do
with the practice of explanation within which the attribution of nonlinguistic
thought is embedded. In the forms of psychological explanation with which we are
most familiar (the standard, belief-desire explanations of the behavior of language-
using, concept-possessing humans) we assume that psychological explanation is
an idealized reconstruction of practical decision-making. In a psychological expla-
nation we cite beliefs and desires such that the agent whose behavior is being ex-
plained could have reasoned (and perhaps even did reason) from those beliefs and
desires to the intention to act in the way that he actually did act. So a proper under-
standing of the practice of giving psychological explanations of the behavior of
nonlinguistic creatures must bring with it a plausible account of how nonlinguistic
creatures fix on a particular course of action.
An adequate account of nonlinguistic thought must accommodate all four of
these dimensions. It must
• explain the metaphysics of nonlinguistic thought
• explain the semantics of nonlinguistic thought
• explain how it is possible for us to identify the content of such thoughts
• explain the decision-making processes of nonlinguistic creatures in a way
that underwrites the practice of psychological explanation
These four dimensions must, of course, be accommodated by a theory of any type
of thought (linguistic or nonlinguistic), and one way of bringing out the particular
difficulties posed by nonlinguistic thought is by looking at how these four sets of
questions are dealt with by two prominent approaches to the nature of thought in
general.3 That will be the task of the next chapter.
The two approaches I will discuss provide very different accounts of the meta-
physical, semantic, epistemological, and explanatory dimensions of thought. They
both share, however, the basic assumption that the only way to understand thought
is through language. One approach, with its origins in some of Frege's writings on
the nature of thought, holds that we can only approach thought through natural lan-
guage. The second approach, inspired by Fodor's representational theory of the
mind, explains thought through the hypothesis of an internal language of thought. I
will use these two accounts as foils in order to develop a sense of the challenges
that face the theorist of nonlinguistic thought.
One reason for highlighting the Fregean conception of thought and the language
of thought hypothesis is that they represent the extreme points on a spectrum of pos-
sible positions with respect to the question of the relative priority of thought and
language. Whereas the Fregean approach to thought ties thoughts ineliminably to
the sentences of a natural language and what it is to understand those sentences, the
language of thought hypothesis completely reverses the order of explanation, sug-
gesting that the very idea of understanding a natural language does not make sense
without assuming a linguistically structured inner representational medium. A
corollary of this is that, whereas the Fregean view stresses the distinctiveness of lan-
guage in a way that makes the possibility of nonlinguistic thought seem deeply prob-
lematic, the Fodorian view goes a long way toward assimilating linguistic thought
and nonlinguistic thought. In contrast to both of these approaches, the positive ac-
count of the nature of nonlinguistic thought that I will be developing in the main
body of the book aims to show both that nonlinguistic thinking has significant com-
monalities with linguistic thought and that there are equally significant limitations to
the types of thought that can be entertained by nonlinguistic creatures.
2
Two Approaches to
the Mature of Thought
Our problem is finding an appropriate conceptual framework for thinking about

the nature and status of psychological explanations in the study of human infants,
nonhuman animals, and early hominids. I have shown in the previous chapter that
such a conceptual framework needs to have four dimensions. It needs to incorpo-
rate, first, an account of the metaphysics of nonlinguistic thinking (an account of
the nature of nonlinguistic thoughts and of what it is for a nonlinguistic creature to
be a thinker); second, an account of the semantics of nonlinguistic thinking (an ac-
count of how the world is represented at the nonlinguistic level); third, an account
of the epistemology of nonlinguistic thinking (explaining how we as observers and
theorists should go about attributing thoughts to creatures that lack a language);
and, fourth, an account of the psychology of nonlinguistic rationality and decision-
making.
In order to appreciate in more detail the requirements imposed within each of
these dimensions it is helpful to examine in some detail two prominent theories of
the nature of thought. The two theories I examine in this chapter are Frege's con-
ception of thoughts as the senses of sentences and Fodor's language of thought hy-
pothesis to the effect that thinking should be understood in terms of the operation
of sentence-like formulae in an internal language of thought. Both approaches start
off from a single basic assumption, which is that the nature of thought can best be
analyzed through the nature of language, but each approach takes a very different
view of the essence of language. In section 2.1 I explain the key elements of
Frege's conception of thoughts as the senses of sentences, using Frege as a foil to
clarify the challenges that confront a theory of nonlinguistic thought under each of
the four dimensions identified in the previous chapter. As I show, taking thoughts
to be the senses of sentences provides a clear account of each of the four di-
mensions of thinking that nonetheless cannot be extended to nonlinguistic crea-
tures. This brings into very clear focus the desiderata for a theory of nonlinguistic
thinking—namely, to provide answers to those questions that, on the Fregean

theory, are answered by taking thoughts to be the senses of sentences. In section
2.2 I move on to the language of thought hypothesis. This hypothesis provides an
account of the nature of thought that is applicable to nonlinguistic creatures and
that accommodates three of the four dimensions. As emerges in section 2.3, how-
ever, the language of thought hypothesis falls short when it comes to the epistemo-
logical dimension. For this reason, even though the language of thought hypothesis
provides a worked-out account of the mechanics of thinking that can be applied to
nonlinguistic creatures, it cannot as it stands serve as a fully satisfying framework
for the explanatory practices of cognitive ethology, developmental psychology,
and cognitive archeology.
In the final section of the chapter I explore the lessons to be learned from these
two approaches to the nature of thought and suggest an alternative way to ap-
proach nonlinguistic thought. The approach that I propose in section 2.3, and that
the remainder of the book develops, takes the epistemological dimension of non-
linguistic thought to be primary.
2.1 The Fregean Approach: Thoughts as the Senses of Sentences

There are genuine questions to be raised about whether the following sketch accu-
rately represents the views of the historical Frege.1 It is certainly the case that
Frege was far more interested in mathematical, particularly arithmetical, thoughts
than in those expressible by means by a natural language. It is also true that Frege
had a rather dismissive attitude to natural language. Perhaps the Fregean view that
I will be discussing is an artifact resulting from the tendency of some of Frege's in-
terpreters to focus on too narrow a range of his writings.21 do not believe this to be
the case. But even if it were it would not matter. The conception of the relation be-
tween thought and language that has emerged from reflection on Frege's writings
represents a powerful and unified theory. It deserves to be taken into account irre-
spective of its exegetical credentials. I will begin by explaining what I take to be
the essentials of the Fregean position and then go on to draw out some of the prin-
cipal constraints that the Fregean account imposes on a theory of nonlinguistic
thought.
By 'thoughts' Frege did not mean individual acts of thinking. He in fact had sur-
prisingly little to say about individual acts of thinking. What he was interested in
was what we now term the contents of individual acts of thinking—what is known,
for example, in an instance of knowledge or believed in an instance of belief. A
Fregean thought is not an act of thinking. Rather, an act of thinking involves a rela-
tion to a thought. In believing, for example, that Paris is the capital of France I
stand in a certain relation to the thought that Paris is the capital of France—a
thought to which any other thinker can become related. These thoughts are abstract
objects that serve not simply as the objects of propositional attitudes such as
knowledge or belief but also as the senses of sentences. In fact, on the understand-
ing of Frege's philosophy of thought that I shall be putting forward, our basic un-
Two Approaches 15
derstanding of thoughts comes from considering them as the senses of sentences. It

is in this sense that we should understand the claim frequently made by Michael
Dummett that the essence of Frege's approach is the idea that the analysis of
thought can proceed only via the analysis of language.
The Fregean theory of thought is a worked-out response to a series of problems
about the nature of thought that it is important to have clearly in mind when think-
ing both about the nature of thought and about the nature of nonlinguistic thinking.
These problems can be mapped on to the four dimensions identified in the previous
chapter.
Let me start with the metaphysical dimension. For Frege, as for just about every
other account of the propositional attitudes, belief is the fundamental propositional
attitude. To have a belief is to commit oneself in thought to a certain state of affairs
holding in the world. The belief is true just if that state of affairs does hold, false
otherwise. Belief itself is a psychological state, not a part of the thought but rather
a relation to the thought. The thought itself is the postulate (for want of a better
word) that the relevant state of affairs holds in the world. To have the attitude of
belief to that thought is to endorse the postulate. To endorse it, of course, is to take
it to be true—and one cannot take to be true anything that is not a candidate for
truth or falsity.
Suppose now that we ask how we are to understand these postulates to the effect
that a certain state of affairs holds in the world. The central idea here is that what
we endorse when we take the attitude of belief to a thought is precisely what we
understand when we understand an assertoric sentence (where an assertoric sen-
tence is one that characterizes the world as being a certain way). The significance
of this emerges when we consider what sort of a thing a thought must be that we
can grasp it. It is no easy matter to explain what it is to grasp something as true.
And yet, if thoughts are to feature in psychological explanation in the manner dis-
cussed in the previous chapter, then we clearly need an account of how a thinker
can grasp thoughts. The basic idea that thoughts are the senses of assertoric sen-
tences gives us a clear way of understanding how we might grasp thoughts. We
grasp thoughts through understanding the sentences that express them. Once we
have grasped a thought in this way we can simply entertain it, where this might be
construed as contemplating it without assigning it a truth-value. We can endorse it
as true, which can be construed as asserting it. Or we can take to it any of the other
propositional attitudes, which can be construed in terms of the different types of
force that can be attached to the sentence.3 Frege's account, therefore, gives a clear
answer to the fundamental metaphysical problem of how we can make epistemic
contact with thoughts—how thoughts can be the sort of things that can feature in
reasoning and have causal effects within the world.
Moving from the metaphysical dimension to the semantic dimension, the most
obvious problems with which Frege is engaged have to do with the structure of
thoughts, and of thought itself. The first problem is the problem of explaining the
composition of individual thoughts. There is a clear sense in which thoughts seem
to be structured entities, made up of elements that can reappear in further thoughts.
This is closely tied to the possibility of thoughts entering into inferential relations
with each other (and hence directly tied to the utility of psychological explanations
of behavior, since beliefs and desires stand in inferential relations to the behavior
that they explain). Some inferential relations are truth-functional. They hold be-
tween thoughts in virtue of their truth-value. So, for example, we know that we can
infer "Socrates is mortal" from "Socrates is mortal and the Argives defeated the
Trojans" because there are no circumstances in which "Socrates is mortal and the
Argives defeated the Trojans" could be true and "Socrates is mortal" false. The in-
ternal structure of the thoughts expressed by these sentences is irrelevant, which is
why in explaining why these inferences are valid we can simply employ preposi-
tional variables and do not need to go into details of what the propositional vari-
ables stand for. Other inferences (those that we might model in the predicate rather
than the propositional calculus) depend on the structure of the individual thought.
Consider, for example, the inference from "Socrates is mortal" to "There is at least
one thing that is mortal." This is a valid inference, in the sense that it is not possi-
ble for the premise to be true and the conclusion false. If Socrates is mortal then at
least one thing is mortal. But the validity of the inference is not in any sense a
function of the truth-values of the relevant sentences. In order to explain why the
inference is valid we need to see the inference as taking the form "a is F, therefore,
at least one thing is F." But this requires us to see the thought expressed by
"Socrates is mortal" as having a structure that we standardly represent as "Fa" in
the predicate calculus, and the thought expressed by "There is at least one thing
that is mortal" as having the structure "3 jcFx." We cannot understand the validity
of inferences of this type without adverting to the structure (or logical form) of the
thoughts in question.4
Let us suppose, then, that the possibility of inferential transitions between
thoughts presupposes some level of structure, whether in the thoughts themselves
or in the way they are determined. How are we to make sense of that structure?
The Fregean approach to thought gives us a clear answer. Thoughts are the senses
of sentences, and the structure of a thought is given by the logical form of the sen-
tence that expresses it—where the logical form of a sentence is given by the appro-
priate logical regimentation of that sentence in Frege's concept-script (roughly, the
second-order predicate calculus). This logical form identifies the constituents of
the thought, that is, the senses of which it is composed. Consider, for example, the
thought expressed by "There is at least one mortal thing." According to Frege, the
proper analysis of the logical form of that sentence identifies it as expressing a
thought whose constituents are (roughly speaking) a first-level functional sense
(corresponding to the predicate "— is mortal") and a second-level functional sense
to the effect that the first-order functional sense has at least one object falling
under it. Of course, the details of Frege's positive account of the composition of
thoughts can, and perhaps ought, to be challenged. Many philosophers, for exam-
ple, would contest the claim that quantifiers should be viewed as functions taking
first-order functions as their arguments. But it is undeniable that the basic idea that
thoughts are the senses of sentences yields both a powerful account of the compo-
Two Approaches 17
sitional structure of thoughts and a practical method for uncovering the composi-
tional structure of any given thought.5
Still on the semantic dimension of thought, there is a further and more overarch-
ing structural isomorphism in play, not simply between a thought and the sentence
expressing it but between thought as a whole and language as a whole. It is widely
held that thoughts have certain structural characteristics as a matter of definition.
Not only are they compositional (made up of recombinable constituents that can
feature in a range of further thoughts in the way I have already considered) but the
range of available thoughts is not in principle limited. The capacities of individual
thinkers may be limited by time, computational constraints, and so forth, but in
principle an indefinite number of thoughts can be constructed from a finite number
of primitives. That is to say, thoughts are generative. An account of thought needs
to provide some explanation of this phenomenon. Taking thoughts to be the senses
of sentences, so that the analysis of thought proceeds via the analysis of language,
offers a clear way of understanding generativity. We can quite simply point to
the mechanisms that secure linguistic generativity—the mechanisms of pronomi-
nal reference that allow one sentence to be embedded within another, the logical
operators that allow sentences to be combined to form complex sentences, and the
ways that complex referring terms (such as definite descriptions) can be formed
from simple predicates.
Let us move on to the epistemological dimension. The problem here is how we
are to understand and attribute thoughts. It is obvious that the practice of psycho-
logical explanation requires the ability to identify the psychological antecedents
of behavior from a third-person perspective. It requires being able to work back-
ward from observed behavior to attribute beliefs and desires with determinate
contents—being able to interpret observed behavior as intended to achieve certain
specifiable ends in the light of the agent's information about the world. But once
again the abstract nature of thoughts makes this enterprise seem deeply problem-
atic. How are we supposed to be able to identify the thoughts to which an agent is
related? How are we supposed to identify the contents of her beliefs and desires? If
thoughts are the senses of sentences, and if belief is in some sense an interioriza-
tion of the linguistic act of assertion, then it is clear how we might proceed (when
we are dealing with linguistic creatures). We can use what people say to work out
what they believe, on the twin assumptions, first, that if someone sincerely assents
to "p" then i it is reasonable to take her to believe that p, and, second, that, ifeople
believe that;? they will generally be disposed to assent to "p." These disquotational
principles are, of course, rules of thumb rather than necessary and sufficient condi-
tions on what it is to believe that. Few would suggest that someone believes that p
if and only if she is disposed to assent to "p." And in practice of course the inter-
pretation of beliefs and the interpretation of speech go hand in hand, since we need
some sort of assumptions about what a person believes before we can determine
how her words might be a guide to what she believes (how else could we interpret,
for example, whether she was being sincere). Nonetheless, taking thoughts to be
the senses of sentences, and the structure of a thought to be given by the structure
of the sentence that expresses it, yields a very clear resolution to the epistemologi-
cal question of how thoughts can be identified and attributed.
The final dimension is the explanatory dimension. Models of psychological ex-
planation go hand in hand with models of practical decision-making. The standard
form of a psychological explanation as applied to language-using humans is "A
0-ed because P1...n" wherewherep qP n report certain psyhological statesf
the subject, most standardly a belief-desire pair. The guiding assumption is that
P1 . n rationalize or make comprehensible why the subject should have acted
the way he did. How do they do this? One standard way of looking at the matter is
expressed in the following passage from Donald Davidson.
If someone acts with an intention then he must have attitudes and beliefs from which,
had he been aware of them and had he the time, he could have reasoned that his act
was desirable. . . . If we can characterize the reasoning that would serve we will, in
effect, have described the logical relations between descriptions of beliefs and desires
and the description of the action, when the former gives the reasons with which the
latter was performed. We are to imagine, then, that the agent's beliefs and desires
provide him with the premises of an argument. (Davidson 1978/1980a, 85-86)
According to this view, which in fact goes back at least as far as Aristotle, a psy-
chological explanation is informative because it points toward (although usually
does not actually specify in any detail) a set of decision-making processes that
could plausibly have generated the target behavior. In the case of our normal day-
to-day psychological explanations it is widely held that we lean heavily on a
shared 'theory of mind,' part of which spells out the inferences that we can expect
to be made by other people with whom we interact and whose behavior we need to
be able to explain and predict. This is why psychological explanations rarely need
to spell out the actual reasoning that might have led from the relevant belief-desire
pair to the action in question. It is obvious to all of us how someone with that
belief-desire pair would have been led to act the way she did. This is partly be-
cause we realize that, had we had that combination of beliefs and desires in similar
circumstances, we would have acted the same way. But it is also because we ap-
preciate the inferential transitions that could have led a rational thinker from that
combination of beliefs and desires to a decision to act in that particular way.
Psychological explanation is in key respects the converse of practical reasoning.
So we need a model not simply of how thoughts can be grasped but also of how
there can be inferential relations between thoughts of the sort that might issue in
behavior.
Once again, the Fregean approach to thought through language provides such a
model—more accurately (since the psychology of reasoning was not Frege's con-
cern), it makes clear how there can be such a model. The notion of logical form is
at the heart of Frege's conception of thought, given that the structure of a thought
is determined by the logical form of the sentence that expresses it. This is the key
to a Fregean account of the psychology of inference. Although inferential relations
in the strict logical sense (what Frege called the laws of thought) hold between
Two Approaches 19
thoughts, we grasp inferential relations between thoughts in terms of inferential re-

lations between logical forms. These relations can be syntactic. That is to say, they
can be understood in terms of rule-governed transitions applying to sequences of
sentences in virtue of their formal structure. Or they can be semantic, holding be-
tween sentences in virtue of their semantic properties. Both ways of understanding
inference have been well studied.
If, therefore, we take thoughts to be the senses of sentences then we have a clear
account of the nature of thought that accommodates the four dimensions identified
in the previous chapter. It is hard to see, however, how this account might be ex-
tended to cover thinking without words. It is true that the general proposal that the
analysis of thought should proceed via the analysis of language is not directly in-
compatible with the existence of nonlinguistic thought. The proposal that a thought
be analyzed through the sentence that expresses it does not entail that any thinker
capable of thinking that thought should be able to express it. There could still be
thoughts without language even if the study of thought has to proceed via the study
of language. Something like this was, after all, Frege's own view. He was explicit
that, although thoughts come for us "clothed in the form of a sentence," it is possi-
ble that other creatures could grasp thoughts directly (although he was doubtless
not thinking about the candidate instances of thinking without words discussed in
the first section of the previous chapter).
The problem comes, however, because it is not enough to provide an account of
what thoughts are and how they should be analyzed. If we are interested in
thoughts from the perspective of psychological explanation (as opposed, for exam-
ple, to the perspective of logic) then our account of what thoughts are has to be ac-
companied by an account of how thoughts can be responsible for action. And this,
as I have shown, seems to require developing the Fregean view in a way that is in-
compatible with the possibility of nonlinguistic thought. If thoughts are the senses
of sentences then it is hard to see (on the assumption that the analysis of thought
has to proceed via the analysis of language) how grasping a thought can be any-
thing other than understanding a sentence, and how practical reasoning is to be
understood except in terms of transitions between sentences. It is sometimes
suggested that we understand a sentence in terms of an independently grasped
thought, but to my knowledge no account has yet been given of what it might be to
grasp a thought independently of understanding a sentence. Even if we bracket
these difficulties, it seems that the relation between the structure and composition
of a thought and the structure and composition of the sentence that expresses it
poses an epistemological problem. We find this problem presented very clearly in
the following passage from Donald Davidson's influential article "Thought and
Talk," where he poses the question of why (in his view) the attribution of thoughts
depends on the interpretation of speech.
The general, and not very informative, reason is that without speech we cannot make
the fine distinctions between thoughts that are essential to the explanations we can
sometimes confidently supply. Our manner of attributing attitudes ensures that all the
expressive power of language can be used to express such distinctions. One can be-
lieve that Scott is not the author of Waverly while not doubting that Scott is Scott; one
can want to be the discoverer of a creature with a heart without wanting to be the dis-
coverer of a creature with a kidney. One can intend to bite into the apple in the hand
without intending to bite into the only apple with a worm in it; and so forth. The in-
tensionality we make so much of in the attribution of thoughts is very hard to make
much of when speech is not present. The dog, we say, knows that its master is at
home. But does it know that Mr Smith (who is his master), or that the president of the
bank (who is that same master) is home? We have no real idea how to settle, or make
sense of, these questions. It is much harder to say, when speech is not present, how to
distinguish universal thoughts from conjunctions of thoughts, or how to attribute con-
ditional thoughts, or thoughts with, so to speak, mixed quantification ("He hopes that
everyone is loved by someone"). (Davidson 1978/1980a, 164)
So, even though the letter of the Fregean approach does not rule out in principle
the attribution of thoughts to nonlinguistic creatures, it does seem to present diffi-
culties for the practices of psychological explanation within which those attribu-
tions are embedded.
In the light of this a theorist concerned with those practices of explanation has a
choice. The first possibility is to seek an alternative account of the nature of non-
linguistic thought that will show how to answer the central questions that the
Fregean account leaves open. He will need to explain how we can understand what
it is to have a belief without appealing to the understanding of sentences or the lin-
guistic act of assertion. He will need to explain how can we understand the infer-
ential transitions by which beliefs and other propositional desires can lead to ac-
tion. And he will need to explain how ethologists, developmental psychologists,
and cognitive archeologists can go about identifying and attributing propositional
attitudes.
In the next section I consider an account of the nature of thought that makes
considerable progress in this direction. This is the complex of ideas associated
with the language of thought hypothesis. The language of thought hypothesis is a
hypothesis about the mechanics of thinking in a way that the Fregean approach
clearly is not. If we follow Frege in distinguishing the logical question of what
thoughts are from the psychological question of how we grasp thoughts and deploy
them in reasoning, then the language of thought hypothesis most definitely ad-
dresses the second question. The central idea of the language of thought hypothesis
is that propositional attitudes are relations to sentences/formulae in an internal lan-
guage of thought. These sentences/formulae represent the thoughts in a way that
allows them to feature in reasoning and decision-making. The language of thought
hypothesis provides a clear alternative to the Fregean account. As I show in sec-
tion 2.3, however, the language of thought hypothesis is seriously limited when
it comes to the epistemological dimension of nonlinguistic thought and is of
little practical use in grounding psychological explanation at the nonlinguistic
level.
But the theorist of nonlinguistic thought does not need to accept the way the prob-
lem has so far been set up. We have been working on the assumption that psy-
chological explanation at the nonlinguistic level is a species of belief-desire psycho-
Two Approaches 21
logy, directly analogous to that which we employ in the psychological explanation

of language-using creatures. But this assumption might be rejected. The theorist
may seek to find differences between the type of thoughts involved in psychological
explanations of the behavior of nonlinguistic creatures and the type of thoughts that
we have been discussing up to now. If these differences are significant enough, then
it may well be that the problems we have been considering cease to loom so large. In
chapter 31 examine a minimalist or deflationaryyway of understanding the thoughtss
that might be attributed to nonlinguistic creatures. On the minimalist view, it is per-
fectly correct to describe nonlinguistic creatures as thinkers, but we need to be care-
ful about how we describe the type of thinking going on. We should not assimilate it
to the thinking engaged in by language-using creatures, or think of it in terms of
belief-desire psychology. When we appreciate the differences between these types
of thinking we see that the apparent incompatibility with the Fregean approach does
not really arise. Nonlinguistic thinking does not involve propositional attitudes—
and, a fortiori, psychological explanation at the nonlinguistic level is not a variant of
belief-desire psychology. So there is no need to give an account of how a nonlin-
guistic creature might believe something—or indeed of how it might grasp a thought
at all. It is not, according to the minimalist view, appropriate to describe nonlinguis-
tic thoughts as having determinate contents. Nor should the types of reasoning en-
gaged in at the nonlinguistic level be modeled on those at the linguistic level. For all
these reasons, then, the failure of the Fregean approach to extend to nonlinguistic
creatures should not be a cause for concern. As I show in chapter 3, however, there
are powerful arguments suggesting that the minimalist conception cannot be the
whole story about nonlinguistic thought.
For the moment, the point to extract is that we have at least established a condi-
tional. If it is indeed the case that psychological explanations applied to nonlin-
guistic creatures are broadly similar in structure and intent to psychological expla-
nations of language-using creatures, then we will need to find answers to our four
key questions. Here they are again.
1. What can we take as our paradigm for grasping a thought (if we cannot
model it as understanding a sentence) and taking a propositional attitude
to it?
2. How should we understand the structure of nonlinguistic thought?
3. How can we go about identifying and attributing propositional attitudes
without the resources provided by linguistic interaction?
4. How can we understand the inferential transitions by which beliefs and
other propositional desires can lead to action (without taking those beliefs
to have linguistic form)?
Of course, it may turn out (as the minimalist view suggests) that nonlinguistic
thoughts are so different in type from linguistic thoughts that these questions do
not need to be answered. I turn to this suggestion in the next chapter. In the next
two sections of this chapter, however, I consider the extent to which the language
of thought hypothesis can deal with these four questions.
2.2 The Language of Thought Hypothesis

At the heart of the Fregean approach to thought is the suggestion that there exists
an isomorphism between the structure of thought and the structure of language.
For Frege and those who follow him it goes without saying that the language in
question must be a public language. The language of thought hypothesis can be
viewed in broad outline as commitment to the fundamental claim about isomor-
phism while lifting the restriction to public languages. The central idea is that
propositional attitudes are relations to sentences in an internal language of
thought—these sentences represent the content of the thought in a way that allows
the content to be deployed in reasoning and decision-making.
The various problems in the mechanics of thinking discussed in the previous
section can all be viewed as variants of the problem of explaining how thoughts, as
abstract objects, can have effects within the physical world—how, for example,
they can be believed and combine with other propositional attitudes to bring about
behavior. On the Fregean view, thoughts are the senses of sentences, and we can
model how they are grasped in terms of the understanding of the sentences that ex-
press them. Belief is modeled on assertion and practical reasoning on the inferen-
tial relations (whether syntactic or semantic) between sentences that are candidates
for assertion. According to the language of thought hypothesis, however, this gets
the direction of explanation completely the wrong way around. In order to under-
stand how thoughts can have effects within the physical world we need to under-
stand the role they play in the generation of behavior, and for that we need to take
them first as the contents of propositional attitudes, and only second as the mean-
ings of sentences. Propositional attitudes are mental states, presumably realized in
(and perhaps even identical to) states of the central nervous system. So it is natural
to think that we must understand the nature of thoughts through understanding the
nature of the propositional attitude states that take them as contents. Here the dis-
tinctive proposal of the classical cognitive science approach emerges. Proposi-
tional attitudes are relations that the thinking subject bears to sentences—but to
sentences in an internal language of thought rather than in a communicable public
language. In the remainder of this section I explore this proposal in more detail,
considering how it might be developed into a theory of nonlinguistic thought along
the dimensions I have been discussing.
Versions of the language of thought hypothesis have frequently been proposed
in the philosophical literature. They go at least as far back as William of Ockham
in the fourteenth century (Panaccio 1992). All these versions share the same basic
idea, namely, that thinking takes place in an internal representational medium that
is in some sense structured like a language. The inner representational medium is
the vehicle for thought. The modern form of the language of thought hypothesis, as
proposed by Jerry Fodor in 1975 and defended by him in many writings since,
is distinctive, however, in providing a detailed and worked-out account of exactly
how thinking might take place in the linguaform inner representational medium.
This account is based on a close analogy between the workings of the mind and the
Two Approaches 23
workings of a computer and is informed by research into the formal properties of

logical systems (metalogic). The core elements of the hypothesis can be identified
in three basic claims.
The first claim is that inferential transitions between thoughts can be understood
in terms of causal interactions between physical structures. This claim has two di-
mensions worth highlighting. The first is that it reflects the robust intentional real-
ism at the heart of the language of thought hypothesis—that is to say, the concep-
tion of prepositional attitudes (and, by extension, the thoughts that are then-
contents) as causally efficacious entities. It may well be that the contents of prepo-
sitional attitudes are abstract objects, but, according to the language of thought hy-
pothesis, if these abstract objects are to play any explanatory role we must be able
to identify physical structures that realize them. This intentional realism is already
a departure from the Fregean approach. Whereas the Fregean approach has nothing
to say about the causal processes that begin with grasping a thought and endorsing
it in belief and end eventually in action, the language of thought hypothesis takes
those processes to be the principal explanandum of a theory of thought. The sec-
ond point is that the language of thought hypothesis is based on a causal concep-
tion of inference. That is, it views practical reasoning as primarily a matter of
causal transitions between thoughts. Again, this is an idea that would be alien to
the Fregean approach.
The distinctiveness of the language of thought hypothesis becomes clearer when
we consider the second key claim, which is that the physical states that realize
thoughts and that stand in the causal relations defining thought are sententially
structured. This in turn can be broken down into two subclaims. The first of these
is that the realizers of thoughts are compositionally structured—that is to say, they
have separable components that can feature in a range of thoughts. The second
subclaim is that these separable components stand in physical relations to each
other that in some sense map onto the logical relations holding between the parts
of the thought that they collectively realize. The physical structure that realizes the
thought is isomorphic to the thought that it represents. It contains elements that
correspond to the parts of the thought and that are connected together in a manner
that directly maps onto the way the elements of the thought are connected together.
The physical structure is a picture of the thought that it realizes.
In the language of thought hypothesis, as in the Fregean approach, it is at the
level of logical form that we must look to identify the structure of the sentence that
gives the genuine structure of the thought. In the Fregean approach the structure of
a thought is given by the logical form of the sentence expressing it—in Frege's
terms, by the translation of the sentence into the Begriffsschrift. In the language of
thought hypothesis, however, there is no gap between the surface structure of the
sentence and its logical form. Sentences in the language of thought represent at the
level of logical form—for the following reason.
One of the key phenomena that the language of thought is called on to explain is
how linguistic comprehension is possible in language-using creatures.6 Since the
general model of linguistic comprehension espoused by defenders of the language
of thought hypothesis is one on which a natural language sentence is understood

by being translated into a sentence in the language of thought, it is hard to see how
the language of thought can admit a distinction between surface structure and deep
structure. There can be no interpreter who will be able to abstract away from sur-
face syntactical features to identify the deep structure of a sentence in the language
of thought. Nor is there any further language into which such a sentence can be
translated. So either language of thought sentences wear their deep structure on the
surface (in the way that sentences in a formal language do) or it becomes very un-
clear how they themselves can have semantic properties.
This requirement that sentences in the language of thought be fully transparent
at the level of logical form is related to the third basic claim in the language of
thought hypothesis, namely, that the causal transitions between physical states re-
spect the rational relations between the thoughts that those physical states repre-
sent. Inferential transitions between sentences in the language of thought are sensi-
tive only to the syntactic features of the relevant sentences, although in a way that
preserves relations of semantic consequence (Fodor 1987, ch. 1, Rey 1997, sec.
8.2). The language of thought theory lays great stress on the idea that inferential
transitions between thoughts can be a function of causal interactions between
physical structures (which of course is what sentences in the language of thought
are when one abstracts from their semantic properties). The obvious and acknowl-
edged inspiration for this basic idea is the metalogical fact that the first-order
predicate calculus is sound and complete (see, for example, Fodor 1987, ch. 1).
That is to say, in every well-formed proof in the first-order predicate calculus the
conclusion really is a logical consequence of the premises, and, conversely, for
every argument in which the conclusion follows logically from the premises and
both conclusion and premises are formulable in the first-order predicate calculus,
there is a well-formed proof. The well-formed-ness of a proof can be checked
purely mechanically, since it is a function solely of the syntactic properties of the
relevant formulae. The language of thought must, therefore, if it is to permit a
comparable form of mapping of syntactic properties onto semantic properties, pos-
sess some analogue of at least the metalogical property of soundness.
The conception of reasoning at the heart of the language of thought hypothesis
is well expressed in the following passage from Fodor that is worth quoting in full.
Computers show us how to connect semantical with causal properties for symbols.
So, if having a prepositional attitude involves tokening a symbol, then we can get
some leverage on connecting semantical properties with causal ones for thoughts.
. . . Here, in barest outline, is how the new story is supposed to go: You connect the
causal properties of a symbol with its semantic properties via its syntax. The syntax
of a symbol is one of its higher-order physical properties. To a metaphorical first ap-
proximation, we can think of the syntactic structure of a symbol as an abstract feature
of its shape. Because, to all intents and purposes, syntax reduces to shape, and be-
cause the shape of a symbol is a potential determinant of its causal role, it is fairly
easy to see how there could be environments in which the causal role of a symbol
correlates with its syntax. It's easy, that is to say, to imagine symbol structures inter-
acting causally in virtue of their syntactic structures. The syntax of a symbol might
Two Approaches 25
determine the causes and effects of its tokenings in much the way that the geometry
of a key determines what locks it will open.
But now, we know from modern logic that certain of the semantic relations among
symbols can be, as it were, 'mimicked' by their syntactic relations: that, when seen
from a very great distance, is what proof theory is about. So, within certain famous
limits, the semantic relation that holds between two symbols when the proposition
expressed by the one entails the proposition expressed by the other can be mimicked
by syntactic relations in virtue of which one of the symbols is derivable from the
other. We can therefore build machines, that have, again within famous limits, the
following property:
The operations of the machine consist entirely of transformations of symbols;

In the course of performing these operations, the machine is sensitive solely to
syntactic properties of the symbols;
And the operations that the machine performs on the symbols are entirely con-
fined to altering their shapes
Yet the machine is so devised that it will tranform one symbol into another if and
only if the propositions expressed by the symbols that are so transformed stand in
certain semantic relations—e.g. the relation that the premises bear to the conclusion
in a valid argument. (Fodor 1987,18-19)
The overall vision is a powerful and clear account of the mechanics of thought.
The account of inference can clearly be extended to yield an account of practical
reasoning applicable to nonlinguistic creatures.
Moreover, the language of thought hypothesis has a clear account of the nature
of belief and other propositional attitudes. Like the Fregean approach, the lan-
guage of thought hypothesis views propositional attitudes as complex entities
made up of an attitude and a content to which that attitude is taken. The content is
not, of course, the relevant sentence in the language of thought (any more than a
thought is to be confused with the natural language sentence that expresses it).
Sentences in the language of thought serve as internal surrogates for the thoughts
with which they are appropriate connected, and the particular attitudes one can
take to thoughts (believing as opposed to entertaining, for example, or desiring as
opposed to fearing) can all be understood in terms of the different functional roles
that their internal surrogates can occupy in the cognitive economy. These different
functional roles can be understood in terms of the different causal interactions into
which that token sentence can feature. So, on the language of thought hypothesis,
entertaining a thought is to be understood in terms of having it available to be used
as, for example, the antecedent of a conditional or as an undischarged assumption
in an argument (in the way that one might, for example, in proof by reductio ad ab-
surdum), where these conditionals and proofs are to be understood as causal inter-
actions holding in virtue of the syntactic properties of sentences in the language of
thought. This is the functional role corresponding to contemplating a thought with-
out committing oneself to its truth or falsity. Believing a thought has a different
functional role. To have a belief with a particular content (the content that p, for
example) is to be disposed to use the sentence in the language of thought that
stands as a surrogate for/? as a premise in practical reasoning (where practical rea-

soning is reasoning that feeds directly into action). It is to be disposed to take the
same attitude toward any language of thought sentence that can be 'inferred' from
p, and so on. Different functional roles can be identified for the other propositional
attitudes.
As far as the three dimensions directly connected with the mechanics of thought
are concerned, therefore, it looks as if the language of thought hypothesis is poten-
tially very useful for thinking about nonlinguistic thought. It should be recognized,
however, that the language of thought hypothesis is itself dependent on a particular
way of understanding nonlinguistic thought. The basic argument for the language
of thought hypothesis is that there are certain types of thinking that would not be
possible unless the thoughts they involved had some sort of linguistic vehicles.
The obvious response to this line of argument is to ask why the linguistic vehicles
in question cannot simply be the sentences of the thinker's language. Perhaps the
most promising way of blunting this objection (see, for example, Fodor 1975) is to
appeal to the thoughts of creatures that do not have any sort of language.7 So, for
example, the language of thought theorist might suggest that we need to postulate
an internal representational medium to make sense of practical decision-making
in animals and young infants—to understand, in roughly the manner of expected
utility theory, how beliefs and desires can issue in action. The point to notice, how-
ever, is that this argument only works on the assumption that psychological expla-
nations of the behavior of nonlinguistic creatures are to be modeled on the belief-
desire psychology we apply to language-using creatures. As I show in more detail
in section 7.1, the conception of practical reasoning in play interprets beliefs as
probability assignments and desires as a preference ordering and sees practical rea-
soning as the computation of the course of action with the maximum expected util-
ity. At no stage is any argument given for the claim that this is the only way to un-
derstand intentional behavior at the nonlinguistic level. And, as I show in the next
chapter, this is clearly not the only way that apparent thinking behavior at the non-
linguistic level should be understood.
Let us, then, draw a provisional conclusion. The language of thought hypothesis
is capable of accounting for the metaphysical, semantic, and explanatory dimen-
sions of nonlinguistic thinking, hence providing a framework for understanding
the type of psychological explanations to be found in ethology, developmental psy-
chology, and cognitive archeology, on the assumption that those psychological ex-
planations are instances of the general pattern of belief-desire explanation. That is
to say, the applicability of the language of thought hypothesis to nonlinguistic
creatures rests on treating them as believers, desirers, and reasoners on an under-
standing of belief, desire, and reasoning that is derived from thinking about
language-users. In this sense, therefore, the language of thought hypothesis is a
hostage to empirical fortune.
But how are we to know whether the behavior of nonlinguistic creatures really
does require belief-desire explanation? Recall that we require a conception of non-
linguistic thought that will underwrite the explanatory practices of the various dis-
Two Approaches 27
ciplines exploring the cognitive abilities of nonlinguistic creatures. This means

that the explanation offered of the nature of nonlinguistic thought must explain
how we, as observers of animal or infant behavior or cognitive archaeologists
striving to reconstruct the mental life of our early ancestors, can reliably attribute
thoughts to explain the behavior of the nonlinguistic creatures we are studying. I
turn in the next section to considering whether the language of thought hypothesis
can satisfy this epistemological requirement.
2.3 The Language of Thought and the Epistemology

of Thought without Language
At first sight the language of thought hypothesis seems impoverished from an epis-
temological point of view. The hypothesis that prepositional attitudes are relations
to sentences in an internal language of thought in such a way that the content of the
thought is given by the semantic properties of the relevant sentences makes it mys-
terious how we can determine what that content is. Suppose, for the sake of argu-
ment, that we adopt a causal covariance account of the content of sentences in the
language of thought (as suggested in Fodor 1987). Then, for a standard belief-
desire explanation of behavior, the contents of the relevant belief and desire will be
determined by the objects and properties in the world that covary in the appropri-
ate way with the constituents of the relevant sentences in the language of thought
(the sentence in the 'belief box' and the sentence in the 'desire box'). But it is far
from clear how we as third-person observers can work out what those contents are
in the case of nonlinguistic creatures.8
There are two distinct questions here. The first is whether the language of
thought hypothesis can be of any immediate epistemological use to practitioners of
the disciplines from which we are taking our lead. How might it help the cognitive
ethologist in the field trying to determine the right way to characterize and explain
the behavior of nonlinguistic creatures to be told that the propositional attitudes of
such creatures consist in sentences in an internal language of thought? The answer
is obvious: not at all. From the viewpoint of the developmental psychologist in the
laboratory or the cognitive ethologist in the field, the language of thought hypothe-
sis is of no immediate epistemological use whatsoever. Nonetheless, this leaves
open the second question, which is whether the language of thought hypothesis
might be of indirect epistemological use. It might be the case that, by studying cor-
relations between brain states and states of affairs in the distal environment, we
can work out what features of the environment a creature is capable of represent-
ing, so that we could arrive, so to speak, at an understanding of the vocabulary of a
given species' language of thought. It is natural in this context to appeal to the re-
sults of magnetic resonance imaging, single neuron studies, and other contempo-
rary methods of studying brain activity. Well-publicized studies such as those that
have led to the discovery of mirror neurons in monkeys (Rizzolati, Fogassi, and
Gallese 2000) might be thought to show that this goal is at least within reach. If we
were able to carry out such a program then we would be able to put the knowledge
thereby acquired to work in the practical business of attributing thoughts to mem-

bers of the relevant species.9 We would know what type of thoughts it would be
possible to attribute to members of a particular species (or to infants at a particular
stage of development).
Clearly, given the limitations of current knowledge about the brain, we are a
long way from being able to provide any such theory. But equally clearly, if we
have reason to think that such a theory will eventually be forthcoming, it would be
foolish to condemn the language of thought hypothesis as a framework for think-
ing about the thoughts of nonlinguistic creatures on epistemological grounds. Un-
fortunately, there are powerful grounds for thinking that it is in principle impossi-
ble to work backward from brain states to states of affairs in the distal environment
in the way that the language of thought hypothesis requires.
The main problem is quite simply that we have no way of accessing the sen-
tences in the language of thought that determine the contents of the creature's be-
liefs and desires.10 Proponents of the language of thought hypothesis often stress
that the notion of sentences in the brain is not to be taken literally. The brain does
not contain little sentences that a neuroscientist might discover by employing pow-
erful enough equipment in the right sort of way. Formulae in the language of
thought exist at a level of functional abstraction and cannot simply be read off
from immediately observable patterns of brain activity. Nonetheless, these sen-
tences are physical structures, albeit physical structures at a functional level sev-
eral layers removed from brute neuronal activity. They have to be physical struc-
tures if the language of thought is be a form of intentional realism (that is, if it is to
allow propositional attitudes to play a genuine causal role in the generation of be-
havior) as well as for the computationalist account of inference in terms of formal
transitions sensitive only to syntactic properties to be coherent. But how are we to
identify these physical structures?
There seem to be two possibilities (Braddon-Mitchell and Fitzpatrick 1990,
Crane 1990). Either we identify the physical structures in terms of their semantic
features or we identify them in terms of their syntactic features. The first possi-
bility is obviously unsatisfactory. It presupposes precisely what is in question,
namely, that we already have some way of working out the content of a sentence in
the language of thought. Yet the second is no less unsatisfactory. Identifying the
syntactic features of the physical structures that are sentences in the language of
thought cannot be done without identifying their semantic features. So we end up
with precisely the same circularity as on the first proposal. Let me explain.
In the case of sentences of a natural language, we can hive off syntax from se-
mantics, and so it seems a natural thought that we can do the same for sentences in
the language of thought. But in the case of natural language we can only hive off
syntax from semantics because we have an understanding of the language that in-
cludes both its syntactic and semantic aspects. We can bring this out by reflecting
on the project of radical interpretation—on how one might go about interpreting a
completely alien language without the benefits of a dictionary, grammar book, or
any relevant collateral information. No radical interpreter could formulate hy-
Two Approaches 29
potheses about the syntax of a completely unfamiliar language without at the same
time forming hypotheses about its semantics. How one takes the words to fit to-
gether depends on what one takes the words to mean. One very basic reason for
this is that the syntax of a natural language involves grammatical categories. It
involves certain expressions being substantives, others verbs, and others being
adjectives or adverbs. One cannot decide what category a given word falls into
without speculating about its semantics. It seems impossible, for example, to dis-
tinguish deciding whether a word refers to an object from deciding whether it is a
noun. So we can only ascribe a given word a particular syntactic role in a language
by according it a certain semantic characterization. Of course, this does not mean
that we cannot have a syntactic understanding of a given language. That would be
absurd. The point is rather that we can only have a syntactic understanding once
we are in a position to identify the syntactic properties of sentences, and we will
not be able to identify those syntactic properties without fixing at least some se-
mantic properties.11
This points to an important equivocation in the concept of syntactic properties
(Devitt 1990). Syntactic properties can be understood in purely physical terms, the
shape of the letters in a word for example. Alternatively, we can understand syn-
tactic properties as relational and functional, on the model of the syntactic proper-
ties of natural language, as having to do with how words are connected with each
other. It is only if syntactic properties are understood in the first sense that a clean
break between syntax and semantics seems appropriate.
But as far as identifying sentences in the language of thought is concerned, we
must understand syntactic properties in the second rather than the first sense. The
syntactic theorist is interested in the functional role of sentences in the language of
thought: how they are causally connected up to each other and to sensory input in a
way that explains behavioral output. The question we need to ask, then, is whether
there are constraints on the 'radical interpretation' of sentences in the language of
thought parallel to the constraints that clearly exist on linguistic radical interpreta-
tion, in such a way that fixing the syntactic properties of particular states will be
impossible without fixing at least some semantic/intentional properties.
Two such constraints spring immediately to mind, both to do with how syntactic
properties are fixed (Braddon-Mitchell and Fitzpatrick 1990, Crane 1990, Sterelny
1990). The first is that any such interpretation requires distinguishing inputs that
do not have a genuine causal role to play (because they are just noise) from those
that do. Some causal connections are functionally relevant, while others are not,
and any satisfactory theory will have to discriminate between these. Second, since
sentences in the language of thought are functional states realized in complex
physical structures, we need to identify causal properties that derive from the
physical instantiation/realization of a particular sentence, rather than from its func-
tional role within the language of thought. The basic point, therefore, is that to
identify sentences in the language of thought we need to distinguish relevant
causal relations from irrelevant ones. It is clear that syntactic properties as nar-
rowly construed (i.e., in physical terms) will be of no use here. There is no reason
to think that there will be any brute physical characteristics demarcating causal
properties that are functionally irrelevant. But syntactic properties in the broad
sense will be of no use either, because we need to discriminate relevant from ir-
relevant causal properties in order to fix the broad syntactic properties. So, if we
are to discriminate just the functionally relevant causal properties, it seems that we
will have to advert to considerations of function and hence to how the animal
needs to represent its environment. Either way, therefore, we will need to fix at
least some semantic properties before being in a position to pick out sentences in
the language of thought in terms of their syntactic properties.12
There is a further point to be taken into account here. In interpreting the type of
correlations at the neural level that might be thought to provide our evidence base
for identifying sentences in the language of thought, we need to make a principled
decision about the level at which we are to look for the neural correlates of cogni-
tion. Are we to look for representation of objects and properties at the level of indi-
vidual neurons, or at the level of populations of neurons? An answer to this ques-
tion cannot simply be read off from patterns of activation in the brain. Let us
suppose, for example, that we find a particular neuron that responds preferentially
to a particular type of stimulus—say, to the face of one's grandmother, to take the
classic example. As Churchland and Sejnowski (1992, ch. 4) have elegantly ar-
gued, such a preferential response would be compatible both with the hypothesis
that the neuron in question is a dedicated grandmother neuron (hence can stand in
a 'proper name' position in sentences in the language of thought) and with the hy-
pothesis that it merely forms a conspicuous part of a population of neurons that
collectively form a vector representation of the grandmother. It turns out that there
are powerful computational reasons for thinking that neural representation is most
likely to take a vector rather than a local form. The number of objects/features that
can be locally represented by a particular population of neurons is a linear function
of the number of neurons in that population, whereas the representative power
varies exponentially with the population when the coding takes a vector form.
Let us suppose, then, that the representational primitives in the brain are vectors
rather than single neurons. This means, of course, that in trying to identify the syn-
tactic properties of sentences in the language of thought we will be operating at the
level of populations of neurons (assuming that it is appropriate to think of popula-
tions of neurons as the bottom-level physical realization of sentences in the lan-
guage of thought). But how can one identify a population of neurons in a way that
might identify it as the constituent of a sentence in the language of thought? How
can one tell, amid the chaos of simultaneously firing neurons in the brain, which
neurons are firing together to provide a vector coding? The only way to identify
populations of neurons is to work out how particular tasks are being performed.
That is to say, by working backward from the particular processing tasks being
performed (hence from what particular populations of neurons might be doing).13
But this, of course, requires starting at the semantic level. It requires starting with
the particular computational task being performed, moving to hypotheses about the
representational tools for carrying the appropriate task out, and then testing those
Two Approaches 31
hypotheses against the observed patterns of activation. Once again it looks very
much as if we can only fix the syntactic properties of sentences in the language of
thought by starting with the semantic properties.
It looks doubtful, therefore, whether the language of thought hypothesis can
help us with the epistemological issues that must be confronted by any account of
thought in the absence of language. It doesn't give us any clues as to how we might
go about attributing thoughts to nonlinguistic creatures. This is not, it must be
stressed, an objection to the language of thought hypothesis as such. As a hypothe-
sis about the machinery of cognition, the language of thought hypothesis is under
no obligation to provide its own epistemology. What it does mean, however, is that
the language of thought hypothesis cannot provide a straightforward solution to
the problem of explaining the thoughts of nonlinguistic creatures. It provides an
account of how nonlinguistic creatures might be thinkers. It offers a theory as to
what the vehicles of nonlinguistic thought might be and meets many of the objec-
tions in principle to the idea of thought without language. But, insofar as it fails to
provide an epistemology for the thoughts of nonlinguistic creatures, it cannot be
the whole story.
2.4 The Road Ahead

As I showed in section 1.2, there are four dimensions to an adequate theory of non-
linguistic thought.
The metaphysical dimension of explaining how nonlinguistic thought is pos-

sible and what its vehicles might be
The semantic dimension of explaining the semantics of nonlinguistic thought
The epistemological dimension of explaining how it is possible for us to iden-
tify the content of such thoughts
The explanatory dimension of elucidating the inferential and decision-making
processes of nonlinguistic creatures in a way that underwrites the practice of
giving psychological explanations of their behavior
This chapter has explored a dialectic between two approaches to providing such a
theory, deriving from two fundamentally different conceptions of thought. On the
one hand, the Fregean approach, based on the principle that thoughts are the senses
of sentences and hence that the study of thoughts can only proceed via the study of
the sentences that express them, finds the notion of thought without language deeply
problematic. Many of the problems that appear insuperable from the perspective of
the Fregean approach seem less threatening if we shift paradigm and adopt a version
of the language of thought hypothesis, centered on the twin principles, first, that we
should study thoughts through studying prepositional attitudes rather than sentences
and, second, that prepositional attitudes should be construed as relations to sen-
tences in an internal language of thought. Yet the language of thought hypothesis
cannot provide us with a wholly satisfactory account, for two reasons.
The first problem with applying a version of the language of thought hypothesis
to nonlinguistic creatures is that it rests on an as yet unsubstantiated commitment
to the applicability of belief-desire psychology to nonlinguistic creatures. The
premise from which we began is that there are various types of behavior identifi-
able in human infants, nonhuman animals, and prelinguistic hominids that cannot
be satisfactorily explained according to existing nonpsychological models such as
innate releasing mechanisms, sensorimotor schemas, or instrumental or classical
conditioning. It looks, for a range of complex behavior types, as if we will need to
turn to psychological explanations that appeal, inter alia, to the way a creature rep-
resents the current environment and to the way those representations interact with
past representations (and representations of the past). But it remains an open ques-
tion whether these representations should be viewed as beliefs or desires, and
whether the type of practical reasoning in which those representations can be de-
ployed at the nonlinguistic level should be modeled, as Fodor suggests, on types of
practical reasoning defined over beliefs and other prepositional attitudes.
The second problem was discussed at some length in the previous section. The
language of thought hypothesis can be of no use when it comes to the epistemo-
logical dimension of nonlinguistic thinking. It is a hypothesis about the mechanics
of thinking, reached by, broadly speaking, transcendental arguments about what
cognitive architecture must be like for thinking behavior to be possible. But it of-
fers no guidance on how we might go about attributing thoughts to nonlinguistic
creatures and hence cannot as it stands serve as a framework for thinking about the
practice of attributing thoughts to nonlinguistic creatures.
The first of these two problems is addressed in chapter 3, where I consider what
I term the minimalist or deflationary account of thinking at the nonlinguistic level.
On the minimalist picture, the type of thinking available to nonlinguistic creatures
is very limited. The distinction between linguistic and nonlinguistic thought is
analogous to the distinction between knowing how to do something and knowing
that something is the case. Nonlinguistic creatures are capable of thinking in the
sense that they can perform various types of goal-directed actions involving repre-
sentations of the environment, but these representations are radically unlike the
thoughts available to language-using creatures. They are not compositionally
structured and do not have determinate contents. In chapter 3 I present a range of
arguments showing that there are important types of nonlinguistic thinking that
cannot be accommodated on the minimalist approach. The upshot of this chapter is
that we cannot avoid or downplay our theoretical commitment to some version of
belief-desire psychology in the absence of language. In the remaining chapters of
the book I offer a positive account of nonlinguistic thought.
Unlike the two approaches I have been considering in this chapter, the account I
offer of thinking without words emphasizes the epistemological and explanatory
dimensions of the problem. The best way of approaching the problem of thought
without language is through providing an epistemological basis for the practice of
attributing thoughts to nonlinguistic creatures and for the psychological explana-
tions within which those attributions take place.
Two Approaches 33
Providing this epistemological basis is a two-stage process. The first stage is fix-
ing an ontology in a way that will allow the theorist to determine what objects a
particular non-language-using creature is capable of thinking about. This is the
equivalent in the realm of thought of one of the principal tasks faced by the radical
interpreter confronting a completely alien language. Just as the radical interpreter
has to determine the ontology of the linguistic community with which he is dealing
in order to know what sort of things can be the semantic values of the expressions
of the language he is trying to interpret, so too must the developmental psycholo-
gist or cognitive ethologist determine how members of a particular species, or in-
fants at a particular stage of development, 'carve up the world' in order to deter-
mine how they think about it. In chapter 4 I suggest that we view particular
ontologies in terms of different types of sensitivity to higher-order structural in-
variants. I show how this provides us with a way of distinguishing the ontologies
of different types of nonlinguistic creature, looking in more detail at the philo-
sophical implications of some of the important work in developmental psychology
referred to at the beginning of the previous chapter.
The second stage in providing an epistemological basis is showing how com-
plete thoughts can be attributed to nonlinguistic creatures. In chapter 5 I show,
using a version of success semantics, how a semantics can be provided for nonlin-
guistic thoughts in a way that both does justice to philosophical constraints on ac-
ceptable theories of content and provides the ethologist or developmental psy-
chologist with a workable method of assigning contents to the beliefs and desires
of nonlinguistic creatures. Existing versions of success semantics have no way of
accommodating the intensional dimension of thought attributions. They do not fix
the complete content of a thought, but simply its truth-conditions. Chapter 5 shows
how success semantics can be supplemented in a way that permits psychological
attributions to take account of how truth conditions are apprehended by the thinker.
In the remainder of this book I do not engage directly with the language of
thought hypothesis. The approach I develop is compatible with certain ways of de-
veloping that hypothesis. Supporters of the language of thought hypothesis can
take the theory put forward in this book as an attempt to provide an epistemology
and semantics for the application of the language of thought hypothesis to nonlin-
guistic creatures. But nothing that I say in the chapters that follow presupposes the
truth of the language of thought hypothesis. The view I develope is that it is per-
fectly possible, and perhaps indeed desirable, to give an account of the episte-
mology and semantics of nonlinguistic thought without venturing into speculations
about cognitive architecture.
3
Minimalist Approaches
to Nonlinguistic Thought
The problem of explaining nonlinguistic thought arises out of the deeply rooted
practices of explaining the behavior of nonlinguistic creatures in psychological
terms. As I showed in section 1.1, developmental psychologists are becoming in-
creasingly happy to talk about young infants' knowledge of the world and their be-
liefssabout how objects behave, while cognitive ethologists explain many types off
behavior in the animal kingdom in terms of desires to bring about certain psycho-
logical states in others, together with beliefssabout how best to satisfy those desiress
in a given environment. Primates and other higher mammals engage in deception
behavior in which they are said to intend to manipulate the psychological states of
their conspecifics through exploiting what they know of their perspective on the
world. Early hominids are said to have a highly developed understanding of the
natural world and of the dynamic properties of various types of object that could
be used for tools.
It is natural to take descriptions such as these at face value, as attributing to
nonlinguistic creatures psychological states rather similar to those that we readily
attribute to each other on the basis of verbal and nonverbal behavior. If we do in-
deed take them at face value then there seems to be little difference between our
everyday practices of psychological explanation and the psychological explana-
tions that we might give of nonlinguistic creatures. Both can be viewed, simpli-
fying somewhat, as instances of belief-desire psychology (more accurately, of
prepositional attitude psychology). I showed in chapter 2, however, that thinking
about the thoughts of nonlinguistic creatures in these terms leads to a range of
difficulties associated with the problem of explaining how nonlinguistic creatures
can bear attitudes to thoughts. There are problems extending to nonlinguistic
creatures the conception of prepositional attitudes as attitudes to thoughts con-
strued as the senses of sentences. The language of thought hypothesis is more
suited to giving an account of the mechanics of nonlinguistic thought, provided
Minimalist Approaches 35
that nonlinguistic thought is taken to be a version of prepositional attitude psy-

chology—but the language of thought hypothesis runs into problems with the
epistemology of nonlinguistic thought. Both ways of taking psychological expla-
nations at the nonlinguistic level to be applications of propositional attitude psy-
chology are problematic.
Perhaps, though, we can avoid these difficulties by challenging the assumption
that belief-desire psychology is the right way to think about psychological expla-
nation at the nonlinguistic level. Perhaps it is a mistake to assimilate thinking at
the nonlinguistic level to thinking at the nonlinguistic level. Perhaps it is a mistake
to take talk of the knowledge, beliefs, desires, hopes, and intentions of nonlinguis-
tic creatures at face value. In this chapter I will consider an approach to nonlin-
guistic thinking that tries to drive a wedge between the type of thinking available
at the nonlinguistic level and the type of thinking that (on this approach) is the ex-
clusive preserve of language-users. According to the minimalist or deflationary
conception of thought without language, researchers in cognitive ethology, devel-
opmental psychology, and cognitive archaeology are absolutely right to describe
nonlinguistic creatures as thinkers. But the type of thinking engaged in by nonhu-
man animals, prelinguistic infants, and the early hominids must be distinguished
from the type of thinking engaged in by language-using creatures.
The minimalist approach is highly congenial to supporters of the Fregean ap-
proach and highly wncongenial to supports of the language of thought hypothesis.
As far as the Fregean approach is concerned, the minimalist approach promises a
way of accommodating the need for psychological explanations of the behavior of
nonlinguistic creatures without having to compromise the basic idea that the types
of thoughts that can feature in propositional attitudes should be understood as the
senses of sentences. As far as the language of thought hypothesis is concerned, the
minimalist approach threatens to undercut a key argument in support of the hy-
pothesis. It is crucial to the language of thought hypothesis that there should be
significant commonalities between linguistic thought and nonlinguistic thought.
This is the most plausible defense against the objection that participation in a pub-
lic language would be quite sufficient for the tasks for which the language of
thought is postulated. It is in many ways quite essential to the language of thought
hypothesis that thought at the nonlinguistic level should be essentially proposi-
tional in nature—or rather, that there should be an essentially propositional com-
ponent to nonlinguistic thought.
The minimalist approach has more sympathizers than systematic exponents, and
much of this chapter will be devoted to trying to expound it in as plausible and
noncontroversial a manner as possible. I start in section 3.1 by sketching out an in-
tuitive contrast between two types of thinking (thinking-how and thinking-that).
In section 3.2 I develop this intuitive contrast to set up the basic framework for a
minimalist account of nonlinguistic thought. Section 3.3 explains how a minimal-
ist might construe the practice of giving psychological explanations of the behav-
ior of nonlinguistic creatures. I offer three arguments against various different
components of the minimalist approach in sections 3.4 to 3.6.
3.1 Thinking-How and Thinking-That: An Intuitive Contrast

In thinking about the minimalist approach to nonlinguistic thought it is helpful
to start with a distinction between what I will call thinking-how and thinking-that,
by analogy with the philosophical distinction initially proposed by Gilbert Ryle
between prepositional knowing-that and nonpropositional knowing-how (Ryle
1949).1
Here are five examples of what I am calling thinking-how:
1. Imagistic reasoning, such as calculating whether a wardrobe will fit into

the space between a chest-of-drawers and the bed. It is plausible on both
introspective and cognitive scientific grounds that calculations of this sort
are carried out by exercising our visual imagination.
2. Empathetic reasoning. This is the sort of reasoning about how people will
behave that results not from applying theories to their behavior or forming
predictions on the basis of their past behavior, but rather from "putting
oneself in their shoes" and imagining how they will respond.
3. Trial and error reasoning. Trial and error reasoning is often driven by a
representation of the goal, but it does not seem to involve the formulation
of explicit hypotheses about how the goal in question is to be achieved.
4. Analogical reasoning. There is a fundamental difference between drawing
an explicit connection between two ideas or two situations and perceiving
an analogy between them. The first can usually be put in a prepositional
form. The second almost always resists being made explicit.
5. The reasoning involved in the exercise of complex bodily skills. The acqui-
sition and exercise of complex skills is a highly cognitive activity, requir-
ing precise calibration of different types of information. But the practical
knowledge that it involves cannot usually be expressed linguistically by
competent practitioners.
This list is not intended to be exhaustive, but it is certainly representative of a broad

category of cognitive ability that has two principal distinguishing characteristics.
In the first place, the various types of thinking-how are most naturally described
as abilities and skills that allow one to modify and adjust oneself to the environ-
ment—as opposed to bodies of information that one acquires about the environ-
ment. They result most typically in responses, rather than in an abstract proposi-
tion that one might take, for example, as the conclusion of an argument. The
responses that emerge need not always be behavioral in a narrow sense. They can
be highly cognitive—as when, for example, one uses empathetic reasoning to ad-
just one's social responses on the basis of how one predicts another person will be-
have. But thinking-how does not seem to take a classical inferential form. Empa-
thetic reasoning and analogical reasoning are forms of pattern recognition that rest
on perceived similarities between ideas and states of affairs, while imagistic rea-
soning seems to involve the manipulation of spatial images and the exercise of vi-
sual imagination. It does not seem right, at least on introspective grounds, to model
the process of thinking-how in terms of inferences carried out on propositionally

encoded representations of the environment.2
The second characteristic of thinking-how is closely connected to the first. The
process of thinking-how cannot easily, if at all, be put into words. Of course, I can
express in language the thought that the wardrobe will fit into the available space
or that the man walking in front of me will swerve to the left, but I cannot express
in language the thinking that leads up to that conclusion. It does not seem right to
describe the processes of thinking-how as defined over states with determinate
contents. The distinction here is in a sense grammatical (and directly analogous to
the distinction initially pointed out by Ryle). The "thinking-how" construction is
naturally completed by an infinitive identifying an activity or process, in contrast
to the "thinking-mat" construction, which requires completion by a sentence after
the "that." The sentence following the "that—" picks out a specific content or
proposition as that which is actually being thought. There is no such content in the
various examples of thinking-how.
Let me try to make the contrast between thinking-how and thinking-that vivid
with a concrete example. It is known that orangutans are fairly successful at track-
ing the larger of two quantities of liquid when the quantities are subjected to a va-
riety of physical transformations (Call and Rochat 1996, 1997). For example, if
presented with two quantities of water of different size that are then poured into
differently shaped containers, they will reliably choose the new container with the
larger quantity of water. How are we to understand the cognitive abilities mani-
fested in successful performance of liquid conservation tasks such as these?
One might explain their success in terms of the practical application of a com-
plex form of prepositional knowledge—an intuitive fluid mechanics containing
certain principles about the relation between volumes of liquid and the height and
width of containers, together with principles about how liquids are conserved over
transformations. These principles would be verbalizable (although not of course by
the orangutans) and would have determinate contents. An intuitive fluid mechanics
thus construed would be on a par with the broadly theoretical naive physics, naive
biology, or naive psychology identified by contemporary researchers in young
children, early hominids, and various sectors of the animal kingdom (particularly
the great apes). It would be understood in terms of tacit mastery of a set of general
principles that are domain-specific and only integrated to a limited extent with
other bodies of information and naive theories.
Alternatively, one might seek to explain the abilities in terms of complex
perceptual skills. These perceptual skills might be either static or dynamic—or in-
deed both. They might, first, involve a perceptual sensitivity to the actual transfor-
mation and a corresponding ability to track a particular quantity over transforma-
tions. That would be a form of dynamic perceptual sensitivity. Alternatively,
the orangutans' abilities might be explained in terms of a static form of perceptual
sensitivity—that is, sensitivity to the correlations between quantity of liquid and
the height and width of containers, correlations that allow two or more differently
shaped containers to be compared. The experimental evidence produced by Call
and Rochat seems to tell convincingly in favor of the static view—what they call
the "direct perceptual estimation strategy." But for present purposes the important
point is that understanding these cognitive skills as a form of perceptual sensitivity
gives us an excellent illustration of nonverbalizable thinking-how involving the
manipulation of spatial images and the exercise of visual imagination rather than
anything classifiable as an inference.
The distinction between thinking-how and thinking-that has been drawn at a
fairly intuitive level. But it should nonetheless be clear how it bears on the issues
raised in the first two chapters. The two approaches to the nature of thought con-
sidered in the previous chapter are conceptions of how we should understand the
nature of thinking-that. Correlatively, the problems we identified in extending each
of those approaches to apparent instances of thinking behavior in nonlinguistic
creatures arise only on the assumption that those thinking behaviors are instances
of thinking-that, rather than thinking-how. Should it turn out that we could inter-
pret those behaviors as instances of thinking-how rather than thinking-that then
those problems would cease to apply. This is, in essence, the proposal of minimal-
ist approaches to nonlinguistic thought, as I will show in more detail in the next
section.
3.2 Minimalist Approaches to Nonlinguistic Thought

These approaches characterize nonlinguistic thought as a form of thinking-how.
Whereas the two approaches canvassed in the previous chapter see the essence of
thought in terms of taking attitudes to propositions, minimalist approaches adopt a
different approach to nonlinguistic thought, taking it to be fundamentally non-
propositional in nature. As remarked earlier, there has been relatively little explicit
working out of alternatives to the propositional approach, either as an approach to
thought in general (but see Cussins 1990 and 1992 for pioneering work in this di-
rection) or as an approach to nonlinguistic thought in particular. I propose to pro-
ceed, therefore, by considering some of the key elements of the propositional ap-
proach to thought and indicating how a theorist might develop a nonpropositional
alternative in each of these dimensions. What will emerge, I hope, is a picture of a
relatively broad (though largely unoccupied) region of conceptual space. The dif-
ferent claims that can be identified within this general region are compatible
but are nonetheless not all mutually entailing. In the final sections of the chapter
I present a range of arguments against the different strands of the minimalist
conception.3
Let us start, then, by reiterating some of the key features of propositional think-
ing. Recall that the standard model of the propositional attitudes imposes a sharp
distinction between attitude and content—between the particular proposition that
is the object of one's mental state and the mental attitude one takes toward it. This
is a distinction that goes back to the beginnings of philosophical logic and the phi-
losophy of language in Gottlob Frege's Begriffsschrift, and it is a distinction thatn thatt
appears to be completely indispensable to a proper account of thought and the
mind. For one thing, as I have already shown, much of our understanding of our-
selves and others rests on our being able to make sense of the idea of a particular
person being able to take different attitudes to the same proposition, as indeed of
different people being able to take different attitudes to the same proposition.
Without this it is hard to see how we could make sense of either the fundamental
continuities holding across a person's life or basic disagreements between people.
Equally significant, an important element in thinking involves entertaining propo-
sitions without taking any sort of attitude toward them. The most obvious example
occurs in conditional thought, which plays such a central role in practical decision-
making. I might, for example, believe that if A is the case then I ought to 0, without
actually believing that A is the case. Of course, the conditional "If A then I ought to
0" might itself feature as part of a modus ponens inference, combining with the
premise that A is the case to yield the conclusion that I ought to 0—in which case I
would be believing that A is the case. But, on the other hand, the conditional might
equally feature in a modus tollens inference, with the rejection of the proposition
that I ought to 0 leading to the rejection of the proposition that A is the case. In this
case I would be entertaining the thought that A is the case without ever believing it.
A second key element of prepositional thinking is that propositions should be
independent of the particular context of thinking. That is to say, it should be possi-
ble to grasp a proposition both without knowing its truth value and without any
contact with the state of affairs that proposition is about.4 The first is one of the
factors that make it possible to take a range of different attitudes to the same
proposition. Some attitudes, such as the attitude of hope for example, seem in
many cases only to be applicable when one does not know the truth-value of that
which one hopes to be the case. As soon as one knows that something either is the
case or is not the case it is no longer appropriate to hope that it is the case. The
same might be said of fear. Many instances of fear involve fear that some state of
affairs might hold, an attitude that ceases to be available if one has determined ei-
ther that the state of affairs does hold or that it does not. The second is what makes
it possible to think not simply about distant regions of space-time but also what
one is not directly presented with.
The third characteristic of prepositional thought is that, for any prepositional at-
titude, it should in principle be possible to find a sentence that will specify the con-
tent of the attitude accurately and without remainder.5 Prepositional attitudes have
contents that are linguistically expressible and that, moreover, have nothing more
to them than can be linguistically expressed. This is closely related to the fourth
characteristic, which is that the contents of prepositional attitudes have a structure
that permits them to feature in inference and that is in some sense isomorphic
to one way of apprehending the structure of the state of affairs that that content
represents.
It will come as no surprise that these four characteristics of propositional
thought map very closely onto the four dimensions in which we have identified the
most serious potential difficulties for the project of explaining the thinking of non-
language-using creatures. Minimalist conceptions of nonlinguistic thought attempt
to account for nonlinguistic thinking in ways that do not involve some or all of
these characteristics of prepositional thought and hence that, if successful, will
allow us to avoid some or all of the related difficulties. Admittedly, as suggested
earlier, it is possible for a minimalist to reject certain elements of the prepositional
conception while retaining others. The four characteristics fall naturally into two
pairs—the first two have to do, broadly speaking, with the relation between con-
tent and context and the second two have more to do with the precise character of
content. So a theorist might, for example, maintain that nonlinguistic thoughts are
not linguistically expressible and lack combinatorial structure, while allowing that
such thoughts can both support the distinction between content and force and be
independent from the context of thinking.6 It should be recognized, however, that
it will be incumbent on such a theorist to give positive accounts of the content-
force distinction and the context-independence of nonlinguistic thought—which
will raise a good proportion of the difficulties raised in the previous chapter.
In thinking about the difference between propositional and nonpropositional
thought, hence about how a minimalist approach might be developed, an obvious
place to start is with the differences between propositional attitudes and percep-
tion. Although some philosophers (most prominently Armstrong 1961) have tried
to assimilate perception to the propositional attitudes by analyzing perception in
terms of dispositions to acquire beliefs, many philosophers take the view that per-
ceptions represent the world in a manner fundamentally different from the way
propositional attitudes represent the world (see, for example, the contributors to
Crane 1992). Perceptual content is not, on this view, a species of propositional atti-
tude content. Theorists in this area tend to stress three features of perceptual con-
tent that militate against considering it on the propositional model.
First, the content of perception seems to be analogue in nature, unlike the con-
ceptual content of propositional attitudes, which is more plausibly seen as digital.
The distinction between analogue and digital representations has (for my pur-
poses) been most perspicuously put by Dretske (1981, ch. 6). Let us take a particu-
lar fact or state of affairs, say the fact or state of affairs that some object s has prop-
erty F. A representation carries the information that s is F in digital form if and
only if it carries no further information about s other than that it is F (and whatever
further facts about it are entailed by the fact that it is F). But whenever a represen-
tation carries the information that s is F in analogue form it always carries addi-
tional information about s. It is plausible that perceptual states represent the world
in analogue form and propositional attitudes in digital form.
Second, the content of perception seems to be unit free (Peacocke 1986). If I
perceptually represent an object as being a certain distance from me I do not usu-
ally represent that distance in terms of a particular unit (in inches, say, as opposed
to centimeters), even though what I represent is a perfectly determinate distance. I
simply represent it as being that distance, where the content of my perception
specifies the distance. Propositional attitudes, however, can only represent dis-
tances (and other comparable quantities) in terms of specific units.
Third, the content of perception is more fine-grained than the content of proposi-
tional attitudes. I can see far more colors than I can name, and discriminate far
more shapes than I have concepts for. My belief that the grass is green has a single
content—but would be the appropriate response to an enormous variety of percep-
tual states.
The debate about whether the content of perception is conceptual or nonconcep-
tual centers around whether a conceptual account can be given of these aspects of
perceptual content (see, for example, Peacocke 1992, McDowell 1994).7 For pre-
sent purposes, these three features of the phenomenology of perception suggest
that the content of perception may be a good place to look for nonpropositional
ways of representing the world. Certainly, of the examples of thinking-how dis-
cussed in the previous section several are explicitly perceptual in form, involving
either occurrent perception (analogical reasoning and trial and error reasoning) or
the exercise of visual imagination (image-based reasoning).
Recall that the first feature of prepositional thought is that it permits a sharp dis-
tinction between content and attitude. Within a prepositional attitude it is possible
to distinguish the particular content from the attitude that is taken to that content,
so that it is possible for different people to take different attitudes to the same con-
tent and the same person to take different attitudes to the same content over time
(and indeed at a time). This sharp distinction between attitude and content is not
available in the case of perception. Partly this is because the content of perception
is perspectival in a way that the content of propositional attitudes (even indexical
prepositional attitudes) simply is not. The way that perception represents the world
is so closely tied to the perceiver's particular point of view that it is difficult to see
how the content of perception could be transferable and shareable in the manner
required by the attitude/content distinction. Perceptions typically feed into the
propositional attitudes by causing perceptual beliefs whose content bears complex
relations to the content of the perceptions on which they are based. Perceptual be-
liefs have contents that abstract away from the perspectival nature of perception to
"repackage" the information in a (digital) form that will permit integration with the
main body of the propositional attitude system.
The second feature of propositional thought is that the proposition in question
should be available to be thought independently of the context of thinking. So, for
example, it must be possible to entertain the thought without assigning it a truth-
value or having any epistemic contact with the state of affairs that would make it
true. The content of perception does not have this type of context-dependence.
There is no analogue in the case of perception for entertaining a thought without
considering whether it is true or false. Of course, we do not always believe what
we perceive, and we sometimes find ourselves in situations where we do not en-
dorse the way perception presents the world as being. Visual illusions are an obvi-
ous example. When we know that the two lines in the Muller-Lyer illusion are in
fact the same length we no longer endorse the perceptual appearance that one is
longer than the other. But it would be wrong to describe this as saying that we in
some sense take an attitude of suspended belief to the perceptual appearance that
one is longer than the other. What happens in such cases is that our perceptual rep-
reservation of the world and our beliefs about the world come apart. The normal
processes by which perceptions lead to beliefs have been overridden by our knowl-
edge of how the illusion works. Nor, of course, is it possible to perceive a state of
affairs with which one is not in any direct sort of epistemic contact. Perceiving is
factive because, as McDowell and others have pointed out, it is a matter of how the
world presents itself to the perceiver.
The third and fourth features of prepositional thinking need to be taken together.
The content of perception cannot be completely and accurately reported in a sen-
tence following a "that—" clause of the type that could give the content of a
propositional attitude, for reasons discussed earlier in the context of the phenome-
nology of perception. Any such formulation of the content of perception would be
compatible with indefinitely many fine-grained perceptual contents. What is speci-
fied by such sentences is the content of the belief to which the perception would
give rise if it was taken at face value (in the way that it is not taken at face value,
for example, in cases of visual illusion where one is aware of the illusion).8 Of
course, it may well be the case that the content of perception can in some way be
specified by a suitably complex sentence—perhaps a sentence that specifies the
perceptible characteristics of every discriminable point in the perceptual field for a
given modality, in the way that the content of a television image might be given by
a sentence that specifies what is going on in each pixel. Let us call this a pixeliza-
tion of the content of perception. It is no part of the present argument to claim that
this is impossible. My point is rather that such a pixelization would not yield a
specification of the right sontof content, where the right sort of content is one thatt
can feature in inferences and can be integrated with the content of propositional at-
titudes. A pixelization has no structure. It is not formulated at a level of description
that captures even the bounded edges and segmentation of the perceptual array.
There is considerable plausibility, then, in the idea that we can develop the intu-
itive contrast between thinking-how and thinking-that into a minimalist account of
nonlinguistic thought by starting out from the nonpropositional content of percep-
tion. In the remainder of this section I will consider how this general strategy
might be implemented. We can begin with Michael Dummett's account of the
types of thought available to nonlinguistic creatures—what he calls protothoughts.
Dummett suggests that the protothoughts of nonlinguistic creatures have a
very circumscribed applicability. They can only occur as integrated with current
activity:
Proto-thought is distinguished from full-fledged thought, as engaged in by human be-

ings for whom language is its vehicle, by its incapacity for detachment from present
activity and circumstances. A human being may be suddenly struck by a thought,
which might be the key to the solution of a mathematical problem or the fact that he
has left some vital document at home: in the latter case he may turn around and go
back for it. Our thoughts may float free of the environment: we may follow a train of
thought quite irrelevant to our surroundings or what we are engaged in doing. An ani-
mal may solve quite complex problems, by a process of thinking about the solution,
as was illustrated by Kohler's chimpanzees, or by the pony which, when confronted

by a cattle-grid, lay down and rolled over it: but its thought, or more exactly, proto-
thought, cannot float free but can occur only as integrated with current activity.
(Dummett 1993,123)
Dummett's emphasis on the context-dependence of protothoughts is very much in

line with the earlier characterization of nonpropositional thinking. So too is his
proposal that protothoughts are essentially tied to the possibilities the environment
affords for action.
The sublinguistic level of proto-thought is essentially spatial, and therefore must be

conceived as operating in our apprehension of what we perceive as having a three-
dimensional shape and occupying a three-dimensional position. But it is also essen-
tially dynamic: it involves the apprehension of the possibilities and probabilities of
movement, and of the effect of impact. For this reason, it incorporates, not merely
perception of position, shape and movement, but also recognition of the gross proper-
ties of material things. It is an immediate feature of even our visual perceptions that
we observe objects as differentiated according to the general type of material of
which they consist: whether they are rigid or flexible, elastic, brittle or plastic, cohe-
sive like a lump of sugar or a heap of grains like caster sugar, solid, liquid or gaseous,
wet or dry, smooth or rough, greasy or clean, and so forth. The reason that we use vi-
sual clues to project these properties, even though unaided vision does not disclose
them, is precisely that they bear on the dynamic possibilities. (Dummett 1993,124)
As one would expect, given their essentially pragmatic and context-bound nature,
the vehicles of protothoughts are much closer to perceptual states than they are to
linguistically expressible propositions. According to Dummett, the vehicles of pro-
tothoughts are "spatial images superimposed on spatial perceptions" (Dummett
1993,123). In perceiving the ambient environment protothinkers visualize the pos-
sible ways it might be transformed, drawing on motor memories and a sense of
their own possibilities for action and reaction.
Protothinkers do not come to a judgment about what the environment contains
or the possibilities it affords, where coming to a judgment implies something that
can be detached from the here-and-now, but nonetheless they perceive the environ-
ment in a way that involves exercising judgment. The parallel to which Dummett
often returns is with the type of thinking involved in skilled behavior in humans:
A car driver or canoeist may have rapidly to estimate the speed and direction of on-
coming cars or boats and their probable trajectory, consider what avoiding action to
take, and so on: it is natural to say that he is engaged in highly concentrated thought.
But the vehicle of such thoughts is certainly not language: it should be said, I think,
to consist in visual imagination superimposed on the visually perceived scene. It is
not just that these thoughts are not in fact framed in words: it is that they do not have
the structure of verbally expressed thoughts. But they deserve the name of "proto-
thoughts" because, while it would be ponderous to speak of truth or falsity in applica-
tion to them, they are intrinsically connected with the possibility of their being mis-
taken: judgment, in a non-technical sense, is just what the driver and the canoeist
need to exercise. (Dummett 1993,122)
Protothoughts can be correct or incorrect. Unlike the correctness conditions of lin-

guistically expressed thoughts, however, their correctness conditions are ulti-
mately to be understood in terms of the success or failure of the actions to which
they give rise.
Dummett does not go into much detail about how the content of protothoughts
should be understood. His talk of "visual imagination superimposed on the visu-
ally perceived scene" and "spatial images superimposed on spatial perceptions"
make clear that he is intending protothoughts to be understood in perceptual terms,
but he does not offer any detailed suggestions as to how the content of pro-
tothoughts should be understood. Adrian Cussins has offered some very interesting
suggestions in this area (Cussins 1990, 1992). Although Cussins and Dummett are
developing very different (and in important respects incompatible) projects, some
of the specific points that Cussins makes about how we can understand experien-
tial content can be seen as deepening Dummett's account of protothoughts.
The central feature of Cussins's account of experiential content is that it should
be understood not in terms of notions such as truth and truth-condition, but rather
in terms of the organism's abilities to act on the perceived environment. What the
organism perceives (the content of its perception) is a distal environment struc-
tured in terms of the possibilities it affords for action.
Contents can be canonically specified with reference to abilities of the organism,

where the abilities are not (or need not be) part of the realm of reference. The realm
of reference is that with respect to which the correctness (e.g. the truth-value) of the
content is determined. Evidently, a subject in thinking of a coffee mug that is full—a
thought which is sustained by the subject's perceiving the mug—is not referring too
abilities to grasp the mug, to track it as it moves, or to be selectively sensitive to
changes in its appearance. Nevertheless, Evans's idea was that the cognitive signifi-
cance of the singular mug content could be captured, and could only be captured, by
the theorist's referring to abilities to grasp the mug or otherwise locate it, to track the
mug through space and time, and to be selectively sensitive (in judgment and action
and memory) to changes in the mug's features. These abilities are not available to the
subject as the content's referent, but they are available to the subject as the subject's
experience-based knowledge of how to act on the object, and respond to it (Cussins
1992, 655)
This conception of an ability-based conception of content offers a very natural way

of developing the idea that protothoughts are context-bound and relativized to the
organism's capacities for action and reaction. On Cussins's view, moreover, this
yields a sharp distinction between the success-governed level of protothoughts and
the truth-governed level of full-fledged thought.9 Protothought does not involve
experience of an objective world.
An objective world is given to a subject if the content presents something as being in-
dependent of the subject's particular abilities and particular location, in space and
time. But given only the realm-of-embodiment-specified content, all the subject (em-
bodied organism) has is an experiential awareness of how to move etc in response to
local changes in its environment. If this were a subject's conception of a referent it
would be a conception of something as not independent of contingent characteristics

of the subject itself. The necessary separation between subject and object would not
have been achieved. (659)
The environment of the protothinker is composed of superimposed features and af-

fordances (see further hereafter) organized in terms of the protothinker's pragmatic
understanding of the environment—what Cussins terms cognitive trails.
A similar contrast is drawn by John Campbell, who distinguishes between two
fundamentally different perspectives on the world (1993, 1994). One perspective,
which he describes in terms of its employment of a primitive physics, is primarily
practical and engaged. The other, which uses an explicit physics, is reflective and
detached. A clearer understanding of what this distinction is emerges from Camp-
bell's discussion of the terms that one might employ in specifying the content of a
practical primitive physics. These terms are, he explains, causally indexical. What
makes terms such as "now" and "here" indexical is the fact that their reference
varies systematically in different contexts, according to the spatiotemporal loca-
tion of the thinker. What makes causally indexical terms indexical is the fact that
their reference varies according to the causal powers of the thinker. Examples are
". . . is too heavy to lift" or ". . . is out of reach." These terms are grasped in
terms of their immediate implications for action and perception: "In these cases, it
is not just that grasp of the term requires the ability to register when it applies. It is
rather that one uses one's grasp of the causal significance of the term in reacting to
recognition that it applies" (1994, 45). So, for example, one would just give up
one's attempts to lift the object in question. A primitive physics, then, is one whose
content is to be specified in causally indexical terms. The content of an explicit
physics, in contrast, is to be specified in terms that are somehow independent of
the practical concerns of the thinker. Campbell describes them in a variety of ways,
as 'disengaged,' 'reflective,' and 'theoretical,' As in the comparable distinctions
drawn by Dummett and Cussins, the level of causal indexicality is governed by the
norms of successful action, as opposed to the norms of truth.
All three authors offer a broadly similar way of understanding nonlinguistic
thought—as fundamentally perceptual in form, essentially pragmatic and context-
dependent, and as falling short in various respects of the full-fledged representa-
tion of objects. On all three accounts the contents of nonlinguistic thought qualify
as nonpropositional in the terms identified earlier and seem far more easily assimi-
lated to what in the previous section I identified as thinking-how. We can, I think,
take all three philosophers to exemplify a minimalist approach that stresses the fol-
lowing four features of nonlinguistic thought.
• Context-bound
• Pragmatic and dynamic
• Perceptually vehicled
• Unstructured
I shall argue hereafter that the minimalist approach cannot be fully adequate to the
task of explaining nonlinguistic thought. In the next section, however, I shall ex-
plain how psychological explanation is to be understood on the minimalist ap-

proach, bringing out how the minimalist approach offers an alternative to proposi-
tional attitude psychology.
3.3 The Minimalist Construal of Intentional Action and Explanation

In order to understand the type of account a minimalist might give of psychologi-
cal explanation we need to be clearer on the general structure of propositional atti-
tude explanation when we are dealing with language-using creatures. In brief, the
aim of such an explanation is to present a combination of beliefs and desires that
will make it intelligible why the action in question should have occurred. The ac-
tion becomes intelligible when any rational agent with those beliefs and desires
and in comparable background conditions could be expected to act in the same
way.
This conception of psychological explanation goes hand in hand with a particu-
lar conception of how intentional actions are generated. There are generally held to
be three separable components in the generation of an intentional action. The basic
motor of an intentional action is a desire (whether a desire for something or a de-
sire that something be the case.) But a desire alone is insufficient to bring about an
action. Desires feed into action when conjoined with instrumental beliefs pointing
to how those desires might be satisfied. These instrumental beliefs themselves
need to be "anchored" in beliefs about the environment (as well as depending on
further background beliefs). Correlatively, then, it is widely held that a satisfactory
psychological explanation will have components corresponding to each of these
three components.
This three-component model yields the standard template of belief-desire expla-
nation, as applicable to language-using creatures. It is almost universal among
philosophers to think that all social interactions are governed by psychological ex-
planations taking this general form—that we are only able to navigate the social
world because we constantly deploy belief-desire explanations to make sense of
what is going on around us. Nothing that I say depends on this claim about the
scope of psychological explanation, and indeed I have argued against it elsewhere
(Bermudez forthcoming-a, forthcoming-d; see also Morton 2002). As far as ordi-
nary social interaction between humans is concerned, I am convinced that rela-
tively little of it is governed by any sort of folk psychological explanation, as op-
posed, for example, to understandings of social roles, social routines, and the
direct perception of emotional states. Psychological explanation and prediction
come into play only when our ordinary methods of understanding other people and
coordinating our behavior with theirs break down.
For present purposes, however, we are interested not in the extent to which
belief-desire psychology controls social interactions among language-users such
as ourselves, but rather with its applicability to nonlinguistic creatures. Are the
psychological explanations given by cognitive ethologists, developmental psy-
chologists, and cognitive archeologists best viewed as forms of belief-desire expla-
nation, or should they viewed in another way? The distinctiveness of the minimal-
ist approach lies effectively in denying that there is any need to attribute beliefs
when explaining the behavior of nonlinguistic creatures. We can, so it is claimed,
understand the behavior of nonlinguistic creatures solely in terms of the goal-
directed exercise of visual imagination.
The minimalist is likely to start by pointing out that the relevant instrumental in-
formation is part of the content of one's current perception of the environment, so
that there is no need for an instrumental belief. If, for example, my desire is for a
drink of water and I see a glass full of a liquid that looks like water in front of me
well within arm's reach, then my reaching out toward the glass will not always de-
pend on a separate instrumental belief to the effect that I will be able to obtain the
glass if I reach out for it. Often I will just be able to see that the glass is within
reach and act accordingly. The possibilities for action are part of what is perceived,
as is stressed in Campbell's notion of causally indexical comprehension and in
Cussins's ability-based specifications of content.
The thesis, associated with J. J. Gibson and the ecological approach to visual
perception, that the content of visual perception includes what Gibson termed af-
fordances offers a further way of extending this basic idea (Bermudez 1998, Gib-
son 1979). An affordance is a resource or support that the environment offers a
particular creature—such as the possibility of shelter or the availability of food.
Although affordances are relativized to particular species, so that the same region
of the environment might offer different affordances to different species, they are
nonetheless objective features of the environment and exist as a function of the
physical properties of the environment. The basic idea behind Gibson's theory of
affordances is that the environment is not perceived in neutral terms. What are per-
ceived are the possibilities that the environment affords for action and reaction, in-
cluding the potential of various locations for providing shelter, concealment, or
nourishment. These affordances are directly perceived in the patterns of light in the
optic flow10—although, of course, creatures need to become "attuned" to the rele-
vant features of the environment. The perception of affordances is a perception of
the instrumental properties of the environment that, the minimalist is likely to
claim, effectively makes the appeal to an instrumental belief nugatory. The ex-
planatory framework of affordances is not compulsory for the supporter of the
minimalist approach to nonlinguistic thought, but it does offer a powerful tool for
making sense of instrumental perception.
So the minimalist account of the intentional actions of nonlinguistic creatures
will see these actions as being brought about by a simple combination of a percep-
tual state and a desire. The animal or infant whose behavior is being explained me-
diately perceives an instrumental property that would allow it to attain a goal—it
perceives, say, not simply the lever but the fact that the lever pressing affords the
possibility of gaining food. Dummett's phrase "the superposition of spatial images
on spatial perceptions" is an apt way of capturing what is going on here. Given that
the animal desires the goal in question, it acts on the mediately perceived instru-
mental property. The animal is hungry and therefore presses the lever.
The minimalist account will be incomplete, however, without some character-

ization of what it is for a nonlinguistic creature to be in a given motivational state.
We should start by noting that motivational states in general can have two different
types of object. That is, one can desire a particular thing, or one can desire that a
particular state of affairs be the case. We can label this the distinction between
goal-desires and situation-desires (with a corresponding distinction between goals
and goal-situations). At the level of verbalizable thought, the distinction can be
marked in terms of two different ways of completing the sentence "X desires —."
When the sentence ascribes a goal-desire it is completed by the name of an object
or by the name of a kind of stuff (e.g., "food"). But when the sentences ascribes a
situation-desire, the sentence is completed by a "that—" clause in which a com-
plete sentence specifies the state of affairs in question.
Goal-desires are more basic than situation-desires. There are many types of
goal-directed behavior that can be interpreted in terms of goal-desires but where it
would be inappropriate to look for a situation-desire. Consider, for example, a bird
returning to a stored cache of seeds. It is known from both ethological studies and
laboratory experiments that species such as chickadees and marsh tits are capable
of hiding extraordinary numbers of seeds in a range of different hiding places and
then retrieving them after considerable periods of time have elapsed (Sherry 1982,
1984). This seems to be goal-directed behavior. It is hard to see, for example, how
it can be attributed to the operation of innate releasing mechanisms, trial and error
learning, or reinforcement learning. The behavior of a marsh tit returning to one of
its many food caches is driven by the combination of a motivational state and a
memory of the particular location and how to get to it.11 But it seems wrong to
hold that what drives the marsh tit is a desire for the food that is in the cache
(which would be a situation-desire). It is driven by a far more general desire for
food. There is no difference, for example, between the desire that drives it to go to
one food cache and the desire that drives it to go to a different one. Nor need it be a
desire for any particular one of the different types of seed or nut from which it de-
rives nourishment. (But see Clayton and Dickinson 1999 on scrub jays.)
Goals are not structured, whereas goal-situations are. Goals are simple ob-
jects, features, or kind of stuff. Qua goals they do not contain distinct parts. Goal-
situations, in contrast, are always complexes, such as facts or states of affairs,
where a complex is an ordered n-tuple of objects, properties, and/or relations. The
simplest sorts of goal-situations are objects possessing particular properties (X de-
sires that it be the case that such-and-such is so-and-so) or objects standing in rela-
tions (X desires that ? stand in relation R to ?).
It is open to the minimalist, therefore, to adopt a very low-key approach to psy-
chological explanation. The minimalist might suggest that a creature's behavior
could be fully explained in terms of a goal-desire and the content of its occurrent
perceptions. This means that all that is really required for a satisfactory psycho-
logical explanation from the minimalist point of view is a specification of the rele-
vant goal-desire. Since one can expect the creature to be suitably perceptually sen-
sitive to its distal environment and the possibilities the environment presents for
action and reaction, there is no need to specify the content of the creature's percep-
tual state. The simple fact that the creature acted in the way it did is a sign that it
perceived the relevant instrumental property. The minimalist will be in a position,
therefore, to work back from the attained (or striven-for) goal to the relevant de-
sire. Once the desire is specified the explanation is complete. What makes the in-
tentional actions of nonlinguistic creatures intentional, on this view, is that they are
minimally goal-directed. Consequently, all an explanation needs to do is to find the
goal. The differences between the minimalist conception and standard belief-
desire models of psychological explanation should be clear.12
It will be helpful to compare the minimalist account as I have developed it with
two earlier accounts of psychological explanation explicitly directed at the project
of explaining the behavior of nonlinguistic creatures and equally intended to be al-
ternatives to standard belief-desire psychology. Charles Taylor has suggested that
the alternative to mechanistic explanation (in the form, say, of appeals to classical
conditioning or innate releasing mechanisms) is what he calls teleological or pur-
posive explanation, the characteristic of which is that events are held to occur
because of what results from them. Here is what Taylor says about the form of a
teleological explanation.
To say that the behavior of a given system should be explained in terms of purpose is,
in part, to make an assertion about the form of the laws, or the type of laws that will
hold of the system. But qua teleological these laws will not be of the kind which
makes behavior a function of the state of some unobservable entity; rather the behav-
ior is a function of the state of the system and (in the case of animate organisms) its
environment; but the relevant feature of system and environment on which behavior
depends will be what the condition of both makes necessary if the end concerned is to
be realized. Thus for instance, we can say that the conditions for a given action, say a
predator stalking his prey, are (1) that the animal be hungry and (2) that this be the
"required" action, i.e. the action in his repertoire which will achieve the result—
catching his next meal. The condition of an event B occurring is, then, not a certain
state ofP [an inner purpose], but that the state of the system S and the environment E
be such that B is required for the end G, by which the system's purpose is defined.
(Taylor 1964, 9-10; my italics)
Taylor is at great pains to show that purposive explanations can be formulated in

ways that do not leave them open to the sorts of criticisms that made behaviorism
seem plausible. But his urge to keep teleological explanations at the level of the
observable and verifiable leaves him with a very impoverished conception of psy-
chological explanation—one that can only be distinguished with difficulty from
certain forms of behaviorism. It is mildly troubling that there is no account of what
makes it the case that a system has a certain end as its goal. It also seems wrong to
demand that a behavior be required to attain the relevant goal. The behavior must
be (perceived to be) a sufficient condition of attaining the end, but it seems unrea-
sonable to demand that it be a necessary condition. There might, after all, be differ-
ent ways of attaining the relevant end in a given environmental situation.
But the real problem is that Taylor's account does not explain how a creature
moves from being in a certain environmental situation to engaging in a particular

behavior. Even if we assume that a particular behavior is required (in those particu-
lar environmental circumstances) to achieve a goal that we have independent rea-
sons to believe the animal has, we still have no understanding of why that behavior
should have emerged at that moment. The animal will only embark on the behavior
if it somehow apprehends that a goal is realizable through that behavior. Or rather,
those behaviors in which we are interested (those for which we are compelled
to offer psychological rather than mechanistic explanations) are precisely those
in which the behavior is not environment driven—and it is precisely these non-
environment-driven behaviors for which Taylor's account leaves an explanatory
gap-
Jonathan Bennett has offered an extension of Taylor's theory that employs the
concept of registration to fill this explanatory gap (Bennett 1976). The basic form
of a Ideological law for Bennett is as follows:
(For all x) ((fix & x registers at time t that F/Gx) —* Fx at time t + d)
Rx is to be read as the claim that suitable enabling conditions are satisfied; F/Gx as
the claim that x is so structured that performing action F at a later time t + d is
causally sufficient for x's obtaining goal G at a still later time. The concept of reg-
istration is offered as a surrogate for perception—as a way of capturing, without
the theoretical baggage that comes with the notion of perception, how features of
the immediate environment might be epistemically accessible. According to Ben-
nett, a creature C registers that P just if C is in a sensory state that is sufficiently
like some P-operative state, where a P-operative state is one that implies that C's
acting on P is noncoincidental (Bennett 1976, 56). The basic idea is that C is per-
ceptually sensitive to P in a core set of environments and acts accordingly. The P-
operative states are those that resemble the sensory states underpinning C's per-
ceptual sensitivity in the core environments.
The concept of registration thus understood fills the gap in Taylor's account. It
gives us a way of understanding why the creature should embark on the relevant
behavior when it does—because it registers the contingency between acting in a
certain way and realizing a certain goal. Whereas on Taylor's account a psycho-
logical explanation should cite simply the relevant goal and the fact that the behav-
ior is a way of bringing about that goal, Bennett more plausibly demands that the
agent register this fact. It is also clear that Bennett's account dovetails very nicely
with the proposal that the instrumental component of intentional action be under-
stood in terms of the perception of affordances. The theory of affordances gives us
a way of understanding just what it would be for a creature to register that per-
forming action F at a later time t + d is causally sufficient for x's obtaining goal G
at a still later time.
Nonetheless, Bennett's basic account must be modified before it gives us an
adequate way of spelling out how a minimalist should understand psychological
explanations. It cannot generally be the case that a creature acts on every affor-
dance that it perceives—but this is clearly the consequence of the basic account.
We need some way of accommodating the thought that a creature might choose be-
tween different affordances as a function of the different goals that it has. Bennett
himself recognizes this and suggests that the basic account be supplemented with a
further clause in the antecedent specifying, essentially, that there is no F*/G* such
that the creature prefers F*/G* to FIG—where preferring F*/G* over F/G would
not simply be a matter of preferring the first outcome to the second, but rather in-
volve a comparison of what Bennett calls "means and goal complexes." Again, the
basic idea is very congenial to the theory of affordances. If the instrumental com-
ponent of an intentional action is in fact to be understood in terms of the perception
of an affordance then it seems absolutely right to hold that, when there is some-
thing like choice going on, the choice is between what we might term courses-
of-action-leading-to-goals, rather than between goals per se.
Let us suppose that all instances of thought without language of the sort identified
by developmental psychology and the other disciplines we are considering can in-
deed be accommodated according to the minimalist conception, as instances
of thinking-how rather than thinking-that. This would be a way of doing justice to
the widespread experimental evidence that nonlinguistic creatures are capable of
genuinely thinking behaviors, while still preserving the distinctiveness and indeed
uniqueness of thought with linguistic vehicles. Only language-using creatures
would be capable, on this view, of thinking that fits the specification of thinking-
that: compositionally structured thoughts with determinate contents would only be
available to creatures who possess a language.
Unfortunately, the minimalist approach of nonlinguistic thought is unsatisfac-
tory—or rather, it cannot provide a sufficient account of nonlinguistic thinking. In
the remaining sections of this chapter I will present three arguments directed at
various strands of the minimalist approach as I have sketched it out. From these ar-
guments there will emerge certain features of nonlinguistic thought that any ade-
quate theory must be able to accommodate. In particular I will argue for two the-
ses. The first thesis is that the instrumental component of intentional action cannot
always be understood at the level of perceptual content. As I will show, in at least
some instances, nonlinguistic creatures behave in ways that involve complicated
instrumental reasoning about the possible consequences of different courses of ac-
tion. This conclusion is argued for in section 3.4 in the argument from planning.
The second thesis is that the representational states of nonlinguistic creatures need
to be viewed as structured states. This conclusion is drawn in the argument from
intelligent action in section 3.5 and in the argument from psychological explana-
tion in section 3.6.
3.4 The Argument from Planning

The minimalist position, as outlined here, is committed to a relatively circum-
scribed conception of nonlinguistic thinking behavior. It is integral to the minimal-
ist position that a nonlinguistic creature's thoughts cannot be divorced from the
here-and-now of the perceived environment. In the words of Michael Dummett,

the thoughts of nonlinguistic creatures "cannot float free but can occur only as in-
tegrated with current activity." (Dummett 1993, 123). To say that protothoughts
can occur only as integrated with current activity is rather metaphorical, however.
It clearly cannot be right that a nonlinguistic creature can act only on a visible
goal—although the anecdotal examples that Dummett gives of Kohler's chimps
and a horse rolling over a cattle grid both happen to be ones in which the goal is
visible. Animals can clearly act on goals that they cannot perceive—the predator
lying in wait for prey is an obvious example. In section 3.3 I presented a model of
psychological explanation that tries to do justice to this while adhering to the basic
minimalist claim that the thoughts of nonlinguistic creatures are circumscribed by
their perceptions of the immediate environment.
The key point is that the content of perception can include factors that are not
themselves directly perceived. For present purposes it will suffice to understand a
creature C immediately perceiving an object O in terms of C standing in such a re-
lation to it as to enjoy an experience as of an O being present in the relevant sen-
sory modality, where an experience as of O being present is an experience that
would allow O to be discriminated from other elements in the perceptual field.13 In
the case of vision the holding of the relevant relation would obviously require that
O was within C's field of vision; that there be no obstructions between O and C;
that C's perceptual apparatus be functioning properly; that environmental condi-
tions be more or less normal; and so forth. It is often correct to describe a creature
as perceiving an object even when there can be no question of immediate percep-
tion. In an important group of cases it is possible to detect the presence of an object
that one cannot immediately perceive from the behavior of things that are immedi-
ately perceptible—as when a creature is alerted to the presence of a predator by the
behavior of conspecifics in a flock. In another important group of cases creatures
can perceive what Jonathan Bennett calls instrumental properties (that is to say, the
fact that the environment is such that a particular course of action will lead to the
satisfaction of a certain goal) even when the goal is not immediately perceptible.
The natural way to understand what I am calling noniinmediate perception is in in-
ferential terms—and this is certainly how it has been understood by many philoso-
phers who have considered the matter.14 Clearly, no defender of the minimalist
theory of nonlinguistic thought can accept such an interpretation of mediate per-
ception. It is hard to see how the notion of inference can be accommodated within
the minimalist framework, since it seems clearly to fall on the thinking-that rather
than thinking-how side of the distinction with which I began the chapter. But,
as I pointed out in section 3.3, Gibson's theory of the direct perception of affor-
dances gives us a good way of understanding how mediate perception might occur
noninferentially.
An illustration of this type of noninferential mediate perception of instrumental
properties can be found in some of the studies Marc Hauser has carried out on the
perception of the functional properties of objects in cotton-top tamarin monkeys
(Hauser 1997). The experiments involved offering the monkeys a series of choices
between two different tools, each of which could be used to access food. The tools
were all cane-like objects with hooks at the end, but they varied across a range of
dimensions. Some of these variations were functional. So, for example, in one ex-
periment the monkey had to choose between a cane with food located on the inside
of the hook (hence easily accessible) and a cane with the food less easily accessi-
ble on the outside of the hook. The tamarins quickly learned to select the tool
yielding the greater affordance of food. This choice pattern was preserved when
the properties of the two canes were systematically varied. Tools with alterations
in noninstrumental properties such as color or texture (which would not affect the
usefulness of the tool for securing the food) were systematically preferred over
tools with alterations in properties such as size or shape, which rendered the tool
less appropriate for securing the food. It looks very much as if the cotton-top
tamarins (who do not use tools in the wild) are learning to become perceptually
sensitive to instrumental properties/Gibsonian affordances.
These brief reflections illustrate how the minimalist is committed to the follow-
ing restrictive thesis about the nature of nonlinguistic thought. Whenever the
thinking behavior of a nonlinguistic creature is directed at something that is not
immediately perceptible, the instrumental representations driving its behavior can-
not be anything more complicated than mediate perceptions. This is a function of
the minimalist emphasis on the content of perception rather than on propositional
attitude content.
Clearly, then, we will have an argument against the minimalist interpretation of
nonlinguistic thought if we provide examples of behaviors in nonlinguistic crea-
tures that do not seem to be interpretable at the level of perceptual content, whether
that content is understood in terms of affordances, or some other form of noninfer-
ential mediate perception (which might, for example, involve the exercise of visual
imagination in reconfiguring the elements of the perceived environment). Of
course, no individual example can be definitive, but it seems to me that there are
enough broad areas in which it is plausible to identify such examples for the mini-
malist claim to be seriously threatened.
Let us start with the notion of an affordance. Although minimalist approaches
are not compelled to employ the notion of an affordance, it is a very natural way of
developing the basic minimalist claim that what a creature perceives are its possi-
bilities for action and reaction in the perceived environment. It is an important part
of the theory of affordances that affordances should reflect what is sometimes
called the "animal-environment fit." It is an ethological commonplace that particu-
lar species (indeed, particular communities within individual species, not to men-
tion individuals at different stages of development) occupy distinct environmental
niches. If explanations appealing to affordances are to be more than just-so stories
there must be a significant correlation between the affordances that a creature can
perceive and relatively stable features of its environment. The perception of affor-
dances is one of the ways that animals become attuned to their environmental
niches. It would seem, therefore, that explanations of behavior appealing to the
perception of affordances will be of limited use when the behavior in question oc-
curs in highly unfamiliar surroundings unrelated to the creature's environmental

niche. Yet much of the evidence for thinking in the absence of language is derived
from studying how nonhuman animals and human infants behave in precisely such
unfamiliar surroundings.
The dishabituation paradigm in developmental psychology is a case in point.
The basic reasoning behind dishabituation experiments is that infants look longer
at events that they find surprising and hence one can read back from what they find
surprising to the expectations that they have about how events should turn out. So
by habituating infants to a given event and then presenting them with events that
differ from that event in specified ways one can detect those features of events to
which they are sensitive. This allows one to identify the general higher-order
physical principles that the infants can detect perceptually. The dishabituation
paradigm has produced much of the evidence of cognitive sophistication that has
revolutionized the study of early infancy (Baillargeon 1995, Gopnik and Meltzoff
1997, Spelke 1990). But many, if not most, dishabituation experiments proceed by
showing infants events that appear to be physically impossible. In Baillargeon's
well-known drawbridge experiments, for example, infants are presented with a
state of affairs that seems to involve a solid drawbridge passing through a solid ob-
ject. What in the 3-month-old infant's environmental niche corresponds to such an
event?
The same holds for much of the laboratory-based research that has been done in
animal learning theory. There is a direct analogy with the infant case, given that
animal cognition is now being studied through versions of the dishabituation para-
digm (Mauser 1998, Munakata et al. 2001).15 Although the behavior of animals in
the wild, as studied by cognitive ethology, is very much suited to explanations of
behavior that stress the ways that animals resonate to features of their environ-
ment, it is much harder to offer genuine explanations of the behavior of animals in
highly artificial environments such as mazes and Skinner boxes. The extensive re-
search that has been carried out on trying to teach animals artificial communication
systems provides an even clearer example. Appeals to the perception of affor-
dances or to any type of mediate perception would seem to be completely vacuous
when trying to explain how a dolphin can learn to respond to sequentially complex
gestural commands (Herman 1980,1986) or how a pygmy chimpanzee can sponta-
neously start to use lexigrams to communicate after having watched experimenters
trying and failing to teach his mother to communicate with them (Greenfield and
Savage-Rumbaugh 1990, Savage-Rumbaugh, Shanker, and Taylor 1998).
Versions of the minimalist approach that depend on the perception of affor-
dances are difficult to apply, therefore, to types of thinking behavior that emerge in
highly unfamiliar environments.16 This is not, of course, an argument against those
versions of the minimalist approach that do not deploy the concept of an affor-
dance. But there is a second problem—one that is not tied to the notion of an affor-
dance. Many types of animal behavior involve acting on goals that are not even
mediately perceptible. Many of these behaviors are not thinking behaviors at all.
So, for example, the minimalist should not be troubled by the fact that birds build
nests to shelter young that have not yet been born or that individual salmon engage
in large-scale migrations. But some are thinking behaviors. Technological manipu-
lation in the higher primates and in the early hominids provides good examples.
Wild chimpanzees, for example, make two different types of wands for dipping
into ant and termite nests from different types of branches. Wands for dipping into
ant swarms are made by stripping the side leaves and leafy stem from a stick sev-
eral feet long. The wands constructed for dipping into termite nests, on the other
hand, are made from vines or more flexible twigs and are considerably shorter.
They also have a bitten end, unlike the ant wands. These tools are extremely spe-
cialized and, as Byrne notes (1995, 97), the wands are often constructed some time
in advance and a considerable distance away from the place where they are going
to be used. Even better evidence for instrumental thinking in tool construction
comes from the archeological record. Complicated tools were being constructed
long before the emergence of language (Gibson and Ingold 1993, Mithen 1996).
As I show in much more detail when I come to consider the nature of nonlin-
guistic reasoning and rationality in chapter 6, examples such as these of techno-
logical manipulation reveal a distinctive type of instrumental thought. In sophisti-
cated forms of tool construction and manipulation a creature is representing the
outcomes of particular courses of action in a manner that we would most naturally
model in terms of a sequence of instrumental statements. In the terms I have been
using, the instrumental properties are perceived in inferential terms as the likely
consequences of a particular course of action. The minimalist account does not
seem able to accommodate this, however. As I have shown, the minimalist is com-
mitted to an account of the perception of instrumental properties in terms of nonin-
ferential mediate perception.
Nor is tool construction the only area that presents problems for the minimalist
approach. There is a basic type of practical decision-making that the minimalist
approach finds very difficult to accommodate. As I shall show in more detail in
chapter 6, all decision-making involves a selection between different courses of
action. But there are two different ways in which different courses of action can be
compared. On the one hand, they can be compared simply qua courses of action.
That is to say, an animal might compare the action of fighting with the action of
fleeing. On the other hand, however, an animal might compare the consequences
of the action of fighting with the consequences of the action of fleeing. These are
two very different types of representation. One involves representations of actions.
These representations are not very complex. They can plausibly be understood at a
purely perceptual level. The other, however, involves representations of contingen-
cies between actions and outcomes. It involves choosing a course of action be-
cause it will lead to a particular goal. This is a far more sophisticated type of repre-
sentation that it does not seem possible to understand in terms of noninferential
mediate perception at all.17
It is arguable that many instances of instrumental conditioning fall into this
category (Dickinson and Balleine 1993, Heyes and Dickinson 1993, 1995). A
related example comes with Rescorla and Skucy's discovery that rats trained to
press a lever for food will cease to press the lever when the schedule is changed so
that the food is delivered whether they press the lever or not (Rescorla and Skucy
1969, Hammond 1980). This seems to involve recognition that the contingency be-
tween lever pressing and food delivery no longer holds. The rats had initially been
pressing the lever in virtue of an instrumental belief that lever pressing would re-
sult in the appearance of food. When the correlation tracked by the instrumental
belief ceased to hold the associated behavior also ceased.
It looks, therefore, as if the minimalist account runs into clearly defined empiri-
cal problems. Nonlinguistic creatures are capable of forms of instrumental reason-
ing considerably more complex than the minimalist account can allow. The con-
clusion to draw from the argument from planning and instrumental belief is that
thinking at the nonlinguistic level cannot be driven by, and limited to, the immedi-
ate environment in the manner prescribed by the minimalist view—even taking
into account the availability on the minimalist construal of a notion of noninferen-
tial mediate perception. The upshot of the argument from planning is that it looks
very much as if psychological explanations of the behavior of nonlinguistic crea-
tures will at least sometimes have to approximate to standard belief-desire expla-
nations by including the ascription of an instrumental belief and/or the assumption
that nonlinguistic creatures engage in inferences about what is not immediately
present.
3.5 The Argument from Intelligent Action

The second argument for the insufficiency of minimalist approaches to nonlinguis-
tic thought starts from the adaptive requirements of the thinkers of such thoughts.
What benefits accrue to a creature in virtue of its capacity to represent the environ-
ment and hence to behave in a thinking rather than unthinking way? The most ob-
vious answer is that the power to represent the environment bestows a flexibility
and range of response that would not otherwise be available—and it is, as I showed
in section 1.2, precisely in order to explain such flexible responses that we need to
appeal to psychological explanations of behavior. But when we think more care-
fully about how the capacity to think generates this flexibility and range of re-
sponse it becomes apparent both that the types of thinking attributed to nonlinguis-
tic creatures must have a distinctive internal structure (Horgan and Tienson 1996)
and that these types of thinking involve a type of reification that is not available at
the purely perceptual level.
The essence of learning is the capacity to extrapolate from previously experi-
enced situations to situations that have not previously been experienced. Such ex-
trapolation involves identifying the salient respects in which two situations are
similar and to strip away the inessential features. The particular features that are
salient will vary, of course, depending on the activities in which the creature is en-
gaged and its particular interests at that moment. With respect to one set of activi-
ties and interests the most salient similarity between two states of affairs might be
that the same individual features in both of them. With respect to a different set of
activities and interests the salient similarity might be that similar properties are in-
stantiated in the two states of affairs. The creature's actions in the second (later)
situation will of course vary depending on which similarities are salient. This is
what generates the flexibility of response characteristic of thinking behavior. A
thinking creature will not always respond to the same situation in different ways,
because its response will depend on which aspects of the situation are perceived as
relevant.
Frequently this flexibility will come only because the creature is capable of rep-
resenting and keeping track of the separable components of the perceived environ-
ment and, subsequently, of acting in ways that reflect this capacity. When this is
the case the creature's representations of the environment must be structured so as
to have separate components that can reappear in, and be extrapolated from, a
range of further representations of the environment. The relevant distinction can be
made in terms of the initial characterization of thinking-how at the beginning of
the chapter. It will be recalled that reasoning by analogy was a prime candidate for
thinking-how—a form of reasoning that is not verbalizable, does not have linguis-
tically expressible structured contents, and is context-bound. The essence of the ar-
gument from intelligent action is that the simple perception of analogies and simi-
larities across different environments and situations is not always sufficient for the
flexibility of response characteristic of thinking behavior. Intelligent action often
requires keeping track of differentially represented features and objects in the per-
ceived environment (see Horgan and Tienson 1996, ch. 3, for a similar argument).
One way of developing this point would be in terms of the level of detachment
from the current context involved in nonlinguistic thought. The essence of the
minimalist approach is that nonlinguistic creatures need only represent the envi-
ronment perceptually, where the content of perception is understood solely in
terms of the possibilities that the environment affords for action and reaction. The
three exponents of the minimalist conception that I have been considering all draw
a sharp distinction between this action-based conception of the content of percep-
tion and a more objective and context-independent conception of the content of
prepositional attitudes. Consider, for example, this passage from Cussins:
An objective world is given to a subject if the content presents something as being in-
dependent of the subject's particular abilities and particular location, in space and
time. But given only the realm-of-embodiment-specified content, all the subject (em-
bodied organism) has is an experiential awareness of how to move etc in response to
local changes in its environment. If this were a subject's conception of a referent it
would be a conception of something as not independent of contingent characteristics
of the subject itself. The necessary separation between subject and object would not
have been achieved. (1992, 659)
A similar point is at the heart of Campbell's conception of the contrast between

causally indexical thought and fully self-conscious thought paradigmatically ex-
emplified only by language-using humans. Causally indexical thought does not
support a conception of objects as individuals persisting through space and time
and entering into causal interactions. Nor does it support the idea of reidentifying
an object (as opposed to recognizing a particular feature or a particular possibility

for acting).
Let us locate this aspect of the minimalist approach in a wider context. In the
broadest sense, to perceive the world as a structured entity is to perceive it as com-
posed of bodies interacting according to certain basic principles within a unified
spatial framework (Strawson 1959). We can follow Quine (1995, ch. 3) in describ-
ing the process of arriving at this sort of perception of the world as one of reifica-
tion. In order to get clear on how this process might work we need to understand
the alternative to a fully reified perspective on the world. One distinction here is
between the recognition of bodies and the recognition of recurrent circumstances
or features. It is possible for a creature to be in a position to recognize that it is
once again in the presence of certain features that it has previously encountered
(such as food or water or danger or shelter) without being able to parse the percep-
tual array into genuine bodies. Let us suppose that the creature encounters a tree
that affords shelter. Then the creature could perceive the tree as affording shelter
without perceiving it as a tree. To perceive a tree as affording shelter one need only
recognize its similarity to other things that have afforded shelter and to act accord-
ingly. All that is required here is appropriate action grounded on some form or
other of perceived similarity. It is often thought that perceiving the tree as a tree re-
quires no more than this. Some such idea is behind most empiricist theories of con-
cept formation and learning. But this neglects an important distinction. Being able
to respond appropriately to the presence of treeness is a matter of appropriate ac-
tion based on perceived similarity. But to perceive a tree as a tree is to perceive it
as an individual thing, as something that persists over time and can be encountered
at different times and in different sensory modalities. It is to be able to reidentify
the tree; to be able to pose the question, for example, whether the tree in front of
one is the same tree one perceived earlier; or to be able to select one tree out from
a group of perceptually similar trees; or to be able to count the number of trees in
one's immediate vicinity.
We can follow Strawson in terming this the distinction between the feature-
placing level of experience and the particular-involving level of experience
(Strawson 1959). At the feature-placing level of experience there is no distinction
to be drawn between one instantiation of, say, foodness and another. A creature at
this level inhabits a world composed of different kinds of stuff. These different
kinds of stuff can be encountered at different places, at different times, and in
different combinations. Some of these encounters and combinations are pre-
dictable, and the creature best able to make these predictions will flourish. But
the regularities on which these predictions are based are simply conjunctions of
features—associations of food with a particular perceptual Gestalt associated with
a certain combination of leaves and branches, for example. In contrast, the creature
that operates at the particular-involving level is able to detect regularities of a com-
pletely different order—regularities that govern the behavior of persisting bodies.
Many of these regularities will be causal, governing the interactions between per-
sisting bodies. Others will be kinematic, governing the possible movements and
behavior of any given body.
The three versions of minimalism that we have been discussing (those offered
by Dummett, Campbell, and Cussins) are all committed to locating nonlinguistic
behavior at the feature-placing level. This follows from the different ways they de-
velop the idea that nonlinguistic thoughts are constrained by the here-and-now and
immediately integrated with action. Types of thinking that are constrained in this
way do not need the mechanisms of individuation and identification that emerge
with the particular-involving conception,. The regularities that drive nonlinguistic
behavior on the minimalist construal are conjunctions of features and circum-
stances, rather than regularities holding over the behavior of persisting bodies. The
particular-involving conception of the world comes with detachment from the
here-and-now and the capacity to represent the world in a manner that is not con-
strained by the requirements of immediate action. The point of the argument from
intelligent action is that many examples of intelligent action in the nonlinguistic
realm require thought at the particular-involving level. Some examples are dis-
cussed in subsequent chapters.
Thinking at the particular-involving level places significant constraints upon
how we think about the content of thought at the nonlinguistic level. The contrast
with perception becomes even clearer. If a creature is representing particulars that
can be individuated and reidentified at subsequent times, then we will need to attri-
bute to it thoughts that have a character much closer to the subject-predicate char-
acter of the thoughts that we attribute to nonlinguistic creatures—thoughts in
which, to take the simplest example, there is a representation of an object in the
subject position and the representation of a property in the predicative position.
This goes far beyond the way in which nonlinguistic thoughts are construed on the
minimalist conception. In particular, and this is a theme that will recur both in the
next argument against minimalism and in the remainder of the book the thoughts
attributed will have an internal structure that perceptions do not have. The ability
to represent a given object as having a given property goes together with the ability
to represent that object as potentially having a range of properties and, moreover,
with the ability to represent that property as holding of a range of further objects.
The stable representations involved are far from the perceptual and sensori-motor
representations postulated by minimalism.
As will emerge in the next two chapters, it is no easy manner either to under-
stand what such thoughts might consist at the nonlinguistic level, or to identify the
grounds on which they can be attributed to nonlinguistic creatures. However, there
are three points that need to be born in mind. The first is that structured thoughts of
this type cannot be accommodated within the minimalist conception of nonlinguis-
tic thinking. The second is that such structured thoughts will need to be attributed
when what is being explained is behavior that rests upon a particular-involving
conception of the world. The third is that nonlinguistic creatures are capable of
such behaviors.
3.6 The Argument from Psychological Explanation

A similar conclusion about the need for structured thoughts at the nonlinguistic
level can be reached by arguing not from the practical requirements of the thinking
creature but instead from the requirements of psychological explanation. In mak-
ing sense of the behavior of a nonlinguistic creature we are often required to attri-
bute to it structured thoughts with a determinate and linguistically expressible con-
tent (or, at least, a content no more indeterminate than we find acceptable in
dealing with language-using creatures).
It is a constraint on explanations in general that they support counterfactuals
(statements about what would happen, or have happened, in different circum-
stances). Let us suppose that the presence of an F-type event is appealed to in order
to explain the occurrence of a G-type event within the context of a particular set of
background conditions that we can term C. Any such putative explanation is sub-
ject to the following explanatory adequacy constraint (EAC):
(EAC) An occurrence of an F-type event in conditions C (causally) explains
the occurrence of a G-type event if and only if the following two conditions are
met: (1) had the F-type event not occurred the G-type event would not have oc-
curred; (2) in any comparable situation in which an F-type event were to occur
a G-type event would also occur.18
Clause 1 is intended to rule out the possibility that the occurrence of the F-type
event is an accidental concomitant of the occurrence of the G-type event that did
not genuinely contribute to its occurrence. If it were an accidental concomitant
then one would expect the G-type event to occur even if the F-type event did not
occur. The second clause is intended to secure some sort of lawlike connection be-
tween the occurrence of F-type events and the occurrence of G-type events, again
to rule out a merely coincidental juxtaposition of F- and G-type events.19
Let us assume (as we need to do even on the minimalist construal) that we are
attributing a desire to explain a particular behavior. The content of the desire we at-
tribute must satisfy the two clauses of EAC. That is to say, it must be the case both
that had the creature in question not had a desire with that content, the behavior
being explained would not have occurred and that in a suitably similar situation in
which the creature has a desire with the same content it would behave in the way it
actually did behave. Suppose we ask how one might go about determining whether
or not clause 1 of the counterfactual constraint is satisfied.20 What would make it
the case that, had the agent not had the hypothesized desire, it would not have
acted in the way it did? How could we evaluate this conditional for truth or falsity?
The content of a desire is given by the state of affairs that serves as its satisfac-
tion-condition, and a state of affairs serves as the satisfaction-condition of a desire
if its being the case would bring about (in the right way) the cessation of the be-
havior to which that desire gave rise. If clause 1 of the counterfactual constraint
does indeed hold, then the behavior would also cease (and hence the desire be sat-
isfied) in any state of affairs that is suitably similar to the state of affairs in which
the creature actually acted. I shall call these the experimental state of affairs and
the candidate state of affairs, respectively. The respects in which the experimental
and the candidate states of affairs are suitably similar are determined, of course, by
the content of the desire. So one would test clause 1 by varying the environment in
various respects and seeing whether the relevant behavior does indeed cease. If the
desire-driven behavior is extinguished (in the right sort of way) by an experimental
state of affairs that differs from the candidate state of affairs in one or more re-
spects then those respects should not be reflected in the specification of the desire.
But, of course, these variations between the target state of affairs and the experi-
mental state of affairs must be variations to which the creature is cognitively sensi-
tive. But this requirement entails that the creature possesses structured motiva-
tional states—desires with contents that are sensitive to variations in possible
satisfaction-conditions.
I can illustrate this point with a simple example. Suppose that we are consider-
ing a bird delivering food to its chicks in the nest (and let us assume, purely for il-
lustrative purposes, that this behavior warrants a belief-desire explanation). The
bird lands at the edge of the nest and delivers food to the nearest chick. The ques-
tion is whether the bird is motivated by the desire to give food to that particular
chick (perhaps because it has kept track of how much that chick has been fed rela-
tive to the other birds) or by a desire to give food to the nearest chick—or by an
even more general desire. It is clear what predictions are made when the counter-
factual constraint is applied to these proposals. Clause 1, as applied to the first can-
didate desire, implies that in the counterfactual situation in which the chick were in
a different position in the nest the parent would seek it out to give it food. Clause 1
applied to the second candidate desire implies that in that counterfactual situation
the bird would behave in the same way as in the target situation. Clause 2 for both
candidate desires suggests that in a situation similar to the target situation the bird
would act in the same way. Clearly, however, the possibility of the counterfactual
constraint applying in this way depends on the bird's desires being sensitive to
separable components of the relevant situations. That is to say, the applicability of
the counterfactual constraint depends on the content of the bird's desire being sen-
sitive to variations in possible satisfaction conditions, hence on its being struc-
tured. The creature must be credited with the ability to detect and recognize com-
mon features of the situation across the experimental and candidate states of
affairs.
This line of thought complements the argument from intelligent action dis-
cussed in section 3.5. The upshot of the earlier argument was that a nonlinguistic
creature's successful navigation of its environment, as manifested in the ability to
learn from experience and to recognize salient similarities across different situa-
tions, requires a more objective way of representing the environment than the
minimalist conception has the resources to provide. It requires something much
like the process that Quine has termed reification, which cannot be accommodated
purely at the level of perceptual content. What I have shown in the argument from
psychological explanation is that a similar approach to nonlinguistic thought ap-
pears to be built into our models of psychological explanation—and, in particular,
into the way we employ counterfactual reasoning to evaluate particular attributions

of desires.
3.7 Taking Stock

Chapter 2 developed some of the principal difficulties posed by the apparent need
to give psychological explanations of the behavior of nonlinguistic creatures.
These difficulties arise primarily because these psychological explanations are un-
derstood as applications of the same belief-desire psychology that is applicable to
language-using creatures. In this chapter I have been looking at an alternative way
of construing the project of explaining the behavior of nonlinguistic creatures in
psychological terms. The minimalist proposal is to take the psychological states at-
tributed in such explanations to be nonpropositional, analyzing them on the model
of perceptual states rather than prepositional attitude states. The thoughts attri-
buted to nonlinguistic creatures on the minimalist approach are context-bound, es-
sentially tied to the creature's capacities for action and reaction, perceptually vehi-
cled, and lacking the constituent structure characteristic of prepositional thought.
This approach to nonlinguistic thought allows us to avoid some of the key prob-
lems that arise in explaining how nonlinguistic creatures can have states with
prepositional content. And, as I showed in section 3.3, it permits a very low-key
approach to psychological explanation. There can be no doubt that the minimalist
account is highly profitable and that these low-key psychological explanations are
frequently applicable. Many types of behavior in nonhuman animals and human
infants are best understood in the, broadly speaking, sensorimotor terms proposed
by the minimalist (whether this is understood in terms of Dummett's proto-
thoughts, Cussins's cognitive trails, or Campbell's causally indexical thoughts—
or, indeed, any other related notion). Many cases of thinking without words (no
doubt, in fact, the majority of such cases) are not best described as involving struc-
tured and context-independent thoughts that are linguistically expressible. I sug-
gested in chapter 1 that thinking behaviors are neither inflexible nor stimulus
driven. This freedom from the constraints of the stimulus and from the rigidity
of fixed behavioral responses can perfectly well be generated by the types of
perception-based thought identified by the minimalist approach. And this percep-
tion-based type of thinking permits both the possibility of misrepresentation and
the cognitive integration that, in section 1.2,1 proposed as the hallmarks of the in-
tentional states featuring in psychological explanations. I will show in more detail
at later stages in the book how the minimalist approach can fruitfully be applied.21
Despite the obvious utility of the minimalist approach, however, I have argued
that it cannot be a complete account of nonlinguistic thought. Various important
types of thinking behavior cannot be understood in the context-bound and essen-
tially perceptual manner proposed by the minimalist account. This is the case, for
example, with those types of thinking that involve instrumental reasoning about
the likely consequences of actions and about how to tailor means to ends. More-
over, we have to see nonlinguistic thoughts as structured in a way that reflects the
creature's capacity to think about the different things and properties that the envi-
ronment contains—both in order to understand how nonlinguistic creatures are ca-
pable of learning and adapting to different environments and in order to identify
the particular beliefs and desires that it is appropriate to identify in a particular
context.
In moving beyond the minimalist framework we need to pay particular attention
to two dimensions of nonlinguistic thought. The first is the dimension of what, fol-
lowing Quine, I have termed reification. That is to say, the way the world is carved
up in thought into reidentifiable particulars, as opposed to clusters of features. This
will be explored in the next chapter, where I will develop a framework for thinking
about different ontologies at the nonlinguistic level. The second is the dimension
of instrumental belief. We need to understand what it is for a nonlinguistic creature
to have beliefs about the consequences of its actions and about the different ways
of achieving a given end. This will be considered in greater detail in chapters 6 and
7. The first step, however, is to get a clearer perspective on what it might mean to
attribute beliefs and desires to nonlinguistic creatures. In the first two sections of
chapter 41 will develop a version of success semantics that will simultaneously ex-
plain what it is for a nonlinguistic creature to have a given belief or a given desire
and how we might go about attributing those beliefs and desires.
4
Ascribing Thoughts
to Nonlinguistic Creatures
Toward an Ontology
It has emerged that we cannot hope to provide an adequate ground for the wide-
spread practices of attributing thoughts to nonlinguistic creatures if we restrict our-
selves to the tools offered by the minimalist approach. The minimalist account of
nonlinguistic thought in terms of perceptually based protothoughts understood pri-
marily in terms of a creature's ability to act on its environment is without doubt all
that we need to understand many of the types of behaviors that were put forward in
chapter 1 as candidates for psychological explanation. The upshot of chapter 3,
however, was that the minimalist approach cannot be a complete account of non-
linguistic thinking. There are types of thinking behavior that require a form of
explanation much closer to prepositional attitude explanation—a form of explana-
tion that will involve the attribution of thoughts. Whereas the minimalist ap-
proach attempts to specify nonlinguistic thought in terms that are fundamentally
perceptual and nonpropositional, it looks as if, for at least some cases of nonlin-
guistic thought, we will have to engage in a limited form of prepositional attitude
psychology, explaining behavior through the attribution of beliefs and desires.
In this chapter and the next I develop a framework for understanding such
attributions.
I will be offering a version of success semantics, based on the idea that the con-
tent of a belief is its utility condition and the content of a desire its satisfaction-
condition. This is outlined in section 4.1. As I show in section 4.2, existing ver-
sions of success semantics generate a significant problem of indeterminacy that is
rather similar to the indeterminacy of radical translation discussed by Quine
(1960). Applying success semantics does not give us any way of distinguishing be-
tween the different thoughts that could be attributed in any given situation. In the
main part of this chapter (sections 4.3 and 4.4) I will be concerned with one di-
mension of this indeterminacy—the ontological indeterminacy arising from the
need to specify how a creature carves up its environment. The issues here intersect,
as I will show, with the points raised in the previous chapter about the need to ex-
plain how nonlinguistic creatures represent and identify particulars.
Section 4.3 considers Quine's account of how a creature can make the move
from experiencing a world composed of features to a world composed of bodies
(the process he terms reification). Quine's own account of reification is heavily
language dependent. He argues, in effect, that reification is achieved through the
linguistic machinery of pronominal reference and quantification. I develop an al-
ternative account of reification based on a different understanding of what it is to
perceive something as a body. The central claim is that perceiving something as a
body is to perceive it as subject to a range of higher-order physical principles. In
section 4.4 I show how this understanding of reification allows us to resolve the
issue of ontological indeterminacy. The essential structure of a creature's ontology
should be understood in terms of the higher-order physical principles to which it is
perceptually sensitive.
4.1 Success Semantics: Problems and Prospects

The idea I propose in this section is that the process of ascribing thoughts to ani-
mals is a form of success semantics, based on the idea that true beliefs are func-
tions from desires to actions that cause thinkers to behave in ways that will satisfy
their desires (Mellor 1991, Whyte 1990). The basic idea of success semantics is
that the content of a belief is given by its utility condition, that is, the condition that
would have to obtain for the various desires with which it is associated to be satis-
fied. In brief, true beliefs cause actions that satisfy desires.
Success semantics originated with some fleeting comments by Frank Ramsey
about the beliefs of a chicken in his 1927 article "Facts and Propositions." He com-
mented that the chicken's belief that a certain type of caterpillar is poisonous
should be equated with the chicken's "abstaining from eating such caterpillars on
account of unpleasant experiences connected with them" (Ramsey 1927[1978],
46). He then went on to generalize the point: "Thus any set of actions for whose
utility P is a necessary and sufficient condition might be called a belief that P,
and so would be true if P, i.e. if they are useful" (46). Although contemporary
proponents of success semantics tend to think that they are merely expanding on
Ramsey's original insight (see, e.g., Mellor 1991, 23), this cannot be right. It
is hard to see how the basic idea that true beliefs are functions from desires to
actions that cause thinkers to behave in ways that will lead to the satisfaction of
those desires can be extracted from Ramsey's comments. For one thing Ramsey's
proposal is clearly committed to a strong form of behaviorism. Ramsey suggests
that we should identify the belief that P with the set of actions whose utility condi-
tion is given by P. But this makes a nonsense out of the idea that true beliefs are
functions from desires to actions. How can a set of actions be a function from
desires to actions? How, for that matter, can a set of actions combine with a desire
to cause an action? Nonetheless, the germ of the idea of success semantics
is clearly there in Ramsey. This is the claim, first, that we need to approach
the truth-conditions of beliefs through their utility conditions and, second, that
these utility conditions should be understood in terms of the satisfaction of
desires.
Two qualifications before going any further. First, I do not think that success se-
mantics can be employed to elucidate the nature of truth in the way that Whyte, for
example, has argued. He suggests that "truth just is the property of a belief that
suffices for your getting what you want when you act on it" (1990, 149). This
seems to invert the true order of explanation. We need an independent characteri-
zation of truth if success semantics is to be genuinely contentful. Second, I am not
putting success semantics forward as a global semantic theory that will apply
equally to the beliefs of language-using creatures and the beliefs of non-language-
using creatures. This is another respect in which I differ from contemporary propo-
nents of success semantics (although not, it should be said, from Ramsey himself).
There are no prospects for giving an account of the complex belief systems of lan-
guage-using creatures in terms of success semantics, for the simple reason that so
many of our beliefs have little direct contact with actions or desires. Success se-
mantics can only be plausibly applied to creatures of whom it is right to say that
the role of doxastic states in their cognitive economy is to secure the satisfaction of
their desires, and that is simply not true of language-using human beings. As I
show in some detail in chapters 8 and 9, the acquisition of language opens up an
enormous range of types of thinking that effectively puts the language-user beyond
the scope of success semantics.
It is sometimes objected that success semantics is threatened by circularity,
since it is impossible to give the contents of desires without presupposing the con-
tents of beliefs (by holding, for example, that the satisfaction-condition of a desire
is that state of affairs guaranteed to obtain if the desire combines with true beliefs
to cause an action). However, there are ways of specifying the content of a desire
that do not involve any such circularity, namely, in terms of the state of affairs that
would cause the desire to cease.1 In thinking about the content of a desire we need
to start with the notion of a satisfaction-condition. The satisfaction-condition of a
desire is a direct analogue of the truth-condition of a belief. Just as the truth-condi-
tion of a belief is the state of affairs that would make it true, the satisfaction-condi-
tion of a desire is the state of affairs that would satisfy it. And just as the content of
a belief is given by the state of affairs that would make it true, the content of a de-
sire is given by the state of affairs that would satisfy it (Whyte 1991).2 What makes
a creature's desire a desire for food is that it would be satisfied by that creature's
obtaining food. What counts as satisfaction? Most straightforwardly, a desire is
satisfied when the behavior to which it is giving rise ceases. Of course, not just any
cessation of behavior will count as a satisfaction of the desire. The behavior to
which my desire for food gives rise may well be brought to a halt if someone offers
me a large sum of money to sit still, but my desire for food would not ipso facto be
satisfied. The cessation of the behavior has to be brought about in the right way. I
shall return to this shortly.
Dretske has objected to accounts of the content of desire in terms of their satis-
faction-conditions. He takes the example of the desire that must be invoked to ex-
plain why a pet rabbit (PJ) licks the spout of a water bottle attached to his cage:
PJ may not actually get what he wants. If the water bottle is empty, he licks the spout
in order to get water, but this goal—getting water—is never reached. This doesn't
prevent it from being PJ's purpose in licking the spout. It doesn't prevent water, or
getting water, from being the goal of that internal state (D) that is now active in pro-
ducing licking movements. For what makes water the goal of this state is not what
will satisfy it (perhaps nothing ever will satisfy it) but the fact that it makes water re-
inforcing, makes water the sort of result that tends to promote the behavior that leads
to it. And what makes this state, and the fact that it is, in this sense, for water, ex-
planatorily relevant to PJ's current licking behavior, is the fact that D was recruited as
a (partial) cause of licking movements because those movements resulted in water.
(Dretske 1988, 128)
Dretske's objection to specifying the content of PJ's desire in terms of its satisfac-
tion-condition is that the desire may well not be satisfied. His competing sugges-
tion is that the content of a desire is given by the state of affairs that reinforces the
actions to which that desire gives rise (provided that the background conditions are
the same). But neither Dretske's objections to the satisfaction-conditions account
nor his alternative are convincing. The problem with making the notion of rein-
forcement central to determining the content of desires is that it is very restrictive.
As Dretske himself recognizes in the final sentence of the quoted paragraph, a rein-
forcement-based account of the content of desire can only accommodate goals that
have been experienced as reinforcers in the past ("the desire that explains the be-
havior is a desire for whatever past result figured as a structuring cause of the be-
havior" [129]). This entails that new desires can emerge only by chance—when a
behavior happens to give rise to a felicitous outcome and the creature in question is
motivated to repeat that behavior (as in Jean Piaget's [1952] notion of a secondary
circular reaction). This result should be avoided if at all possible, and it is fortunate
therefore that his objection to the satisfaction-condition account is less than con-
vincing. The undeniable possibility that a desire may not be satisfied should be no
more of an obstacle to specifying its content in terms of what would satisfy it than
the possibility that a belief might not be true should stand in the way of specifying
its content in terms of what would make it true. The point is not what will satisfy
the desire, but what would satisfy it (and not what does make it true, but what
would make it true).
Yet the notion of reinforcement does have a role to play in determining the con-
tent of desire. As Whyte has noticed (1991), we can employ it to solve the problem
mentioned earlier, namely, the problem of specifying the right sort of circum-
stances in which the cessation of desire-generated behavior will indicate the con-
tent of the desire. As he puts it, "a desire is satisfied if and only if its going away
reinforces the disposition to act in the way that caused it to go away" (1991, 68).
Using the notion of reinforcement in this way does not restrict the possible con-
tents of desires to goals that have been experienced as reinforcers in the past, thus
avoiding the central problem in Dretske's theory.
This account of the satisfaction-conditions of desires reveals a further reason for

thinking that success semantics cannot be extended beyond the realm of the non-
linguistic. The model only works for relatively simple desires. In fact it works best
for basic appetites (and it is no accident of course that published discussions of
success semantics tend to concentrate on these). There is a very clear sense in
which we can identify when the behavior associated with the desire for food actu-
ally ceases and hence work backward to its satisfaction-condition. But one of the
many cognitive advantages made possible by language acquisition is precisely that
of supporting desires for which this straightforward model of cessation and satis-
faction is completely inappropriate. To give just one example, I will show in chap-
ters 8 and 9 that acquisition of a language makes possible critical reflection on
one's own thinking. A creature capable of critical thinking about its own desires is,
ipso facto, capable of second-order desires about the desires it should have. But
only the simplest and most straightforward second-order desires can be understood
in terms of the states of affairs that would bring about their cessation. Some sec-
ond-order desires are targeted on specific first-order desires and hence can be un-
derstood as states that cease when the relevant first-order desires come into being.
But many second-order desires are open-ended. They are desires not about the spe-
cific desires one wants to have but about the sort of person one wants to be. Sec-
ond-order desires such as these do not come to a halt, and their satisfaction is not
all-or-nothing.
Taken as a theory applicable solely to the behavior of nonlinguistic creatures,
however, success semantics has several conspicuous advantages. The first is epis-
temological. Success semantics provides simultaneous answers to two questions—
the question of what a nonlinguistic belief consists in and the question of how we
are to go about identifying the content of that nonlinguistic belief. As I showed in
chapter 2, no acceptable theory of the thoughts of nonlinguistic creatures can sepa-
rate the constitutive dimension from the epistemological dimension.3 This was one
of the principal shortcomings of the language of thought theory. In contrast, the
two dimensions are inextricably linked in success semantics. The utility condition
of a belief in a particular situation is completely open to the third-person perspec-
tive of the ethologist or developmental psychologist. Moreover, success semantics
does not fall foul of the basic difficulty that strikes competing theories such as
causal covariance semantics (Fodor 1987) and teleosemantics (Dretske 1988, Mil-
likan 1984). The characteristic of these theories is that they work backward from
vehicle to content, characterizing content in terms of a relation between the vehicle
and features of the distal environment in a way that requires being able to identify
the vehicle independently of identifying the content. As I showed in chapter 2
(section 2.3), such theories are prone to serious epistemological difficulties, be-
cause we cannot identify a representational vehicle without fixing its semantic
properties.
A second advantage of success semantics is that it provides a clear way of cap-
turing how an adaptive creature is in tune with its environment without making im-
plausible claims at the level of the vehicle of representation. According to the vari-
ous versions of teleosemantics, for example, the attunement of a creature to its en-
vironmental niche is a direct function of the fact that various elements of the sub-
personal representational system have evolved to track certain features of the distal
environment. This runs into some of the problems identified in the previous chap-
ter for the minimalist theory of nonlinguistic thought. In particular, many types of
intelligent thinking behavior arise in novel situations where evolutionary notions
of proper function do not seem to get a grip. Attunement to the environment arises
at the level of the organism, rather than at the level of subpersonal representational
vehicles. That is to say, an organism can be attuned to the environment in a way
that will allow it to operate efficiently and successfully, even though there has not
been selective pressure for sensitivity at the subpersonal level to the relevant fea-
tures of the distal environment. Successful attunement to a particular situation can
emerge from the combination of different heuristics and representational abilities
for which the selective pressures were completely different. In the extreme it might
be, as in the cases brought to our attention by Stephen Jay Gould, be a complete
by-product of adaptation to something else (Gould and Lewontin 1979).
Perhaps the principal advantage of success semantics, however, is that it deliv-
ers truth-conditions (via utility conditions). Success semantics starts at the level of
the complete thought, rather than at the level of individual content constituents.
Causal covariance theories have primarily addressed the question of how individu-
als and properties are mentally represented, but no accounts have been offered of
how these representations combine to generate a thought. The defining feature of a
thought is that it is trath-evaluable. Thoughts are things that can be either true or
false because they represent states of affairs as holding. These states of affairs are
then- truth-conditions. Yet it is clear that the simultaneous tokening in the mind of a
property representation and an object representation is not sufficient to generate
the thought that the object in question has the property in question—hence to de-
termine the relevant truth-condition. This can be seen most easily when we are
dealing with relational properties that are asymmetric. Simultaneous tokening of
the mental representation of a two-place relational property and two object-repre-
sentations leaves completely undetermined which object-representation is to oc-
cupy which place in the property-representation. This is, of course, the traditional
problem of the unity of the proposition, which might be described in more up-to-
date terms as the binding problem for thoughts. There is no such problem for suc-
cess semantics, because it starts from complete thoughts/truth-conditions.
4.2 Success Semantics and Radical Interpretation

The utility condition of a belief, as I have so far discussed the notion, is simply a
state of affairs—namely, the state of affairs whose holding will bring about the sat-
isfaction of desires with which that belief is conjoined. The satisfaction-condition
of a desire is equally a state of affairs—namely, the state of affairs that will extin-
guish in the right sort of way the behavior to which the desire has given rise. But
identifying these states of affairs is only the first step in the process of giving a
psychological explanation at the nonlinguistic level. We also need to know how

the creature in question registers that state of affairs—what it takes to be the con-
stituents of that state of affairs and how it might reidentify them in different con-
texts. When we are dealing with language-using creatures with whom we believe
that we share a conceptual framework, we can presume that their perspective on
the world is broadly similar to our own. So we are able to move without any prob-
lem from identifying the relevant utility condition or satisfaction-condition to
specifying the content of the appropriate belief or desire. Our specification of the
content simply employs the conceptual apparatus that we ourselves would employ
to characterize the relevant state of affairs. This, of course, is because we can pre-
sume that our fellow language-users apprehend a world composed of roughly the
same type of objects that we perceive and obeying the same kind of laws.
But there are no grounds for making a comparable assumption when we are
dealing with the beliefs and desires of nonlinguistic creatures. An important source
for the resistance that many theorists have expressed to the ascription of beliefs
and desires to nonlinguistic creatures is the thought that any such ascription will
inevitably and illegitimately use our own conceptual framework and ways of cate-
gorizing the world in characterizing the psychological states of creatures who may
well have a completely different way of categorizing the world. And it is surely
correct that even once we have identified a utility/satisfaction-condition, it will be
misleading to specify the relevant state of affairs in a manner that unquestioningly
presupposes our own conceptual framework. When we are faced with the behavior
of nonlinguistic creatures and attempt to move backward from that behavior to
what they believe and desire, we are in many respects in the position that Quine
envisaged a field linguist being in when confronting a completely unfamiliar lan-
guage about whose syntax and semantics he has absolutely no clue (Quine 1960).
The field linguist's task is to translate the language into his own language. He
starts, of course, with sentences that seem the closest to factual observations on the
perceptible environment, trying to correlate utterances with the states of affairs on
which they seem to be commenting.
The sentences first and most surely translated in such a case are ones keyed to present
events that are conspicuous to the linguist and his informant. A rabbit scurries by, the
native says "Gavagai," and the linguist notes down the sentence "Rabbit" (or "Lo, a
rabbit") as tentative translation, subject to testing in further cases. (Quine 1960, 29)
The testing is required to rule out alternative available translations. The native might
have been commenting on the presence of food, or on the presence of an animal, or
on the rabbit's color. It seems plausible that these ambiguities can be resolved by
considering other situations in which "Gavagai" is uttered, as well as what the native
says when confronted with things that are animals but not rabbits, as well as non-
rabbit-like objects that happen to be the same color as the original rabbit:
The general law for which he is assembling instances is roughly that the native will
assent to "Gavagai?" under just those stimulations under which we, if asked, would
assent to "Rabbit?" and correspondingly for dissent. (Quine 1960, 30)
Quine's principal point is that there is no way of determining how to move beyond
this general law. Quite apart from the local semantic indeterminacies that the radi-
cal translator can resolve in the manner I have considered, there are global onto-
logical indeterminacies that do not lend themselves to the same treatment. The
simple fact that there are certain sentences that inspire identical patterns of assent
and dissent from translator and native across a range of stimulus situations (so that
they are what Quine calls stimulus synonymous) does not guarantee even the coex-
tensiveness of the relevant terms:
For, consider "Gavagai." Who knows but what the objects to which this term applies
are not rabbits after all, but mere stages, or brief temporal segments, of rabbits? In ei-
ther event the stimulus situations that prompt assent to "Gavagai" would be the same
as for "Rabbit." Or perhaps the objects to which "Gavagai" applies are all and sundry
undetached parts of rabbits: again the stimulus meaning would register no difference.
When from the sameness of stimulus meanings of "Gavagai" and "Rabbit" the lin-
guist leaps to the conclusion that a gavagai is a whole enduring rabbit, he is just tak-
ing for granted that the native is enough like us to have a brief general term for rab-
bits and no brief general term for rabbit stages or parts. (Quine 1960, 51-52)
The problem here is one of whether terms are common nouns that divide their ref-
erence in a standard way (e.g., "rabbit") or whether they divide their reference in a
nonstandard way or whether they fail to divide their reference at all. This is, of
course, an ontological issue. The problem is how to determine whether to translate
into a language reflecting a particular set of ontological categories.
Quine himself takes the view that this ontological indeterminacy cannot be re-
solved—and he argues from it to the general indeterminacy of linguistic meaning.
There are serious questions to be asked about whether radical translation could
possibly be subject to an ontological indeterminacy as radical as Quine suggests
and, of course, about the inference from indeterminacy of translation to indetermi-
nacy of linguistic meaning. Gareth Evans, for example, has argued that certain
types of combination and negation of predicates would be impossible to under-
stand in a language whose general terms failed to divide their reference over physi-
cal objects (Evans 1975; and see Wright 1997 for further discussion). Be that as it
may, it is clear that a parallel set of questions is raised by the interpretation of the
behavior of nonlinguistic creatures. The developmental psychologist, animal be-
haviorist, or cognitive ethologist is in exactly the position of an idealized radical
translator, with the utility/satisfaction-condition corresponding to the field lin-
guist's stimulus situation. There are many stimulus synonymous ways of specifying
the utility condition—many different ways of describing the given state of affairs
that are compatible with its being such as to result in the satisfaction of the crea-
ture's desires. Precisely the questions that Quine raises for the radical translator
arise for the ethologist or animal behaviorist, even once a utility condition has been
identified within the ethologist's own ontological framework. These are questions
about how to specify that utility condition to reflect the way the creature carves up
the environment it experiences—how it digitalizes the analogue information de-
rived from perception.
The issue of ontological indeterminacy is most conspicuously a matter of the

types of singular terms and common nouns that can be employed in ascribing
thought-contents to nonlinguistic creatures—which will of course have ramifica-
tions for the predicational structure and compositionality of the language em-
ployed to attribute contents. The determination of an ontology will, if it is indeed
possible, allow an initial removal of indeterminacy. It will place us in a position to
discount as expressions of the thought in question all those sentences whose singu-
lar terms and common nouns do not pick out individuals of the appropriate onto-
logical category. But is it really possible to determine an ontology for nonlinguistic
creatures?
Competing philosophical orthodoxies push us in different directions. For
philosophers influenced by Quine, there can be no such determination of an ontol-
ogy—radical indeterminacy is ineliminable. For another group of philosophers,
the ontology of mind-independent persisting objects standing in causal, spatial,
and temporal relations is in some sense a necessary condition of genuine thought
(Evans 1975, Strawson 1959). This view is essentially neo-Kantian. Both views
militate against the determination of an ontology for nonlinguistic creatures—the
first for obvious reasons and the second because it seems frankly incredible that
animals and nonlinguistic infants should carve up the world in exactly the same
way that language-using adult humans do.
As I bring out in the remainder of the chapter, however, there is a false opposi-
tion here. It is not the case that the only determinate ontologies are object-centered
in the strong sense of involving mind-independent persisting objects standing in
causal, spatial, and temporal relations to each other. There is a spectrum of possi-
ble ontologies approximating in different degrees to the full-fledged ontology with
which we are familiar from the neo-Kantian tradition in philosophy. Different on-
tologies will be appropriate for different species, as well as, of course, for different
developmental stages within a single species. Identifying an ontology will be the
first step in capturing, at the level of sense, the determinate contents of nonlinguis-
tic thoughts.
4.3 Two Types of Reification

As emerged in the previous chapter, there is an important distinction between fea-
ture-placing and particular-involving levels of experience. The transition from fea-
ture-placing to particular-involving experience takes place in many species, in-
cluding our own. I shall, following Quine, call this transition the process of
reification. The details of how this move might take place have not received much
attention from philosophers, although there are important issues concerning the
preconditions of reification and the cognitive abilities on which it depends. The
most developed theory in this area is Quine's account in various writings (1960,
1974, 1995) of what he terms the psychogenesis of reification. Quine sketches out
an account of how reification might occur in terms that are decidedly language-
dependent. In this section I explore this account, explain why it raises problems for
the theory of nonlinguistic thought, and then offer an alternative. As I show at the
end of this section, the proposed alternative will offer a way of thinking about dif-
ferent ontologies at the nonlinguistic level and working toward a resolution of the
indeterminacy problem identified in the previous section.
The key claim in Quine's account is that reification involves more than the per-
ception of clusters of features. Clearly, it is a necessary condition for perceiving
the world as divided up into bodies that a perceiver be capable of recognizing that
features are juxtaposed in a single place—that the perceiver be capable of respond-
ing to the presence of combinations of features as well as to individual features. To
perceive a body is to perceive a compact clustering of visually detectable features
and to recognize that that cluster is associated with further features detectable
in different sensory modalities. A body is a bundle of properties. But a body is
not just a bundle of properties. A body is a thing that has certain properties. The
simple clustering of colocated features can be immediately perceived, but to get
genuine reification there needs to be an understanding (which may or may not be
purely perceptual) of a form of coinstantiation stronger than simple spatiotemporal
coinstantiation.
It might help to focus the issue here by drawing a comparison with a well-
known difficulty in the neurophysiology of perception. It has been established that
different perceptual features are processed in different areas of the visual cortex
(Zeki 1993). This raises the question (the so-called binding problem) of how and
where those different streams of processing are brought together in a way that will
permit our phenomenological experience of unified objects (for a review see Treis-
man 1996). That the binding process can go wrong is shown by the possibility of
inducing illusory conjunctions in which a feature of one object is perceived as a
feature of another. Such illusory conjunctions occur when exposures are brief and
subjects are prevented from focusing their attention on each object in turn. In one
experiment (Treisman and Schmidt 1982) subjects were briefly presented with an
array of colored letters bounded on each side by a digit and told in advance that the
task would be to identify the flanking digits. The array was displayed for 300 ms—
just long enough for the subjects to attend to the digits and succeed in identifying
them. When the subjects are asked to report on the letters between the flanking dig-
its they regularly mix up shapes and colors, so that the letters are correctly identi-
fied but with the wrong colors. The subjects correctly pick out all the relevant fea-
tures in the perceptual array—the full range of shapes and visually presented
colors. But they put them together (bind them) in the wrong way. One way of de-
scribing the distinction between a body and a bundle of properties would be to say
that this featural binding problem is not the only binding problem. A solution to the
featural binding problem will still only give us a bundle of features. It would not
be enough to underwrite our experience of a world of objects. Something more
needs to be added. A further binding problem needs to be solved.
The point emerges from a different angle if we consider the type of generaliza-
tions that can be entertained at the feature-placing level—what Quine calls obser-
vation categoricals: associations of features along the lines of "Where there's yel-
low, there's food" or "When there's growling, danger is near." It is clear, as Quine
remarks (1995, 25), that observation categoricals occur at the most basic level of
inductive learning—and, moreover, can be completely association-driven. There is
a clear difference between these feature-based generalizations and the correspon-
ding particular-involving generalizations: "Yellow things provide food" or
"Growling things are dangerous." Most important, the truth of the relevant obser-
vation categorical is entirely compatible with the falsity of the associated particu-
lar-involving generalization. It may well be that the presence of yellow is an infal-
lible sign of the presence of food, but this is perfectly compatible with some
yellow things not themselves being edible but always being accompanied by edi-
ble things. Similarly, growling might reliably signal danger even though the dan-
ger does not issue from the creature that is doing the growling. Yet this distinction
is unavailable to creatures operating solely at the feature-placing level. What
would make it available? How might a creature make the shift from observation
categoricals to particular-involving generalizations? In Quine's view the shift is
made with the acquisition of the linguistic apparatus of quantification.
The crucial leap to reification of ravens can be achieved by just improving our near
approximation by changing "there" to "it": "Whenever there is a raven, it is a black
raven"; "Whenever there is a raven it is black"; "All ravens are black".
The pronoun "it" is a vital new link between the component observation sentences
"Raven" or "Lo, a raven" and "Black" or "Lo, black." It posits common carriers of
the two traits, ravenhood and black. The carriers are ravens, bodies. I see this
pronominal construction as achieving objective reference. (1995, 27)
According to Quine, then, reification is achieved by acquiring the linguistic tools

that will allow the perceiver/thinker to move beyond juxtapositions of features and
to bind those bundles together into bodies. These tools are not, as one might imme-
diately think, the apparatus of proper names and common nouns. A child can em-
ploy a proper name to refer to a feature, and mere employment of a common noun
is not an index of reification—as is shown by Quine's own example of the sentence
"Whenever there is a raven there is a black raven," which can be used in a way that
simply associates the features of blackness and ravenhood, rather than quantifying
over the domain of ravens. Proper names only achieve reification when they are
brought within the ambit of pronominal reference and quantification.
Quine's suggestive remarks make very clear that the theorist of nonlinguistic
thought needs at the very least to show that reification can take place in the ab-
sence of the linguistic machinery associated with pronominal reference and quan-
tification. It is important to get clear on the precise terms of this challenge. The
outcome of reification is the perception of an articulated and structured world. We
need good evidence for thinking that nonlinguistic creatures can achieve this out-
come—hence that it cannot be dependent on the linguistic machinery of pronomi-
nal reference, or indeed of any other type.
This might seem an impossible task. How can one tell whether a creature is per-
ceiving an agglomeration of features rather than a reified body? The answer
emerges when one reflects that reified bodies obey certain principles and behave in
certain regular ways. Clearly, it is a necessary condition of perceiving the world as

divided up into objects that a perceiver be capable of recognizing that features are
juxtaposed in a single place—that the perceiver be capable of responding to the
presence of combinations of features as well as to individual features. To perceive
a body is to perceive a compact clustering of visually detectable features and to
recognize that that cluster is associated with further features detectable in different
sensory modalities. But perceiving objects is not just a matter of how one parses
the perceptual array. It is also a matter of the expectations one has about how those
compact clusters of visually detectable features will behave, both when one is per-
ceiving them and when one is not perceiving them.
A world articulated in terms of reified bodies is a world that obeys certain basic
physical regularities. The physical regularities under discussion here are not the
simple particular-involving generalizations that I have already discussed. They are
the higher-order physical regularities that govern, and indeed make possible, those
particular-involving generalizations by circumscribing what it is for something to
count as a particular or a body at all. Bodies tend by and large to have certain
higher-order physical properties in common. They tend, for example, to have
sharply defined edges relative to their environment. They usually maintain certain
forms of perceptual constancy relative to the perceiver when the perceiver moves
or when they move themselves. Their properties can be detected in different per-
ceptual modalities. They do not pop in and out of existence but move through
space-time on a single connected path. They are resistant to pressure and have a
certain mass that governs their interactions with other individuals. No more than
one body can be in one place at a time. They are susceptible to gravity and interact
with each other through transmitting energy. And so forth. These are higher-order
similarities that are not immediately perceptually salient but to which a creature
must be sensitive if it is properly to be described as perceiving and acting within a
suitably reified and structured world.
Earlier I suggested that a creature might be operating at what, following Straw-
son, I called the feature-placing level of experience even though its perceptual sys-
tems had solved the binding problem, as it is standardly understood within cogni-
tive neuroscience. We are now in a much better position to get a grip on what this
might amount to. Suppose we have a creature that perceives the world as com-
posed of bounded segments whose boundaries map onto the boundaries of the ob-
jects in the perceived environment, and moreover that it clusters perceptible prop-
erties in more or less the right way. That is to say, to put it in Triesman's terms, the
representation of redness gets bound to the representations of size and shape in a
way that matches the distribution of the relevant features in the environment. It
seems to me that this does not imply any perceptual sensitivity to the higher-order
physical regularities that govern the behavior of objects. The binding can take
place without there being any such sensitivity.
Here is a very simple illustration. One of the characteristic regularities govern-
ing the behavior of objects is that they tend to maintain their shape and size over
relatively short periods of time. I shall call this the object-property of constancy
(I'll say more about this terminology shortly). It is clear what is involved in per-
ceptual sensitivity to the object-property of constancy. The perceptual systems of
normal perceivers automatically compensate for variations in the size and shape of
the image on the retina produced by a particular distal object—the well-known
phenomena of size and shape constancy. After a very early age, normal humans,
for example, do not see a person diminishing in size as he moves away from
them—what they see is a person who retains a constant size moving into the dis-
tance. Similarly, we are all capable of extracting a single perceptual form as an ob-
ject's shape despite variations in its orientation. But it does not seem to be part of
solving the binding problem that the perceptual system should be sensitive to
shape constancy and size constancy. We can appreciate this by thinking about how
we would expect a perceptual system that is not sensitive to shape constancy and
size constancy to operate. We would not expect such a system to produce the to-
tally chaotic representations that would result if it had completely failed to solve
the binding problem. Nor would we expect it to come up with the sort of illusory
conjunctions that we considered earlier—illusory conjunctions in which, say, color
features and size features are mismatched. All we would expect is that the system
would not make the sort of compensations that we make. An organism with such a
perceptual system would not see a retreating object as retaining its size—nor
would it perceive a rotating object as retaining its shape. Nonetheless, it would still
see a persisting cluster of coinstantiated features.
An example will help make the point. There is strong experimental evidence
that size constancy is an acquired feature of infant perceptual systems. The experi-
mental paradigm was originally developed by Thomas Bower (1964, 1966), al-
though it turned out that it proved impossible to replicate his early findings of size
constancy as early as 8 weeks of age. I will briefly sketch it out, as it is important
for the point I want to make. The paradigm involved operant conditioning. Infants
were conditioned to respond to visually presented cubes of a particular size. The
reinforced behavior was head movement, and the reinforcing was the appearance
of the infant's mother.
The infants were conditioned to respond when a cube 30 cm in diameter was
presented 1 m in front of them (see fig. 4.1). The question that the experimenters
posed was how this response would be generalized—on the standard assumption
that a generalized response reflects a perceived similarity. Would the response gen-
eralize across similarities in retinal image size? Or would it generalize across sim-
ilarities in real size, thus showing that the infants were compensating for differ-
ences in size of the retinal image and deploying a form of size constancy? It turns
out (McKenzie, Tootell, and Day 1980) that size constancy is present at 6 months
although not much before.
Consider what happens in experiments that fail to detect size constancy in
young infants, such as those of McKenzie and Day (1972). Using a version of the
paradigm just described, McKenzie and Day conditioned infants between 6 and 20
weeks to one object at a set distance. They found that infants failed to generalize
their response when an object was presented at a different distance. The simple
Figure 4.1 (a) Apparatus used by Bower (1964, 1966) to test for depth perception and size
constancy, (b) Test stimuli showing true sizes, distances, and corresponding retinal image.
Reproduced by kind permission of Blackwell Publishers from Bremner 1988, p. 78.
point I want to extract is that the original process of conditioning would only work
if the infant perceptual system had solved the binding problem—that is to say, if
the features of the cube were all represented in a suitably integrated cluster. The
very possibility of showing that infants fail to detect shape constancy requires as-
suming that the infant perceptual systems have solved the binding problem. Yet,
since perceptual sensitivity to shape constancy is integral to the perception of an
object, it seems clear that solving the binding problem does not guarantee object
perception. There is more to perceiving an object than perceiving clusters of coin-

stantiated features.
What we need to do, therefore, to test whether reification can emerge in the ab-
sence of the linguistic machinery of quantification, is to investigate the extent to
which nonlinguistic creatures can properly be described as perceptually sensitive
to reified bodies. As Quine's work shows, one way of understanding the move
from bundles of features to reified bodies is through the idea that bodies are bun-
dles of features with a single common bearer. We might term this the traditional
metaphysical approach to individuation, in honor of its distinguished ancestry
within philosophy. If we adopt the metaphysical approach then something like
Quine's account seems very persuasive, because it is hard to see how a creature
might get a grip on the idea of a bundle of features having a common bearer with-
out explicitly positing the existence of a common bearer—and hard to see how
such explicit positing could take place without the machinery provided by the lin-
guistic tools of pronominal reference and quantification. But Quine's thesis looks
less compelling if we shift from viewing reified bodies as bundles of features with
a single common bearer to viewing them as bundles of features that obey certain
general higher-order physical principles. I will term this the higher order principles
approach to individuation. In the next section I will show how this can be put to
work to resolve the ontological indeterminacy associated with the ascription of
thoughts to nonlinguistic creatures.
4.4 Dishabituation and Higher-Order Principles

How might we employ the higher-order principles approach to individuation to in-
vestigate the process of reification in nonlinguistic creatures? The dishabituation
paradigm in developmental psychology has produced striking and suggestive re-
sults in the study of nonlinguistic infants (Baillargeon 1995, Gopnik and Meltzoff
1997, Spelke 1990) and nonhuman animals (Hauser 1998, 2001, Munakata et al.
2001). The reasoning behind dishabituation experiments is that nonlinguistic crea-
tures look longer at events that they find surprising and hence that one can read
back from their looking-times to the expectations that they have about how events
should turn out. So, by habituating nonlinguistic creatures to a given event and
then presenting them with events that differ from that event in specified ways, one
can detect those features of events to which they are sensitive. What has emerged
from experiments with the dishabituation paradigm on very young infants (long
before even the first stages of language acquisition) is that sensitivity to certain
very basic physical principles starts at birth. The process of reification is far ad-
vanced long before the acquisition of language. The dishabituation paradigm has
also been used to investigate the perceptual expectations of higher primates, partic-
ularly rhesus monkeys and cotton-top tamarins (Hauser et al. forthcoming). The re-
sults of these experiments provide strong evidence that reification precedes lan-
guage acquisition in phylogenetic as well as ontogenetic terms (Munakata et al.
2001). In this section I will explore some of the key findings about infant cognition
achieved through the dishabituation paradigm. The extension of these findings to

nonhuman animals will be considered in the next section.
Elizabeth Spelke has suggested that human infants perceive individuals as spa-
tially connected bodies that retain their connectedness as they move. She maintains
that the perceptual universe of infants as young as 3 months is governed by what
she calls the principle of cohesion, according to which surfaces lie on a single body
if and only if they are connected (Spelke 1990, Spelke and Van de Valle 1993). It is
evidence for the principle of cohesion, for example, that infants do not appear to
perceive the boundary between two objects that are stationary and adjacent, even
when the objects differ in color, shape, and texture. Spelke's dishabituation experi-
ments have shown that infants perceive two objects as one if they are spatially con-
tiguous, even if they do not display the sort of Gestalt properties that adults take to
define the boundaries of stationary objects—such as color similarities, smoothness
of edges, and so forth. That is, in experimental set-ups such as those in figure 4.2
young infants find the righthand scenario in the two test conditions more surprising
than the lefthand scenario.
This is particularly striking in condition (d), where the conflicting visual proper-
ties in the two halves of the stationary object might suggest to many adult viewers
that the righthand scenario is more likely.
Infants seem perceptually sensitive to principles governing the movement of
bodies. Unsurprisingly (since contiguity and connectedness dominate in their per-
ception of stationary objects), infants perceive a single object moving when two
adjacent objects move together (Spelke, von Hofsten, and Kestenbaum 1989). It
Figure 4.2 A schematic depiction of displays from an experiment on infants' perception of

object boundaries. Reproduced by kind permission of Blackwell Publishers from Spelke
and Van de Walle 1993,135.
also follows from the cohesion principle that infants will not amalgamate two ob-
jects separated by a visible gap even if they are moving together. They do, how-
ever, perceive a single object when the two objects are separated in depth. Spelke
explains this by suggesting that infant perceptions are governed by the principle of
contact, according to which surfaces move together if and only if they are in con-
tact. So, if two partly hidden surfaces move together and there is no visible gap
making it impossible for them to be in contact, the infants will expect them to con-
tinue to behave as if they were two surfaces of a single object.
It is another basic principle governing the behavior of enduring bodies (as op-
posed to sensations or sounds, for example) that they should continue to exist
when unperceived. There is some evidence from infant behavior that they have at
least the beginnings of a grip on this principle. Some of this evidence is anecdotal.
One might think that infants reveal sensitivity to this higher-order principle when
they move their head one way to regain sight of an object they lost sight of when
they moved their head in the opposite direction. Some of it is more experimental.
For example, 5-month-old infants will continue to reach for a previously seen ob-
ject when the room lights are put out (Hood and Willats 1986), although this may
well be due to an executive inability to inhibit an already launched prepotent re-
sponse. Experiments have also shown that infants will look in anticipation toward
the far end of a screen when an object disappears at the other edge, and will show
surprise when it does not appear at the appropriate time (Moore, Borton, and
Darby 1978). But perhaps the most clear-cut evidence in this area has come from a
famous set of dishabituation experiments carried out by Renee Baillargeon (1987).
In one experiment, Baillargeon habituated 4.5-month-old infants to a screen rotat-
ing 180° on a table, rather like a drawbridge. She then, in full view of the infant,
placed a stationary object behind the screen so that it was completely occluded by
the time the screen had been raised 60°. She was interested in whether the infants
would show surprise at trials on which the screen rotated 180° even though the ob-
ject had been placed behind it, as opposed to the trials in which the screen came to
a halt just at the point where it would be expected to encounter the invisible object.
It turned out, and has proved to be a very robust finding, that the infants were more
surprised by the first type of trial. It appears that they have some expectations
about the physical properties of objects even when those objects are currently in-
visible. Moreover, the drawbridge experiments show an appreciation that objects
have properties of solidity and impenetrability.
The dishabituation paradigm was deliberately designed as an alternative to the
more traditional Piagetian methods of testing infant's cognitive capacities as a
function of their motor capacities. It was felt that these methods could not ade-
quately distinguish performance failures due to lack of sensorimotor coordination
from those due to cognitive capacities not being sufficiently developed. Nonethe-
less, looking at infants' actions and reactions can provide rich evidence for the
richness of their perceived environment. The idea that individuals are bundles of
features that can be detected in different sensory modalities emerges very early on,
for example, and is manifest in infant behavior. Almost as soon as they are capable
of coordinating their movements, infants will move an object they hold into their
line of sight (Piaget 1954). Also appearing very early on in infancy is the capacity
for fairly accurate perception of whether objects are in reach, something that
clearly depends on perceiving the object as a locus of both visually and factually
identifiable properties (Field 1976, Von Hofsten 1982).
A final source of evidence in this area comes from experiments on infant percep-
tion of shape and size constancy. Bodies tend to preserve their shape and size over
relatively short time intervals. Since bodies move relative to perceivers (and per-
ceivers relative to bodies) the size of the retinal image is no index of the real size
of the body. The way perceptual systems automatically correct for this is known as
size constancy. Shape constancy is a similar phenomenon. Clearly, if infants are to
be properly described as perceiving a world articulated into bodies they must be
sensitive to the constancies that govern object perception in adults. As I have
noted, there is evidence that infants are sensitive to size constancy by about 6
months (McKenzie, Tootell, & Day 1980). Shape constancy, in contrast, has been
demonstrated not simply in young infants (Bower 1966, Caron, Caron, and Carl-
son 1979) but also in newborns (Slater and Morrison 1985).
The experimental evidence clearly suggests, therefore, that even young infants
are perceptually sensitive to a significant proportion of the higher-order physical
regularities to which bodies are subject. It seems right to conclude from this that
they inhabit a reified and articulated perceptual universe. The immediate conclu-
sion to draw is that nonlinguistic creatures are perfectly capable of perceiving a
structured world. It seems unlikely that the perceptual universe of young infants is
structured according to criteria of identity that are at all comparable to those em-
ployed by adult humans. It looks very much as if the principles of individuation
that young infants employ are criteria of identity for bodies in general, rather than
criteria of identity for particular bodies. This is consistent with experimental evi-
dence showing that, although infants younger than 9 months will show surprise
when a body disappears behind a screen and fails to emerge on the other side, they
seem unmoved if a different object emerges on the other side (Bower 1982). This
is not particularly surprising. But it does highlight the importance of clarifying
the differences between the way linguistic creatures apprehend the world and the
way nonlinguistic creatures do. I shall return to this in chapters 8 and 9, where I
will argue that there are radical differences between linguistic and nonlinguistic
cognition.
There is a further conclusion to draw from this examination of reification in
human ontogeny, and this brings me back to the discussion at the end of the last
section. I proposed there that we must be sensitive to the range of different ontolo-
gies that exist throughout the animal kingdom and through the course of infant de-
velopment. There is no one way that nonlinguistic creatures carve up the world.
What is now emerging is a way of making this basic thought concrete and determi-
nate. We can understand variations in ontology across the animal kingdom and the
course of infant development in terms of differential sensitivity to the set of basic
physical principles that govern the behavior of objects. What makes the ontology
of one species different from that of another, what makes it the case that they carve
the world up in different ways, is that they are sensitive to different higher-order
physical principles. In the next section I turn to making this basic thought more
precise.
4.5 From Object-Properties to Ontology

In the previous section I argued in support of the higher-order principles approach
to individuation. What makes it the case that a creature perceives agglomerations
of spatiotemporally coinstantiated properties and features as bodies is that it per-
ceives them as subject to certain higher-order physical principles. The results of
the dishabituation experiments surveyed in the previous section give us a clear il-
lustration in the case of human infants of what this perceptual sensitivity amounts
to. In this section I will explore how this conception of individuation can be em-
ployed to determine the ontology of nonlinguistic creatures, thus allowing us to
take the first step in resolving the indeterminacies of success semantics. The basic
idea is that we can understand a creature's ontology in terms of the particular set of
physical principles to which it takes bodies to be subject. The problem, of course,
is how we are to go about identifying the relevant set of physical principles—the
physical principles that form what I will term the physical parameter set. The easi-
est case is one in which, as with the infant example, the relevant physical parame-
ter set is largely a subset of our own physical parameter set. In this case we can
simply work backward from our own case, employing what I will term the canoni-
cal object-properties strategy—essentially subtracting from our own physical pa-
rameter set. But, as I have already shown in passing and will explore further below,
the infant physical parameter set contains higher-order principles that do not fea-
ture in our own. So the canonical object-properties strategy needs to be supple-
mented what I term the deviant object-properties strategy.
I shall start with the canonical object-properties strategy. The basic idea here (as
I discussed in The Paradox of Self-Consciousness})is to start with the propertiess
that are constitutive of something qualifying as an object. These are what I will call
canonical object-properties. The following are plausible candidates for the status
of canonical object-properties:
• The property of following a single continuous trajectory through space-

time
• The property of continuing to exist when unperceived
• The property of having a determinate shape
• The property of being internally unified
• The property of only being able to undergo a fixed set of changes
• The property of being impenetrable
• The property of being apt to fall when unsupported
• The property of being internally causally connected
• The property of having mass
• The property of posing resistance to the touch

• The property of having its state of motion or rest explicable in terms of me-
chanical forces acting on it
• The property of causally influencing other objects
The list of canonical object-properties is not fixed and rigid. There may well be
some objects that lack one or more object-properties, and it is possible that certain
object-properties will be reducible to others (as the property of presenting resist-
ance to the touch might be reducible to the property of impenetrability). Some of
the candidate object-properties might not be expected to apply to all objects at all
times. Objects that are lighter than air, for example, will not be apt to fall when un-
supported. Even more borderline are those objects, such as birds, that are capable
of overcoming the effects of gravity. Yet part of what it is to be perceptually sensi-
tive to an object-property is to be perceptually sensitive to the exceptional cases in
which certain object-properties cease to hold.
To each of these object-properties there obviously corresponds a higher-order
physical regularity of the type discussed in the previous section. So, for example, in
virtue of the fact that having a determinate shape is an object-property, there is a
higher-order physical principle that bodies have determinate shape. The proposal I
made in The Paradox of Self-Consciousnessswas that we can understand the wayy
prelinguistic infants carve up the world in terms of the object-properties (and correl-
ative higher-order physical principles) to which they are perceptually sensitive.
How might perceptual sensitivity to these higher-order physical principles be mani-
fested? Some will be directly manifested in the manner revealed by the dishabitua-
tion experiments. So, for example, perceptual sensitivity to the object-property of
being apt to fall when unsupported (and correlative physical principle that bodies
tend to fall to the ground when unsupported) might be manifested in surprise if an
object resting on a supporting surface stays where it is when the supporting surface
is removed. Sensitivity to other object-properties will be manifested indirectly. Con-
sider the object-property of having a determinate shape. One would not expect an in-
fant sensitive to this object-property to be surprised by a thing that lacks it. After all,
they might regularly be expected to encounter smoke rings, shadows, clouds, and
the like. But they might manifest their sensitivity to the object-property by showing
surprise when an object that lacks it nonetheless appears to possess other object-
properties, such as, for example, the object-property of impenetrability.
It seems possible, therefore, to construct an account of the ontological develop-
ment of the human infant in terms of perceptual sensitivity to an increasing range
of object-properties. That sensitivity to certain object-properties emerges at more
or less constant stages in human ontogeny seems undeniable. It would appear, for
example, that sensitivity to the object-property of causally influencing other bodies
emerges in some form at around the age of 7 months (Leslie 1984). However (and
this is something I failed to realize in The Paradox of Self-Consciousness), this ac-
count cannot be a complete account, even in the case of infants. Not all the "prop-
erties" to which infants are perceptually sensitive are canonical object-properties.
What is sometimes called the infant's naive physics is not simply an impoverished
version of our own. This is stressed in recent work by Alison Gopnik and Andrew
Meltzoff on the development of physical knowledge during infancy and early
childhood (Gopnik and Meltzoff 1997, particularly ch. 4).
Gopnik and Meltzoff identify a crucial difference between the criteria that
young infants (before the age of about 9 months) and adults deploy to determine
object continuity. For infants, movement information dominates information about
features and properties, so that their principal criterion for whether or not some-
thing is the same object is simply that it should maintain a single observable trajec-
tory, irrespective of any alterations that there might be in its general appearance.
This is one important reason why infants will not show surprise when one object
disappears behind a screen and a completely different object emerges on the other
side. For adults, on the other hand, as Gopnik and Meltzoff elegantly point out,
featural constancy can frequently trump continuity of trajectory in determining
whether or not two perceptions are taken to be perceptions of a single object:
As adults we individuate and reidentify objects by using both place and trajectory in-
formation and static-property information. We also use property information to pre-
dict and explain appearances and disappearances. If the same large, distinctive white
rabbit appears in the box and later in the hat, I assume it's the same rabbit, even if I
don't immediately see a path of movement for it. In fact, I infer an often quite com-
plex invisible path for the object. If I see the green scarf turn into a bunch of flowers
as it passes through the conjuror's hand while maintaining its trajectory, I assume it is
a different object. On the other hand, if an object changes its trajectory, even in a very
complex way, while maintaining its properties, I will assume it is still the same ob-
ject. (Gopnik and Meltzoff 1997, 86)
This example shows that it is a matter of some delicacy to characterize exactly

what object-property the infants are tracking. They are not tracking the object-
property of following a single, continuous trajectory through space-time. Rather
they are tracking a deviant object-property of spatiotemporal continuity that can be
best identified through its associated higher-order physical principle, namely, that
everything following a single trajectory through space-time is a continuously ex-
isting body. This is in contrast to the higher-order physical principle associated
with the canonical object-property of spatiotemporal continuity, which is that all
continuously existing bodies follow a single trajectory through space-time.
In the example just considered there is a canonical object-property on which the
deviant object-property can be modelled. The difference between them is essen-
tially one of direction of entailment. But there are more extreme examples of de-
viant object-properties elsewhere in the animal kingdom, where one might expect
to find an almost complete divergence from canonical object-properties. Some of
these will be a function of radically different sensory modalities—echolocation in
bats and dolphins, for example, or electric field sensitivity in various species of
freshwater and marine fishes or magnetic field sensitivity in pigeons. Others will
be a function of the different weightings between sensory modalities that hold for
other species. The overwhelming importance of olfaction in the cognition of ca-
nines and rodents is a case in point. Different species have evolved to be sensitive
to different types of information, and there is serious work to be done identifying
the relevant types of object-properties in a way that will allow us to move toward
an understanding of their ontology.
Most approaches to studying object representation in nonhuman animals have un-
derstandably been based on search- and behavior-based criteria. An elegant experi-
ment recently carried out by Marc Hauser and collaborators (2001) has offered very
suggestive evidence that cotton-top tamarins have a "gravity bias" analogous to that
of 2.5- to 3-year-old human infants—that is to say, they expect objects to fall down
in a straight line irrespective of their trajectory. An adult cotton-top tamarin was pre-
sented with an opaque S-shaped tube arranged so that the bottom opening was sig-
nificantly horizontally displaced from the top opening and with a box directly be-
neath each opening. The experiment was to drop an object down the tube in order to
see where the monkey would search for it. It turned out that the monkey consistently
searched in the box beneath the top opening (the entry point) rather than that beneath
the bottom opening (the exit point)—even immediately after having seen the exper-
iment carried out with a transparent rather than an opaque tube. When the tube is
transparent the monkey searches in the correct box (presumably because it has been
able visually to track the object as it falls through the tube), but as soon as the opaque
tube is reinstated the original behavior reappears.
A different set of experiments has illustrated that rhesus monkeys possess this
gravity bias to an extent that overrides sensitivity to the object-properties of so-
lidity and impenetrability (Hauser 2001). In these experiments rhesus monkeys are
placed in front of a table with two boxes, one of which is on top of the table and
the other below it (see fig. 4.3, step 1). The setup of boxes and table is hidden by an
occluder (as in step 2) and a food reward is dropped behind the occluder. The aim
of the experiment is to see where the monkey searches for the food. The food
reward has of course ended up in the top box, but it turns out that rhesus monkeys
almost invariably search in the lower box—even though to have ended up in that
box the food reward would have had to pass through both the upper box and the
table-top.4
There is considerable scope for applying the dishabituation paradigm to nonlin-
guistic creatures. Steps have already been taken toward using preferential looking
time to tap into the perceptual expectations of free-ranging rhesus monkeys. Mu-
nakata and collaborators (including Elizabeth Spelke) have explored the expecta-
tions that rhesus monkeys have about object motion. It turns out (Munakata et al.
2001) that monkeys appear to be more surprised by distinct objects moving to-
gether than they are by distinct objects moving conjointly. Just like human infants,
they do not amalgamate together objects between which they can see a spatial gap,
even when those objects are moving together.
Another potential model for sensitivity to object-properties in nonhuman ani-
mals is provided by research into the visual discriminatory abilities of pigeons
using the familiar paradigm of conditioning to color slides containing a variety of
visual scenes (see Walker 1983, 254-266, for a brief survey). It has been found
Figure 4.3 A schematic depiction of an experimental paradigm testing gravity bias in

rhesus monkeys. Reproduced by kind permission from Hauser 2001, 86.
that pigeons are able to distinguish scenes containing people from scenes not con-
taining people (Herrnstein and Loveland 1964); scenes with a particular individual
from scenes without that individual (Herrnstein, Loveland, and Cable 1976);
scenes with trees from those without trees (Cerella 1979); scenes with pigeons
from scenes with other birds (Poole and Lander 1971). It would be interesting to
investigate whether these experimental techniques might be adapted to plot dis-
criminatory abilities at the level of object-properties.
It is useful to draw a parallel with what Thomas Nagel termed objective phe-
nomenology in his well-known article "What is it like to be a bat?" Philosophical
attention has focused on the negative claims that Nagel makes in that article—
claims about the inadequacy of science for explaining conscious experience,
whether our own or that of other species, "from the inside"—neglecting the posi-
tive suggestions that he makes about the possibility of delineating the general
structural features of experience in a way that will allow them to be scientifically
illuminated.
At present we are completely unequipped to think about the subjective character of
experience without relying on the imagination—without taking up the point of view
of the experiential subject. This should be regarded as a challenge to form new con-
cepts and devise a new method—an objective phenomenology not dependent on em-
pathy or the imagination. Though presumably it would not capture everything, its
goal would be to describe, at least in part, the subjective character of experiences in a
form comprehensible to beings incapable of having those experiences. (Nagel 1979,
178-179)
Objective phenomenology is not concerned with the individual what-it-is-likeness
of experience, but rather with the general structure of experience shared by mem-
bers of a given species at a particular stage of development.
The strategy I am proposing in this chapter provides a concrete way of imple-
menting Nagel's proposal in this passage. Nagel calls for a way of understanding
how things are from a creature's point of view that does not depend on being able
to imagine how things are from that perspective. One way (the only way?) of doing
this is to work out the object-properties to which a creature is perceptually sensi-
tive. The developmental psychological work on object perception gives us an in-
sight into how the world is experienced from the infant's point of view, but it nei-
ther depends on nor brings us one iota closer to an understanding of what it is like
to be a 6-month-old human infant. The same point holds with even more force in
the case of creatures that are far more exotic relative to adult language-using hu-
mans than young infants are. We can, I think, get some sort of a grip on the object-
properties to which dolphins are sensitive, even though we will never be able to
imagine what it is like to be a creature that navigates the world primarily through
echolocation. To take a simple example, it is known that harbor porpoises (Pho-
caena phocaend) and Amazon river dolphins (Inia geoffrensis)s)an avoid sub-b-
merged wires that are only 1 mm in diameter (Evans 1973). Clearly some sort of
sensitivity to the object-properties of impenetrability and determinate shape is im-
plicated here. Further research will no doubt be able to tell us more.
5
Ascribing Thoughts to
Nonlinguistic Creatures
Modes of Presentation
There is a problem of indeterminacy in ascribing thoughts to nonlinguistic crea-

tures. There seem in any given situation to be indefinitely many different thoughts
that could be attributed to the nonlinguistic agent. How are we to narrow them
down to get either a unique content-ascription or, more plausibly, the sort of local-
ized indeterminacy that is perfectly permissible in our normal interactions with
other linguistic creatures? According to the approach I have adopted, the content of
a belief is determined by its utility condition, where a utility condition is a state of
affairs in the distal environment. But there are indefinitely many different ways in
which that utility condition can be described. Which description should we take as
the correct specification of the content of the belief? The problem of "narrowing
down" has several dimensions. One problem derives from ontological indetermi-
nacy. This was discussed in the previous chapter, where I showed how we might
go about attributing different ontologies to different categories of nonlinguistic
creature. The proposal that has emerged is that we can make progress calibrating a
creature's ontology in terms of the particular set of higher-order physical principles
that it takes to govern the behavior of bodies to which it is perceptually sensitive.
This gives us the resources with which to start narrowing down the range of possi-
ble content specifications.
Determining an ontology, however, can be no more than the first stage in resolv-
ing the problem of indeterminacy. At several points in this book I have emphasized
the importance of distinguishing the level of sense from the level of reference. At
the level of reference a utility condition is essentially a state of affairs. A single
state of affairs can be apprehended by a creature in many different ways, and in
order to explain how its representations of the environment cause it to behave in
the way it does, we must specify not simply the state of affairs it is representing but
also how it apprehends that state of affairs. Success semantics, as so far formu-
lated, offers us no way of choosing between the different modes of presentations
Modes of Presentation 89
under which a single state of affairs can be apprehended. If all we have to go on is

the idea that a belief is a function from desires to actions that tends to ensure the
satisfaction of desires with which it is associated, the best that success semantics
can offer us is a range of sentences reflecting the different ways that single state
of affairs might be apprehended. If we are to use success semantics as a tool to
capture the content of nonlinguistic thoughts, it needs to provide us with a single
sentence—a sentence that gives the determinate content of the belief in question.1
The question, therefore, is: How are we to go about eliminating those sentences
that do not correctly specify the mode of presentation under which the utility con-
dition is apprehended?
I shall term this the mode of presentation problem. This is, in effect, the problem
of explaining how we can give a canonical specification of a utility condition,
where a canonical description is one that reflects the way the utility condition is
apprehended by the creature in question. Such a canonical specification will be a
specification of a complete thought. The simplest type of thought is one that can be
specified in an atomic sentence. We can term this an atomic thought. It is a thought
to the effect that a certain individual (relative to a particular ontology) has a certain
property. In the case of atomic thoughts there will be two aspects to the canonical
specification of the utility condition, corresponding to the nominative and predica-
tive thought-constituents. Correspondingly, there will be two dimensions to the
mode of presentation problem. The first dimension is specifying a notion of mode
of presentation that can be deployed in giving the nominative element of a
thought-ascription—that is, to spell out what it is for a nonlinguistic creature to ap-
prehend a reified particular under one mode of presentation rather than another.
This is the topic of section 5.1. The second dimension of the mode of presentation
problem is explaining how the notion of a mode of presentation might be applied
in the predicative element of a thought-ascription—that is to say, how a given
property can be apprehended in different ways. This is explored in section 5.2. In
the final section of the chapter I provide an overview of how success semantics can
be deployed to make sense of the attribution of thoughts to nonlinguistic creatures.
5./ Individuals, Modes of Presentation, and the Sortal Problem

The utility condition of a belief is a state of affairs construable in purely exten-
sional terms. It can be adequately specified even though it does not reflect the or-
ganism's perspective on the world. Yet, although the extensional utility conditions
of a belief might be what secures the satisfaction of the desires with which that be-
lief is associated, we cannot understand why a creature acts the way it does, or pre-
dict how it will act, simply with reference to a purely extensional utility condition.
The notion of a belief's utility condition has two roles. On the one hand it has to be
such that true beliefs will cause actions that satisfy desires. For this the purely ex-
tensional notion will do. On the other hand, it has to explain why the believer acted
the way it did (or to predict the way the believer will act). And for this a purely ex-
tensional characterization will not suffice. It is clear that a creature might see an
object one way and not be impelled to act, whereas seeing it in a different way
leads immediately to action. The same animal can be seen as a threat or not. A
water hole might be an affordance of water at one time of day and an affordance of
danger at another. Different actions will be appropriate to different ways of appre-
hending the animal or water hole.
Authors who have proposed versions of success semantics have tended not to
address this problem. We can take Robert Stalnaker, whose version of success se-
mantics (1987) is put forward with particular reference to the scope it offers for ex-
plaining the behavior of nonlinguistic creatures, as a stalking horse. Here is Stal-
naker's account of desire and belief
To desire that p is to be disposed to act in ways that would tend to bring it about that
p in a world in which one's beliefs, whatever they are, are true. To believe that/? is to
be disposed to act in ways that would tend to satisfy one's desires, whatever they are,
in a world in which p (together with one's other beliefs) were true. (1987, 15)
Stalnaker's comments elsewhere in the chapter make clear that he is not adopting a
purely dispositionalist, or behaviorist, account of belief and desire. Beliefs and de-
sires are the internal states, whatever they might be, that cause the appropriate type
of actions. So, for Stalnaker, as for the version of success semantics spelled out in
the previous chapter, true beliefs cause actions that satisfy desires, and desires
cause actions, which are successful when beliefs are true.
Stalnaker's version of success semantics is linked to a particular version of pos-
sible worlds semantics. In straightforward possible world semantics the content of
a belief is given by the possible situations in which it is true, so that the contents of
beliefs are functions from possible worlds to truth-values. In the success semantics
version, the content of a belief is given in terms of the possible situations in which
the relevant desires with which it is conjoined are satisfied. This will be its utility
condition. In psychological explanation, therefore, specifying the content of a
belief is specifying the range of possible situations that collectively make up the
utility condition of the belief. But Stalnaker is explicit that, in specifying this
utility condition, we do not need to make any reference to how that utility condi-
tion is apprehended by the creature whose behavior we are trying to explain:
It is essential to rational activities such as deliberation and investigation that the par-
ticipants represent alternative possibilities, and it is essential to the role of beliefs and
desires in the explanation of action that the contents of those attitudes distinguish be-
tween the alternative possibilities. The particular ways in which alternative possibili-
ties are represented, or the particular means by which distinctions between them are
made, are not essential to such activities and explanations, even if it is essential that
the possibilities be represented, and the distinctions be made, in some way or other.
(Stalnaker 1987, 23)
It is a crucial element in Stalnaker's possible worlds approach to prepositional atti-

tude explanation that we as theorists should distinguish between the nature of a
proposition and the way that proposition is determined. So, what makes it the case
that a particular set S of possible situations is the utility condition for a belief is
that the believer is somehow capable of representing the set of relevant possible
situations (which need not, of course, be all the situations that could arise) and di-
viding that set into two parts, one of which corresponds to S. Stalnaker is clear that
there is some psychological story to be told about how the set of possible outcomes
is partitioned into the two subsets that jointly determine the utility condition. He is
equally clear, however, that the psychological mechanisms that will be appealed to
in explaining how this partitioning takes place do not form part of the belief. The
belief is simply a function from possible states of the world to truth-values.
In fact, Stalnaker maintains that this (broadly speaking) extensional approach to
belief-content yields particular advantages for psychological explanations of the
behavior of nonlinguistic creatures. It allows it to be the case, for example, that
two creatures with wildly differing conceptual repertoires could entertain the same
proposition, provided that, for some limited subset of the total possible ways in
which things could turn out, they both partition that subset in the same way. There
is no requirement that they partition the subset in the same way for the same rea-
sons. And this means that we can deploy our own highly complex language and the
conceptual scheme that goes with it to characterize the beliefs of creatures whose
cognitive perspective on the world might be vastly more primitive than our own:
When we ascribe attitudes to an animal, what we presuppose is simply that it has

some mechanism for representing and distinguishing between alternative situations.
It may be a very crude one, recognizing only gross differences, and containing no
representation at all of possible situations radically different from the way things are.
The conceptual space necessary to account for its capacities and dispositions might
consist of just a few relevantly different "ways things might be." But our language,
with its complex structure and capacity to make more subtle discriminations and de-
scribe more distant possibilities, may be used to distinguish between the few alterna-
tive possibilities represented by the dog without thereby attributing our concepts to it.
Because propositions do not mirror the structure of sentences that express them, it is
possible to use sophisticated, semantically complex sentences to ascribe attitudes to
creatures with very limited cognitive capacities. (Stalnaker 1987, 63)
There are clear advantages to understanding belief-contents in the way that Stal-
naker understands them. If we follow him in sharply distinguishing the nature of
the proposition from the way it is determined then we may well succeed in closing
the gap between the beliefs of language-using creatures and the beliefs of non-
language-using creatures (particularly if one follows him in holding that identity of
belief-content can be determined relative to a particular set of possible worlds,
rather than relative to all possible worlds).
But there is a price to pay. The notion of belief-content thus obtained is no
longer a powerful tool for psychological explanation. What is psychologically mo-
tivating, what feeds directly into action (hence what needs to be identified if action
is to be explained) is more likely to be found at the level at which a particular
proposition is determined, rather than at the level of the proposition itself. What
makes it the case that I act on the basis of my belief (in the light of my desires) is
that I have a way of categorizing the relevant class of possible situations that al-
lows me to go about working out whether or not the actual situation falls into that
class. Suppose we want to explain why the gazelle acts on its belief that all lions
are dangerous in a given situation. It will not be sufficient to say that the gazelle is
related to a proposition that divides a range of possible situations up so that none
of them contain a lion without that lion being dangerous. It might well be the case
both that the gazelle is related to that set of possible situations and that the situa-
tion confronting it falls within the set without the gazelle acting appropriately. The
gazelle will only act if (to put it in Stalnaker's terms) its general belief is such as to
allow it to identify the situation it perceives as falling within the range of possible
situations. And this is a matter of how the general belief is determined, rather than
of its content. It is a function of how the gazelle represents lions and how it dis-
criminates situations in which lions occur from those in which they are absent—
and of course of how it represents danger and discriminates situations that are dan-
gerous from those that are not.
It would seem, therefore, that our version of success semantics needs to be fur-
ther supplemented by showing how it can capture the mode of presentation under
which utility and satisfaction conditions are apprehended. The first stage in ex-
panding success semantics in this direction is to explain what it might be for a non-
linguistic creature to apprehend an object in a particular way. The ontological is-
sues discussed in the previous chapter go some way toward addressing this issue.
Part of what it is to apprehend an object is to apprehend it as something that obeys
certain basic physical principles, and this, as I showed earlier, can be understood at
least in part in terms of perceptual expectations about how that object might be-
have. So spelling out the particular ontology in terms of which a nonlinguistic
creature carves up the world goes some way toward resolving the first dimension
of the mode of presentation problem.
There is a problem here, however. Part of what it is to apprehend an object in
a particular way is that it should be apprehended as a member of a particular
category—that it should be apprehended as falling under a particular sortal. Yet it
is hard to see how the ontological considerations broached in the previous chapter
fully address this issue. The proposal that the object-properties to which nonlin-
guistic creatures are perceptually sensitive determines the way they carve up the
world is directly analogous to saying that we carve the world up into objects (the
difference consisting, of course, in the range of object-properties to which we are
sensitive). And, while it is true that we do carve the world up into objects, it might
be argued that this is an artifact of certain other principles of organization and clas-
sification rather than a bottom-level fact about our experience of the world. So, for
example, it might be suggested that the concept objecttis simply not a sortal con--
cept and hence does not give us a way of carving up the world at all. Part of what it
is for something to be a sortal concept is that it should provide criteria for indi-
viduating and counting the number of objects falling under that sortal at a given
moment. So anybody competent with the concept chair might be expected to have
the resources to count the number of chairs present in a room. Yet it is often sug-
gested that the concept objecttis not like that at all (Wiggins 1997). It is impossible
to count the number of objects in a room until we know what is to count as an ob-
ject, and we will not be in a position to know that until we have more determinate
sortal concepts in play. The concept object, on this view, is no more than a pseu-
dosortal. So, a fortiori, the nonlinguistic equivalent of the concept object (that is to
say, the way of carving up the perceived environment determined by the range of
object-properties to which a creature is perceptually sensitive) cannot function in
anything like the manner of a sortal concept either. So we have not in fact got any
closer at all to making progress on the mode of presentation problem.
There are several points here. First, it is clear that some sort of principle of indi-
viduation and counting is required if we are to speak of the apprehension of par-
ticulars at all. That is one of the principal differences between the feature-placing
level of experience and the particular-involving level of experience. So if the argu-
ment is to be faulted, the faults must lie elsewhere. The problem comes, I think,
with the argument that the machinery of object-properties does not provide us with
criteria for counting and individuation. It is just false that it is indeterminate how
many objects there are in a room—or rather that there is any radical indeterminacy
that cannot be resolved in a principled way, depending on the requirements of the
particular situation. The qualification is important. Consider a lamp—an artifact
often mentioned in this context. Is a lamp a simple object? Or should we rather
view it as a complex object composed of a stand, an electric bulb, and a lamp-
shade? It is likely that there is no fixed answer. In some contexts one will answer
one way and in others the second way. Everything depends on which object-
properties are more salient in a particular context. It is clear, for example, that if
following a single continuous path through space-time is the salient object-prop-
erty then the lamp will count as a simple object. If, on the other hand, internal ho-
mogeneity and internal unification are the salient object-properties then one might
well wish to view it as a composite object composed of three simple objects. And it
is easy to imagine contexts in which one object-property would be more salient
than another. So, for example, if one is preparing an inventory of the house to give
to the movers, it is likely that spatiotemporal continuity would be dominant. After
all, what matters is that the lamp should turn up at the end of the move, not that its
three components should. But if, on the other hand, one is a lamp designer trying
to perfect a lamp design and entertaining the possibility of a different combina-
tion of base and stand, it is natural to think that the lamp would be classified as
a complex object and that the emphasis would be on coordinating the separate
components.
The issue here is really one about the relations between parts and wholes. What
in one context will be treated as parts of a single object will in other contexts
emerge as objects in their own right—and, of course, vice versa. Many cases of al-
leged indeterminacy are ones in which we are unsure how to count objects with de-
tachable parts. But this is properly described as context-dependence rather than in-
determinacy. The appearance of indeterminacy comes about in part because, when
we are invited to imagine ourselves, say, counting the objects in a room, not
enough of the context is filled in for us to have a handle on how the part-versus-
whole issue is to be resolved. Moreover, a form of fallacious thinking (what we

might call the subtraction fallacy) increases the appearance of indeterminacy.
Some objects are such that it is clear that, if a part of them is removed, one will still
be left with an object. So, for example, if I remove a painting from its frame, I am
left with the frame, which is, of course, an object. It would be wrong to conclude
from this that it makes sense to suppose that there were two objects occupying the
space on the wall where the framed painting used to hang.2 In most contexts it
would be more appropriate to think that there was one object that could come apart
into two parts, each of which could in certain circumstances properly be consid-
ered an object (although neither would be identical to the original object). But, of
course, if one did think that the framed painting could count as two objects, then it
would be much easier to think that there would be unavoidable indeterminacy
when one sets about counting the number of objects in a room.
One would expect, then, that just as the concept object can function in many
cases as a sortal concept that would permit individuation and counting, so too can
the nonlinguistic analogue of the concept object (that is, the implicit ontology de-
termined by the range of object-properties to which a creature is perceptually sen-
sitive). Nonetheless, perceptual sensitivity to a given range of object-properties is
not sufficieentto explain the particular manner in which an object is apprehended byy
a creature. The most primitive form of categorization is grounded in perceived
similarity. Seeing an object in a particular manner (under a particular mode of
presentation) is in large part a matter of seeing similarities between it and other ob-
jects. And so for a creature to apprehend an object under a particular mode of pre-
sentation is both a matter of apprehending it relative to a particular ontology (as
determined by the basic processes of reification and carving up the perceived envi-
ronment into discrete units obeying certain high-level physical principles) and a
matter of perceiving similarities between that object and other objects.
The notion of perceived similarity here is being understood in a broad sense, to
include both featural similarity and higher-level similarities of the sort that might
best be captured through the concepts of instrumental properties and affordances
(as discussed in chapter 3). So, for example, a nonlinguistic creature might identify
a conspecific on, broadly speaking, featural grounds—that is to say, as falling
within a class of creatures sharing a common set of immediately perceptible physi-
cal properties, which might be detectable in a range of different modalities. On the
other hand it might be sensitive to similarities that run counter to such immediately
perceptible featural similarities. Things that have no little or no immediately
salient featural similarities can nonetheless resemble each other in respect of being
potential prey—or for that matter potential predators. Part of what it is for a non-
linguistic creature to apprehend something as a predator (under the predator mode
of presentation) is that it should be sensitive to similarities between that thing and
other potential predators, similarities that can at best only be mediately perceived.
Sensitivity to these mediately perceptible similarities will manifest itself in same-
ness of behavioral response, so that things that are perceived under the same mode
of presentation will elicit the same types of response (in appropriate contexts).
And, of course, from the third-person perspective of the ethologist or developmen-

tal psychologist, similarity of behavioral response is the principal clue to identify-
ing the mode of presentation under which a particular object is apprehended. But it
is the fact that things of a certain type (relative to the perceiver) are perceived
under the same mode of presentation that grounds this sameness of response,
rather than the sameness of response constituting their being perceived under the
same mode of presentation.
To draw the strands of this section together, the idea of apprehending individuals
under particular modes of presentation is applicable to nonlinguistic creatures to
the extent that they apprehend reified individual as objects* of certain types. The
object* locution is intended to serve as a place-holder, to do justice to the fact
that different types of nonlinguistic creature will carve their environment up in
different ways as a function of being perceptually sensitive to different object-
properties. The essence of perception under a particular mode of presentation
comes because different nonlinguistic creatures will perceive different types of
similarity between these objects*. Any object* can stand in a range of similarity
relations, correlated with properties that are both mediately and immediately per-
ceivable. Different such properties will be salient for a given creature at different
times, or for different creatures at a given time. As a consequence, a creature will
be capable of acting in different ways vis-a-vis a given object at different times,
and different creatures will be capable of acting differently toward the same object
at the same time. The particular way a nonlinguistic creature acts will be a function
of the particular similarity-based mode of presentation that dominates at that mo-
ment. Incorporating this understanding of mode of presentation into the way we
specify utility conditions brings us closer to meeting the requirements of psycho-
logical explanation. The next step is to extend the notion of mode of presentation
from the apprehension of objects to the apprehension of properties, thus making
available the idea of a complete utility condition being apprehended under a given
mode of presentation.
5.2 Structured Thoughts and Psychological Explanation

In discussing the minimalist approach to nonlinguistic thought in chapter 3 I
showed that the practice of giving psychological explanations of the behavior of
nonlinguistic creatures rests crucially on those thoughts being structured. Psycho-
logical explanations can only be predictively useful and genuinely explanatory
if the thoughts that they involve are composed of distinguishable thought-
constituents that can feature in further thoughts. Integral to genuine explanations
are predictions about how the agent would behave in relevantly similar situations,
under the guiding assumption that there are significant constancies in the way the
agent represents the environment across these situations. An animal will continue
to pursue the same goal when it encounters it in two different environments if and
only if it recognizes it—recognition that is reflected in the fact that there is a single
content-constituent across the two cases. Nor, of course, do representational con-
stancies hold simply across perceptual states. They are required if instrumental be-
liefs are to be anchored in perceptions of the environment, as well as providing
links between motivational states and perceptual-doxastic states. Behavioral con-
stancies and representational constancies go hand in hand. But the availability of
these representational constancies is dependent on the existence of structured
thoughts. The key epistemological question for the theorist of nonlinguistic
thought is how one goes about attributing such structured thoughts.
The discussion in the previous section took us part of the way toward answering
this question by illustrating what it might be for a nonlinguistic creature to appre-
hend particular objects* under distinct modes of presentation. But, although dis-
cussions of the notion of sense and mode of presentation in the context of natural
language have tended to concentrate on proper names, it is clear that we will
not have the materials for understanding the machinery of complete thoughts
unless the notion of mode of presentation can be applied to the apprehension of
properties—and hence can feature in the predicative component of a thought. In
this section I will explore how this might be achieved, by working in some detail
through an example of how a theorist trying to apply success semantics to explain
a particular behavior might resolve a particular type of indeterminacy arising pri-
marily with respect to the predicative component of a thought.
Let us imagine, then, an animal behaviorist or a cognitive ethologist faced with
a particular action that she is trying to explain. Discussion in earlier chapters has
shown how such a person might begin. We can assume that the observer has estab-
lished that the behavior is indeed one that requires psychological explanation. The
behavior displays an appropriate degree of flexibility of response to the immediate
environment and does not seem to be an instance of a fixed behavior pattern, or
the result of a straightforward pattern of instrumental conditioning. The first step
would be to work out what goal the animal was pursuing when it acted the way it
did. Was the animal foraging, pursuing a mate, look for a nest location, escaping a
predator? Once the goal has been identified with reasonable confidence, the next
step is to identify the relevant utility condition—the state of affairs whose holding
would, in normal circumstances, bring about the satisfaction of the relevant moti-
vational state. The utility condition, of course, may or may not actually hold, de-
pending on whether the action is or is not successful. It is no use, however, de-
scribing this utility condition relative to our own conceptual scheme. It needs to be
described in a way that reflects how it is apprehended by the creature in question.
Let us suppose that we have a range of possible content-ascriptions for our can-
didate behavior—content ascriptions reflecting what we know of the agent's ontol-
ogy. These content-ascriptions are, of course, different descriptions of the relevant
utility condition. The variation between them will not be at the level of the indi-
vidual objects* that are picked out by the singular terms and common nouns. We
can expect these to remain constant across the different descriptions. Let us sup-
pose also that we have some understanding, along the lines suggested in the previ-
ous section, of the particular mode of presentation under which those objects* are
being apprehended. In order to move beyond this to select a canonical specification
of the utility condition, we need a principled way both of identifying the properties
that the creature represents these individuals as bearing, and/or the relations in
which it represents them as standing, and of identifying the particular way it appre-
hends those properties and relations.
It is at this point that we must turn back to the considerations that lead us to the
need to ascribe structured thoughts—and in particular the idea that behavioral con-
stancies require representational constancies of the sort reflected by identical
content-constituents featuring in a range of different thoughts. The key idea here is
that the content-constituents that reflect the properties and relations apprehended
by a creature must be projectable. They must be capable of performing genuine ex-
planatory work when embedded in different thoughts operative in different con-
texts. How can this be determined?
I will describe the procedure in the abstract and then give a concrete example of
how it might be applied. We will need to identify comparable sets of sentences de-
scribing the utility conditions for other behaviors salient to the target behavior—
such as, for example, how the animal acts in different but structurally similar envi-
ronments, or how it acts when the instrumental contingency between behavior and
goal ceases to hold, or how it acts in similar situations when it no longer has the de-
sire in question. If the original set of sentences describing our target behavior had n
different predicative content constituents, then each set of sentences describing the
utility conditions of salient behaviors must also have the same range of predicative
components. Clearly, all members of the original set of sentences (which we can
term the target-set) are possible candidates for the psychological explanation of the
target behavior—more accurately, they specify the content of beliefs that could have
caused the target behavior. However, the same does not hold for all members of the
sets of sentences describing the utility conditions of salient behaviors (the salient-
sets). Some of these sentences will quite simply not be appropriate for the behaviors
with which they are matched. Let us assume, for simplicity's sake, that the sentences
in the target-set and their equivalents in the salient-sets are atomic sentences and
hence differ in only one content-constituent. This will, of course, be the predicative
component (since, as mentioned earlier, we can assume constant singular terms and
common nouns across the target-set and salient-sets). When we have a common
content-constituent across a target-set and salient-set that is potentially explanatory
of the target behavior but is not potentially explanatory of the salient behavior, we
shall say that the content-constituent fails to project. Conversely, of course, content-
constituents that project are ones that are common to the target-set and the poten-
tially explanatory subset of a given salient-set.
In order to determine which of the descriptions of the utility condition of the tar-
get behavior captures how that state of affairs is apprehended by the agent, we look
for the member of the target-set that has the most projectable content-constituent.
That is to say, we look for the member of the target-set that has the greatest number
of genuinely explanatory counterparts in the salient-sets. This will give us the
mode of presentation of the relevant property or relation in virtue of giving the
simplest and most powerful explanation of how the animal behaves.
For a simplified illustration of how this might work, suppose that our target be-
havior is a rat's success in retrieving food in a cross-shaped maze. The rat has suc-
ceeded in learning how to run the maze and retrieve the food. Applying our stan-
dard methodology, we begin by identifying the motivational state driving the rat's
behavior. Pretty obviously this will be a desire for food—a goal-desire rather than
a situation-desire, in my earlier terminology. The next step is to identify the utility
condition of the accompanying belief. The desire for food is satisfied because the
food is in the location in which it in fact is. I shall call this location P. The utility
condition, therefore, is that there be food at location P. As far as the utility condi-
tion of the belief is concerned it is immaterial how this location is coded and de-
scribed. Any frame of reference can be used to fix its location. Hence there will be
a range of different sentences correctly describing that location by plotting it rela-
tive to different coordinate systems. These sentences will be atomic sentences, es-
sentially attributing to the food the property of being at a particular location.3 The
sentences will differ only in their predicative components, and these differences
will reflect different ways of coding the location of the food. Each different coding
will reflect a difference at the level of sense, rather than at the level of reference.
The actual spatial location of the food is constant, but there is a range of different
ways that the rat might have grasped it. It is clear that we do not apprehend spatial
locations tout court. We apprehend them relative to different frames of reference
and coordinate systems. So too for the rat. The rat has learned how to navigate the
maze and remembers the way through. Only one of the atomic sentences correctly
captures the way the rat's memory codes the spatial location of the food. How are
we to identify which sentence it is?
The rat might be coding the location of the food as the end-point of a particular
series of bodily movements. That is, the rat might remember the particular set of
movements that took it to the food originally. This would not, strictly speaking, be
coding the location of the food in spatial terms at all, since there is no question of
identifying coordinates relative to a frame of reference. It would, however, be the
interpretation of the rat's behavior favored by hard-line behaviorists. There are fur-
ther distinctions that might be made deriving from different ways that a sequence
of movements might be apprehended—in terms of joint torques and muscle flex-
ions and contractions, for example, as opposed to a succession of individual move-
ment end-points. We can put these to one side.
Suppose, on the other hand, that the rat does code the location as a place defined
by its geometrical relations to other places. In this case there are different coordi-
nate systems it might be employing. It might, for example, be employing a coordi-
nate system based on an egocentric frame of reference whose origin is given by its
own bodily position when it originally began to run the maze. So it might fix the
location of the food using egocentric spatial predicates such as "to my right," "in
front of me," and so forth. This type of coordinate system is similar to the coding
of spatial location in terms of a chained sequence of bodily movements in at least
the following sense. However the maze is moved relative to the environment, the
coded spatial location of the food within the maze will remain constant relative to
the rat's starting point. Nonetheless, one would expect the two ways of coding to
lead to different behaviors if, for example, some of the spatial distances involved
were changed. Consider a very simple cross-shaped maze, and suppose that the
food is hidden in the eastern branch while the rat starts from the end of the south-
ern branch. Would a rat that has learned to retrieve food in this maze be able to re-
trieve food hidden in the same maze location in another cross-shaped maze in
which the southern branch happened to be twice as long? It seems likely that if the
rat were coding in terms of a sequence of bodily movements, it would not succeed.
Repeating the sequence of bodily movements will not get it to the food, since those
movements would take it only halfway up the southern branch. This would not
hold, however, for a rat that had coded the requisite lefthand turn at the end of the
southern branch in a body-relative coordinate system. The actual distance to be
traveled down the branch is of secondary importance on the egocentric system in
comparison with the body-relative direction of the turn at the end of the branch.
If the rat is not using an egocentric coordinate system and is not coding the loca-
tion in terms of a chained series of movements then it must be employing an allo-
centric frame of reference, a frame of reference centered on fixed points in the en-
vironment rather than on its own body. There seem to be two types of allocentric
coordinate systems it might be employing. It might be employing a coordinate sys-
tem centered on points external to the maze that can be seen from within the maze.
Or it might be employing a coordinate system centered on points internal to the
maze. The first, but not the second, allows the coded spatial location of the food to
vary with certain movements of the maze relative to the extramaze environment.
Both coordinate systems differ from the first two ways of coding spatial location
by allowing the coded spatial location of the food relative to the rat to remain con-
stant when the rat's starting position is changed. If the rat codes the location of the
food relative to, say a light bulb hanging over the section of the maze where the
food is, then it will still be able to find the food when it starts from a different part
of the maze. The same can hold if the rat has coded the location of the food relative
to a set of landmarks within the maze. In contrast, if the rat's spatial coding takes
the form either of a chained sequence of bodily movements or of a body-centered
coordinate system then changing the starting point of the maze will render it inca-
pable of returning to the location where it originally found the food.
The difference between these four different ways of understanding the predica-
tive component of the thought should be understood at the level of sense. At the
level of reference there is just one property—namely, the property of being located
at the relevant point in space.4 The utility condition of the belief (that is, the state
of affairs whose holding would secure the satisfaction of the rat's hunger) is the
state of affairs of the food's being located at that particular place (having the prop-
erty of being located at that particular place). The four different possible frames of
reference in terms of which the location of the food can be coded by the rat in the
maze represent four different modes of presentation under which the rat can appre-
hend the property of an object's being located at that place. The point here is that
thinkers and agents can only think about and act on points in space relative to a
frame of reference on which they are able to locate themselves.5 A location needs
to be represented in a way that will allow the agent to act relative to that location.
Descriptively adequate and predictively powerful psychological explanations need
to take account of the particular frame of reference employed by the thinking agent
whose behavior is being explained.
At the level of sense, therefore, there are four different ways of apprehending
the property of being located at that particular place—the property such that the
fact that it holds of the food determines the utility condition of the belief.6 This
gives us four different candidate beliefs about the location of the food, roughly
along the following lines:
(la) Food is located at the end-point of movements M1 n
(Ib) Food is located at coordinates (x, y) in egocentric space
(Ic) Food is located at coordinates (x', y') in maze space
(Id) Food is located at coordinates (x", y") in environmental space
We can break these four down to yield four possible content-constituents, corre-
sponding to the following four relational predicates:
(a') n—located at the end-point of movements—"
(b') n—located at egocentric coordinates (x, y)—"
(c') n—located at maze coordinates (x', y')—"
(d') "—located at environmental coordinates (x", y")—"
The simple behavior of the rat in the cross-shaped maze is neutral between all four
descriptions. The rat that learns to run the maze and successfully retrieves the food
when placed at the beginning of the maze could be employing any one of these dif-
ferent ways of coding the location of the food.
If we are to discriminate between them, we need to manipulate the parameters to
try to find out which specifications of the utility condition are suitably projectable,
in the sense discussed earlier. We need to identify a further set of navigation be-
haviors (what I earlier termed salient behaviors) that modify some of the relevant
parameters in a way that will discriminate between the different ways that the rat
might be apprehending spatial location. The guiding assumption here, of course, is
that the rat will employ the same general means of coding the location of the food
in the salient behaviors as in the target behavior. This seems plausible, and allows
us to capitalize on the fact that each sentence in the target-set generates different
predictions about how the rat will behave if, once it has learned to run the maze, it
is then placed in a differently configured maze. According to the chained motor
program interpretation, corresponding to sentence (la) and content-constituent (a'),
the rat will make the same series of movements that had taken it to the food in the
original maze. According to the egocentric coordinates interpretation, as reflected in
sentence (Ib) in the target-set and corresponding content-constituent (b'), the rat
will move to the same position relative to its starting-point as in the original maze.
The maze space view—in sentence (Ic) and content-constituent (c')—predicts
that the rat will go to the same position in maze space, while the environmental
space view—(Id) and (d')—predicts that the rat will go to the same position in
environmental space.
What we need to do, therefore, is modify the maze in a way that will generate a
series of further salient behaviors and allow us to see which of the four content-
constituents (a')-(d') is suitably projectable. There is, in fact, a relatively well un-
derstood experimental literature that can be interpreted as doing precisely that
(Gallistel 1990). In the following I shall adopt a slight simplification in describing
the experimental work on the coding of spatial location in rats. The experiments
depend on comparing performance in different mazes. Performance can be calcu-
lated in terms of the number of trials required to learn how to run the maze—or as
a function of the percentage of successful trials. It is essentially a statistical mea-
sure calculated on the performance of groups of rats, not always involving the
same rats in different mazes. Some experiments involve training rats to run a par-
ticular maze and then either changing the configuration of the maze (by blocking
some of the routes, for example, or by rotating the maze as a whole) or starting the
rats off in a different section of the maze from the one they started from in the
training runs. Other paradigms train groups of rats to run mazes in ways designed
to draw on different ways of coding spatial location. In this second type of experi-
ment what is important is the relative speed with which the respective groups of
rats learn to run the maze. I shall describe the experiments, however, as if they in-
volved simply training a single rat in one maze and observing its performance on a
single trial in another maze—even though it is highly unlikely that any single rat
has ever been run through the all of the mazes I am about to describe.
Recall that the behavior we are trying to explain is the successful performance
of a rat in a cross-shaped maze. The rat starts in one of the four arms of the maze,
while the food is located in a different arm. One obvious modification of the ex-
perimental setup would be to keep the location of the food constant while starting
the rat from a different arm. This experiment was famously carried out by Tolman
and collaborators in one of the first experiments to present a significant challenge
to behaviorist theories of location coding. Their variation was to train the rat start-
ing from one branch of the maze and then to observe how it behaved when started
from the opposite branch of the maze (Tolman, Ritchie, and Kalish 1946). I shall
term this the reversed start paradigm. As is well known, the rat successfully
reached the goal. Clearly, the rat would not have succeeded had it been coding the
location of the goal in terms of a chained motor program. Nor would it have suc-
ceeded if it had been coding the location of the food in egocentric behavioral space
coordinates. Let us suppose that in the original maze the rat began from the south-
ern arm, with the food reward located at the eastern arm. To obtain the reward the
rat must turn to the right at the center of the cross. In the experimental situation, in
contrast, the rat begins at the northern arm and to obtain the food has to make a left
turn at the center of the cross. On both the motor program and behavioral space
views, the rat would not turn left when coming from the north but would repeat the
original right turn, ending up at the western arm. Both of these codings would have
the rat making, from its own point of view, the same turn from the opposite branch
of the maze as it did from the original branch, and this would place it in the wrong
branch for the food.
In terms of our original candidate beliefs, the content constituents in (la) and
(Ib) fail to project to the reversed start paradigm. This leaves us with the two re-
maining beliefs. So, the candidate beliefs here are as follows.
(2c) Food is located at maze coordinates (x', y')
(2d) Food is located at environmental coordinates (x", y")
The predicative component in each of these candidate beliefs does indeed project
to the reversed start paradigm. We have already succeeded, therefore, in making a
preliminary separation of our four original candidate content-constituents. Yet we
still need to disambiguate between the two candidates that remain in play.
A further classic parameter variation is to alter the shape of the maze. So, in-
stead of starting the rat off from the opposite point of a cross-shaped maze, one
might train the rat in a T-shaped maze and then see how it performs when the maze
is shifted 180° (Tolman, Ritchie, and Kalish 1947). Extrapolating from behavior in
the cross-shaped maze, one might expect the rat to turn in the opposite direction to
the training schedule in order to return to the same "place." But this is not what
happens. The rats tend to make the same turn as in the training schedule. So it
might be appropriate to think that the rats are merely repeating the same sequence
of bodily movements. This supports the following.
(3a) Food is located at the end-point of movemn

They could also be returning to the same point in behavioral space, which supports:
(3b) Food is located at egocentric coordinates (x, y)
But it could of course be that the rats are returning to the same maze-location.
Turning the maze around 180° does not alter any of the geometrical relations inter-
nal to the maze. This would support:
(3c) Food is located at maze coordinates (x', /)
The rats cannot, however, be returning to the same location in environmental
space, because the geometrical relations between points within the maze and
points outside the maze are altered by moving the maze 180°. In this third behav-
ior, therefore, we have identified three projectable content-constituents.
For each of these behaviors we can, in our simplified paradigm, identify the pos-
sible content-constituents. We can plot them out and make a table as follows.
Behavior 1 (a) (b) (c) (d)

Behavior 2 (c) (d)
BehaviorS (a) (b) (c)
So, we have four possible ways of spelling out the utility condition of the rat's be-
havior in the first maze, depending on how we take the rat to be coding the location
of the food. We assume that the rat's coding of the place where the food is located
remains constant. As the table shows, there is only one way of spelling out the
mode of presentation under which the rat is apprehending the target place that
projects successfully across all three of the behaviors. We should conclude, there-
fore, that the rat is coding the location of the food within a coordinate system
whose frame of reference is determined by fixed points within the space of the
maze.
This gives us the resources to identify the content-ascription in the first behavior
that projects in an explanatory way. The sentence expressing a thought containing
content-constituent (c') should be taken to express the determinate content because
it allows the identification of a common content-constituent across the range of
behaviors—and we need a common content-constituent because the relevant de-
sires and cognitive background conditions remain constant across the behaviors.
Some caveats should be registered at this point. There is no reason to think that
a rat will always code locations in the same way. It is well documented, for exam-
ple, that rats can orient themselves in mazes by using nongeometric cues (Cheng
1986). Rats learn surprisingly quickly to run a radial maze in which the arms are
randomly permuted between each trial, provided the arms are featurally highly dis-
tinct. Similarly, it seems that the salience of landmarks in the nonmaze environ-
ment affects whether rats code locations on a maze-centered or environment-
centered frame of reference (see Cheng 1986, Gallistel 1990, ch. 6). The experi-
menter needs to ensure as close a match as possible between the target behavior
and the experimentally induced salient behaviors. Only thus can genuine pro-
jectability be established.
Nonetheless, and bearing these caveats in mind, working through this example,
in conjunction with the discussion in the previous section, shows how a theorist
might go about resolving the inevitable indeterminacies generated by applying
success semantics to a target behavior. These indeterminacies arise at the level of
sense. They derive from the different ways a creature might apprehend a given
utility condition. Success semantics alone cannot provide a determinate specifica-
tion of a utility condition. At best it will leave us with a choice between different
descriptions of the relevant state of affairs. Unless we break these descriptions
down into separable content-constituents and consider which of those content-
constituents are projectable in the manner described in this section, we will be un-
able to narrow the list of candidate beliefs down in the way required to respect the
requirement that a psychological explanation reflect the way a creature apprehends
its environment.
It should not be thought, however, that this type of disambiguation is possible
only in a laboratory environment. I have concentrated on the example of experi-
mental work on rat navigational abilities because it presents in particularly clear
form the basic structure of the strategy to be adopted in specifying a utility condi-
tion at the level of sense. But there is no reason at all why a similar strategy should
not be adopted in a more ecologically valid context to attribute beliefs to animals
in the wild. What I am calling salient behaviors can be identified by observation.
Salient behaviors involve the variation of parameters, and this is of course possible
without actively manipulating those parameters. Equally, it is likely in practice that
mixed strategies can be profitably pursued, deploying research done in experimen-
tal contexts on conspecifics to guide the attribution of thoughts to animals in the
wild. Indeed, it should be possible to deploy the salient behaviors of members of
other species, either in captivity or in the wild—provided that due account is taken
of the possibility that cross-species generalizations may be invalid.
5.3 Success Semantics and Nonlinguistic Thought: An Overview

The upshot of the discussion of minimalism in chapter 3 was that an important
class of nonlinguistic behavior requires explanation in terms of propositional atti-
tude psychology. The types of nonpropositional thinking explored by different
forms of the minimalist approach cannot explain all types of nonlinguistic behav-
ior. In this context, then, it is important to stress that success semantics is intended
to provide a means of extending propositional attitude psychology to nonlinguistic
creatures. It is proposed as a way of attributing beliefs and desires in the absence of
language, and as such, of course, it needs to satisfy the basic desiderata that hold
for any theory of propositional thinking, as opposed to the more imagistic and non-
propositional types of thinking discussed in the context of the minimalist concep-
tion in chapter 4.1 considered four such desiderata in 3.2.
Propositional thinking involves a sharp distinction between attitude and con-

tent—between the particular proposition that is the object of one's mental
state and the mental attitude one takes toward it.
Propositions must be independent of the particular context of thinking. A
theory of propositional thinking must make it possible for a thinker to have,
for example, mistaken beliefs and to entertain a thought without knowing its
truth-value.
For any propositional attitude, it should in principle be possible to find a sen-
tence that will specify the content of the attitude accurately and without
remainder.
The contents of propositional attitudes have a structure that permits them to
feature in inference and that is in some sense isomorphic to one way of appre-
hending the structure of the state of affairs that that content represents.
It will be helpful to use these four characteristics of propositional thinking to struc-

ture an overview of success semantics.
Success semantics, unlike the two models of propositional thinking considered
in chapter 2, is driven primarily by the relation between thinking and behavior. As
such it can be viewed as a form of philosophical functionalism.7 Beliefs, according
to success semantics, are causal functions from desires to actions. The content of a
belief is its utility condition, where a utility condition is the state of affairs whose
holding would result in the satisfaction of desires with which that belief is associ-
ated. True beliefs are such as to cause actions that satisfy desires. Clearly, desires
cannot on pain of circularity be understood as functions from true beliefs to ac-
tions, and success semantics has a different account of the content of desires. A
particular desire has the content that it does in virtue of its satisfaction-condition,
where the satisfaction-condition of a desire is the state of affairs whose holding
leads to the cessation of the behavior to which the desire gives rise. The state of af-
fairs has to be such, moreover, as to bring about the cessation of the relevant be-
havior in a manner that will make the creature more likely to pursue that behavior
in the future.
Nonetheless, even though success semantics offers different accounts of the
content of belief and the content of desire, it is able to satisfy the first two desid-
erata of a theory of propositional thinking. It is clearly the case that a creature can
bear different attitudes to the same content, since a single state of affairs can be
both the utility condition of a belief and the satisfaction-condition of a desire.
Moreover, the accounts of what it is to have the attitude of belief to that state of af-
fairs and the attitude of desire toward it are distinct, though related. As far as the
second feature of propositional thinking is concerned, the thoughts attributed
through success semantics are both susceptible to error and divorced from the im-
mediate context of thinking. Whereas the minimalist approach to nonlinguistic
thinking has difficulty explaining how there can be thoughts about states of affairs
that are not in some sense part of the "here-and-now," success semantics faces
no such restriction. The characterization of belief and desire is not in any sense
context-bound or limited to states of affairs that can be mediately or immediately
perceived. All that is required for a state of affairs to be the object of a belief or a
desire, respectively, is that its holding should result either in the satisfaction of the
desire(s) with which that belief is conjoined, or that it should be such as to extin-
guish in a reinforcing manner a particular pattern of behavior. There are no con-
straints of either spatial or temporal contiguity. Nor, of course, are beliefs guaran-
teed to be true or desires guaranteed to be satisfied. The satisfaction-condition of a
desire is the state of affairs that would satisfy it, and the utility condition of a belief
is the state of affairs that would result in the satisfaction of desires with which it is
conjoined. Success semantics allows us to work backward from behavior to beliefs
and desires without assuming that the behavior is always successful.
It would seem, then, that success semantics offers a way of understanding the
propositional attitudes that respects the basic features of the distinction between
content and attitude. As such it provides a response to some of the potential prob-
lems that emerged in the context of the Fregean conception of thoughts as the
senses of sentences. It will be remembered that these problems were also ad-
dressed by the language of thought hypothesis. However, as I showed in chapter 2,
the language of thought hypothesis has difficulties with the epistemological di-
mension of an account of thought. It provides no guidance on how we might go
about attributing thoughts to nonlinguistic creatures in the context of psychologi-
cal explanation. The discussion of success semantics in the previous two chapters
should have brought out, in contrast, the extent to which success semantics pro-
vides not simply an account of the nature of prepositional attitudes but also a prac-
tical framework for the attribution of those thoughts. Understanding the contents
of beliefs in terms of utility conditions and the contents of desires in terms of satis-
faction conditions clearly offers a way of working backward from behavior to the
thoughts that gave rise to it, since the content of any given propositional attitude
is determined by its behavioral ramifications. The constitutive account of what
nonlinguistic thoughts are yields a method for identifying and attributing such
thoughts.
However, there is a significant problem in applying the method of success se-
mantics. The objection that is perhaps most frequently raised to the ascription of
thoughts to nonlinguistic creatures, hence to the practices of psychological expla-
nation within which such ascriptions take place, is that it is impossible to ascribe
thoughts with suitably determinate contents (Davidson 1975). This is linked to the
third and fourth features of propositional thinking identified earlier. Together these
add up to the requirement that, for any propositional attitude, it should in principle
be possible to find a sentence that will specify the content of the attitude accu-
rately, without remainder and in a way that makes clear the structure of the thought
and hence its potential inferential role. How are we to do this for success seman-
tics? The guiding idea in success semantics is that the content of a belief is to be
given in terms of the state of affairs that would bring about the satisfaction of de-
sires with which it is associated. But there are indefinitely many ways that such a
state of affairs can be characterized. How are we to identify the correct specifica-
tion (or group of specifications)? This is the principal epistemological challenge
confronting the theorist of nonlinguistic thought. Most of my discussion of success
semantics has been devoted to showing how this epistemological challenge can be
met.
In the previous chapter I considered the first stage in meeting this challenge—
the stage of working out how the creature in question (or rather, the species and
stage of development of which it is a representative) carves up the world in percep-
tion and thought. I suggested that this might be done by working out the higher-
order physical principles to which the creature is sensitive, hence demarcating an
ontology for the creature whose behavior one is trying to explain, interpret, or pre-
dict. I showed in chapter 4 that different species (and individuals at different stages
of development within a given species) will reifyythe distal environment in differ--
ent ways, where a reified environment is one that is parsed into clusters of features
that the perceiver/thinker expects to behave in certain ways. We can, therefore,
make a start on resolving the indeterminacy by identifying the expectations that
the creatures will have about the behavior of bodies in its environment. This will
allow us to have a much clearer sense of the vocabulary in terms of which thoughts
should be attributed—and in particular, how the singular terms and common nouns
in thought ascriptions should be understood.
This can be no more than the first stage, however, because it reflects only the
way a particular type of creature carves up its environment. This ontology will, of
course, be shared by all conspecifics at the appropriate stage of development. This
is not enough for a resolution of the indeterminacy. Such a resolution will come
only if we have a way of obtaining a canonical description of a utility condition,
where a canonical description is one that reflects the way the utility condition is
apprehended by the particular creature in question. Such a canonical description
will be a specification of a complete thought, of the sort that can feature in psycho-
logical explanations of why that particular creature behaved the way it did in a par-
ticular context. A complete thought specified in this way will incorporate consider-
ations of cognitive significance.
Let us assume that we are dealing with the simplest type of thought, namely, an
atomic thought that can be specified in an atomic sentence. These are thoughts to
the effect that a certain individual (relative to a particular ontology) has a certain
property. In the case of an atomic thought, there will be two aspects to the canoni-
cal specification of the utility condition, corresponding to the nominative and
predicative thought-constituents. The first will be a specification of how the indi-
vidual in question is apprehended by the thinker—the mode of presentation in
terms of which it is given. And the second will be a specification of the property
that is being attributed to the individual. Once these two components have been
specified, then we can properly be described as having provided a canonical speci-
fication of a utility condition.
As far as the nominative component of the structured thought is concerned, it is
not enough simply to pick out an object (or what I termed an object*) in terms of
its general ontological characteristics. In order to explain the role that preposi-
tional attitudes play in psychological explanation, we also need to specify the
mode of presentation under which that individual is apprehended. The general on-
tological framework within which a creature experiences the world forms part of
its perspective on the world, but it does not provide the full explanatory power of
the notion of a mode of presentation. In particular, it does not help us to understand
how two different creatures might react differently to the same situation, or how a
single creature might react differently to the same situation at different times. We
need some sort of analogue at the nonlinguistic level of the notion of cognitive sig-
nificance as it features in psychological explanations of the behavior of language-
using creatures. I suggested in section 5.1 that we could move toward resolving the
mode of presentation problem by considering perceived similarities—in particular
the similarities that a creature perceives between the individual in question and
other individuals. The cognitive significance attached to an apprehended indi-
vidual can be understood in terms of the similarity-classes within which that indi-
vidual features. Different creatures will have different similarity-spaces, hence dif-
ferent ways of grouping things as, say, predators. Such perceived similarities (or,
for that matter, failures to perceive similarities) will manifest themselves in behav-
ior and are a crucial part of explaining what is distinctive in each individual crea-
ture's way of apprehending the world.
The notions of mode of presentation and cognitive significance are not only ap-
plicable to the nominative component of a thought, however. We also require some
understanding of how we might go about determining how a given property can be
apprehended in different ways—of how the notion of a mode of presentation might

be applied to properties. The principal issue here is how we interpret the way a
nonlinguistic creature apprehends the structure of a state of affairs. The strategy I
propose for thinking about how this extension might be achieved picks up on a
theme that has recurred at several points during this book—the interdependence
between the possibility of psychological explanation and the ascription of struc-
tured thoughts. The guiding idea is that the description of a utility condition can
be effected in different ways to yield a range of different predicative content-
constituents—that reflect different ways that a creature might be apprehending the
relevant properties in a state of affairs. The "correct" content-constituent is the one
that best projects across an appropriate range of actual and counterfactual patterns
of behavior. In the previous section I showed in some detail how this method
might be applied to explain the particular mode of presentation under which a rat
in a maze apprehends the property of being located at a particular place.
In terms of the overall project of providing a framework for understanding the
practices of giving psychological explanations of the behavior of nonlinguistic
creatures, I have reached the following position. The discussion of the minimalist
account in chapter 3 pointed the way toward an important way of understanding
many types of thinking behavior in nonlinguistic creatures. We can formulate the
majority of our psychological explanations at the nonlinguistic level in terms of
the sort of context-bound and perception-based representations of the environment
developed in, for example, Dummett's conception of protothoughts, Cussins's cog-
nitive trails, or Campbell's causally indexical thoughts. But, as I showed in chapter
3, the minimalist approach does not have the resources to accommodate all the
types of psychological explanation required at the nonlinguistic level. Some such
explanations will need to couched within the framework of propositional attitude
psychology. It is the problem of how to understand this type of explanation that I
have been pursuing in this and the previous chapter. The success semantics ap-
proach has shown what it is to attribute beliefs and desires at the nonlinguistic
level and how we should understand the contents of those beliefs and desires. I
have, therefore, made considerable progress on what in chapter 1 were identified
as the metaphysical, semantical, and epistemological dimensions of an account of
nonlinguistic thinking. I have, however, only made partial progress on what I
termed the explanatory dimension, namely, the task of elucidating the explanatory
practices in which the attribution of those beliefs and desires is embedded. What, if
anything, are the analogues in the nonlinguistic case for the considerations of ra-
tionality that operate so centrally in psychological explanations of the behavior of
language-using creatures? How do beliefs and desires in nonlinguistic creatures
translate into action? What, if any, processes of practical reasoning are we suppos-
ing at the nonlinguistic level? These and related questions will be addressed in the
following two chapters.
6
Rationality without Language
In the previous two chapters I proposed a way of attributing thoughts to non-

linguistic creatures. This has taken us a considerable way toward addressing the
problems identified in earlier chapters. It is becoming clearer how nonlinguistic
creatures might have, and how we might go about, identifying thoughts that have
determinate contents, are genuinely structured, and reflect the way the relevant
state of affairs is apprehended by the creature in question. But this still leaves us
some way short of an adequate theoretical framework for the practices of psycho-
logical explanation that we find in, for example, contemporary developmental psy-
chology, cognitive ethology, and cognitive archeology. In this chapter and the next
I complete the account by considering how we should understand the twin notions
of rationality and practical reasoning at the nonlinguistic level.
I begin in section 6.1 by explaining why it is imperative to give an account of ra-
tionality at the nonlinguistic level and why the task is a difficult one. The problem
is that the standard, inference-based models of rationality that seem appropriate at
the linguistic level are not applicable to nonlinguistic creatures. In sections 6.2 to
6.41 distinguish three different ways that behavior at the nonlinguistic level can be
characterized as rational. In 6.5 I discuss the implications of this tripartite distinc-
tion for psychological explanation at the nonlinguistic level.
6.1 Theoretical and Practical Rationality

A successful practice of psychological explanation requires more than the simple
attribution of thoughts. Psychological explanations operate by attributing preposi-
tional attitudes (typically a combination of beliefs and desires) that rationalize the
behavior being explained (Davidson 1963). The governing principle of the expla-
nation is that it would be rational for a creature with that combination of beliefs
and desires (and no significantly countervailing beliefs and desires) to act in the
way that it did in fact act.
The requirement here is not, of course, that the relevant action be completely ra-
tional in a full or everyday sense of the term. What matters is that it be rational
from the point of view of the agent. That is to say, the performance of the action
should make sense in the light of the agent's beliefs and desires. But it may be that
those beliefs, or some subset of them, are irrational in a way that makes the action
itself irrational. It may be helpful to distinguish explicitly between internal and ex-
ternal rationality. Assessments of internal rationality are relative to an agent's dox-
astic and motivational states, taking those states as given, while assessments of ex-
ternal rationality include assessments of the doxastic states underlying the action.
To say that an action is externally rational is to say that it is in some sense appro-
priate to the circumstances in which it is performed, where those circumstances in-
clude the agent's motivational states—with different theories of external ration-
ality interpreting the type of appropriateness involved here in different ways.
The internal rationalizing connection between beliefs, desires, and actions al-
lows the attribution of thoughts and desires to be genuinely explanatory. Beliefs
and desires cause behavior qua beliefs and desires (that is to say, in virtue of their
content) because their contents rationally dictate a single course of action—or a
limited number of possible courses of action. In the absence of such a rationalizing
connection, there would be no reason why a belief-desire pair with those particular
contents should cause that particular action.1 Clearly, therefore, any application of
psychological explanations to nonlinguistic creatures must rest on the appropriate-
ness of applying criteria of rationality to nonlinguistic creatures. But how are we to
extend the notion of rationality to nonlinguistic creatures?
The basic problem is that the models of rationality we possess are not easily
generalized to nonlinguistic creatures. Consider, for example, the influential ac-
count of practical rationality offered by Donald Davidson—what we might term
the inference-based conception of practical rationality:
was desirable. . . . If we can characterize the reasoning that would serve we will, in
provide him with the premises of an argument. (Davidson 1978,1980a, 85-86)
Let us suppose that practical reasoning is argument-like in the way that Davidson
and many others have suggested. Then it straightforwardly follows that we can
only explain and predict the behavior of other creatures to the extent that we can
understand the inferential relations between descriptions of their beliefs, desires,
and so forth on the one hand and their actions on the other. And it seems natural to
think that our understanding of the inferential relations between propositional atti-
tudes and actions depends on their conforming to the dictates of what might be
termed procedural rationality—that is, sensitivity to certain basic principles of de-
ductive and inductive inference (although such conformity is a necessary rather
than a sufficient condition). Obvious examples are the familiar deductive princi-
Rationality without Language 111
pies of modus ponens, modus tollens, contraposition, and so forth, together with
such basic principles of probability theory as that the probability of a conjunction
can never be greater than the probability of its conjuncts; that the probability of a
hypothesis and the probability of its negation should add up to 1; and so on.
There are two principal obstacles to extending this inference-based conception
of rationality to nonlinguistic creatures. The first obstacle concerns the structure of
the vehicles of nonlinguistic thought. We understand inference in formal terms—in
terms of rules that operate on representations in virtue of their structure. But we
have no theory at all of formal inferential transitions between thoughts that do not
have linguistic vehicles. Our models of formal inference are based squarely on
transitions between natural language sentences (as codified in a suitable formal
language). To be clear on the problem, we need to remember that there is an im-
portant distinction between two different ways thoughts can be structured. They
can be structured at the level of their vehicles or they can be structured at the level
of their contents. In arguing against the minimalist conception of nonlinguistic
thought in chapter 3 and in the semantics I sketched out in chapters 4 and 5, I
made the case for viewing nonlinguistic thoughts as having structured contents—
contents that contain distinguishable content-constituents capable of featuring in
further thoughts. Clearly, it is a necessary condition on there being formal inferen-
tial transitions between contentful thoughts that those thoughts should have struc-
tured contents. Nonetheless, it is not a sufficient condition. Formal rules of in-
ference do not operate on thought-contents but rather on the vehicles of those
contents. They are syntactic rather than semantic. It is easy to lose sight of this
when dealing with inferential transitions between sentences in formal languages,
because there is a structural isomorphism between the logical structure of the sen-
tence that expresses a given thought and the structure of the thought-content that it
expresses. In the case of formal language sentences, it doesn't really matter
whether one considers the structure of the vehicle or the structure of the content,
since they are isomorphic. But in the case of nonlinguistic thought (as hi the case
of sentences in natural languages) the distinction becomes important. The contents
of nonlinguistic thought are indeed linguistically expressible and have a commen-
surate degree of structure. That is what makes them instances of thinking-that,
rather than thinking-how. But it is far from clear that the vehicles of nonlinguistic
thought are linguistically structured in a way that would make it possible to apply
formal rules of inference to them. This is a theme that will be discussed in more
detail in the next two chapters.
It might be objected that "everyday rationality" of the type that psychological
explanations aim to track is not in fact formal. That is to say, everyday rationality
proceeds by seeing immediate connections between ideas, rather than by mechani-
cally applying formal rules of inference. This, broadly speaking, intuitionist line of
thought goes back to Locke and has been promoted more recently by Jonathan
Lowe (1993). The development of formal codifications of reasoning, according to
Lowe, involves maintaining reflective equilibrium between the deliverances of
everyday rationality and the deductive consequences of particular formal systems.
Someone attracted to the intuitionist conception of everyday rationality would be

quite justified in mistrusting an argument from a syntactic understanding of infer-
ence to the need for sentential vehicles for the thoughts between which inferences
take place. But it is not clear that such a theorist can avoid the need for sentential
vehicles altogether. What exactly are the two things between which we intuit con-
nections, if not sentences? It is hard for us to make much sense of what Locke and
his contemporaries found so easy to take for granted, namely, that we can directly
perceive connections between thoughts. The most natural way to gloss the idea of
directly perceived connections between thoughts is in terms of directly perceived
connections between the truth of one sentence and the truth of another.2 And with
this we are back with linguistic vehicles.
In any case, as Davidson himself makes very clear, the explanatory power of the
rationality connection in ordinary folk psychological explanations depends on the
fact that the person whose behavior is being explained is at least in principle capa-
ble of carrying out the relevant reasoning—as he puts it, "had he been aware of
them and had he the time, he could have reasoned that his act was desirable."
Without this we would have nothing like a genuine explanation. But to be capable
of reasoning requires mastery of the relevant inferential principles—that is to say,
mastery of the canons of what I have termed procedural rationality. And it is here
that the inference-based conception breaks down. Nonlinguistic creatures are
not reasoners in anything like the sense required for them to be rational on the
inference-based conception of rationality.3
Although students of animal behavior have sometimes suggested that nonlin-
guistic creatures are capable of mastering certain basic formal principles of infer-
ence, these claims are plausible only when the notion of mastering a formal princi-
ple of inference is taken in such an etiolated sense that it no longer really counts as
inference at all. Some of the claims made about Ronald Schusterman's work with
sea lions will illustrate this. It is frequently claimed that Schusterman has trained
his sea lions to understand and apply the logical principle of the transitivity of
identity. Here is a representative report:
Researchers have so far discovered that the essentials of Aristotelian logic are acces-
sible to at least one other species: the sea lion. The first task presented to the sea lions
was to learn that two icons are equivalent: X = Y. Next they were taught that Y and Z
were equivalent: Y = Z. Then they were asked if X and Z were equivalent: does X =
Z? Sea lions readily mastered this logical train. (Gould and Gould 1994, 176)
When we look more closely at the experimental paradigm, however, it becomes

clear that there is a certain creative license in this description. The initial training,
which Gould and Gould describe in terms of teaching the animal that two icons are
equivalent, really amounted only to rewarding the sea lion when it chose the sec-
ond icon (Y) shortly after being presented with the first icon (X). There seems no
reason why this form of learning should be described in terms of a logical concept
such as the concept of identity. Not only is the learned behavior a purely condi-
tioned response, but it is hard to know how even to interpret the suggestion that the
two icons might be identical, given that the icons used in the experiment had dif-
ferent illustrations on them.
It is true that there is something significant going on here. The sea lions learn to
associate X and Z even though they have never been exposed to the conjunction of
the two icons. This is interesting because it seems clearly to contradict the basic
claim of classical conditioning theory, which is that learned associations must rest
on reinforcement. Classical conditioning theory predicts that the sea lions would
fail the test, on the grounds that they had not been exposed to any reinforcement of
the association between the first and third icons. But, understood in these terms,
the sea lion performance seems much closer to the well-documented phenomenon
of sensory preconditioning in rats (Rizley and Rescorla 1972). In sensory precon-
ditioning rats are exposed to two pairings. The first pairing is of a light and a tone,
while the second is of the same light and an electric shock. If, once the precondi-
tioning has been completed, the rats are exposed to the tone on its own, they will
manifest the same aversive behavior that they were conditioned to show to the
light (in virtue of its association with the electric shock). Just as the sea lions had
never been exposed to the conjunction of icon X and icon Z, the rats had never
been exposed to the conjunction of tone and electric shock. Yet in both cases the
response was generalized without further training. It might be appropriate to de-
scribe this phenomenon using some phrase such as the "transitivity of association,"
but there seems to be no sense in which implicit grasp of any logical principle is
involved.
It is striking that Schusterman and his coworkers do not make any explicit
claims about the concept of identity. In, for example, Kastak, Schusterman, and
Kastak 2001, the experiments are described as showing that sea lions are capable
of equivalence classification, where this is understood as the capacity to classify
groups of physically dissimilar stimuli that are related by a relation R possessing
the formal properties of reflexivity (i.e., every member of the group bears that rela-
tion to itself), symmetry (if a bears relation R to b then b will bear relation R to a),
and transitivity (if a bears relation Rtob and b bears relation R to c then a will bear
relation R to c). Any relation satisfying these formal properties is an equivalence
relation. It is true, of course, that the logical concept of identity is an equivalence
relation—but it hardly follows that sea lions that have shown themselves capable
of equivalence classification will ipso facto have mastered the concept of identity.
There are various equivalence relations in play in the different experimental para-
digms that have been taken to illustrate equivalence classification, but none of
them has anything to do with the concept of identity. The simplest equivalence re-
lation is the relation of having-the-same-reinforcement-history-as (i.e., all stimuli
that have been positively reinforced will be classified together, as will all stimuli
that have been negatively reinforced). Classification according to this very
straightforward equivalence relation is not, in fact, unique to sea lions. It has been
demonstrated in pigeons (Vaughan 1988). An example of a more complex equiva-
lence relation would be the relation of having-the-same-function-as. It turns out
that sea lions (although as yet no other species) are capable of classifying stimuli
ordered by this equivalence relation. A recent survey article (Schusterman, Kastak,

and Kastak 2002) suggests that sea lions are able to classify according to equiva-
lence relations of being a conspecific or being related. None of these equivalence
relations, however, has anything to do with the equivalence relation of identity, as
a relation that holds between every thing and itself.
This is no place for a survey of claims that have been made for logical compe-
tence in nonlinguistic creatures, but it seems to me that all such claims can be fitted
into one of either three categories. In the first category fall claims such as those
made by the Goulds about Schusterman's sea lions experiments. The key charac-
teristic here is that appeal is made to mastery of logical principles in order to ex-
plain forms of behavior that can be far more easily explained more parsimoniously.
There is obviously little aid here for an extension of the inference-based concep-
tion to nonlinguistic creatures.
In the second category certain forms of cognitive ability are wrongly described
as logical. As an example we can take David Premack's well-known experiments
using a same-different paradigm on chimpanzees (Premack and Premack 1983).
Premack trained his chimpanzees by presenting them with an initial symbol fol-
lowed by a pair containing that symbol and a second, different symbol. They were
positively rewarded for selecting the second symbol. Although at first reinforce-
ment was required for each new pairing, the chimpanzees gradually began to re-
spond correctly without any reinforcement. It may well be right to interpret this be-
havior in terms of mastery of the concepts same and different,t,but there is surely noo
logical principle involved here. The experiment says nothing about the capacity of
chimpanzees to master formal principles of reasoning. Again, this will be of little
use to proponents of an inference-based conception of practical reasoning.
In the third category of claims about the logical competence of nonlinguistic crea-
tures we can place what might be termed "as-if attributions of certain patterns of
reasoning. As has become well known from studies of animal foraging behavior, it is
possible to model certain aspects of animal behavior by making the heuristic as-
sumption that the animal is performing complex cost-benefit calculations. Here is
how Krebs and Kacelnik describe the bare bones of the framework they propose for
studying patterns of animal behavior, such as those displayed by a foraging robin:
We shall use the metaphor of the animal as a 'decision-maker.' Without implying any
conscious choice, the robin can be thought of as 'deciding' whether to sing or to feed,
whether to feed on worms or on insects, whether to search for food on the grass or on
the flower bed. We shall see how these decisions can be analyzed in terms of the costs
and benefits of alternative courses of action. Costs and benefits are ultimately mea-
sured in terms of Darwinian fitness (survival and reproduction), and may, in many in-
stances, be measured in terms of some more immediate metric such as energy expen-
diture, food intake or amount of body reserves. Analyzing decisions in terms of their
costs and benefits cannot be done without also taking into consideration physiologi-
cal and psychological features that might act as constraints on an animal's perform-
ance. The fitness consequences of decisions, and the various constraints that limit an
animal's options, can be brought together in a single framework using optimality
modeling. (Krebs and Kacelnik 1991)
The guiding assumption of optimal foraging theory is that animals should optimize
the net amount of energy obtained in a given period of time. So acquired energy is
the benefit in the cost-benefit analysis. In the case of a foraging bird, for example,
faced with the "decision" of whether to keep on foraging in the location it is in or
to move to another location, the costs are the depletions of energy incurred through
flight from one location to another and during foraging activity in a particular loca-
tion. The cost-benefit analysis can be carried out once certain basic variables are
known, such as the rate of gaining energy in one location, the energy cost of flying
from one location to another, and the expected energy gain in the new location. It
turns out that optimality modeling makes robust predictions of foraging behavior
in birds such as starlings (Stumus vulgaris) and blue tits (Parus major). Cowie's
study of great tits foraging in an experimental environment containing sawdust-
filled cups with mealworms hidden inside showed that the amount of time a given
bird spent at a given cup could be accurately predicted as a function of the travel
time between patches and the quantity of mealworms in the cup (Cowie 1977).
Similarly, Kacelnik has shown that adult starlings foraging for their young behave
in ways predicted by the marginal value theorem of optimal foraging theory, on the
assumption that the relevant currency is net energy gain to the family (Kacelnik
1984). Foraging starlings collect a beakload of food from a foraging site before re-
turning to their nests. Obviously, diminishing returns will set in during the forag-
ing excursion, as the bird will be less efficient at gathering new food the fuller its
beak is with the food it has already gathered. The marginal value theorem is a
quantitative way of predicting the adjustments that a creature will make in its for-
aging behavior to compensate for these diminishing returns.
Of course, as Krebs and Kacelnik make plain in the quoted passage, there is no
suggestion that the great tits or starlings really are carrying out complex calcula-
tions about how net energy gain can be maximized within a particular set of pa-
rameters and background constraints. It is a crucial tenet of optimal foraging
theory that the optimizing behavior is achieved by the animal following a set of
relatively simple rules of thumb or heuristics, which are most probably innate
rather than learned. So, for example, a great tit might be hard-wired to move on to
the next tree after a certain number of seconds unsuccessful foraging in one tree.
Hence the phrase "as-if' reasoning. Evolution has worked in such a way (at least
according to the proponents of optimal foraging theory) that foraging species have
evolved sets of heuristic strategies that result in optimal adaptation to their eco-
logical niches. This optimal adaptation can be mathematically modeled in terms of
what is ultimately a sophisticated version of expected utility theory, but the behav-
iors in which it manifests itself do not result from the application of such a
theory—any more than a bird's capacity to fly reflects any mastery on its part of
the basic principles of aerodynamics. Once again, there is no scope here for under-
standing how the inference-based conception of practical rationality can be ex-
tended to nonlinguistic creatures.
It would seem, therefore, that Davidson's strategy of treating practical ration-
ality as a form of theoretical rationality is doomed to failure. It is true that I have so
far argued explicitly only for the thesis that we cannot extend a workable notion of
procedural rationality to nonlinguistic creatures. Despite the views of some
philosophers (Stein 1996) and many psychologists of reasoning (particularly those
studying the Wason selection task and other experimental paradigms testing mas-
tery of the rules that govern conditional reasoning), mastery of the basic principles
of inference needs to be clearly distinguished from what might be termed norms of
good reasoning. These are principles that govern the processes of thinking: weigh-
ing up the evidence for and against a particular proposition; judging the likelihood
of a particular event; changing one's beliefs and probability assignments in re-
sponse to changes in the available evidence; and so forth. The norms of reasoning
go to make up what I have elsewhere termed the domain of epistemic rationality
(Bermudez 2001). It is also true, as I show in the final chapter (section 9.1), that
there is some scope for the relevant types of belief formation and revision at the
nonlinguistic level. But it is hard to see how the epistemic dimension of theoretical
rationality can be of any use in making sense of practical rationality.
Clearly, therefore, if we are to identify a sense of nonlinguistic rationality in the
practical sphere, it is no use approaching the issue via procedural rationality in the
way Davidson does. We need a specific account of how nonlinguistic creatures
might be practically rational. In line with the general methodology adopted in this
book, I will broach the issue from an epistemological and operational dimension,
starting with the following questions. In what circumstances would it be appropri-
ate to describe the behavior of a nonlinguistic creature as rational? What opera-
tional criteria might there be for nonlinguistic rationality? In the next three sec-
tions I will identify three different types of behavior, each of which can be
described as rational in a different sense. For reasons that will emerge hereafter,
only the second and third of these can count as rational in the right sense to ground
the practice of psychological explanation.
6.2 Level 0 Rationality

Let me start with a basic datum. There is no need for psychological explanations of
animal behavior when we are dealing with tropistic behaviors such as those pro-
duced by reflexes, innate releasing mechanisms (such as imprinting mechanisms),
or classical conditioning. In such situations we can explain the behavior in terms of
a lawlike connection between stimulus and response. Whenever the relevant stim-
ulus is encountered the same response will emerge. We do not need to postulate
intermediary representational states between sensory transducers and behavioral
output.
Nonetheless, there is a sense in which tropistic behavior of this form can be de-
scribed as rational. This is the sense of the word 'rational' on which 'rational'
means something like 'adaptive' or 'conducive to survival' (Dawkins 1986). There
are criteria by which it is appropriate to judge the rationality of even the simplest
tropistic forms of behavior. Take a behavior that has evolved as an adaptive re-
sponse to potentially harmful events on the body surface, such as the eye-blink re-
sponse to puffs of air. This is clearly adaptive, and the adaptiveness carries over
even to behaviors in which the eye-blink response is itself part of a larger condi-
tioned process. If there is reliable advance warning of the arrival of the uncondi-
tioned stimulus (e.g., the sound of a tone preceding the puff of air) then there are
adaptive advantages in activating the response before the unconditioned stimulus
actually appears.
I will call this level 0 rationality. It has two characteristic features, as follow.
1. It is not grounded in any process of decision-making.
2. It is applicable not to particular behaviors but to the presence (either in the
organism or in the species) of a particular tendency or disposition.
The first feature should be self-explanatory, but some comments are required on
the second feature. We can only apply the notion of rationality when there is a
space of alternatives. A rational behavior has to be one that is performed rather
than some other behavior that could have been performed. But of course there is no
such space of alternatives at the level of individual tropistic behaviors. The space
of alternatives exists only at the level of the genetically determined disposition to
behave in a certain way. Putting the point in more familiar philosophical terms,
level 0 rationality applies only to behavior types and not to behavior tokens.
This type of rationality is not confined to tropistic behaviors. We find it exempli-
fied in foraging behavior. As we saw in the previous section, many species have
evolved to follow simple behavioral rules when foraging for food. According to
optimal foraging theory, in many cases these rules are such that individuals of the
relevant species are maximally adapted to their environment. So, how might one
apply the notion of rationality to the feeding patterns of a redshank (Tringa
tetanus)! Redshanks are shorebirds that dig for worms in estuaries at low tide. It
has been noticed that they sometimes feed exclusively on large worms and at other
times feed on both large and small worms. The hypothesis put forward by
Stephens and Krebs to explain this behavior involves what they call the principle
of lost opportunity (Stephens and Krebs 1986). In essence, although a large worm
is worth more to the redshank in terms of quantity of energy gained per unit of for-
aging time than a small worm, the costs of searching exclusively for large worms
can have deleterious consequences, except when the large worms are relatively
plentiful. If the large worms are rare then it will obviously take much longer to find
one—time during which the redshank is not only expending valuable energy but
also losing opportunities to gain energy from smaller worms. So redshanks only
ever forage exclusively for large worms when there are plenty of large worms
around.
It seems perfectly reasonable to say that this is rational behavior on the part of the
redshank. But of course this is not to say that whenever the redshank switches from
a restricted search strategy (only large worms) to an unrestricted search
strategy (both large and small worms) it is behaving rationally. That would not be
right, because at the level of the individual foraging behavior there is no space of al-
ternatives. The redshank is "following" a relatively simple algorithm, and there
seems no sense in which it could fail to follow it—unless, of course, the foraging al-
gorithm was trumped by another algorithm and the bird ceased foraging, as it might
do if a predator was detected. The space of alternatives exists at the level of the hard-
wired algorithm. What is rational is not the redshank's behavior on a particular oc-
casion but rather the fact that it has evolved in such a way as to follow an algorithm
that allows it to switch from a restricted search strategy to an unrestricted search
strategy—as opposed, for example, to always following the same strategy.
As the example of optimal foraging theory makes clear, for a behavior pattern to
count as rational in the level 0 sense, there must be not only a contrast space of
alternative behavior patterns but also a normative standard against which those
behavior patterns can be assessed. The normative "currency" by which to judge for-
aging behavior is some form of maximization of energy gain. But there are questions
to be raised about how widely the energy calculation should extend. In certain cases
(such as the redshank example just considered) it makes sense to restrict the calcula-
tion to the foraging bird. In other cases, such as that of the starlings foraging for their
young, the most accurate predictions come when one works on the basis of the rate
of energy procurement for the family as a whole. Nor, of course, is maximization of
energy gain the only available currency. Another obvious normative standard for
evaluating particular patterns is that they should facilitate predator avoidance—or
that they should not hinder the animal from rinding a mate.
A distinction needs to be made between short-term criteria of rationality of the
type I have just been considering and long-term criteria (see Dawkins 1986 for a
similar distinction between short-term and long-term rationality). Rate of energy
procurement is a short-term criterion of rationality. The accepted currency for
long-term calculations is fitness. There are different ways of calculating fitness.
Strictly speaking, fitness is a matter of relative quantities of specific genes in the
gene pool over time, but in practical terms this can often be measured in terms of
the reproductive output of individual animals. This is individual lifetime fitness,
calculated as the product of the length of time the animal survives and the average
number of offspring it produces during each year of its life. In the not uncommon
situation in which an animal sacrifices its own individual lifetime fitness in a way
that increases the lifetime fitness of other animals that share many of the same
genes by descent, the appropriate currency is inclusive fitness. Inclusive fitness is
arrived at by adding the individual lifetime fitness of the donor animal to the sum
of the individual lifetime fitness of the animals helped by the donor, each dis-
counted by the probability of their sharing genes with the donor animal.
There are, therefore, many different criteria according to which level 0 ration-
ality can be assessed. The basic distinction is between short-term and long-term
criteria, but within each grouping there are further distinctions. This is important
for the following reason. One objection likely to be raised against the very idea of
level 0 rationality is that it is completely Panglossian. Every form of behavior that
is best analyzed at this level will come out as rational, simply as a function of the
way the notion is defined. Any suitable behavior pattern that exists does so because
it has proved more adaptive than the other potential behavior patterns available to
natural selection at a given moment in evolutionary time. But then, it looks very
much as if any extant behavior pattern will prove to be rational in the level 0 sense
simply in virtue of having been selected. This would deprive the notion of level 0
rationality of sense and point, since it would make it impossible for there to be a
thing as level 0 irrationality.
However, evolutionary fitness is not the only criterion by which the level 0 ra-
tionality of a particular behavior pattern can be assessed. Indeed, in many cases it
seems that fitness (whether individual or group) is not really an appropriate cur-
rency. It is true that the overarching currency of natural selection is fitness—but for
that reason fitness is too coarse-grained a tool for thinking about the ways that par-
ticular behavior patterns help a creature fit into its ecological niche. Short-term cri-
teria are far more helpful than long-term criteria in that respect. It may well be the
case that a particular behavior pattern is rational in the long-term sense of having
been the most fitness-promoting of the alternatives available for natural selection
to choose between without being rational in the short-term sense. Let us suppose
that there is an optimality threshold for each of the different aspects of an animal's
existence—foraging, mate selection, nest-building, rearing of young, food avoid-
ance, and so forth. The fitness of animals of a particular species may well depend
so crucially on striking the right balance between the various different activities in
which it engages that performance above the optimality threshold on one dimen-
sion could not be achieved without compromising the balance. So, for example,
performance above the rationality threshold on foraging might only be possible at
the price of performance on predator avoidance too far below the threshold to be
sustainable. This might well be a case in which, assuming that we were interested
solely in short-term criteria of level 0 rationality, none of a creature's specific be-
havior patterns would qualify as level 0 rational. It is more likely, of course, that
long-term fitness would be served by a combination of behavior patterns, some of
which fall below the optimality threshold while others are safely above it.
Level 0 rationality is not always a matter of optimization or maximization.
There are good examples of level 0 rationality in the signaling strategies that are
widespread in the animal kingdom (Bradbury and Vehrencamp 1998) and in par-
ticular in what might be termed two-way information transfers. These are informa-
tional transactions in which information is transferred in two directions (as op-
posed to one-way information transfers such as mating displays). We find such
transactions most typically in situations where animals are in conflict over food,
prospective mates, or breeding sites. Rather than resort to the direct use of force to
resolve the conflict, many animals employ threat-display signals to come to a non-
violent consensus as to which is the stronger. One feature of two-way information
transfers is that each sender can modify its signal in the light of the signal it re-
ceives from the other participant in the exchange. Conflict resolution is not, of
course, the only sphere in which we find two-way information transfers. Mating
displays and territorial signaling provide examples in which, unlike many in-
stances of conflict resolution, the signaling behaviors of the two participants are
not symmetrical. One feature that all such information exchanges have in common
is that they lend themselves to being modeled game-theoretically (Lewis 1969,

Skryms 1996). Evolutionary game theory can help make sense of why particular
signaling exchanges between sender and recipient should have become stabilized.
We can see both participants' signaling behavior as strategies adopted by particular
roles within a population (where a role might be male vs. female, young males vs.
older males, and so forth) that have become stabilized because each is the optimal
response to the other within the range of alternatives made available by a particular
evolutionary context. They are both evolutionary stable strategies (Maynard Smith
1982).
To summarize, level 0 rationality is a type of rationality appropriate to behavior
types rather than behavior tokens. It does not involve any genuine decision-
making. The behavior types that qualify for level 0 rationality will most often be
hard-wired, ranging from simple reflexes such as the eye-blink response to innate
releasing mechanisms. They are properly described as rational because there is a
normative theory, such as for example some version or other of optimal foraging
theory, under which they come out as the best response among the range of alter-
native patterns of behavior available at that time. There are, broadly speaking, two
different ways of determining whether or not a particular strategy is a best re-
sponse. It might qualify in virtue of maximizing some particular currency. Or it
might qualify as being part of what game theorists call an equilibrium strategy—
that is, a strategy such that neither party can benefit by deviating from it. In either
case there remains the question of what the currency is in terms of which gains and
benefits are being calculated, and as I have shown there are a range of potential
currencies, short-term and long-term.

Discussions of nonlinguistic rationality, particularly on the part of philosophers,
frequently assume that there is no middle ground between what I have termed level
0 rationality and the sophisticated types of rationality that we find in language-
users and that can be modeled by formal logic, rational choice theory, game theory,
and so forth. This does not seem right. Level 0 rationality is limited in two ways.
First, it does not involve a recognizable process of decision-making. Second, it is
not applicable to behavior tokens but only to behavior types. We can easily see that
there is room in logical space for at least one intermediate conception of ration-
ality. Such an intermediate form of rationality would be one that was subject to one
of these limitations but not the other. The abstract possibility is illustrated in the
following table.
Applicable to Behavior Tokens Decision-Making
Level O X X
Level 1 V X
Is this possibility realized? It seems to me that it is. A given tropistic behavior

token cannot be rational because there is no sense in which it is selected from
a range of alternatives. Consequently it cannot involve a process of decision-
making. But it does not follow from this that any behavior that is properly de-
scribed as having been selected from a range of alternatives must involve a process
of decision-making.
Let me take a simplified example fairly common in the animal kingdom. Imag-
ine an animal confronted with another potentially threatening animal. The animal
has two possible courses of action—fight or flee. There is a clear sense in which
one of the two courses of action could be more rational than the other. Roughly
speaking, it will be in the animal's best interests either to fight or to flee. And it
seems that in such a situation there need be no process of decision-making. The
animal might just "see" that fighting is the appropriate response. Or it might just
"see" that fleeing is appropriate. The theory of affordances developed by J. J. Gib-
son (1979) gives us a way of making sense of this as a form of direct perception.
Gibson's theory is that perception is not neutral. It is not just a matter of seeing
various objects that stand in spatial relations to each other. It involves seeing our
own possibilities for action—seeing the possibilities that are "afforded" by the en-
vironment. If this is right then we can see how a given behavior might be selected
from a range of alternatives in a way that does not involve a process of decision-
making. The comparison of affordances does not require a process of decision-
making. Nonetheless it is assessable according to criteria of rationality. I shall call
this the sphere of level 1 rationality. I shall offer a detailed example of how the no-
tion of level 1 rationality might be deployed in section 7.1. We will see there how
different courses of action can be compared without a process of decision-making.
The principal difference between level 0 rationality and level 1 rationality is in
the location of the contrast space of alternative possible courses of action. Assess-
ments of rationality are only applicable when the relevant course of action is prop-
erly described as having been selected from a range of alternatives. In level 0 ra-
tionality the appropriate contrast space is between different behavior patterns or
tropistic mechanisms that might have been selected by evolution, whereas in level
1 rationality the contrast space is a range of different possible courses of action
available to the organism at the relevant time.
One would expect there to be a range of possible currencies in terms of which a
particular behavior token might be assessed for level 1 rationality, much like the
range of currencies operative for level 0 rationality. Level 1 rationality is subject to
the same distinction between short-term and long-term criteria, and within each
group there will be different criteria pushing in different directions. In the case of
level 0 rationality we are considering patterns of behavior that are most probably
instinctive, but the sources of behaviors assessable for level 1 rationality are far
more fluid. We can imagine an animal being motivated to act in a way that would
reduce its individual lifetime fitness (calculated as the product of survival and fe-
cundity) and yet that counts as level 1 rational. The most obvious case would be
one where an animal (a member of a species such as the vervet monkey, which has
a highly developed system for alerting conspecifics to the presence of different

types of predator) fails to signal the approach of a predator and instead flees the
scene. It might well be that fleeing as opposed to warning decreases both indi-
vidual lifetime fitness (by making other members of the group less likely to coop-
erate in future) and inclusive fitness (by decreasing that individual's chances of
finding a mate) while yet counting as level 1 rational according to other criteria,
such as for example the criterion of the individual's long-term survival.4
As with level 0 rationality, there will be two different ways that the notion of
level 1 rationality might apply. A particular behavior token might qualify as level 1
rational in virtue of maximizing units of some relevant currency. Alternatively, it
might qualify in virtue of being an equilibrium strategy in game-theoretical
terms—although, unlike the level 0 case, the relevant strategy will be behaving in
that particular way in that particular context, rather than a generalized behavior
pattern. Although in practice it will not always be clear whether a given behavior is
to be judged in terms of level 0 or level 1 rationality, one might expect the sort of
situations in which issues of level 1 rationality arise to be correlated with greater
amenability to game-theoretical analysis, for the following reason. The type of be-
haviors that it is appropriate to assess for level 0 rationality are highly invariant be-
haviors, instances of rigid behavior patterns that one might expect to be repeated
whenever the environment is suitably configured. It is when this invariance breaks
down that we move to assessing the rationality of behavior tokens rather than be-
havior types. But one would expect the invariance to break down far more fre-
quently in cases involving interanimal interaction (such as conflict resolution) than
in those where the interaction is simply between animal and environment, and it is
the former that are susceptible to game-theoretic analyses.
Thinking back to chapter 3, it will be apparent that the notion of level 1 ration-
ality is very much in the spirit of the minimalist conception of nonlinguistic
thought. According to the minimalist view, the thinking of nonlinguistic creatures
cannot be significantly divorced from the here-and-now of the perceived environ-
ment, and whenever the thinking behavior of a nonlinguistic creature is directed at
something that is not immediately perceptible, the instrumental representations
driving its behavior cannot be anything more complicated than mediate percep-
tions of the sort provided by Gibsonian affordances. If, therefore, a genuine notion
of rationality is applicable to thinking behavior as the minimalist understands it, it
can only be either level 0 rationality or level 1 rationality, depending on whether
the notion of rationality is applicable at the level of behavior types or behavior to-
kens. As I showed in chapter 3, however, the minimalist conception of nonlinguis-
tic thought is too restrictive. This in itself is enough to suggest that there must be a
further intermediate level of rationality. That suggestion is strongly reinforced by
the fact that one of the decisive arguments against the minimalist conception was
essentially that there are important types of behavior in nonlinguistic creatures that
cannot be understood in terms of mediate perceptions such as Gibsonian affor-
dances. In the next section I will show how a further notion of rationality can be
developed to deal with these more complex behaviors.

What is the next level of rationality? Clearly it will be one in which both criteria
are satisfied. Thus:
Applicable to Behavior Tokens Decision-Making
Level O X X
Level 1 V V
Level 2 V X
But it is not immediately obvious how such a notion of rationality can be applied
at the nonlinguistic level. It is far from straightforward to see how nonlinguistic
creatures can properly be described as decision-makers. There is no room for an
inference-based conception of decision-making—whether inference is taken
strictly as a matter of formally characterizable operations defined over syntactic
vehicles, or more broadly as a matter of the immediate perception of entailment re-
lations between thoughts.5 But then how else could decision-making be under-
stood? Unless we can develop an alternative conception of decision-making, it
looks very much as if we will have to restrict the notion of nonlinguistic rationality
to level 0 and level 1 rationality.
Let us look once again at the requirements. I have already shown how in behav-
iors assessable in terms of level 1 rationality it is appropriate to describe an animal
as behaving in a particular way within a contrast space of alternative possible
courses of action. It is natural and convenient to formulate this in terms of Gib-
son's theory of affordances. So one might say that an animal perceives a range of
different courses of action afforded by the environment and acts on one of these
perceived affordances. Why is this not properly described as decision-making? To
appreciate why not, it is important to make a distinction between two different
ways that different courses of action can be compared. On the one hand, they can
be compared simply qua courses of action. That is to say, an animal might compare
the action of fighting with the action of fleeing. It is this sort of comparison that is
at play in the forms of behavior that are assessable for level 1 rationality. The pos-
sibility of such comparison requires simply representations of actions. These rep-
resentations are not very complex. They might (if some form of Gibson's theory is
correct) be understood at a purely perceptual level. It is perfectly possible, and in-
deed highly likely, that the choice between such action-representations can be
made on relatively simple and more-or-less noncognitive grounds. One way of in-
terpreting the whole basis of instrumental conditioning is as a process of attaching
a certain positive valence (through reinforcement) to the representation of a par-
ticular action, so that the animal being conditioned is motivated to act on that ac-
tion-representation rather than one of the others afforded by the immediate envi-
ronment. It is for this reason, I think, that the notion of decision-making is best not
applied to cases in which courses of action are compared qua courses of action.
Such cases do indeed involve a form of choice—or perhaps 'selection' would be a
better word. But the choice or selection is not made on the right sort of grounds to
qualify as decision-making.6
What then is involved in genuine decision-making? The minimal requirement is
that the selection of a particular course of action from the contrast space of alterna-
tive possible courses of action should be made on consequence-sensitive grounds.7
That is to say, genuine decision-making involves a selection between different
possible courses of action that is grounded on an assessment of the likely conse-
quences that those different possible courses of action will have. Deciding is not
simply selecting. It is selecting for a reason. There is an important distinction here.
In many cases of instrumental conditioning (but not all, as I shall shortly show) it
is correct to say that the relevant action is performed because of its consequences.
So, for example, what explains the pigeon's pressing the lever is that it will result
in delivery of food from the cartridge. It is the association of that behavior with
those consequences that has created the positive valence that leads the pigeon to
carry out the action of lever pressing. But to say that an action is being performed
on consequence-sensitive grounds implies far more than its simply being per-
formed because of its consequences. It implies that the agent has made an assess-
ment of those consequences, on the basis of a belief about the outcome that that ac-
tion is likely to have (and, most likely, a comparison of that outcome with the
likely outcomes of other possible courses of action). Decision-making only takes
place, in other words, at a level where instrumental beliefs are available.
The distinction between simple selection (as in level 1 rationality) and genuine
decision-making of the sort that only becomes available with level 2 rationality
can be put in terms of the representation of contingencies. Some of these contin-
gencies hold between an action and an anticipated outcome-situation. But there
does not always need to be representation of the action itself. It is natural to think
that one important factor motivating an animal to act on a particular represented
contingency is its recognition that the outcome-situation would satisfy one of
its desires—in the terminology introduced in chapter 3, the anticipated outcome-
situation would be a goal-situation. So in level 2 rationality we encounter once
again a familiar pattern of psychological explanation—that is to say, psychological
explanation in terms of a belief-desire pair linked by an instrumental belief about
how a desire might be satisfied in a particular context.
A consequence of this is that level 2 rationality can be assessed in a fundamen-
tally different way from the other two levels of rationality that I have considered.
Whereas both level 0 and level 1 rationality can only be understood in terms either
of maximization of a given currency (such as rate of energy procurement, indi-
vidual lifetime fitness, and so forth) or of strategies that are in game-theoretic equi-
librium, level 2 rationality admits a fundamentally different type of assessment.
Since the crucial element in level 2 rationality is the way action is grounded in in-
strumental beliefs about the outcomes of those actions, it is clear that in an impor-
tant respect the level 2 rationality of an action will depend on the "match" between
action and background beliefs. In this sense the level 2 rationality of a particular
action will be a function of:
1. The accuracy of the instrumental belief

2. The extent to which the action in question is a suitable implementation of
the instrumental belief
There is fortunately no need at present to explain how both criteria can be devel-
oped into a full-blown theory of level 2 rationality.8 There is enough on the table
already for the central problem to be clear.
The principal question is how we should understand instrumental beliefs and the
representation of contingencies at the nonlinguistic level. It is open to a skeptic to
accept the distinction between level 1 and level 2 rationality but to deny that be-
havior amenable to assessment in terms of level 2 rationality is available at the
nonlinguistic level. How, it might be asked, can contingencies be represented and
compared except through inferential processes of the sort that we have seen not to
be available to nonlinguistic creatures? How can we make any sense of a creature
acting on the basis of a comparison of contingencies unless we take it to be choos-
ing the course of action with the highest likely benefits when each course of ac-
tion's benefits are discounted by their probability? And surely that requires it to be
representing the different potential outcomes, the utility it attaches to each of them,
and its estimate of the likelihood that each outcome will occur—as well, of course,
as performing the necessary calculations?
In line with the methodology adopted elsewhere in this book, I propose to make
a start on this question by asking how we might understand the distinction between
level 1 rationality and level 2 rationality in operational terms. What evidence
might there be that a creature is representing the consequences of two or more dif-
ferent courses of action, rather than the actions themselves? Once we have a clear
set of operational criteria in view it will be easier (in the next section) to offer a
constitutive account of level 2 rationality. The particular forms of reasoning in-
volved in level 2 rationality will be discussed further in chapter 7.
It is easiest to see how we might detect when a creature is not representing the
contingency between action and consequence—that is to say, how we might detect
when criteria of level 1 rationality should be applied rather than criteria of level 2
rationality. For example, if behavior were driven by the representation of an instru-
mental contingency then one would expect the behavior to cease in the face of re-
peated evidence that the contingency no longer holds. If the animal persists in the
response then we have prima facie evidence that the contingency is not being rep-
resented. A classic experiment has been done by Hershberger (1986), who set up a
graphic illustration of such a reversed contingency for chicks. In the experimental
setup their food source retreated from them at twice the rate they walked toward it
but advanced toward them at twice the rate they walked away from it. Even after
100 trials, the chicks only succeeded in obtaining the food 30 percent of the time,
which has been taken clearly to indicate that they were failing to represent the two
relevant contingencies (that walking backward causes the food to advance and
walking forward causes the food to retreat).9
It is a minimal operational criterion, therefore, for actions being grounded on
representations of contingencies that the action should not be persisted in once the
animal has been confronted with evidence that the contingency ceases to hold. It is
relatively easy to see what would count as evidence that this operational criterion
is not being met. But what evidence might there be for thinking that it is being
met? Some suggestive discoveries have been made by Rescorla and Skucy, and
further developed by Hammond. They found that rats that have been trained to
press a lever for food will cease to press the lever when the schedule is changed so
that the food is delivered whether they press the lever or not (Rescorla and Skucy
1969; Hammond 1980). This seems to involve recognition that the contingency be-
tween lever-pressing and food delivery no longer holds. The rats had initially been
pressing the lever in virtue of an instrumental belief that lever-pressing would re-
sult in the appearance of food. When the correlation tracked by the instrumental
belief ceased to hold, the associated behavior also ceased.
Perhaps the most obvious source of relatively clear-cut evidence of level 2 ra-
tionality in nonlinguistic creatures is tool manufacture and tool-using behavior.
The means-end domain of tool construction and use is deeply tied up with the rep-
resentation of contingencies. Of course, not all such behavior is evidence for level
2 rationality. Many types of tool construction are relatively hard-wired and, al-
though capable of refinement and improvement through learning, seem best evalu-
ated according to the criteria associated with level 0 or level 1 rationality. The con-
struction of dams by beavers is a case in point, much closer to nest building than to
deliberate tool-based manipulation of the environment. A good example, though,
of what seems to be the genuine representation of contingencies comes with the
way chimpanzees in the wild manufacture tools for particular purposes (Byrne
1995, 96-97). Wild chimpanzees make wands for dipping into ant swarms by
stripping the side leaves and leafy stem from a stick several feet long. The wands
constructed for dipping into termite nests, on the other hand, are made from vines
or more flexible twigs and are considerably shorter. They also have a bitten end,
unlike the ant wands. It is sometimes remarked that such tool construction is
purely innate. Gould and Gould (1998, 55) suggest, for example, that there is no
genuine thought involved in what they call termite-fishing because even chim-
panzees born in captivity make a habit of putting long thin things into holes. This
neglects, however, the specialized nature of the different tools constructed. It is not
just a matter of dipping a long thin stick into a narrow hole—the long thin stick
needs to be constructed differently depending on what sort of hole it is going into.
Nor, moreover, does the wand construction seem to be a form of trial-and-error
learning. Byrne notes (97) that the wands are often constructed some time in ad-
vance and a considerable distance away from the place where they are going to be
used.
Perhaps our best source of information for tool manufacture comes from archeo-
logical studies of the tools constructed by prelinguistic hominids. In human evolu-
tion the construction of complex tools long predated the emergence of language
(Gibson and Ingold 1993). The fossil record suggests that hand-axes, the character-
istic tool of early Homo habilis, first appeared about 1.4 million years ago—long
before the evolution of language that (even on the most optimistic scenario) could
not have occurred before the speciation of archaic Homo sapiens about one million
years later. Considerable technical skill is required to make a hand-axe. Since the
hand-axe is symmetrical, the flakes need to be removed from alternate sides. Each
nodule is different, with different stresses and fracture lines, and the toolmaker
needs to keep in mind a specific goal and adjust his blows accordingly. The force
of the blows needs to be precisely calculated. The entire process is highly compli-
cated and dependent on constant feedback and revision. A highly developed form
of instrumental rationality is at work here, feeding into action.
The next major event in the early evolution of archaic Homo sapiens was the
emergence of the Levallois flake, a characteristic tool that only emerged in the
Middle Paleolithic period (Mithen 1996a). (Figure 6.1 shows the process of mak-
ing a Levallois flake.) These Levallois flakes were then incorporated into more
complex tools. Spears were made, for example, by hafting flakes onto wooden
shafts—a process that involves the extraction of resin, the production of lashing
materials, and so forth. It is hard to see how such complex forms of tool construc-
Figure 6.1 The making of a Levallois flake. To make a Levallois flake one must remove
flakes from the surface of a core to leave a series of ridges on a domed surface (1-3, which
will then guide the removal of the final pointed flake. A striking platform is prepared per-
pendicular to the domed surface of the core (4) and the Levallois flake removed by a single
blow. From a drawing by Margaret Matthews in The Prehistory of the Minddby Steven
Mithen, published by Thames and Hudson, Ltd.
tion, including the combination of tools to make further tools, could be possible
without explicit representation of contingencies. Here, it seems, we are well within
the realm of level 2 rationality.
6.5 Nonlinguistic Rationality and Psychological Explanation

In section 6.11 argued that it would be futile to try to extend to nonlinguistic crea-
tures what I termed the inference-based conception of practical reasoning. Accord-
ing to Davidson and many others, the rationalizing connection between the beliefs
and desires cited in a psychological explanation and the action they explain is de-
rived from the possibility of constructing an argument with those beliefs and de-
sires as premises and a description of the action as conclusion. Of course, it is not
just the abstract possibility that such an argument could be constructed that is im-
portant. It is essential that the agent should herself be capable of constructing the
argument, although she need not actually have done so on the occasion in question.
But, as I argued, we have no understanding of how nonlinguistic creatures can
construct arguments, or assess validity in any other way. So the inference-based
conception of practical reasoning has to be abandoned.
What I have shown in the last few sections is that this does not mean that the no-
tion of rationality cannot be applied at the nonlinguistic level. Quite the contrary.
There are three distinct and useful senses in which the behavior of nonlinguistic
creatures can be properly assessed for rationality. The behavior of nonlinguistic
creatures can be assessed according to three different sets of norms—those of level
0 rationality, level 1 rationality, or level 2 rationality. Each set of norms is appro-
priate for different types of behavior. Behaviors that are instances of innate releas-
ing mechanisms and other fixed and invariant behavior patterns are best assessed
according to the criteria of level 0 rationality. The assessment of rationality at-
taches in this case to the type under which the behavior falls, rather than to the
token behavior instantiated on a particular occasion. When it makes sense to sup-
pose that the action performed was one of a range of possible courses of action
open to the creature in question, it becomes assessable according to the norms of
level 1 rationality. When the criteria for genuine decision-making are met, then the
norms of level 2 rationality come into play.
Two questions arise at this point. First, one might ask which of the three levels
of rationality I have identified is appropriate for the project of providing psycho-
logical explanations of the behavior of nonlinguistic creatures. Are psychological
explanations available all the way down the ladder of rationality, or is there a pri-
vileged level or levels of rationality below which psychological explanation is not
possible? Second, one might ask what the relation is between the way rationality is
assessed at level 2 and the way it is assessed according to the normative theories of
rationality that might plausibly be taken as normatively binding on language-using
agents. How, if at all, do the normative criteria of rationality that we apply in our
everyday psychological explanations and predictions differ from those that we
might apply to nonlinguistic creatures? I will take these questions in order.
It is obvious, more or less as a matter of definition, that there is no scope for psy-
chological explanation of behaviors for which considerations of level 0 rationality
are appropriate. It is equally obvious that psychological explanations can be appro-
priate when we are dealing with behaviors assessable for level 2 rationality. The
principal question is whether we can have psychological explanations in situations
where the appropriate criteria of rationality are those of level 1 rationality. There is
a straightforward line of argument setting out to establish that level 2 rationality is
required for psychological explanation. The argument runs as follows. Giving a
psychological explanation is saying that an animal has acted in a certain way be-
cause of its beliefs and its desires. More precisely, it is to say that the combination
of its beliefs and desires explains its actions. An animal has certain beliefs about its
environment and also certain desires. But how do these come together to bring
about action? Only through the representation of contingencies between actions
and their outcomes. Only when an animal forms the belief that a certain course of
action will lead to the satisfaction of a desire. But this is an instrumental belief
about the consequences that an action is likely to have—hence the behavior to
which it gives rise falls squarely within the domain of level 2 rationality.
The problem with this line of argument is that the fact that a certain course of
action will bring about the satisfaction of a desire may be immediately perceptu-
ally manifest—as when, for example, a food reward is in plain view. An explicit
belief is not always required. Any psychological explanation will always have an
instrumental component, but that component need not take the form of an instru-
mental belief. In fact, to revert to some of the points that emerged in our earlier
discussion of the minimalist approach, instrumental beliefs really only enter the
picture when two conditions are met. The first is that the goal of the action should
not be immediately perceptible and the second is that there should be no immedi-
ately perceptible instrumental properties (that is to say, the creature should not be
capable of seeing that a certain course of action will lead to a desired result). The
fact, however, that one or both of these conditions is not met does not entail that
we are dealing with an action that is explicable in nonpsychological terms.
The basic requirement for psychological explanation is negative (as emerged in
chapter 1). An action requires psychological explanation just if its occurrence could
not have been predicted solely from knowledge of the environmental parameters
and sensory input. That is to say, the need for psychological explanation arises only
in situations where the connections between sensory input and behavioral output
cannot be plotted in a lawlike manner. Clearly, however, it is perfectly possible for a
situation to qualify even if the goal of the action is immediately perceptible—or, for
that matter, if the distal environment contains immediately perceptible instrumental
properties. In such situations the instrumental component of the psychological ex-
planation will most likely be part of the content of perception.
It should not be thought, however, that instrumental beliefs (and with them level
2 rationality) come into play only when there is no immediately perceptible goal. It
is possible for a goal to be directly in view and yet for it to be far from apparent
how it is to be gained. In such a situation one would expect the instrumental com-
ponent to take the form of an instrumental belief. Some of the classic examples of
instrumental reasoning in animals fall into this category. Kohler's (1925) chim-
panzees could clearly see the bunch of bananas that was hung out of their reach.
They just could not immediately see how to reach the bunch, until they formed an
appropriate instrumental belief (which, depending on the chimpanzee, was either
that stacking boxes one on top of the other would bring the bananas within reach,
or that two sticks could be joined together to knock the bananas down, or that
standing on a box would bring the bananas within reach of the stick). Another, less
anecdotal example comes from Bernd Heinrich's (2000) experiments with hand-
reared ravens. Pieces of meat were hung by string from their perches, too far for
them to reach from the perch and too securely tied to be accessible in flight. Pour
out of the five ravens eventually worked out different ways of pulling up the string
and obtaining the meat. As with the chimpanzee example, the goal was clearly in
view. It is tempting to think that the difference was made by an instrumental belief
about how that goal might be obtained.
What is important, therefore, for the applicability of psychological explanation
is that there be an instrumental component in the psychological states that give rise
to the particular action. This instrumental component can be part of the content of
perception (in what I have termed level 1 rationality) or it can take the form of a
separate instrumental belief (in level 2 rationality). In order to appreciate the im-
portance of this instrumental component it is helpful to return to the distinction be-
tween internal and external rationality briefly introduced earlier (see section 6.1).
An action is internally rational when it makes sense relative to an agent's beliefs
and desires, while an action is externally rational when it makes sense relative to a
given set of environmental parameters that include the agent's desires but not his
beliefs. From the viewpoint of psychological explanation the first of these is para-
mount. Agents often act on the basis of poorly supported inductive generalizations
and inaccurate assessments of the situation. What is important in explaining their
behavior is how those generalizations and assessments get translated into action.
The question of how they might have acted optimally in that situation is not rele-
vant. But assessments of internal rationality only make sense when there is an in-
strumental component—only with the instrumental component in play can one
properly evaluate the appropriateness of an action relative to the agent's beliefs.
Psychological explanation, therefore, is only applicable when we are dealing with
behaviors reflecting either level 2 rationality or an instance of level 1 rationality in-
corporating an instrumental component.
Moving now to the second question, how are we to understand the relation be-
tween level 1 and level 2 rationality, on the one hand, and the type of rationality
that we think of as governing everyday commonsense psychological explanation,
on the other? There are two subquestions to be separated out here. The first con-
cerns the norms that govern ascriptions of rationality at the different levels; the
second concerns the reasoning that generates the relevant behaviors. How should
we compare the reasoning implicated hi behaviors assessable for level 1, level 2,
and commonsense psychological rationality? The answer to the first question is
relatively straightforward. There is little difference in the norms that govern and
guide ascriptions of rationality at all three levels. Level 1 rationality is governed by
norms of maximizing expected amounts of a particular currency, as well as by the
norm of maintaining interanimal strategies that are in game-theoretic equilibrium.
There is nothing here alien to the norms of folk psychological rationality—
although the range of available currencies is much greater in folk psychological ra-
tionality, and there is scope for issues such as incommensurability that are hard to
imagine arising at the nonlinguistic level. The difference is one of degree rather
than kind. Something similar holds for the norms governing level 2 rationality. As-
sessments of level 2 rationality are determined by two factors, first the accuracy of
the appropriate instrumental belief and second the appropriateness of the action
relative to that instrumental belief. It is hard to see how the norms governing folk
psychological rationality could be any different from these.
The crucial differences come, not at the level of the norms governing the ascrip-
tions of rationality, but rather at the level of the reasoning that leads up to action.
We can illustrate this by comparing level 2 rationality with the inference-based
conception of practical rationality introduced through the passage from Davidson
quoted at the beginning of the chapter. As just observed, the overarching norms of
rationality governing ascriptions of rationality are the same at the two levels,
namely, that the instrumental belief should be accurate and that the action should
be appropriate relative to that instrumental belief. The difference comes in the way
"appropriateness" is calculated in the two cases. In essence, the inference-based
conception of practical rationality understands the appropriateness of an action
relative to an instrumental belief in terms of the possibility of constructing a valid
argument from the instrumental belief (and associated beliefs and desires) to a de-
scription of the action. As I showed in the first section of this chapter, no such un-
derstanding of appropriateness is available at the nonlinguistic level, because the
notion of formal inference is not applicable at the nonlinguistic level. So how
should appropriateness be understood at the nonlinguistic level?
It is tempting to put the point in terms of consistency, so that an appropriate ac-
tion was one that is consistent with the instrumental belief, and understanding an
action as appropriate is a matter of understanding it as consistent with the relevant
beliefs, perceptions, and background desires, but this would be unsatisfactory for
two reasons. The first is that many ways of acting in a manner consistent with an
instrumental belief do not involve acting on it. Not all of these would properly be
described as appropriate. Second, the notion of consistency can be viewed in two
ways, neither of which is applicable at the nonlinguistic level. Consistency might
be viewed as primarily an inferential matter, so that a set of beliefs, desires, and in-
tentions is consistent just if its members do not jointly entail a contradiction. But
this sort of consistency-based understanding of the appropriateness of a course of
action to a set of beliefs and desires would be little improvement over an
inference-based conception, which I have already shown to be inapplicable at the
nonlinguistic level. Things are no better if consistency is viewed in semantic
terms, so that a set of beliefs, desires, and intentions is consistent just if they can be
true/satisfied together. Understanding consistency is a matter of understanding the

consistency of a set of sentences—hence requires linguistic vehicles.
At the nonlinguistic level the appropriateness of a course of action relative to a
set of beliefs and desires, including an instrumental belief, can really only be un-
derstood as a matter of a creature straightforwardly acting on a given instrumental
belief in the light of its other beliefs and desires. It is to be assumed, in other
words, that an instrumental belief will be immediately acted on, unless there are
significant countervailing considerations—perhaps a potential threat from a preda-
tor, or the overwhelming energy cost of carrying out the relevant action. At the
nonlinguistic level, issues of rationality get a grip, therefore, in two ways. The first
way is through the accuracy of the instrumental belief—the rationality of the con-
templated course of action as a means of obtaining the required goal. The second
way is through the translation of the instrumental belief into action—relative to the
creature's other beliefs, desires, and needs.
As far as reasoning is concerned, level 2 rationality depends on what one might
term a straightforward practical syllogism—that is to say, an immediate translation
of an instrumental belief into action. A preliminary answer to the second question,
therefore, is that there are significant differences between level 2 rationality as ap-
plied to nonlinguistic creatures and the forms of rationality that we deploy in ex-
plaining and predicting the behavior of other language-using creatures. These are
differences, however, in the types of reasoning available at the different levels,
rather than in the overarching norms that govern ascriptions of rationality. The
overarching norms remain constant. What varies is the complexity of the decision-
making processes that lead up to the action whose degree of rationality is being as-
sessed. The question of the specific types of reasoning available at the nonlinguis-
tic level will be discussed in greater detail in the next chapter.
7
Practical Reasoning and Protolocjic
In the previous chapter I showed how the notion of rationality might be deployed in
psychological explanations of the behavior of nonlinguistic creatures. I identified
three different ways of characterizing behavior at the nonlinguistic level as rational.
Two of these conceptions of rationality are relevant to the project of psychological
explanation. These are what I termed level 1 and level 2 rationality. The distinctive
feature of level 1 rationality is that, although it applies to individual behaviors, it is
not properly describable as involving a process of decision-making. In this respect it
contrasts with level 2 rationality, which depends crucially on the existence of instru-
mental beliefs about how a given end might be achieved in a particular context.
In this chapter I will complete the account of psychological explanation by ex-
ploring how the notion of practical reasoning might be applied at the nonlinguistic
level. In section 7.1 I explore the idea, familiar from discussions of the language
of thought hypothesis, that practical reasoning should be understood in decision-
theoretic terms, both when we are dealing with linguistic and with nonlinguistic
creatures. I will show that the decision-theoretic model is not required for the
explanation of behaviors that are rational in either the level 1 or the level 2 sense.
Nonetheless, certain forms of practical reasoning are available at the nonlinguistic
level. In section 7.2 I show how it is possible to identify at the nonlinguistic level
certain very basic forms of protological thinking. These basic forms of protologi-
cal thinking include what I term protonegation and protoconditional reasoning.
The second of these in particular can help us to understand the instrumental beliefs
in behavior assessable according to the norms of level 2 rationality.
7.1 The Decision-Theoretic Model of Practical Reasoning

There is a basic distinction, drawn in this book as indeed in every discussion of
nonlinguistic thought, between brute reaction and intelligent, thinking behavior.
Thinking behavior involves flexible and variable responses to the environment. It

supports content-involving counterfactuals. It cannot be predicted on the basis of
knowledge of what is happening at the sensory periphery. And so on. These opera-
tional criteria of course all point toward the intuitive idea that thinking behavior re-
sults from deliberation on the environment as represented in the light of back-
ground representations and motivational states.
One powerful way of fleshing out this intuitive idea is the proposal that decision-
making at the nonlinguistic level might be decision-theoretic in nature. This con-
ception of nonlinguistic reasoning has traditionally been associated with the lan-
guage of thought hypothesis—on the plausible assumption that, if the actual process
of decision-making is decision-theoretic, then the various components of the
decision-making process (specifications of possible outcomes, calculations of pref-
erences, assessments of probability, and so forth) will need to be represented in a
language-like medium. In line with the general strategy of neutrality adopted in
chapter 2,1 will prescind from questions that have to do with the vehicle of non-
linguistic thinking. My concern in this section is with the basic claim that the lan-
guage of thought theory sets out to explain the idea that we can only make sense of
practical reasoning in decision-theoretic terms as a process of maximizing expected
utility.
Fodor (1975, 28-29) offers the following schematic model of practical decision-
making (see also Rey 1997).
1. A given creature finds itself in a certain situation S.

2. It believes that a certain set of behavioral options, Bj . . . Bn, is available
inS.
3. The creature predicts the probable consequences of performing each
of those behavioral options by computing a series of conditionals of the
form: If Bf is performed in S then consequences C(. will occur with a certain
probability.
4. A preference ordering is assigned to the consequences.
5. The creature's choice of behavior is determined as a function of the as-
signed preferences and probabilities.1
There is a basic distinction to be drawn between descriptive theories of decision-

making and normative theories of rationality. Descriptive theories are psychologi-
cal theories about how decisions are reached and courses of action fixed on,
whereas normative theories concern the rationality of the processes of decision-
making and of the eventual course of action. It is clear that Fodor offers his model
of practical reasoning as a descriptive psychological theory. He does not see it as
providing a normative theory of rationality, since it has nothing to say about the ra-
tionality or otherwise of the creature's beliefs about possible options and out-
comes; its assessments of the likelihoods of those outcomes; and its assignment of
subjective utilities.
In contrast, in chapter 6 I offered an account of nonlinguistic rationality in
Practical Reasoning and Protologic 135
which the theory of expected utility functioned as a normative theory. The norma-
tive criteria by which level 0 and level 1 rationality are to be judged include vari-
ous versions of expected utility theory. The variation comes because of the differ-
ent ways that utility can be understood at this level—in terms of rate of energy
procurement, individual lifetime fitness, inclusive fitness, and so on. In neither
level 0 nor level 1 rationality, however, should it be assumed that the normative
criteria in any way reflect the decision-making that goes on in behavior that is as-
sessed for rationality at the relevant level. Decision-making, properly speaking,
enters the picture only with level 2 rationality.2 There are ways of comparing pos-
sible courses of action and choosing between them that do not require anything
recognizable as a process of decision-making. As I pointed out in the previous
chapter, the affordances of a particular course of action can be manifest in the con-
tent of perception. The leopard does not look at the antelope and work out that if it
pursues the antelope it will have a high probability of securing a considerable
amount of food. When it sees the antelope it sees the affordance of food. In situa-
tions such as these, the instrumentality of a particular course of action is manifest
in the content of perception—and of course a single perception can reveal several
different potential courses of action.
It will be helpful to work through an example of behavior assessable according
to the norms of level 1 rationality. Consider the roaring contests of red deer stags
(Clutton-Brock and Albon 1979). During the rutting season red deer compete with
each other for the control of groups of females. Actually fighting each other is
risky and exhausting for both winner and loser. So fighting contests are frequently
replaced by roaring contests, where the winner is the loudest and most sustained
roarer and the loser backs down. A male's capacity for roaring is a good index of
his strength and consequently of how he would have fought had he been called
upon to do so. The dispute is frequently resolved by the roaring contest. However,
if the roaring rates are more or less similar and fail to identify a "winner," it is
usual for the contest to shift to a second stage in which the deer perform a parallel
walk that seems to provide information about body height and antler size (Clutton-
Brock and Albon 1979). If that also turns out equal then the deer lock antlers and
proceed to fight.
In the abstract, and for the purposes of normative evaluation, the contest can be
modeled as a series of choices. The behavioral options for each stag at each choice
node are simple and can be represented diagrammatically (see fig. 7.1). The first
choice node represents the decision to embark on the roaring contest at all and the
options are: Roar or Back Down. Presumably there will be some stage during the
initial roaring contest during which the stag has to decide whether to continue roar-
ing or to back down. Hence: Continue Roaring or Back Down. If at any stage dur-
ing the roaring contest the other stag backs down then the overall competition is
over. If, however, neither participant backs down then at the end of the roaring
contest each stag will face the choice of terminating the contest or entering the
next stage: Back Down or Parallel Walk. During the parallel walk stage, each stag
has to decide whether to continue walking or to back down: Continue Walking or
Figure 7.1 The choice nodes in a model of red deer roaring contests.
Back Down. At the end of the parallel walk stage, the choice is between retreating
and progressing to the pushing contest: Back Down or Lock Antlers.
It is clear that some version of normative expected utility theory can be de-
ployed to determine the respective payoffs at each node in the choice-tree. The
stags are competing for groups of females, and it is natural to think that the rele-
vant currency will be some form of inclusive fitness. In calculating the expected
utility of the different behavioral options at each point in the choice-tree, we need
to take into account the overall benefits in terms of inclusive fitness of controlling
the group of females, since control of the group will be one of the two outcomes
for each of what might be termed the proactive strategies: Roar, Continue Roaring,
Parallel Walk, Continue Walking, Lock Antlers. For the first four of these proactive
strategies, the second of the two outcomes will be straightforward—essentially a
controlled retreat without physical damage. This can straightforwardly be trans-
lated into the currency of inclusive fitness. Things are slightly more complicated
at the final decision-node, since the second outcome of the Lock Antlers strategy
is a damaging defeat that is potentially life-threatening and definitely fitness-
decreasing. The nonproactive strategies (essentially Back Down at each choice
node) all have a single possible outcome, rather similar to the second outcome of
the first four proactive strategies.
Once we have determined the payoffs for the different outcomes associated with
each strategy available at the relevant decision-nodes, the next step is to derive an
expected utility calculation by calculating the probability of each of the potential
outcomes. The calculation will not be easy, since the utility of the different out-
comes in the early stages is partly a function of what can reasonably be expected to
happen if proactive strategies are adopted in the later stages. Nonetheless, it is
manageable. It is relatively clear, therefore, how we might develop and apply a
normative framework for thinking about red deer roaring contests. But recall that
we are investigating whether we need to model choice behavior in a way that takes
the normative theory by which we assess the rationality of a course of action as a
structural guide for the descriptive theory. The question, really, is whether the ap-
plicability to the red deer stages of normative criteria based on considerations of
expected utility depends on our viewing the deer as themselves making calcula-
tions of expected utility.
It is clear how we, as observers, assign probabilities to the relevant outcomes.
We might employ our knowledge of the respective ages of the two stags, of how
successful they have each been in previous contests, of how damaged they are
from fights in the past, and so on. Obviously, in most cases neither stag has access
to this type of information. So how can they calculate probabilities at all? The
short answer is that they do not need to. The relevant information is directly pres-
ent in the content of perception. The overarching rationale for the two early stages
of the contest is that they each allow the participant to pick up information about
the likely outcome of an aggressive conflict. A male's capacity for roaring is a
good index of his strength, and consequently of how he would fight were he called
on to do so—since roaring uses the thoracic muscles that are also employed in
fighting. And, of course, it is not just that the male who roars loudest will have the
greater probability of defeating his opponent in an actual physical contest than the
male who loses the roaring contest. The male who roars the loudest will also have
a higher probability of being able to defend the group of females. Both types of
information are directly available in the content of auditory perception. The same
holds for the parallel walk section of the contest, although in a different sensory
modality. The parallel walk transmits information about the relative heights
and muscle sizes of the two participants—information that is highly relevant to
the expected utility calculation. The adaptive aim of the parallel walk section is,
of course, that it allows both participants to pick up this information simply by
looking—hence without having to engage in any sort of conflict.
The red deer example shows, I think, how we can place a wedge between the
normative and descriptive dimensions of our theories of practical decision-making
as applied to the behavior of nonlinguistic creatures. From a normative point of
view we can understand the roaring contests within the framework of expected
utility theory. We can also employ this framework for predictive purposes. Given
our knowledge of the different possible strategies and the payoffs and probabilities
of their respective outcomes, we can (on the plausible assumption that both par-
ticipants are suitably sensitive to the relevant parameters) predict how far down the
choice-tree each stag will be prepared to travel. But we can do this without having
to say that the stags are doing anything comparable to calculations of expected
utilities. For the stag the behavioral options are manifest; Roar or Back Down, and
so on. The contrast-space of alternative courses of action does not need to be repre-
sented propositionally. What determines which course of action is selected? The
most obvious candidate is the information about likely outcomes that is transmit-
ted once the roaring contest has been entered into. This is information about likely
outcomes, but it is not necessarily registered as information about likely outcomes.
The animal just sees what to do by comparing its roar with its rival's roar, or the re-
spective shoulder heights and musculatures revealed in profile.
It looks, therefore, as if a large category of behaviors for which psychological
explanations seem appropriate, hence that involve genuine choice, can be modeled
in a way that does not implicate anything like the machinery of practical decision-
making envisaged by Fodor. But I have so far confined myself to what in the
previous chapter I identified as level 1 rationality. Surely, the defender of the
decision-theoretic model is likely to object, consequence-sensitive reasoning of the
decision-theoretic type is implicated in level 2 rationality. I defined level 2 ration-
ality as involving genuine decision-making and grounded in the representation of
contingencies. Is this not decision-theoretic?
To say that the decision-making that generates behavior assessable at level 2 ra-
tionality is consequence-sensitive in form is not, however, to say that it takes the
form of a calculation of expected utility in the manner implied by Fodor's psycho-
logical model. The point that I stressed in discussing level 2 rationality and the
concomitant practical reasoning is that the consequence-sensitive dimension enters
the picture via instrumental beliefs rather than via complex calculations of prefer-
ence ordering and utility. Creatures (both linguistic and nonlinguistic) can act on
the basis of a comparison of contingencies without calculating which course of ac-
tion in the contrast-space of possibilities has the highest likely benefits when each
course of action's potential benefits are discounted by their probability. The seduc-
tiveness of the decision-theoretic model of practical reasoning comes from the idea
that consequence-sensitive comparison of alternative courses of action requires a
two-stage partition of the space of possibilities—an initial partition into differ-
ent available courses of action followed by a further partition, for each of those
courses of action, into its different potential outcomes. But this is not, of course,
the only possible model. Once again the conflation of the normative and the de-
scriptive threatens to mislead. It is certainly true that, when we are assessing the
rationality of a particular course of action, we need to consider the full range of
possible outcomes. We cannot come to a view about whether a creature acted ra-
tionally in a particular situation without taking into account both the potential
benefits that might have motivated it and the potentially disastrous consequences
that ought perhaps to have put it off. But the creature in question does not need to
make a comparable comparison and calculation. It is quite possible that all it needs
to motivate its action is an instrumental belief to the effect that acting in that par-
ticular way is the only way of deriving the benefits in question.3
Let us distinguish then between explicit decision-theoretic reasoning and instru-

mental reasoning. Only the first requires the sort of comparative calculations en-
visaged by Fodor. The point and applicability of the contrast becomes very clear
when we consider the examples of level 2 rationality put forward in chapter 6. The
construction and use of tools loomed large there, and it is hard to see how these be-
haviors are best modeled in terms of the two-stage partition assumed in the de-
scriptive application of the expected utility model. It seems far more appropriate to
describe the chimpanzee as whittling down a twig because it believes that this will
enable it to use the twig to extract termites from the termite nest than as settling on
that course of action because it has calculated that it offers the highest expected
utility. Again, the normative and descriptive dimensions come apart. We, as theo-
rists, are perfectly capable of assessing the rationality of the chimpanzee's behav-
ior by applying some form of the two-stage partition and calculating not simply the
likely outcomes, costs, and benefits of the tool construction but also the likely out-
comes, costs, and benefits of the other activities in which the chimpanzee could
have been engaged at that time. But the chimpanzee doesn't have to do all that.
The same holds for the construction of hand-axes by early hominids. The machin-
ery that we employ to assess the rationality of tool construction and manipulation
is not a reconstruction of the processes of practical reasoning implicated in those
behaviors.
But why should we treat a behavior so grounded as genuinely thinking behavior
and hence as a candidate for assessments of rationality at all? Do we not pay a
heavy price for prising apart the descriptive and the normative by so diluting the
type of practical reasoning at stake? Two points can be made in response. The first
is that it is simply not the case that no comparison is involved. The environment af-
fords many possible courses of action, and creatures operating at level 2 do indeed
fix on a single member of the relevant contrast-space. Tool construction is hardly a
tropistic behavior. There may even be more than one instrumental belief in play. Of
course, if there is more than one instrumental belief then it is likely that some com-
parison of goals and outcomes will be required, but there is no reason to think that
this will require complicated machinery of the sort postulated by Fodor. The sec-
ond point is that, although the relevant behaviors do most likely involve compar-
isons of different possible courses of action, that is not why they are being treated
as intentional behaviors and consequently as candidates for psychological explana-
tion. The intentionality of the behavior derives from two considerations. The first
is the creature's sensitivity to the consequences of its behavior and to the contin-
gency (or lack of it) between behavior and desired outcome. This explains why we
need to assume that the creature is genuinely representing the contingency and
the instrumentality. The second factor is that the behavior does not seem to be the
product of trial and error. This rules out taking the representations involved at the
level of thinking-how and consequently rules out analysis in terms of the minimal-
ist conception of nonlinguistic thought discussed in chapter 3.
It would seem, therefore, that at the nonlinguistic level we have models of prac-
tical reasoning and deliberation that do not implicate the complicated inferences
and calculations that Fodor suggests are required for any form of thinking behav-
ior. The next section will explore these models, and the types of protological think-
ing that they involve, in more detail.
7.2 Inference and Protologic

In section 6.11 suggested that there is no hope of applying an inference-based con-
ception of rationality at the nonlinguistic level, and in the previous section it
emerged that the decision-theoretic model is equally inappropriate for characteriz-
ing the practical reasoning of nonlinguistic creatures. This poses a challenge. The
notions of thought and reasoning are inextricably linked. There is no sense in
which there can be thought without reasoning, and many of the arguments put for-
ward in earlier chapters to show, for example, that nonlinguistic thoughts have
genuine structure and determinate contents have hinged on claims about the role
that such thoughts must play in reasoning. So we clearly need an account of the
particular types of reasoning that are available at the nonlinguistic level.
The issue becomes particularly pressing when we look ahead to the conclusions
of the next two chapters. In chapter 8 I show that there are definite limits to the
scope of nonlinguistic thinking. Certain types of thinking are in principle only
available to creatures who dispose of a language. The fundamental reason for the
restriction is that, for reasons that will be explored in the final two chapters,
thoughts can only be the objects of further thoughts when they are linguistically
vehicled. Intentional ascent (that is to say, thinking about thoughts) requires the
possibility of semantic ascent. Some thoughts that involve intentional ascent are
explicitly targeted on first-order thoughts. This is the case, for example, with those
forms of psychological understanding that involve attributing thoughts to others
(see section 9.3). But intentional ascent is also involved (or so I shall argue) in
types of thinking involving operators, such as tense operators and modal operators,
that govern entire thoughts. If this line of argument is sound then it rules out think-
ing about the reasoning of nonlinguistic creatures in terms of elementary logical
concepts. By 'elementary logical concepts' I mean those logical concepts associ-
ated with the standard connectives in the prepositional calculus—the concepts of
conjunction, negation, and disjunction and the material conditional. As the term
"prepositional calculus" suggests, these concepts are adverbial in precisely the
same way as the concepts associated with modal and tense operators. They apply
to complete thoughts (propositions). It is complete thoughts that are negated, con-
joined, disjoined, and connected hi the manner characteristic of the material condi-
tional. Hence their application is a process of intentional ascent that requires se-
mantic ascent.
The challenge, therefore, is to identify forms of reasoning at the nonlinguistic
level and then explain them in a manner that does not require the reasoner to de-
ploy elementary logical concepts. In identifying these types of reasoning we
should be guided, as usual, by thinking both about how they might be manifested
in behavior and about how they might be attributed to explain behavior. In this sec-
tion I shall discuss three such basic types of reasoning that it would be natural
(when thinking about language-using creatures) to characterize in terms of mastery
of certain primitive basic logical concepts. As I will show, there is an alternative
way of understanding them at the nonlinguistic level.
The first type of reasoning can be described as reasoning from an excluded alter-
native. This is the type of inference that takes a creature from recognition that one
of an incompatible pair of states of affairs holds to the recognition that the other
does not hold. Here is an example. Imagine a creature that has learned that the lion
and the gazelle will not be at the watering-hole at the same time and, moreover, is
in a position to see that the gazelle is drinking happily at the watering-hole. The
creature can conclude with confidence that the lion is not in the vicinity. This type
of reasoning is one of the ways a creature can learn about what is not immediately
perceptible. One can see easily, for example, how this sort of inference could be
life-preserving for a creature who is just as threatened by the lion as the gazelle is.
It is natural to formalize it in the prepositional calculus as an instance of disjunc-
tive syllogism (the transition from "A or B" and "not-A" to "B," where A stands
for "The gazelle is not at the water-hole" and B for "The lion is not at the water-
hole").
A second such way of moving beyond the here-and-now comes with straightfor-
ward conditional reasoning of the form standardly termed modus ponens—that is,
the reasoning that takes one from recognition that there is a conditional depend-
ence between two states of affairs (the second will be the case if the first is the
case) and a recognition that the first state of affairs is indeed the case to the conclu-
sion that the second state of affairs is the case. Conditional reasoning of this type is
deeply implicated in a range of different activities. The detection of patterns of be-
havior seems closely bound up with the possibility of conditional reasoning. A
creature that knows that if the gazelles see the lion they will run away and that rec-
ognizes (on the basis of its understanding of the gazelles' visual perspective, as
will be discussed in section 9.3) that the lion will shortly be detected by the
gazelles is in a position to predict that the gazelles will soon take flight.
The third fundamental type of inference is also based on recognition of a condi-
tional dependence between two states of affairs—but in this case (formalized in
terms of modus tollens) the reasoning proceeds from recognition that the second
state of affairs does not hold to recognition that the first state of affairs is not the
case. So, for example, to stick with the gazelles, an observer (perhaps a fellow
predator) who is too far away to have a view about the visual perspective of the
gazelles can infer from the fact that they are happily feeding where they are that
they have not yet seen the lion.
In standard propositional logic these three fundamental forms of inference are
understood in terms of the three propositional operators of disjunction, negation,
and the material conditional. What I have called reasoning from an excluded alter-
native is standardly understood in terms of the inference schema of disjunctive syl-
logism (Either A or B, not-A. Therefore B), with the relevant operators being dis-
junction and negation. The two types of conditional reasoning are both understood
in terms of the material conditional (If A, then B). All three operators are functions
from propositions to propositions and hence take complete thoughts as both argu-
ments and values. Clearly, given my comments in the previous chapter about the
inference-based conception of rationality (and anticipating the argument from in-
tentional ascent to be given in the next chapter), if we are to find analogues of
these three types of reasoning at the nonlinguistic level then we will need to find
ways of understanding them on which they do not involve propositional operators.
We can make a start on this by simplifying the problem. Reasoning from an ex-
cluded alternative can be understood as a form of conditional reasoning, rather
than in terms of disjunctive syllogism. Let A stand for the sentence "The gazelle is
not at the watering-hole" and B for the sentence "The lion is not at the watering-
hole." Then the disjunction "A or B" is truth-functionally equivalent to the condi-
tional "If not-A, then B" (that is to say, "If the gazelle is at the watering-hole, then
the lion is not at the watering-hole"). Both sentences will be true just if it is not the
case that the gazelle and the lion are both at the watering-hole.4 The process of rea-
soning from an excluded alternative will involve a grasp on the one hand of the
conditional "If not-A, then B" and on the other of the antecedent of that condi-
tional (where that antecedent is in some sense negative). All we need, therefore, is
to find a way of understanding analogues for negation and the conditional that do
not operate on complete thoughts.
Let us start with negation. Modern, that is to say post-Fregean, logic is founded
on the idea that, as far as the fundamental logical form of sentences is concerned,
the linguistic act of negation applies essentially to sentences—and correlatively, at
the level of thought, that negation is a logical operation on propositions. It may
seem that, in the sentence "Socrates is not wise" (and still more so in the sentence
"Socrates is unwise"), a particular property, the property of wisdom, is being held
not to apply to Socrates. However, the surface form of natural language sentences
is deceptive. The negation operator actually applies at the level of the sentence
"Socrates is wise" rather than at the level of the predicate "— is wise."5 The sen-
tence "Socrates is unwise" is a sentence that is true just if the sentence "Socrates is
wise" is false. Many philosophers have thought that this obscures an important dis-
tinction. The sentence "Socrates is wise" can be false in circumstances in which
Socrates does not exist—such as now, for example. Yet these are not, many have
thought, circumstances in which it would be appropriate to say that Socrates is un-
wise. One way of putting the point would be to say that, whereas the two sentences
"Socrates is wise" and "Socrates is unwise" are contraries (i.e., they cannot both
simultaneously be true), the two sentences "Socrates is wise" and "It is not the case
that Socrates is wise" are contradictories (i.e., one or other of them must be true).
This, in fact, is how the distinction between predicate negation and sentential
negation was originally put by Aristotle in the Prior Analytics (1.46). Aristotle
insisted, and in this he was followed by almost all logicians until Frege, that there
is a fundamental logical difference between negating a sentence and negating a
predicate.
There is no need to go into the question of whether the distinction between
predicate negation and sentential negation is a genuine logical distinction (as op-
posed, for example, to a distinction in the pragmatics of ordinary language best ac-
commodated at the level of conversational implicature)—or the related question of
whether a Fregean or an Aristotelian account of negation is a better way to under-
stand the way negation operates in ordinary language.6 For present purposes the
important point is that the distinction between predicate negation and sentential
negation gives us a way of understanding negation (or rather, protonegatiori) at
the nonlinguistic level as involving a thought with a negative predicate (subject
to the qualifications to be noted in the next paragraph)—as opposed to the truth-
functional construction of a complex thought. In terms of understanding thought at
the nonlinguistic level, the problem of understanding how a creature without lan-
guage can be capable of negation becomes the problem of how a creature without
language can think thoughts in which the predicate component is one rather than
the other of a parr of contraries. The task becomes one of understanding how the
nonlinguistic creature can grasp pairs of concepts that are contraries—the concepts
of presence and absence, for example, or of safety and danger, or of visibility and
invisibility. To return to the observer at the water-hole, we should understand the
thought that the gazelle is not at the water-hole as the thought that the gazelle is ab-
sent from the water-hole—rather than as the denial of the thought that the gazelle
is at the water-hole. Such a thought would be the contrary of the thought that the
gazelle is at the water-hole—but it would not be constructed from that thought.
It is often put forward as an objection in principle to the idea of predicate nega-
tion that there is no principled way of distinguishing positive predicates from neg-
ative predicates. To take Frege's example, which of the pair of contrary predicates
'mortal' and 'immortal' is positive and which negative? Is 'immortal' negative, on
the grounds that predicating immortality of something is denying it mortality? Or
is 'immortal' positive, on the grounds that mortality is a lack of immortality? There
is clearly no way of deciding this matter. But as far as protonegation is concerned,
there is no need to classify one of a pair of contrary properties as positive and the
other as negative. The important point is simply that it is understood (on the part of
the protonegater) that no thing can simultaneously be characterized in terms of the
two contrary properties. This is sufficient to permit primitive versions of the two
basic types of inference involving negation that I identified earlier. The first type of
inference involves reasoning from an excluded alternative. Consider the earlier ex-
ample of the gazelle and the lion at the watering-hole. We are trying to characterize
how a creature might reason from the thought that the gazelle is at the watering-
hole to the thought that the lion is not at the watering hole. The reasoning here can
be assimilated to standard conditional reasoning by treating the central premise as
a conditional—namely, the conditional that if the gazelle is at the watering-hole
then the lion is not at the watering-hole. The notion of protonegation shows how
this can be understood without deploying prepositional negation. The conditional
in question becomes "If the gazelle is present (at the watering-hole) then the lion
is absent (at the watering-hole)." Grasping this thought (apart from the need, to
be explored further hereafter, to develop a nonlinguistic analogue of the truth-
functional conditional operator) is a matter of understanding that presence and ab-

sence are contrary concepts. Any creature that understands that presence and ab-
sence are contrary concepts will also be able (again subject to a satisfactory ac-
count being given at the nonlinguistic level of conditional reasoning) to undertake
reasoning approximating to modus tollens. Starting with the conditional "If the
gazelle is present (at the watering-hole), then the lion is absent," such a creature
will be able to protonegate the consequent by forming the thought that the lion is
present and hence to arrive at the protonegation of the antecedent (namely, "The
gazelle is absent"). This inference is not valid in virtue of its form in the way an in-
stance of modus tollens is valid in virtue of its form—that is to say, it is not an in-
stance of a valid argument-schema. But it is of course valid in the semantic
sense—that is to say, its premises cannot be true and its conclusion false.
It might be objected to the proposed understanding of protonegation that the
very idea of contrary properties is parasitic on propositional negation—hence that
it will not be possible for a creature to grasp the idea that two properties might be
contraries without a grip on the notion of propositional negation. Two properties
are contrary just if it is not possible for them both to be true of a given thing at a
given time—from which it appears to follow that the concept of contrariety pre-
supposes the concept of propositional negation.7 What this line of objection neg-
lects, however, is that a creature can master pairs of contrary concepts (such as the
concepts of presence and absence) and deploy those concepts in inferences using
protonegation without a full understanding of the notion of contrariety. It is no
more plausible to think that the effective deployment of contrary concepts requires
a theoretical grasp of the concept of contrariety than it is to demand that the effec-
tive deployment of number concepts requires a theoretical grasp of the concept of
number.
The notion of protonegation that these reflections yield is obviously very differ-
ent from the truth-functional understanding of negation that dominates contempo-
rary logic. But this is not a disadvantage. It may well be the case that, as Frege em-
phatically argued, there is no room in logic and the analysis of language for
anything analagous to what I am calling protonegation. That would hardly entail,
however, that protonegation cannot be deployed at the nonlinguistic level. Frege's
arguments in his essay "Negation" (1918-19a) are all devoted to trying to show
the absurdity of starting with a complete thought and then construing negation as
an operation on a component of that thought. Nothing like that is involved in pro-
tonegation, however, for the simple reason that protonegation is not an operation
on complete thoughts. It may be that protonegation leaves no traces at the level
of full-fledged linguistic thought—that the relation of contrariety drops out of the
picture in the way that certain principles of infant naive physics drop out of the
picture in the normal course of development (as I showed in chapter 3). Protonega-
tion may be a ladder that is thrown away on the way to mastery of the full-fledged
concept of negation as a truth-functional operation on complete thoughts. We
might take as a model Quine's proposal (1974, 89-92) that children learning the
relative pronoun begin by treating relative clauses as self-contained grammatical
entities (effectively complex predicables) before learning that relative pronouns in

fact operate as conjunctions paired with simple pronouns.8 Or it may be, as Aristo-
tle originally suggested, that doing justice to the full complexity of negation re-
quires accommodating the distinction between contradictories and contraries and
hence allowing that both predicates and sentences can be negated. Were that the
case, then there would less of a radical discontinuity between thought at the lin-
guistic level and thought at the nonlinguistic level.
This discussion of protonegation still leaves us with an important challenge. We
need an analogous way of understanding how some precursor of the conditional
operator can operate at the nonlinguistic level. The conditional operator is a truth-
functional propositional operator forming a complex thought from two thoughts in
such a way that the complex thought is true in all circumstances except those in
which the first component thought is true and the second component thought is
false. Clearly this requires intentional ascent and hence (by the arguments to be ex-
plored in the next chapter) is unavailable at the nonlinguistic level. Equally clearly,
the strategy of finding an intrapropositional precursor for the propositional opera-
tor that led to the concept of protonegation holds no promise in the case of the con-
ditional. Conditional thought clearly links two different things. But if those things
cannot be complete thoughts, then what can they be?
The proposal I would like to explore is that we look for the sources of condi-
tional reasoning in a primitive form of causal reasoning. Whereas conditional rea-
soning involves a propositional operator establishing a truth-functional relation be-
tween complete thoughts, causal reasoning works on the basis of a causal
condition holding between one state of affairs and another. Since causal relation-
ships do not hold between complete thoughts, an understanding of causality pre-
supposes no intentional ascent and hence does not require language.
Philosophers frequently reflect on the relations between causation and condi-
tionals, and it is often suggested that a proper understanding of conditionals will be
an ingredient of an adequate account of causation.9 At the very least, a causal ex-
planation of a particular event entails certain conditional predictions about what
would happen in suitably similar situations.10 It is clear that, in the order of analy-
sis, thought and talk about conditionals is more fundamental than thought and talk
about causation. No one has ever proposed that we understand conditionals in
terms of causation. Conditional sentences and conditional thoughts assert the exis-
tence of dependence relations, and causation is just one of a range of possible de-
pendence relations. Nonetheless, the order of acquisition frequently fails to dupli-
cate the order of analysis. It is highly plausible on experimental and observational
grounds that the capacity for causal cognition is very widespread in the animal
kingdom and available at a very early stage in human development (Dickinson and
Shanks 1995 and the other essays in Sperber 1995)—which is exactly what one
would predict on evolutionary grounds. The ability to detect certain types of causal
regularity and to distinguish genuine causal relations from accidental conjunctions
has obvious survival value. Causal dependence relations are directly observable,
highly salient, and pragmatically significant in a way that no other dependence
relations are. It seems plausible both that causal relations should be more primitive
than conditional relations and that a creature arrives at an understanding of condi-
tional dependence by abstracting away from the more familiar and everyday rela-
tion of causal dependence. Perhaps the child's first step toward an understanding of
conditional dependence is observing that a certain relation holds between the truth-
values of two separate thoughts when a causal relation holds between the states of
affairs that those thoughts characterize. Once this first step (which of course in-
volves intentional ascent and hence requires language) has been taken, it is a rela-
tively straightforward matter to notice that there are other types of situation (and
correlatively other types of dependence relation) that share that feature. And thus
the abstract concept of conditional dependence is grasped.
How might causality be understood at the nonlinguistic level? There are certain
aspects of the full-fledged concept of causation that clearly cannot be available at
the nonlinguistic level. The full-fledged understanding of causation has a modal di-
mension that comes with the thought that a cause is sufficient for the effect it
brings about, and this is, in effect, the thought that it is not possible for the cause to
occur without the effect occurring. Yet if we accept my earlier suggestion that
modal thinking involves a type of intentional ascent and hence requires semantic
ascent, the notion of sufficiency is not available at the nonlinguistic level. Different
theorists will view this with different degrees of concern. Some analyses of causa-
tion take the idea that a cause is sufficient for its effect as central (e.g., Mackie
1965). Other accounts do not. If we follow Mellor (1995) in holding that what
makes it the case that one fact c causes another fact e is that the conditional proba-
bility of e given c is greater than the conditional probability of e given not-c, then
cases in which causes really are sufficient for their effects cease to be central to the
understanding of causation. But there is no suggestion that nonlinguistic creatures
can have a full understanding of causation. The proposal is simply that (at least
some) nonlinguistic creatures have a basic capacity to track causal relationships
holding between events or facts and that this basic capacity allows them to engage
in a primitive form of conditional reasoning.
All accounts of causation, from David Hume's pioneering account onward, are
agreed that certain forms of regularity are at the heart of the notion. And it seems
overwhelmingly plausible that the core of the understanding of causation, or more
accurately protocausation, at the nonlinguistic level will be based on the register-
ing of regularities in the distal environment. It is easy to see where this type of un-
derstanding might originate. On the one hand, it seems plausible to take a sensi-
tivity to environmental regularities to be a basic part of the innate endowment of
any creature capable of learning about the environment. On the other, one might
expect any creature to be peculiarly sensitive to regularities between its own ac-
tions and ensuing changes in its immediate environment. Of course, however, as
regularity theories of causation have been forced to acknowledge, there are many
regularities that are not causal, and it is in the capacity to distinguish genuinely
causal regularities from accidental regularities that one might expect differences
between different species of nonlinguistic creature and, for that matter, different
stages of development within any given species. We know from dishabituation

studies of infant perceptual expectations about object behavior that even when
they are only a few months old infants show surprise at examples of "action at a
distance," and many developmental psychologists have suggested that their under-
standing of the physical world is governed by the principle that objects can only
interact causally when they are in physical contact (Spelke 1990).
The regularities to which nonlinguistic creatures are sensitive (unlike those that
are usually stressed in regularity analyses of causation) need not be exceptionless.
Indeed, there are very good evolutionary reasons why one would expect causal
cognition to be sensitive to probabilistic regularities (Brunswik 1943). Decision-
making in the wild is decision-making under uncertainty, and no creature that
waited for an exceptionless regularity would fare well in evading predators and ob-
taining food. This is another reason not to be concerned about the unavailability at
the nonlinguistic level of the idea that causes are sufficient for then" effects. It may
well be the case that every apparent example of probabilistic (or indeterministic)
causation at the macro level can be explained in terms of hidden variables, so that
what looks like a probabilistic regularity is really a manifestation of a deeper un-
derlying exceptionless regularity. If this were the case then the only reasons for
adopting a probabilistic understanding of causation would be apparent examples of
indeterministic causation at the microphysical level. But the emphasis is very dif-
ferent when it comes to nonlinguistic creatures' understanding of protocausation.
As far as the practicalities of foraging and reproducing are concerned, the apparent
probabilistic regularities are paramount.
Protocausal understanding tracks relationships, which can be either determinis-
tic or probabilistic, between states of affairs. This is why an understanding of pro-
tocausation is available at the nonlinguistic level. It also explains why primitive
versions of certain fundamental inference forms are available at the nonlinguistic
level. Let us return to the three basic inference forms. We are looking for ana-
logues at the nonlinguistic level of the basic inference forms of disjunctive syllo-
gism, modus ponens and modus tollens. The basic logical operations involved here
are negation and the conditional. If negation at the nonlinguistic level is under-
stood in terms of the mastery of pairs of contrary predicates, as suggested in the
earlier discussion of protonegation, and if we view the relevant conditionals as
protocausal conditionals holding between states of affairs rather than between
complete thoughts (e.g., between modes of presentation of those states of affairs),
then we have all we need for nonlinguistic analogues of our three basic forms of
inference. Modus ponens can be understood straightforwardly in terms of a causal
conditional together with an understanding, which may take the form of a percep-
tion or a memory, that the antecedent holds. The consequent will straightforwardly
be detached. We can view modus tollens in terms of the combination of a causal
conditional with the protonegation of the consequent of that causal conditional re-
sulting in the detachment of the protonegation of the antecedent. As I showed ear-
lier, the disjunctive syllogism "A or B, not-A, therefore B" can be understood in
terms of a causal conditional with not-A as its antecedent and B as its consequent.
By way of concluding this section, I shall relate this discussion of the forms of
inference available at the nonlinguistic level to the earlier discussions of nonlin-
guistic rationality. The first point to make is that nothing that has been said about
these types of inference is incompatible with the claim that what I termed the
inference-based conception of practical reasoning is unavailable at the nonlinguis-
tic level. The mere fact that some types of inference (or protoinference) can be de-
ployed by nonlinguistic creatures does not mean that we can understand practical
reasoning at the nonlinguistic level in terms of the inference-based conception. Let
us look again at the passage from Davidson in which this conception of practical
reasoning (and, concomitantly, of rationality) is sketched out.
was desirable. . . . If we can characterise the reasoning that would serve we will, in
provide him with the premises of an argument. (Davidson 1978/1980a, 85-86)
The proposal is to understand an agent's practical reasoning in terms of the reason-

ing that might have led him, starting with his beliefs and desires, to fix on the rele-
vant course of action. There are two principal reasons why the three basic forms of
protoinference I have been considering will not support this form of reasoning.
One of these reasons is a consequence of my earlier argument that intentional as-
cent requires semantic ascent, while the other has to do with the distinctive and re-
stricted character of nonlinguistic protoinferences.
I shall start with the general reason. Davidson's conception of practical reason-
ing has the practical reasoner starting with a particular belief-desire pair (or, more
realistically, a particular set of beliefs and pro-attitudes) and reasoning from those
to an intention to act in a certain way. One might expect this type of reasoning to
involve, at the very least, reflection on what outcomes would be the best way of
satisfying those relevant desires or pro-attitudes and on the likelihood of different
courses of action generating those outcomes. Yet reflection of this type seems a
clear example of second-order reflection on one's own motivations and on the in-
formation one possesses about the world. In order to be able to engage in this form
of practical reasoning, one needs to be able to take one's beliefs and desires as ob-
jects of thought—to think about the consequences that follow from one's desires in
the light of one's beliefs. And, if the argument in 8.2 is sound, such intentional as-
cent requires semantic ascent and hence cannot be available at the nonlinguistic
level.
Even putting this general worry to one side, there is a more specific reason block-
ing the application of the inference-based conception of practical reasoning at the
nonlinguistic level. The first is that protoinferences at the nonlinguistic level are not
made in virtue of their form. Creatures who engage in, for example, proto-modus
tollens need not have any grasp of the form of an inferential transition as truth-
preserving. In fact, they cannot have any such grasp, since that would involve
second-order reflection on the evidential relations between propositions and hence
semantic ascent (which explains why, as I showed in chapter 5, there is no evidence
that nonlinguistic creatures have any grasp of logical concepts). Connected with this
is the fact that any given creature will only be able to make a limited number of pro-
toinferences. This is particularly clear in the case of protonegation, which can only
be deployed by a creature that has at its disposal the appropriate pair of contrary
concepts. A nonlinguistic thinker can only negate those thoughts with a predicative
component for which it can generate a contrary. That is the limitation of proto-
negation—as opposed to propositional negation, which permits the negation of any
thought that can be entertained.
In terms of the three levels of nonlinguistic rationality identified in chapter 5,
the protoinferences I have been discussing offer a further perspective on behavior
that is a candidate for level 2 rationality. The key feature of practical reasoning
that is a candidate for level 2 rationality is that it involves instrumental beliefs—
instrumental beliefs that might feature in thinking about how to manipulate tools
or about how an action might lead to a particular outcome. These instrumental be-
liefs can be understood as causal conditionals connecting a particular type of ac-
tion with a particular type of outcome, and the three different types of protoinfer-
ence offer different ways that these instrumental beliefs can feed into action. Not
all causal conditionals, of course, are instrumental beliefs. A causal conditional
relates states of affairs, and there is no need for either of those states of affairs to
involve an action on the part of the agent. Causal conditionals can track regulari-
ties holding between events in the physical world as well as regularities in the be-
havior of other agents, and so one would expect some form of protoinference
to feature in behavior that is a candidate for level 1 rationality (although not in be-
havior to be judged by the norms of level 0 rationality). It may well be that
proto—modus ponens is more widespread than either of the two protoinferences in-
volving protonegation—simply because creatures might plausibly be expected to
become capable of tracking regularities between states of affairs before they are
capable of grasping pairs of contrary concepts. Certainly the empirical evidence
strongly suggests that rudimentary forms of causal cognition are very widespread
in the animal kingdom and operative early on in ontogeny (Sperber 1995).
8
Language and Thinking about Thoughts
At various points in this book I have examined different conceptions of the relation
between linguistic and nonlinguistic thought, and correspondingly of the scope and
limits of nonlinguistic thought. These different conceptions fall naturally into two
broad groups. At one extreme are views that seek to drive as much of a wedge as
possible between linguistic and nonlinguistic thought. In chapter 3, for example, I
considered the minimalist conception of nonlinguistic thought, according to which
nonlinguistic creatures are capable only of an etiolated and imagistic type of think-
ing, the vehicles of which are "spatial images superimposed on spatial percep-
tions." If the minimalist conception is correct then it will provide us with a very
clear sense of the differences between linguistic thought and nonlinguistic thought.
So too would a view that goes naturally with the minimalist conception—the view,
namely, that considerations of sense and mode of presentation can get no grip in
the absence of language. But much of this book has been devoted to arguing that,
initial appearances to the contrary, there is indeed room at the nonlinguistic level
for the ascription of thoughts that have determinate content and compositional
structure—and, more strongly, that such thoughts are indispensable for the purpose
of psychological explanation. So where does this leave us? Are there any genuine
differences at all between linguistic thought and nonlinguistic thought?
The principal claim of this chapter is that there is an important class of thoughts
that is in principle unavailable to nonlinguistic creatures. In the first section I
briefly discuss some of the claims that have been made about how language can
function as a cognitive tool. It emerges that many of these functions do not actually
require a full-fledged language. If language is to have a distinctive, unique, and es-
sential cognitive function, so that the difference between linguistic and nonlinguis-
tic cognition is qualitative rather than merely quantitative, then we must look else-
where. In section 8.21 develop my own account of the contribution that language
can make to cognition—and explain why it is a contribution that can only be made
Language and Thinking about Thoughts 151
by language. The central claim is that all thinking that involves intentional ascent
(roughly, all thinking that involves thinking about thoughts) requires the capacity
for semantic ascent (roughly, the capacity to think about words). In the next chap-
ter I will go on to explore the practical implications of this result.
8.1 Language as a Cognitive Tool

Andy Clark (1998) has usefully plotted out six fundamental respects in which lan-
guage can function as a tool, enhancing, extending, and facilitating thought and
cognition), as follows.1
1. Memory augmentation. The acquisition of a public language offers us
powerful means of systematically storing data (not simply in written lan-
guage but also in communicated oral traditions).
2. Environmental simplification..Applyingglinguisticc labels is one way that
the perceived environment can be broken down into persisting objects and
properties.
3. Coordination. Language permits the mutual control of attention and re-
source allocation in coordinated activities.
4. Transcending path-dependent learning. The learning patterns of linguistic
creatures are not constrained by the particular cognitive paths they have
followed, since linguistic communication allows ordinary learning pat-
terns to be circumvented and short-circuited.
5. Control loops. Language allows us to create control loops for our future
behavior (by writing down plans that might be too complicated to keep in
one's head, for example) as well as to register and respond to the instruc-
tion and regulation of others.
6. Data manipulation and representation. "Extended intellectual arguments
and theses are almost always the product of brains acting in concert with
multiple external resources. These resources enable us to pursue manipula-
tions and juxtapositions of ideas and data which would quickly baffle the
un-augmented brain" (Clark 1998,173).
These are all, no doubt, important ways that language possession serves to extend
and enhance thinking. But they do not, I think, mark types of thinking that are in
principle available only to language-using creatures. As a first step in plotting out
the scope of nonlinguistic thought, it will be useful to explore how several of the
functions that Clark identifies can be carried out by nonlinguistic processes.
By "nonlinguistic process" I mean one or other of two things. On the one hand
a process can be straightforwardly nonlinguistic in virtue of not relying on any
symbols at all.2 On the other a process can be nonlinguistic even though it is de-
pendent on symbols, simply because it does not rely on the right sort of symbols.
Of course, to describe a process as nonlinguistic in this second sense is hostage
to a principled distinction between linguistic and nonlinguistic symbol systems,
and I shall have some more to say about that shortly. First, though, consider how
far one can get with the prosthetic functions of language at an entirely nonsym-
bolic level.
The mutual control of attention and resource allocation in coordinated social
activities does not require the intervention of language. Coordination requires a
degree of communication, but this communication can perfectly well be nonsym-
bolic (hence, uncontroversially, nonlinguistic). It is well known, for example, that
human infants engage from a very early age in sustained periods of coordinated ac-
tivity with their caregivers. The process has aptly been called one of affect attune-
ment (Stern 1985). It is a process of exploring and communicating emotional states
through changes in facial expression, vocalizations, and gesture.
Affect attunement is the performance of behaviors that express the quality of feeling
of a shared affect state without imitating the exact behavioral expression of the inner
state. . . . The reason attunement behaviors are so important is that true imitation
does not permit the partners to refer to the internal state. It maintains the focus of at-
tention on the forms of the external behaviors. Attunement behaviors, on the other
hand, recast the event and shift the focus of attention to what is behind the behavior,
to the quality of feeling that is being shared. (Stern 1985, 142)
In affect attunement, emotional states are communicated by being shared. The in-
fant's participation in this sort of communication is intentional (as, of course, is the
caregiver's), and the process of affect attunement is a type of coordinated activity
in which there is a mutual control of resource allocation. Nonetheless, the process
is not one of symbolic communication, because facial expressions and vocaliza-
tions are not symbols for the emotional states that are being communicated. The
link between a scowl and a feeling of disagreeableness, for example, is expressive
rather than symbolic. The behavioral manifestations of emotions and feelings can-
not be divorced from the emotions and feelings that are being manifested. That is
to say, the link between them is not arbitrary and conventional, which is the
essence of the symbolic.
A further example of coordination without symbols comes with the well-studied
dances carried out by worker honeybees (von Frisch 1967).3 Although the details
of the dances vary among the seven known species of the genus Apis, the basic
principles are similar. Worker bees returning to the nest site after successful forag-
ing land on the horizontal top of the nest and perform what has come to be known
as a waggle dance to inform their fellow workers of the location of the food source.
In the waggle dance the bee flies in a figure-of-eight pattern, moving its abdomen
back and forth laterally on the straight line in between the two circles. Although
there is of course a degree of error, the waggle dances communicate information
about the direction, distance, and desirability of the food source. Each of these
three dimensions of variation is correlated with a dimension of variation in the
dance. The angle of the dance relative to the position of the sun indicates the direc-
tion of the food source. The duration of a complete figure-of-eight circuit indicates
the distance to the food source (or rather the flying time to the food source, because
it increases when the bees would have to fly into a headwind). And the vigor
of the dance indicates the desirability of the food to be found. Variation in all
three environmental features is preserved in the relevant dimensions of the bee

dance.
The bee dance is an even clearer example than affect attunement of coordination
in resource allocation—and, as in the affect attunement case, it does not seem ap-
propriate to describe the coordination as resting on symbolic communication. The
bee dance is nonsymbolic because it does not involve arbitrary signals. To appreci-
ate the point, and in particular the distinction between arbitrary and iconic signals,
it is worth making a small digression into the theory of signals. There can be no
transfer of information without rules for the encoding and decoding of signals
(Green and Marler 1979). Viewed in broad and abstract terms, the rules for the en-
coding of a signal govern the transformations that lead from stimuli received at the
sensory surfaces to signaling behavior. The simplest such rules yield deterministic
links from input to output, as when the signaling behavior is what is sometimes
called an 'instinctive behavior pattern' or an 'innate releasing mechanism.' Newly
hatched herring gulls, for example, are particularly sensitive to the sensory input
correlated with the length, movement, and coloration of the adult herring gull's
bill. A transformation rule maps this onto the chick's characteristic signaling be-
havior, vigorous pecking at the adult's bill tip (Tinbergen 1973). In this case the
decoding signal is equally straightforward. The adult herring gull responds by
feeding the chick. More complex signaling behaviors might involve a series of
rules governing a series of transformations. In affect attunement, for example,
there will be rules governing the extraction of various qualitative and quantitative
aspects of vocal exclamations and gestures by the partner in the exchange; rules
mapping those aspects onto an amodal 'common currency'; and rules that then
generate responsive gestures or vocalizations (Stern 1985).
In an iconic signal system there is, for each dimension of variation in the rele-
vant environmental feature (the distance of the food source in the honeybee case,
for example), a single transformation rule mapping variation in that parameter onto
variation in the relevant dimension of the signal. The receiver of the signal will de-
code the signal by an inverse mapping.4 Signs are arbitrary, as opposed to iconic,
when no such general and continuous rules exist to map variations in environmen-
tal parameters onto variations in signals. Each member of the signal set is linked to
a particular environmental feature by specific assignments that can only be charac-
terized one by one. Within the framework set by the distinction between arbitrary
and iconic signals, it seems relatively uncontroversial, first, that the bee dances are
iconic rather than arbitrary signaling practices and, second, that no practice that
does not involve arbitrary signals can plausibly be described as linguistic.
The bee dances, then, clearly illustrate how coordination in resource allocation
and communal activity can be achieved at the nonlinguistic level. They also illus-
trate how control loops can be created without language and, moreover, how path-
dependent learning can be transcended. The whole point of the bee dances is that
the vast majority of worker bees do not need to seek the food for themselves. The
worker bees that remain in the hive can profit from the explorations of a small
number of bees to circumvent the standard search procedures.
Imitative learning is perhaps the most fundamental way of overcoming path-

dependence through permitting the social transmission of knowledge. Imitation is
extremely widespread in the animal kingdom and is present in human infants more
or less from birth (Meltzoff and Moore 1977, 1983). The most straightforward
form of imitative learning is simple mimicry, of the sort that can be seen in the so-
cial transmission of bird songs and in neonatal imitation of facial expressions. In
many species, individual birds deprived (e.g., by neonatal deafening or by being
raised in isolation) of the opportunity to imitate the songs of conspecifics develop
either only very rudimentary version of the standard songs or radically abnormal
songs (see, e.g., Marler 1970). At a more sophisticated level we can identify vari-
ous types of mimetic learning, of different degrees of sophistication. It is known
from studies of chimpanzee tool use, for example, that chimpanzees are skilled
at learning about the dynamic properties of objects from observing them being
manipulated by others (Boesch and Boesch 1992). There is some debate about
whether tool-using chimpanzees are actually imitating patterns of behavior
(Tomasello 1996), as opposed to picking up the dynamic affordances of objects
and learning about features of the environment, but these clearly illustrate some
form of the social transmission of knowledge.
The cognitive significance of mimetic learning within phylogeny has been rec-
ognized by Merlin Donald in his speculative reconstruction of the emergence of
the modern mind (Donald 1991). Donald suggests that a long period of mimetic
culture served as the bridge between the episodic form of cognition characteristic
of the apes and earliest hominids and the language-based culture that emerged after
the transition from the Middle Paleolithic to the Upper Paleolithic. He associates
mimetic culture with a wide range of social cognitive phenomena that collectively
served as an adaptive foundation for the evolutionary emergence of language. The
enormous development in techniques of tool manufacture characteristic of the
Middle Paleolithic period must have been linked with new means for the transmis-
sion of information across a community. From what we know of contemporary
hunter-gatherer societies, the process of tool manufacture is very much a social
process, and clearly dependent on successful communication among the partici-
pants (Reynolds 1993). If we follow the majority of archeologists in dating the
emergence of linguistic communication after the Middle Paleolithic (Bicker-
ton 1990, 1996, Liebermann 1984) then it is hard to see any alternative to group
mimetic learning as the crucial cognitive mechanism underpinning complex tool
manufacture. Mimesis can also plausibly be viewed as the foundation for the social
expression of emotion—again a form of nonlinguistic communication that circum-
vents the path-dependence of language.
Returning to Clark's list of the instrumental cognitive functions served by lan-
guage, I have already shown in some detail that at a very basic level the function of
environmental simplification does not require any sort of communication, sym-
bolic or otherwise. Human infants are born able to parse the perceived environ-
ment into object-like segments that are taken to obey certain basic higher-order
physical principles. This is probably not enough, however, for the type of environ-
mental simplification that Clark intends. His view, I suspect, is that the environ-
mental simplification that language provides applies to a perceived environment
that is already parsed into objects or object-like entities. Language does not enable
us to perceive an environment composed of discrete and continuously existing ob-
jects in the way that many philosophers have argued.5 What it allows us to do is to
impose a simplified pattern that will allow us to make sense of the discrete and
continuously existing objects that we encounter in perception and action.
Even when the function of environmental simplification is understood in this
sense, however, it still seems clear that it can be achieved at the nonlinguistic level.
It is at this point that it becomes pressing to say something about quite how
"linguistic" and "nonlinguistic" are being understood here. There has been consid-
erable discussion among philosophers, linguists, and psychologists as to what
constitutes a language. Savage-Rumbaugh has argued, for example, that language
emerges with the appearance of symbolic communication (Savage-Rumbaugh
1986). Most linguists, on the other hand, think that some form of grammar and
syntactic structure is essential for a symbol system to count as a genuine language
(Chomsky 1980). The majority of philosophers have followed the linguists in this
respect (Bennett 1976), although David Lewis (1983) has proposed in a rather
more inclusive spirit that languages are simply functions from expressions onto
meanings. Confronted with this it is hard not to be pulled in two directions. On the
one hand it is tempting to think that there is little point in trying to arbitrate in this
debate. It is unclear what criteria we should use to judge any of these proposals. If
we are trying to capture the essence of human language then it is clear where the
answer lies—but our project is broader than that. The ordinary meaning of the
word "language" is too vague to help us. After all, ordinary language allows us to
talk freely of the language of architecture and the language of the body. But, on the
other hand, it seems impossible to investigate the nature and scope of nonlinguistic
thought without a demarcation line between the linguistic and the nonlinguistic.
I will adopt a middle way between these two extremes. A genuine language must
allow for the formation of complex symbols from simple symbols. The essence of
language is the combination of symbols with each other to express thoughts, taking
thoughts to be complex entities that can be assessed for truth or falsity. The possi-
bility of truth appears only with complex symbols. It emerges only when a state of
affairs is asserted to be the case, and this requires, at a bare minimum, the combi-
nation of a naming symbol and a predicate symbol. A lexicon of simple symbols
will not support communicative utterances that are assessable for truth or falsity
(although they may, of course, be adequate to perform speech acts such as issuing
injunctions). So complex symbolic systems allow the formation of complex sym-
bols, symbols whose meaning is determined by the meaning of the symbols of
which they are composed. That is to say, complex symbol systems display what is
often known as compositionality, the possibility of recombining the members of a
finite number of simple symbols to produce a range of complex symbols. This
characteristic of complex symbol systems is frequently stressed in both philo-
sophical and psychological discussions of language. It tends to be taken as a uni-
tary phenomenon, but compositionality can be displayed in two fundamentally dif-

ferent ways, corresponding to two fundamentally different types of complex
symbol systems.
In the first category, the category of sequentially complex symbol systems, com-
plex symbols are formed in a purely additive or successive manner. Correlatively,
understanding a sequentially complex symbol is a matter of successively under-
standing the simple symbols that make it up. A good example of such a sequen-
tially complex symbol system comes with the communication systems sponta-
neously developed by very young children bom deaf but whose parents do not use
sign language to them. Such children tend without instruction to develop elemen-
tary signing systems employing combinations of two signs (Goldin-Meadow
1979). The two-sign combinations typically involve a pointing gesture combined
with an action sign. An example might be pointing at a door combined with a turn-
ing gesture to signify that it should be opened. Pygmy chimpanzees, or bonobos
(Pan paniscus), have acquired similar abilities (Greenfield and Savage-Rumbaugh
1990). It is clear that we are dealing here with the formation of complex symbols.
In the given example, it is the formation of a complex symbol communicating an
injunction—the injunction to open the door. But the mode of formation is ex-
tremely simple. The syntactic rules of the symbol system are such that the action
sign (the turning gesture) can only be conjoined with a sign that picks out an ob-
ject. There is no way of using the syntactical rules to build up further complex
symbols.
In what might be termed a hierarchically complex symbol system, on the other
hand, the logical and semantic relations between the components of a complex
symbol cannot in every case be read off from the sequential ordering of simple
symbols. The basic feature of a hierarchically complex symbolic system is that it
possesses a hierarchically organized compositional semantics operating in such a
way that understanding a complex symbol will not always be a matter of succes-
sively understanding the simple symbols that make it up. To get from a sequen-
tially complex to a hierarchically complex symbol system we must lift the basic re-
striction that complex symbols can only be formed by combining a predicate
symbol and the appropriate number of naming symbols. Hierarchically complex
symbol systems possess a variety of mechanisms that allow the formation of com-
plex symbols. Operators analogous to the definite and indefinite articles in English
will allow the formation of complex names (i.e., definite and Indefinite descrip-
tions), while reiterable operators that apply to complete sentences will allow the
formation of complex symbols compounded from sentences. The logical constants
are a case in point, but so too are the operators indicating possibility and necessity
and indeed tense. I will have more to say about these in the next chapter (particu-
larly in sections 9.4-9.7).
Returning to the matter at hand, it seems plausible that the function of environ-
mental simplification identified by Clark can be carried out by "labeling" environ-
mental features with a simple symbol system. Environmental simplification does
not require complex symbols (although of course it would be greatly enhanced by
a complex symbol system) and hence does not require a language. A classic exam-
ple of how this might work (and one that provides a further example of how the in-
strumental functions I have already considered can be effected at the nonlinguistic
level) comes with the alarm calls of vervet monkeys (Cheney and Seyfarth 1990).
Vervet monkeys have three alarm calls, each geared to a different category of
predator—eagle, leopard, and snake. When vervets hear a particular alarm call
they do not display an indiscriminate fear or avoidance response, but rather behave
in ways appropriate to the predator to whose presence they have been alerted.
When they hear the eagle alarm call, for example, they look up and scan the sky.
Vervets also have two further calls, the "wrr" and "chutter" calls, which communi-
cate the nearby presence of strange groups of vervets. According to Cheney and
Seyfarth, vervet monkeys are more likely to issue alarm calls when there are close
kin in the vicinity (1990, ch. 5). Something similar holds for the food signals is-
sued by macaque monkeys, who seem to scan for members of their social group
before issuing a food signal (Hauser and Marler 1993). Even if we assume that the
vervet alarm calls are symbolic, they do not constitute a complex symbol system,
since there is no scope for the formation of complex symbols. Yet it is clear that the
alarm calls are serving a function of environmental simplification, and indeed per-
mit the creation of control loops for regulating each individual's future behavior
and the behavior of other members of the group.
Nor is it only in primates that we find such behaviors. A classic ethological ex-
ample of environmental simplification and control loop creation comes with the
ways that different species of birds hide caches of seeds at specific locations.
Clark's nutcrackers are known to deploy multiple cues to identify the locations of
nut caches. Experiments have shown that they are able to reidentify nut stores even
when experimenters manipulate the most obvious landmarks (Balda and Turek
1984). In creating food stores, birds both simplify the environment in terms of
markers that allow them to reidentify the caches and create control loops for future
behavior in terms of a structure of intercache trajectories that will determine their
movements when retrieving food.
As far as Clark's original list of six instrumental functions is concerned, there-
fore, it looks as if we can make sense of at least four of them operating at the non-
linguistic level. And the two that are left (memory augmentation and data manipu-
lation) do not seem very likely candidates for marking a distinctive type of
cognition only made available by the acquisition of language. There are all sorts of
ways that nonlinguistic creatures can augment their limited memory capacities,
some of which I have already considered in thinking about how path-dependent
learning might be transcended at the nonlinguistic level. And data manipulation is
a characteristic of all thought. The only differences between linguistic and nonlin-
guistic thought in either of these dimensions are likely to be differences of degree.
It would seem, therefore, that if there indeed are any fundamental differences of
type between linguistic and nonlinguistic cognition, they will lie elsewhere. In the
next section I will turn to a fundamentally different type of argument for the dis-
tinctiveness of linguistic cognition.
8.2 Intentional Ascent and Semantic Ascent

In the same article in which he puts forward the six instrumental functions of lan-
guage that I considered in section 7.1, Andy Clark offers the following conjecture
about the role of language in underpinning the distinctiveness of human cognition.
Perhaps it is public language that is responsible for a complex of rather distinctive
features of human thought—viz., our ability to display second-order cognitive dy-
namics. By second-order cognitive dynamics I mean a cluster of powerful capacities
involving self-evaluation, self-criticism and finely honed remedial responses. Exam-
ples would include: recognizing a flaw in our own plan or argument, and dedicating
further cognitive efforts to fixing it; reflecting on the unreliability of our own initial
judgements in certain types of situation and proceeding with special caution as a re-
sult; coming to see why we reached a particular conclusion by appreciating the logi-
cal transitions in our own thought; thinking about the conditions under which we
think best and trying to bring them about. The list could be continued, but the pattern
should be clear. In all these cases we are effectively thinking about our own cognitive
profiles or about specific thoughts. (Clark 1996,177)
He explains how language makes these types of thought available in the following
terms.
It is easy to see in broad outline how this might come about. For as soon as we for-
mulate a thought in words (or on paper), it becomes an object both for ourselves and
for others. As an object it is the kind of thing we can have thoughts about. In creating
the object we need have no thoughts about thoughts—but once it is there, the oppor-
tunity immediately exists to attend to it as an object in its own right. The process of
linguistic formulation thus creates the stable structure to which subsequent thinkings
attach. (Clark 1996, 177)
It seems to me that the basic idea Clark is putting forward is essentially correct.
My aim in this section of the chapter will be to formulate the basic idea more pre-
cisely and to provide an argument to back it up.
Clark's explanation of the role language plays in second-order cognitive dynam-
ics is incomplete, at least for my purposes. For one thing, all he really offers is an
account of how, given that we have language, we are able to engage in second-
order cognitive dynamics—whereas what we need is an argument that second-
order cognitive dynamics can only be undertaken by language-using creatures.
This links up directly with a more fundamental worry. The natural way to derive an
argument for the necessity of language from Clark's suggestions would be to claim
that language is required for thinking about our own thoughts. But this claim is
hardly uncontroversial. It would be instantly denied, for example, by language of
thought theorists. It is an integral part of the language of thought hypothesis that
reflexive thinking is available in the language of thought. Once again we find our-
selves caught in the familiar grey area between public languages and the language
of thought. We need to investigate not simply whether some form of language is
required but what type of language that must be.
The basic premise from which we need to start here is that reflexive thoughts can
only be possible if the target thoughts have vehicles that allow them to be the objects
of further thoughts. The question, then, is what form these vehicles must take.
Broadly speaking, the candidates fall into two broad categories. They might be ei-
ther personal-level vehicles or subpersonal vehicles. Any argument to show that re-
flexive thinking requires the target thoughts to be vehicled in a public language will
have to establish, first, that the vehicles must be at the personal level and, second,
that the only available vehicles at the personal level are public language sentences.
The plausibility of the first step in the argument emerges when we reflect that re-
flexive thinking will paradigmatically involve a direct and conscious cognitive ac-
cess to the target thoughts. As the first passage quoted from Clark at the beginning
of this section makes clear, typical examples of reflexive thinking might include
evaluating evidential and inferential relations between thoughts. Such reflexive
thinking involves having the target thoughts in mind—entertaining them con-
sciously and considering how they relate to each other logically and evidentially.
Yet it seems clear that we do not have the appropriate sort of direct and conscious
cognitive access to subpersonal states. Second-order cognitive dynamics is a mat-
ter of the conscious regulation and policing of one's own thoughts-and we do not
consciously regulate and police sentences in the language of thought. It might well
be the case that certain types of hypothesis testing and refinement do take place at
the subpersonal level. Something like this happens, according to Fodor, when we
learn a language. Nothing I say is incompatible with that proposal, since my claim
is simply that such processes would not count as instances of second-order cogni-
tive dynamics. Nonetheless, one might want to question the proposal on other
grounds, such as the availability of alternative models of language acquisition, par-
ticularly those developed within the connectionist tradition (for a survey of recent
work see McLeod et al. 1998).
It is hard to see how a defender of the language of thought hypothesis could
maintain that we do have conscious access to sentences in the language of thought,
given that the language of thought hypothesis is a hypothesis about subpersonal
cognitive architecture, not about the medium of conscious thought. It is generally
accepted that subpersonal states are inferentially insulated from the conscious
processes of cognitive evaluation and self-criticism. This is precisely the distinc-
tion between the personal and subpersonal levels (see, for example, Stich 1978). A
fortiori, therefore, it follows that the vehicles of the thoughts that are the objects of
what Clark calls second-order cognitive dynamics cannot be sentences in a subper-
sonal language of thought.6
There are versions of the language of thought hypothesis, however, that are not
put forward purely as hypotheses about subpersonal cognitive architecture and
hence are not directly blocked by this line of argument. Some authors have sug-
gested that there might be a modular internal language in which thoughts are
encoded for conscious consideration. This general approach would sit well with
(although neither entails nor is entailed by) higher-order thought theories of con-
sciousness (Rosenthal 1991). It seems to me, however, that this proposal lacks
phenomenological plausibility. All the prepositional thoughts that we consciously
introspect, whether thoughts that come unbidden into our minds or thoughts that
we consciously instigate and consider in solving a problem, take the form of sen-
tences in a public language. We do, of course, engage in various types of nonsen-
tential thinking—what, in chapter 3,1 termed thinking-how in contrast to thinking-
that—and it is only mistaking nonpropositional thinking-how for propositional
thinking-that that leads to the view that we can be aware of "pure" thoughts with-
out any sentences featuring among the contents of introspection.7 There are certain
types of problem that we solve by manipulating mental images and exercising the
visual imagination. And we are, of course, conscious of bodily sensations, emo-
tional feelings, and other such qualitative states (although these are not properly
described as types of thinking at all). But we are not, I think, ever conscious of
propositional thoughts that do not have linguistic vehicles. When we are conscious
of propositional thoughts we are conscious of imaged sentences.8 What we intro-
spect when we introspect our propositional thoughts in the manner required for the
processes of second-order cognitive dynamics is inner speech.
This brings me to the second step of the argument. What needs to be shown is
that public language sentences are the only possible personal-level vehicles for
thoughts that are to be the objects of reflexive thinking. The conclusion so far is
that sentences in the language of thought are not appropriate vehicles. The only
way to proceed is by elimination—showing that there are no viable alternatives
to public language sentences. How else might the representation relation work?
There seem to be two possibilities. On the one hand representation might be se-
cured symbolically through the complex symbols of a natural language (complex
symbols being required since what are being represented are thoughts about states
of affairs). A thought would be represented, therefore, through its linguistic expres-
sion and would appear as a potential object of thought qua linguistic entity. On the
other hand, representation might be secured in an analogue manner, through some
kind of pictorial model. On this conception of the vehicles of thought, which we
find developed in different ways in mental models theory in the psychology of rea-
soning (originally proposed in Craik 1943 but most comprehensively developed
in Johnson-Laird 1983) and in the conception of mental maps put forward by
Braddon-Mitchell and Jackson (1996), the vehicle of a thought is a pictorial repre-
sentation of the state of affairs being thought about.9
The idea of structural isomorphism is at the heart of both theories. Both mental
models and maps are spatially isomorphic with what they represent. The relations
(or at least some of them) holding between elements of the mental model/map can
be mapped on to the relations holding between objects in the represented state of
affairs. In this way representation is secured through the relations of exemplifica-
tion and resemblance. The mental model/map represents a state of affairs by exem-
plifying the structure of that state of affairs—that is to say, by itself possessing a
structure that resembles (at some suitable level of abstraction) the structure of the
represented state of affairs. It does so, however, in a way that does not have an in-
dependently identifiable structure corresponding to the state of affairs exemplified.
Braddon-Mitchell and Jackson put the point clearly.
There is no natural way of dividing a map at its truth-assessable representational

joints. Each part of a map contributes to the representational content of the whole
map, in the sense that had that part of the map been different, the representational
content of the whole would have been different. Change the bit of the map of the
United States between New York and Boston, and you change systematically what
the map says. This is part of what makes it true that the map is structured. However,
there is no preferred way of dividing the map into basic representational units. There
are many jigsaw puzzles you might make out of the map, but no single one would
have a claim to have pieces that were all and only the most basic units. (Braddon-
Mitchell and Jackson 1996, 171)
We need, therefore, to distinguish weak and strong senses in which a representa-

tional vehicle might be structured. In the weak sense there is structure whenever a
structural isomorphism can be identified between the vehicle and what it repre-
sents. In the strong sense, however, structure requires the existence of basic repre-
sentational units combined according to independently identifiable combinatorial
rules. Natural language sentences (or for that matter sentences in the language of
thought) are clearly structured in the strong sense, whereas mental maps/models
only possess structure in the weak sense.
In mental models and maps, the representation relation will be secured through
some combination of isomorphic resemblance and exemplification holding prima-
rily between the model/map as a whole and the represented state of affairs as a
whole (and only derivatively at the level of the distinguishable elements of the
map/model and state of affairs, respectively). There are important questions to ask
about the nature, and even the possibility, of resemblance and exemplification-
based representation—questions that go back to early criticisms of the British Em-
piricists (and that have arisen more recently in criticisms of the picture theory of
the proposition offered by Ludwig Wittgenstein in the Tractatus Logico-Philo-
sophicus). I shall prescind from these, however. There is a fundamental problem
with either version of the analogue proposal as an account of how it is possible
for thoughts to have vehicles that allow them to become the objects of further
thoughts.
Second-order cognitive dynamics involves sensitivity to the inferential relations
between thoughts, and we do not yet have an understanding of how images can be
inferentially connected to each other. The problem once again derives from the in-
timate relation between inference and structure explored in earlier chapters. There
is a sense in which mental models and maps are structured, since they contain ele-
ments that can feature in further mental models/maps. Nonetheless, they do not
seem to be structured in the right sort of way to permit the reflexive type of
second-order cognitive dynamics under discussion. It will be helpful to take maps
and mental models separately.
A canonical example of second-order cognitive dynamics might be coming to
see the evidential basis for a particular belief and then evaluating the inferential
transition made on that basis. (This would be an example of what in section 9.1 I
will term reflective doxastic modification.) An example might be the realization
that one has made an overly rash inductive generalization, or a faulty deductive in-
ference. It is perfectly easy to see how there could be some very basic forms of in-
ferential transition between maps. Such transitions might be modeled on broadly
associationist lines, and it is the possibility of such transitions that enables maps to
serve as guides to action. What is not possible, however, is for such transitions to
be understood and evaluated in terms of either deductive validity or probabilistic
support. Those very features of maps (their analogue nature and structural isomor-
phism with what they represent) that make them so useful for guiding action serve
to make them inappropriate for the type of inferential evaluation characteristic of
second-order cognitive dynamics. In order for such evaluation to take place, the
maps must be interpreted in broadly prepositional terms. We must interpret one
map as expressing one proposition and the second as representing a further propo-
sition, and then evaluate the inferential relations (be they deductive, inductive, or
probabilistic) between those two propositions. Once again, our only understanding
of how to do this rests on the two propositions being linguistically formulated.
Braddon-Mitchell and Jackson do not directly address this issue, but they
do offer the following explanation of how maps can evolve over time in what is
clearly intended to be an analogy with inferential transitions between linguistically
vehicled representations.
Maps are physical entities whose structure can govern the way they evolve over time.
When cartographers update maps or put two maps together to make one that incorpo-
rates all the information in a single map, these operations are governed in part by the
structures of the maps they are working on. And in order to find a target, rockets use a
kind of internal map that gets continually updated as new information comes in. In
these rockets, later maps are causal products of earlier maps plus what comes in via
the rocket's sensors. Hence map theorists can tell an essentially similar story to lan-
guage of thought theorists about how thoughts evolve over time as a function of their
prepositional objects. (Braddon-Mitchell and Jackson 1996, 173)
There is a fundamental disanalogy here, however. The issue is not really about
how thoughts evolve over time. In a very important sense individual thoughts quite
simply do not evolve over time. It is systems of thought that evolve, and they do
so as a function of the inferential relations between the thoughts that compose
them. These inferential relations hold between distinct thoughts, and nothing that
Braddon-Mitchell and Jackson say in this short passage gives us any way of under-
standing how we should understand inferential relations between distinct thoughts
at the level of mental maps. The process of combining maps has only very limited
analogies with the process of inferring one thought from another. We do not have,
for example, any idea what a conditional map might look like—and consequently
little understanding of how conditional reasoning might take place at the level of
mental maps. A fortiori, therefore, there seems no sense in which we can under-
stand second-order cognitive dynamics as applying to the inferential transitions
between mental maps.
The situation is somewhat more complicated with mental models theory. How
can mental models not be suitable targets for the type of reflexive thinking charac-
teristic of second-order cognitive dynamics, given that mental models are explic-
itly proposed as providing a unified account of deductive, probabilistic, and modal
reasoning (Johnson-Laird 1999)? It is important to be clear, however, about the
precise claims of mental models theory. Mental models theory is proposed as an al-
ternative to the so-called mental logic theory (Rips 1994), according to which rea-
soning is an exclusively syntactic matter grounded in formal rules of inference.10
The key idea of mental models theory is that arguments are evaluated by the con-
struction of mental models of the relevant premises. An argument is judged to be
deductively valid if the conclusion holds in all the constructed models of the prem-
ises, probabilistically valid if it holds in most of the models, and so forth. The
claim made is that this model of reasoning provides a better explanation of the pat-
terns discovered in the experimental study of how subjects reason than the idea
that reasoning involves the manipulation of sentential representations according to
formal rules. Some of these patterns involve systematic susceptibility to formal
fallacies, which it is obviously difficult to explain on the mental logic approach.
Others are simply patterns in the time taken to carry out certain inferences, which
advocates of the mental models approach claim is directly correlated with the
number of models that the reasoner needs to construct (Johnson-Laird and Byrne
1991).
Mental models are indeed supposed to be structurally isomorphic to the states of
affairs described in the premises. But like mental maps, their structure is deriva-
tive. It is derived from the premises that they are modeling. The models are con-
structed from constituents and properties that feature in the premise being mod-
eled. And those premises are of course linguistic entities. It would be a mistake to
think that mental models theory construes inference in terms of transitions be-
tween mental models—any more than a model-theoretic approach to the sentence
calculus construes inference in terms of transitions between truth tables. Mental
models theory construes inference as a matter of transitions between sententially
encoded propositions. What is distinctive about it is that it construes those transi-
tions between sententially encoded propositions as taking place in virtue of rela-
tions between analogue representations of the states of affairs portrayed in those
sententially encoded propositions (as opposed to formal relations holding between
the syntactic structures of the relevant sentences). Mental models theory is not a
genuine alternative to the sentential conception of reasoning, since the whole idea
of a mental model only makes sense within the framework of the sentential con-
ception. Mental models theory offers a particular way of developing the sentential
conception, not of supplanting it. Mental models are not the vehicles of inference,
but rather, as their name suggests, models of those inferences.
By a process of elimination, therefore, we have reached the conclusion that
thoughts can only be the objects of the type of reflexive thinking in which thoughts
are the objects of thought if they have natural language vehicles. This is not, of
course, to say that we cannot deploy mental maps and mental models. It is clear
that we do, and it is highly likely that nonlinguistic creatures do as well. Nor is it to
say that mental models and mental maps cannot be the objects of thought. It seems
clear that we can think about mental models and mental maps as well as think by
means of them. The point is that we cannot use mental maps or mental models for
thinking about thoughts in the manner demanded by second-order cognitive dy-
namics. Natural language sentences are the only proxies that will permit thoughts
to function as the objects of thought in this manner. To put the matter in the form of
a slogan, there can be no intentional ascent without semantic ascent. We think
about thoughts through thinking about the sentences through which those thoughts
might be expressed. The significance of this thesis depends, however, on the types
of thinking that constitutively involve intentional, and hence semantic, ascent. We
began this section with Clark's programmatic suggestion that only language-users
can be capable of the types of cognitive self-criticism and self-monitoring that he
terms second-order cognitive dynamics. In the next chapter I will work toward a
more wide-ranging classification of the scope and limits of nonlinguistic thought.
9
The Limits of Thinking without Words
In the previous chapter I showed that a certain type of thinking is unavailable at the
nonlinguistic level. Forms of thinking that involve thinking about thought (hence
taking particular thoughts as the objects of thought) are only available to creatures
participating in a public language. Thoughts can only be the objects of further
thoughts if they have suitable vehicles, and the only suitable vehicles are public
language sentences. Intentional ascent requires semantic ascent. In this chapter I
consider the practical implications this has for the scope and limits of nonlinguistic
thought. I am concerned in particular with the following two questions. What types
of thinking are in principle unavailable to nonlinguistic creatures? What sort of
primitive precursors might there be at the nonlinguistic level for types of thinking
that involve intentional ascent in their full-fledged form?
It will turn out that intentional ascent is a broader category than immediately ap-
pears. There are two types of intentional ascent, which might be termed explicit
and implicit intentional ascent respectively. The first type includes all those forms
of thought that involve metarepresentation, or the ability to have thoughts that take
further thoughts as their direct objects. I consider examples of metarepresenta-
tional thinking in the first three sections of this chapter. In section 9.11 explain the
distinctive type of belief revision available only to language-using creatures, con-
trasting it with the more primitive ways that nonlinguistic creatures can modify
their beliefs and behavior in the light of changing evidence, whether supporting or
countervailing. Section 9.2 discusses how the possibility of a certain type of
higher-order desire involves intentional ascent (the taking of a cognitive attitude
toward a particular desire) and hence depends on language. In section 9.3 I discuss
the relation between language possession and what is often called theory of mind. I
argue that the attribution of beliefs, certain types of desires (what I earlier called
situation-desires), and propositional attitudes in general is only available to lin-
guistic creatures—again because it involves thinking about thinking. A primitive
type of psychological explanation, involving the attribution of goal-desires and

perceptual states, is nonetheless possible at the nonlinguistic level.
Metarepresentational thinking, however, is not the only form of thinking that in-
volves intentional ascent. Intentional ascent can also occur implicitly. Implicit in-
tentional ascent does not involve taking thoughts as the objects of further thoughts.
Rather, it involves embedding thoughts to create more complex thoughts whose
contents are states of the world, rather than one's own mental states (as when one
engages in second-order cognitive dynamics) or someone else's mental states (as
when one is attributing beliefs and desires to others). In section 9.4 I explain in
the abstract how and why embedding requires intentional ascent with particular
reference to the truth-functional prepositional operators and to the formation of
quantified thoughts. I go on to suggest that tense and modal operators may best be
viewed as analogous to adverbial modifiers, qualifying complete thoughts, with
the result that tense-based thinking involves a form of intentional ascent. Section
9.5 explores the different ways of thinking about generality available at the lin-
guistic and nonlinguistic levels and shows how a limited type of general thought
is available at the nonlinguistic level. In section 9.6 I discuss the role that the
acquisition of the relative pronoun might play in making possible the move from
domain-specific cognition to domain-general cognition.
9.1 Belief Revision and Second-Order Cognitive Dynamics

The distinctiveness of linguistic cognition was introduced in the previous section
through Clark's notion of second-order cognitive dynamics. Here again is how
Clark characterizes this notion:
By second-order cognitive dynamics I mean a cluster of powerful capacities involv-

ing self-evaluation, self-criticism and finely honed remedial responses. Examples
would include: recognizing a flaw in our own plan or argument, and dedicating fur-
ther cognitive efforts to fixing it; reflecting on the unreliability of our own initial
judgements in certain types of situation and proceeding with special caution as a re-
sult; coming to see why we reached a particular conclusion by appreciating the logi-
cal transitions in our own thought; thinking about the conditions under which we
think best and trying to bring them about. (1999, 177)
In this section I delineate more precisely the scope of the distinctive type of reflec-
tion that is made available by the presence of language. It turns out that this dis-
tinctive type of reflection has two broad components—the availability of a distinc-
tively second-order variety of belief revision and the availability of second-order
desires.
In order to appreciate what is really distinctive about linguistic cognition, two
distinctions are needed. The first concerns the relation between behavior and its ef-
fects. Many types of behavior (indeed, plausibly, all types of behavior that are not
tropistic, generated by innate releasing mechanisms, or classically conditioned) are
driven by what might broadly be termed outcome-sensitivity. Animals and infants
The Limits 167
will modify all but the simplest behaviors according to their outcomes. This is the
essence of trial-and-error learning, as well as being an indispensable adaptive trait.
But these are two different types of outcome-sensitivity. We can term them practi-
cal outcome-sensitivity and doxastic outcome-sensitivity, respectively. A creature
is practically outcome-sensitive when it modifies its behavior directly as a function
of exposure to evidence of that behavior's success or failure. A creature is doxasti-
cally outcome-sensitive, on the other hand, when it modifies its behavior as a func-
tion of modifications in its beliefs, having modified its beliefs in response to evi-
dence of the truth or falsity of those beliefs.
Practical outcome-sensitivity can be illustrated through the distinction between
classical conditioning and instrumental conditioning. In classical conditioning, a
neutral stimulus (e.g., the sound of a bell) is followed by an unconditioned stimu-
lus (e.g., the presentation of food) that elicits a reaction (e.g., salivation). The out-
come of classical conditioning is that the conditioned response (the salivation)
comes to be given to the conditioned stimulus (the sound of the bell) in the absence
of the unconditioned stimulus. In instrumental or operant conditioning, the presen-
tation of the reinforcing stimulus is contingent on the animal making a particular
behavioral response (such as pecking a lever). If the behavioral response does not
occur, the reinforcing stimulus is withheld. Classically conditioned behavior is not
outcome-sensitive in any interesting sense, since it is not the behavior that is rein-
forced. The conditioning is a function of the strength of the association between
the conditioned response (e.g., the sound of the bell) and the unconditioned stimu-
lus (e.g., the presentation of food). The process of conditioning is not determined
by anything the animal actually does.
Instrumental conditioning, on the other hand, is clearly outcome-sensitive. It is
the behavior itself that is reinforced. Nonetheless, most varieties of instrumental
conditioning are not doxastically outcome-sensitive. The category of instrumen-
tally conditioned behavior is not fully homogenous, and, although a case can be
made for regarding certain types of instrumental conditioning as resting on instru-
mental beliefs (see, e.g., Dickinson and Balleine 1993, Heyes and Dickinson 1993,
and the experiments by Rescorla and Skucy, and Hammond discussed hereafter),
the vast majority of instrumentally conditioned behaviors are not best interpreted
in terms of beliefs about the contingency between behavior and outcome.
Clear examples of doxastic outcome-sensitivity come with behaviors exempli-
fying what in chapter 61 termed level 2 rationality. Level 2 rationality involves in-
strumental beliefs about the contingencies between courses of action and outcomes
and no creature can properly be credited with such instrumental beliefs unless
those beliefs are evidentially sensitive to the contingency in question. We consid-
ered a clear example of this sensitivity in the lever-pressing experiments. Rats
trained to press a lever to obtain food will cease lever-pressing when the schedule
is changed so that the food appears irrespective of whether the rat presses the lever
or not. They are sensitive to the evidential connection between the action of lever
pressing and the appearance of the food and modify their behavior accordingly
(Rescorla and Skucy 1969, Hammond 1980). Creatures (such as Hershbergers's
chicks who persist in moving toward a food-source that retreats from them at twice
the rate at which they approach it) clearly lack this sensitivity (see section 6.4
above). Similarly, a direct sensitivity to the outcomes of action is an integral part
of tool construction and use. One cannot devise tools without being sensitive to
their aptness or not for the job for which they are being devised. Supporting and/or
countervailing evidence needs to be taken into account in modifying the tools.
As the example of level 2 rationality shows, doxastic outcome-sensitivity is not
simply a matter of behavior modification resulting from changes in beliefs. If, for
example, a creature stops stalking one kind of prey because it has noticed an easier
quarry within closer reach then its behavior has been modified as a consequence
of changes in its (perceptual) beliefs. Nonetheless, it has not displayed doxastic
outcome-sensitivity because the beliefs in question are not beliefs about the out-
come of the behavior and the means by which that outcome is to be achieved. Nor
are the relevant changes in belief the result of changes in the balance of evidence
for the original belief (as they would be if, for example, the creature noticed that
what it had taken for one kind of prey was really another). Bearing this in mind,
many forms of behavior exemplifying level 1 rationality are going to come out as
practically but not doxastically outcome-sensitive. I showed in chapter 6 that a
wedge needs to be driven between the normative and descriptive dimensions of
our theories of practical decision-making as applied to behaviors falling within the
general category of level 1 rationality. We as observers can model level 1 choice
behavior in terms of some normative theory such as expected utility theory, which
has a manifestly instrumental dimension (given that the calculation of expected
utility rests on the relevant utilities and probabilities of the different possible out-
comes consequent on any given behavioral option). But this does not mean that the
practical decision-making resulting in the behavior itself involved any instrumen-
tal reasoning—nor, a fortiori, any degree of doxastic outcome-sensitivity. A clear
example of this emerged in the red deer roaring contexts. The louder a red deer
stag roars, the stronger it is—since the same thoracic muscles are employed in
roaring and fighting. To the extent that a deer acts because it recognizes that it is
roaring louder than its antagonist, it will be acting in a way that can be analyzed in-
strumentally as maximizing its expected utility. But it is not acting the way it does
because it has engaged in any consequence-sensitive reasoning. Therefore, even if
during the roaring contest the stag discovers that its antagonist can in fact roar
louder than initially appeared and consequently withdraws from the roaring con-
test, this would still not count as doxastic outcome-sensitivity.
It is clear, therefore, that doxastic outcome-sensitivity, although relatively cir-
cumscribed within the nonlinguistic realm, is not the preserve of language-using
creatures. To see what is epistemically distinctive about the realm of the linguistic,
we need to make a further distinction between two different types of doxastic
outcome-sensitivity. Doxastic outcome-sensitivity involves, as I have shown, be-
havioral modification as a consequence of doxastic modification—and not simply
doxastic modification per se but rather the modification of beliefs about the instru-
mental connection between a particular course of action and an intended outcome.
The Limits 169
There are two different ways that the relevant doxastic modification can take place.
The examples I have been considering up to now are of what might be termed di-
rect doxastic modification. That is to say, beliefs are modified in direct response to
changes in the structure and nature of available evidence. When it becomes appar-
ent to the rat, for example, that the food reward will appear irrespective of whether
or not it presses the lever, this new piece of evidence has a direct effect on the rat's
belief about the dependence of food on lever-pressing, and that original belief is
accordingly modified. A similar process might occur during the process of tool
construction—when the stone fractures one way rather than another, the experi-
enced stone-knapper might revise his instrumental beliefs about the appropriate
way to detach flakes to construct a hand-axe. What is characteristic of both these
examples is that perceptually registered countervailing evidence impacts immedi-
ately on the belief in question. The instrumental belief controlling the behavior is
instantly modified in response to the perceived discrepancy between actual out-
come and envisaged outcome.
In direct doxastic outcome-sensitivity, therefore, there is no need for the rat or
the tool-maker to reflect on the relations of evidential support holding between
what they perceive and what they believe. Such reflective changes of belief are, in
contrast, characteristic of the second type of doxastic modification (which we can
term reflective doxastic modification). The distinction can be put in terms of the
different norms governing each type of belief revision. Direct doxastic modifica-
tion can be seen as a process of maintaining and/or restoring coherence within the
belief system (and, of course, between the belief system and the deliverances of
perception). What matters is that the belief system should be coherent and inte-
grated enough for the creature in question to preserve a unified epistemic perspec-
tive on the world and to be able to act accordingly. There should be no tension
(more realistically: as little tension as possible) between the evidence that is avail-
able to a creature and the beliefs that that evidence supports. When there is tension
between evidence and beliefs, modifications are made to restore coherence.
Reflective doxastic modification, on the other hand, is governed by the norm of
truth rather than the norm of coherence. Coherence is a desideratum but it is not
the sole desideratum. What matters above all is that the evidence should be such as
makes the beliefs that it supports likely to be true. Reflective doxastic modification
concerns itself explicitly with the logical and probabilistic relations between evi-
dence and beliefs (as well, of course, as between the individual beliefs within a
belief system). At the reflective level a belief might be rejected or modified in the
absence of countervailing evidence or tension with existing beliefs—it might, for
example, be rejected simply because the believer recognizes that it is not war-
ranted by the evidence. Although the notions of warrant and justification can be ap-
plied in an attenuated sense at the level of direct belief revision (most prominently
in a broadly externalist sense according to which, roughly speaking, a set of beliefs
is warranted to the extent that it is produced by reliable mechanisms and modified
according to principles that tend to preserve truth and eliminate error), there is no
sense in which the extent to which their beliefs are warranted or justified can be an
issue for creatures operating solely at that level. That is to say, internalist notions
of warrant and justification can get a grip only at the reflective level, because these
notions can be applied only to thinkers capable of explicit reflection on the rela-
tions between thoughts and perceptions and between thoughts and other thoughts.1
The link between reflective doxastic modification and language should be clear,
given the discussion in the previous chapter. It is only possible to evaluate and re-
flect on the extent to which one belief implies another (more precisely: the extent
to which acceptance of one belief commits one to acceptance of another), or the
extent to which a belief is supported by a particular type of evidence, if one is able
explicitly to hold those beliefs in mind. And, as I discussed at some length earlier,
thoughts can only be the objects of further thoughts in the way that this requires
if they have linguistic vehicles. It is clear, moreover, that reflective belief revi-
sion will involve explicit consideration of the formal logical relations between
thoughts—and we have no understanding of the logical relations between thoughts
except when those thoughts have linguistic vehicles.
The discussion so far has concentrated on processes of belief revision. But once
the mechanisms of reflective belief revision are in place, broader possibilities
emerge for reflective self-monitoring and epistemic self-criticism. Reflective sub-
jects can evaluate beliefs without changing them. They can distinguish between
those beliefs that are candidates for revision and those that are not candidates for
revision. They can plot out the evidential connections between the different com-
ponents of their belief systems, working out what rests on what and which beliefs
are bearing the greatest weight. Reflective subjects can also identify and learn from
their past mistakes, settling on epistemic policies that are designed to avoid those
mistakes. The notion of second-order cognitive dynamics that Clark originally in-
troduced includes all these high-level forms of epistemic self-monitoring. They are
all rooted, however, in the distinctive type of belief revision that I have termed re-
flective doxastic modification—itself a particular type of outcome-sensitivity. The
process of emancipation from the doxastic here-and-now is a gradual one.
9.2 Higher-Order Desires

Nonlinguistic creatures are incapable of intentional ascent. I have shown the role
that intentional ascent plays in certain types of belief revision and, more broadly,
in epistemic self-monitoring. Are there comparable phenomena in the realm of de-
sire? It is useful at this point to advert to Frankfurt's influential distinction (1971)
between first-order and second-order desires. In the terminology I have been
adopting, first-order desires can be either goal-desires or situation-desires. They
are first-order because they are directed at objects and/or properties (in the case of
goal-desires) or states of affairs (in the case of situation-desires). Second-order de-
sires, in contrast, are situation-desires that take first-order desires as their objects.
A second-order desire is a desire to the effect that one should have a particular
first-order desire.2
When we consider the structure of second-order desires, it becomes clear why
The Limits 171
an important class of second-order desires should be language-dependent. The

content (that is to say, the satisfaction-condition) of a second-order desire is that
one should have a particular first-order desire. The content of the first-order desire
is embedded within the content of the second-order desire. It is, so to speak, the
target of the second-order desire. But then it looks very much as if second-order
desires involve intentional ascent and therefore, by the argument of section 8.2, the
capacity for semantic ascent. Still, some qualification is required. In particular we
need to distinguish cases in which the target first-order desire is a goal-desire from
cases in which it is a situation-desire. The satisfaction-condition for a second-order
desire embedding a goal-desire is the state of affairs in which the desirer stands in
an extensionally specifiable desiring relation to a particular object or property. This
does not seem to involve intentional ascent in the strict sense of the word, since the
content of the target desire is specified only at the level of reference. There is no
embedded thought, which there clearly is when the target desire is a situation-
desire. The satisfaction-condition of a situation-desire is the state of affairs in
which the desirer stands in the desiring relation to the state of affairs in which she
has a desire with a determinate structured content at the level of sense.3
Of course, the capacity to form second-order desires is not an isolated capacity.
Just as reflective belief revision comes as part of a package of strategies and abili-
ties for epistemic self-assessment and self-monitoring, so too should the capacity
for second-order desires be viewed as part of a complex set of strategies and abili-
ties for determining the sort of person one is going to be. The second-order desirer
is capable not simply of assessing her first-order desires but also of evaluating her
first-order desires one by one and indeed of comparing and weighting them. New
possibilities of decision-making open up. The second-order desirer can reflect on
desires and preferences in a way that will allow her to adopt something like the
primitive version of decision theory that I considered at the beginning of chapter 7.
One would expect, moreover, that this type of reflection would be linked with a
particular type of self-awareness—the awareness of oneself as an agent with dis-
tinctive projects, goals, and aims. The agent who can reflect on her own desires
will also be able to see how they fit together, which ones are more fundamental,
and which are more basic to her character. The analogy with the epistemic case is
very close. In each category the basic capacity for intentional ascent is linked with
reflective abilities that have no analogue at the nonlinguistic level.
9.3 Intentional Ascent and Understanding Other Minds

It is natural to think that the conclusions of the previous two sections can be ex-
tended from the intrasubjective case to the intersubjective case. I have argued that
intentional ascent, and therefore language, is required for reflection on one's own
beliefs and desires. Might a parallel argument hold for reflection on the beliefs and
desires of other subjects and agents? Is it possible for nonlinguistic creatures to
participate in practices of attributing psychological states to their conspecifics, or
indeed to any other creatures? I have shown that nonlinguistic creatures can have a
surprisingly rich psychological life. But to what extent can they understand the
psychological lives of others? Unsurprisingly, it turns out that everything depends
on which psychological states are in question. In this section I argue that only a
very limited range of psychological attributions is possible at the nonlinguistic
level. In essence, all that is available at the nonlinguistic level is an understanding
of the perceptual states of other creatures, an understanding of their desires, and,
relatedly, an understanding of the goal-directed nature of intentional action.
In the light of the preceding discussion, it is not hard to see why a very broad
class of psychological attributions should be unavailable to nonlinguistic creatures.
To attribute, for example, a belief or a situation-desire to another creature is e:3sen-
tially to view that creature as standing in a particular relation to a thought—the re-
lation of believing the thought to be true or the relation of desiring that the state of
affairs characterized in the thought come to pass. Clearly, therefore, the attribution
of a belief requires thinking about a thought. It is a canonical form of intentional
ascent that requires being able to "hold a thought in mind." As such, it is immedi-
ately susceptible to the argument sketched out in section 8.2 to the effect that in-
tentional ascent requires the possibility of semantic ascent and hence is only avail-
able to language-using creatures. A belief can only be attributed if the though): that
is the content of the belief can be represented by the attributer, which requires that
it have a vehicle—and, as I have shown, the only candidate vehicles are linguistic.
The argument applies to the attribution of all mental states that are traditionally
known as propositional attitudes—that is, to the attribution of all those mental
states that involve the subject's taking a particular attitude (the attitude of hoping,
say, or fearing) to a thought. This is closely connected, of course, with two familiar
facts about ascribing propositional attitudes. The first is that propositional attitude
ascriptions create opaque contents—they are intensional. A creature may believe a
particular state of affairs to be the case under one "mode of presentation" (under
one way of characterizing it) but not under another. Opacity is one fundamental
reason why we need to take the objects of propositional attitudes to be thoughts,
rather than the states of affairs or situations that are the truth-conditions of those
thoughts. The second familiar fact (and the reason for the opacity of propositional
attitude ascriptions) is the possibility of error. One can believe things that are not
true and one can desire things that do not exist (that, for example, will not come
into existence unless the desire is satisfied). How can the object of belief be a situ-
ation or state of affairs if it is possible to believe things that are not in fact the case?
What could the object of belief be other than a thought? For these two reasons,
then, there can be no attribution of propositional attitudes without intentional
ascent.
So any mental state that involves taking an attitude to a thought can only be at-
tributed by language-using creatures (although of course it can perfectly well be
had by nonlinguistic creatures). If there are any mental states that can be attributed
by nonlinguistic creatures, therefore, they can only be mental states that do not in-
volve taking an attitude to a thought—and that, in turn, means that they can neither
be opaque nor admit the same type of possible error that we find in the canonical
The Limits 173
prepositional attitudes. It is easy to identify one candidate on the basis of the

distinction between goal-desires and situation-desires. The distinction is effec-
tively between desire construed as a prepositional attitude (in situation-desires,
which are attributed via that-clauses picking out the thought that is the object of
desire) and the more fundamental goal-desires that are directed not at thoughts but
rather at objects or features. It will be recalled that goal-desires, unlike situation-
desires, are extensional. There is no reason, therefore, why nonlinguistic creatures
should not be able to attribute goal-desires to other agents. The argument from in-
tentional ascent cannot get a grip, since goal-desires are relations between a sub-
ject and an object/feature, rather than between a subject and a proposition.
The ability to attribute goal-desires goes hand in hand with a basic understand-
ing of intentional, that is to say goal-directed, behavior. Although there will be
many different degrees of complexity in goal-directed behavior, depending on the
richness of the desires and beliefs by which it is driven, a creature that is capable
of attributing goal-desires will be able to make the basic distinction between pur-
poseful behaviors on the one hand and random movements and instinctive reac-
tions on the other. A purposive action is an action for which a motivating goal-
desire can be identified.
Goal-desires cannot be the only mental states identified and attributed by non-
linguistic creatures. It is hard to see, for example, how a goal-desire can be at-
tributed to a creature without some evidence of the information that the creature
possesses about its environment. At the bare minimum this information will be
perceptual. To know what goal-desire might be motivating a creature at a given
moment we need to know, first, what end it is pursuing and, second, how it might
reasonably expect that end to be realized by its current behavior. Both of these re-
quire knowing to which features of its environment the creature is perceptually
sensitive. If, therefore, a nonlinguistic creature is to be able to attribute goal-
desires to a fellow creature, it must be able to formulate hypotheses about what
that creature is perceiving.
This presents us with a puzzle. Reports of perception are usually taken to bear
significant structural commonalities with prepositional attitude reports. Except for
those philosophers adopting some version of the epistemic theory of perception
(according to which perception can be reductively analyzed in terms of the acqui-
sition of dispositions to belief), there is a consensus that the content of perception
cannot straightforwardly be assimilated to the content of belief (Crane 1992,
Dretske 1981, Peacocke 1983, 1992). Nonetheless, the acknowledged differences
between the content of belief and the content of perception are not usually taken
to stand in the way of reporting the content of perception with "that—" clauses
specifying the perceiver's relation to a thought—the thought in question being the
propositional content of the belief to which the perception might be expected
to give rise.4 But this, by the familiar argument from intentional ascent, seems
to place the understanding of perception beyond the reach of nonlinguistic
creatures—and therefore (by the argument of the previous paragraph) bars them
from identifying and attributing goal-desires.5
The resolution of the puzzle comes with a distinction between two different
types of perception and, correspondingly, between two different types of percep-
tual report. I will call the two types of report SS reports and ES reports, respec-
tively.6 SS reports have the following three characteristics. They
1. Take nonpropositional complements
2. Create transparent contexts
3. Are made true by direct perceptual relations between perceivers and
particulars
ES reports differ from SS reports in all three dimensions. ES reports
1. Take propositional complements
2. Create opaque contexts
3. Are made true by perceptually based epistemic relations between per-
ceivers and propositions
Many philosophers endorse the idea that SS reports and ES reports characterize
different modes of perception. In Dretske's terms, SS reports describe what can be
termed simple seeing, while ES reports characterize epistemic seeing.7 The basic
idea is that simple seeing and epistemic seeing are dissociable components in nor-
mal visual perception. Dretske, in the book that put this sort of distinction on the
map, explicitly suggests that what we see in simple seeing (or what he calls
nonepistemic seeing) "is a function solely of what there is to see and what, given
our visual apparatus and the conditions in which we employ it, we are capable of
visually differentiating" (1969, 76). That is to say, the distinction between simple
seeing and epistemic seeing is phenomenologically salient. Dretske explicitly
draws a parallel with the conception of the sensory given that was at the core of the
sense-datum theory of perception. Whereas the sense-datum theorist believed, or is
alleged to have believed, that we can strip away from the rich manifold of percep-
tual experience to reach a level at which what is really seen are colored expanses,
the simple seeing theorist holds that what we really reach after such a process is
correctly characterized by an SS report.
For present purposes we can remain neutral on the question of whether this dis-
sociability thesis is correct at the level of perception itself. It may or may not be
the case that we can strip away the epistemic dimension from the perceptions of
creatures capable of epistemic seeing in a way that will allow us to characterize
these perceptions entirely in nonepistemic terms. It is certainly true, however, that
the dissociability thesis is perfectly correct at the level of perceptual reports. An
episode of epistemic seeing can perfectly well be characterized, albeit incom-
pletely, by an SS report. The SS report can be viewed as specifying the objects of
perception—I am taking the objects of perception here to be whatever it is that
stands at the other end of the causal chain leading to perceptual states. On this re-
port, an SS report simply states that the perceiver stands in an appropriate percep-
tual relation to a certain object—the relation that will allow her to discriminate that
object from the perceptual background. It has nothing to say about how that object
The Limits 175
is perceived (about its mode of presentation), which of course is why SS reports

create transparent contexts.
The significance of this should be clear. The argument from intentional ascent
shows that nonlinguistic creatures are not capable of the sort of understanding of
vision that might be expressed through ES reports, since ES reports require speci-
fying a perceiver's relation to a thought. But this is perfectly compatible with non-
linguistic creatures being capable of SS reports and, correlatively, of thinking
about the direct perceptual relations between other creatures and objects in the
world. Nor are SS reports, and the understanding that goes with them, quite as re-
stricted as initially appears. Although many of those who have made the distinction
between simple seeing and epistemic seeing have specified the former in terms of
relations purely between a perceiver and an object, there is no reason why this has
to be the case. The relatum of simple seeing can be a state of affairs where a state
of affairs might minimally be construed as an object exemplifying a property or
two or more objects exemplifying a particular relation.8 It seems natural to suggest
that a nonlinguistic creature can include in the complement of an SS report pretty
much anything it can think about directly—that is, anything it can think about di-
rectly it can think about as the object of another's perceptual state.
The availability of SS reports to nonlinguistic creatures allows them to engage
in a primitive form of psychological explanation. Psychological explanation, as
traditionally construed, is not available to nonlinguistic creatures since it depends
on the attribution of beliefs and situation-desires. Nonetheless, a combination
of SS reports and the attribution of goal-desires will permit nonlinguistic creatures
to make some progress on explaining and predicting the behavior of their con-
specifics and potential predators. If one knows what a creature desires and has
some sense of its perceptual sensitivity to the environmental layout (as well of
course as an understanding of its motor capabilities) one can expect to be able to
predict its behavior with some success.
To take stock, then, it appears that propositional attitude ascriptions are beyond
the capacity of nonlinguistic creatures. The argument from intentional ascent
shows us that language is required for all psychological attributions that specify a
thinker's relation to a thought. The only psychological attributions that escape the
argument from intentional ascent are attributions of goal-desires and the SS reports
that characterize simple seeings (and the equivalent, of course, in other sensory
modalities). Nonlinguistic creatures, therefore, are capable only of a highly cir-
cumscribed understanding of other minds that falls far short of what is often de-
scribed as possession of a theory of mind.
It should be noted that this restrictive interpretation of the "mind-reading" abili-
ties of nonlinguistic creatures is compatible with much recent research into the ex-
tent to which nonhuman primates can properly be described as possessing a "theory
of mind." There are well-documented examples of primate behavior that some
prominent students of animal behavior have thought can only be interpreted as ex-
amples of interpersonal deception (de Waal 1982, Premack and Woodruff 1978, and
some of the essays in Byrne and Whiten 1988). But the consensus opinion among
primatologists is that a more parsimonious interpretation of these behaviors is to be

preferred (see, e.g., Gomez 1996, Hauser 2000, and Povinelli 1996).9 Many of the
examples of what has come to be termed tactical deception (Byrne 1995) can be un-
derstood as the manipulation not of another's prepositional attitudes but simply of
their visual perspective. Here is an example of a tactical deception in a troupe of ba-
boons in Ethiopia that lends itself to such an interpretation:
An adult female spent 20 min in gradually shifting in a seated position over a dis-
tance of about 2 m to a place behind a rock about 50 cm high where she began to
groom the subadult male follower of the group—an interaction not tolerated by the
adult male. As I was observing from a cliff slightly above [the animals] I could judge
that the adult male leader could, from his resting position, see the tail, back and
crown of the female's head, but not her front, arms and face: the subadult male sat in
a bent position while being groomed, and was also invisible to the leader. The leader
could thus see that she was present, but probably not that she groomed. (Report by
Hans Kummer, quoted in Byrne 1995, 106)
The behavior of the female baboon, assuming that it is indeed to count as an in-
stance of tactical deception, does not seem to require assuming an intention to ma-
nipulate the beliefs of the alpha male (e.g. an intention to bring it about that he be-
lieve that she is not grooming the subadult male). What she is doing is profiting
from an understanding of the alpha male's visual "take" on the situation to escape
detection. The female baboon needs only to appreciate the alpha male's line of
sight and the fact that he would be prevented from seeing the subadult male by the
intervening rock. This seems firmly at the level of simple seeing rather than epis-
temic seeing.
It is interesting to note that a recent survey article draws a basic distinction
between two different levels of understanding vision that can be mapped without
too much effort onto the distinction between simple seeing and epistemic seeing
and that can be put to work to distinguish between different ways of understanding
the experimental and ethological evidence for primate "mind-reading." Daniel
Povinelli (1996) has distinguished three different types of knowledge about visual
perception and its mechanisms. The first level is a simple sensitivity to the pres-
ence of eyes and eyelike stimuli. This sensitivity is exploited, for example, by the
many species of moth and butterfly that are patterned to resemble eyes. Such sensi-
tivity is clearly widespread throughout the animal kingdom and is too primitive to
concern us. The second level seems closely correlated with an appreciation of
nonepistemic seeing. Povinelli describes it as an understanding of perception as a
cognitive connection between organisms and world. The third level seems much
closer to an understanding of epistemic seeing. This is "the understanding that in
addition to linking an individual's mental state of attention to the external world,
visual perception also alters one's internal experiences, states of knowledge and
belief (Povinelli 1996, 313). Within this classification the standard understanding
of primate social deception would place it squarely at the third level, while the pro-
posal being developed in this section is that it can be understood at the second
level.
The Limits 177
Considerable laboratory research has recently been carried out on primate un-
derstanding of visual perspective and the direction of gaze. Some of this research
has been negative. There is solid experimental evidence that chimpanzees are not
very good at understanding the visual perspective of humans (Povinelli and Eddy
1996a, 1996b). In particular, when given the choice of begging for food from
an experimenter who can see them from one who cannot, they are only capable of
making very crude discriminations. They successfully distinguish between, for ex-
ample, an experimenter facing them and one whose back is turned toward them,
but they do not seem to understand the difference between an experimenter who is
directly looking at them and one who cannot see them because he has a bucket
over his head. Some experimenters have taken this to cast doubt on chimpanzee
understanding of vision and visual perspective. Others have pointed out, however,
that the paradigm is a highly unnatural one for primates, not simply because it
involves humans rather than conspecifics but because the natural situation for
chimpanzees is one of competition for scarce resources, rather than cooperation or
collaboration (Hauser et al. 1993). Recent investigation of chimpanzees' under-
standing of the visual perspective of conspecifics in competitive situations shows
that chimpanzees can monitor the visual perspective of a competitor and informa-
tion to guide their own actions (Hare, Call, and Tomasello 2001, Hare et al. 2000).
One group of experiments involved pairs of chimpanzees, one subordinate and one
dominant, placed in separate rooms and separated by occluders (Hare, Call, and
Tomasello 2001). Experiments hid food in full view of the subordinate chimpanzee
who could expect that the dominant chimpanzee would be successful in competing
for any food that both had seen hidden. The various conditions manipulated the
dominant chimpanzee's visual access to the location of the food—in some he saw
where the food was hidden, in others not, while in a third condition he was misin-
formed about the location of the food because it was moved after he had initially
seen it being placed. In all conditions the subordinate was able to monitor the
dominant's visual access to the location of the food. The subordinate chimpanzee
was consistently able to make use of this information to its own advantage, refrain-
ing from directly competing with the dominant chimpanzee for food about whose
location the dominant had accurate information and preferentially retrieving food
items about which the dominant was either uninformed or misinformed.
A natural way of interpreting these experimental data, in conjunction with the
ethological data that originally gave rise to them, is as manifesting a form of social
understanding reliant on the use of information about vision and goal-directed ac-
tion, rather than on the attribution of beliefs, desires, and other propositional atti-
tudes. If the argument of this chapter is correct then this form of social understand-
ing is all that is available at the nonlinguistic level and is sufficient to explain the
various forms of social deception, social interaction, and social cognition that have
been identified in nonlinguistic creatures. This proposal is line with that offered by
primatologists such as Tomasello and Call (1997) and Whiten (1996), who com-
plain that the existing discussions of primate social cognition are unnecessarily po-
larized between interpretations of primate social behavior in terms of full-fledged
belief-desire psychology on the one hand and unthinking reinforcement on the

other.
9.4 Embedding and Intentional Ascent

The role played by intentional ascent is very clear in the three types of thinking I
have considered. Thoughts are part of what is being thought about in second-order
cognitive dynamics, in second-order desires, and when one speculates about the
mental states of others. When, for example, one attributes a belief to another crea-
ture, one is effectively attributing to that person a relation to a thought. Similarly,
in evaluating a piece of reasoning in the manner characteristic of second-order
cognitive dynamics, one is considering the logical or probabilistic relations hold-
ing between the thoughts of which it is made up. It does not make sense to suppose
that either of these types of thinking could be engaged in without the capacity for
intentional ascent—and hence, by the argument of the previous chapter, without
the possibility of semantic ascent. It would be a mistake, however, to conclude that
intentional ascent is involved only when thoughts are explicitly targeted on other
thoughts—that is to say, when thoughts are part of the content of what is being
thought. I bring out in this section how the capacity for intentional ascent is re-
quired for all thinking that involves compound thoughts with further thoughts em-
bedded in them, irrespective of whether those embedded thoughts feature in the
content of thinking.
We can start with a basic class of compound thoughts—namely, those involving
the basic logical connectives, such as disjunction, conjunction, and the material
conditional. Consider a disjunctive thought of the sort that might be expressed in
the sentence "A or B." What is it to be capable of entertaining such a thought? It is
to be capable of understanding that a certain relation holds between two
thoughts—the relation of their not both being false (and, on some understandings
of 'or,' their not both being true). Of course, the disjunctive thought itself is not a
thought about the two thoughts expressed by A and B. It is a thought about the
states of affairs that serve as the truth-conditions for the thoughts. Nonetheless, the
disjunctive thought is not available to be thought by any creature not capable of
thinking about how the truth-value of one thought might be related to the truth-
value of another thought.10 The same holds of the other truth-functional preposi-
tional operators. Understanding a truth-functional operator is understanding how
it serves to form compound thoughts whose truth-value is a function of the truth-
values of their parts. No creature that was not capable of thinking about thoughts
could have any understanding of truth-functional compound thoughts.11
The example of the truth-functional prepositional operators shows that a par-
ticular type of thinking can presuppose the capacity for intentional ascent, even
though the thoughts instantiating that type of thinking are not themselves thoughts
about thoughts. Compound thoughts formed through the truth-functional preposi-
tional operators involve intentional ascent implicitly rather than explicitly. Are
there other types of complex thoughts of which the same can be said? Extrapolat-
The Limits 179
ing from the truth-functional operators suggests that a similar argument will
apply to complex thoughts formed by operators that can only be understood via
the notion of truth—and this, of course, can hold for operators that are not truth-
functional. Let me offer a conjecture with respect to modal operators (in the sense
of 'modal' on which modal operators are those attributing possibility and neces-
sity). It seems very plausible that our understanding of necessity cannot be di-
vorced from the understanding of truth. Our primary understanding of the notion
of necessity may well be in terms of a prepositional (or, for that matter, a senten-
tial) operator, and a proposition is necessary just if it is true in every possible situ-
ation (in every possible world). If this is the case then modal thoughts would pre-
suppose intentional ascent, and with it semantic ascent, no less than compound
thoughts formed by means of the truth-functional prepositional connectives.
Modal thinking is in many respects similar to tensed thinking. The logical sys-
tems that have been developed to display the logical structure of tensed thought are
analogous to those that have been developed to elucidate the logical structure of
modal thought. They are all what might be termed adverbial extensions of non-
modal and nontensed logics.12 That is to say, they can be derived from nonmodal
and nontensed logics by the application of operators that function as adverbs.13
The following passage is a classic statement of this approach as applied to tense
logic.
If an expression constructs a sentence out of one other sentence it is an adverb or an

adverbial phrase, like 'not' or 'It is not the case that' or 'allegedly' or 'It is alleged
that' or 'possibly' or 'It is possible t h a t ' . . . I want to suggest that putting a verb into
the past or future tense is exactly the same sort of thing as adding an adverb to the
sentence. 'I was having my breakfast' is related to 'I am having my breakfast' in ex-
actly the same way as 'I am allegedly having my breakfast' is related to it, and it is
only a historical accident that we generally form the past tense by modifying the
present tense, e.g. by changing 'am' to 'was,' rather than by tacking on an adverb. In
a rationalized language with uniform constructions for similar functions we would
form the past tense by prefixing to a given sentence the phrase 'It was the case that'
or 'It has been the case that' and the future tense by prefixing 'It will be the case that.'
(Prior 1968, 7, quoted in Evans 1985)
The central implication of the adverbial nature of the tense and modal operators is
that, where "T" is an operator of the relevant class and "p" the name of a sentence,
the meaning of "Tp" will be a function of the meaning of "p" and the meaning of
the relevant operator.
Perhaps it is equally the case that our understanding of temporal order is inextri-
cably linked with our understanding of truth, so that understanding tense is a mat-
ter of understanding that propositions can have different truth-values at different
tunes. On this view, our basic understanding of temporal relations is not an under-
standing of temporal relations holding between events or states of affairs. It is,
rather, an understanding of the temporal relation between the truth of different
propositions. We understand the idea that event A took place earlier than event B
by understanding that it was true that event A took place before it was true that
event B took place. This is because thinking about the temporal relations between
events requires thinking about events holding at particular times, and the under-
standing of an event holding at a time other than the present is parasitic on the un-
derstanding of a particular proposition being true at a particular time. Thinking
about temporal relations (or so I am suggesting) requires being able to think about
the possibility of propositions being true or false at different times. It may well be
that the capacity to think about possibility, to think about time, and to think about
the truth-values of propositions are all interdependent and interlinked. If this were
so, then modal and tensed thinking would be just as unavailable to nonlinguistic
creatures as thinking involving the truth-functional prepositional connectives.
This suggestion about the language-dependence of tensed thinking enables us to
make sense of a puzzling transition in the prehistory of the human race. The con-
sensus among archeologists and students of human evolution is almost universal
that the crucial stage in human cognitive evolution occurred about 40,000 to
35,000 years ago, with the transition from the Middle Paleolithic to the Upper
Paleolithic. This transition involved a sudden explosion in tool technology and
social/cultural organization, with the emergence for the first time of forms of life
that are recognizably congruent with those of modern humans. With the transition
to the Upper Paleolithic come the first decorative objects; the first really com-
pelling evidence for totemistic/religious behavior, as revealed in burial practices
and totemic representations; sophisticated hunting strategies that capitalize on sea-
sonal migrations and fluctuations in animal numbers; and far more complex forms
of tool production that seem to have drawn on detailed knowledge of natural his-
tory to tailor tools for particular hunting tasks, as opposed to the more general-pur-
pose tools of the Lower and Early Paleolithic (for overviews see Donald 1991, ch.
8, Mellars 1996, andMithen 1996a, ch. 9). From this point on, the rate of cognitive
evolution accelerated exponentially. It is, of course, for this reason that many cog-
nitive archeologists have identified this transition as involving the emergence of a
recognizably human language.
One striking feature of the type of cognition emerging in the Upper Paleolithic
is the way it exemplifies a novel conception of time. It is a type of thinking reflect-
ing a conception of time that goes beyond a simple ability to discriminate events
that are taking place in the present from those that have taken place in the past or
that will take place in the future. There are ways of being oriented toward the past
or the future that are available to relatively simple organisms—most simply, in the
form of expectations about future events and memories of past events. But these
need to be sharply distinguished from modes of thought that depend on thinkers
being able to locate themselves within a temporal narrative.14 What the fossil
record tells us about the earlier hominids strongly suggests that they behaved in
ways that involved distinguishing the past, the present, and the future. But there is
nothing to suggest patterns of behavior available only to creatures capable of nar-
rative thought. In contrast such patterns of behavior seem extremely widespread in
the Upper Paleolithic.
A very clear illustration comes with the hunting strategies that emerged during
The Limits 181
this period. Archeologists studying the Upper Paleolithic have found a vastly in-
creased sensitivity to seasonal changes in the density and movements of animals.
Whereas in the Middle Paleolithic hominids tended to hunt single animals in a
very opportunistic way (or indeed perhaps only to scavenge on their remains), the
later hominids appear to have shown far more long-range planning in their forag-
ing behavior. They prepared tools long before they needed them, designing them
for specific prey and specific hunting techniques. Their excavated campsites seem
to have been prepared for long occupations and were systematically moved to
intercept prey or harvest food resources (Binford 1989, Mellars 1996, Mithen
1996a). There is also evidence that, at least after the earlier stages of the Upper
Paleolithic, the targets shifted from isolated animals to large herds of animals like
reindeer and red deer, which were ambushed on their migration paths at particular
times of year (Mithen 1990).
If, as I have suggested, tensed thinking is only available to language-using crea-
tures, then it is easy to see why the evolution of a recognizable human language
should have been connected with the emergence of these types of cognition and
behavior in the transition from the Middle to the Upper Paleolithic. These forms of
hunting behavior presuppose a type of narrative thinking that is quite simply un-
available in the absence of language. And this is, in fact, consistent with influential
views about when natural language emerged in the course of evolution (Mithen
1996).
I have so far in this section considered several ways that intentional ascent
might be involved implicitly in types of thinking that are not explicitly targeted on
further thoughts. I argued that the formation of compound thoughts by means of
the truth-functional propositional connectives presupposes the capacity for inten-
tional ascent and suggested, more tentatively, that similar capacities may be pre-
supposed by the ability to entertain thoughts involving temporal or modal notions.
In the remainder of this section 1 discuss a further way that the capacity for inten-
tional ascent can be presupposed by different types of thinking. The types of think-
ing that I have been considering up to now have involved the formation of com-
plex thoughts by what we can think of as adverbial operators on propositions, such
as the truth-functional connectives and the modal and tense operators. I turn now
to a different way of forming complex thoughts.
Consider the inference form of existential generalization. This is the pattern of
inference instantiated by the transition from "Fa" to " BxFx"—that is to say, from
an atomic sentence to the effect that a named individual has a given property to the
general proposition that at least one individual has that property. The logical oper-
ations involved in this transition are clear enough (Dummett 1973). The first is
breaking down the atomic sentence into two components, a predicative component
and a nominative component (or, in Fregean terms, a function and an argument).
Once the internal structure of the atomic sentence is manifest, the next operation is
to replace the nominative component with a variable. The final operation is to bind
that variable with an existential quantifier. This sequence of logical operations
gives us an important clue as to what is involved in a subject's being able to under-
stand the existential quantifier in a manner that permits existential generalization.

In order to understand how a given sentence can imply an existential generaliza-
tion, a thinker needs to be able to view it as being composed in such a way that the
nominative component can be replaced by an arbitrary name (hence by a variable).
The sentence needs to be "broken down" in thought before the existential quanti-
fier can be applied. But this breaking down in thought of an atomic sentence pre-
supposes the capacity for intentional ascent. It involves holding the thought in
mind and determining its structure in a way that creates a space for the variable
that will be bound by the existential quantifier. The same point holds, mutatis mu-
tandis, for thinking involving the universal quantifier (as 1 will discuss further in
section 9.5).
It would seem, then, that there are two principal ways that the capacity for inten-
tional ascent can be presupposed in the generation of complex thoughts. First, the
formation of compound thoughts by means of the truth-functional prepositional
operators presupposes the capacity for intentional ascent because grasping a com-
plex thought is a matter of grasping the way in which its truth-value is determined
by the truth-value of its constituent thoughts, and truth is a notion that applies pri-
marily to thoughts. Second, complex thoughts involving quantification presuppose
the capacity for intentional ascent because mastery of the mechanisms of quantifi-
cation requires the ability to determine the structure of a thought. I suggested also
that the capacity to form complex thoughts involving modal and tense operators
may well also be dependent on the capacity for intentional ascent, but this sugges-
tion will play no further role in the argument.
If this line of argument is well founded, then it follows that logic requires
language—a conclusion reached on very different grounds by Jonathan Bennett
(1976). So what are the consequences for reasoning at the nonlinguistic level? I
have already explored some relevant terrain in chapter 7, where I showed how
nonlinguistic creatures could be capable of certain forms of protoinference that do
not involve intentional ascent. Conditional reasoning, it was suggested, can be un-
derstood in causal terms, while protonegation can be understood in terms of mas-
tery of contrary concepts (as opposed to the ability to apply a truth-functional op-
erator to propositions). This still leaves open an important question, however. If
thinking involving quantifiers presupposes the capacity for intentional ascent,
hence is unavailable to nonlinguistic creatures, should we conclude that nonlin-
guistic creatures cannot be credited with any form of general thoughts? This ques-
tion will be addressed in the next section.
The proposed analysis of quantificational thinking raises a further question. I
have argued that the key feature in quantificational thinking is the ability to break
down a thought in such a way as to open up a space for a variable. Given the
highly plausible suggestion, already discussed in chapter 4, that the variable of
quantification is the analogue of the relative pronoun in natural language, it is natu-
ral to wonder whether the ability to break down thoughts that comes with the ca-
pacity for intentional ascent might not have cognitive implications that I have not
The Limits 183
yet discussed. In section 9.6 I shall propose that this type of intentional ascent
makes possible (and may well be required for) the integration of different types of
domain-specific thinking.
9.5 Two Types of General Thought

In the previous section I argued that quantificational thinking requires language.
This conclusion seems to entail that nonlinguistic creatures cannot have general
thoughts. This section argues that this conclusion only follows on one way of un-
derstanding generality.
Philosophers have distinguished two ways of thinking about generality (Ben-
nett 1977, Lewis 1969). The distinction goes back at least as far as the twelfth
century—we find it in Abelard's distinction between generality in sensu composite
and generality in sensu diviso. If I believe, in sensu composite, that every F is a G
then the content of my belief is a general proposition to the effect that everything
that is an F is also a G. There may, of course, be individual things that are F but
that I do not believe to be G—since I may be unaware that they are in fact Fs, for
example. Nonetheless, as far as I am concerned it is a general truth that nothing
can be an F without also being a G. So, for example, I might believe in sensu com-
posite that all tigers are dangerous even though, unaware that an albino tiger is still
a tiger, I might find myself in front of a tiger and fail to recognize that it is danger-
ous. If, on the other hand, I believe in sensu diviso that every F is a G, the content
of my belief is not a general proposition at all. Rather, what I have is a certain dis-
position to form singular beliefs. Whenever I encounter an F, or indeed whenever I
think about one, I tend to come to the view that it is a G. It would be evidence
against ascribing to me the belief in sensu diviso that every tiger is dangerous if I
were to fail to identify the albino tiger as dangerous.
The distinction is sometimes termed the distinction between collective gener-
ality (in sensu composite) and distributive generality (in sensu diviso). It can most
easily be viewed as a distinction of scope. Universal quantifiers will feature in the
attribution of both types of general belief, but in the case of general beliefs in
sensu diviso, the universal quantifier will fall outside the scope of the belief opera-
tor, while in the case of general beliefs in sensu composite, it will fall within the
scope of the belief operator.
It is plausible to interpret the distinction between general beliefs in sensu com-
posite and general beliefs in sensu diviso as entailing that only general beliefs in
sensu composite take quantified propositions as their objects. This in turn implies
that general beliefs in sensu composite are only available to creatures that have the
capacity for intentional ascent (for reasons brought out in the previous section). An
analogous point holds, of course, for existential beliefs (and is, in effect, the dis-
tinction between de re and de dicto existential beliefs). A nonlinguistic creature
can have a belief to the effect that some F is a G in the etiolated sense that there is
some F of which it believes that it is a G (with the existential quantifier falling out-
side the scope of the belief operator). Only a language-using creature, however,
can have a belief that is correctly characterized with an existential quantifier
falling within the scope of the belief operator.
Of course, the plausibility of this thesis about the language-dependence of a dis-
tinctive kind of thinking about generality stands or falls with the original distinc-
tion between two types of general belief. Some authors have put forward what ap-
pears to be an intermediate conception of general beliefs, suggesting that there is
really only one type of belief about generality that fits into neither the distributive
nor the collective mold. Braithwaite, for example, offers an account of belief ac-
cording to which believing that p is a matter of behaving as if p were true. He ex-
tends this to general beliefs as follows:
A third advantage of my doctrine [the first two advantages being (1) that it lends itself
to an account of degrees of belief and (2) that it can serve as the basis for an account
of justified inductive belief] is that it assists in one of the most subtle problems of
logic, that of the analysis of general propositions. For action appropriate to belief in a
general proposition does not present any special problems. 'I am disposed to act ap-
propriately to every P being Q' means that, whenever I am disposed to act appropri-
ately to a thing's being P, I am disposed to act appropriately to its being Q. (Braith-
waite 1932-33, 39-40)
Although Bennett, for example, suggests that Braithwaite is offering an account of

distributive generality (Bennett 1977, 104 n. 3), there is a crucial difference be-
tween Braithwaite's proposal and the distributive construal. Braithwaite does not
suggest that the general belief that all Fs are G requires behaving, with respect to
everything that is in fact an F, as if it were a G. Rather, he requires simply that the
believer should behave with respect to everything that he takes to be an F as if it
were a G. Failing to respond appropriately to the dangerousness of an albino tiger
would not, therefore, disqualify me from believing that all tigers are dangerous
(hence the belief would not be correctly characterized with a universal quantifier
outside the scope of the belief operator in the way that general beliefs in sensu di-
viso would be correctly characterized).
But it is far from clear that Braithwaite has really given us a satisfying account
of general beliefs. One problem with his view is that it lacks the appropriate coun-
terfactual implications. A believer might be disposed to behave with respect to
everything that she hi fact takes to be an F as if it were a G without being disposed
to behave with respect to everything that she might take to be an F as if such things
were Gs. A genuinely universal belief to the effect that all Fs are Gs is completely
incompatible with the possibility that something might be taken to be an F and yet
not be believed to be a G—but Braithwaite's account seems to leave this possi-
bility open. Of course, though, a simple modification will rescue the theory. All we
need require is that the believer be disposed to behave with respect to everything
that she is disposed to believe to be an F as if it were a G. But this brings us to the
real problem. Braithwaite is suggesting not only that creatures can have fully gen-
eral beliefs even though they do not have the resources to mark that generality in
The Limits 185
any way (either at the linguistic level or at the level of thought) but, more strongly,
that there are no general beliefs for which markers of generality are required. And
it is this second claim that must be challenged. It cannot be the case that there is no
significant cognitive difference between a creature that has a generalized Braith-
waitean disposition, on the one hand, and a creature for which such a disposition
is grounded in an explicitly formulated general thought of the type expressible by
a universally quantified sentence. The difference is easily capturable when we
consider the logical form of the appropriate belief ascriptions. I have already
shown that general beliefs in sensu diviso should be characterized in the following
manner:
For all x, if jc is an F then j8 believes (is disposed to act as if) x is a G
General beliefs in the Braithwaite style are specifiable in a broadly similar manner,
with a slightly expanded antecedent.
For all jc, if AT is an F and is taken by p to be an F, then y3 believes (is disposed to
act as if) A: is a G
But this still falls short of general beliefs in sensu composite specifiable as follows:
(3) /3 believes that, for all x, if x is an F then x is a G.
Nothing could persuade me that there is no significant difference between (2)
and (3).
It would seem, then, that there is a real and genuine distinction to be drawn be-
tween two types of thinking about generality. Although one of these ways of think-
ing about generality (that which involves contents containing quantified general
beliefs) is restricted to creatures capable of intentional ascent (and hence of seman-
tic ascent), the distinction offers a type of thinking about generality that is indeed
available at the nonlinguistic level. As such, it provides a primitive analogue for
quantificational thinking in the same manner that the forms of protoinference dis-
cussed in chapter 7 provide primitive analogues of the truth-functional preposi-
tional connectives.
9.6 From Domain-Specificity to Domain-Generality

The hypothesis of domain-specificity has become increasingly popular among
theorists in cognitive archeology, evolutionary psychology, and developmental
psychology. (Barkow, Cosmides, and Tooby 1992, Hirschfeld and Gelman 1994,
Mithen 1996; but see Fodor 2000 for criticism of extreme versions of the hypothe-
sis). This section suggests that the apparent domain-specificity of various types of
nonlinguistic thinking may in fact reflect a structural limitation on the types of
thinking available at the nonlinguistic level.
The basic tenet of the hypothesis of domain-specificity is that certain fundamen-
tal types of cognitive activity are carried out by modular systems that have evolved
to deal with particular types of problem and particular types of situation. Popular
candidates for domain-specific modules include the interpersonal competences in-

volved in social interactions; the basic principles about objects and their interac-
tions that are usually collectively labeled naive physics; and an intuitive grasp of
folk biology and natural history. These modules operate on a highly selective and
domain-specific set of inputs with a fixed and limited amount of background infor-
mation. For present purposes, there is systematicity within each module, but not
across modules. An example of this failure of systematicity can be found in the
archeological record—namely, the failure of early hominids to integrate their prac-
tical abilities in tool construction with their detailed knowledge of natural history
in order to produce hand-axes for specific purposes. In the Middle Paleolithic, for
example, we find what seem to be highly developed tool-making skills existing
side by side with a subtle and advanced knowledge of the natural environment, but
it is not until the Upper Paleolithic that we see these two bodies of knowledge
being integrated in the form of tools specially designed for dealing with different
plants and animals, together with hunting strategies that are tailored to the habits of
specific animals (Mithen 1996a).
The proposal I want to consider is that language is required for the integration of
domain-specific modules.1S The argument here hinges on the distinctive role of the
relative pronoun in permitting the construction of relative clauses. This was briefly
discussed in chapter 4 when I considered Quine's account of the psychogenesis of
reification. I rejected Quine's argument that reification depends on the linguistic
mechanisms of quantification and the relative pronoun (the natural language
equivalent of the bound variable of quantification). But nonetheless I saw no rea-
son to dispute Quine's general claim about the enormous cognitive significance
of the linguistic device of the relative pronoun. The crucial feature of the relative
pronoun in this respect is that it permits the formation of relative clauses. One way
of thinking about relative clauses is as a way of distinguishing within a sentence
between the object that is the logical subject of the sentence (what the sentence
is about) from what the sentence says about that object (Quine 1974, sec. 24,
1995).16 In English, for example, from a sentence like "The red deer comes to the
water just before nightfall" we can extract the relative clause "that comes to the
water just before nightfall," which can be used to characterize other animals, or
be embedded in further sentences, and so forth. Once the relative clause has thus
been constructed and detached from the original sentence, the information it pro-
vides is available in a form that can be employed in a variety of different contexts.
Consider a simplified model of domain-specific cognition in which a thinker can
think about two separate domains, each of which contains two objects and four
properties. The properties from one domain can only be applied in thought to the
objects in that domain, and no object or predicate features in more than one do-
main. Suppose that a is an object in the first domain and G a predicate in the sec-
ond domain. How might such a creature come to be capable of the thought that
Gal Only be forming a conception of what it is for something to be a G that is no
longer tied to the objects in the second domain. That is to say, only by forming a
conception of what it is for an arbitrary object to be a G. And it is this that the in-
The Limits 187
(reduction of the relative pronoun makes possible. The relative clause is the most
basic way of extrapolating complex predicates from complete sentences in a form
that will allow them to be applied to objects falling under other cognitive do-
mains—and the defining feature of domain-general cognition is that objects from
one domain can be thought about in terms formerly associated only with objects
from another domain, as in totemic art when an artifact is attributed the properties
of an animal, or as in advanced tool construction when the design of a tool is
specifically tailored to properties of the intended prey.
Let us assume, therefore, that the integration of different types of domain-
specific cognition must involve some cognitive mechanism that operates in a man-
ner analogous to the relative pronoun to distinguish within thoughts between what
the thought is about and what the thought affirms of that object. At this point the
close relation between the relative pronoun and the variable of quantification be-
comes relevant once again. One might expect this cognitive mechanism to operate
in a manner very similar to that involved in extracting an existential generalization
from a given sentence—namely, by breaking the thought down in such a way that
the predicative component is detached from the nominative component. In the case
of quantified thought, the result of this "decomposition" of the thought is to allow
the insertion of a variable bound by a quantifier into the place occupied by the
name. In the case of the transition from domain-specific to domain-general think-
ing, however, the result of the "decomposition" is to make it possible to insert an
arbitrary name into the place of the name. In both cases, however, the reliance on
the capacity for intentional ascent is clear. What is required is the ability to hold a
thought in mind in order to identify and manipulate its structure. And this, by the
argument of chapter 8, is only possible when that thought has a linguistic vehicle.
In the first seven chapters of the book I defended at some length the thesis that it
can be appropriate and correct to attribute to nonlinguistic creatures thoughts that
have determinate contents, are compositionally structured, and reflect the mode of
presentation under which the creature in question apprehends the immediate envi-
ronment. Not only is the ascription of such thoughts often mandated by the re-
quirements of psychological explanation, but the explanatory practices within
which such thought-ascriptions take place can be embedded within viable concep-
tions of nonlinguistic reasoning and nonlinguistic rationality.
The principal claim of the two final chapters has been that there are significant
limitations to the cognitive abilities of nonlinguistic creatures. Certain types of
thinking are in principle only available to creatures that dispose of a language. The
fundamental reason for the restriction developed in chapter 8 is that thoughts can
only be the objects of further thoughts when they have linguistic vehicles. Inten-
tional ascent (that is to say, thinking about thoughts) requires the possibility of se-
mantic ascent. The argument from intentional ascent to semantic ascent places sig-
nificant restrictions on the scope of thinking without words, Many types of
thinking involve intentional ascent explicitly, because they are directly targeted on
first-order thoughts. This is the case, for example, in the types of reflection associ-
ated with second-order cognitive dynamics (section 9.1) and most forms of higher-
order desire (section 9.2). Explicit intentional ascent is also involved in all forms
of psychological understanding that involve attributing thoughts to others (section
9.3). In section 9.41 showed that intentional ascent can also be involved implicitly
even in types of thinking that are not explicitly targeted on first-order thoughts—
that is to say, in types of thinking whose contents do not involve further thoughts.
These are types of thinking into which a creature cannot enter without the capacity
for intentional ascent. The principal example discussed in section 9.4 was logical
thinking involving the truth-functional prepositional connectives and quantifiers
(although it was also suggested that modal and tensed thought might fall into this
category). In section 9.5 I identified a type of general belief that does not presup-
pose intentional ascent and hence is available to nonlinguistic creatures. The final
section of the chapter explored the role of language hi making possible the transi-
tion from domain-specific to domain-general cognition. It was argued that the
mechanisms required for domain-general thinking are very similar to those that
make possible quantificational thought.
The picture that has emerged of the scope of nonlinguistic thinking provides
principled grounds for thinking that many cognitive abilities that have traditionally
been taken to be uniquely human are indeed unique to language-using humans.
Only language-using creatures can be logical thinkers, monitor their own proces-
ses of belief formation and argument, and reflect on the desires that they want to
have. Only language-using creatures are capable of attributing thoughts to other
creatures. Yet the gulf between linguistic and nonlinguistic thought should not be
exaggerated. Many of these uniquely human cognitive abilities have analogues at
the nonlinguistic level. Logical thinking may be the preserve of language-users,
but forms of general belief and types of protoinference are available at the nonlin-
guistic level. Nonlinguistic creatures cannot monitor their own processes of belief
formation, but they are nonetheless capable of sophisticated forms of belief revi-
sion. The attribution of thoughts is not possible at the nonlinguistic level, but there
are still relatively complicated ways that nonlinguistic creatures can think about
the perceptions and desires of other creatures and hence in which they can explain
and predict behavior in, broadly speaking, psychological terms. The cognitive sep-
aration between creatures that have language and creatures that do not is very real.
But it is a separation between two types of thinking—between two ways of repre-
senting the social and physical environment—rather than between thought and the
absence of thought.
Afterwordd
The conception of nonlinguistic thought developed in this book has been formu-
lated with a view primarily to the requirements of, and constraints imposed by,
those disciplines studying the cognitive abilities of nonlinguistic creatures,
namely, cognitive ethology, developmental psychology and cognitive archeology. I
have stressed, not simply the nature and content of the thoughts that might be at-
tributed to nonlinguistic creatures, but also the practices of explanation within
which those attributions take place. Indeed, the accounts eventually reached of the
nature and content of nonlinguistic thinking were formulated in the light of par-
ticular ways of understanding those practices of explanation. The plural here is im-
portant. Different types of explanation are appropriate for different types of behav-
ior at the nonlinguistic level, and those types of explanation bring with them
different and distinctive ways of understanding the thoughts that lie behind the be-
haviors in question.
The most obvious contrast is between, on the one hand, those explanations that
presuppose some nonlinguistic analogue of belief-desire psychology and, on the
other, explanations that appeal to non-propositional perceptual states. We do not
have to make a general choice between understanding the thoughts of nonlinguis-
tic creatures on the model of prepositional attitudes and understanding them as es-
sentially perceptual in form (in the manner proposed by the minimalist approach
discussed in chapter 3). Each model is suitable for different types of behavior.
Most examples of thinking behavior in nonlinguistic creatures should no doubt be
accommodated in more or less the way that the minimalist proposes—as involv-
ing, in Dummett's phrase, the superposition of spatial images upon spatial percep-
tions. Behaviors for which the minimalist approach is appropriate are character-
ized by being circumscribed by the here-and-now. The information on which the
animal or infant acts is available in the content of perception. No abstraction away
from the environmental context is required. There is, in short, no form of reasoning
involved. This does not mean, however, that there is no sense in which behavior of
this type can be assessed for rationality. The notion of rationality is inextricably
linked with the practice of psychological explanation. An explanation works by
making an action comprehensible and, broadly speaking, an action is made com-
prehensible when it becomes clear why it was a rational thing for the agent in
question to carry out, in the light of their motivations and the information at their
disposal. But rationality does not require reasoning. There are different ways in
which the notion of rationality can be deployed with respect to nonlinguistic crea-
tures, and what I termed level-1 rationality is clearly applicable in the context of
minimalist explanations. Minimalist explanations are appropriate for behaviors
that can be understood as responses to perceptually available information, and the
rationality of those responses can be understood according to a range of normative
criteria even though the responses in question are not reasoned responses.
Yet, the minimalist approach has its limitations. There are behaviors that out-
strip the bounds of the here-and-now in a way that the minimalist approach cannot
accommodate. These behaviors have in common a type of consequence-sensitivity
that involves instrumental reasoning about the outcomes of actions—reasoning
that exploits the particular connections between actions and outcomes. It is with
behaviors of this type that we come to the domain of propositional attitude psy-
chology. We need to attribute beliefs and desire to make sense of these behaviors.
This is so for two reasons. The first is that the types of mental state invoked in the
minimalist mode of explanation do not support instrumental reasoning of the ap-
propriate type. The second is that the form of explanation appropriate to conse-
quence-sensitive behavior requires mental states with contents that have structure
(that is to say, that are made up of constituents that can feature in further thoughts)
and whose contents can be linguistically expressed in a relatively determinate
manner and without remainder. These reasons are of course related. We have diffi-
culty making sense of the idea of reasoning defined over states lacking determinate
contents and intrinsic structure.
The proposal to apply propositional attitude psychology to nonlinguistic crea-
tures brings its own peculiar difficulties. In particular there is the problem of ex-
plaining both how the beliefs and desires of nonlinguistic creatures can have deter-
minate and linguistically expressible contents and how we can find out what those
contents are. The version of success semantics proposed in this book tackles these
questions simultaneously. The requirements of successful action fix the content of
belief in such a way that what a creature believes can be worked out from how it
behaves and what it wants to achieve. Of course, the way in which a creature be-
haves needs to be understood broadly, as involving not just how it actually did be-
have but how it would behave in different circumstances. And, in order to know
what a creature believes about the world one needs to have an idea of the sorts of
discriminations that it is capable of making between categories of objects, of the
similarities to which it is sensitive and of the general expectations that it possesses
Afterword 191
about how bodies behave. This sort of understanding of how a creature carves up
its environment is not easy to gain, but we have seen that there are ways in which it
can be gained, drawing upon existing research paradigms for studying infants and
animals.
The beliefs and desires attributed to nonlinguistic creatures feature in instru-
mental reasoning. Yet the forms of reasoning available at the nonlinguistic level
are fundamentally different from those in terms of which reasoning at the linguis-
tic level is usually understood. Animals and infants are not capable of applying for-
mal rules of inference defined over propositions. Not only is the evidence usually
taken to reveal mastery of formal principles of inference in nonlinguistic creatures
highly inconclusive, but there are good grounds for thinking that there can be no
logic without language. These grounds derive from the general line of argument,
developed in chapter 8, to the effect that intentional ascent requires semantic as-
cent. Public language vehicles are required for thoughts to be the objects of further
thoughts, and (as I argued in section 9.4) it is not possible to understand either the
truth-functional logical connectives or the mechanisms of quantification without
being capable of intentional ascent of this type. So, we cannot understand the way
in which a nonlinguistic creature's beliefs and desires can lead to action by think-
ing about how it might reason deductively or inductively from those beliefs and
desires to a particular intention to act. In chapter 7 I offered an alternative way of
understanding reasoning at the nonlinguistic level, in terms of what I termed proto-
inference. Proto-inferences are not made in virtue of their form. Rather, they trade
on a creature's mastery of pairs of contrary concepts and on its understanding of
causal connections between states of affairs. The mechanisms of proto-inference
allow analogues at the nonlinguistic level of fundamental forms of reasoning, such
as reasoning from an excluded alternative (the analogue of disjunctive syllogism)
and what I termed proto-modus ponens and proto-modus tollens.
The question of nonlinguistic rationality reflects the more general issue of the
differences between linguistic and nonlinguistic thought. The argument that in-
tentional ascent requires semantic ascent imposes serious limits on the range of
thoughts available at the nonlinguistic level. Most obviously, it means that animals
and infants cannot be capable of attributing propositional attitudes in the interests
of psychological explanation, since attributing a mental state effectively involves
identifying a relation between an individual and a thought. Nor are nonlinguistic
creatures capable of the type of second-order cognitive dynamics that involves ex-
plicitly reflecting on the inferential connections between thoughts and the likeli-
hood of their truth, since this requires explicitly taking thoughts as the objects of
thoughts. This does not entail, however, that no forms of psychological under-
standing or belief revision are available at the nonlinguistic level. As we saw in
chapter 9 there are primitive forms of psychological understanding that do not in-
volve attributing propositional attitudes. These forms of psychological understand-
ing exploit perceptual connections between agents and states of affairs and an
understanding of the goals of intentional action. Similarly, there are relatively
sophisticated forms of belief revision that do not involve the type of metarepresen-
tational thinking characteristic of second-order cognitive dynamics. This, of
course, is precisely what one would expect. As I have stressed in earlier work
(Bermudez 1998), high-level cognitive abilities do not emerge ex nihilo. They
arise from a basis of lower-level cognitive precursors that are both ontogenetically
and phylogenetically primitive.
The overall picture that has emerged of nonlinguistic thought strikes a balance
between the two conceptions of the relation between thought and language with
which we began in chapter 2.1 have stressed that thinking in general by no means
requires the vehicle of a public language. Moreover, we can attribute to nonlin-
guistic creatures full-fledged thoughts that are the objects of prepositional atti-
tudes, in addition to the non-propositional and essentially perceptual types of
thinking envisaged by the minimalist that we considered in chapter 3. As we saw
in chapters 4 and 5 the full-fledged thoughts we can attribute to nonlinguistic crea-
tures are composed of distinguishable components that can feature in further
thoughts. These thoughts reflect the creature's ontological perspective on the
world—the way in which it carves up the environment into bounded individuals
and the object-properties to which it is sensitive. They have determinate contents
that must be specified by means of a "that-"clause. And they reflect the aspectu-
ality and intensionality of the modes of presentation under which objects and prop-
erties are thought about. Nonetheless, in opposition to the conception of the rela-
tion between language and thought favored by language of thought theorists, we
cannot simply analyze the thinking of creatures who do not possess a public lan-
guage by crediting them with an internal language of thought that has the expres-
sive capacity of a public language. The argument that intentional ascent requires
semantic ascent places significant limitations on the expressive power of nonlin-
guistic thought. And hence, even if it is thought that nonlinguistic thought must be
realized in a language of thought, that language of thought must be significantly
different to that hypothesized to account for the thinking of nonlinguistic creatures.
The fact of the matter, however, is that we have little idea of what the vehicle of
nonlinguistic thought might be. At the level of the vehicle, nonlinguistic thoughts
are rather similar to scientific unobservables (Sellars 1956/1997). We can see their
effects but (as yet) only speculate as to the intrinsic nature of their vehicles. They
provide us with a powerful and indispensable tool for understanding the behavior
of nonlinguistic creatures, a tool that we can employ even though we do not clearly
understand the physical basis for it. What I have tried to bring out in this book,
however, is that we can ascribe thoughts with structured contents to nonlinguistic
creatures and employ those thoughts to give intentional explanations of their be-
havior without knowing what the vehicles of those thoughts are. Knowing what the
vehicle of nonlinguistic thoughts are is not necessary to give intentional explana-
tions of the behavior of nonlinguistic creatures. And nor is it sufficient. Even if we
were convinced by arguments such as those put forward by proponents of the lan-
guage of thought hypothesis this would not help us at all in determining the con-
Afterword 193
tents of those thoughts and how they might be employed in psychological explana-
tion. What we need, and what I have tried to show that we can attain, is a way of
working backwards from behavior to genuinely explanatory structured representa-
tions. This will ultimately yield a conceptual foundation for the various disciplines
that are committed to giving psychological explanations of the behavior of nonlin-
guistic creatures.
Notes
Chapter 1
1. Interestingly, however, it has been suggested that some processes of instrumental con-
ditioning are not in fact completely nonpsychological. See for example Heyes and Dickin-
son 1993 and Dickinson and Balleine 1993.
2. The remainder of this section draws on section 4.2 of Bermudez 1998.
3. In the following, unless there is explicit indication to the contrary, I will be using
"linguistic thought" to abbreviate "the thoughts of language-using creatures." It should
not be thought that linguistic thoughts are necessarily linguistically expressed by the thinker.
Chapter 2
1. Dissenting interpretations of Frege of varying stripes will be found in, for example,
Baker and Hacker 1984, Makin 2000, Weiner 1990, and Sluga 1980.
2. The interpreter most often mentioned in this context is Michael Dummett, who has
done more than anyone both to promote Frege as a serious philosopher and to promote the
idea that Frege's fundamental contribution to philosophy is the idea that we can only ana-
lyze thought through analyzing language.
3. For illuminating discussion of Frege's notion offeree see Dummett 1973, ch. 10.
4. This claim has been challenged by the tradition of possible worlds semantics. Accord-
ing to Stalnaker (1987), for example, the objects of prepositional attitudes are sets of possi-
ble worlds and as such do not have any constituent structure. He writes, for example, that
"belief-desire explanations do presuppose that propositions stand in logical relations such as
entailment, and one can generalize about entailment relations by describing propositions in
terms of the way they are or might be determined. But as the possible worlds analysis of
propositions makes clear, the assumption that propositions stand in logical relations does
not imply that they have linguistic structure or logical form" (60-61).
A proper consideration of Stalnaker's position would take us too far afield, but it is worth
noting that he does make room for structure in the way a thought (or proposition) is deter-
mined. The basic idea is that thinkers have a way of thinking about Socrates and the property
of mortality that can be combined to determine a particular set of possible worlds (namely, all
and only the possible worlds in which Socrates is mortal). Similarly, thinkers have a way of
thinking about the property of mortality and the existential quantifier that can be combined to
determine another set of possible worlds (namely, all and only the possible worlds in which at
least one thing is mortal). The inference is valid just if the first set of possible worlds is a sub-
set of the second. There are definite advantages to this way of thinking about inference (some
of which will be considered later on in this book). For the moment I need note simply that
structure has not been banished. It has merely been relocated—from the thought itself to the
way in which that thought is determined.
5. There are nonetheless difficulties with Frege's suggestion that the structure of a
196 Notes to Pages 23-36
thought is given by the structure of the sentence that expresses it. For further discussion see
Bell 1987, Dummett 1991a, and Bermudez 2001a.
6. In fact, one of the key arguments for the language of thought hypothesis is that we
cannot explain the possibility of language learning without assuming that there is a lan-
guage of thought (Fodor 1975). The argument is, in essence, that language learning is a
process of hypothesis formation and testing and hence requires a language-like medium in
which those hypotheses can be formulated and evaluated. Ex hypothesi this cannot the lan-
guage being learned. Hence, Fodor concludes, it must be a language of thought at least as
expressively powerful as the language being learned.
7. Defenders of the language of thought hypothesis do have an alternative line of re-
sponse. They can claim that the process of language learning is essentially a process of hy-
pothesis formation and testing that requires postulating a language of thought. This is
Fodor's view. It is unlikely to be widely accepted, however. There is little evidence that lan-
guage learning is anything like what Fodor describes.
8. The remainder of this section develops arguments originally presented in Bermudez
1995c.
9. A possible model for this type of explanation (although not one developed from the
perspective of the language of thought theory) comes with the speculations about the rela-
tion between mirror neurons and empathetic understanding of conspecifics to be found in
Gallese and Goldman 1998.
10. This problem arises for any theory of the semantics of sentences in the language of
thought. Nothing in the argument to follow hinges on the earlier decision to consider causal
covariance theories rather than, say, Ideological theories.
11. This interdependence of syntax and semantics clearly poses the question of how lan-
guages are acquired in human development. Even if, as Chomsky and others have sug-
gested, there is an innate basic syntactic competence, the problem still arises for the acquisi-
tion of semantics. It is widely accepted that prosodic cues (e.g., the fact that changes in
frequency and lengthening indicate major clausal and phrasal boundaries) and a general,
probably innate knowledge of the general characteristics of clauses and phrases play a sig-
nificant part in language acquisition (Stromswold 2000). But it is hard to see how these
could suffice without the operation of some sort of 'semantic bootstrapping,' where seman-
tic bootstrapping is precisely the process of using the interdependence of semantics and syn-
tax to work toward a semantic understanding (Pinker 1994). There are very real questions
about the extent of the semantic knowledge required, although all parties are agreed that ar-
guments from the poverty of the stimulus are less compelling for the acquisition of seman-
tics than for the acquisition of syntax.
12. Compare Braddon-Mitchell and Fitzpatrick 1990, 14.
13. There is a nice example of how this might be done in Churchland and Sejnowski
1992, ch. 4,183-188. They show (using a neural network model of how the shape of a figure
can be computed from the patterns of shading on its surface) how individual neurons that
seem to be responsive to particular stimuli (hence that look very much as if they are grand-
mother neurons) can be interpreted as forming parts of vectors processing information that
does not in any straightforward way map onto the individual features to which the indi-
vidual neurons are responsive.
Chapter 3
1. There are also analogies with the distinction made by cognitive psychologists be-
tween declarative memory (remembering-that) and procedural memory (remembering-
how). See, e.g., Squire, Knowlton, andMusen 1993.
Notes to Pages 37-54 197
2. It is important to distinguish this proposal from a superficially similar claim. Theo-

rists concerned with the subpersonal underpinnings of thinking-how have debated whether
imagistic representations exist at the subpersonal level. Although certain well-known exper-
imental results (Shephard 1982) seem to suggest that the subpersonal vehicles of certain
personal-level types of reasoning are imagistic in form, other theorists have argued that the
results are indeed compatible with versions of the language of thought theory (see Block
1991 and Tye 1991 for discussion from a philosophical point of view). But this debate is or-
thogonal to that with which I am currently concerned. Thinking-how is proposed as a per-
sonal-level phenomenon—and it is quite possible that the subpersonal vehicles of thinking-
how might be propositionally encoded sentences in a language of thought. For present
purposes I am taking "subpersonal" to be synonymous with "at the level of the organism"
and hence not confined to persons as standardly understood. For further discussion of the
relevance of the person/subpersonal distinction to psychology see Bermudez 2000b and the
other essays collected in Bermudez and Elton 2000.
3. I am very grateful to David Chalmers for suggesting this way of presenting and en-
gaging with the minimalist approach.
4. This is why the various types of indexical thought have traditionally posed problems
for the propositional approach to thought. For further discussion see Kaplan 1989 and Perry
1977 and 1979.
5. Once again, of course, indexicals pose significant difficulties (Perry 1977, 1979).
These are not, however, difficulties that have led theorists to propose indexical thought as an
example of nonpropositional thought in the sense I am discussing.
6. It is also open to a theorist, at least in principle, to maintain linguistic expressibility
and structure while denying the independence of content from context and the availability
of the content/force distinction. But it is difficult to see what might motivate such a position.
7. For a short introduction to the terms of the debate see Bermudez forthcoming-c. As I
explained in the preface, the position I am developing in this book is intended to be neutral
on the general question of conceptual versus nonconceptual content.
8. For discussion of what it is to take a perception "at face value" see Peacocke 1999
and Bermudez 2002.
9. Cussins does not himself use these terms, but this way of putting the point is, I think,
in the spirit of his account.
10. The theory of affordances has most frequently been applied to vision, but the thesis
extends to the other sensory modalities (see Turvey 1996 for an ecological approach to hap-
tic perception).
11. For a discussion of the extent to which the abilities of marsh tits and chickadees to
hide and retrieve food is really due to place-memory rather than the deployment of simple
heuristics and rules of thumb, see the articles by Sherry cited in the text and Shettleworth
and Krebs 1986.
12. These differences are slightly diminished if the minimalist chooses to specify the
contents of nonlinguistic desires in terms of situation-desires, rather than goal-desires, since
a situation-desire is much closer to a propositional attitude than a goal-desire. Nonetheless,
even with situation-desires there is still no need to move beyond perception in psychologi-
cal explanation.
13. For a more detailed discussion of immediate perception and the issues that it raises in
the epistemology of perception see Bermtidez 2000a.
14. See, for example, Armstrong 1961. There is an instructive discussion of the inferen-
tial approach to mediate perception in the first chapter of Jackson 1977.
15. The use of dishabituation experiments in infant and animal cognition will be dis-
cussed in more detail in chapter 4.
16. Nonetheless, some learned behaviors, such as the tool-use in cotton-top tamarins
(Hauser 1997) discussed in the previous section, can be accommodated in terms of the di-
rect perception of instrumental properties.
17. This distinction will be discussed further in chapter 6 in the context of rationality at
the nonlinguistic level.
18. This is adapted from Baker 1995, 122. Baker is more cautious than I am. She holds
EAC to be a sufficient but not a necessary condition of a genuine explanation. My second
clause is also slightly different from hers. She has: (2) given that an F-type event did occur,
the occurrence of a G-type event was inevitable. The difference is not important, however,
as my clause 2 can be seen as a way of making plainer what inevitability might mean in this
context.
19. Of course, EAC is roughly formulated, and there is an extensive literature devoted to
making it more precise and less susceptible to counterexamples, but it is quite adequate for
present purposes. A version of EAC is sometimes adopted as the main plank of counterfac-
tual analyses of causation (e.g., Lewis 1973). Nothing I say here depends on the acceptabil-
ity or otherwise of such counterfactual analyses.
20. I will discuss desires here purely for the sake of convenience—because we already
have in play an understanding of how to go about establishing the content of a desire. It
should be clear that the following argument could be applied equally effectively to beliefs.
21. See particularly the discussion of level 1 rationality in sections 6.3 and 7.1.
Chapter 4
1. In any case, worries about circularity could easily be dispelled through the mecha-
nism of ramsification (Lewis 1972). Ramsification makes a virtue out of the interdepend-
ence of belief and desire attributions.
2. In fact, the content of both beliefs and situation-desires is given by the state of affairs
as apprehended in a particular way. I shall return to this later; it is not important for the mo-
ment.
3. Readers of Peacocke 1999 will recognize this as an instance of what he calls the In-
tegration Challenge. For further discussion see Bermudez 2001d.
4. Various control experiments rule out the obvious explanations of the monkey behav-
ior, such as their having a bias toward the bottom box (as a function, perhaps, of food al-
ways being on the ground). The details can be found in Hauser 2001.
Chapters
1. This is an idealization, of course. In practice, thoughts are not that fine-grained (nor
do we need them to be for the purposes of psychological explanation and prediction). A sub-
set of the descriptions will be adequate. The essential point is that some have to be excluded
in order to narrow the field down until we have the type of localized indeterminacy that we
are happy to accept in our ordinary social interactions.
2. There is a clear example of the subtraction fallacy in a well-known argument that has
been used to argue for a range of theses in metaphysics, from four-dimensional conceptions
of physical objects to the relativity of identity. The argument hinges on the claim that, since
there is an object in existence at a time t at which, for example, I cut off my hand, there must
have been an object in existence before time t corresponding to my physical body minus my
hand. This claim is, of course, completely unwarranted. See Heller 1997, ch. 1, for a de-
ployment of the claim in defense of four-dimensionalism about objects.
3. Clearly, these sentences cannot strictly speaking be atomic sentences, since food is
not a particular and hence not something that will be picked out by a singular term. I will
put these difficulties to one side, however, as they do not affect the principal issue here,
which is how we should view the predicative component of the sentences in the target-set.
4. This should not taken to imply the existence of an absolute space determining ab-
solute spatial position—merely that there is some privileged coordinate system that fixes
spatial location at the level of reference.
5. For further discussion see Campbell 1993, Evans 1982, and (with particular reference
to nonlinguistic creatures) Bermudez 1998, ch. 8.
6. It is natural to ask whether these four proposed ways of understanding how the loca-
tion of the food is apprehended by the rat exhaust the range of possibilities. As will be ex-
plored further below, rats are very capable of deploying nongeometrical cues in situations
where the geometrical structure of their environment is systematically perturbed, but there
is considerable evidence that rats tend to use geometrical information when that information
is available (Gallistel 1990, ch. 6, O'Keefe and Nadel 1978). Moreover, the experimental
paradigms under discussion are designed to rule out the possibility that the rats are using
nongeometrical cues. If, as this suggests, the only candidate modes of presentation of the lo-
cation of the food are geometrical, then there are reasonable grounds for thinking that there
are no further possibilities other than the four canvassed here. I am assuming, following
Gallistel 1990, that the frames of reference deployed by rats (and within the animal king-
dom generally) are metric systems.
7. As discussed in chapter 4, however, success semantics is most plausible when it is de-
veloped without the global explanatory pretensions of philosophical functionalism. I am of-
fering success semantics solely as a theory of nonlinguistic thought. There are, I think, no
prospects for using it to explain all types of propositional thinking.
Chapter 6
1. This conception of psychological explanation is held widely but not unanimously. Ac-
cording to the theory of intentional icons developed by Ruth MUlikan, intentional states such
as beliefs and desires should be understood in functional terms—in terms primarily of why
they have come about and what jobs they are designed to do. A corollary of her approach to
mental states is to downplay the rational connections holding between beliefs, desires, and
other propositional attitudes. Psychological explanation, as Millikan construes it (in, for ex-
ample, Millikan 1984, 1986) is a form of functional explanation, much closer to explanation
in biology than to psychological explanation as traditionally conceived. Millikan's general
approach has been sympathetically applied to the domain of cognitive ethology in chapter 6
of Allen and Bekoff 1997. There is much to be said for the view that psychology is a branch of
biology, and it is also true that much explanation within psychology and ethology is func-
tional rather than causal/predictive, but the cases that concern me in this book do not fit Mil-
likan's framework, which works best for simple, subpersonal mechanisms.
2. This would effectively amount to a conception of everyday reasoning taking semantic,
rather than syntactic, validity as fundamental.
3. In fact, as I show in chapters 8 and 9, there is an even more fundamental reason why
nonlinguistic creatures cannot carry out the relevant reasoning. The type of practical syllo-
gism envisaged by Davidson involves taking one's own beliefs and desires as premises.
However, taking one's own thoughts as objects of thought in this way is only possible if
they have linguistic vehicles—intentional ascent requires semantic ascent, as I argue in sec-
tion 8.2. This is a further reason why the limited forms of protoinference available at the
nonlinguistic level (see chapter 7) are irrelevant to the inference-based conception of practi-
cal reasoning.
4. It is known that vervet monkeys are sensitive to whether other individuals in the
group are reliable sources of information about predators (Cheney and Seyfarth 1990).
5. As I will show in the next chapter, there is room at the nonlinguistic level for what I
term protoinference, but it is a long way from that required for the inference-based concep-
tion of rationality.
6. Of course, this is to a certain extent a stipulation as to how words are to be used, but
there is some justification for it in our ordinary ways of speaking. If I act completely ran-
domly to resolve a deadlock in a situation where I really cannot see any advantage in acting
one way rather than another, it would seem natural to describe this as my selecting a course
of action without actually deciding on it.
7. There should be no implication that, for example, ethical deontologists are not advo-
cating a method of decision-making. My comments should be understood as limited to "de-
cision-making" as it might plausibly be identified in a nonlinguistic context. It seems clear,
however, that making decisions on deontological grounds is not possible for creatures that
are not capable of making decisions on consequence-sensitive grounds. Moreover, as I shall
show in chapters 8 and 9, the type of reflection accompanying deontological decision-mak-
ing is not available to nonlinguistic creatures.
8. It is worth pointing out, though, that the normative criteria of level 2 rationality incor-
porate elements of both internal and external rationality, in the sense introduced in section
6.1. The requirement that the relevant instrumental belief be accurate is clearly an external
requirement. It is quite possible that an action will fail to qualify as rational in this external
sense, even though it clearly satisfies the second, internal criterion. Arguably this would be
enough for it to qualify as rational in our everyday sense—and, as I suggested, internal ra-
tionality is what is important for psychological explanation. In any case it is also possible to
apply the normative criteria appropriate for level 0 and level 1 rationality to level 2 behav-
iors. We can assess the types of instrumental reasoning involved in, say, tool construction
from the viewpoint of expected utility theory.
9. This at least is the interpretation favored by Heyes and Dickinson (1993). Their inter-
pretation, which is formulated in the context of a more overarching attack on ethological at-
tributions of prepositional attitudes in nonlaboratory conditions, has been challenged (Allen
and Bekoff 1995). One point Allen and Bekoff make is that the chick's behavior might be
underwritten by a more general instrumental belief to the effect that approaching objects
permits access to them, and there are good reasons why this belief might be robust in the
face of the chick's experiences with the mirror—it is not always rational to revise deep-
seated beliefs in the face of countervailing evidence. This point is well taken, and that may
well be the correct interpretation of the chick's behavior. But the more general point stands,
which is that instrumental beliefs involve a general sensitivity to contingencies, and where
no such sensitivity can plausibly be identified, the attribution of an instrumental belief must
be revoked. What Allen and Bekoff do, in effect, is offer a way that the behavior of the
chicks can be interpreted as suitably contingency-sensitive. For further discussion see
Heyes and Dickinson's (1995) reply to Allen and Bekoff.
Chapter 7
1. Fodor's model cannot be quite right as it stands. It is not (usually) the case that each
behavioral option will have only one outcome, and what the creature will have to compute,
for each behavioral option, are the likelihoods of the principal different outcomes that could
occur. Intuitively, a creature will need to consider not simply the most desirable outcome
that might be consequent on acting in a given way but also the less desirable and indeed
positively undesirable outcomes that might also occur. Each of these outcomes will have a
different utility. The decision-maker will then need to weight the likelihoods of each differ-
ent outcome by its desirability. The sum of these calculations will yield an expected utility
for that behavioral option. The final stage will be simply to select the behavioral option with
the greatest expected utility.
2. I shall say more about the particular type of inferences that might be deployed in
level 2 rationality in section 7.2.
3. In the next section I will look in more detail at the form these instrumental beliefs
might take—and at the inferences within which they might feature.
4. I am taking 'or' in its inclusive sense, according to which a disjunction remains true if
both its disjuncts are true. The exclusive sense of 'or' can be defined by adding to the inclusive
sense the further requirement that the two disjuncts not both be true. This requirement can it-
self be given a conditional reading. "Not-(P and Q)" is equivalent to "If P, then not-Q."
5. For an extended discussion of this view of negation see Frege 1918-19a.
6. For extended discussion of these and related matters see Sommers 1982 and Grice
1989. Bochvar 1981 provides a formal development of the idea that there are two funda-
mentally different types of negation.
7. And of course this also brings in modal notions, which, as I will show in the final
chapter, may well not be available at the nonlinguistic level.
8. This is what Geach has called the Latin prose theory of relative pronouns, on which,
for example, the sentence "Any man who owns a car drives it" is assimilated to "Any man,
if he owns a car, drives it." See Geach 1967/68.
9. The most radical proposal in this area is that singular causal claims of the form
"Event c caused event e" can be analyzed in terms of counterfactual conditional claims of
the type "Had event c not occurred, event e would not have occurred." A counterfactual
theory of causation is proposed in Lewis 1973.
10. These predictions are most frequently viewed as involving (non-truth-functional)
subjunctive conditionals (about what would or would not happen), rather than truth-func-
tional indicative conditionals. There is a useful discussion of causation and conditionals in
the first chapter of Mellor 1995.
Chapter 8
1. As I will show hereafter, these six respects do not exhaust Clark's conception of the
contribution that language can make to cognition.
2. I take it as uncontroversial that no system of communication can count as linguistic
unless it involves symbols. The characteristic of symbols is that they are arbitrary as op-
posed to iconic signals (see further in the main text hereafter).
3. The bee dances have also received attention from philosophers. See Bennett 1964.
4. The definition of iconic signals just posed is significantly different from that sug-
gested by Charles Sanders Peirce, who is usually credited with having first formulated the
distinction between iconic and arbitrary signals (or rather, in his terms, between iconic and
symbolic signs). Peirce defines an icon as a "Sign that represents its Object in resembling it"
(1991, 270). He does not impose any requirement that continuous variation in the object be
matched by continuous variation in the signal, and consequently takes as iconic signs that
would not count as iconic on the criteria just outlined. For example, he holds that linguistic
predicates are iconic as well as logical and mathematical proofs. Peirce's theory of signs is
discussed in chapters 4 and 6 of Hookway 1985.
5. Consider, for example, the following two passages from Michael Dummett's book on
Frege. "Our apprehension of reality as decomposable into discrete objects is the product of
our application to an originally unarticulated reality of the conceptual apparatus embodied
in our language" (Dummett 1973,505). And: "Our ability to discriminate, within reality, ob-
jects of any particular kind results from our having learned to use expressions, names or
general terms, with which are associated a criterion of identity which yields segments of re-
ality of just that shape: we can, in principle, conceive of a language containing names and
general terms with which significantly different criteria of identity were associated, and the
speakers of such a language would view the world as falling apart into discrete objects in a
different way from ourselves... For Frege, the world does not come to us articulated in any
way; it is we who, by the use of our language (or by grasping the thoughts expressed in that
language), impose a structure on it" (1973, 503-504).
6. Nothing I say is intended to be incompatible with the substantive claims made by the
theorists in the language of thought tradition. In particular, it may well be the case (as Fodor
suggests) that the understanding of public language sentences involves in some sense trans-
lating them into the language of thought. My point is simply that reflexive thinking (in the
sense in which I am understanding it, namely, as involving, e.g., the evaluation of epistemic
links between propositions) is directed at the public language sentences rather than at the
sentences in the language of thought that give their meaning.
7. The point was well put by Wittgenstein, who has plausible things to say about what
is going on when it seems to one that one is introspecting a thought that is not sententially
vehicled: "What happens when we make an effort—say in writing a letter—to find the right
expression for our thoughts?—This phrase compares the process to one of translating or de-
scribing: the thoughts are already there (perhaps were there in advance) and we merely look
for their expression. This picture is more or less appropriate in different cases.—But can't
all sorts of things happen here?—I surrender to a mood and the expression comes. Or a. pic-
ture occurs to me and I try to describe it. Or an English expression occurs to me and I try to
hit on the corresponding German one. Or I make a gesture and ask myself: What words cor-
respond to this gesture? And so on." (Wittgenstein 1953, sec. 335). For further discussion of
Wittgenstein's complex views on the relation between language and thought see Budd 1989,
chs. 5 and 6.
8. Peter Carruthers, who proposes that all domain-general cognition consists in the for-
mation and manipulation of linguistic representations at the level of what Chomsky terms
logical form, nonetheless thinks that we cannot be conscious of these "stripped down" lin-
guistic representations (Carruthers 1996, forthcoming). We are only conscious of sentences
with the full complement of natural language phonological and structural features from
which the level of logical form is an abstraction.
9. I will count mental models theory as a conception of pictorial models even though
the two notions cannot be straightforwardly mapped onto each other. It is true that mental
models, as proposed by Johnson-Laird, are intended to be semiperceptual states that resem-
ble the situations they represent. Nonetheless, there is a crucial ambiguity in mental models
theory. On the one hand the manipulation of mental models is supposed to take place in
working memory, the contents of which are generally thought to be open to conscious ac-
cess and report. On the other hand it seems clear that introspection will have little role to
play in deciding the issue between mental logic and mental models as far as the psychology
of reasoning is concerned. To the extent that mental models theory is a theory of the subper-
sonal mechanisms of thought, it will be correspondingly of less use as a theoretical account
of how thoughts might be vehicled in a way that would allow them to be the objects of fur-
ther reflexive thinking.
10. The difference between the mental logic and mental models theories is frequently
compared to that between proof-theoretic and model-theoretic approaches in logic.
Chapter 9
1. The distinction between direct and reflective doxastic modification is linked to the
distinction drawn by Gilbert Harman between foundationalist and coherentist modes of be-
lief revision (see Harman 1986). Harman's book contains much useful discussion of the dif-
ferent principles operating in the two different modes of belief revision. He does not take a
view, however, on the relation between the coherentist/foundationalist distinction and the
divide between the linguistic and the nonlinguistic.
2. Frankfurt makes a further distinction between second-order desires and second-order
volitions, where a second-order volition is a desire not simply to have a particular first-order
desire but also for that first-order desire to be one's will. I will use the term "second-order
desire" to include second-order volitions.
3. As I noted in chapter 3, this type of second-order desire can be accommodated
within the general paradigm of success semantics. I suggested there that those second-order
desires that are targeted on specific first-order desires can be understood as states that cease
when the relevant first-order deskes come into being.
4. This is perfectly compatible with rejection of the epistemic theory of perception.
The proposal is not that the propositional report says all that there is to say about the con-
tent of perception. One might think that perceptual states carry information in an analogue
rather than digital manner, for example, while nonetheless thinking that the informa-
tion carried can be propositionally reported. Nor, of course, does the suggestion that the
propositional content of perception corresponds to the content of the perceptual belief to
which it might be expected to give rise imply that perceptions are nothing more than dis-
positions to acquire belief. The canonical statement of the epistemic theory is Armstrong
1961.
5. It is worth noting that, strictly speaking, perceptual reports can only have one of the
two characteristic features of propositional attitude reports. Clearly, the "that—•" clauses re-
porting the contents of perception are intensional, but since perception is factive (that is, one
can only perceive what is in fact the case), perceptual reports cannot be false. All this
means, of course, is that we sometimes have to refer to "seemings-to-perceive" rather than
to perceivings tout court.
6. The basic distinction here is made in various places, including Dretske 1969 and the
final chapter of Jackson 1975. In the characterisation of SS reports and ES reports I am
drawing on a more recent paper by Kevin Mulligan (1998).
7. As is frequently the case in the philosophy of perception, the modality of vision is
really what is under discussion.
8. Mulligan (1998) suggests that the characterization of simple seeing should employ
the machinery of tropes rather than properties.
9. Michael Tomasello (2000) has explicitly argued that (to put it in my terms) the ca-
pacity for intentional ascent is unique to humans and hence that the mind-reading interpreta-
tion of nonhuman apes cannot be correct. He does not, however, offer an argument for why
intentional ascent should not be available to nonhumans. The argument from intentional as-
cent to semantic ascent would no doubt be congenial to him.
10. As discussed in chapter 7, analogues of disjunction are available at the nonlinguistic
level—but these analogues do not employ anything like operators on propositions.
11. Of course, there are philosophers who think that the primary bearers of truth and fal-
sity are sentences rather than thoughts. This would provide an even quicker way of arguing
from truth-functional connectives to the need for semantic ascent.
12. I am taking 'extension' in a nontechnical sense here. But of course both modal and
tense logics are extensions of classical logic in the technical sense that they incorporate the
vocabulary of classical logic and have the same theorems and inferences involving only that
vocabulary.
13. The general approach has been challenged. See, for example, Evans 1985. Bui: these
challenges have not been accompanied by any alternative construals of tense logics.
14. My proposal bears certain resemblances to Donald's (1991) description of the Upper
Paleolithic in terms of mythic culture. See also the discussion of different ways of under-
standing time in Campbell 1994.
15. A similar proposal is made in Carruthers forthcoming, although on very different
grounds.
16. Geach has pointed out (1962, sees. 71-73) that this way of thinking about relative
clauses does not extend easily to certain tricky cases. I am inclined to respond to this as
Quine does (1974, 91) by distinguishing between the most plausible account of modem En-
glish and the most plausible account of the evolution of language.
References
Aiello, L. C.1996. Hominine preadaptations for language. In Mellars and Gibson 1996.
Allen, C., and M. Bekoff. 1995. Cognitive ethology and the intentionality of animal behav-
iour, Mind and Language 10, 313-328.
Allen, C., and M. Bekoff. 1997. Species of Mind. Cambridge, MA: MIT Press.
Armstrong, D. M. 1961. Perception and the Physical World. London: Routledge and Kegan
Paul.
Armstrong, D. M. 1962. Bodily Sensations. London: Routledge and Kegan Paul.
Armstrong, D. M. 1997. A World of States of Affairs.s.Cambridge: Cambridge Universityy
Press.
Baier, A. 1979. Mind and change of mind. Midwest Studies in Philosophy 4, 157-176.
Baillargeon, R. 1987. Object permanence in 3.5- and 4.5-month-old infants. Developmental
Psychology 23, 655-64.
Baillargeon, R. 1995. Physical reasoning in infancy. In Gazzaniga 1995.
Baker, G. P., and P. M. S. Hacker. 1984. Frege: Logical Excavations. Oxford: Oxford Uni-
versity Press.
Baker, L. R. 1995. Explaining Attitudes: A Practical Approach to the Mind. Cambridge:
Cambridge University Press.
Balda, R. P., and R. J. Turek. 1984. Memory in birds. In Roitblat, Bever, and Terrace
1984.
Barkow, J. H., L. Cosmides, and J. Tooby. 1992. The Adapted Mind: Evolutionary Psy-
chology and the Generation of Culture. Oxford: Oxford University Press.
Barwise, J., and J. Perry. 1983. Situations and Attitudes. Cambridge, MA: MIT Press.
Bechtel, W., and J. Abrahamsen. 1991. Connectionism and the Mind. Oxford: Blackwell.
Bekoff, M., C. Allen, and G. M. Burghardt. 2002. The Cognitive Animal. Cambridge, MA:
MIT Press.
Bell, D. 1987. Thoughts. Notre Dame Journal of Formal Logic 28, 36-50.
Bennett, J. 1964. Rationality. Indianapolis. Hackett.
Bennett, J. 1976. Linguistic Behaviour. Cambridge: Cambridge University Press.
Bermudez, J. L. 1994. Peacocke's argument against the autonomy of nonconceptual con-
tent. Mind and Language 9, 203-218.
Bermtidez, J. L. 1995a. Aspects of the self (John Campbell's Past, Space and Self), Inquiry
38, 1-15.
Bermiidez, J. L. 1995b. Nonconceptual content: From perceptual experience to subpersonal
computational states. Mind and Language 10, 333-369.
Bermudez, J. L. 1995c. Syntax, semantics and levels of explanation. Philosophical Quar-
terly 45, 361-367.
Bermudez, J. L. 1997. Practical understanding versus reflective understanding, Philosophy
and Phenomenological Research 67, 635-641.
Bermudez, J. L. 1998. The Paradox of Self-Consciousness. Cambridge, MA: MIT Press.
206 References
Bermudez, J. L. 1999. Naturalism and conceptual norms. Philosophical Quarterly 49,

77-85.
Bermudez, J. L. 2000a. Naturalized sense data. Philosophy and Phenomenological Re-
search 61, 353-374.
Bermtidez, J. L. 2000b. Personal and subpersonal: A difference without a distinction. Philo-
sophical Explorations 2, 63-82.
Bermudez, J. L. 2001a. Frege on thoughts and their structure, Philosophiegeschichte und
logische Analyse 4, 87-105.
Bermudez, J. L. 2001b. Normativity and rationality in delusional psychiatric disorders.
Mind and Language 16, 457^193.
Bermudez, J. L. 2001c. Nonconceptual self-consciousness and cognitive science. Synthese
129,129-149.
Bermudez, J. L. 2001d, Review of C. Peacocke, Being Known. Philosophical Psychology
14,250-256.
Bermudez, J. L. 2002a. Sources of self-consciousness. Proceedings of the Aristotelian Soci-
ety 102, 87-107.
Bemuidez, J. L. 2002b. Nonconceptual content. In The Macmillan Encyclopedia of Cogni-
tive Science.
Bermudez, J. L. Forthcoming-a. The domain of folk psychology. In O'Hear forthcoming.
Bernuidez, J. L. Forthcorning-b. Evans and the sense of T'. In Grush forthcoming.
Bermudez, J. L. Forthcoming-c. The Philosophy of Psychology: A Contemporary Introduc-
tion. London: Routledge.
Bermtidez, J. L., and M. E. Elton (Eds.). 2000. Personal and Subpersonal: Essays on Psy-
chological Explanation [Special issue]. Philosophical Explorations 3, 63-82.
Bermudez, J. L., and A. Millar (Eds.). 2002. Reason and Nature: Essays in the Theory of
Rationality. Oxford: Oxford University Press.
Bickerton, D. 1990. Language and Species. Chicago: Chicago University Press.
Bickerton, D. 1996. Language and Human Behaviour. London: UCL Press.
Binford, L. R. 1989. Isolating the transition to the cultural adaptations: An organizational
approach. In Trinkhaus 1989.
Black, A. H., and Prokasy, W. F. (Eds.) 1972. Classical Conditioning II: Current Research
and Theory. New York: Appleton Century Crofts.
Block, N. 1991. Imagery. Cambridge, MA: MIT Press.
Bochvar, D. A. 1981, On a three-valued logical calculus and its application to the analysis of
the paradoxes of the classical extended functional calculus. History and Philosophy
ofLogic2,87-ll2.
Boden, M. (Ed.). 1990. The Philosophy of Artificial Intelligence. Oxford: Oxford University
Press.
Boesch, C., and H. Boesch. 1992. Transmission aspects of tool use in wild chimpanzees. In
Ingold and Gibson 1992.
Bower, T. G. R. 1966. The visual world of infants. Scientific American 215, 80-92.
Bower, T. G. R. 1982. Development in Infancy. San Francisco: Freeman.
Bradbury, J. W., and S. L. Vehrencamp. 1998. Principles of Animal Communication. Sun-
dcrland, MA: Sinnauer.
Braddon-Mitchell, D., and J. Fitzpatrick. 1990. Explanation and the language of thought.
Synthese 83, 3-29.
Braddon-Mitchell, D., and F. Jackson. 1996. Philosophy of Mind and Cognition. Oxford:
Blackwell.
Braithwaite, R. B. 1932-33. The nature of believing. Proceedings of the Aristotelian Society
33,129-146.
References 207
Bremner, J. G. 1988. Infancy. Oxford: Blackwell.

Broome, J. 1991. Utility. Economics and Philosophy 1, 1-12.
Brown, C. R., M. B. Brown, and M. L. Shaffer. 1991. Food-sharing signals among socially
roraging cliff swallows. Animal Behavior 42, 551-564.
Brunswik, E. 1943. Organismic achievement and environmental probability. Psychological
Review 50, 255-272.
Budd, M. 1989. Wittgenstein's Philosophy of Psychology. London: Routledge.
Byrne, R. W. 1995. The Thinking Ape. Oxford: Oxford University Press.
Byrne, R. W., and A. Whiten (Eds.). 1988. Machiavellian Intelligence: Social Expertise and
the Evolution of Intellect in Monkeys, Apes and Humans. Oxford: Blackwell.
Call, J., and P. Rochat. 1996. Liquid conservation in orangutans (Pongo Pygmaeus) and hu-
mans {Homo sapiens)'. Individual differences and perceptual strategies. Journal of
Comparative Psychology 110, 219-232.
Call, J., and P. Rochat. 1997. Perceptual strategies in the estimation of physical quantities by
orangutans (Pongo Pygmaeus). Journal of Comparative Psychology 111, 315—329.
Campbell, J. 1993. The role of physical objects in spatial thinking. In N. Eilan, R. Mc-
Carthy, and B. Brewer (Eds.), Spatial Representation. Oxford: Oxford University
Press.
Campbell, J. 1994. Past, Space and Selff. Cambridge, MA: MIT Press.
Carey, S., and R. Gelman (Eds.). 1991. The Epigenesis of Mind: Essays on Biology and
Cognition. Hillsdale, NJ: Erlbaum.
Caron, A. J., R. F. Caron, and V. R. Carlson. 1979. Infant perception of the invariant shape
of objects varying in slant. Child Development 50, 716-721.
Carruthers, P. 1996. Language, Thought and Consciousness. Cambridge: Cambridge Uni-
versity Press.
Carruthers, P. 2000. Phenomenal Character. Cambridge: Cambridge University Press.
Carruthers, P., and J. Boucher (Eds.). 1998. Language and Thought: Interdisciplinary
Themes. Cambridge: Cambridge University Press.
Carruthers, P., and P. K. Smith. 1996. Theories of Theories of Mind. Cambridge: Cambridge
University Press.
Cerella, J. 1979. Visual classes and natural categories in the pigeon. Journal of Experimen-
tal Psychology: Human Perception and Performancee5, 68-77.
Cheney, D. L., and R. M. Seyfarth. 1990. How Monkeys See the World. Chicago: University
of Chicago Press.
Cheng, K. 1986. A purely geometrical module in the rat's spatial representation. Cognition
23, 149-178.
Child, W. 1994. Causality, Interpretation and the Mind. Oxford: Oxford University Press.
Chomsky, N. 1958. Review of B. F. Skinner's Verbal Behavior. Language 35, 26-58.
Chomsky, N. 1980. Rules and Representations. Oxford: Blackwell.
Churchland, P. M, 1989. The Neurocomputational Perspective: The Nature of Mind and the
Structure of Science. Cambridge, MA: MIT Press.
Churchland, P. S. 1986. Neurophilosophy: Towards a Unified Science of the Mind-Brain.
Cambridge, MA: MIT Press.
Churchland, P. S., and T. Sejnowski. 1992. The Computational Mind. Cambridge, MA: MIT
Press.
Clark, A. 1996. Dealing in futures: Folk psychology and the role of representations in cog-
nitive science. In McCauley 1996.
Clark, A. 1998. Magic words: How language augments human cognition. In Carruthers and
Boucher 1998.
Clark, A., and A. Karmiloff-Smith. 1993. The cognizer's innards: A psychological and
208 References
philosophical perspective on the development of thought. Mind and Language 8,

487-519.
Clayton, N. S., and A. Dickinson. 1999. Memory for the contents of caches by Scrub Jays.
Journal of Experimental Psychology: Animal Behavior Processes 25, 82-91.
Glutton-Brock, T. H., and S. D. Albon. 1979. The roaring of red deer and the evolution of
honest advertisement. Behaviour 69,145-170.
Cohen, L. J. 1992. An Essay on Belief and Acceptance. Oxford: Oxford University Press.
Corballis, M., and E. G. L. Lea. 1999. The Descent of Mind: Psychological Perspectives on
Hominid Evolution. Oxford: Oxford University Press.
Cosmides, L., and J. Tooby. 1994. Origins of domain-specificity: The evolution of func-
tional organization. In Hirschfeld and Gelman 1994.
Cowie, R. 1977. Optimal foraging in great tits (Parus Major). Nature 268,137-139.
Craik, K. 1967. The Nature of Explanation. Cambridge: Cambridge University Press.
Crane, T. (Ed.). 1990. The language of thought: No syntax without semantics. Mind and
Language 5, 187-212.
Crane, T. 1992. The Contents of Experience. Cambridge: Cambridge University Press.
Cussins, A. 1990. The connectionist construction of concepts. In Boden 1990.
Cussins, A. 1992. Content, embodiment and objectivity: The theory of cognitive trails. Mind
101,651-688.
Dancy, J. 1988. Perceptual Knowledge. Oxford: Oxford University Press.
Davidson, D. 1963. Actions, reasons and causes. Journal of Philosophy 60, 685-700.
Reprinted in Davidson 1980a.
Davidson, D. 1967a. Causal relations. Journal of Philosophy 64, 691-703. Reprinted in
Davidson 1980a.
Davidson, D. 1967b. Truth and meaning. Synthese 17, 304-323. Reprinted in Davidson
1980b.
Davidson, D. 1969. The individuation of events. In N. Rescher (Ed.). Essays in Honor of
Carl G. Hempel. Dordrecht: Reidel. Reprinted in Davidson 1980a.
Davidson, D. 1970. Mental events. In L. Foster and J. W. Swanson, Experience and Theory.
Amherst: University of Massachusetts Press. Reprinted in Davidson 1980a.
Davidson, D. 1975. Thought and talk. In S. Guttenplan (Ed.), Mind and Language. Oxford:
Oxford University Press. Reprinted in Davidson 1980b.
Davidson, D. 1978. Intending. In Y. Yovel (Ed.), Philosophy of History and Action. Dor-
drecht: Reidel. Reprinted in Davidson 1980a.
Davidson, D. 1980a. Essays on Actions and Events. Oxford: Oxford University Press.
Davidson, D. 1980b. Essays on Truth and Interpretation. Oxford: Oxford University Press.
Davis, S. (Ed.). 1992. Connectionism: Theory and Practice. New York: Oxford University
Press.
Dawkins, M. S. 1986. Unravelling Animal Behaviour. Harlow, UK: Longman.
Dennett, D. 1979. The Intentional Stance. Cambridge, MA: MIT Press.
Dennett, D. 1981. How to change your mind. In Brainstorms. Brighton, UK: Harvester
Press.
Dennett, D. 1996. Kinds of Minds. New York: Basic Books.
De Sousa, R. 1971. How to give a piece of your mind: Or, the logic of belief and assent.
Review of Metaphysics 25, 52-79.
Devitt, M. 1990. A narrow representational theory of the mind. In Lycan 1990.
Dickinson, A. 1980. Contemporary Animal Learning Theory. Cambridge: Cambridge Uni-
versity Press.
Dickinson, A., and B. Balleine. 1993. Actions and responses: The dual psychology of be-
References 209
haviour. In N. Eilan, B. Brewer, and R. McCarthy (Eds,), Spatial Representation.

Oxford: Blackwell.
Dickinson, A., and D. Shanks. 1995. Instrumental action and causal representation. In Sper-
ber, Premach, and Premach 1995.
Donald, M. 1991. Origins of the Modern Mind. Cambridge, MA: Harvard University Press.
Dretske, F. 1969, Seeing and Knowing. London: Routledge.
Dretske, F. 1981. Knowledge and the Flow of Information. Cambridge, MA: MIT Press.
Dretske, F. 1988. Reasons in a World of Causes. Cambridge, MA: MIT Press.
Dummett, M. 1973. Frege: Philosophy of Language. London: Duckworth.
Dummett, M. 1989. Language and communication. In A. George (Ed.), Reflections on
Chomsky. Oxford: Blackwell.
Dummett, M. 199la. More on thoughts. In Frege and Other Philosophers. Oxford: Oxford
University Press,
Dummett, M. 1991b. Thought and perception: The views of two philosophical innovators.
In Frege and Other Philosophers. Oxford: Oxford University Press.
Dummett, M. 1993. The Origins of Analytical Philosophy. London: Duckworth.
Evans, G. 1975. Identity and predication. Journal of Philosophy 72, 343-363.
Evans, G. 1981. Understanding demonstratives. In Parret 1981.
Evans, G. 1982. The Varieties of Reference. Oxford: Oxford University Press.
Evans, G. 1985. Is tense logic based on a mistake? In Collected Papers. Oxford: Oxford
University Press.
Evans, W. E. 1973. The echolocation by marine delphids and one species of freshwater dol-
phin. Journal of the Acoustic Society of America 54, 191-204.
Field, J. 1976. Relation of young infants' reaching behaviour to stimulus distance and so-
lidity. Developmental Psychology 12, 444-448.
Fisette, D. 1998. Consciousness and Intentionality: Models and Modalities of Attribution.
Derdrecht: Kluwer.
Fodor, J. 1975. The Language of Thought. Cambridge, MA: Harvard University Press.
Fodor, J. 1983. The Modularity of Mind. Cambridge, MA: MIT Press.
Fodor, J. 1986. Why paramecia don't have mental representations. Midwest Studies in Phi-
losophy 10, 3-23.
Fodor, J. 1987. Psychosemantics. Cambridge, MA: MIT Press.
Fodor, J. 2000. The Mind Doesn 't Work That Way. Cambridge, MA: MIT Press.
Fodor, J., and B. P. McLaughlin. 1995. Connectionism and the problem of systematicity:
Why Smolensky's solution doesn't work. In Macdonald and Macdonald 1995.
Fodor, J., and Z. Pylyshyn. 1988. Connectionism and cognitive architecture: A critical
analysis. Cognition 28, 3-71.
Frankfurt, H. 1971. Freedom of the will and the concept of a person. Philosophical Review
68, 5-20.
Prankish, K. 1996. Natural language and virtual belief. In P. Carruthers and P. K. Smith
(Eds.), Theories of Theories of Mind. Cambridge: Cambridge University Press.
Frege, G. 1918-19b. Negation. Translated by P. Geach and R. H. Stoothoff in B. McGuiness
(Ed.), Collected Papers on Mathematics, Logic, and Philosophy. Oxford: Blackwell.
(1984).
Frege, G. 1918-19a. Thoughts. Translated by P. Geach and R. H. Stoothoff in B. McGuiness
(Ed.), Collected Papers on Mathematics, Logic, and Philosophy. Oxford: Blackwell.
(1984).
Frisch, K. von. 1967. The Dance Language and Orientation of Bees. Cambridge, MA: Har-
vard University Press.
210 References
Gallese, V., and A. Goldman. 1998. Mirror neurons and the simulation theory of mind-
reading. Trends in Cognitive Science 3, 493-501.
Gallistel, C. R. 1990. The Organization of Learning. Cambridge, MA: MIT Press,
Gallistel, C. R. 2000. The replacement of general purpose learning models with adaptively
specialised learning modules. In Gazzaniga 2000.
Gallistel, C. R., A. L. Brown, S. Carey, R, Gelman, and F. C. Keil. 1991. Lessons from
animal learning for the study of cognitive development. In Carey and Gelman
1991.
Gazzaniga, M. S. (Ed,). 1995. The Cognitive Neurosciences. Cambridge, MA: MIT Press.
Gazzaniga, M. S. (Ed.). 2000. The New Cognitive Neurosciences. Cambridge, MA: MIT
Press.
Geach, P. T. 1962. Reference and Generality. Ithaca: Cornell University Press.
Geach, P. T. 1967/68. Identity. Review of Metaphysics 21, 3-12.
Gibson, J. J. 1979. The Ecological Approach to Visual Perception. Boston: Houghton
Mifflin.
Gibson, K. R., and T. Ingold. 1993. Tools, Language and Cognition in Human Evolution.
Cambridge: Cambridge University Press.
Goldin-Meadow, S. 1979. Structure in a manual communication system developed without
a language model: Language without a helping hand. In H. A. Whitaker (Ed.), Stud-
ies in Neurolinguistics, Vol. 4. New York: Academic Press.
Gomez, J. L. 1996. Non-human primate theories of (non-human primate) Minds: Some is-
sues concerning the origins of mind-reading. In P. Carruthers and P. K. Smith (Eds.),
Theories of Theories of Mind. Cambridge: Cambridge University Press.
Gopnik, A., and A. Meltzoff. 1997. Thoughts, Theories and Things. Cambridge, MA: MIT
Press.
Gould, J. L., and C. J. Gould. 1994. The Animal Mind. New York: Scientific American
Library.
Gould, J. L., and C. J. Gould. 1998. Reasoning in animals. Scientific American 9, 52-59.
Gould, S. J., and R. Lewontin. 1979. The spandrels of San Marco and the panglossian para-
digm: A critique of the adaptationist programme, Proceedings of Royal Society of
London B 205, 581-598.
Green, S., and P. M. Marler, 1979. The analysis of animal communication. In P. M. Marler
and J. C. Vanderbergh (Eds.), Handbook of Behavioral Neurobiology: Vol 3. Social
Behavior and Communication. New York: Plenum Press.
Greene, E. 1987. Individuals hi an osprey colony discriminate between high and low quality
information. Nature, 329,239-241.
Greenfield, P. M. 1991. Language, tools and brain: The ontogeny and phytogeny of hierar-
chically organised sequential behavior. Behavioral and Brain Sciences 14, 531-551.
Greenfield, P. M., and E. S. Savage-Rumbaugh. 1990. Grammatical combination in pan
paniscus: Processes of learning and invention in the evolution and development of
language. In S. Taylor-Parker and K. R. Gibson (Eds.), Language and Intelligence in
Monkeys and Apes. Cambridge: Cambridge University Press.
Grice, H. P. 1989. Studies in the Ways of Words. Cambridge, MA: Harvard University
Press.
Griffith, C. R. 1943. Principles of Systematic Psychology. Urbana: University of Illinois
Press.
Griffin, D. 2001. Animal Minds: From Cognition to Consciousness. Chicago: University o
Chicago Press.
Grush, R. Forthcoming. Essays on Gareth Evans. New York: Oxford University Press.
Guttenplan, S. (Ed.). 1975. Mind and Language. Oxford: Oxford University Press.
References 211
Hale, B., and C. Wright. 1997. A Companion to the Philosophy of Language. Oxford:
Blackwell.
Hammond, L. J. 1980. The effect of contingencies upon appetitive conditioning of tree-
operant behavior. Journal of the Experimental Analysis of Behavior 34, 297—304.
Hare, B., J. Call, B. Agnetta, and M. Tomasello, 2000. Chimpanzees know what con-
specifics do and do not see. Animal Behaviour 59,771-785.
Hare, B., J. Call, M. Tomassello. 2001. Do chimpanzees know what conspecifics know?
Animal Behaviour 61, 139-151.
Harman, G. 1972. Logical form. Foundations of Language 9, 38-65.
Harman, G. 1986. Change in View. Cambridge, MA: MIT Press.
Hauser, M. D. 1997. Artifactual kinds and functional design features: What a primate under
stands without language. Cognition 64, 285-308.
Hauser, M. D. 1998. Expectations about object motion and destination: Experiments with a
nonhuman primate. Developmental Science 1, 31-38.
Hauser, M. D. 2000. Wild Mind: What Animals Really Think. London: Penguin Books.
Hauser, M. D. 2001. Searching for food in the wild: A nonhuman primate's expectations
about invisible displacement. Developmental Science 4, 84-93.
Hauser, M. D., and P. Marler. 1993. Food calls in rhesus macaques (Macaco Mulata), I & D
Behavioral Ecology 4,194-205, 206-212.
Hauser, M. D., T. Williams, J. D. Kralik, and D. Moscovitz. 2001. What guides a search for
food that has disappeared: Experiments on cotton-top tamarins. Journal of Com-
parative Psychology.
Heinrich, B. 2000. Testing insight in ravens. In C. Heyes and L. Huber (Eds.), The Evolution
of Cognition. Cambridge, MA: MIT Press.
Heller, M. 1997. The Ontology of Physical Objects. Cambridge: Cambridge University Press.
Herman, L. M. 1980. Cognitive characteristics of dolphins. In L. M. Herman (Ed.),
Cetacean Behavior: Mechanisms and Functions. New York: Wiley.
Herman, L. M. 1986. Cognition and language competencies of bottlenosed dolphins. In
Schusterman, Thomas, and Woods 1986.
Herrnstein, R. J., and D. H. Loveland. 1964. Complex visual concepts in the pigeon. Sci-
ence 146,549-551.
Herrnstein, R. J., D. H. Loveland, and C. Cable. 1976. Natural concepts in pigeons. Journal
of Experimental Psychology: Animal Behaviour Processes 2, 285-302.
Hershberger, W. A. 1986. An approach through the looking-glass. Animal Learning and Be-
havior 14,443^*51.
Heyes, C., and A. Dickinson. 1993. The intentionality of animal action. In Davies 1993.
Heyes, C., and A. Dickinson. 1995. Folk psychology won't go away: Response to Allen and
Bekoff, Mind and Language 10, 329-332.
Heyes, C., and B. G. Galef. 1996. Social Learning in Animals: The Roots of Culture. New
York: Academic Press.
Heyes, C., and L. Huber. 2000. The Evolution of Cognition. Cambridge, MA: MIT Press.
Hintikka, J. 1983. The Game of Language. Dordrecht: Reidel.
Hirschfeld, L. A., and S. A. Gelman. 1994. Mapping the Mind: Domain-Specificity in Cog-
nition and Culture. Cambridge: Cambridge University Press.
Hood, B., and P. Willats. 1986. Reaching in the dark to an object's remembered position:
Evidence for object permanence in 5-month-old infants. British Journal of Develop-
mental Psychology 4, 57-65.
Hookway, C. 1985. Peirce. London: Routledge.
Horgan, T, and J. Tienson. 1992. Structured representations in connectionist systems. In
Davis 1992.
212 References
Morgan, T., and J. Tienson. 1996. Connectionism and the Philosophy of Psychology. Cam-
bridge, MA: MIT Press.
Hornsby, J. 1997.Simple MindenessCambridge, Ma:Mit Pre
Ingold, T., and K. R. Gibson. 1992. Tools, Language and Intelligence: Evolutionary Impli-
cations. Oxford: Oxford University Press.
Jackson, F. 1977. Perception. Cambridge: Cambridge University Press.
Jacob, P. 1997. What Minds Can Do: Intentionality in a Non-intentional World. Cambridge:
Jeffrey, R. 1990. The Logic of Decision. Rev. 2nd ed. Chicago: University of Chicago Press.
Johnson-Laird, P. 1983. Mental Models. Cambridge: Cambridge University Press.
Johnson-Laird, P. 1999. Mental models. In Wilson and Keil 1999.
Johnson-Laird, P., and R. M. J. Byrne. 1991. Deduction. Hillsdale, NJ: Erlbaum.
Kacelnik, A. 1984. Central place foraging in starlings (Sturnus vulgaris). Journal of Animal
Ecology 53, 283-299.
Kaplan, D. 1989. Demonstrations. In J. Almong et al. (Eds.), Themes from Kaplan. Oxford:
Oxford University Press.
Karmiloff-Smith, A. 1993. Beyond Modularity. Cambridge, MA: MIT Press.
Kastak, C. R., R. J. Schusterman, and D. Kastak. 2001. Equivalence classification by Cali-
fornia sea lions using class-specific reinforcers, Journal of the Experimental Analy-
sis ofBehaviour 76, 131-158.
Kim, J. 1984. Supervenient and epiphenomenal causation. In Kim 1993.
Kim, J. 1993. Supervenience and Mind. Cambridge: Cambridge University Press.
Kohler, W. 1925. The Mentality of Apes. New York: Harcourt Brace.
Krebs, J. R., and N. B. Davies. 1991. Behavioural Ecology: An Evolutionary Approach.
Oxford: Blackwell Scientific.
Krebs, J. R., and A. Kacelnik. 1991. Decision-making. In Krebs and Davies 1991.
Lea, S. E. G. 1984. Instinct, Environment and Behaviour. London: Methuen.
Leslie, A. M. 1982. The perception of causality in infants. Perception 11, 173-186.
Leslie, A. M. 1984. Infant perception of a manual pick-up event. British Journal of Devel-
opmental Psychology 2, 19-32.
Lewis, D. 1969. Convention. Cambridge, MA: Harvard University Press.
Lewis, D. 1972. Psychophysical and theoretical identifications. Australasian Journal of Phi-
losophy 50, 249-258.
Lewis, D. 1973. Causation. Journal of Philosophy 70, 556-567.
Lewis, D. 1983. Language and languages. In Philosophical Papers, Vol. 1. Cambridge:
Lieberman, P. 1984. The Biology and Evolution of Language. Cambridge, MA: Harvard
University Press.
Lowe, E. J. 1993. Rationality, deduction and mental models. In Manktelow and Over
1993.
Luce, R. D., and H. Raiffa. 1957. Games and Decisions: Introduction and Critical Survey.
New York: Wiley,
Lycan, W. G. (Ed.). 1990. Mind and Cognition. Oxford: Blackwell.
McCauley, R. N. 1996. The Churchlands and Their Critics. Oxford: Blackwell.
Macdonald, C. 1995. Introduction: Classicism versus connectionism. In Macdonald and
Macdonald 1995.
Macdonald, C., and G. Macdonald. 1995. Connectionism: Debates in Psychological Expla-
nation. Oxford: Blackwell.
McDowell, J. 1994. Mind and World. Cambridge, MA: MIT Press.
References 213
McKenzie, B. E., and R. H. Day, 1972, Object distance as a determination of visual fixation
in early infancy. Science 178, 1108-1110.
McKenzie, B. E., H. E. Tootell, and R. H. Day. 1980. Development of visual size con-
stancy during the first year of human infancy. Developmental Psychology 16,
163-174.
Mackie, J. L. 1965. Causes and conditions, American Philosophical Quarterly 2, 245
264.
Mackintosh, N. 1983. Conditioning and Associative Learning. Oxford: Oxford University
Press.
McLeod, P., K. Plunkett, and E. T. Rolls. 1998. Introduction to Connectionist Modelling o
Cognitive Processes. Oxford: Oxford University Press.
Makin, G. 2000. The Metaphysicians of Meaning: Frege and Russell on Sense and Denota-
tion. London: Routledge.
Malcolm, N. 1977. Thoughtless brutes. In Thought and Knowledge. Ithaca, NY: Cornell
University Press.
Maloney, J. J. 1989. The Mundane Matter of the Mental Language. Cambridge: Cambridg
University Press.
Manktelow, K. L, and D. E. Over (Eds.). 1993. Rationality: Psychological and Philosophi
cal Perspectives. London: Routledge.
Marler, P. 1970. A comparative approach to vocal learning: Song development in white
crowned sparrows. Journal of Comparative and Physiological Psychology 71 (Sup
plement), 1-25.
Marler, P. 1991. The instinct to learn. In Hirschfeld and Gelman 1994.
Maynard Smith, J. 1982. Evolution and the Theory of Games. Cambridge: Cambridge Uni-
versity Press.
Mellars, P. 1996. Symbolism, language and the Neanderthal mind. In Mellars and Gibson
1996.
Mellars, P., and K. Gibson. 19%. Modelling Early Human Minds. Cambridge: McDonald
Institute Monographs.
Mellor, D. H. 1991. Matters of Metaphysics. Cambridge: Cambridge University Press.
Mellor, D. H. 1995. The Facts of Causation. London: Routledge.
Meltzoff, A. N., and M. K. Moore. 1977. Imitation of facial and manual gestures by huma
neomnates. Science 198, 75-78.
Miles, H, L. 1990. Cognitive foundations for reference in a signing orangutan. In Taylor-
Parker and Gibson 1990.
Millikan, R. 1984. Language, Thought and Other Biological Categories. Cambridge, MA:
MIT Press.
Millikan, R. 1986. Thought without laws. Philosophical Review 95,47-80.
Mitchell, R., and N. Thompson. 1986. Deception. Buffalo, NY: SUMY Press.
Mithen, S. 1990. Thoughtful Foragers: A Study of Prehistoric Decision-Making. Ca -
bridge: Cambridge University Press.
Mithen, S. 1996. The Prehistory of the Mind. London: Thames and Hudson.
M011er, A. P. 1988. False alarm calls as a means of resource usurpation in the great tt Parus
Major. Ethology 79, 25-30.
Montague, R. 1974. Formal Philosophy. New Haven: Yale University Press.
Moore, M. K., R. Borton, and B. L. Darby. 1978. Visual tracking in young infants: Evidenc
for object identity or object permanence? Journal of Experimental Child Psychology
25, 183-198.
Morgan, C. L. 1894. An Introduction to Comparative Psychology. London: Walter Scott.
214 References
Morgan, C. L. 1903. An Introduction to Comparative Psychology. New rev. ed. London:

Walter Scott.
Morton, A. 2002. The Importance of Being Understood: Folk Psychology as Ethics. Lon-
don: Routledge.
Mulligan, K. 1998. Perception, particulars and predicates. In Fisette 1998.
Munakata, Y, L. R. Santos, E. S. Spelke, M. D. Hauser, and R. C. O'Reilly. 2001. Visual
representation in the wild: How rhesus monkeys parse objects. Journal of Cognitive
Neuroscience 13, 44-58.
Munn, C. A. 1986. The deceptive use of alarm calls by sentinel species in mixed-species
flocks of neotropical birds. In Mitchell and Thompson 1986.
Nagel, T. 1979. What is it like to be a bat? In Mortal Questions. Cambridge: Cambridge
University Press.
O'Hear, A. (Ed.). Forthcoming. Minds and Persons. Cambridge: Cambridge University
Press.
O'Keefe, J., and L. Nadel (Eds.). 1978. The Hippocampus as a Cognitive Map. Oxford: Ox-
ford University Press.
Panaccio, C. 1992. Les mots, les concepts et les chases. Paris: Vrin.
Parrel, H. 1981. Meaning and Understanding. Berlin: De Gruyter.
Peacocke, C. 1979. The problem of the essential indexical. Nous 13, 3-21.
Peacocke, C. 1983. Sense and Content. Oxford: Oxford University Press.
Peacocke, C. 1986. Analogue content. Proceedings of the Aristotelian Society Suppl. Vol.
60, 1-17.
Peacocke, C. 1992. A Study of Concepts. Cambridge, MA: MIT Press.
Peacocke, C. 1999. Being Known. Oxford: Oxford University Press.
Peirce, C. S. 1991. Peirce on Signs: Writings on Semiotics. Chapel Hill: University of
Carolina Press.
Perry, J. 1977. Frege on demonstratives. Philosophical Review 86,474 - 497.
Pettit, P. 1993. The Common Mind: An Essay on Psychology, Society and Politics. Oxford:
Oxford University Press.
Piaget, J. 1952. The Origins of Intelligence in the Child. New York: Basic Books.
Piaget, J. 1954. The Construction of Reality in the Child. New York: Basic Books.
Pinker, S. 1994. The Language Instinct. London: Penguin Books.
Poole, J., and D. G. Lander. 1971. The pigeon's concept of pigeon. Psychonomic Science
25,157-158.
Povinelli, D. J. 1996. Chimpanzee theory of mind. In Carruthers and Smith 1996.
Povinelli, D. J., and T. J. Eddy. 1996a. Chimpanzees: Joint visual attention. Psychological
Science 1, 129-135.
Povinelli, D. J., and T. J. Eddy. 1996b. Factors influencing young chimpanzees' recognition
of attention. Journal of Comparative Psychologyy110, 336-345..
Premack, D., and A. J, Premack. 1983. The Mind of an Ape. Hillsdale, NJ: Erlbaum.
Premack, D., and G. Woodruff. 1978. Does the chimpanzee have a theory of mind? Behav-
ioural and Brain Sciences 1, 515-526.
Prior, A. N. 1968. Papers on Time and Tense. Oxford: Clarendon Press.
Quine, W. V. O. 1960. Word and Object. Cambridge, MA: MIT Press.
Quine, W. V. O. 1974. The Roots of Reference. Cambridge, MA: Harvard University Press.
Quine, W. V. 0.1995. From Stimulus to Science. Cambridge, MA: Harvard University Press.
Ramsey, F. P. 1927[1978]. Facts and propositions. In Foundations: Essays in Philosophy,
Logic, Mathematics and Economics. London: Routledge.
Ramsey, W. 1992. Connectionism and the philosophy of mental representation. In Davis
1992.
References 215
Rescorla, R. A., and J. C. Skucy. 1969. Effect of response-independent reinforcers during

extinction. Journal of Comparative and Physiological Psychology 67, 381-389.
Rescorla, R. A., and A. R. Wagner. 1972. A theory of Pavlovian conditioning: variations in
the effectiveness of reinforcement and non-reinforcement. In Black and Prokasy
1972.
Rey, G. 1997. Contemporary Philosophy of Mind. Oxford: Blackwell.
Reynolds, P. C. 1993. The complementation theory of language and tool use. In Gibson and
Ingold 1993.
Rips, L. 1994. The Psychology of Proof. Cambridge, MA: MIT Press.
Ristau, C. A. 1991. Aspects of the cognitive ethology of an injury-feigning bird, the piping
plover. In Ristau 1991.
Ristau, C. A. 1991. Cognitive Ethology: The Minds of Other Animals. Hillsdale, NJ: Erlbaum.
Rizley, R. C., and R. A. Rescorla. 1972. Associations in second-order conditioning and sen-
sory preconditioning, Journal of Comparative and Physiological Psychology 81,
1-11.
Rizzolati, G., L. Fogassi, and V. Gallese. 2000. Cortical mechanisms subserving object
grasping and action recognition: A new view on the cortical motor functions. In
Gazzaniga 2000.
Roitblat, H. L., T. G. Bever, and H. S. Terrace. 1984. Animal Cognition. Hillsdale, NJ:
Erlbaum.
Rosenthal, D. M. 1986. Two concepts of consciousness. Philosophical Studies 49, 329-359.
Rosenthal, D. M. 1993. Thinking that one thinks. In M. Davies and G. W. Humphreys
(Eds.), Consciousness. Oxford: Blackwell.
Rumelhart, D. E., and J. L. McClelland. 1986. On learning the past tenses of English verbs.
In Rumelhart, McClelland, and the PDP Research Group 1986b.
Rumelhart, D. E., J. L. McClelland, and the PDP Research Group. 1986a. Parallel Dist-
ributed Processing: Explorations in the Microstructure of Cognition: Vol. 1. Founda-
tions. Cambridge, MA: MIT Press.
Rumelhart, D. E., J. L. McClelland, and the PDP Research Group. 1986b. Parallel Dis-
tributed Processing: Explorations in the Microstructure of Cognition: Vol. 2. Psy-
chological and Biological Models. Cambridge, MA: MIT Press.
Ryle, G. 1949. The Concept of Mind. New York: Barnes and Noble.
Sainsbury, R. M. 2002. Departing from Frege. London: Routledge.
Samuelson, P. 1938. A note on the pure theory of consumers' behaviour, Economica 5,
61-71.
Savage-Rumbaugh, E. S. 1986. Ape Language: From Conditioned Response to Symbol.
New York: Columbia University Press.
Savage-Rumbaugh, E. S., S. Shanker, and T. Taylor. 1998. Apes, Language and the Human
Mind. Oxford: Oxford University Press.
Schusterman, R. J., J. A. Thomas, and F. G. Woods (Eds.). 1986. Dolphin Cognition and
Behaviour: A Comparative Approach. Hillsdale, NJ: Erlbaum.
Schusterman, R. J., C. R. Kastak, and D. Kastak. 2002.The cognitive sea lion: Meaning and
memory in the lab and in nature. In Bekoff, Allen, and Burghardt 2002.
Sellars, W. 1956/1997. Empiricism and the Philosophy of Mind. Cambridge, MA: Harvard
University Press.
Shephard, R. 1982. Mental Images and Their Tranformations. Cambridge, MA: MIT Press.
Sherry, D. F. 1982. Food storage, memory and marsh tits. Animal Behaviour 30, 631-633.
Sherry, D. F. 1984. Food storage by black-capped chickadees: Memory for the location and
content of caches. Animal Behaviour 32, 451^64.
Shettleworth, S. J., and J. R. Krebs. 1986. Stored and encountered seeds: A comparison of
216 References
two spatial memory tasks. Journal of Experimental Psychology: Animal Behaviour

Processes 12, 248-256.
Skinner, B. F. 1957. Verbal Behaviour. New York: Appleton-Century-Crofts.
Skryms, B. 1996. Evolution of the Social Contract. Cambridge: Cambridge University
Press.
Slater, A. M., and V. Morrison. 1985. Shape constancy and slant perception at birth. Percep-
tion 14, 337-344.
Sluga, H. 1980. Gottlob Frege. London: Routledge Kegan Paul.
Smolensky, P. 1988. On the proper treatment of connectionism. Behavioral and Brain Sci-
ences 11, 1-74. Reprinted in Macdonald and Macdonald 1995.
Smolensky, P. 1991. Connectionism, constituency and the language of thought. In Loewer
andRey 1991.
Smolensky, P. 1995. Constituent structure and explanation in an integrated connectionist/
symbolic cognitive architecture. In Macdonald and Macdonald 1995.
Sommers, F. 1982. The Logic of Natural Language. Oxford: Clarendon Press.
Spelke, E. S. 1990. Principles of object perception. Cognitive Science 14, 29-56.
Spelke, E. S., C. Hofsten, and R. Kestenbaum. 1989. Object perception and object-directed
reaching in infancy: Interaction of spatial and kinetic information for object bound-
aries. Developmental Psychology 25,185-196.
Spelke, E. S., and G. A. Van de Valle. 1993. Perceiving and reasoning about objects: In-
sights from infants. In N. Eilan, R. McCarthy, and B. Brewer (Eds.), Spatial Repre-
sentation. Oxford: Blackwell.
Sperber, D., D. Premach, and A. J. Premack (Eds.). 1995. Causal Cognition. New York: Ox-
ford University Press.
Squire, L. R., B. Knowlton, and G. Musen. 1993. The structure and organisation of memory.
Annual Review of Psychology 44,453-^496.
Stalnaker, R. 1987. Inquiry. Cambridge, MA: MIT Press.
Stein, E. 1996. Without Good Reason: The Rationality Debate in Cognitive Science. Oxford:
Clarendon Press.
Stephens, D. W., and J. R. Krebs. 1986. Foraging Theory. Princeton, NJ: Princeton Univer-
sity Press.
Sterelny, K. 1990. A Representational Theory of Mind. Oxford: Blackwell.
Stern, D. 1985. The Interpersonal World of the Infant. New York: Academic Press.
Stich, S, 1978. Beliefs and subdoxastic states. Philosophy of Science 45,499-518.
Strawson, P. F. 1959. Individuals. London: Methuen.
Stromswold, K. 2000. The cognitive neuroscience of language acquisition. In Gazzaniga
2000.
Taylor, C. 1964. The Explanation of Behaviour. London: Routledge Kegan Paul.
Taylor-Parker S., and K. R. Gibson. 1990. Language and Intelligence in Monkeys and Apes.
Cambridge: Cambridge University Press.
Terrace, H. S., L. A. Pettito, F. J. Sanders, and T. G. Bever. 1979. Can an ape create a sen-
tence? Science 206, 891-900.
Tinbergen, N. 1951. The Study of Instinct. Oxford: Oxford University Press.
Tinbergen, N. 1973. The Animal in Its World. Cambridge, MA: Harvard University Press.
Tolman, E. C., and C. H. Honzik. 1930a. "Insight" in rats. University of California Publica-
tions in Psychology 4, 215-232,
Tolman, E. C., and C. H. Honzik. 1930b. Introduction and removal of reward, and maze
learning in rats. University of California Publications in Psychology 4, 257-275.
Tolman, E. C., B. F. Ritchie, and D. Kalish. 1946. Studies in spatial learning II: Place learn-
ing versus response learning. Journal of Experimental Psychology 36, 221-229.
References 217
Tolman, E. C., B. F. Ritchie, and D. Kalish. 1947. Studies in spatial learning V: Response
learning versus place learning via the non-correction method. Journal of Experimen-
tal Psychology 37, 285-292.
Tomasello, M. 1996. Do apes ape? In Heyes and Galef 1996.
Tomasello, M. 2000. Two hypotheses about primate cognition. In Heyes and Huber 2000.
Travis, C. 1994. On constraints of generality. Proceedings of the Aristotelian Society 94,
165-188.
Treisman, A. 1996. The binding problem. Current Opinion in Neurobiology 6, 171-178.
Treisman, A., and H. Schmidt. 1982. Illusory conjunctions in the perception of objects.
Cognitive Psychology 14, 107-141.
Trinkhaus, E. (Ed.). 1989. The Emergence of Modem Humans: Biological Adaptations in
the Later Pleistocene. Cambridge: Cambridge University Press.
Turvey, M. T. 1996. Dynamic touch. American Psychologist 51, 1134-1152.
Tye, M. 1991. The Imagery Debate. Cambridge, MA: MIT Press.
Tye, M. 1992. Visual content and visual qualia. In T. Crane (Ed.), The Contents of Percep-
tion. Cambridge: Cambridge University Press.
Tye, M. 1995. Ten Problems of Consciousness. Cambridge, MA: MIT Press.
Van Gelder, T. 1990. Compositionality: A connectionist variation on a classical theme. Cog-
nitive Science 14, 355-384.
Vaughan, W., Jr. 1988. Formation of Equivalence Sets in Pigeons. Journal of Experimental
Psychology: Animal Behaviour Processes 14, 36-42.
Von Hofsten, C. 1982. Foundations for perceptual development. Advances in Infancy Re-
search 2, 241-261.
de Waal, F. 1982. Chimpanzee Politics: Power and Sex among Apes. London: Jonathan
Cape.
Walker, S. 1983. Animal Thought. London: Routledge and Kegan Paul.
Weiner, J. 1990. Frege in Perspective. Ithaca, NY: Cornell University Press.
Whyte, J. T. 1990. Success semantics. Analysis 50, 149-157.
Whyte, J. T. 1991. The normal rewards of success. Analysis, 51, 65-73.
Wiggins, D. 1997. Sortal concepts: A reply to Xu. Mind and Language 12, 413-421.
Wilson, R. A., and F. C. Keil. 1999. The MIT Encyclopedia of Cognitive Science. Cam-
bridge, MA: MIT Press.
Wimmer, H., and J. Perner. 1983. Beliefs about beliefs: Representation and constraining
function of wrong beliefs in young children's understanding of deception. Cognition
13, 103-128.
Wittgenstein, L. 1953. Philosophical Investigations. Translated by G. E. M. Anscombe. Ox-
ford: Blackwell.
Wright, C. 1997. The indeterminacy of translation. In Hale and Wright 1997.
Wynn, T. 1993. Layers of thinking in tool behavior. In Gibson and Ingold 1993.
Zeki, S. 1993. A Vision of the Brain. Oxford: Oxford University Press.
Index
Affect attunement, 152 Bonobos (pointing behavior in), 156

Affordances, 47, 50-51, 52-54, 90, 94, Borton, R., 80
121,122,123,139,197 n.10 Bower, T., 76-77, 81
Allen,C.,4,199n.l,200n.9 Bradbury, J. W., 119
Allocentric coordinate systems, 98-103 Braddon-Mitchell, D., 28, 29, 160, 196 n.12
Archeology, cognitive, 4-5, 35, 46, 109, Braithwaite, R. B., 184-185
189 Brunswik, E., 147
Argument from intelligent action, 56—59, Budd, M., 202, n.7
61 Byrne, R. M. J., 163
Argument from planning, 51-56 Byrne, R. W., 4, 55, 126, 175
Argument from psychological explanation,
60-62 Cable, C., 86
Aristotle, 18,142,143,145 Call, J., 37-38, 177
Armstrong, D. M., 40, 197 n.14, 203 n.4 Campbell, J., 45,47, 57, 59, 108, 199 n.5,
204 n. 14
Baboons (tactical deception in), 176 Carlson, V. R., 81
Baillargeon, R., 3, 54, 78, 80 Caron,A. J., 81
Baker, G. P., 195 n.l Caron, R. R, 81
Baker, L. R., 198 n.l8 Carruthers, P., 202 n.8, 204 n.15
Balda, R. P., 157 Causation. See Protocausation
Balleine.B., 55, 166, 195 n.l Cerella, J., 86
Barkow.J. H., 185 Chalmers, D., 197 n.3
Bee dances. See Honeybees Cheney, D. L., 157, 200 n.4
Begriffsschrift, 23, 38 Cheng, K., 103
Bekoff, M., 4, 199 n.l, 200 n.9 Chickadees (food caching), 48, 197 n.ll
Belief revision. See Doxastic modification Chimpanzees
Bell, D., 196 n.5 tool manufacture in wild, 55
Bennett, J., 50-51, 52, 155, 183, 184, 201 understanding of visual perspective,
n.3 177-178
Bermudez, J. L., 46, 47, 116, 192, 195 n.2, Chomsky, N., 155, 202 n.8
196 n.5,196 n.8, 197 n.2, 197 n.7,197 Churchland, P. S., 30, 196 n.13
n.8,197n.l3,198n.l3, 199 n.5 Clark, A., 151, 158,159, 166, 201 n.l
Bickerton, D., 154 Clark's nutcrackers (caching behavior in),
Binding problem, 73, 75 157
object perception and, 76-78 Clayton, N. S., 48
Blue tits (foraging behavior in), 115 Cognition, neural correlates of, 30-31
Bochvar, D. A., 201 n.6 Cognitive archeology. See Archeology,
Boesch, C., 154 cognitive
Boesch, H., 154 Cognitive ethology. See Ethology, cognitive
220 Index
Cognitive integration, 8 Eddy, T. J., 177

Cognitive trails (Cussins), 45 Egocentric coordinate systems, 98-103
Compositionality, 17, 23, 72, 150, 155, Elton, M., 197 n.2
187 Epistemic seeing, 174-175
Conditioning, 113 Equivalence classification, 113-114
classical, 7, 116, 166 Ethology, cognitive, 4, 5, 35, 46, 53-54, 71,
instrumental, 7, 55-56,123,124,166 109,189
Connectionism, 159 Evans, G., 71, 72,179,199 n.5, 204 n.l3
Consequence-sensitivity, 124,138-139, Evolutionary game theory, 120, 122
168. See also Outcome-sensitivity Eye-blink response, 116-117
Content, nonconceptual, ix-x, 41
Control loops, 151 Fitness, 118-119,121-122
Corballis, M., 4 inclusive, 118, 122, 136
Cosmides, L., 185 Fitzpatrick, J., 28, 29, 196 n.12
Cowie, R., 115 Fodor, J., 9-10, 12, 27, 68,134,138,139,
Craik, K. J. M., 160 159,185,196 n.6, 200 n.l. See also
Crane, T., 28, 29, 40, 173 Language of thought hypothesis
Cussins, A., 38, 44-45, 47, 57, 59, 108, Fogassi, L., 27
197 n.9 Force/content distinction, 15, 39,40,41,
105
Darby, B. L., 80 Frankfurt, H., 203, n.2
Davidson, D. 18,19-20,106,110-116, Frege, G., 12, 14-21, 23, 38, 142,143, 181,
128,199 n.3 195 n.5, 201 n.5
Dawkins.M. S.,4,116,118 Dummett on, 15,195 n.3, 201 n.5
Day, R. H., 76 force/content distinction, 15
Deception behavior, 4,176 thoughts vs. acts of thinking, 14-15
Desire see also Force/content distinction;
content of, 60, 66-68 Fregean approach to understanding
goal-desires vs. situation-desires, 48-49, thought
170, 173 Fregean approach to understanding thought,
higher-order, 68, 165, 170-171 14-21, 35,105
Developmental psychology. See Psychol- difficulties acomrnodating nonlinguistic
ogy, developmental thought, 19-20
Devitt, M., 29 Frisch, K. von, 152
Dickinson, A., 48, 55,145,166,195 n.l, Functionalism, 104-105,199 n.7
200 n.9
Dishabituation paradigm, 5, 54, 78-80, Gallese, V., 27,196 n.9
147 Gallistel, C. R., 5,101,103,199 n.6
Disjunctive syllogism, 141, 147, 191 Gavagai, 70-71
Dolphins Geach, P. T., 201 n.8, 204 n.16
echolocation, 84 Gelman, S. A., 185
response to gestural commands, 54 Generality, 166
sensitivity to submerged wires, 87 in sensu composite (collective), 183-185
Donald, M., 5,154,204 n.14 in sensu diviso (distributive), 183-185
Doxastic modification, 165,168-170 Generativity, 17
direct, 169 Gibson, J. J., 47, 52-54, 121, 122, 123
reflective, 161-162,169-170 Gibson, K., 4,55,126
Dretske, R, 40,66-67,68,173,174, 203 n.6 Goldman, A., 196 n.9
Dummett, M., 15,42-44,45,47, 52, 59, G6mez, J. C., 176
108, 181, 189, 195 n.2, 201 n.5 Gopnik,A.,3,54,78,84
Index 221
Gould, C.J., 112,114 Kaplan, D., 197 n.4

Gould, J.L., 112, 114 Kastak, C. R., 113, 114
Gould, S. J., 69 Kastak, D., 113, 114
Greenfield, P. M., 54, 156 Kestenbaum, R., 79
Grice, H. P., 201 n.6 Knowlton, 196 n.l
Griffin, D., 4 K6hler,W.,43, 52, 130
Griffith, C. R., 6 Krebs, J. R., 114-115, 117,197 n.ll
Hacker, P. M. S., 195 n.l Lander, D. G., 86

Hammond, L. J., 56, 126, 167 Language
Hare, B., 177 as a cognitive tool, 150,151
Harman, G., 203 n.l definition of, 155-156
Hauser, M. D., 4, 5, 52-54, 78, 85-86, 157, emergence of, 4, 55,154,180-181,196
176, 177, 198n.l6, 198 n.4 n.ll
Heinrich, B., 130 required for attribution of prepositional
Heller, M, 198 n.2 attitudes, 172-173
Herman, L., 54 required for logic, 182, 191
Herrnstein, R. J., 86 required for modal thought, 179
Hershberger.W.A., 125, 167 required for quantified thought, 181-183
Heyes, C., 55, 166, 195 n.l, 200 n.9 required for reflective doxastic modifica-
Hirschfeld, L.A., 185 tion, 170
Hofsten, C. von, 79, 81 required for reflexive thinking, 158-164
Honeybees (worker dances in), 152-153 required for tensed thought, 179-180
Hood, B., 80 Language of thought hypothesis, 12, 20,
Hookway, C., 201 n.4 22-31,35,159-160,192-193
Horgan, T., 56, 57 decision-theoretic model of practical
Hume, D., 145 reasoning and, 134
epistemology of, 27-31
Illusory conjunctions, 73 key elements of, 23-26
Imitation, 154 linguistic comprehension and, 23-24,
Ingold, T., 55, 126 196 nn. 6, 7
Innate releasing mechanisms, 4, 7, 8, 48, reasoning and, 24-25
116, 166 syntax and semantics, 24—25, 28-31
Instrumental component of intentional Lea, E. G. L., 4, 8
action, 46-51, 51-56, 63, 129-130, Learning, 56-59
131-132, 139 Leslie, A. M., 83
Intentional ascent, 140, 142, 148, 151, 165, Levallois flake, 127-128
187, 191 Level 0 rationality, 116-120, 123, 128, 129
attributions of perception and, 173-175 norms of, 118-120,135
higher-order desires and, 170-171 Level 1 rationality, 120-122,123,124,168
implicit, 166, 178 minimalist conception of nonlinguistic
involved in attribution of prepositional thought and, 122, 190
attitudes, 172-173 norms of, 121-122, 131,135
requires semantic ascent, 158-163 psychological explanation and, 129-130
Level 2 rationality, 123-128,129-132
Jackson, R, 160, 197 n.14, 203 n.6 distinguished from level 1 rationality,
Johnson-Laird, P., 160,163 125-128
doxastic outcome-sensitivity and,
Kacelnik,A.,114 167-168
Kalish, D., 101, 102 norms of, 124-125, 131-132, 135
222 Index
Level 2 rationality (continued) Modal operators, 166

protologic and, 149 Mode of presentation problem, 89-108
reasoning in, 131-132 nominative, 89, 92-95, 96,107
Lewis, D., 120,155,198 n.19,198 n.l, 201 perceived similarity and, 94-95
n.9 predicative, 89, 96-103,107-108
Lewontin, R., 69 M011er, A. P., 4
Liebermann, P., 154 Monkeys, cotton-top tamarin
Linguistic thought gravity bias in, 85
psychological explanation and, 18-19 sensitivity to functional properties,
epistemology of, 17-18 52-53,198 n.16
Fregean understanding of, 13, 14—21 Monkeys, rhesus, gravity bias in, 85
language of thought hypothesis and, 13 Moore, M. K., 80,154
metaphysics of, 15 Morgan, L. See Morgan's canon
semantics of, 15-17 Morgan's canon, 6-7
structure of, 15-17 Morrison, V., 81
Locke,!., 111-112 Morton, A., 46
Logical concepts, 156,191 Miiller-Lyer illusion, 41
not available to nonlinguistic creatures, Mulligan, K., 203 nn.6, 8
140-141,178-183 Munakata, Y., 5, 54,78, 85
Logical form, 18, 24 Munn, C. A., 4
Loveland, D. H., 86 Musen, G., 196 n.l
Lowe, E.J., 111
Nadel, L., 199 n.6
McDowell,!., 41,42 Nagel, T., 86
McKenzie, B. E., 76, 81 Naive physics. See Physics, naive
McLeod, P., 159 Narrative thinking, 166, 180-181
Mackie,!. L., 146 Negation
Magnetic resonance imaging, 27 predicate vs. prepositional, 142-143,
Makin, G., 195 n.l 144-145
Marler, P., 154, 157 see also Protonegation
Marsh tits (food caching), 48, 197 n.l 1 Neural correlates of cognition. See Cogni-
Maynard Smith,!., 120 tion, neural correlates of
Mellars, P., 4 Nonconceptual content. See Content, non-
Mellor, D. H., 65, 146, 201 n.10 conceptual
Meltzoff, A., 3, 54, 78, 84, 154 Nonlinguistic thought
Mental logic theory, 163 epistemology of, 11, 32-33
Mental maps theory, 160-162 example of rat maze behavior, 98-103
Mental models theory, 160-161, 162-164, inference and, 11-12, 140-149
202 n.9 limits to, 140,165-188,191-192
Metarepresentation. See Intentional ascent ontology of, 33, 82-87
Millikan, R. G., 68, 199 n.l principal problems concerning, 10-12,
Mimesis. See Imitation 21,31
Minimalist theory of nonlinguistic thought, problematic on Fregean understanding of
21, 32, 34-63,64,108,150,189-190 thought, 19-20
arguments against, 51-62 as protothought, 42-45
level 1 rationality and, 122, 190 rationality of, 109-132
psychological explanation according to, relation to belief-desire psychology,
46-51. 10-11, 21, 26, 32, 34-35, 64 189-190
See also Protothought semantics of, 11, 33
Mithen, S., 4-5, 55, 127, 181, 185, 186 structure in, 57,95-97,100-103
Index 223
see also Minimalist theory of nonlin- Principle of cohesion, 79

guistic thought; Rationality; Success Principle of contact, 80
semantics Prior, A. N., 179
Projectability. See Structured representa-
Object*, 95,96 tions
Object concept (as sortal), 92-94 Pronominal reference, 74, 78
Objective phenomenology, 86-87 Prepositional thought. See Thought,
Object perception prepositional
in infancy, 79-81 Protocausation, 146-147
in nonhuman animals, 85-87 Proto-inference. See Protologic
scepticism about, 92-94 Protologic, 140-149, 188. See also Proto-
Object properties, 75, 82-87, 92-93 causation; Protonegation.
Observation categoricals, 73-74 Protonegation, 143-145, 147, 149
O'Keefe, J., 199 n.6 Protothought, 42-46, 52, 64
Ontological indeterminacy, 71-72, 82-87, Campbell on, 45, 62
106 Cussins on, 44-45,62
Optimal foraging theory, 114-115 Dummett on, 42-44, 62
heuristics in, 115, 117-118 Psychological explanation
Orangutans (liquid conservation tasks), cognitive integration and, 9, 62
37-38 counterfactuals in, 60-61
Outcome-sensitivity, 166-168 distinctive features of, 7-10
practical vs. doxastic, 167-168 language of thought hypothesis and, 22
minimalist construal of, 46-51
Paradox of Self-Consciousness, The, 82, 83 misrepresentation and, 9-10, 62
Path-dependent learning, 151 practical decision-making and, 18
Peacocke, C., 40-41,173,197 n.8, 198 n.3 primitive form available to nonlinguistic
Peirce, C. S., 201 n.4 creatures, 175-178
Perception rationality and, 109-110, 128-132,
analogue nature of, 40 190-191
direct vs. indirect, 52 structured representations and, 95-97,
feature-placing vs. particular-involving, 108
58-59, 93 utility conditions and, 90-92
fineness of grain, 40—41 Psychology, developmental, 3, 5, 35,46,
instrumental dimension of, 52-56, 57 71,109,189
nonpropositional nature of, 40-42 Pygmy chimpanzees (learning lexigrams),
two types of report, 174-175 54
unit-free nature of, 40
Perceptual constancy, 75-78, 81 Quine, W. V. O., 58, 61, 63, 64, 78,144,
Perry, J., 197 nn. 4, 5 186, 204n.l6
Physics, naive, 4, 37, 186 radical interpretation, 64, 70-72
Piaget, J., 7, 67, 81
Pigeons Radical interpretation, 70-72
equivalence classification in, 113-114 Ramsey, F. P., 65
visual discrimination in, 85-86 Rationality
Pinker, S., 196 n. 11 inference-based conception of, 110-116,
Poole, J., 86 128,142,148-149,191
Possible worlds semantics, 90-92, 195 n.4 internal vs. external, 110, 130, 200 n.8
Povinelli, D., 176, 177 intuitionist conception of, 111-112
Premack,A. J., 114 normative vs. descriptive dimensions,
Premack, D., 114, 175 134-138
224 Index
Rationality (continued) Savage-Rumbaugh, E. S., 54, 155, 156

procedural, 110, 112 Schmidt, H., 73
see also Level 0 rationality; Level 1 Schusterman, R. J., 112-114
rationality; Level 2 rationality; Sea lions (equivalence classification in),
Reasoning 112-114
Rats (maze behavior), 98-103 Second-order cognitive dynamics, 158,
different frames of reference, 98-100 159,161, 166, 170, 188, 191-192
utility condition of, 99-103 Sejnowski, T., 30,196 n.13
Rats (sensitivity to change in conditioning Sellars.W., 192.
schedule), 55-56, 125-126 Semantic ascent. See Intentional ascent
Rats (sensory preconditioning in), 113 Sense. See Fregean approach to understand-
Ravens (instrumental reasoning in), 130 ing thought; Mode of presentation
Reasoning problem
analogical, 36, 57 Seyfarth, R. M., 157, 200 n.4
conditional, 141, 145-146 Shanker, S., 54
decision-theoretic model of, 133-140 Shanks, D., 145
empathetic, 36 Shephard, R., 197 n.2
from excluded alternative, 141, 142, Sherry, D.R, 48, 197 n.ll
143-144 Shettleworth, S. J., 197n.ll
imagistic, 36,41, 42^3 Signalling strategies, 119-120
normative vs. descriptive dimensions, Signals
134-138 arbitrary, 153
trial and error, 36 iconic, 153
see also Protocausation; Protonegation; Simple seeing, 174-175
Rationality; Visual imagination Single neuron studies, 27
Red deer stags (roaring contests of), Skryms, B., 120
135-138,168 Skucy, J. C., 55-56, 126, 167
Redshank (feeding patterns of), 117-118 Slater, A. M., 81
Reification Sluga, H., 195 n.l
higher-order principles account of, 65, Sommers, R, 201 n.6
75-78, 94, 106 Spelke, E., 3, 54, 78, 85, 147
Quine's account of, 56, 58-59, 61, 63, Sperber, D., 145, 149
65, 72-73 Squire, L., 196 n.l
pronominal reference and, 74, 78 Stalnaker, R., 90-92, 195 n.4
two accounts of, 72-78 Starlings (foraging behavior in), 115
Relative clauses, 144-145, 186-187, 204 Stein, E., 116
n.16 Stephens, D. W., 117
Relative pronoun, 166, 186-187, 201 n.8. Stern, D., 152
See also Pronominal reference; Relative Stich, S., 159
clauses Strawson, P. R, 58, 72, 75
Rescorla, R. A., 55-56, 113, 126, 166 Stromswold, K., 196 n.ll
Rey, G., 134 Structured representations, 57, 59, 61, 190
Rips, L. J., 163 projectability and, 97, 100-103
Ristau, C., 4 psychological explanation and, 95-97,
Ritchie, B. R, 101,102 108
Rizley,R. C., 113 vehicle vs. content, 161
Rizzolati, G., 27 Subtraction fallacy, 94, 198 n.2
Rochat, P., 37-38 Success semantics, 33, 63, 64, 88-89,
Rosenthal, D. M., 159 90-92, 190-191
Ryle, G., 37 advantages of, 68-69
Index 225
characterisation of, 65-69 Tinbergen, N., 7, 8

not an account of truth, 66 Tolman, E. C., 101, 102
overview of, 104-108 Tomasello, M., 154, 177, 203 n.9
radical interpretation and, 70-72 Tooby,J., 185
restricted to nonlinguistic creatures, 66, Tool manufacture and use, 126-128, 139
68 chimpanzees, 55, 126, 154
see also Mode of presentation problem; cotton-top tamarins, 52-53
Utility condition early hominids, 126-128, 180, 186
Symbols, 152 Tootell, H. E., 76, 81
arbitrary, 155-156 Treisman, A., 73, 75
complex, 155-156,160 Turek, R. J., 157
Symbol systems Turvey, M.T.,197n.lO
hierarchically complex, 156
sequentially complex, 156 Upper Paleolithic, 180-181, 186
Utility condition, 65-66, 68, 69-70. See
Taylor, C., 49-50 also Success semantics
Taylor, T., 54 canonical specification of, 96-97,
Teleosemantics, 68 100-103,107
Tense operators, 166 indeterminacy in specifying, 88-89,
Theory of mind, 165,175-178 106-108
Thinking about thinking. See Metarepresen- intensional rather than extensional,
tation; Thought, reflexive 89-92
Thinking-how, 160 psychological explanation and, 90-92
contrast with thinking-that, 36-38 see also Mode of presentation problem
Thinking-that. See Thinking-how
Thought and language Van de Walle, G. A., 79
isomorphism between, 16-17,22 Vaughan,W., 113
relative priority of, 12 Vehicle. See Thought, vehicles of
Thought, compound, 178-183 Vehrencamp, S. L., 119
Thought, domain-general, 185-187 Vervet monkeys (signaling behavior in),
Thought, domain-specific, 5, 37, 185-187 121-122, 157
Thought, modal, 179, 188 Visual imagination, 36,41, 47, 53. See also
Thought, prepositional Reasoning, imagistic
distinct from perception, 40-42
key features of, 38-39, 104 Waal, R de, 175
Thought, quantifed, 181-183, 183-185, 187 Walker, S., 85
Thought, reflexive Wason selection task, 116
language required for, 158-164 Weiner.J., 195 n.l
varieties of, 165-188 Whiten, A., 175,177
see also Intentional ascent Whyte, J. T., 65,67
rthough tenced 170 180 Wiggins, D., 92-94
Thought, vehicles of, 111-112,159-164, Willats, P., 80
192-193 Wittgenstein, L., 202 n.7
structure in, 161 Woodruff, G., 175
Thought without language. See Nonlinguis- Wright, C., 71
tic thought
Tienson, I, 56, 57 Zeki, S., 73

Bermúdez, J.L., Thinking Without Words PDF

Uploaded by

Copyright:

Available Formats

Bermúdez, J.L., Thinking Without Words PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bermúdez, J.L., Thinking Without Words PDF

Uploaded by

Copyright:

Available Formats

Thinking without Words

PHILOSOPHY OF MIND SERIES

The Conscious Mind

Deconstructing the Mind

The Human Animal

Minds and Bodies

Consciousness and Cognition

Jose Luis Bermudez

Copyright © 2003 by Oxford University Press, Inc.

Oxford is a registered trademark of Oxford University Press

Library of Congress Cataloging-in-Publication Data

versions of the discussion of the language of thought hypothesis were delivered at

to me to be fundamentally correct, and the discussion of the differences between

Note to the Reader xv

1 The Problem of Thinking without Words 3

1.1 Morgan's canon and psychological explanations

1.2 The parameters of the problem 10

2 Two Approaches to the Nature of Thought 13

2.1 The Fregean approach: Thoughts as the senses

2.2 The language of thought hypothesis 22

2.3 The language of thought and the epistemology

2.4 The road ahead 31

3 Minimalist Approaches to Nonlinguistic Thought 34

3.1 Thinking-how and thinking-that: An intuitive

3.2 Minimalist approaches to nonlinguistic thought 38

3.3 The minimalist construal of intentional action

3.4 The argument from planning 51

3.5 The argument from intelligent action 56

3.6 The argument from psychological explanation 60

3.7 Taking stock 62

4 Ascribing Thoughts to Nonlinguistic Creatures:

4.1 Success semantics: Problems and prospects 65

4.2 Success semantics and radical interpretation 69

4.3 Two types of reification 72

4.4 Dishabituation and higher-order principles 78

4.5 From object-properties to ontology 82

5 Ascribing Thoughts to Nonlinguistic Creatures:

5.1 Individuals, modes of presentation and

5.2 Structured thoughts and psychological explanation 95

5.3 Success semantics and nonlinguistic thought:

6 Rationality without Language 109

6.1 Theoretical and practical rationality 109

6.2 Level 0 rationality 116

6.3 Level 1 rationality 120

6.4 Level 2 rationality 123

6.5 Nonlinguistic rationality and psychological

7 Practical Reasoning and Protologic 133

7.1 The decision-theoretic model of practical reasoning 133

7.2 Inference and proto-logic 140

8 Language and Thinking about Thoughts 150

8.1 Language as a cognitive tool 151

8.2 Intentional ascent and semantic ascent 158

9 The Limits of Thinking without Words 165

9.1 Belief revision and second-order cognitive dynamics 166

9.2 Higher-order desires 170

9.3 Intentional ascent and understanding other minds 171

9.4 Embedding and intentional ascent 178

9.5 Two types of general thought 183

9.6 From domain-specificity to domain-generality 185