Forms of Thought

FORMS OF THOUGHT
Forms of thought are involved whenever we name, describe, or

identify things, and whenever we distinguish between what is, might
be, or must be the case. It appears to be a distinctive feature of human
thought that we can have modal thoughts, about what is possible or
necessary, and conditional thoughts, about what would or might be
the case if something else were the case. Even the simplest thoughts
are structured somewhat like sentences, containing referential and
predicative elements, and studying these structures is the main task of
philosophical logic. This clear and accessible book investigates the
forms of thought, focusing on and drawing out the central logical
notions of reference, predication, identity, modality, and condition-
ality. It will be useful to students and other readers interested in
epistemology and metaphysics, philosophy of mind and language,
and philosophical logic.
e . j . l o w e is Professor of Philosophy at the University of Durham.

His publications include Subjects of Experience (Cambridge, 1996),
The Possibility of Metaphysics (1998), An Introduction to the Philosophy
of Mind (Cambridge, 2000), The Four-Category Ontology (2006),
Personal Agency (2008), and More Kinds of Being (2009). He is a
General Editor of the series Cambridge Studies in Philosophy.
FORMS OF THOUGHT
A Study in Philosophical Logic
E. J. LOWE
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, São Paulo, Delhi, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge cb2 8ru, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9781107001251
© E. J. Lowe 2013
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.
First published 2013
Printed and bound in the United Kingdom by the MPG Books Group
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data

Lowe, E. J. (E. Jonathan)
Forms of thought : a study in philosophical logic / E.J. Lowe.
pages cm
Includes bibliographical references (pages) and index.
isbn 978-1-107-00125-1 (Hardback)
1. Logic. I. Title.
bc71.l69 2013
160–dc23
2012041306
isbn 978-1-107-00125-1 Hardback
Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party internet websites referred to
in this publication, and does not guarantee that any content on such
websites is, or will remain, accurate or appropriate.
Contents
List of figures page vii

List of tables viii
Preface ix
Acknowledgements xii
1 Introduction 1
part i reference and predication

2 Individuation, reference, and sortal terms 11
3 Two styles of predication – dispositional and occurrent 31
4 Ontological categories and categorial predication 50
part ii identity
5 What is a criterion of identity? 69
6 Identity conditions and their grounds 94
part iii modality

7 Identity, vagueness, and modality 119
8 Necessity, essence, and possible worlds 139
v
vi Contents
part iv conditionality
9 The truth about counterfactuals 163
10 Conditionals and conditional probability 182
Bibliography 205
Index 210
Figures
Figure 3.1 The ontological square, version i page 34

Figure 3.2 The ontological square, version ii 35
Figure 3.3 The ontological square, version iii 36
Figure 3.4 The ontological square, version iv 38
vii
Tables
Table 3.1 Logical symbolism for the four-category ontology page 36
viii
Preface
I have given this book the subtitle A Study in Philosophical Logic in

recognition of Bertrand Russell. It was Russell who gave philosophical
logic its name, in Our Knowledge of the External World, saying that its
business is to extract our knowledge of the logical forms of propositions
from its ‘concrete integuments’ and to ‘render it explicit and pure’ (see the
passage quoted at the beginning of Chapter 1 of the present book).
Although in practice that task must be approached by investigating the
structure of sentences in natural language – since it is in such sentences
that our thoughts are clothed and communicated – the underlying aim is
to reveal the forms of thought, at least to the extent that thoughts are
propositional in character and thus capable of standing in logical relations
to one another.
Propositional thought is always complex and structured, even when it
involves only ‘simple’ or ‘atomic’ propositions, and this is why propos-
itional thoughts can always stand in logical relations to other such
thoughts. Identifying the forms of atomic thoughts is, then, the first task
of philosophical logic, and only having completed that should we endeav-
our to reveal the forms of more complex thoughts, including compound
thoughts which contain subordinate thought-contents as proper parts.
Atomic thoughts contain both referential and predicative constituents, so
that an inquiry into the nature of reference and predication is an essential
first step in philosophical logic. But some predications have a special
importance from a logical point of view, especially predications of identity
and predications of necessity and possibility – modal predications. That
being so, an investigation of the notions of identity and modality is
another essential step that philosophical logic must take. And where
compound thoughts are concerned, the compounding relations between
subordinate thought-contents need to be understood, the most important
of these – in view of its intimate connection with logical inference – being
conditionality. These, accordingly, are the central logical notions that will
ix
x Preface
be discussed in this book: reference, predication, identity, modality, and
conditionality. Much that needs to be said concerning the forms of thought
can be said in terms of these notions. Other important logical notions that
I shall not focus on in this book are those of negation, existence, and truth –
and, although I have views about all of these three notions, I shall reserve a
full examination of them for another occasion, discussing them only in
passing in the present book.
However, a complicating factor for present purposes is that none of the
logical notions that I do discuss in this book can be understood entirely
independently of each other: for instance, there are intimate connections
between reference and identity and between modality and conditionality.
Hence, although different chapters of this book are primarily devoted to
each of the five core notions mentioned above, not everything to be said
about each notion will be confined to the chapters devoted to it. Even so,
I have endeavoured to make each chapter of the book relatively self-
standing, so that it is intelligible for a reader who does not have time to
study the book as a whole and just wants to focus on one particular issue.
As a consequence, some chapters inevitably include a certain amount of
recapitulation of matters discussed more fully elsewhere; but I hope that
readers will find this preferable to a much more extensive use of cross-
reference between chapters, which would have required them to turn quite
frequently to other chapters in the course of their reading in order to
follow certain discussions or lines of argument.
I have noted already that this book does not pretend to offer a discussion
of all the logical notions needed for a full characterization of the forms of
propositional thought and an exhaustive account of all the logical relations
in which propositional thoughts can stand to one another. But one
omission may strike some readers as being strange, namely the lack of
much discussion – save in Chapter 10 – of the propositional attitudes, such
as belief. Now, it is undoubtedly true that important logical questions arise
concerning the validity of logical inferences in which the conclusion and
some of the premises are sentences involving propositional attitude verbs,
because a sentence of the form ‘S believes that p’ – for example –
apparently provides a non-extensional context for the embedded sentence
‘p’, in which co-referring terms cannot necessarily be substituted for one
another salva veritate. However, while the notion of extensionality is
undoubtedly a logical one – and one which I admittedly do not discuss
in this book – it must surely also be acknowledged that the notion of belief
as such is not a logical but a psychological one, and this, at root, is why I do
not have much to say about it in this book. My assumption is that
Preface xi
philosophical logic, although it is centrally concerned with the nature of
propositional thoughts, is concerned with them solely insofar as such
thoughts have truth-evaluable propositional contents, and is not at all
concerned with such thoughts qua psychological states of thinkers: that,
rather, is a task for the philosophy of mind.
Of course, there are some interesting formal and semantic analogies
between propositional attitude expressions and modal expressions –
between, for instance, sentences of the form ‘S believes that p’ and
sentences of the form ‘It is possible/necessary that p’. However, I do
discuss modality extensively in this book, especially in Part iii, and even
urge, in Chapter 9, that modal expressions in everyday language sometimes
call for ‘epistemic’ interpretations. Nonetheless, I have serious doubts
about the idea that there could be a logic of belief – ‘doxastic’ logic – on
a par with modal logic, treating ‘it is believed that’ as a quasi-logical
sentential operator on a par with ‘it is necessary that’. Purely logical
notions, such as the five core notions focused on in this book, are ‘topic-
neutral’, since logic and reasoning are applicable to any subject matter,
whether it be in the domain of the physical, the psychological, the social,
or indeed the abstract (as in mathematics). This, incidentally, is not to
deny that an important distinction may be drawn between theoretical and
practical reasoning, with the former providing guidance concerning what
we should believe in the light of the evidence available to us, while the latter
provides guidance concerning how we should act in the light of our goals
and needs. But the same purely logical notions are applicable on both sides
of this divide. Theoretical reasoning, in my view, does not require a logic of
belief, in the sense that a putative ‘doxastic’ logic would constitute this.
This is a book aimed primarily at professional philosophers and graduate
students in philosophy, although I have tried to write it in a style that
makes it clear and accessible also to middle- and upper-level undergraduate
students of philosophy with a suitable background in logic, metaphysics,
and the philosophy of language. It is partisan, in the sense that I resolutely
defend certain controversial positions on the issues that I discuss, but
I attempt to conduct this defence in a fair-minded spirit, giving due weight
to the force of opposing views. Most of these positions I have defended in
print previously, and the chapters of this book draw to varying extents on
earlier work of mine. At the same time, I have naturally changed my
opinions about many matters over the years and this book represents only
my current views about the topics that it covers.
Acknowledgements
The chapters of this book draw in some measure on previously published

papers of mine, as follows, and I am grateful in all cases to the publishers
and editors concerned for permission to reuse the material in question in
this way. Chapter 2: ‘Individuation, reference, and sortal terms’, in
A. Raftopoulos and P. Machamer (eds.), Perception, Realism, and the
Problem of Reference (Cambridge University Press, 2012), pp. 123–41.
Chapter 3: ‘Modes of exemplification’, in B. Langlet and J.-M. Monnoyer
(eds.), Gustav Bergmann: Phenomenological Realism and Dialectical Ontol-
ogy (Frankfurt: Ontos Verlag, 2009), pp. 173–91. Chapter 4: ‘Categorial
predication’, Ratio 25 (2012), pp. 369–86. Chapter 5: ‘What is a criterion of
identity?’, Philosophical Quarterly 39 (1989), pp. 1–21. Chapter 6: ‘How are
identity conditions grounded?’, in C. Kanzian (ed.), Persistence (Frankfurt:
Ontos Verlag, 2008), pp. 73–89. Chapter 7: ‘Identity, vagueness, and
modality’, in J. L. Bermúdez (ed.), Thought, Reference, and Experience:
Themes from the Philosophy of Gareth Evans (Oxford University Press,
2005), pp. 290–310. Chapter 9: ‘The truth about counterfactuals’, Philo-
sophical Quarterly 45 (1995), pp. 41–59. Chapter 10: ‘Conditional probabil-
ity and conditional beliefs’, Mind 105 (1996), pp. 603–15, and ‘What is
conditional probability?’, Analysis 68 (2008), pp. 218–23.
Some of the material in this book was first presented in public in the
form of conference or research seminar papers, and I am grateful to the
many people – too many to be named individually here – who offered
criticisms and suggestions for improvement on those occasions. I am
indebted to Hilary Gaskin and Anna Lowe of Cambridge University Press
for all their expert advice, encouragement, and assistance. Finally, I am also
very grateful to a number of anonymous readers for the Press who
provided very helpful suggestions both concerning the original book
proposal and regarding the penultimate draft of the manuscript.
xii
chapter 1
Introduction
Some kind of knowledge of logical forms, though with most people it

is not explicit, is involved in all understanding of discourse. It is the
business of philosophical logic to extract this knowledge from its
concrete integuments, and to render it explicit and pure.
Bertrand Russell (1914)
As I mentioned in the Preface, I have given this book the subtitle A Study
in Philosophical Logic in recognition of Bertrand Russell, who coined the
term ‘philosophical logic’ in the passage quoted immediately above.1 It is
unfortunate, in my view, that many philosophers now seem to have
forgotten the origin of this term and, instead of using it in Russell’s very
useful sense, take it to mean instead something like the philosophy of logic(s),
which is at once broader and narrower than what, I think, Russell primar-
ily had in mind: broader inasmuch as the philosophy of logic(s) is con-
cerned, inter alia, with evaluating consistency and completeness proofs for
various systems of formal logic – that is, with metalogic – and with
adjudicating between different rival systems of formal logic (for instance,
different formal systems of modal logic); and narrower inasmuch as the
philosophy of logic(s) is less concerned with what may aptly be called the
logic of natural language, as opposed to systems of formal logic which utilize
artificial symbolic languages. As I understand Russell, the primary aim of
philosophical logic is to reveal the forms of thought, to the extent that
thoughts are propositional in character and thus capable of standing in
logical relations to one another, and this requires it to focus on thought as
it is most naturally expressed, in the shape of sentences of one or another
natural language. As I explained in the Preface, it is in pursuit of this aim
that I have singled out the focal topics of the book’s remaining nine
chapters – namely the central logical notions of reference, predication,
1
Bertrand Russell, Our Knowledge of the External World (London: George Allen and Unwin, 1922),
p. 53.
1
2 Introduction
identity, modality, and conditionality. It is my belief that much – though by
no means all – that needs to be said concerning the forms of thought can
be said in terms of these key notions.
I shall keep the rest of this Introduction brief, restricting it to a short
outline of the contents of the remaining chapters of the book, but I refer
readers once more to the Preface for a statement of my primary intentions
and guiding thoughts in writing the book. The remainder of the book is
divided into four Parts, dealing respectively with the topics of reference and
predication (Chapters 2 to 4), identity (Chapters 5 and 6), modality (Chap-
ters 7 and 8), and conditionality (Chapters 9 and 10).
1 reference and predication

In Chapter 2, ‘Individuation, reference, and sortal terms’, I argue –
contrary to the adherents of most versions of the so-called ‘direct’ theory
of reference – that singular reference to an individual cannot in general be
secured by a thinker without that thinker’s grasping, at least implicitly, a
criterion of identity which that individual satisfies, where such a criterion is
linked to a family of general terms of the ‘sortal’ variety. This is a claim that
I have defended elsewhere, notably in my Kinds of Being: A Study of
Individuation, Identity, and the Logic of Sortal Terms (Oxford: Blackwell,
1989) and more recently in my ‘Sortals and the Individuation of Objects’,
Mind and Language 22 (2007), pp. 514–33. Here I argue afresh for the
claim. My defence of the claim does not, however, commit me to the truth
of a so-called ‘descriptive’ theory of reference, as such a theory would
normally be understood. Moreover, I distinguish my version of the claim –
which I call ‘categorialism’ – from a more demanding and consequently
less credible version, called ‘sortalism’. According to my version, singular
thought about an individual is available only to a thinker who at minimum
grasps – even if only implicitly and somewhat imprecisely – to which
ontological category the individual in question belongs, thereby allowing
that the thinker may be seriously in error concerning any specific sortal
concept under which that individual falls.
Chapter 3 is entitled ‘Two styles of predication – dispositional and
occurrent’. In this chapter, I am concerned solely with what are sometimes
called material predications, as opposed to formal predications. As
I understand this distinction, predications of the latter kind predicate
merely ‘formal’ properties and relations, such as existence and identity, of
their subjects, whereas predications of the former kind predicate ‘material’
properties and relations – that is, genuine universals – of their subjects. The
Introduction 3
implication is that material predications, as I understand them, are exist-
ence-committing: they commit the person who makes such a predication to
the existence of the relevant universal. Hence, this is a thoroughly anti-
nominalist view where such predications are concerned. However, I also
believe, in line with other previous work of mine, that material predica-
tions further subdivide exhaustively and exclusively into two sub-kinds:
dispositional predications and occurrent predications. This is an idea exten-
sively discussed in my previously mentioned book, Kinds of Being, and
more recently in my The Four-Category Ontology: A Metaphysical Founda-
tion for Natural Science (Oxford: Clarendon Press, 2006) and More Kinds of
Being: A Further Study of Individuation, Identity and the Logic of Sortal
Terms (Malden, MA and Oxford: Wiley-Blackwell, 2009), the latter being
a revised and extended version of Kinds of Being. But my treatment of the
topic in Chapter 3 goes considerably beyond these earlier treatments in
important new ways. In it, I also take the opportunity to correct some
misconceptions that critics of my position have fallen prey to.
In Chapter 4, ‘Ontological categories and categorial predication’,
I return to two important topics involved, either explicitly or implicitly,
in the preceding two chapters: the notion of an ontological category and
the notion of formal predication. My view is that a predication in which
an entity is said to belong to a certain ontological category is one of the
formal kind, the implication being that ontological categories should not
be thought of as being high-level universals and, correspondingly, that
categorial concepts should be thought of as being ‘formal’ rather than
‘material’ ones. Strictly speaking, then, such categories should not be
included in an inventory of what there is: they do not belong to the
existential content of reality and are not ‘entities’ of any kind – although
this in no way compromises the mind-independent truth of categorial
predications. In the course of accommodating the notion of categorial
predication, I criticize the ontological presuppositions of the type of
formal predicate logic that contemporary philosophers have inherited
from the founders of modern quantificational logic, notably Gottlob
Frege and Bertrand Russell, and propose some major reformations.
This carries further forward the task, begun in my Kinds of Being, The
Four-Category Ontology, and More Kinds of Being, of constructing a system
of formal logic which perspicuously reflects the neo-Aristotelian onto-
logical presuppositions of my own preferred system of categorial
ontology, which identifies four fundamental ontological categories – those
of individual substance, substantial kind, individual mode or accident, and
universal attribute.
4 Introduction
Here it is perhaps worth mentioning again that I regard existence as
being a ‘formal’ property, in the same sense in which I take identity to be a
‘formal’ relation. I also contend that the notion of existence, like that of
identity, is a primitive and unanalysable one. The existence predicate,
I maintain, is a so-called ‘first-order’ predicate, not a ‘second-order’ one,
as Frege, Russell, W. V. Quine and very many modern analytic philoso-
phers suppose. (It is largely a matter of taste whether we say ‘second-order’
or ‘second-level ’ here, so my choice of the former is not meant to be
significant.) It is important in this connection not to confuse the claim,
which is certainly still very widely held, that ‘– exist(s)’ is not a first-order
predicate, on a par with ‘– run(s)’ or ‘– eat(s)’, with the claim, often
associated with Kant, that ‘– exist(s)’ is not a predicate at all. The latter
view is scarcely credible, if taken literally. The former view, however, arises
from the widespread doctrine that the logical form of propositions
ascribing existence is quantificational – so that, for instance, ‘Tigers exist’
should be understood to be logically equivalent to, or indeed analysable
as, ‘Something is a tiger’. Quantifier phrases, as they are standardly
construed by philosophical logicians, have the logical status of second-
order predicates – that is, predicates of (first-order) predicates. However,
in my view, existence is not properly expressed by a quantifier – the
tendentiously named ‘existential’ quantifier, standardly symbolized by
‘9’. ‘– exist(s)’ really is a first-order predicate, on a par, as far as logical
syntax is concerned, with ‘– run(s)’ and ‘– eat(s)’. Nonetheless, because the
existence predicate is a formal rather than a material one, it would be
wrong in my view to suppose that existence is a real universal – and hence
wrong to suppose that existence is something that itself exists. There is
nothing at all paradoxical in saying this: indeed, on the contrary, to say
that existence exists should strike most philosophers as absurd.
2 identity
Chapter 5, ‘What is a criterion of identity?’, looks in more depth at the
notion of such a criterion that was first introduced in Chapter 2. This
chapter is based on my paper of the same title which appeared in Philo-
sophical Quarterly 39 (1989), pp. 1–21. I have retained its original title for
this chapter and have revised it only where it deviates from my current
views on its topic, because it has been widely referred to in the intervening
years and I therefore thought it appropriate to make it available, in a form
as close as possible to its original one, in the present volume. The only
significant way in which I have changed my mind about things said in the
Introduction 5
original version involves certain matters covered in Chapter 2, concerning
the manner in which children might be equipped to form identity-
judgements about perceptible material objects in their immediate environ-
ment. Accordingly, I have now brought what I say in Chapter 5 into line
with what I say in Chapter 2 on the matters in question.
In Chapter 6, ‘Identity conditions and their grounds’, I advance from
the more semantically oriented concerns of Chapter 5 to explicitly meta-
physical ones, where questions of identity are at issue. Assuming, in line
with the conclusions of Chapter 5, that entities of different kinds very
often possess different identity conditions – determining, for instance,
what possible changes they can intelligibly be supposed to persist through
over time – the question arises as to the source or ground of these
conditions. One view which I resolutely reject in this chapter is the idea
that these conditions have a purely conceptual basis and are to that extent
the workmanship of the human mind, as John Locke might have put it.
Instead, I argue in favour of a metaphysically realist view of how identity
conditions are grounded, according to which their source lies in the very
essences of the entities concerned, with ‘essence’ being construed in a
realist and broadly neo-Aristotelian fashion consonant with the neo-
Aristotelian categorial ontology espoused in earlier chapters. A very
important aspect of my own account of essence – whether or not it is
faithful to Aristotle himself in this respect – is that I deny that essences
are themselves entities of any kind. In other words, I take the concept of
essence to be, in the terminology introduced earlier, a formal rather than
a material one.
3 modality
Chapter 7 is entitled ‘Identity, vagueness, and modality’. In this chapter
I challenge the widely held view that predications of identity can never be
vague or indeterminate in respect of their truth-value and never be contin-
gent, other than as a consequence of features of the language in which we
express them – that is to say, that the source of such vagueness or
contingency can never be ontological, as opposed to semantic or epistemic,
in character. Here I focus on two very well-known attempts to uphold
each aspect of this widely held view, namely Gareth Evans’s attempted
proof that there cannot be ‘vague objects’ and the alleged proof of the
necessity of identity that is attributable, independently, to Saul Kripke
and Ruth Barcan Marcus. These two supposed proofs are interestingly
parallel in certain important respects and both, in my view, suffer from
6 Introduction
essentially the same underlying fault, which renders each of them subtly
question-begging. The vagueness question is particularly important, from a
metaphysical point of view, because if my opponents are correct it is
difficult to see how our common-sense ontology of ‘ordinary objects’, such
as tables and horses – Aristotle’s ‘primary substances’ – could be held to
reflect the true nature of mind-independent reality. Instead, we would
seem to be driven to endorse a much more ‘revisionary’ and ‘sparse’
ontology, acknowledging the reality only of ‘simple’ material objects, such
as the fundamental particles posited by physics, or indeed the reality only
of a single material object – the physical cosmos as a whole – as some
extreme ontological monists maintain that we should.
Chapter 8, ‘Necessity, essence, and possible worlds’, focuses solely upon
the semantics, logic, and metaphysics of modality. Very commonly in
recent times – thanks especially to the seminal work of Saul Kripke on
the foundations of modal logic – the notion of a necessarily true propos-
ition is explicated in the following way: such a proposition, it is said, is one
that is true in every possible world. However, this explication is no clearer
than the key notion of a ‘possible world’ upon which it draws. In this
chapter, I argue that this notion is thoroughly obscure and really of no use
at all in explicating either the notion of necessity or the metaphysical
ground of necessary truth. Instead, I appeal for these purposes once more
to a neo-Aristotelian notion of essence, building on recent work of mine on
this theme in, for instance, my paper ‘Two Notions of Being: Entity and
Essence’, in Robin Le Poidevin (ed.), Being: Developments in Contemporary
Metaphysics (Cambridge University Press, 2008). The conception of
essence that I defend is, as I say, a neo-Aristotelian one, in stark contrast
with the current mainstream conception, which attempts to define essence
in terms of necessity, rather than vice versa. In defending this approach,
I consciously draw upon insights that are to be found in Kit Fine’s
important recent work on the topic of essence and modality, although
my own views on these matters do not exactly coincide with his in every
important respect.
4 conditionality
In Chapter 9, ‘The truth about counterfactuals’, I develop a distinctive
account of the logic and semantics of counterfactual conditionals which
departs in important respects from all other existing accounts, most
notably the highly influential account of David Lewis. Of course,
the interpretation of conditionals quite generally is notoriously
Introduction 7
controversial – much more so than that, say, of conjunctive or disjunc-
tive propositions. It is still hotly debated, for example, whether condi-
tionals fall into two logically distinct classes – indicative conditionals
and subjunctive conditionals – and equally hotly debated whether all
indicative conditionals are so-called material conditionals. Another
much-disputed question is whether the notion of conditionality, at least
in the case of indicative conditionals, is explicable in terms of the
notion of conditional probability, rather than vice versa – a matter to
which I turn in the final chapter of the book. In the present chapter,
I argue in defence of a unified theory of conditionals, embracing both
indicatives and subjunctives, which explicates them in terms of a
generalized notion of necessity – this notion admitting various more
specific modal interpretations dependent on context. One very import-
ant implication of the account is that the logic of conditionals, including
counterfactuals, is reducible to a variety of standard modal logic. This
chapter is essentially a revised and updated version of my paper of the
same title, ‘The Truth about Counterfactuals’, Philosophical Quarterly
45 (1995), pp. 41–59, although the system of conditional logic that
I defend was first aired much earlier, in my ‘A Simplification of the
Logic of Conditionals’, Notre Dame Journal of Formal Logic 24 (1983),
pp. 357–66. As with Chapter 5, I thought it best to restrict revisions
here to a necessary minimum, because the original paper has been quite
widely referred to since it first appeared in 1995. One reason why
I consider the work of this chapter to be particularly important is that
it can be drawn upon to challenge a view that has recently gained some
currency, according to which our knowledge of modal truths, quite
generally, can be explicated in terms of our knowledge of counterfactual
conditionals. I believe the very reverse of this to be the case, precisely
because I consider the logic of conditionals to be reducible to a variety
of modal logic.
As I have just indicated, Chapter 10, ‘Conditionals and conditional
probability’, is ultimately motivated by the question whether the notion
of conditionality – the notion canonically expressed by the logical connect-
ive ‘if’ – is explicable in terms of the notion of conditional probability, as the
latter is standardly understood in the mathematical theory of probability.
A positive answer to this question has been very ably defended by Dorothy
Edgington, whose work consequently poses a serious threat to my own
attempt to frame a unified theory of conditionals which draws instead
upon modal notions and standard modal logic. In this chapter, I argue that
Edgington’s position is unsustainable and that, in fact, the correct
8 Introduction
direction of explanation is from the notion of conditionality to the notion
of conditional probability, not vice versa. In the process of arguing for this,
I subject the standard ratio-based definition of conditional probability to a
number of criticisms and propose in place of it a definition of conditional
probability which is framed in explicitly conditional terms – and hence in
terms fully consonant with my own unified theory of conditionals.
part i
Reference and predication
chapter 2
Individuation, reference, and sortal terms
In this, the first substantive chapter of the book, I want to defend a thesis
that I call categorialism regarding the individuation of objects, in the
cognitive sense of the term ‘individuation’.1 Individuation in this sense –
which is to be distinguished from individuation in the metaphysical
sense2 – is the singling out of an object in thought. According to categori-
alism, a thinker can single out an object in this way only if he or she grasps,
at least implicitly, some categorial concept under which he or she conceives
the object in question to fall – such a concept being one that supplies a
distinctive criterion of identity for objects conceived to fall under it.
Plausible examples of such categorial concepts would be the concepts of
an animal, a material artefact, and (what I shall call, for want of a better
term) a geographical prominence.
Categorialism, thus, is a more liberal doctrine than sortalism – the latter
doctrine maintaining that an object can be singled out in thought only
when conceived of as falling under some specific sortal concept, such as the
concept of a cat, a table, or a mountain. As these everyday examples
illustrate, categorial concepts are more abstract than any of the more
specific sortal concepts that fall within their range of application: animal,
for instance, is more abstract than either cat or dog, and geographical
prominence is more abstract than either mountain or island. All sortal
concepts falling within the range of application of the same categorial
concept are, it seems clear, necessarily associated with the same criterion of
identity, but they evidently differ with respect to the more specific sortal
persistence conditions governing objects that fall under them. These sortal
persistence conditions – which impose restrictions on what varieties of
1
I have defended this thesis before: see, especially, my ‘Sortals and the Individuation of Objects’, Mind
and Language 22 (2007), pp. 514–33. Here I want to strengthen and extend the arguments of that
paper.
2
For more on this distinction, see my ‘Individuation’, in M. J. Loux and D. W. Zimmerman (eds.),
The Oxford Handbook of Metaphysics (Oxford University Press, 2003).
11
12 Reference and predication
natural change an object can be supposed to survive while continuing to
fall under the relevant sortal concept – are for the most part discoverable
only empirically, whereas criteria of identity proper are most plausibly
classified as relatively a priori metaphysical principles.
In the present chapter, I shall offer some arguments in support of
categorialism and then go on to inquire whether these arguments can be
extended from the domain of singular thought to that of singular linguistic
reference: that is, I shall inquire whether it can reasonably be contended
that a speaker cannot successfully refer to an object by means of a proper
name unless he or she grasps, at least implicitly, that the name’s referent
falls under a certain categorial concept, which supplies a criterion of
identity for the referent. This contention conflicts, of course, with the
assumptions of any purely ‘direct’ theory of reference, to the extent that it
makes an object’s known satisfaction of some broadly descriptive specifi-
cation a necessary – albeit not a sufficient – condition for successful
linguistic reference to that object.
1 sortal, categorial, and transcategorial terms

In what follows, I shall talk pretty much interchangeably of terms and
concepts, except when it is important to distinguish between constituents of
language and constituents of thought. For much of the time, however, it
will be more convenient to speak of terms, as these are obviously more
immediately identifiable, being words or phrases occurring in natural or
formal languages. Sortal terms – a locution coined by John Locke3 – are
nouns or noun phrases denoting putative sorts or kinds of objects.4 They
are also sometimes called substantival general terms.5 Familiar examples
would be the terms ‘cat’, ‘table’, and ‘mountain’. They differ from adjec-
tival general terms, such as ‘white’, ‘square’, and ‘steep’, in having not only
criteria of application but also criteria of identity associated with their use.
This is reflected in the fact that such terms are count nouns, not merely in
the purely grammatical sense, but also in the more robust sense that there
are principles governing their correct use in counting or enumerating
objects to which they apply. A criterion of application tells us to which
objects a general term applies and thereby fixes its extension. A criterion of
3
See John Locke, An Essay Concerning Human Understanding, ed. P. H. Nidditch (Oxford: Clarendon
Press, 1975), iii, iii, 15.
4
See also P. F. Strawson, Individuals: An Essay in Descriptive Metaphysics (London: Methuen, 1959),
p. 168; and David Wiggins, Sameness and Substance Renewed (Cambridge University Press, 2001).
5
See P. T. Geach, Reference and Generality, 3rd edn (Ithaca, NY: Cornell University Press, 1980).
Individuation, reference, and sortal terms 13
identity tells us what conditions need to be satisfied by objects to which a
general term applies if those objects are to be identical with one another.
Since objects can be counted only if some principle is supplied determin-
ing whether or not certain objects to be included in the count are identical
with or distinct from one another, criteria of identity are presupposed by
principles for counting. For example, an instruction to count the cats
living in someone’s house only makes sense given a principled way to
determine whether a cat encountered at one time and place in the house is,
or is not, the same cat as a cat encountered at another time and place in the
house. Clearly, different criteria of identity and principles for counting
apply to objects of different sorts – for instance, to cats as opposed to
mountains. Equally clearly, there are no criteria of identity or principles for
counting that apply to white objects, say, or to square objects, purely
insofar as those objects are white or square. After all, both cats and
mountains can be white, but when we count white cats we apply different
principles from those we apply when we count white mountains, and this
reflects a difference between the criteria of identity associated with the
sortal terms ‘cat’ and ‘mountain’.
Sortal terms fall into hierarchies of subsumption. For instance, ‘cat’ is
subsumed by ‘mammal’, which is in turn subsumed by ‘vertebrate’.
Equally, ‘cat’ itself subsumes, for instance, ‘Siamese cat’. Every sortal term
within a given hierarchy of subsumption is necessarily governed by the
same criterion of identity. At the top of any given hierarchy of subsump-
tion is a categorial term: in the case of the hierarchy to which ‘cat’ belongs,
the term in question is ‘animal’ (or, perhaps, ‘living organism’). This is the
highest term in the hierarchy which shares the same criterion of identity of
all the sortal terms below it in the hierarchy. Any general term that has a
still more general criterion of application than this categorial term, in that
it also applies to objects describable by sortal terms belonging to other
hierarchies of subsumption, is a transcategorial term. Thus, for example,
‘material object’ is a transcategorial term because it applies to objects such
as cats, but also to objects such as mountains, even though the sortal terms
‘cat’ and ‘mountain’ belong to different hierarchies of subsumption.
Because a transcategorial term does not belong to any single hierarchy of
subsumption, there can be no specific criterion of identity, or principle for
counting, associated with its use. As we shall shortly see, it is important not
to confuse the criterion of identity associated with a sortal term with
another kind of principle governing such a term, namely one specifying
the sortal persistence conditions of objects to which the sortal term applies.
These are the conditions that an object must continue to satisfy if the
sortal term is to continue to be applicable to it, and these conditions can
very often differ for different sortal terms within the same hierarchy of
subsumption, or subsumed by the same categorial term. Thus, for
instance, the persistence conditions of cats cannot simply be identified
with those of mammals in general and, equally evidently, cats and frogs,
say, have different persistence conditions, even though ‘cat’ and ‘frog’ are
both subsumed by the categorial term ‘animal’. It is only because all
animals share the same criterion of identity that we can make sense of
narratives, whether fictional or scientific, in which an animal of one sort
supposedly undergoes metamorphosis into an animal of another sort, while
remaining one and the same individual animal. And it is because objects
that are not describable by the same categorial term – such as cats and
mountains – do not share the same criterion of identity that we cannot
make sense of narratives in which, for example, an individual cat survives
transmutation into a mountain. A cat could conceivably be replaced by a
cat-shaped mountain, but the two could not conceivably be one and the
same object: the cat could not continue to exist ‘as’ a mountain.
2 criteria of identity and sortal persistence

conditions
Now, something more precise needs to be said concerning criteria of
identity and sortal persistence conditions. A criterion of identity is a
principle expressing a non-trivial logically necessary and sufficient condi-
tion for the identity of objects of a given sort or kind, φ.6 Formally, such a
principle may be stated as follows:
(CIφ) 8x8y((φx & φy) ! (x ¼ y $ Rφxy))
Or, in plain English: for any objects x and y, if x and y are φs, then x is
identical with y if and only if x stands to y in the relation Rφ. To avoid
triviality, we must insist that Rφ – which may be called the criterial relation
for φs – is not simply the relation of identity itself. Rφ must, of course, be
an equivalence relation defined on objects of the sort or kind φ – that is to
say, it must be reflexive, symmetrical, and transitive, and either hold or fail
to hold between any pair of objects of the sort or kind φ. A paradigm
example of a criterion of identity is provided by the axiom of extensionality
6
I shall say much more about criteria of identity in Chapter 5. See also my More Kinds of Being:
A Further Study of Individuation, Identity and the Logic of Sortal Terms (Malden, MA and Oxford:
Wiley-Blackwell, 2009), pp. 16–28.
of set theory, according to which if x and y are sets, then x is identical with y
if and only if x and y have the same members – so that sameness of
membership is the criterial relation for sets. But sets, of course, are abstract
objects. It is rather harder to provide completely uncontentious examples
of criteria of identity for concrete objects of any sort. This should not
surprise us, since our grasp of such criteria is typically implicit rather than
explicit and the criteria themselves are often open to question and revision
in the light of philosophical argument. The criteria implicit in the everyday
use of sortal terms are, moreover, often somewhat vague and shifting. This
would be a defect in a formal language, such as the language of mathemat-
ics, but can hardly be complained about where everyday discourse is
concerned.
Vagueness in our everyday criteria of identity has the consequence that
some everyday questions of identity lack determinate answers, but the vast
majority do not. Consider, for instance, the everyday criterion of identity
for mountains, which is plausibly something like this:
(CIM) For any objects x and y, if x and y are mountains, then x is identical with y if
and only if x and y have the same peak.
Here we are taking mountains to be regions of terrain that are elevated
above their surroundings and which possess a peak – that is, a highest
point. It may be worried that the concept of a peak or highest point is in
some sense more sophisticated than that of a mountain and that this
somehow compromises (CIM)’s claim to be a criterion of identity for
mountains. However, in the first place, I would not wish to claim that
anyone who grasps the concept of a mountain must have an explicit grasp
of the concept of a peak; and, in the second place, it seems clear that the
concept of a peak or highest point is at least implicitly presupposed by that
of a mountain, whereas the reverse is plainly not the case, since many
things other than mountains can possess a highest point. Undoubtedly,
(CIM) is a rather rough-and-ready definition that professional geographers
might take issue with, but it will serve for purposes of illustration.
Now, (CIM) is clearly incapable of resolving some questions of
mountain-identity. For instance, if we have a region of terrain that is
elevated above its surroundings but in which two approximately equally
high points are unsurpassed by any other, with a saddle-shaped dip
between them, (CIM) doesn’t really help us to decide whether what we
have here is a single mountain or two mountains separated by a shallow
valley. But this doesn’t mean that (CIM) is worthless as a criterion of
identity for mountains, since in the vast majority of cases it does supply a
determinate answer to questions of mountain-identity. The same lesson
may be drawn by reflecting on the everyday criterion of identity for
animals, which – to echo Locke7 – is plausibly something like the
following:
(CIA) For any objects x and y, if x and y are animals, then x is identical with y if
and only if x and y participate in the same life.
It may sometimes be hard to determine whether we have a case of two
animals that are vitally connected to one another – as in a case of conjoined
twins, or in a case of a mother and her unborn child – or just a single
animal. And this is because the notion of ‘sameness of life’ is to a certain
extent vague. But other cases are clear-cut: for instance, a rat and a flea that
lives in its fur are clearly two distinct animals according to (CIA), which is
as it should be.
A further lesson that (CIM) and (CIA) serve to reinforce is a point
mentioned earlier, namely that objects belonging to sorts that are governed
by different criteria of identity cannot intelligibly be identified with one
another, with the consequence that one cannot intelligibly suppose that –
to use again our earlier example – a cat could survive a process of
metamorphosis which left it existing ‘as’ a mountain. This is because cats,
being animals, have their identity determined by the relation of sameness of
life, but mountains are simply not living things and consequently cannot
be identified with anything that is essentially alive. Here it may be objected
that I am just assuming without argument that, indeed, any animal is
essentially an animal and so essentially alive. I confess that I am indeed
making this assumption, although it seems to me to be an entirely
reasonable one. To reject it is to suppose, in effect, that ‘animal’ is not,
after all, a categorial term. To suppose that an animal could survive being
changed into a mountain is to suppose that the sortal terms ‘animal’ and
‘mountain’ are both subsumed by some single higher-level categorial term
which supplies a common criterion of identity for both animals and
mountains. But what could this putative categorial term be? It could not
be ‘material object’, since that is pretty clearly transcategorial and supplies
no single criterion of identity for any of the objects to which it applies. The
term applies, after all, to anything that is both an object and composed of
matter. But, it seems clear, there is no single criterion of identity governing
all such objects – bearing in mind here that a genuine criterion of identity
must supply a non-trivial criterial relation for the objects that it governs.
7
See Locke, An Essay Concerning Human Understanding, ii, xxvii, 4.
If this claim – that ‘material object’ is a transcategorial term – is not
immediately obvious to some philosophers, I suspect it is because they may
be prone to confuse it with another term which is, pretty clearly, a
categorial term, namely ‘hunk of matter’. A hunk of matter – or what
Locke called a ‘body’, ‘mass’, or ‘parcel’ of matter – is a quantity of matter
collected into a cohesive whole, which can remain intact under the
impression of an external force and undergo motion as a result. Borrowing
from Locke’s account,8 it is not too difficult to state a plausible criterion of
identity for hunks of matter, as follows:
(CIH) For any objects x and y, if x and y are hunks of matter, then x is identical
with y if and only if x and y are composed of the same material particles
bonded together.
(CIH) has the plausible implication that if some material particles are
removed from a certain hunk of matter, then what remains is, strictly
speaking, a different hunk of matter. The fact that we do not always speak
strictly in such circumstances is not to the point, since most ordinary
speakers can readily be persuaded to agree that they are speaking loosely if
they say that the loss of a few particles leaves us with just the same hunk;
otherwise, indeed, unscrupulous dealers could exploit purchasers of gold
ingots without threat of challenge, by regularly rubbing off a few gold
particles between receiving their fee and delivering the goods. Here it may
be objected that the purchasers are only interested in buying a certain
quantity of gold, not a particular piece or hunk of it. However, a ‘piece’ of
gold just is a certain quantity of gold gathered together into a connected
whole, so the distinction has no bearing on the case. Even so, we should
once again acknowledge a certain amount of harmless vagueness in (CIH),
arising from the fact that it is not always perfectly clear whether or not a
certain ‘material particle’ (itself a somewhat vague term) is ‘bonded’ to
others in a certain hunk of matter with sufficient cohesion to qualify as
being a material part of that hunk.
This much, in any case, is perfectly clear: that no animal is to be
identified with any hunk of matter, nor is any mountain to be so identified,
despite the fact that animals and mountains are both material objects. The
truth, rather, is that, at any time at which it exists, an animal coincides with
a certain hunk of matter, as does a mountain. Thus, if Oscar is a certain cat
existing now, then Oscar now coincides with a certain hunk of matter: but
Oscar is not identical with that hunk, because if some material particles are
8
See Locke, An Essay Concerning Human Understanding, ii, xxvii, 3.
removed from it a different hunk of matter will then coincide with Oscar,
but Oscar will remain the same cat, provided that the removal of those
particles does not terminate Oscar’s life. Similarly, Mount Everest pres-
ently coincides with a certain very large hunk of matter. But if that hunk of
matter were to be transported intact to Australia, this would not be a way
of moving Mount Everest to Australia. Rather, it would be a way of
destroying Mount Everest, since the removal process would have left the
Himalayas without that elevated region of terrain that is Mount Everest.
So far in this section, I have said a good deal about criteria of identity,
but nothing yet about sortal persistence conditions. Sortal persistence
conditions are the conditions that an object of a given sort must comply
with in order to continue to exist as an object of that sort. Clearly, the
criterion of identity governing a given sort of objects must be complied
with by any such object if it is to continue to exist at all, but this is not
enough for its persistence as an object of that sort, since its compliance with
the criterion of identity is compatible with its ‘metamorphosis’ into an
object of another sort governed by the same criterion. A simple example to
illustrate this point is the following. Mount Everest is, obviously, currently
a mountain: but if sea levels were to rise dramatically, it could be trans-
formed into an island. Mountains and islands are different sorts of ‘geo-
graphical prominence’ – to coin a phrase for the categorial term applicable
to them both – since an island is necessarily surrounded by water, whereas
a mountain (a terrestrial mountain, at any rate, as opposed to an undersea
one) is necessarily not. A slightly more controversial example involving
animals is provided by the case of a caterpillar and the butterfly into which
it is eventually transformed. It is apparently the case that, in the chrysalis
stage, the internal structure of the caterpillar is completely destroyed and
reorganized, making this unlike the simple development of an amphibian,
for example, from its larval to its adult phase. Nonetheless, throughout the
metamorphosis of the caterpillar into the butterfly, the same life continues,
making the butterfly the same animal as the caterpillar, even if it is not, in
the relevant sense, the same sort of animal. This example is, of course, a
scientific one which has a basis in empirical fact. But fictional examples of
a similar kind are familiar from folklore, as in the story of the frog prince,
in which a human being is transformed into an amphibian. We can make
sense of this story, even while dismissing it as pure fiction, because we can
intelligibly suppose the prince and the frog to participate in the same
uninterrupted life.
Sortal persistence conditions, to the extent that they go beyond the
demands of criteria of identity, are only discoverable by empirical means.
But in order to discover those conditions empirically, we must first be able
to grasp the relevant criteria of identity. For example, it is only because we
already know or assume that sameness of life is the criterial relation for
animal identity that we can then go on to determine whether, when a
caterpillar is transformed into a butterfly, this is to be classified as an
individual’s surviving a change of sort or merely as its surviving a change of
phase within the same sort. Consequently, criteria of identity, although
they obviously do not lack empirical content, have the status of ‘frame-
work principles’ rather than mere empirical discoveries. Revising or
amending a criterion of identity is, thus, more in the nature of a methodo-
logical or indeed a philosophical exercise than is revising or amending an
account of the sortal persistence conditions governing a sort or kind, at
least where natural kinds are concerned. A great deal more can and should
be said about such matters, but enough has now been said for the purposes
of the present chapter.
3 two notions of individuation

At the beginning of this chapter, I said that I would be defending a certain
thesis regarding the individuation of objects, which I call categorialism. But
in order to make this thesis clear, it is first necessary to distinguish between
two different, albeit related, notions of individuation. The notion of
individuation with which I am chiefly concerned at present is a purely
cognitive notion. Individuation in this sense is a kind of cognitive achieve-
ment, namely the successful singling out of an object in thought. Categorialism
is, then, the doctrine that a thinker can single out an object in this way
only if he or she grasps, at least implicitly, some categorial concept under
which he or she conceives the object in question to fall. When a thinker
thus singles out an object in thought and has a specific thought about that
very object, we have a case of singular, or de re, thought about that object.
And it seems clear that at least sometimes thinkers do have such singular
thoughts. I shall have much more to say about such thoughts shortly.
Before that, however, I need to say something about the other notion of
individuation.
In addition to the notion of individuation in the cognitive sense, we
have the notion of individuation in the metaphysical sense. Individuation
in this sense has nothing to do with thinkers or thoughts – except, of
course, when we are talking about the metaphysical individuation of
thinkers and thoughts themselves, for these, like objects of any other kind,
are subject to principles of individuation. A principle of individuation, in
the metaphysical sense, is a principle telling us how objects of a certain sort
are singled out in reality, as opposed to how they are singled out in thought.
Such a principle tells us what makes an object of a certain sort the very
individual that it is, as opposed to any other individual of the same or
indeed of any other sort. Unsurprisingly, then, a principle of individuation
is closely related to a criterion of identity, although the two notions do not
perfectly coincide. A criterion of identity for objects of a sort φ tells us what
it takes for such objects to be the same φ or different φs. But it doesn’t
necessarily tell us what makes such an object the particular φ that it is. For
example, our criterion of identity for mountains, (CIM), tells us that
mountain A is the same mountain as mountain B if and only if A and B
have the same peak. But it doesn’t appear that a particular mountain, such
as Mount Everest, is individuated by its peak – not, at least, by its peak
alone. To know which mountain Mount Everest is, we need to know not
only which peak is its peak, but also, and more importantly, its geograph-
ical extent. Only when we know that do we know where Mount Everest
‘begins’ and other mountains in the region ‘leave off’. That is to say, only
then do we know how Mount Everest is singled out in reality from other
surrounding mountains, and indeed from the air moving about it and the
animals living on its slopes.
I have just spoken of knowing which mountain Mount Everest is.
Someone who does know this can certainly individuate Mount Everest
in the cognitive sense – can single that mountain out in thought and have
singular thoughts about that very mountain. So, a thinker’s knowledge of
the principle of individuation governing an object, which is intimately
bound up with its criterion of identity, is clearly at least part of a sufficient
condition for that thinker’s being able to single out that object in thought.
The example of sets may provide a useful illustration once more. A set, it
seems clear, is individuated – in the metaphysical sense – by its members,
so that in this case, at least, there is a coincidence between the criterion of
identity for sets and their principle of individuation. However, in order to
know which set a given set is, it does not suffice merely to know that this
set, like any other set, is individuated by its members. What does suffice is
to know not only this but also which objects are in fact the members of the
set in question. We may call a set’s members its individuators. And the
point of the example is to show that a sufficient condition for a thinker’s
being able to single out an object in thought is his or her knowing not only
the object’s principle of individuation, but also its individuators. Mere
knowledge of the object’s individuators does not suffice, without know-
ledge that they are its individuators, which requires knowledge of the
object’s principle of individuation. A child might know which objects are
the members of a certain set – for instance, that 2, 3, 5 and 7 are the
members of the set of prime numbers smaller than 10 – but without its
knowing that these numbers are what individuate that set, it is doubtful
whether the child could really be said to know which set that set is, since it
might be under the mistaken impression that another set could have these
same members.
However, I want to make it quite clear that I am not presuming at this
point that a thinker can single out an object in thought only if he or she
knows which object it is and knows this by knowing not only its principle
of individuation but also what its individuators are. I am merely saying
that knowledge of the latter kind suffices for a thinker’s ability to single out
an object in thought. I also believe that thinkers certainly do sometimes
have knowledge of this kind and are, in virtue of it, able to single out
certain objects in thought. The question, however, is whether such know-
ledge is necessary for that achievement and, if not, what condition is
necessary. Categorialism maintains that one necessary condition is the
thinker’s grasp – which may only be implicit – of a categorial concept
under which he or she conceives the object in question to fall.
4 object perception and singular thought

The case discussed in the previous section, of a thinker’s singling out in
thought a particular set – the set of prime numbers smaller than 10 – is
special in that the object in question is an abstract object, and hence not
one that can be perceived by the senses. It may well be supposed that
singling out in thought an abstract object like this is particularly
demanding intellectually, for precisely this reason. By the same token, it
may be supposed that if an object can be perceived by a thinker, the
thinker’s perception of the object provides him or her with a way of
singling it out in thought which is intellectually much less demanding –
so much so that in cases like this categorialism is a most implausible
doctrine. Of course, this presumes that categorialism does not extend
beyond thought to perception itself: that is, it presumes that it is false to
maintain that a subject cannot even perceive an object without grasping
some categorial concept under which he or she conceives the object in
question to fall. But perceptual categorialism, as we may call this more
extreme doctrine, is a most implausible thesis. It would imply, for instance,
that animals lacking categorial concepts – as, quite plausibly, many higher
mammals do – cannot perceive objects in their environment.
For my own part, I am content to deny perceptual categorialism.
I favour a causal theory of object perception according to which – to a
first approximation – a subject S perceives an object O if and only if certain
features of S ’s perceptual experience, discriminable by S, are differentially
causally dependent on certain properties of O.9 A feature F of S’s experi-
ence is differentially causally dependent on a property P of O just in case
variations in P would cause corresponding variations in F. So, for example,
by this account I currently see a brown wooden table standing in front of
me if, say, variations in its shape, orientation, or hue would give rise to
corresponding variations in certain features of my current visual experi-
ence, discriminable by me. If variations in properties of the table would
give rise to no variations in my current visual experience, or only to
variations indiscriminable by me, then, I think, I do not see the table.
This account of object perception imposes no direct epistemic requirement
on such perception: a subject is not required, by this account, to know
what object he or she perceives, nor even that he or she is perceiving an
object. At most the account defines object perception in such a way that a
subject’s perception of an object makes it possible for the subject to know
what he or she is perceiving, and that he or she is perceiving it, provided
that he or she meets certain other requirements for such knowledge, yet to
be specified. For instance, it might be insisted, as a minimum requirement,
that a subject cannot know what he or she is perceiving, or that he or she is
perceiving it, unless he or she is attending to the object in question.
I should also make it clear that, in saying that the relevant variations in
the subject’s experience should be discriminable by the subject, I am not
smuggling in an epistemic requirement on object perception. For I do not
require the subject to be able to form discriminatory judgements regarding
features of his or her perceptual experience – only to be sensitive to relevant
variations in his or her perceptual experience. A subject could be tested for
such sensitivity without being asked whether or not he or she could notice
any experiential difference of a relevant kind. This is still only a rather
sketchy account of object perception, but it will suffice for the discussion
that is to follow.
Categorialism is likely to be challenged on the basis of examples such as
the following. A subject, S – perhaps an infant or a young child – is
confronted with an object, O, which S sees. Perhaps, furthermore, S directs
his or her attention towards O and, as O moves, S tracks O’s movement
with his or her eyes, continuing to direct his or her attention towards O.
9
See my Subjects of Experience (Cambridge University Press, 1996), pp. 102–17.
Now let the question be asked: can S, in virtue of this sort of perceptual
connection with O, have singular thoughts specifically about O – thoughts
about O such as that it is white or that it is furry? Of course, we must
presume that S can indeed think and, moreover, can have thoughts of the
form ‘X is white’ and ‘X is furry’. Our question is only whether, in order to
have such singular thoughts about O in particular, all that is necessary in
addition to these general cognitive capacities is that S should have the kind
of perceptual connection with O that has just been described. A good
many theorists would, I am sure, answer this question positively. But a
categorialist would not. A categorialist will insist that, unless S conceives of
O in a certain way, even if only implicitly, and is broadly correct in so
conceiving of O, S’s perceptual connection with O does not enable S to
have singular thoughts about O. The conceptual requirement in question
is this: S must, at least implicitly, conceive of O as falling under a certain
categorial concept. This requirement will be met if, say, S conceives of O as
being an animal. Thus, if S is perceptually connected with O in the
described fashion, S conceives of O as being an animal, and O is an animal,
then S is thereby put in a position to have singular thoughts about O. But
it is not necessary that S should explicitly conceive of O as being an animal.
It suffices that S should conceive of O as being some specific sort of animal,
such as a cat. And in that case it is not necessary that O should actually be a
cat, only that it be some kind of animal. Thus S can be mistaken about
what sort of animal O is and still be put in a position to have singular
thoughts about O. But, according to categorialism, S cannot be put in a
position to have singular thoughts about O, even if S is perceptually
connected with O in the described fashion, if S either fails, even implicitly,
to conceive of O as falling under any categorial concept at all or else
misconceives the putative object of his or her perception as being some-
thing falling under a categorial concept which does not apply to O.
It seems to me that this contention is susceptible of proof, or at least to
the nearest thing to proof that is normally available in questions of
philosophy. Let us suppose that O is, in fact, a cat. That is to say, let us
suppose that S is perceptually connected to a certain cat in the described
fashion: S sees the cat, S directs his or her attention towards the cat, and
S tracks the cat’s movements with his or her eyes, continuing to direct his
or her attention towards the cat. Call the cat in question ‘Oscar’. But let us
also suppose that S does not conceive of Oscar as being a cat, nor indeed as
falling under any sortal or categorial concept. Then the anti-categorialist is
presented with the following difficulty. As we have already observed,
whenever a certain animal exists in a certain place at a certain time, it
coincides with a certain hunk of matter, which is numerically distinct from
the animal in question. Now, the perceptual connection that S has with
respect to Oscar, the cat, S will very likely also have with respect to a
certain hunk of matter – namely the hunk of matter composing Oscar
during the perceptual episode in question. I say ‘very likely’ because, of
course, it is at least possible that different hunks of matter should compose
Oscar at the beginning and end of the episode – if, for example, Oscar
should lose some hair or ingest some food during the course of that
episode. However, let us set aside this possibility for the time being and
return to it later, since it will turn out to be an important one.
The problem for the anti-categorialist is now to explain why S’s percep-
tual episode should put S in a position to have singular thoughts specific-
ally about Oscar, as opposed to the hunk of matter – call it ‘Hunk’ – that,
as we are now supposing, composes Oscar throughout the episode. If the
anti-categorialist maintains, in the light of this challenge, that S is put in a
position to have singular thoughts about both Oscar and Hunk, then
another problem arises: what makes one of those thoughts specifically a
thought about Oscar and another specifically a thought about Hunk?
A genuinely singular, or de re, thought must determinately be a thought
about a particular, uniquely identifiable object. I am not insisting that the
thinker of the thought must be able to identify that object – to do so would
clearly be question-begging in the present context – only that there must
be some determining factor which makes the thought in question a thought
about the particular object in question. But, as far as I can see, the anti-
categorialist has no resources with which to say what this determining
factor could be, since in a case such as that just sketched two different
objects are symmetrically related to the thinker in the only way that the
anti-categorialist requires for singular thought about them. By contrast, the
categorialist has a tie-breaker for this kind of situation. According to
categorialism, what makes Oscar the object of one of S’s thoughts is the
fact that S conceives of one of the objects being perceived as being a cat –
or at least as being an animal or some sort of animal – and is correct in
doing so.
A discussion is now needed of the more complicated case in which, say,
Oscar loses some hair or ingests some food as it moves through S’s field of
view. If S continues to track Oscar, can’t the anti-categorialist maintain
that this is why S is put in a position to have singular thoughts specifically
about Oscar? For, of course, in this case there is no single hunk of matter
that S is tracking. Moreover, the anti-categorialist may now go on to claim
that the original example, in which Oscar does not change in material
composition, only presents a special case of this more complicated one.
The anti-categorialist may urge that what puts S in a position to have
singular thoughts specifically about Oscar in all such cases, complicated or
simple, is a fact expressible by the subjunctive conditional that S would
continue to track Oscar if Oscar were noticeably to lose or gain some
appropriate component matter, such as some hair or some food. However,
far from this suggestion serving to aid the anti-categorialist cause, I think
that it effectively concedes victory to categorialism. For recall that I insisted
only that S should at least implicitly conceive of the object of perception –
in this case, Oscar – as falling under some categorial concept which does in
fact apply to that object, namely, animal. I don’t require S to be able to
articulate this concept or its associated criterion of identity explicitly. But if
S is demonstrably capable of perceptually tracking a particular animal
through various changes to its material composition which are specifically
permitted by the criterion of identity for animals, but not by the criteria of
identity governing other categorial concepts, including that of hunk of
matter, then this provides strong evidence that S does in fact possess an
implicit grasp of the categorial concept of an animal. And bear in mind
here that not just any change of material composition is compatible with
the continuing existence of an animal.
Indeed, we can now see more clearly why it is categorial concepts that
are crucial to the capacity for singular thought about objects encountered
in perception, for it is precisely these concepts that are most intimately tied
to criteria of identity. In cases in which two initially coinciding objects,
such as a cat and a hunk of matter, go their separate ways in the course of
some extended perceptual episode, the correct way to track those different
objects will be determined precisely by their respective criteria of identity –
whence it is reasonable to conclude that subjects who succeed in correctly
tracking those objects exhibit at least an implicit grasp of the criteria of
identity governing those objects, and thereby an implicit grasp of the
categorial concepts under which they fall and which determine those
criteria of identity. In short, the very perceptual tracking considerations
that the anti-categorialist is prone to regard as defeating the claims of
categorialism in fact serve to support those claims, when these consider-
ations are given their proper scope.
What if it should turn out that S – who, remember, we have supposed
to be an infant or young child – doesn’t consistently and comprehensively
track Oscar in all kinds of circumstances, in accordance with the criterion
of identity for animals, but still clearly does so in certain limited kinds of
circumstances? Suppose, for instance, that S tracks Oscar successfully
through the loss of some hair and the ingestion or excretion of food matter,
or through changes in the dispositions of Oscar’s limbs and tail, but that S
fails to do so through periods in which Oscar is curled up and asleep. Then
I think we should say that S hasn’t fully grasped, even only implicitly, the
adult concept of an animal and its associated criterion of identity, perhaps
because S doesn’t yet understand that an animal can still be alive when it is
asleep. But I still think that S is evincing the implicit grasp of some
categorial concept with an associated criterion of identity – a concept
which we might describe as being an infantile precursor to the fully fledged
adult concept of an animal (a precursor concept which might well be
innate in humans). And we may deem that this grasp is sufficient for S to
have singular thoughts about Oscar, at least with respect to the limited
kinds of circumstances in which the infantile criterion does not signifi-
cantly come apart from the adult one in terms of its implications for the
identity of the tracked object. A suitably nuanced version of categorialism
should certainly be prepared to admit such qualifications as these to its
basic doctrine.
5 categorialism and linguistic reference

From this point forward, I am going to assume that categorialism is
correct. But categorialism is a thesis about singular thought, not about
linguistic reference. Even so, it may – indeed, I think it does – have
important implications for linguistic reference. According to currently
dominant ‘causal’, or ‘direct’, or ‘anti-descriptivist’ theories of reference,
all that is required for a speaker’s use of a proper name or natural kind
term, N, to be referentially successful – that is, for the speaker to use N to
refer successfully to N’s referent – is that the speaker’s use of N should
stand in the right kind of causal-historical relation to a ‘reference-fixing’
event in which N was first introduced into the speaker’s speech-
community.10 The ‘right kind’ of causal-historical relation consists, sup-
posedly, in a chain of usage-acquisition, where each speaker S in the
chain – apart, of course, from the first – uses N with an intention to refer
to the same thing as did the speaker S 0 from whom S acquired the use of N.
The first speaker in the chain is the person who first introduced N into the
speech-community. According to this sort of account – which, to be fair,
has only been sketched in a very bare-bones way here – successive speakers
10
The locus classicus for this sort of view is, of course, Saul A. Kripke, Naming and Necessity (Oxford:
Blackwell, 1980).
in the chain of acquisition do not need to associate with N any descriptive
information which applies to N’s referent in order to be able to use N
successfully to refer to that referent. Hence, in particular, it is not the case,
on this view, that the referent of N, as used by a given speaker, is
determined by a certain definite description which the speaker associates
with N.
Now, I am happy to concede that there may be a relatively attenuated
and anodyne notion of ‘reference’ for which this sort of account is broadly
acceptable. But if categorialism is a correct doctrine concerning singular
thought, such an account of linguistic reference cannot be the whole story.
This is because we must acknowledge that, at least sometimes, speakers use
sentences containing proper names and natural kind terms to express
singular thoughts. Suppose that some friends of my neighbours have
acquired a new pet cat, which they decide to call ‘Oscar’. And suppose
that I overhear a conversation between my neighbours and their friends
during the course of which they say, amongst other things, that Oscar is
white and beautiful, but never say anything that reveals, either explicitly or
implicitly, that Oscar is a cat. According to the causal-historical theory of
linguistic reference, I am now in a position to refer successfully to Oscar,
provided I use the name ‘Oscar’ with the intention to refer to whatever it
was that my neighbours and their friends were referring to when they used
the name in the conversation that I overheard. The only constraint on my
referential success with the name ‘Oscar’ is supposed to be that I should
intend to use it to refer to the same thing that my neighbours and their
friends were using ‘Oscar’ to refer to. But ‘thing’ is not a categorial term: it
is a transcategorial term, like ‘object’. It carries with it no specific criterion of
identity. Consequently, this supposed ‘constraint’ on my successful use of
the name ‘Oscar’ is very slim indeed. As far as I am concerned, Oscar
might be almost any kind of thing whatever. Indeed, if I were honest
I would have to confess that I have no idea what kind of thing Oscar is.
Now, as I say, I am relatively content to acknowledge that there is some
sense in which I am referring to Oscar, even when I say that I have no idea
what kind of thing Oscar is. But the more important question, I think, is
whether, in the scenario just sketched, I can intelligibly be said to be able
to use the name ‘Oscar’ in sentences to express singular thoughts about
Oscar. I don’t see how I can, given that categorialism is a correct doctrine
concerning singular thought. For categorialism requires that, in order for
me to have a singular thought about Oscar, I must at least conceive of
Oscar as falling under some categorial concept, and correctly so conceive of
him – so that, since Oscar is in fact a cat, I must at least conceive of Oscar
as being an animal of some sort. But in the envisaged scenario I do not
conceive of Oscar in this way, nor do I conceive of Oscar as falling under
any other categorial concept.
Here it may be urged that this just shows that categorialism must be
wrong, on the grounds that I can at least use the sentence ‘I have no idea
what kind of thing Oscar is’ to express a singular thought of mine about
Oscar. But that response is question-begging in the present context, given
that I have just spent a good deal of time defending categorialism. Merely
to assert that such a sentence may be used to express a singular thought
about Oscar is no rebuttal of my previous arguments. At most it can be
demanded of me that I give a plausible account of what sort of thought
such a sentence could be supposed to convey, assuming the truth of
categorialism. But this is surely not difficult. The obvious answer, which
I think is a plausible one, is that ‘I have no idea what kind of thing Oscar
is’ is implicitly metalinguistic. The thought that I am conveying by it is the
thought that I have no idea what kind of thing the referent of the name
‘Oscar’ is. Here it might be objected that this only postpones the problem,
because I have just replaced the name ‘Oscar’ by the singular noun phrase
‘the referent of the name “Oscar”’. But, of course, even by the lights of the
causal-historical theory of reference, this definite description should not be
assimilated to a proper name. Rather, we can explain its standard use in
terms of something like Russell’s theory of descriptions; that is, in quanti-
ficational terms. Indeed, another way of expressing the thought conveyed
by my use of the sentence ‘I have no idea what kind of thing Oscar is’
would be something like this: the name ‘Oscar’ refers to something, but I’ve
no idea what kind of thing it is. And, of course, the ‘it’ in ‘I’ve no idea what
kind of thing it is’ is not playing a referential role, but rather the role of a
variable of quantification, bound by the preceding quantifier ‘something’.
So, my contention is that, when a speaker uses a sentence containing a
proper name or natural kind term to express a singular thought about the
referent of that name or natural kind term, he or she must at least possess
implicit knowledge of what categorial concept the referent falls under. This
knowledge will obviously be descriptive knowledge, of a certain kind, about
the referent. But in saying this I am not simply returning to the ‘descrip-
tive theory of reference’, which causal-historical accounts of reference
sought to overturn. For I am not contending that the descriptive know-
ledge of the referent that the speaker must possess in order to use the
proper name or natural kind term in sentences expressing singular
thoughts about the referent must be descriptive knowledge which deter-
mines precisely which thing is that referent. On the other hand, nor am
I arguing against the latter view. Rather, at this point I would rather keep
an open mind about the matter.
Another important application of categorialism to the theory of linguis-
tic reference concerns the notion of a ‘reference-fixing event’, of the kind
involved, according to the causal-historical account of reference, when a
proper name or natural kind term is first introduced into a speech-
community. Let us return to the story of my neighbours’ friends’ pet
cat. It was said that the friends decided to call their new pet cat ‘Oscar’.
But to do that, they needed to have singular thoughts about Oscar. It may be
that they first saw Oscar in a pet shop and, looking at him, decided there
and then to call him ‘Oscar’. Perhaps one of the friends said to the other,
‘Let’s buy him and call him “Oscar”’. But this required the friend to use
the pronoun ‘him’ in a sentence expressing a singular thought about the cat
in front of them. Consequently, according to categorialism, this friend had
to conceive of that cat at least as being an animal of some sort.
What, though, if these people were seriously mistaken about the nature
of the shop they had entered. Suppose that they thought it was a china
shop and that what they were looking at was a very lifelike china ornament
of a cat (perhaps Oscar was fast asleep at the time). Wouldn’t they still
have succeeded in having singular thoughts about that cat and have
succeeded in naming him ‘Oscar’? It might appear that common sense is
on the side of those who answer this question positively, but we need to
think the matter through more carefully. What entitles us to assume that
what the friends would have succeeded in deciding to call ‘Oscar’ was the
cat that they were looking at, given that they didn’t conceive anything that
they were looking at to be a cat? It won’t do to answer here that the thing
that they were looking at was in fact a cat, because this presumes that there
was just one thing that they were looking at. But, as was remarked in an
earlier section, whenever an animal of any sort is located at a certain place
at a certain time, it coincides with a certain hunk of matter, which is
distinct from the animal in question. So there were – at least – two things
that the friends were looking at, a cat and a hunk of matter coinciding with
that cat. Why, then, should we suppose that, despite their mistake, they
succeeded in calling the cat ‘Oscar’ rather than the hunk of matter? To this it
may be replied that they certainly didn’t intend to be naming a hunk of
matter. So the suggestion is that, by default, what they succeeded in
naming was the cat. But that suggestion is question-begging in the present
context, for it assumes that they did indeed succeed in naming something –
and yet whether they did so is precisely the question now at issue. In any
case, how can something be named by default in this supposed fashion?
How can an intention not to name one thing somehow bring it about that
another thing is named?
The suggestion that, in the scenario just described, my neighbours’
friends succeeded in naming a cat ‘Oscar’, despite being under the impres-
sion that they were naming a china ornament, also faces the following
difficulty. Suppose that the friends never discovered their mistake, because
they were suddenly called away and never managed to return to the shop.
And suppose that, later, reminiscing with my neighbours about the inci-
dent and expressing their regret at not having been able to make their
purchase, they passed the name ‘Oscar’ on to my neighbours. And suppose
that my neighbours also believed that ‘Oscar’ was the name of a certain
china ornament that their friends had intended to purchase. According to
preceding arguments of this section, my neighbours are not in a position to
express singular thoughts about the cat that their friends actually saw, when
using sentences containing the name ‘Oscar’. And that seems right. If they
say such things as ‘Oscar would probably have been broken by now,
because our friends are so clumsy with ornaments’, it surely makes no
sense to suppose that they are expressing such thoughts about a certain cat.
But if that is the right thing to say about my neighbours’ use of the name
‘Oscar’ – that it simply fails to refer – then it is surely also the right thing to
say about their friends’ use of it, even though it was they who introduced
the name in the first place. Certainly, this is the verdict of categorialism
and I consider it to be a virtue, not a failing, of the doctrine that this is so.
Finally, I should note that although I have been concentrating on the
reference of proper names in this section, I intend all of its lessons to apply
also to natural kind terms, although the adjustments to the account that
would be needed to extend it to these would take up more space than the
present chapter allows.
chapter 3
Two styles of predication – dispositional

and occurrent
In the preceding chapter, I devoted a considerable amount of discussion to

categorial terms and concepts, although my focus was exclusively on
categories of objects. Here I want to expand this discussion much more
widely to embrace other categories of entities, including properties. Now, in
recent years, I have been developing a neo-Aristotelian system of categorial
ontology which I call the four-category ontology.1 In this system,
exemplification – understood as a formal ontological relation between
particulars and universals – is not regarded as a primitive formal ontological
relation but, rather, as having two different species, the dispositional and
the occurrent, each of which is analysable with the aid of two formal
ontological relations which are regarded as primitive: instantiation and
characterization. Corresponding to this ontological distinction with regard
to exemplification, there is a logical distinction to be drawn between
dispositional and occurrent predication. In the present chapter, I want to
develop this line of thought in greater detail and at the same time clear up
some difficulties that a number of my critics have claimed to find in my
account of these matters.
Before proceeding, I should mention that the distinction between
dispositional and occurrent predication appears to be clearly marked in
natural language by certain syntactic differences, so that my ontological
distinction between dispositional and occurrent exemplification is partly
intended to explicate these syntactic differences. Most notably, verb-
phrases in natural language exhibit not only differences of tense, but also
differences of what grammarians call aspect. For instance, the verbs in the
sentences ‘This chemical substance dissolves in water’ and ‘This chemical
substance is dissolving in water’ are both present-tensed, but the first has a
‘habitual’ aspect whereas the second has a ‘continuous’ aspect. We
1
See especially my The Four-Category Ontology: A Metaphysical Foundation for Natural Science
(Oxford: Clarendon Press, 2006).
31
understand the first as attributing a habitude or disposition to the subject of
the sentence and the second as attributing an ongoing activity to that same
subject – an activity which constitutes the manifestation or exercise of the
disposition in question. Indeed, the first sentence may be paraphrased as
‘This chemical substance is water-soluble’, in which a dispositional adjective
is explicitly used to describe the subject – such adjectives typically being
formed from the stem of a verb plus a suffix of the form ‘-able’, ‘-ible’, or ‘-
uble’, as in ‘breakable’, ‘fusible’, and ‘(dis)soluble’. It is also notable that
when the subject of a sentence whose verb has the ‘habitual’ aspect is a
kind of thing or stuff – rather than a particular instance of such a kind – as
in ‘Sodium chloride dissolves in water’, the sentence in question expresses
what philosophers of science would call a lawlike generalization, or natural
law (in this case, a chemical law). These points will all be seen to be highly
relevant to the discussion that follows.
1 the four-category ontology and its logic

The four basic ontological categories of my system are these: (1) substan-
tial universals, (2) non-substantial universals, (3) substantial particulars,
and (4) non-substantial particulars – or, less long-windedly and more
memorably, kinds, attributes, objects, and modes.2 The latter terminology,
although convenient, is, I concede, not entirely perspicuous. For one
thing, the term ‘attribute’ might be thought to embrace only monadic
universals, whereas I am happy to include relational universals in my
ontology. For another, the terms ‘object’ and ‘kind’ both have common
uses in metaphysics that are much broader than mine are intended to be.
Thus ‘object’ is sometimes used as a synonym for the all-purpose term
‘entity’, while ‘kind’ is often used as an alternative to ‘type’, in the sense
in which the latter figures in the so-called ‘type–token distinction’.
(In other words, ‘type’ is often used in a way in which it is pretty much
interchangeable with ‘universal’, thus ignoring entirely my own distinc-
tion between substantial and non-substantial universals.) Finally, ‘mode’
may strike many as archaic-sounding in comparison with the currently
more fashionable term ‘trope’ – although I would urge that the former is
in fact more, rather than less, perspicuous than the latter, since it is
2
I should perhaps stress, if it is not already sufficiently obvious, that the basic ontological categories of
which I am now speaking lie at a much higher level of abstraction than the categories that were the
chief concern of Chapter 2, such as the category of animals and the category of material artefacts. The
latter categories are only sub-categories of the basic category of substantial particulars, inasmuch as all
individual animals and material artefacts qualify as ‘individual substances’.
Two styles of predication – dispositional and occurrent 33
appropriately suggestive of the idea that properties, whether they are
universals or particulars, are rightly to be thought of as ways of being.
(The term ‘trope’ has, furthermore, a standard literary use which has no
connection whatever with its current use in metaphysics – a fact that can
only serve to render the latter use confusing.)
Since, however, any choice of terminology in this area of metaphysics is
bound to involve some departure from common usage, it may help at this
point if I supply some everyday examples of items that at least appear to fall
into the four categories that I have in mind when I present my ontological
system. I must stress that these can only be taken, at this stage at least, to
be apparent examples because, although I certainly want to defend the
four-category ontology as a foundation for metaphysical inquiry, I do not
want to be committed without argument to supposing that everyday
language and thought provide us with incontestably correct illustrations
of its applicability to the real world. We should be prepared to allow that
the four categories are best illustrated, in fact, only by entities postulated in
advanced scientific theories, rather than by those assumed in our
‘common-sense’ ontology. So, with this caveat in mind, here are some
putative examples of the four categories. A particular table, rock, or dog
would, then, be an example of something belonging to the category of
object, as I conceive of it. Such items are more traditionally known as
individual substances. Corresponding examples of the category of kinds, as
I conceive of it, would be the kinds table, rock, and dog of which the
foregoing objects are, respectively, particular instances. Examples of the
category of attribute would be the properties, conceived as universals, of
brownness, hardness, and furriness that are exemplified, respectively, by
those objects. And, finally, examples of the category of mode would be
the particular instances of those universals possessed by those objects: the
table’s particular brownness, the rock’s particular hardness, and the dog’s
particular furriness.
At least implicitly, I have already drawn on both of my aforementioned
primitive formal ontological relations – instantiation and characterization –
in introducing and describing the foregoing examples. For, first of all, both
kinds and attributes are instantiated by, respectively, the objects and modes
that are their particular instances – for example, the kind table by particular
tables and the attribute brownness by particular brownnesses. And, second,
in my terminology, to say that an object possesses a mode of a certain
attribute is just another way of saying that that mode characterizes – or, if
one prefers an older expression, inheres in – the object in question. Indeed,
I want to go further and say that, likewise, characterization is a relation in
Kinds characterized by Attributes
instantiated by exemplified by instantiated by
Objects characterized by Modes
Figure 3.1. The ontological square, version i
which, at the level of universals, attributes stand to kinds. I shall say much
more about this in due course, but already we have the materials to
construct, in Figure 3.1, a version of the ontological square: a diagram that
is enormously useful for the purposes of depicting the formal ontological
relationships in which items belonging to the four different categories
stand to one another.
It may be observed that the upper level of the ontological square,
occupied by kinds and attributes, is the level of universals, while the lower
level of the square, occupied by objects and modes, is the level of
particulars. Similarly, we can call the left-hand side of the square, occupied
by kinds and objects, the side of subjects and the right-hand side of the
square, occupied by attributes and modes, the side of properties – subjects
being entities that are characterized in various ways and properties being
entities that characterize in various ways. Indeed, using this terminology,
we could speak of the four fundamental ontological categories depicted in
the square, beginning at the bottom left-hand corner and proceeding
clockwise around it, as being those of particular subjects, universal subjects,
universal properties and particular properties. In that case, however, it should
be clearly understood that the expressions ‘universal’, ‘particular’, ‘subject’
and ‘property’ do not themselves signify ontological categories as such but
are, rather, cross-categorial terms, just as the all-purpose ontological term
‘entity’ is. Figure 3.2 is another version of the ontological square depicting
this aspect of the four-category ontology.
At this point, it will be useful for me to introduce some logical
symbolism, for in what follows we shall be concerned quite as much with
the logic as with the metaphysics of the four-category ontology. Standard
first-order predicate logic with identity deploys only a single class of
constants and variables – objectual ones – and a way of representing
Universals
Kinds Attributes
Subjects Properties
Objects Modes
Particulars
Figure 3.2. The ontological square, version ii
explicitly only a single formal ontological relation, identity. In a logic that is

capable of representing perspicuously all of the metaphysically important
features of the four-category ontology, we need, however, four different
classes of constants and variables, together with ways of representing
explicitly three different primitive formal ontological relations – not just
identity, but also instantiation and characterization.
So this is what I propose. We shall use, as is already customary, lower-
case letters from the beginning of the Roman alphabet, a, b, c, . . . as object
constants and lower-case letters from the end of the Roman alphabet, x, y,
z, . . . as object variables. For kind constants we shall use lower-case letters
from the beginning of the Greek alphabet, α, β, γ, . . ., and for kind
variables we shall use lower case letters from the end of the Greek alphabet,
φ, χ, ψ, . . . For attribute constants we shall, in mimicry of existing custom,
use the upper-case Roman letters F, G, H, . . . and for attribute variables
we shall use upper-case letters from the end of the Roman alphabet, X, Y,
Z, . . . Finally, for mode constants we shall use the lower-case Roman letters
f, g, h, . . ., while for mode variables we shall use the lower-case Roman
letters r, s, t, . . . As for the three primitive formal ontological relations, we
shall, as is customary, represent identity by the equality sign, ‘¼’, and
supplement this with the slash, ‘/’, to represent instantiation. Finally, we
shall represent characterization by simple juxtaposition of appropriately
chosen constants or variables. Thus, for example, ‘a/β’ and ‘f/G’ say,
respectively, that object a instantiates kind β and that mode f instantiates
Table 3.1. Logical symbolism for the four-category ontology
Objects Kinds Attributes Modes
constants a, b, c, . . . α, β, γ, . . . F, G, H, . . . f, g, h, . . .
variables x, y, z, . . . φ, χ, ψ, . . . X, Y, Z, . . . r, s, t, . . .
Instantiation: a/β, f/G

Characterization: af, βG
dispositional exemplification
Kinds Attributes
Objects Modes
occurrent exemplification
Figure 3.3. The ontological square, version iii
attribute G, while ‘af ’ and ‘βG’ say, respectively, that object a is charac-
terized by mode f and that kind β is characterized by attribute G.
The foregoing proposals are presented in more convenient tabular form
in Table 3.1.
Since I do not regard exemplification as being a primitive formal onto-
logical relation, I do not need an undefined symbol to represent it.
Exemplification, as Figure 3.1 implies, is a relation between objects and
attributes. Or, more exactly, there are two different relations of exemplifi-
cation between objects and attributes, corresponding to the two different
routes from the bottom left-hand corner of the ontological square (the
object corner) to the upper right-hand corner (the attribute corner). For
reasons which will become more apparent in due course, I call these two
different species of exemplification dispositional and occurrent
exemplification, which may be depicted on the ontological square as in
Figure 3.3.
Using the expressions ‘D[a, F]’ and ‘O[a, F]’ to say, respectively, that
object a exemplifies attribute F dispositionally and that object a exemplifies
attribute F occurrently, I propose that we may define these two species of
exemplification as follows:
D[a, F] ¼df 9φ(φF & a/φ)
O[a, F] ¼df 9r(ar & r/F)
In other words, an object a exemplifies an attribute F dispositionally just in
case a instantiates some kind that is characterized by F, while an object a
exemplifies an attribute F occurrently just in case a is characterized by some
mode that instantiates F.
2 cross-categorial ontological dependencies

Before I say any more about the crucial issue of the two different species of
exemplification, I want to present yet another version of the ontological
square, this time one which represents the metaphysically significant relation-
ships of ontological dependency that are characteristic of entities belonging to
the four different ontological categories.3 First of all, then, I need to point out
that, true to the Aristotelian spirit in which the ontological square is con-
ceived, the four-category ontology embodies an immanent realist view of
universals, according to which it is an essential feature of any universal that
it has particular instances, which provide the ground of its existence.
According to this view, there are, then, no uninstantiated universals and every
universal stands in a relationship of existential dependence to its particular
instances. I shall call the relationship in question weak existential dependence.
I call it ‘weak’ for the following reason: although, according to this view, a
universal must have particular instances, which constitute the ground of its
existence, a universal does not depend for its very identity on the particular
instances that it happens to have, which is a purely contingent matter. Thus,
for example, although the attribute brownness (assuming there to be such an
attribute) must have particular instances in the form of the particular brown-
nesses of various brown objects, that very same attribute could have existed
even if those particular brownnesses had not, provided that other particular
brownnesses had existed – for example, if other objects had been brown.
Next, I want to say that the modes of an object stand to that object in
another relationship of existential dependence, which I shall call strong
existential dependence – ‘strong’ because in this case the modes do depend
for their very identity on the object that they characterize. Thus, for example,
what distinguishes one particular brownness from another exactly resembling
3
For more on the notion – and the varieties – of ontological dependence, see my The Possibility of
Metaphysics: Substance, Identity, and Time (Oxford: Clarendon Press, 1998), Chapter 6.
brownness is, precisely, the object that possesses it, or in which it ‘inheres’ – the
implication of this being that modes cannot be ‘transferred’ from one object
to another and cannot exist ‘unattached’ to any object. Clearly, since one
object must possess many different modes, whereas the same mode cannot be
possessed by many different objects, the relationship of strong existential
dependence between modes and their object is a many–one relation.
Finally, I want to say that a kind stands in a one–many relationship of strong
existential dependence with its various attributes: that is, that one kind must
be characterized by many different attributes and that it depends for its very
identity on the attributes that characterize it (or, at least, on those of its
attributes that are ‘part of its essence’, if a distinction may be drawn between
these and its ‘accidental’ attributes). Thus, for example, I would want to say
that the kind electron depends for its identity on the specific attributes of
charge, spin, and rest mass that characterize that kind of fundamental physical
particle. (Electrons carry unit negative charge, have a spin of one-half and have
a certain specific rest mass, differing from all other kinds of fundamental
physical particle in at least one of these respects.)
Figure 3.4, then, is a version of the ontological square which represents
the foregoing relationships of ontological dependency between items
located at the different corners of the square, where a solid-headed arrow
signifies strong existential dependence and an open-headed arrow signifies
weak existential dependence.
It will be noticed that, in Figure 3.4, each corner of the ontological square
differs from every other corner in respect of the number and type of arrows
that proceed from or lead to it. The object corner has two arrows, one of each
type, leading to it. The kind corner has two arrows, one of each type,
proceeding from it. The attribute corner has one solid-headed arrow leading
to it and one open-headed arrow proceeding from it. And the mode corner has
Kinds Attributes
one-many
many-one
Objects Modes
Figure 3.4. The ontological square, version iv

one open-headed arrow leading to it and one solid-headed arrow proceeding
from it. Thus objects are represented as being the least dependent of all entities,
in keeping with another Aristotelian sentiment, namely that individual sub-
stances are ontologically more basic than entities belonging to any other
ontological category, insofar as they constitute the ultimate ground of all
being. Moreover, because the four corners of the ontological square can be
unambiguously identified in terms of the number and type of arrows leading
to or proceeding from them – in other words, because the four categories can
be unambiguously differentiated from one another in terms of the character-
istic relationships of ontological dependency in which their members stand to
the members of the other categories – the four-category ontology is not
subject to the kind of objection that Frank Ramsey and others have raised
against simpler systems of categorial ontology which appeal merely to the
subject–property or universal–particular distinctions.4 This objection is that
the advocates of these systems cannot explain, in a non-question-begging way,
what distinguishes each of the categories in such a system from another
category in the same system and thus what renders each category unique.
We could call this the problem of categorial uniqueness – and the important
point is that the four-category ontology has the distinct advantage over some
of its apparently more parsimonious rivals of being able to solve it.
Incidentally, before I move on from the subject of the ontological
square, here is a useful mnemonic to remind one of the order of its four
corners, starting, in appropriately Aristotelian fashion, with the bottom
left-hand object corner: OcKhAM – Objects, Kinds, Attributes, and Modes.
There is some irony in this mnemonic, of course. William of Ockham
himself, who did not favour realism concerning universals, would certainly
not have approved of the four-category ontology. And, indeed, many
present-day ontologists would no doubt attempt to wield Ockham’s
notorious razor against my system, accusing me of multiplying entities
beyond necessity. However, one of the objectives of my work on the four-
category ontology has been precisely to show how that ontology, while it is
undoubtedly richer than many of its popular rivals – such as the pure trope
ontology, the ontology of objects and attributes, and the ontology of
objects and modes, each of which limits itself to only one or two corners
of the ontological square – deserves our allegiance on account of its
superior capacity to resolve a wide range of metaphysical problems, such
as the problem of categorial uniqueness, which we have just discussed, and
4
See F. P. Ramsey, ‘Universals’, in his The Foundations of Mathematics and Other Logical Essays
(London: Kegan Paul, 1931). For further discussion, see my The Four-Category Ontology, Chapter 7.
the so-called inference problem concerning the relationship between laws of
nature and particular matters of fact. This latter problem, indeed, is the
one that I shall focus on next.
3 laws of nature and the inference problem

As I have already indicated, in my view all kinds are characterized by certain
attributes – and laws of nature, by my account, consist precisely in such
facts of characterization solely involving universals.5 All objects, on the
other hand, are characterized by modes that instantiate certain attributes –
and it is in their being so characterized that particular occurrent facts or
states of affairs consist. By contrast, particular dispositional facts or states of
affairs consist in objects instantiating kinds that are characterized by certain
attributes – in other words, they consist in objects being subject to certain
laws governing the kinds to which they belong. What Bas van Fraassen and
others call ‘the inference problem’ does not afflict this account of laws, as it
does those of David Armstrong, Michael Tooley, and Fred Dretske –
accounts which, like mine, locate laws in the domain of universals.6 The
complaint, as it applies to Armstrong’s theory, is just this. According to
Armstrong, a law of nature (at least, in the simplest sort of case) consists in
the obtaining of a second-order relation of natural necessitation, N,
between first-order universals or attributes – for example, F and G – which
may be represented by a statement of the form ‘N(F, G)’ (‘F-ness necessi-
tates G-ness’, as we might express it in English).7 And then he contends,
crucially, that ‘N(F, G)’ entails the corresponding universal generalization
concerning particulars, ‘8x(Fx ! Gx)’ (in plain English, ‘Everything that is
F is G’ or ‘All Fs are Gs’). But now the problem is that no account is
provided of how ‘N(F, G)’ entails ‘8x(Fx ! Gx)’, since there appears to be
no appropriate formal relationship between two sentences of these forms.
As I have just remarked, the inference problem does not arise for my
account of laws. This is because, according to my account, what laws entail
regarding the domain of particulars are solely dispositional, not occurrent, facts
or states of affairs – and such entailments are formally valid in the system of
sortal logic that I endorse. In the simplest case, a law has the form ‘βF ’, where
‘β’ denotes a kind and ‘F ’ an attribute. A corresponding particular disposi-
tional fact has the form ‘9φ(φF & a/φ)’, where, as was explained earlier, ‘φ’ is a
variable ranging over kinds, ‘a’ denotes an object, and ‘/’ signifies the formal
5
See further my The Four-Category Ontology, especially Part iii.
6
See Bas C. van Fraassen, Laws and Symmetry (Oxford: Clarendon Press, 1989), Chapter 5.
7
See D. M. Armstrong, What Is a Law of Nature? (Cambridge University Press, 1983).
ontological relation of instantiation. (The quoted formula here expresses,
thus, the fact that a is disposed to be F – in other words, that a exemplifies
F dispositionally. Recall here our earlier definition of ‘D[a, F]’ as being
equivalent to ‘9φ(φF & a/φ)’.) The crucial point, then, is that ‘βF ’ entails
‘8x(x/β ! 9φ(φF & x/φ))’ – and thus, via the definition of ‘D[a, F]’, entails
‘8x(x/β ! D[x, F])’ – the proof of which is elementary: see Proof [1] in the
Appendix at the end of this chapter. In other words, if the law that kind β is
characterized by attribute F obtains, then it follows that every object instanti-
ating β has the disposition to be F. But, crucially, it does not follow that every
such object is occurrently F. Laws, then, fully determine how objects are
disposed or tend to appear or behave, but not how they actually appear or
behave. Consequently, they do not entail universal regularities (Humean
‘constant conjunctions’) amongst such appearances or behaviour, in the
way that the Armstrong–Tooley–Dretske account of laws proposes and which
critics like Van Fraassen profess – rightly, I think – to find mysterious.
4 conditional laws and the problem of

idiosyncratic dispositions
However, some further complications now need to be discussed, in the
light of criticisms of my system alluded to earlier. As some critics have
pointed out, it seems implausible, at first sight, to suppose that every
particular dispositional fact is grounded in some law governing a kind to
which the object possessing the disposition belongs.8 After all, objects of
the same kind often seem to have different and sometimes even entirely
idiosyncratic dispositions. For instance, a particular dog, Fido, might
conceivably be the only dog in the world disposed to drink whisky: but
it is clearly not a law governing canine behaviour that dogs drink whisky.
This apparent difficulty can, however, be overcome by recognizing the
irreducibly conditional character of many laws. That is to say, the funda-
mental form of many laws is not, as we have hitherto been supposing,
simply something like ‘βF ’, but rather something like ‘β(G ! F)’.
Consider the following simple example. A body of frozen water – a piece
of ice – is not disposed to evaporate, but a body of boiling water is.
(Change the example if you do not like it: I use it purely for illustrative
purposes.) Yet both are bodies of the same kind – water. Ice and boiling
8
Amongst the people who have raised this point, either in print or in correspondence, are Ryan
Wasserman, Ludger Jansen, and David S. Oderberg. I am particularly indebted to David Oderberg
in this regard.
water are not, after all, different kinds of stuff, just the same kind of stuff
in different physical forms. The change that happens when ice melts is a
phase change, not a substantial change. Consequently, however, we
should not affirm it as a law governing the kind water that water
evaporates – rather, what we should affirm is that water evaporates if, or
when, it is boiling, and this is consistent with our also affirming that water
does not evaporate if, or when, it is frozen. Both laws apply at all times to
any body of water, whether or not it happens to be frozen or boiling (or
indeed neither). What entitles us to say that a piece of ice is not disposed
to evaporate, whereas some boiling water is, is not, then, that they
instantiate different kinds governed by different unconditional laws, but
rather that the single kind that they both do instantiate – water – is
governed by two different conditional laws, and that while one of these
bodies of water happens to satisfy the antecedent condition of one of
these laws, the other body happens to satisfy instead the antecedent
condition of the other law.
Casually speaking, we may indeed say that the sentences ‘Boiling water
evaporates’ and ‘Frozen water does not evaporate’ express laws. But it is at
least potentially misleading to express the laws in question in those ways,
since the complex sortal terms ‘boiling water’ and ‘frozen water’ are not
denotative of different kinds. We do better to express the laws in question
conditionally, as earlier: ‘Water, when it is boiling, evaporates’ and ‘Water,
when it is frozen, does not evaporate’. An adequate system of sortal logic
will reflect this by admitting logical operations on predicates as well as on
whole sentential formulas.9
So how do the foregoing considerations enable us to deal with the
example of Fido, the dog that is idiosyncratically disposed to drink whisky?
They do so in the following way. It is surely reasonable to suppose that
Fido’s peculiar condition is lawfully explicable, rather than just being
miraculous. Presumably, Fido – perhaps on account of some feature of
his past experience or training – has some property, X, such that it is a law
that dogs, if they have X, drink whisky. Fido just happens to be peculiar in
being the only dog to have X.10
9
See further my More Kinds of Being: A Further Study of Individuation, Identity and the Logic of Sortal
Terms (Malden, MA and Oxford: Wiley-Blackwell, 2009), Chapter 11. Note especially, in this
connection, that ‘β(G ! F)’ must not be taken to be logically equivalent to ‘βG ! βF ’: for example,
‘Water evaporates if, or when, it is boiling’ is plainly not equivalent to ‘If water boils, then water
evaporates’.
10
Note that I am assuming here that Fido’s peculiarity is not the upshot of some genetic mutation,
because in that case there might well be grounds for saying that Fido differs in kind from other dogs
It will be helpful at this point to introduce a distinction between
essential and accidental dispositions – and to say that Fido possesses
only an accidental disposition to drink whisky. Our earlier definition of
‘D[a, F]’ is, in these terms, really only a definition of essential disposition-
ality. So let us make that explicit by using the formula ‘DE[a, F]’ to say
that a is essentially disposed to be F, defining this exactly as we earlier
defined ‘D[a, F]’:
DE[a, F] ¼df 9φ(φF & a/φ)
Then let us use the formula ‘DA[a, F]’ to say that a is accidentally disposed
to be F, defining this as follows:
DA[a, F] ¼df 9X(O[a, X] & ~DE[a, X] & 9φ(φ(X ! F) & a/φ))
In other words, we say that a is accidentally disposed to be F just in case
there is some attribute, X, such that a occurrently exemplifies X – but is
not essentially disposed to be X – and for some kind, φ, that is instantiated
by a, it is a law that φs are F if they are X. Drawing on the preceding
definition of essential dispositionality, we can simplify this to:
DA[a, F] ¼df 9X(O[a, X] & ~DE[a, X] & DE[a, (X ! F)])
That is to say, a is accidentally disposed to be F just in case a occurrently
exemplifies some attribute, X, such that, although a is not essentially
disposed to be X, a is essentially disposed to be F if it is X.
Here it may be asked why I include the conjunct ‘~DE[a, X]’ in the
definiens of ‘DA[a, F]’. The answer should be evident. It is plausible to
suppose that ‘DE[a, X]’ and ‘DE[a, (X ! F)]’ together entail ‘DE[a, F]’, so
that without this conjunct our definition would allow an object a to be
both essentially and accidentally disposed to be F, which seems absurd. In
fact, this entailment holds only under certain assumptions – notably,
under the assumptions that an object cannot instantiate more than one
kind and that two laws of the form ‘βG’ and ‘β(G ! F)’ together entail one
of the form ‘βF’. See Proof [2] in the Appendix for a proof that the
entailment holds under these particular assumptions, which I shall take
to be correct for present purposes.
This, however, raises a further question. It might be supposed, prima
facie, that ‘O[a, F]’ entails ‘DE[a, F]’ – that if an object a exemplifies an
attribute F occurrently, then a is essentially disposed to be F, rendering our
and that he belongs to a kind, φ, such that it is a law that φs drink whisky. I am grateful to Frédéric
Nef for raising this issue in conversation.
definition of ‘DA[a, F]’ inconsistent. Clearly, we must for this very reason
deny that such an entailment holds – and, indeed, it seems reasonable to
maintain instead that ‘O[a, F]’ entails only the disjunction ‘DE[a, F] _
DA[a, F]’: that if an object a is occurrently F, then a is either essentially or
accidentally disposed to be F. But aren’t we now faced with a regress
problem? For, if the latter entailment holds, then, clearly, given our
original definition of accidental dispositionality, so does the following:
‘DA[a, F]’ entails ‘9X(DA[a, X] & 9φ(φ(X ! F) & a/φ))’. The implication
seems to be that an object can possess one accidental disposition only on
condition of possessing another one.
Fortunately, it seems clear that this potential regress is not inevitably
infinite. For notice that we don’t have it that ‘DA[a, F]’ entails ‘9X(X 6¼
F & DA[a, X] & 9φ(φ(X ! F) & a/φ))’. So it is not in fact strictly true,
given our assumptions so far, that an object can possess one accidental
disposition only on condition of possessing another one. We can, then,
have a case in which ‘DA[a, F]’ is true even though the following is also
true: ‘8X((DA[a, X] & 9φ(φ(X ! F) & a/φ)) ! X ¼ F)’. In such a case,
it would clearly follow that this is true: ‘DA[a, F] & 9φ(φ(F ! F) &
a/φ)’, which is equivalent to ‘DA[a, F] & DE[a, (F ! F)]’. For a proof
of this, see Proof [3] in the Appendix. However, there is no contradic-
tion lurking here, because ‘DA[a, F] & DE[a, (F ! F)]’ is not at all
absurd, in constrast with ‘DA[a, F] & DE[a, F]’. For, presumably, it is a
merely trivial logical truth that any object a is essentially disposed to be
F if it is F. In other words, our definitions of accidental and essential
dispositionality allow that sometimes, at least, it is true that an object a is
accidentally disposed to be F simply because a truth of the following
form obtains: ‘O[a, F] & ~DE[a, F]’ – for this certainly entails the
definiens of ‘DA[a, F]’, given the triviality of the aforementioned truth.
See again the Appendix, Proof [4], for a proof of this. In other, and
presumably more usual, cases, of course, ‘DA[a, F]’ will be true even
though ‘O[a, F]’ is not true.
5 conditional laws and the principle

of instantiation
Finally, I want to point out that an important advantage of recognizing the
prevalence of conditional laws is that it renders more plausible the principle
of instantiation that is central to a neo-Aristotelian immanent realist view of
universals like mine – especially for those philosophers, such as Armstrong
and myself, who not only adhere to this principle but also hold that we
should decide which universals to admit into our ontology largely by seeing
which universals science invokes in the laws that it postulates to explain
empirical phenomena. According to the principle of instantiation, every
existing universal must have, at least at some time and some place, a
particular instance: there are no uninstantiated universals. Now, for a
conditional law of the form ‘β(G ! F)’ to obtain, the principle of
instantiation requires only that each of the universals β, F, and
G should have instances – not that there should be an instance of β
that also occurrently exemplifies either G or F. For example, the
principle allows that it may be a law that water, when it is boiling,
evaporates – that boiling water evaporates – without requiring there to
exist, at any time or place, a body of water that is boiling. Imagine, thus,
a universe in which water exists but temperatures never rise high
enough to reach its boiling point. Provided that some liquids sometimes
actually boil and evaporate in this universe, it can still be true, consist-
ently with the principle of instantiation, that water, if or when it is
boiling, evaporates.
Now, clearly, many laws are expressed by sentences which do not
reveal their underlying conditional form – ‘Ice does not evaporate’ is a
simple example of this. ‘Water boils’ is, of course, another – for this
must be taken as an elliptical way of stating the conditional law that
water boils if, or when, its temperature and the atmospheric pressure
reach certain levels (for instance, when its temperature reaches a hun-
dred degrees Celsius at sea-level pressure).11 But it may well be that very
many more laws than we intuitively suppose are really conditional in
form, thereby reducing considerably the number of universals that we
need to include in our ontology and, correspondingly, reducing the
burden that is imposed by the principle of instantiation. Here, then, is
another example of how the four-category ontology, despite being – or,
rather, in virtue of being – more complex than some of its rivals, can
deal better than they do with certain long-standing metaphysical
problems.
11
Note, indeed, that if we held ‘Water boils’ to be a law, then, given our previous assumption that
‘Water, if it is boiling, evaporates’ is a law, we would be committed – via the principle that ‘βG &
β(G ! F)’ entails ‘βF ’ – to holding that ‘Water evaporates’ is a law, which we have already denied to
be the case. This confirms that ‘Water boils’ should be taken to be an elliptical expression of a law
which is really conditional in form.
Appendix: proofs
The simple proofs included in this appendix all use the familiar tree method of
proof,12 whereby it may be shown that a set of premises, Г, entails a proposition
p by assuming Г and ~p and showing that a logical inconsistency must result.
An ‘x’ at the foot of a branch indicates that it contains such an inconsistency –
that is, that the branch includes both a certain proposition and that propos-
ition’s negation – and if all branches terminate in this way the proof is complete.
[1] Proof that ‘βF ’ entails ‘8x(x/β ! 9φ(φF & x/φ))’
βF
~"x(x/β ® $φ(φF & x/φ))
$x~(x/β ® $φ(φF & x/φ))
$x(x/β & ~$φ(φF & x/φ))
a/β & ~$φ(φF & a/φ)
a/β
~$φ(φF & a/φ)
"φ~(φF & a/φ)
"φ(φF ® ~a/φ)
βF ® ~a/β
~βF ~a/β
x x
12
See, for example, Colin Howson, Logic with Trees (London: Routledge, 1997).
[2] Proof that ‘DE[a, G] & DE[a, (G ! F)]’ entails ‘DE[a, F]’
DE[a, G ] & DE[a, (G ® F )]

~DE[a, F]
$φ(φG & a/φ)

$φ(φ(G ® F ) & a/φ)
~$φ(φF & a/φ)
βG & a/β
γ(G ® F) & a/γ
βG
a/β
a/γ
β=γ (i)
β(G ® F) & a/β
β(G ® F)
βG & β(G ® F)
βF (ii)
βF & a/β
$φ(φF & a/φ)

x
Notes
(i) This line follows from the preceding two, given, as was proposed in the
main text, that an object cannot instantiate more than one kind.
(ii) This line follows from the preceding one, given, as was proposed in
the main text, that two laws of the form ‘βG’ and ‘β(G ! F)’ together
entail one of the form ‘βF ’.
[3] Proof that ‘DA[a, F]’ and ‘8X((DA[a, X] & 9φ(φ(X ! F) & a/φ)) !
X ¼ F)’ together entail ‘9φ(φ(F ! F) & a/φ)’
DA[a, F ]
"X((DA[a, X] & $φ(φ(X ® F ) & a/φ)) ® X = F )
~$φ(φ(F ® F) & a/φ)
$X(O[a, X ] & ~DE[a, X ] & $φ(φ(X ® F ) & a/φ)) (i)
O[a, G ] & ~DE[a, G] & $φ(φ(G ® F ) & a/φ)
O[a, G ]
~DE[a, G ]
$φ(φ(G ®F ) & a/φ)
DA[a, G ] DE[a, G ] (ii)

x
(DA[a, G ] & $φ(φ(G ® F ) & a/φ)) ® G = F
~(DA[a, G ] & $φ(φ(G ®F ) & a/φ)) G=F
$φ(φ(F ® F ) & a/φ)

x
~DA[a, G ] ~$φ(φ(G ®F ) & a/φ)
x x
Notes
(i) This line follows from ‘DA[a, F]’, given the definition of the latter in
the main text.
(ii) This line follows from ‘O[a, G]’, given, as was contended in the main
text, that ‘O[a, G]’ entails ‘DE[a, G] _ DA[a, G]’.
[4] Proof that ‘O[a, F] & ~DE[a, F]’ entails ‘9X(O[a, X] & ~DE[a, X] &
9φ(φ(X ! F) & a/φ))’ – the definiens of ‘DA[a, F]’
O[a, F ] & ~DE[a, F ]

~$X (O[a, X ] & ~DE[a, X ] & $φ(φ(X ® F ) & a/φ))
$φ(φ(F ® F ) & a/φ)
O[a, F ] & ~DE[a, F ] & $φ(φ(F ® F ) & a/φ)
$X (O[a, X ] & ~DE[a, X ] & $φ(φ(X ® F ) & a/φ))

x
chapter 4
Ontological categories and categorial predication
I have said a good deal in previous chapters about the notion of an

ontological category, but now I want to examine more closely the forms
of sentence that we deploy in speaking about such categories. When we say
of something that it ‘is an object’, or ‘is an event’, or ‘is a property’ – just to
cite a few examples – we are engaging in what I propose to call categorial
predication: we are assigning something to a certain ontological category.
Ontological categorization is clearly a type of classification, but it differs
radically from the types of classification that are involved in the taxonomic
practices of empirical sciences, as when a physicist says of a certain particle
that it ‘is an electron’, or when a zoologist says of a certain animal that it ‘is
a mammal’, or when a meteorologist says of a certain weather phenom-
enon that it ‘is a hurricane’. Classifications of the latter types presuppose
that the items being classified have already been assigned to appropriate
ontological categories, such as the categories of object, species, or event. But
what do categorial predications mean? How are their truth-conditions to be
determined and how can those truth-conditions be known to be satisfied?
Do they have truthmakers? Questions like these are amongst those that will
be addressed in the present chapter. As in the preceding chapter, my most
important guiding light in such matters will be Aristotle, although what
follows is in no sense intended to be an exercise in the scholarly exegesis of
Aristotelian texts.
1 fantology; or, ‘ontology lite’

Most philosophers today who have been brought up in the analytical
tradition have been exposed, at a formative period of their thinking, to
the formalism of first-order predicate logic with identity. This has
equipped them with a certain conception of reference and predication
which is, from the point of view of serious ontology, extremely thin and
superficial. It is a view which embodies – to invoke Barry Smith’s apt
50
Ontological categories and categorial predication 51
term – all the myths of ‘Fantology’: the idea that the most basic form of
atomic proposition is one that may be symbolized as ‘Fa’, where ‘F ’ is the
predicate and ‘a’ is a singular term, or ‘individual constant’ (the logical
counterpart of a proper name).1 The only further elaboration of this that is
countenanced is to admit relational predicates with any finite number, n,
of ‘places’, giving us as the most general form of an atomic proposition
‘Rna1a2 . . . an’. And the only ‘relation’ that is given any special formal
recognition is the dyadic relation of identity, with its own dedicated
symbol, ‘¼’, as in ‘a1 ¼ a2’. Sometimes, a formal recognition is also
accorded to the monadic existence predicate, as in ‘E!a’, but this is generally
analysed in terms of the particular (or, more tendentiously, ‘existential’)
quantifier, ‘9’, together with identity, as being equivalent to ‘9x(x ¼ a)’.
And that, basically, is the sum total of the formal machinery of standard
first-order predicate logic that serves to represent anything remotely ‘onto-
logical’ in character: it is ‘ontology lite’.
One point I am aiming to make here is that there are many more
ontological distinctions that we need to be able to make that go beyond
either the distinction between object (or ‘individual’) and property or that
between existence and identity. It just isn’t good enough to say, with
W. V. Quine, that the fundamental question of ontology is ‘What is
there?’, and that its most concise answer is ‘Everything’.2 Ontology is
concerned above all with the categorial structure of reality – the division
of reality into fundamental types of entity and their ontological relations
with one another. The object/property distinction is very probably one
such distinction that any system of categorial ontology should recognize,
and identity is one such relation, but very plausibly there are many others
besides these.
Note that, on the now standard view – basically Quine’s, which is a
development of Frege’s and Russell’s – we do not even get an ‘ontological
commitment’ to properties and relations out of ‘first-order’ languages, since
the latter do not involve quantification into predicate position. For that we
need, supposedly, a second-order language, where we can say things of the
form ‘9F(Fa)’ and the like. But this then apparently treats ‘properties’ (the
‘values’ of second-order variables) as second-order objects, of which yet
higher-order properties may further be predicated. So, on this view, the
1
See Barry Smith, ‘Of Substances, Accidents and Universals: In Defence of a Constituent Ontology’,
Philosophical Papers 26 (1997), pp. 105–27, and ‘Against Fantology’, in M. E. Reicher and J. C. Marek
(eds.), Experience and Analysis (Vienna: HPT & ÖBV, 2005).
2
See W. V. Quine, ‘On What There Is’, in his From a Logical Point of View, 2nd edn (Cambridge,
MA: Harvard University Press, 1961).
object/property distinction is really just a relative one, with an nth-order
object being an (n–1)th-order property, for all n > 1. Hence all entities are
‘objects’ on this view, but there are different ‘orders’ of objects, starting
with first-order ones which are not ‘properties’ of anything. And maybe we
can even discern an echo here, however weak, of the Aristotelian notion of
a ‘primary substance’, which is not ‘said of’ anything – of which much
more anon. (Quine himself, of course, was sceptical about including
‘properties’ in our ontology – at least, properties conceived as ‘universals’,
as opposed to items identifiable as sets of first-order objects – on the
grounds that he could see no principled way to individuate them,
rendering them vulnerable to his dictum ‘No entity without identity’.)
The next pernicious aspect of the ‘standard’ view is this: it accommo-
dates no notion of ‘property’ other than as something – though exactly what
is often left obscure – that ‘corresponds’ to a predicate, as in ‘Fa’, where ‘F ’
supposedly expresses a ‘property’ of a. This is despite the fact that we know
that, on pain of contradiction, not every predicate can denote or express a
property – this simply being a consequence of one version of Russell’s
paradox. Take the predicate ‘– is non-self-exemplifying’, which seemingly
applies, for example, to the first-order property of being green (a ‘first-order’
property because it is a property of first-order objects, such as apples and
leaves). ‘Being green (greenness) is not green’ certainly seems to be true,
whence it seems that we can conclude that ‘Being green is non-self-
exemplifying’ is also true. If the example is not liked, another can easily
replace it. But we know that there can be no (second-order) property
(property of a first-order property) of being non-self-exemplifying, since if
there were it could plainly be neither self-exemplifying nor non-self-
exemplifying, giving us a contradiction.
We are also now in the territory of Frege’s notorious paradox of the
concept (that is, first-order property) horse, which he contended was not an
object because it is not ‘saturated’ – the apparent implication being that
the object that we do denote by the singular term ‘(the property of) being a
horse’ is not what is expressed by the predicate ‘– is a horse’.3 The best that
the standard view can do at this point, it seems, is to say that for every
‘property’ of order n – ‘property’ in the sense of semantic value of a
predicate – there is a corresponding proxy-object of order (n þ 1), which
is the semantic value of a corresponding singular term. If that is right, then
it turns out that the object/property distinction isn’t even straightforwardly
3
See Gottlob Frege, ‘On Concept and Object’, in Translations from the Philosophical Writings of
Gottlob Frege, 2nd edn, ed. and trans. P. T. Geach and M. Black (Oxford: Blackwell, 1960).
relative, as was suggested earlier. Rather, we have a series of objects of
ascending ‘orders’ and, distinct but in parallel with that, a series of corres-
ponding ‘properties’. The scheme is something like the following – where,
listed in each column of the table, are typical expressions whose semantic
values are the ‘objects’ and ‘properties’ of successively higher ‘orders’:
Objects Properties
1st order ‘Dobbin’ ‘– is a horse’

2nd order ‘Being a horse’ ‘– is a first-order property’
3rd order ‘Being a first-order property’ ‘– is a second-order property’
4th order ‘Being a second-order property’ ‘– is a third-order property’
Et cetera
This scheme is organized so as to enable us, supposedly, to assign appro-

priate ‘semantic values’ to the semantically interpretable parts (subjects and
predicates) of sentences such as the following:
(1) Dobbin is a horse
(2) Being a horse is a first-order property
(3) Being a first-order property is a second-order property
Et cetera
Of course, as well as affirming, for example, (2) – ‘Being a horse is a first-
order property’ – we are also supposed to be able to affirm ‘Being a horse is a
second-order object’, since the foregoing table displays that alleged fact by
listing ‘Being a horse’ in the second row under the ‘Objects’ column. One
might suppose that this would entitle us to conclude that a first-order
property is (identical with) a second-order object: but that is problematic,
given Frege’s contention that the object/property (or object/concept)
distinction is mutually exclusive, on the ground that properties, but not
objects, are ‘unsaturated’ entities. This just shows how intractable the
‘paradox’ is, at least given Fregean assumptions.
But what, really, are the ‘semantic values’ of predicates – properties –
supposed to be? On one view – not Frege’s, clearly, but maybe Quine’s –
they are just the ‘extensions’ of those predicates: the sets of things to which
they apply, such as the set of all (actually existing) horses in the case of the
predicate ‘– is a horse’. This would make the semantic value of that
predicate an object, however, since sets are pretty clearly objects by any
reasonable account. On another view – neither Frege’s nor Quine’s – the
semantic value of such a predicate is instead a certain kind of function:
namely, a function from ‘possible worlds’ to sets of objects existing in those
worlds.4 Thus the semantic value of the predicate ‘– is a horse’, on this
view, is a function from possible worlds to the sets of horses existing in
those worlds (and Quine would reject the view because he rejects ‘possible
worlds’). This, in the current technical jargon, assigns an intension, rather
than just an extension, as the ‘semantic value’ of this predicate. But, fairly
evidently, a ‘function’, at least as this is normally understood by math-
ematical logicians, is itself just a special kind of set-theoretical entity and so
a certain kind of abstract object – not the kind of ‘unsaturated’ entity that
Frege took properties (or ‘concepts’) to be. However, these entanglements
take us too far from our current purpose, save to illustrate once more the
baroque qualities of ‘Fantology’ and its insouciance about questions of
serious ontology. Its adherents exhibit no genuine interest in understand-
ing the real nature of properties, if such entities there be.
However, one important further application of the foregoing scheme of
objects and properties of different ‘orders’ is worth mentioning, and it
concerns the notion of existence. As was indicated earlier, ‘Dobbin exists’ is
standardly analysed as ‘9x(x ¼ Dobbin)’, and here ‘9x(x ¼ –)’ may be
regarded as denoting or expressing a first-order property – the property,
possessed by Dobbin and indeed by all other existing objects, of being
identical with something. But we can, supposedly, also re-parse ‘9x(x ¼
Dobbin)’ by treating the expression ‘¼ Dobbin’ as being, in effect, a sign
for the quite different first-order property of being identical with Dobbin.
This being done, ‘9x(x –)’ may then be taken to express the second-order
property of having at least one (first-order) instance, which is here being
predicated of the first-order property of being identical with Dobbin. Thus
re-parsed, ‘9x(x ¼ Dobbin)’ should really be understood as having the
logical form ‘G2(F1)’, with ‘F1’ denoting the first-order property of being
identical with Dobbin and ‘G2’ the second-order property of having at least
one instance, so that the whole sentence may be re-translated into (rather
barbaric) English as ‘Being identical with Dobbin has at least one instance’.
4
Frege himself does, in his own way, treat properties (‘concepts’) as functions, but as functions from
objects to truth-values, and he accordingly regards functions as ‘unsaturated’ entities: see ‘Function
and Concept’, in Geach and Black (eds.), Translations from the Philosophical Works of Gottlob Frege.
Russell speaks instead of ‘propositional functions’, conceived as functions from objects to
propositions: see ‘Propositional Functions’, in Bertrand Russell, Introduction to Mathematical
Philosophy (London: George Allen and Unwin, 1919). But neither view is any more attractive than
the views now under discussion in this paragraph.
But once again, of course, the singular term ‘being identical with Dobbin’
now has to be taken to denote a second-order object, not the first-order
property that is the semantic value of the predicate ‘– is identical with
Dobbin’, at least if we follow Frege in these matters.
Now, at this point I want to cry out that all of this is completely insane
from an ontological point of view that aspires to any seriousness, being
driven entirely by the constraints of a particular style of logical formalism
and the ramshackle ontology that typically accompanies it. We need to sort
out our ontology properly first, and only then shape our formal logic to fit
it, not vice versa. And the first step towards sanity here is to abandon the
idea that there is something special and sacrosanct about the ‘atomic’
logical form ‘Fa’ – Fantology. Fantology, which originates from the
systems of formal logic newly developed by Frege and Russell around the
beginning of the twentieth century, does implicitly rest on certain onto-
logical assumptions, but on rather weak and ill-thought-out ones –
assumptions which seemed to matter little when they were overshadowed
by the sheer logical power of those formal systems. It weakly reflects, thus,
the object/property distinction, whose historical roots lie in traditional
Aristotelian substance ontology – ultimately, in fact, in Aristotle’s early
work, the Categories.5 But in the Categories, Aristotle does not assume a
simple dichotomy between ‘substance’ (or ‘object’) and ‘property’. Rather,
he introduces a more complex fourfold ontological scheme by way of two
key formal notions: those of ‘being said of a subject’ and ‘being in a
subject’. Somewhat obscure though these notions may initially seem to
be, on further investigation they in fact bear rich ontological fruit and
valuable insights into the proper relations between logic and ontology. It is
a worthwhile project, then, to try to clarify them in terms rather more
familiar to present-day metaphysicians, whereupon a comparison between
Fantology and traditional Aristotelian categorial ontology will prove to be
quite revealing.
2 aristotelian categorial ontology and its logical

formalization
I turn now to the foregoing task: that of explicating the ‘being said of ’/
‘being in’ distinction and its application by Aristotle in his characteriza-
tions of the most basic ontological types figuring in his fourfold categorial
scheme, these types being (1) primary substance, (2) secondary substance,
5
See Aristotle, Categories and De Interpretatione, trans. J. L. Ackrill (Oxford: Clarendon Press, 1963).
(3) property or attribute and (4) individual accident or mode (to use some
familiar Scholastic nomenclature).6 First of all, then, being said of is clearly
indicative of predication, while being in is indicative of what would, long
after Aristotle’s time, come to be called inherence. Now, Aristotle’s primary
substances in the Categories are described by him as being neither said of nor
in a subject – in other words, they are not predicable of anything, nor do
they exist ‘in’ anything as ontological ingredients or constituents. Being
neither ‘of’ nor ‘in’ other things, they are thus in neither sense ontologic-
ally dependent beings, and this indeed is why primary substances are taken
by him to be the entities that are ontologically most fundamental. By
contrast, Aristotle’s secondary substances – the species and genera of
primary substances – are, according to him, ‘said of’ but not ‘in’ a subject,
thus sharing one kind of ontological independence with primary sub-
stances but not another. Thus, for example, in affirming that Dobbin is
a horse, we are predicating the species horse of the primary or individual
substance Dobbin. But, on Aristotle’s view, this species isn’t ‘in’ the
individual substance, as an ontological ‘constituent’ of the latter – that
is, as some entity numerically distinct from that substance but one which,
nonetheless, somehow helps to constitute it as the particular substance that
it is. Next, we have items in the category ‘both said of and in a subject’,
which gives us a contrast between the predicate ‘– is a horse’ and, say, the
predicate ‘– is warm-blooded’. The latter expresses a property or attribute of
Dobbin, which he shares with all other individual substances of the same
species (all other horses) – shares, it seems, as an ingredient or constituent
in his nature or being (his ‘essence’).7 Finally, there are the items that are ‘in
a subject but not said of a subject’, which are generally taken to be a
primary substance’s ‘individual accidents’ or ‘modes’ – items such as the
particular whiteness of Dobbin, as opposed both to the universal whiteness
that he shares with all other white primary substances and to the particular
whitenesses of other white primary substances.
It will be noted that all predicables belong either to the category of
secondary substances or to the category of attributes and that all items in
6
I use a slightly different terminology for these categories in Chapter 3 – a terminology that
I personally prefer – but here I am more concerned to be faithful to the nomenclature of Aristotle
and his Scholastic successors.
7
Here I am, for the sake of simplicity, glossing over an important distinction between properties in
the strictest sense, which are necessarily shared by all primary substances of the same species – by all
individual horses, for instance – and what might be called ‘general accidents’, which are shared by
some but not all such primary substances, an example being Dobbin’s whiteness (since not all horses
are white). I take it that, for the Aristotelian, both warm-bloodedness and whiteness are ‘in’ Dobbin,
but only the former is necessarily ‘in’ all individual horses.
these categories are universals rather than particulars – all particulars
belonging either to the category of primary substances or to the category
of modes. Thus, on this account, although modes are in one sense
‘properties’ of primary substances, they are not predicable of them, which
may sound odd to the ears of present-day metaphysicians. And yet it does
seem to be borne out by what we actually say in English and other natural
languages. When, for instance, we say that Dobbin is white, we are making
no reference to his individual whiteness, even if it is because this individual
whiteness ‘inheres’ in him that whiteness (the universal) is predicable of
him. (Incidentally, it is precisely because present-day metaphysics is
equivocal about the status of ‘properties’, sometimes treating them as
universals and sometimes as particulars in the guise of ‘tropes’, that
I generally prefer to use the term ‘attribute’ to denote items that are ‘both
said of and in a subject’.)
Much more can and should be said about all this, but already we can see
that we have here a much richer ontology than anything that is offered by
Fantology and one that is, despite being categorially more complex,
ontologically far less baroque and extravagant. For example, we have no
grounds now for believing in a potentially infinite hierarchy of ‘orders’ of
objects and properties. Thus warm-bloodedness is said of a subject – it is a
‘predicable’ – but is not itself a subject, in the relevant sense of ‘subject’. Of
course, the word ‘warm-bloodedness’ can be made the grammatical ‘sub-
ject’ of a verb: but that is not the ontological conception of a subject, which
is that of a substance (whether primary or secondary). So, the sentence
‘Warm-bloodedness is a property of horses’, say, shouldn’t be understood
as predicating the (pseudo-)property or attribute of being a property of
horses of the (pseudo-)subject warm-bloodedness. Rather, it is just a round-
about way of saying ‘Horses are warm-blooded’, which expresses a general
truth about the secondary substance or species horse, holding in virtue of
that species’ essence. To regard warm-bloodedness as a subject – a quasi-
substance – would simply and literally be a category mistake, on the
Aristotelian view. Thus, on this approach, we need have no truck with
‘second-order logic’ (at least as it is ordinarily conceived) and other such
formal monstrosities. And we aren’t faced with Frege’s hideous ‘paradox’ of
the concept horse. For that paradox is really just an artefact of an impover-
ished logical formalism and its misconceived ontological assumptions.
So, what would a better logical formalism look like? First of all, if we are
going to follow the Aristotle of the Categories, we shall obviously need four
distinct classes of ‘material’ (that is, non-formal or non-logical) expressions,
not just the two (‘F ’ and ‘a’) of standard first-order predicate logic, in order
to denote (1) primary substances, (2) secondary substances, (3) properties
or attributes, and (4) individual accidents or modes. Let us then adopt the
following notation for this purpose:8
(1) a, b, c, . . . denote primary substances

(2) α, β, γ, . . . denote secondary substances
(3) F, G, H, . . . denote attributes
(4) f, g, h, . . . denote modes
Again, if we are going to follow the Aristotle of the Categories, we need

different devices for expressing ‘saying of’ (predication) and ‘being in’
(inherence), in place of the single device for expressing ‘predication’ that
we find in standard first-order logic. And indeed I am happy to follow
Aristotle here too, partly for purposes of illustration, but also because
I largely agree with him.9 So, to this end, let us simply use post-positioning
to represent predication, as in standard first-order logic, giving us, for
example, ‘βa’ and ‘Fa’ as ways to symbolize ‘Dobbin is a horse’ and
‘Dobbin is warm-blooded’ respectively (where ‘β’ ¼ ‘horse’, ‘a’ ¼
‘Dobbin’ and ‘F ’ ¼ ‘warm-blooded’). And let us additionally use pre-
positioning to represent inherence, giving us, for example, ‘af ’ and ‘aG’ as
ways to symbolize ‘This whiteness is in Dobbin’ and ‘Whiteness is in
Dobbin’ respectively (where ‘f ’ ¼ ‘this whiteness’, ‘a’ ¼ ‘Dobbin’ and
‘G ’ ¼ ‘white(ness)’). Note that, with this scheme, we can represent
‘Dobbin is white’ and ‘Whiteness is in Dobbin’ as ‘Ga’ and ‘aG’ respect-
ively, reversing the positions of ‘G’ and ‘a’. But, very plausibly, two such
sentences are logically equivalent, even if they are not synonymous, so that
for logical purposes we may discard formulas of the form ‘aG’ as superflu-
ous. Here is the scheme laid out in tabular form, followed by the formation
rules for constructing ‘atomic’ sentences:10
8
Compare Table 3.1 of Chapter 3, which adopts the same symbolism.
9
See my The Four-Category Ontology: A Metaphysical Foundation for Natural Science (Oxford:
Clarendon Press, 2006). Although I broadly follow Aristotle in that book, I do not there deploy
his being said of/being in distinction, preferring instead to make use of a three-way distinction
between instantiation, characterization, and exemplification. I still prefer the latter approach –
which I adopt also in Chapter 3 – but am using this opportunity to explore further an approach that
is closer to Aristotle’s own.
10
It should be noted, incidentally, that these formation rules differ from those provided in Table 3.1 of
Chapter 3, since in that chapter I was appealing to my own preferred formal ontological relations of
instantiation, characterization, and exemplification, instead of employing Aristotle’s being said of/
being in, or predicable/inherent distinction.
1. Subjects 2. Predicables 3. Inherents
Primary substances Secondary substances Attributes

a, b, c, . . . α, β, γ, . . . F, G, H, . . .
Secondary substances Attributes Modes
α, β, γ, . . . F, G, H, . . . f, g, h, . . .
Rule 1. Any item in column 1 can have something in column 2

predicated of it, this being represented by post-positioning the former
item to the latter: thus, ‘βa’, ‘Ga’, ‘αβ’ and ‘Fα’ – as in ‘Dobbin is a
horse’, ‘Dobbin is white’, ‘Horses are mammals’ (or ‘A/The horse is a
mammal’) and ‘Mammals are warm-blooded’, where ‘β’ ¼ ‘horse’,
‘a’ ¼ ‘Dobbin’, ‘G’ ¼ ‘white’, ‘α’ ¼ ‘mammal’ and ‘F ’ ¼ ‘warm-
blooded’. (Note that the definite or indefinite article in ‘A/The horse
is a mammal’ is logically redundant and would not, of course, have
any equivalent in Latin and many other languages.)
Rule 2. Any item in column 1 can have something in column 3 inherent
in it, this being represented by pre-positioning the former item to the
latter: thus, ‘aG’, ‘af ’, ‘αF ’ and ‘βf ’ – as in ‘White(ness) is in
Dobbin’, ‘This whiteness is in Dobbin’, ‘Warm-blooded(ness) is in
mammals’ and ‘This whiteness is in horses’, where ‘G’ ¼ ‘white
(ness)’, ‘a’ ¼ ‘Dobbin’, ‘f ’ ¼ ‘this whiteness’, ‘F ’ ¼ ‘warm-blooded
(ness)’, ‘α’ ¼ ‘mammal’ and ‘β’ ¼ ‘horse’. As implied above, we take
‘warm-blooded’ and ‘warm-bloodedness’ to be equivalent for onto-
logical purposes, the difference in form being merely a grammatical
peculiarity of English. And, once more, we take ‘White(ness) is in
Dobbin’ and ‘Warm-blooded(ness) is in mammals’ to be logically
equivalent, respectively, to ‘Dobbin is white’ and ‘Mammals are
warm-blooded’, rendering formulas of the forms ‘aG’ and ‘αF ’
redundant for logical purposes. The only odd case is the last, ‘This
whiteness is in horses’, for how, it might be asked, can a mode ‘inhere’
in a species? One answer might be that it does so just as long as some
individual member of the species, such as Dobbin, has this whiteness
inhering in it. Alternatively, we might simply want to exclude this last
case as not well formed and restrict accordingly the second formation
rule just stated (Rule 2).
Observe that these formation rules give us just the following six types of
‘atomic’ sentences: Fa, αb, Gβ, αβ, af and αg. The first type predicates an
attribute of a primary substance, the second predicates a secondary
substance of a primary substance, the third predicates an attribute of a
secondary substance, the fourth predicates a secondary substance of
another secondary substance, the fifth expresses the inherence of a mode
in a primary substance and the sixth expresses the inherence of a mode in
a secondary substance. (For reasons just explained, the first type also
serves to express the inherence of an attribute in a primary substance and
the third type also serves to express the inherence of an attribute in a
secondary substance.) As just mentioned, we might want to exclude the
sixth type and allow only the first five. There is nothing sacrosanct, of
course, about this notation, and others could have been used quite as
well. But it is interesting to note that, if we restrict our attention to just
the first five types, we can see that the four basic classes of ‘material’
terms occur in them with the following frequencies: secondary substance
terms (α, β) four times, primary substance terms (a, b) three times,
attribute terms (F, G) two times and mode terms (f, g) just once. Whether
that rather neat distribution has any significance is hard to say. In saying
that just these types of atomic sentences are well formed, other combin-
ations of terms are by implication excluded, such as ‘GF ’ and ‘f β’: one
attribute cannot be predicated of or inhere in another attribute, nor can a
mode be predicated of a secondary substance or a secondary substance
inhere in a mode.
Of course, this gives us, so far, only a way to represent formally ‘atomic’
propositions. There is a lot more expressive power that we still need to
cater for in order to express, for instance, truths of existence and identity.
Here we may follow existing practice, however, and use the symbols ‘E!’
and ‘¼’ respectively for these purposes. But we also need quantifiers – at
least a particular and a universal quantifier – although for this purpose too
we may as well again follow existing practice and use the symbols ‘9’ and
‘8’. However, we shall not adopt the usual assumption that existence can
be ‘analysed’ in terms of ‘9’ and ‘¼’. And another appropriate diversion
from standard practice would be to favour so-called restricted quantifiers
for most purposes. For instance, in order to represent the sentence ‘Some
(individual) horses are white’, we shall use a formula such as ‘(9x: αx)(Fx)’,
where ‘α’ ¼ ‘horse’ and ‘F ’ ¼ ‘white’. Similarly, in order to represent the
sentence ‘Some (species of) mammals are viviparous’, we shall use a
formula such as ‘(9φ: βφ)(Gφ)’, where ‘β’ ¼ ‘mammal’ and ‘G’ ¼ ‘vivipar-
ous’. (It will be noticed, incidentally, that I am here adopting the conven-
tion of using x, y, z, . . . as variables ranging over primary substances and φ,
χ, ψ, . . . as variables ranging over secondary substances.)
My reason for favouring restricted quantifiers for these purposes emerges
most clearly in the case of universal generalizations. Consider, for instance,
the true sentence ‘All (species of) mammals are warm-blooded’. This I prefer
to represent by a formula such as ‘(8φ: βφ)(Hφ)’, where ‘β’ ¼ ‘mammal’ and
‘H’ ¼ ‘warm-blooded’. This, I think, is greatly preferable to a formula such
as ‘8φ(βφ ! Hφ)’, which uses unrestricted quantification over secondary
substances. In fairly plain English, the difference is, very roughly, between
‘Any mammalian species is warm-blooded’ and ‘Any species, if it is mam-
malian, is warm-blooded’. But one problem with the latter formulation
arises when we consider what sort of sentence qualifies as an instance of this
sort of generalization. The sort of sentence that qualifies is one such as
‘If (the species) mountain is mammalian, then it is warm-blooded’ – or,
more colloquially, ‘If mountains are mammals, then they are warm-
blooded’ – which I would represent by a formula such as ‘(βγ ! Hγ)’,
where ‘γ’ ¼ ‘mountain’. The latter clearly is entailed by ‘8φ(βφ ! Hφ)’, by
an application of the logical rule of universal instantiation. But the antece-
dent of ‘If mountains are mammals, then they are warm-blooded’ – ‘Moun-
tains are mammals’ – is very hard to make any sense of. Indeed, it seems to
constitute a category mistake: not, indeed, one involving the four most basic
categories of the Aristotelian scheme, but one involving two different
subcategories of secondary substances. Mammals (that is, mammalian
species, such as the horse and the rabbit) belong to the subcategory of
biological species, whereas mountains belong to the subcategory of geological
species – and it apparently makes no sense even to entertain the ‘thought’
that mountains are mammals; that is, that a species of geological structure is
a species of living organism. No such absurdity is entailed by my preferred
formula, ‘(8φ: βφ)(Hφ)’. This, in conjunction with a formula of the form
‘βγ’, entails one of the form ‘Hγ’. For instance, ‘All (species of) mammals
are warm-blooded’ together with ‘The horse is a (species of) mammal’
entails ‘The horse is warm-blooded’. But the additional premise here,
‘The horse is a (species of) mammal’, is evidently perfectly uncontentious
and indeed just expresses an essential truth about horses.
I noted above in passing that I follow the convention of using x, y, z, . . .
as variables ranging over primary substances and φ, χ, ψ, . . . as variables
ranging over secondary substances. For the sake of completeness, however,
we need also variables ranging over attributes and modes.11 But in saying
11
Such variables are provided for in Table 3.1 of Chapter 3, but again it must be remarked that Table 1
of Chapter 3 employs different formation rules for constructing atomic sentences from those now
being utilized.
this we must be careful to remember that the latter are not subjects (that is,
they are not substances, either primary or secondary). We can have names
for them and variables ranging over them, but that should not lead us to
treat them as quasi- or pseudo-substances, which is the implicit mistake of
those philosophers and logicians who think that ‘second-order’ logic, by
quantifying into predicate position, incurs ontological commitment to a
new class of ‘objects’, over and above the ‘first-order’ objects that are the
supposed values of ‘first-order’ variables. This, I think, is just a horrible
ontological muddle on their part. Properties, in the form of both attributes
(universal properties) and modes (particular properties), should certainly
be accorded a place in any sensible ontology, but it is wrong to reify or
hypostatize them. This is because they are essentially ‘inherent’ entities,
always being ‘in’ a subject (substance) – or, as we might otherwise put it,
always being only aspects of substances, or ‘ways substances are’, never
substances in their own right.
Note, incidentally, that the formal logical language sketched above is in
fact only classifiable as a ‘first-order’ language in the standard sense, despite
the fact that it includes names for and variables ranging over properties, in
the shape of both attributes and modes. This is because it does not involve
‘quantification into predicate position’ in the standard sense. (Moreover, in
model-theoretic terms, it does not invoke a domain which includes all
subsets of the domain of first-order objects quantified over by a standard
first-order language and hence a domain whose cardinality is necessarily
greater than that of the latter, even if there are infinitely many such first-
order objects; a domain of quantification for a formalized language like
mine could perfectly well include only a denumerable infinity of entities,
so long as it included some entities belonging to each of the four basic
ontological categories.) Now, the latter phenomenon – quantification into
predicate position – is exemplified in a formula of so-called ‘second-order’
logic such as ‘9F(Fa)’. But, in standard predicate logic, the ‘F ’ in ‘Fa’ is
supposed to represent a predicate, understood as an ‘incomplete’ expression
such as ‘– is white’. By contrast, ‘Fa’ in my formalization of Aristotelian
categorial ontology serves to express the proposition that the attribute
F(ness) inheres in, or is predicable of, the primary substance a. ‘F ’ and ‘a’
here are thus to be thought of as two terms, each naming an entity
belonging to a certain ontological category. In standard predicate logic,
‘F ’ is not a term in this sense at all, since it doesn’t serve to name any entity
but just represents what remains of a complete predicative sentence when a
name is removed from it – as, for example, ‘– is white’ is what remains
when the name ‘Dobbin’ is removed from the sentence ‘Dobbin is white’.
Another way to make this point is to say that, in the standard formalism,
the ‘F ’ in ‘Fa’ has an implicit ‘is’ of predication built into it, whereas in my
formalism ‘F ’ simply denotes a certain attribute and its predicability of a is
represented formally not by a further symbol (although this could certainly
be done), but rather by means of the post-positioning convention whereby
‘a’ is placed immediately after ‘F ’.
3 categorial predication: its form, meaning, and use

I come now properly to the most important topic of this chapter, categorial
predication, for which the preceding two sections have provided a necessary
preliminary. The system of formal logic whose language I have been
constructing is meant to be one which respects and reflects certain funda-
mental categorial distinctions of an ontological nature. But now we have to
consider how we can speak explicitly of such categorial distinctions, by
extending the expressive power of our formalized language. So far, these
categorial distinctions have been only implicit in the language, being
embodied in our choice of symbol types and our ways of representing
predication and inherence. A categorial statement, however, will be one
which explicitly assigns some entity to a specific ontological category; and
in our present system, of course, we have four such categories: those of
primary and secondary substance, attribute and mode. (But we should again
recall that these are just the basic categories of the system, which need by
no means exclude further subcategories of these basic ones.)
So consider, for example, a statement such as ‘Dobbin is a primary
substance’, or ‘The horse is a secondary substance (species)’. On the face of
it, the expression ‘– is a primary substance’ is a predicate, which says
something of Dobbin. (That, as we have seen, is at least the now standard
conception of what a ‘predicate’ is.) But on our currently preferred
Aristotelian view of predication, predicables are what are ‘sayable’ of sub-
jects. So, does there exist a predicable that is said of Dobbin by the
statement ‘Dobbin is a primary substance’? If so, then that predicable will
have to be either a secondary substance or else an attribute: for these and
only these are things that may be ‘said of’ a subject. One suggestion, then,
might be that there is a species (or, rather, a very high-level genus) – that of
primary substance – which can be ‘said of’ Dobbin, very much as the
species horse and the genus mammal can be ‘said of’ Dobbin. An alternative
suggestion is that there is a highly abstract attribute – that of being a
primary substance – which can be ‘said of’ Dobbin, very much as the
attributes being warm-blooded and being herbivorous can be ‘said of’
Dobbin. But neither suggestion is preferable to the other and both are in
fact unattractive (and perhaps even incoherent).12
The solution is to reject both suggestions. This, however, requires us to
recognize a certain ambiguity in the notion of ‘saying of’ or predication. In
one sense – the sense hitherto to the fore in our discussion of the
Aristotelian system – the notion of predication is a relational one. In this
sense, in predication one thing is ‘said of’ another thing, with each of these
things belonging to an appropriate ontological category. For example, an
attribute is said of a substance, either primary or secondary. Or a secondary
substance is said of a primary substance. Or one secondary substance is said
of another secondary substance. But then there is and must be another,
non-relational notion of ‘saying of’ or predication, where this includes
assigning an item to a certain ontological category. (Another plausible case
is that of predicating existence of something, since it is highly doubtful that
existence is properly conceived as a property or attribute of anything; if it
were, then it ought to make sense to say that existence exists, and yet it
scarcely does seem to make sense to say this.)
One characteristic of a statement involving categorial predication is that if
it is ‘formally correct’, then it should be necessarily true. A perspicuous
formalized language should respect this requirement. Suppose, thus, that
we introduce the formal ontological predicates ‘P’, ‘S’, ‘A’ and ‘M’ into the
formalized language that was developed in section 2. These are to express,
respectively, the English predicates ‘– is a primary substance’, ‘– is a
secondary substance’, ‘– is an attribute’ and ‘– is a mode’. Then, to distin-
guish categorial predication from (what we might aptly call) material predi-
cation (which we have chosen to express by the device of post-positioning),
let us use superscription for the former. Thus, for example, ‘Dobbin is a
primary substance’ will be formalized as ‘aP’, where ‘a’ represents ‘Dobbin’.
And then our point is that such a statement will be necessarily true if and
only if it is well formed, as it is in this case: that is, it will be necessarily true if
and only if the categorial superscript matches the symbol-type to which it is
attached. In the present case, ‘a’ is a symbol for an individual or primary
substance and hence matches the superscript ‘P’. By contrast, a formula such
as ‘FP’, representing a statement such as ‘Whiteness is a primary substance’,
is just not well formed in this system and hence necessarily false.
12
Consider, thus, the proposal that ‘primary substance’ denotes a genus to which all primary
substances belong. Then it turns out that, since all genera belong to the category of secondary
substance, the sentence ‘Primary substance is a secondary substance’ must be in some sense true. But
I find it very hard to make any clear sense of this.
But how, it might now be asked, could there be any real use for such
statements of categorial predication, given that the categorial distinctions
are already built into the symbolism of the formalized language (as they are
not, perhaps, in a natural language such as English)? The answer is that we
want our language to be capable of talking about pure ontology. For that,
we need also names and variables which are categorially neutral, in order to
say things such as ‘Every primary substance has at least one mode inherent
in it’. Thus, using ‘e’ (for ‘entity’) as a new type of ontologically neutral
variable, we could express the last-mentioned sentence formally in this
manner: ‘8e1(e1P ! 9e2(e2M & e1e2))’. (Here we are using unrestricted
quantifiers, of course, and the proposal would be that these are only to
be used in statements of pure ontology; note also that, in the formula just
stated, ‘e1e2’ must be construed as expressing inherence rather than predica-
tion, given the formation rules and the typing of e1 and e2 as P and M
respectively.) Statements of pure ontology would all be like this and in this
way we could envisage the construction of a formal, axiomatizable theory of
pure ontology, which would constitute an a priori science analogous to
various branches of pure mathematics. In the formal theory of pure
ontology, no specific entity of any category would be referred to, such as
Dobbin or whiteness: all statements would concern the categories them-
selves and relationships obtaining between their members purely in virtue
of their categorial status, as in the case of the sample statement cited above.
Of course, for present purposes I am assuming that the ‘correct’ formal
theory of pure ontology will be a characteristically ‘Aristotelian’ one, of the
kind sketched earlier. But that assumption is not vital to the notion of pure
ontology as such. Indeed, one can envisage alternative (or even just
different) systems of pure ontology, just as there are different branches of
pure mathematics. (Some systems of ontology, for instance, include the
basic category of event, whereas in the ‘Aristotelian’ ontology there is no
room for such entities save in the guise, perhaps, of modes of primary
substances.) However, one should not take the analogy with pure math-
ematics too far, since the latter consists of theories which do make
reference to specific entities of certain types, such as the natural numbers,
whereas pure ontology is perfectly general or ‘topic-neutral’ in its subject
matter.
To repeat an earlier point of great importance, categorial predications
are – as Wittgenstein might at one time have remarked – true, when they
are true, simply in virtue of their ‘logical grammar’. Thus, ‘aP’ can be seen
to be true simply by inspection of its logical form. In that sense, such a
truth has and requires no ‘truthmaker’, if by a ‘truthmaker’ we mean some
entity which, by existing, makes it true. ‘aP’ doesn’t even require the
existence of the primary or individual substance a to make it true: thus,
‘Dobbin is a primary substance’ can be known to be a true – indeed, a
necessarily true – categorial predication whether or not Dobbin is known
to exist. I do want to allow, of course, that from ‘aP’ we may validly infer
‘9e(eP & e ¼ a)’, and vice versa. Thus, I am happy to allow that ‘Some
primary substance is (identical with) Dobbin’ is just a long-winded way of
saying ‘Dobbin is a primary substance’. But recall that I am rejecting the
claim that ‘Some primary substance is (identical with) Dobbin’ is logically
equivalent to ‘Dobbin exists and is a primary substance’ or, more generally,
that ‘9e(eP & e ¼ a)’ is logically equivalent to ‘(E!a & aP)’. Dobbin’s
existing is no doubt logically equivalent to some existing primary sub-
stance’s being (identical with) Dobbin, but not just to Dobbin’s being a
primary substance, since the latter is just an a priori truth arising from an
ontological necessity concerning the correct ontological categorization of
any such item as Dobbin is conceived to be, whether or not Dobbin
actually exists.
part ii
Identity
chapter 5
What is a criterion of identity?
In Chapter 2, I made extensive appeal to the notion of a criterion of

identity, in connection with our ability to individuate and refer to particu-
lar objects both at a time and across time. However, the complaint is often
heard that, while the phrase ‘criterion of identity’ is frequently used in
contemporary philosophical writings, relatively rarely is any attempt made
to spell out its intended meaning at all precisely, or to defend the cogency
of the notion that it is supposed to convey.1 There may also seem to be a
suspicious dearth of specific and plausible examples of such criteria in the
literature. Since, as I have already made clear, I consider the notion of a
criterion of identity to be a vitally important one, both for the philosophy
of logic and language and for metaphysics, I want, in this chapter, to try to
dispel some of the obscurity and to allay some of the suspicions that have
beset this notion.2 Two particularly important issues that I shall address are
these: first, in stating a criterion of identity for individuals of a given sort, is
it legitimate to quantify over a domain including precisely individuals of
that sort? In short, are ‘impredicative’ criteria of identity acceptable? And,
second, must it be the case that criteria of identity for at least some sorts of
individuals can be stated, at least in principle, in terms which involve
neither reference to nor quantification over individuals of any sort? The
thought behind a positive answer to this second question would be that
putative criteria of this kind are needed to underpin our most primitive
1
See, for example, P. F. Strawson, ‘Entity and Identity’, in H. D. Lewis (ed.), Contemporary British
Philosophy, Fourth Series (London: George Allen and Unwin, 1976), reprinted in his Entity and
Identity and Other Essays (Oxford: Clarendon Press, 1997).
2
In this chapter, I shall in part – but only in part – be covering the same territory as I did earlier in
Chapter 2. But here I shall be approaching the notion of a criterion of identity with much greater
emphasis on its history and its logical involvements, thereby extensively supplementing the
characterization of this notion that I provided in Chapter 2, with a view to making its
philosophical credentials more secure. Another important difference between the present chapter
and Chapter 2 is that, in the present chapter, I lay no emphasis, as I did in Chapter 2, on the
distinction between sortal and categorial terms, focusing here almost exclusively on the former.
69
70 Identity
acts of reference to individuals and thus ultimately ground our very
understanding of the notions of individuality and reference. My answers
to these two questions will be ‘yes’ and ‘no’ respectively.
I should mention that my discussion of criteria of identity in this
chapter will keep a much closer eye on the history of this topic than was
the case in Chapter 2, as well as looking in much more depth at certain
logical problems to which the notion of a criterion of identity appears to
give rise, including questions concerning the logical form of such criteria,
which has certainly been subject to much debate.
1 the fregean thesis

Gottlob Frege is usually given credit for introducing the notion of a
criterion of identity into philosophical discourse, on the strength of certain
sections of the Grundlagen.3 My own feeling is that this accolade – while
warranted in regard to terminology – fails to do justice to John Locke, who
grasped the key point that that in which identity ‘consists’ for things of one
sort (say, parcels of matter) may not be the same as that in which it
‘consists’ for things of another sort (say, living organisms).4 What assuredly
can be credited to Frege, however, are first a certain very general thesis
apparently tying the sense of any singular term to a criterion of identity,
and second one very specific and frequently exploited example of such a
criterion. In fact, though, both the scope of the thesis and the significance
of the example are open to debate.
Frege’s general thesis – to which I shall hereafter refer as the ‘Fregean
thesis’ – is framed by him in the following words:
If we are to use the symbol a to signify an object, we must have a criterion for
deciding in all cases whether b is the same as a, even if it is not always in our power
to apply this criterion.5
3
See Gottlob Frege, Die Grundlagen der Arithmetik (1884), translated as The Foundations of Arithmetic
by J. L. Austin (Oxford: Blackwell, 1953), §§62 ff. The implication in question may be found in
Michael Dummett, Frege: Philosophy of Language, 2nd edn (London: Duckworth, 1981), pp. 73
and 545.
4
Press, 1975), ii, xxvii. ‘Consist’ is Locke’s own choice of word: see §9.
5
Frege, The Foundations of Arithmetic, §62. The word which Austin translates as ‘criterion’ is
Kennzeichen. By contrast, Ludwig Wittgenstein uses the word Kriterium throughout the
Philosophical Investigations, which may cast some doubt on the close affinity mooted by Dummett
in Frege: Philosophy of Language, p. 73. Of course, precisely what Wittgenstein himself meant by
‘criterion’ is the subject of a vast secondary literature, to which I have no wish to contribute here.
What is a criterion of identity? 71
The reason why I say that the intended scope of this thesis is debatable is
that §62 of the Grundlagen, in which it appears, is the first of a group of
sections introduced by the heading ‘To obtain the concept of Number, we
must first fix the sense of a numerical identity’, and §62 itself opens with
the question ‘How, then, are numbers to be given to us, if we cannot have
any ideas or intuitions of them?’ – all of which might suggest that the thesis
in question is intended to be restricted to symbols used to signify abstract
rather than concrete, perceptible objects. One commentator who certainly
does not interpret the intended scope of the thesis in this narrow way,
however, is Michael Dummett, who expressly regards it as intended to
apply to all significant singular terms, and indeed endorses the thesis under
this liberal interpretation.6
2 frege’s example of directions and parallelism

Before examining the content and merits of the Fregean thesis, we need to
look at the oft-cited illustrative example of a criterion of identity that Frege
deploys – the example concerning parallelism and the directions of lines.
In §64, Frege remarks,
The judgement ‘line a is parallel to line b’ . . . can be taken as an identity. If we do
this, we obtain the concept of a direction, and say: ‘the direction of line a is
identical with the direction of line b’.
What Frege appears to be suggesting, in the light of the general thesis just
enunciated by him, is that in order to use an expression of the form ‘the
direction of line a’ to refer to an object (a direction), we need at least to
supply a logically necessary and sufficient condition for the truth of
identity statements in which such expressions flank both sides of the
identity sign – a condition which, moreover, can be expressed without
reference (either explicit or implicit) to directions. He is further saying that
just such a condition can be specified in terms of the parallelism of the lines
whose directions are to be identified. Generalizing, this provides us with
the following model of a criterion of identity, where ‘f()’ is a functional
expression, ‘x’ and ‘y’ are individual variables, and ‘R’ expresses a certain
equivalence relation defined over the domain to which the values of those
variables belong:
(A) 8x8y(f (x) ¼ f (y) $ Rxy)
6
See Dummett, Frege: Philosophy of Language, pp. 73 ff. and 545 ff.
72 Identity
Now, one problem with this model is that it seems to be applicable only in
the case of a special class of singular terms – those formed by means of
functional expressions like ‘the direction of’. (Other examples might be
‘the shape of’ and ‘the colour of’, the former indeed being explicitly
mentioned by Frege himself.) Proper names, strictly so called, clearly do
not belong to this class – for example, ‘London’ and ‘Napoleon’. Neither
do certain other sorts of singular terms, such as personal pronouns.
Surprisingly, however, this has not prevented some writers from trying
to force all criteria of identity into the mould of Frege’s example, even
though it is not clear that Frege himself would have endorsed this. An
instance of this approach may be found in a paper by Timothy
Williamson.7 Thus we find Williamson supposing that the form of a
criterion of personal identity is to be extracted from the schema (A) by
allowing the variables ‘x’ and ‘y’ to range over person-stages and stipulating
that ‘f (x)’ means ‘the person of whom x is a stage’. But what reason at
all have we to suppose that persons must be like directions in being objects
to which we need (primarily, at least) to refer by means of functional
expressions? Directions are, if you will, essentially directions of something –
in Frege’s treatment, they are directions of lines. But persons are not – or,
at any rate, are not at all obviously – essentially persons of anything at all.
Indeed, the supposed parallel between ‘the direction of line x’ and ‘the
person of whom x is a stage’ seems entirely spurious, even if one counten-
ances – as I do not – such objects as ‘person-stages’. For an expression of
the latter form is, on the contrary, more naturally assimilable to one such
as ‘the line of which x is the direction’. In short, what is being overlooked is
a certain order of ontological dependency seemingly implicit in Frege’s
discussion of the example of directions: the parallelism of lines can provide
a criterion of identity for the directions of lines only because directions are
ontologically – and indeed conceptually – dependent on lines, in a way in
which lines are not on directions.8 But this immediately raises a difficulty
for anyone seeking to extend schema (A) to names of what we might, in an
Aristotelian vein, call (primary) substances, since these – assuming them to
7
See Timothy Williamson, ‘Criteria of Identity and the Axiom of Choice’, Journal of Philosophy 83
(1986), pp. 380–94.
8
This might perhaps be questioned, at least when directions are thought of in a perceptual, subject-
centred way (as when one compares the directions of two objects as seen from a certain location).
Obviously, however, Frege is not thinking of directions in this sense. But, in any case, the real issue is
not whether Frege was right about directions but whether the account that he gives of them, right or
wrong, can legitimately be extended to objects such as persons.
exist – are precisely objects standing in no such relationship of ontological
dependency with other objects.9
3 a return to the fregean thesis

Having seen something of the limitations of Frege’s best-known example
of a criterion of identity, we may return to the general Fregean thesis that a
singular term can be significantly employed only in association with some
criterion of identity – where by a ‘criterion of identity’ let us not, on
account of the limitations just adverted to, presume exclusively to mean a
principle conforming to schema (A) above, but merely, as Frege himself
puts it, a rule ‘for deciding in all cases whether b is the same as a’, where ‘a’
and ‘b’ are certain singular terms.
Unfortunately, this looser and more informal characterization of a
criterion of identity threatens to be altogether too liberal. Thus it is likely
to be objected that, understood in this light, the Fregean thesis is almost
trivially true, since it may be urged that Leibniz’s principle of the identity
of indiscernibles provides just such a rule – and provides it for all singular
terms, quite irrespective of their logical form and of the nature of the
objects that they designate.10 However, whatever else emerges from Frege’s
discussion of criteria of identity, it does at least emerge that such criteria, as
he understands them, certainly are to be conceived of as very often differing
for singular terms designating objects of different sorts, and most import-
antly are expected to cast light on the meaning of certain general terms
describing those objects – general terms such as ‘number’, ‘direction’, and
‘shape’. This is a feature of Frege’s treatment which I shall explore more
thoroughly a little later.
Now, this semantic connection between singular and general terms is
evident in the case of the examples expressly discussed by Frege, because in
their case a species of singular term is explicitly formed with the aid of a
general term – for example, ‘the direction of line x’ with the aid of the
general term ‘direction’. When the Fregean thesis is advanced as applying
universally to singular terms, however, the connection must be assumed in
9
For more on this conception of substance, see my The Possibility of Metaphysics: Substance, Identity,
and Time (Oxford: Clarendon Press, 1998), Chapter 6.
10
Baruch Brody argues along very much these lines in his Identity and Essence (Princeton University
Press, 1980), Chapters 1–3. Brody contends that Leibniz’s principle (coupled with the standard
logical axioms of identity) provides all that we could need by way of a criterion of identity, and
accordingly dismisses attempts to frame distinctive identity criteria for different sorts of things.
Naturally, I regard this contention as being entirely misconceived, for reasons which will become
plain.
74 Identity
many cases to be only implicit. Thus adherents of the full-blown thesis will
typically urge, say, that the proper name ‘Napoleon’ is semantically linked
to the sense of the general term ‘man’ (or perhaps ‘person’) – the linkage
being provided precisely by the criterion of identity associated with the
proper name ‘Napoleon’, this also being associated with the use of all other
singular terms designating individual men. And – to complete the picture –
a sortal or substantival general term will be precisely one which, like ‘man’,
has thus associated with it, as indeed a component of its very sense, some
unique criterion of identity for the use of singular terms designating objects
describable by that general term. By contrast, an adjectival general term –
to use P. T. Geach’s terminology – will be one which, like ‘red thing’,
supplies no criterion of identity for instances falling under it.11
All of this suggests another general formulation of the notion of a
criterion of identity which is at once less limiting than that provided by
schema (A) and more specific than the loose and informal characterization
quoted earlier from Frege – a formulation which makes quite explicit the
sortal relativity of identity criteria. This is, moreover, a formulation which
has in fact found a certain amount of favour in the literature. According
to this school of thought, the general form that a criterion of identity will
take is the following:12
(B) 8x8y((φx & φy) ! (x ¼ y $ Rxy))
where ‘φ’ is a general term of the sortal or substantival variety and ‘R’ again
expresses a certain equivalence relation.13 A paradigm example of a criter-
ion of identity conforming to this schema is provided by the axiom of
extensionality of set theory: for the latter may be taken as telling us that if x
and y are sets, then x is identical with y if and only if x and y have the same
members. Of course, sets are once again abstract objects, like directions, and
11
The expressions ‘substantival’ and ‘adjectival’, used in this way, are introduced by P. T. Geach – in
emulation of Aquinas – in his Reference and Generality, 3rd edn (Ithaca, NY: Cornell University Press,
1980), p. 63. In similar vein, P. F. Strawson distinguishes between what he calls sortal and
characterizing universals: see his Individuals: An Essay in Descriptive Metaphysics (London:
Methuen, 1959), p. 168.
12
In Chapter 2, I adopted the convention of subscripting ‘R’ with the sortal symbol ‘φ’, to emphasize
the fact that different sortals are very often associated with different criterial relations. Here I shall
take that point for granted.
13
In what follows, I shall speak of ‘type-(A)’ and ‘type-(B)’ identity criteria. It has, however, now
become customary to use the terms ‘two-level identity criterion’ and ‘one-level identity criterion’ for
this purpose: for further discussion, see my ‘One-Level versus Two-Level Identity Criteria’, Analysis
51 (1991), pp. 192–4. I shall avoid this terminology here because I think that it is potentially
misleading in certain ways.
this may prompt some doubts about the possibility of formulating criteria
of identity in the mould of schema (B) that are applicable to concrete
objects, such as men. But that (B) is an advance on (A) is suggested by the
fact that, whilst we can apparently recast Frege’s criterion of identity for
directions along the lines of schema (B) – by saying that if x and y are
directions, then x is identical with y if and only if x and y are directions of
lines that are parallel14 – we cannot, it seems, similarly recast the foregoing
criterion of identity for sets along the lines of schema (A). (We can of
course say, along these lines, such things as that the set of which x is the
sole member is identical with the set of which y is the sole member if and
only if x is identical with y: but that, obviously, does not take us very far.)
Even so, the superiority of schema (B) is perhaps not altogether clear-cut,
because when Frege’s criterion of identity for directions is recast along the
lines of schema (B), it is no longer the case that the logically necessary and
sufficient condition for the identity of directions is expressed without
reference to – or, more accurately, without quantification over – directions.
This is a point to which I shall return later in more general terms, when we
look into the question whether this feature of type-(B) criteria of identity is
a fatal defect in them. To anticipate my eventual conclusion, I do not
think that it is.
4 type-(b) identity criteria and the problem

of circularity
Now, can we plausibly specify criteria of identity conforming to schema
(B) in which the general term ‘φ’ describes objects of some concrete, as
opposed to abstract, sort or kind – such as men, trees, or ships, as opposed
to numbers, sets, or directions? (Precisely how to define the concrete/
abstract distinction is not at all easy, but it will suffice here to take concrete
objects to be ones which are, at least in principle, perceptible and ostend-
ible.15) Many philosophers are quite sanguine about this possibility, myself
included – although some seem to think that the task is rather easier than
I do. One philosopher who is clearly committed to the possibility is
14
How, precisely, one might express this reformulated criterion in logical notation is a matter for some
debate, but one obvious option is the following: 8x8y((direction(x) & direction(y)) ! (x ¼ y $
9w9z(line(w) & line(z) & of(x, w) & of(y, z) & parallel(w, z)))), where ‘of ’ expresses the relation
between a direction and a line that obtains when the former is the direction of the latter – so that the
‘of ’ relation is the converse of the relation between a line and a direction that obtains when the
former ‘has’ the latter.
15
I say much more about the concrete/abstract distinction in my ‘The Metaphysics of Abstract
Objects’, Journal of Philosophy 92 (1995), pp. 509–24.
76 Identity
Donald Davidson, whose well-known criterion of identity for events is
quite clearly formulated along the lines of schema (B). According to
Davidson, if x and y are events, then x is identical with y if and only if x
and y have the same causes and effects.16 The analogy between this and the
criterion of identity for sets stated earlier seems quite obvious. And yet
there is in fact an important disanalogy between the two which is, in my
view, fatal to Davidson’s proposal.
Briefly, the problem is this. For Davidson, causes and effects are themselves
events, so that in making criterial for the identity of events the sameness of
their causes and effects, Davidson is apparently involved in a vicious
circularity.17 No such circularity is involved in the criterion of identity for
sets stated earlier, since that only made the sameness of their members
criterial for the identity of sets – and, while sets may themselves be members
of sets, so too may objects which are not sets (except, of course, in so-called
‘pure’ set theory of the Zermelo–Fraenkel (ZF) type, about which I shall say
more in a moment). Moreover, where sets do have sets as members, it will at
least ultimately be possible (by repeated applications of the criterion) to
settle a question of set identity by reference to the identity of set-members
which are not themselves sets – assuming, at least, the truth of the axiom of
regularity (otherwise known as the Axiom der Fundierung). Even in ‘pure’ ZF
set theory, the criterion will serve the same purpose, because it serves to
ensure that there is just one empty set at the base of the hierarchy of sets. But
observe here that a similar defence of Davidson’s criterion of event identity is
not available: first, because there is no reason to suppose that there is
precisely one uncaused event and, second, because there is no reason to
think that anything corresponds in the case of events to the hierarchy of sets.
In short, what is lacking is an axiomatic theory of events providing for
Davidson’s criterion the sort of framework that axiomatic set theory provides
for our criterion of set identity.18
16
See Donald Davidson, ‘The Individuation of Events’, in his Essays on Actions and Events (Oxford:
Clarendon Press, 1980), p. 179. Later, Davidson withdrew his support for this criterion in response to
some criticisms by W. V. Quine: see his reply to Quine in Ernest LePore and Brian McLaughlin (eds.),
Actions and Events: Perspectives on the Philosophy of Donald Davidson (Oxford: Blackwell, 1985); and, for
further discussion, my A Survey of Metaphysics (Oxford University Press, 2002), pp. 224–8.
17
Davidson denies that there is any formal circularity in his proposed criterion. And that is certainly
correct if we state the criterion as follows: 8x8y((event(x) & event(y)) !(x ¼ y $ (8z(cause(z, x) $
cause(z, y)) & 8z(effect(z, x) $ effect(z, y))))). But – as we shall soon see – this is not to the point.
A criticism similar to mine is developed by J. E. Tiles in his Things that Happen (Aberdeen
University Press, 1981), pp. 1 ff.
18
See further my ‘Impredicative Identity Criteria and Davidson’s Criterion of Event Identity’, Analysis
49 (1989), pp. 178–81.
Perhaps the point against Davidson can be made more graphically by
saying that the set-theoretical principle which is more truly analogous to
his proposed criterion of event identity is the principle that if x and y are
sets, then x is identical with y if and only if x and y include exactly the same
sets – which is an undoubtedly valid principle, since every set includes itself
(that is, has itself as a subset) and mutually inclusive sets are identical.
Now, the latter principle will patently not serve as a criterion of identity for
sets, precisely because it transparently already presupposes an account of the
identity conditions of sets. The corresponding presupposition in David-
son’s proposed criterion is not perhaps quite so transparent as this, but
emerges once it is noted that to say that x and y have the same causes and
effects is – for Davidson, at least, since the only notion of causation that he
admits is that of event causation – just to say that the same events are causes
of both x and y and that the same events are effects of both x and y.
The important lesson to issue from this, then, is that, in formulating a
criterion of identity in conformity with schema (B), the greatest care must
be taken not to presuppose already, in framing the criterial relation R, an
account of φ identity, where ‘φ’ is the general term whose associated
criterion of identity is being sought. The problem is that such a presuppos-
ition may be buried fairly deeply, so that no explicit or formal circularity
can be discerned in the statement of the proposed criterion. However, the
next question that we need to address is whether such a presupposition
might somehow be built into the very nature of any type-(B) identity
criterion, with fatal consequences for all criteria of that type.
5 do type-(b) identity criteria necessarily fall

prey to a circularity problem?
I raise this question because I anticipate at this stage an objection along the
following lines, which picks up a point touched upon earlier. I have urged –
or so it might seem – that a criterion of identity for φs formulated on the
model of schema (B) should not be such that the condition expressed on
the right-hand side of the biconditional, of the form ‘Rxy’, already presup-
poses for its interpretation an account of φ identity.19 But it may be
objected against me – and indeed against schema (B) as a model of identity
19
In point of fact – and this is crucial – I have not urged precisely this: I have urged only that the
criterial relation, R, should not presuppose an account of φ identity, not that the criterial condition,
Rxy, should be (fully) interpretable independently of a grasp of φ identity. The significance of this
distinction will emerge shortly.
78 Identity
criteria quite generally – that there will inevitably be just such a presuppos-
ition, precisely insofar that, in order to interpret the expression ‘Rxy’, we
shall need to be in a position to identify individual φs as the values of the
variables ‘x’ and ‘y’. We are returning then to the point acknowledged
earlier, that type-(B) criteria of identity carry reference to φs – or, more
accurately, involve quantification over φs – in their expression of the
criterial condition for φ identity. And the crucial question is: does this
matter?
To make the matter more specific, suppose that we are attempting to
formulate a criterion of identity for men and that, having framed one along
the lines of schema (B), we go on to try to apply our criterion in a
particular instance involving the use of two singular terms – as it might
be, ‘Napoleon’ and ‘Bonaparte’. According to our criterion, we purport to
know that if a and b are men, then a is identical with b if and only if a
stands to b in the criterial relation R (whatever that might be). But, it may
be objected, unless we already understand the use of the singular terms ‘a’
and ‘b’ – and thereby grasp whatever criterion of identity it is that governs
that use, in accordance with the Fregean thesis – we cannot be in a
position to assess the truth or falsehood of the proposition that a stands
to b in the relation R. Hence our alleged criterion of identity cannot in fact
be correct, since our very capacity to understand and apply it already
presupposes a grasp of whatever criterion of identity it is that governs the
use of a singular term to refer to an individual man.
6 what is the primary purpose of a criterion

of identity?
In order to defend schema (B) against this sort of objection, we need to
examine more searchingly just what purpose a criterion of identity should
be expected to serve. What kind of semantic information is such a criterion
supposed, primarily, to convey? Information about the sense of a singular
term belonging to a certain class, or information about the sense of a
certain general term? From Frege’s remarks in the Grundlagen, one might
initially suppose the former, though I think that in fact, even by his
account, the latter is more properly true. Of course, the two kinds of
information are not independent: what is at issue, however, is a question of
priority.
Let us return to Frege’s example concerning directions. We are told that
the identity statement ‘The direction of line a is identical with the direc-
tion of line b’ is to be understood as being logically equivalent to ‘Line a is
parallel to line b’ – and this is presented, ostensibly, as conveying semantic
information about the use of a singular term of the form ‘the direction of
line x’. Yet it is clear from the context of Frege’s remarks that his primary
aim is to cast light on the meaning of the general term ‘direction’. For what
he is doing in these sections of the Grundlagen is casting about for a way of
defining the concept of number, and using the concept of direction as a
heuristic aid in his search. What he is specifically investigating is the
possibility of defining ‘number’ contextually by laying down a definitional
equivalent to the proposition ‘The number which belongs to the concept F
is the same as that which belongs to the concept G’. In fact, he subse-
quently makes it clear that such an approach is inadequate, because it fails
to help to determine the truth conditions of various other sorts of propos-
itions whose expression may involve the use of the general term ‘number’.
But this is not to say that genuine and useful semantic information
concerning the concept of number or direction is not, after all, conveyed
by laying down a criterion of identity for numbers or directions: plainly,
indeed, it is, even if the information falls short of providing a complete
explication of the concept in question.
Now, to the extent that a criterion of identity for directions conveys
vital semantic information concerning the general term ‘direction’, it also
conveys vital semantic information concerning any singular term of the
form ‘the direction of line x’, simply because the sense of the general term
is a component of the sense of any such singular term. Could it be, then,
that what Frege meant by his key thesis – that ‘if we are to use the symbol a
to signify an object, we must have a criterion for deciding in all cases
whether b is the same as a’ – was effectively no more than this: that when a
singular term is formed with the aid of a sortal or substantival general term,
then simply because (as it may be supposed) a component of the sense of
the latter is a certain criterion of identity, that criterion must be grasped in
order to understand the correct application of the singular term? If so, then
it is not obvious that the thesis is intended to extend to singular terms not
formed in this way, such as proper names strictly so called – although this
conclusion can, of course, be reached if it is conceded, as many self-styled
‘Fregeans’ would doubtless willingly concede, that the sense of any proper
name incorporates that of some sortal or substantival general term.
Seen in this light, the Fregean thesis is primarily one concerning the
semantics of a certain category of general terms – sortal or substantival
general terms – although it has implications also for the semantics of
singular terms just to the extent that such terms incorporate as a compon-
ent of their sense, either explicitly or implicitly, the sense of some sortal
80 Identity
term. That all singular terms in fact meet this condition is a plausible view.
Singular terms such as ‘the direction of line a’, and indeed ‘line a’, do so
quite explicitly, as does one such as ‘the man standing next to me’. And
proper names arguably do so implicitly, since we are standardly introduced
to their correct use either with the aid of singular terms of the previous
sort – as when one is told ‘John Smith is the man standing next to me’ – or
else with the aid of demonstratives used in conjunction with sortal terms,
as when one is told ‘That man is John Smith’.20 In short, it is plausible to
claim that reference to individuals is only ever secured through the
application of general concepts of the sortal or substantival variety.
7 in defence of type-(b) identity criteria

Now, what bearing does all of this have on the objection raised earlier to
criteria of identity formulated in conformity with schema (B)? The objec-
tion turned on the fact that, in a type-(B) criterion of φ identity, reference
to individual φs – or, at least, quantification over them – is not excluded
from the expression of the criterial condition on the right-hand side of the
biconditional. And it was observed that no such complaint could be raised
against type-(A) criteria, of which Frege’s own criterion of identity for
directions is an instance. One question that we must therefore address is
this: was it important for Frege’s purposes that no reference be made to
directions in stating the criterial condition for the identity of directions?
And if so, why? The answer is that initially it was important, because
initially Frege was holding out the hope that a complete definition of the
concept of direction might be supplied contextually by the provision of a
criterion of identity for directions. Now, obviously, in a definition, the
definiendum must not be included in the definiens, so that it was at least
important for Frege at this point that the term ‘direction’ should not
appear on the right-hand side of the biconditional. And this term would
apparently have to appear there if explicit reference to directions were to be
20
This is not to imply that a proper name thus introduced acquires for the auditor just the sense of the
singular term through whose aid the introduction is made – that, for example, ‘John Smith’ acquires
for the auditor the same sense as ‘the man standing next to the speaker’. The most that is being
suggested is that the auditor, if he is to grasp the correct use of the proper name thus introduced,
must grasp that a component of its sense is that of the general term which features in the singular
term through whose aid the introduction is made – that, for example, ‘John Smith’ incorporates in
its sense the sense of the general term ‘man’. We shall see later that provision must also be made for
the introduction of proper names by means of simple demonstratives unattached to sortal terms –
although, at the same time, we shall see that this does not materially affect the view of proper names
being advocated here.
made on the right-hand side of the biconditional, given that the canonical
way to refer to a direction is by means of a singular term of the form ‘the
direction of line x’. But, furthermore, it should be evident that, even if
such explicit use of the term ‘direction’ were to be avoided, a circularity
fatal to Frege’s proposed contextual definition would still be present even
by virtue of any unavoidable quantification over directions on the right-
hand side of the biconditional – because an understanding of the defin-
ition would then still presuppose a grasp of the very concept supposedly
being defined.
However, here we must remember that Frege in fact goes on to reject the
suggestion that a contextual definition of the concept of direction can be
supplied by the provision of a criterion of identity for directions, for
reasons mentioned earlier. These reasons make it clear that this approach
cannot provide for a complete definition of ‘direction’, even though such a
criterion will convey partial semantic information concerning the concept
of direction. But since the hope that a criterion of identity for directions
might supply a complete definition of the concept of direction has now
been given up, the rationale behind the requirement that such a criterion
should make no reference to and involve no quantification over directions
on the right-hand side of the biconditional has disappeared. Hence the
apparent advantage of type-(A) criteria over type-(B) criteria has vanished.
Furthermore, it is clear that the unnecessarily stringent requirement satis-
fied by type-(A) criteria is precisely the source of their limitations observed
earlier. Such criteria are available only in the case of sorts of objects, such as
directions, to which reference may often be eliminated in favour of
reference to objects of other sorts, upon which they are ontologically
dependent – and hence not in the case of objects of an ontologically
independent sort.
But, if all of this is so, what becomes of my earlier criticism of
Davidson’s criterion of event identity on the grounds of its circularity?
That objection still stands, because what was being objected to was not
merely the reference to, or quantification over, events in the expression of
the criterial condition for event identity, but rather the fact that the very
criterial relation R invoked in that condition is itself specified in terms
involving event identity. It may well be that reference to, or quantification
over, events is ineliminable in this context, but that is no reason to suppose
that the identity conditions of events can only be expressed in terms
making appeal to the identity or diversity of events related to them in certain
ways – any more than the corresponding thing follows in the case of sets.
I conclude, then, that type-(B) criteria of identity can be absolved from any
82 Identity
general charge of vicious circularity, even though Davidson’s particular
proposal cannot be thus absolved.
It may be felt, however, that I have still not adequately tackled the challenge
to type-(B) identity criteria posed in section 5 above, particularly as illustrated
in the second paragraph of that section. There it was urged, on behalf of an
objector, that unless we already understand the use of certain singular terms,
‘a’ and ‘b’, and thereby grasp whatever criterion of identity it is that governs
that use, we cannot be in a position to assess the truth or falsehood of the
proposition that a stands to b in the alleged criterial relation R – and hence
that our very capacity to understand and apply a type-(B) principle already
presupposes a grasp of the relevant criterion of identity, which cannot
therefore be expressed by that principle. To this I would reply that it is a
misconception to suppose that a criterion of identity is a principle whose
primary purpose is that of being invoked to settle particular questions of
identity or diversity concerning individuals. Such a criterion is not – at least as
I understand the notion of an identity criterion, and, I believe, as Frege did
also – an epistemic or heuristic principle for the discovery of particular truths of
identity. (That Frege thought likewise is suggested by his remark, quoted
earlier from §62 of the Grundlagen, that it need not always be in our power to
apply a criterion of identity.) Rather, as I have tried to explain, criteria of
identity are primarily semantic principles – albeit ones with important meta-
physical implications – whose grasp is essential to a proper understanding of
general terms of a certain kind, namely sortal or substantival general terms.
I can concede that, in view of this, ‘criterion’ is perhaps not the most apt term
for such a principle, but its use has unfortunately become entrenched and
indeed partly reflects the prevalence of confusion on this score. The crucial
point, then, is that what a type-(B) criterion of identity is intended to convey –
namely an aspect of the sense of a certain sortal term – can be sufficiently
grasped by a person not already equipped to deploy singular terms governed
by that criterion, provided at least that this person understands in general what
it is to make singular reference to or to quantify over individuals. This proviso
and its implications will be our next topic for discussion.
8 can the fregean thesis hold with complete

universality?
A question which now becomes pressing is this: if we are to settle upon
schema (B) as providing the canonical form of a criterion of identity, can
we simultaneously endorse the Fregean thesis as holding quite universally
for all singular terms? That is to say – and assuming that all singular terms
incorporate in their sense the sense of some sortal or substantival general
term – can it be the case that every such general term has associated with it,
as a component of its very sense, some unique criterion of identity that can
be formulated in accordance with schema (B)? Of course, we might seek to
make this true by definition, by stipulating that only a general term
meeting this condition qualifies as a sortal. But that would only serve to
obscure what is in fact an interesting and difficult question – namely
whether there are any general concepts which we apply by individuating
and identifying particulars as falling under them and yet which are such
that this application is entirely criterionless. The question is not the merely
superficial one of whether we are always aware of employing criteria of
identity in our application of certain general terms for these purposes, but
rather the deeper one of whether there always exist such criteria whose –
perhaps only implicit – employment by a subject can always in principle be
elicited. We shall see shortly that empirical psychological findings are
certainly relevant to this question.
Now, it is at least arguable that there are and must be ‘criterionless’
sortals and hence that the Fregean thesis does not hold with complete
universality – not, at least, when a criterion of identity is understood as a
principle conforming to schema (B). One line of argument for this would
be the following. A criterion of identity conforming to schema (B) can,
clearly, only be grasped by someone who at least understands what it is in
general to make singular reference to particular objects, because such a
criterion employs the apparatus of quantification with individual variables.
Therefore – it may be urged – a grasp of such a criterion cannot be a
requirement of every significant use of a singular term to refer to a
particular object. For, it may be said, what acquiring a grasp of a criterion
of identity for φs does is to help to confer, upon someone who previously
lacked it, an ability to make singular reference to particular φs. And this it
does by helping to confer upon that person an understanding of that
general concept of a φ that is incorporated in the sense of any singular
term capable of being used to refer to a particular φ.21 But, it may be
maintained, only someone already conversant with the practice of singular
reference to objects of some sort is in a position to be helped in this way to
21
This is a slight oversimplification because, for instance, someone possessing the general concept of
an animal but not that of a dog is still in a position to make singular reference to particular dogs by
means of singular terms incorporating in their sense only the general concept of an animal – for
example, a singular term such as ‘the animal in that cage’. This is because ‘dog’ shares the same
criterion of identity as ‘animal’. But, for that very reason, the oversimplification is not inimical to the
line of argument that I am developing. See Chapter 2 for further discussion relevant to this point.
84 Identity
extend the range of his objects of singular reference – and consequently, it
seems, his familiarity with that practice must ultimately be grounded in the
application of criterionless general concepts. In short, if this line of
reasoning is correct, it cannot be through a grasp of type-(B) criteria of
identity that a speaker or thinker first engages in the practice of singular
reference.
9 dummett on criteria of identity

Now, it would appear that Michael Dummett, for one, would agree with
this last statement – and yet not agree that there are criterionless sortals,
nor that the Fregean thesis fails to hold with complete universality.22 This
is because he would not agree that all criteria of identity must be express-
ible in conformity with schema (B) – nor even, indeed, in conformity with
schema (A). And this in turn is because he holds the view that at least some
criteria of identity must be such that a grasp of them does not already
presuppose a general capacity to employ singular terms of some sort. For
he holds that a capacity to employ the linguistic apparatus of singular
reference is one of the things that a child must acquire on the basis of a
more primitive level of linguistic achievement which altogether excludes
the use of this apparatus23 – and that the acquisition proceeds precisely
through the progressive grasp of certain primitive criteria of identity which,
obviously, would not require for their articulation the apparatus either of
singular reference or of quantification with individual variables.
What would a primitive ‘Dummettian’ criterion of identity appeal to
and how might it be articulated – what would be its logical form? I should
make it clear that ‘primitive’, in this context, is my choice of epithet, not
Dummett’s own. Dummett himself is disconcertingly obscure about these
matters, although we may assume that he would ultimately seek to invoke
certain spatiotemporal patterns of form, continuity, and persistence
amongst non-particularized ‘features’ of a subject’s perceptual environ-
ment.24 However, a central element of his position is that acquiring a
grasp of such a criterion at least involves acquiring an ability to make, in
appropriate circumstances, statements of the form ‘This is the same φ as
22
My ensuing discussion of Dummett’s position is based on my understanding of what he says in his
Frege: Philosophy of Language, chiefly Chapter 16, and in his The Interpretation of Frege’s Philosophy
(London: Duckworth, 1981), Chapter 11.
23
See Dummett, The Interpretation of Frege’s Philosophy, p. 216.
24
For an elaborate attempt to develop principles of this sort, see Tiles, Things that Happen. Compare
also Strawson, Individuals, pp. 202 ff. and pp. 214 ff. – although Strawson, like Dummett, declines
to go into details.
that’ – accompanied, if need be, by appropriate ostensive gestures – where
‘φ’ is a sortal term.25 Thus at one point he mentions the case of a child
pointing simultaneously to the head and tail of a cat and saying ‘This is the
same cat as that’. And, while accepting that it is not by learning to do this
correctly that a child starts to learn the use of the word ‘cat’, Dummett
nonetheless clearly thinks that acquiring an ability to make such identifi-
cations with the aid of simple demonstratives constitutes an important
staging post in the progression towards a full competence to make singular
reference to objects – objects such as cats. It is enough to know this much
about Dummett’s doctrine to enter into the following discussion.
It is evidently vital to the viability of any position such as Dummett’s
that the use of a sentence involving a demonstrative may be significant and
unambiguous even though no act of singular reference is effected through
its utterance. It is also essential to his position that a sortal term may be
significantly employed in a way which does not presuppose a grasp of
whatever criterion of identity is associated with it – a thesis which Dum-
mett does in fact also uphold but which I shall not seek to challenge here.26
Now, Dummett does explicitly maintain that a statement of the form
‘That is F ’ – where ‘F ’ may or may not be a sortal – may be significant and
unambiguous even though no identifiable object is singled out by the
demonstrative ‘that’, of which the property F is then predicated.27 He even
finds support for this contention in the opening sections of Wittgenstein’s
Philosophical Investigations, concerning ostensive definition.28 I find this
contention unattractive and the evidence of support for it in those sections
of the Philosophical Investigations tenuous. Wittgenstein, it seems to me, is
saying rather that in certain contexts it is quite unnecessary for a speaker to
supplement a demonstrative by a sortal expression in order to effect an
unambiguous act of singular reference intelligible to his intended audience
because the context of utterance itself – including here such things as the
activities that the speaker and audience may be co-operatively engaged in,
such as playing a game of chess – suffices to eliminate any possible
ambiguity.29
25
See Dummett, The Interpretation of Frege’s Philosophy, p. 217.
26
See Dummett, Frege: Philosophy of Language, p. 537, where he says, ‘I can understand when it is right
to say, “That is a book”, before knowing any criterion for the identity of books’.
27
See Dummett, Frege: Philosophy of Language, pp. 572 ff.
28
See Dummett, Frege: Philosophy of Language, pp. 577 ff.
29
See Ludwig Wittgenstein, Philosophical Investigations, trans. G. E. M. Anscombe (Oxford:
Blackwell, 1958), § 31.
86 Identity
This aside, Dummett’s contention concerning statements of the form
‘That is F ’ is clearly thought by him to be related quite intimately to
certain suggestions of Strawson’s concerning the possibility of a ‘feature-
placing’ language altogether lacking the apparatus of singular reference.30
The implication here is that there is a use of a statement of the form ‘That
is F ’ in which it merely conveys the sort of information that might be
conveyed more artificially by saying something like ‘It is F over there’,
where the ‘it’ is no more referential than the ‘it’ in ‘It is raining’.31 I submit
that this is not a correct observation about the adult use of demonstratives
in English, which use is, I believe, directed only at singular reference.32
Furthermore, the highly speculative thesis that there is a stage in an infant’s
linguistic development at which its use of demonstratives does and must
operate like this – a thesis to which Dummett seems to be committed – is,
I believe, neither borne out by empirical evidence nor defensible by a priori
philosophical argument. For what work has been done by developmental
psychologists such as T. G. Bower on early infant perception and related
motor activity strongly suggests in fact that, from their very earliest
months – and certainly before any significant level of linguistic ability
has been achieved – human infants perceptually individuate discrete
objects in their environment and do so in a way which indicates the
exercise of an innate cognitive capacity.33 That being so, however, there
is simply no reason to suppose that, in order to teach a child the correct use
of the linguistic apparatus of singular reference (including the referential
use of demonstratives), one must graft this onto a more primitive level of
linguistic achievement which altogether excludes the use of that apparatus.
The teacher can simply rely on the child’s innate cognitive capacity to
30
See Dummett, The Interpretation of Frege’s Philosophy, p. 217, and Strawson, Individuals, pp. 202 ff.
There is indeed in general a close affinity between Dummett’s and Strawson’s views concerning the
conceptual underpinnings of singular reference, although Strawson seems less committed to making
any empirically significant claim about actual human language mastery or acquisition. I have chosen
to concentrate on Dummett’s account partly for this reason, but more because of his explicit
espousal of what I have called the (full-blown) Fregean Thesis.
31
Compare Dummett, Frege: Philosophy of Language, p. 577.
32
It may be felt that certain uses of demonstratives in questions clearly falsify this claim – for instance,
questions of the type ‘What is that?’ (uttered, perhaps, in conditions of poor visibility), where it
seems that, far from a singular reference being made, one is being sought. I think that various
responses are available to me here, such as that reference is being made to a phenomenal appearance
assumed by the speaker to be intersubjectively perceptible, and that what is really being requested is
a description of the cause of the appearance. However, the issue is not crucial to the points that
I shall go on to develop in this chapter, so I shall not dwell on it here. I do go into such matters more
fully in my ‘Sortals and the Individuation of Objects’, Mind and Language 22 (2007), pp. 514–33.
33
See T. G. Bower, Development in Infancy (San Francisco: Freeman, 1974) and A Primer of Infant
Development (San Francisco: Freeman, 1977).
individuate perceptually at least some of the very same objects to which the
teacher himself may make demonstrative reference for teaching purposes.
And this is where Wittgenstein’s remarks, cited earlier, are very important.
For if the teacher, ostending an object and saying ‘That is F ’ or ‘That is a’,
in order to begin to teach the child the correct use of the predicate ‘F ’ or
the singular term ‘a’, always needed to supplement the demonstrative – or,
in the earliest stages, the bare ostensive gesture – with a sortal term already
understood by the child, then clearly his task would be impossible. What
renders it achievable, however, is not – as Dummett would have it – that
demonstratives and ostensive gestures may be significantly used in certain
contexts without serving to make singular reference, but rather that the
teacher can rely at least sometimes upon the contextual disambiguating
factor of an identity between the object that he is ostending and that on
which the child’s perceptual attention is concurrently fixed – that is, upon
the fact that he and the child are simultaneously individuating the same
object, he ostensively and linguistically and the child perceptually.
The question now arises, if the foregoing account is broadly correct, of
whether the innate individuative capacities that I am ascribing to prelin-
guistic infants are capacities whose exercise necessarily involves, on the part
of the infant, the application of criteria of identity, grasped in some non-
linguistic mode of thought – or, at least, in a mode of thought which does
not involve the understanding of any fragment of a natural language, such
as English. To propose this would be to differ from Dummett not least on
the score of whether all humanly employed criteria of identity are
language-based and culturally transmitted. It would also be to reject as
otiose the notion of what I earlier called a ‘primitive Dummettian criterion
of identity’, supposedly possessing a logical form different from that of
either schema (A) or schema (B). But it would not, of course, commit one
to a rejection of the Fregean thesis in its full-blown form, since it would
merely imply that some of our criteria of identity are innate, not learned –
not that some of our sortal terms and concepts are entirely criterionless.
However, there are philosophers who would, I am sure, endorse precisely
the latter view, where an infant’s deployment of certain sortal concepts is
concerned. They would propose that an infant’s most basic cognitive
capacities to individuate certain perceptible objects in its physical environ-
ment depend not on an implicit intellectual grasp of any principle deserv-
ing of the title ‘criterion of identity’, but simply on certain sorts of sensory
and motor behaviour, which are evolved features of the human brain and its
control mechanisms – in particular, perceptual tracking behaviour and
correlated motor responses such as reaching and grasping. On this sort of
88 Identity
account, the infant’s most primitive acts of object individuation are, thus,
not at bottom mental acts in any intellective sense, but instead essentially
bodily ones. And then the claim would be that an infant incapable of such
primitive bodily acts of object individuation could not subsequently be
trained in the correct use of the referential apparatus of any natural
language.
Now, at one time I had considerable sympathy for a view of this sort,34
partly because I was persuaded by the line of argument sketched in section
8 above for the necessity of there being at least some humanly graspable
sortal concepts that are criterionless, as part of an account of how human
subjects can acquire mastery of the linguistic apparatus of singular
reference.35 And it seemed to me then that the sort of view just outlined
would very naturally and plausibly serve this apparent theoretical need.
However, it has since become evident to me that the ‘perceptual tracking’
model of infant object individuation is deeply flawed, because it cannot
really manage to explain what it has to explain – namely how prelinguistic
infants can succeed in unambiguously singling out certain discrete objects in
their physical environment. Since I have discussed these matters exten-
sively in Chapter 2 above, I shall say no more about them now. As for the
line of argument that led me into sympathy with the ‘perceptual tracking’
model, that too now seems to me to be unpersuasive. For, once again, we
can plausibly appeal to the innateness of certain relevant cognitive capaci-
ties. Only if one assumed that the cognitive capacities essential for mastery
of the linguistic apparatus of singular reference had to be acquired from
experience and training would there be pressure to suppose that these must
be grafted onto some more basic set of cognitive capacities. But if a
capacity for singular thought about discrete environmental objects can be
taken to be innate, as the work of Bower and others suggests, then,
evidently, nothing more cognitively ‘basic’ need be presumed in order to
explain the subsequent mastery of that linguistic apparatus.
Let me now return briefly to Dummett. His position, it seems clear, is
founded at least partly on his open espousal of what he himself calls the
34
See the original paper on which much of this chapter (of the same title) is based, my ‘What Is a
Criterion of Identity?’, Philosophical Quarterly 39 (1989), pp. 1–21.
35
I should mention that I do have certain quite different reasons, of a logico-metaphysical kind, for
supposing that at least some sortal concepts must be criterionless, notably the concept of a person:
see my More Kinds of Being: A Further Study of Individuation, Identity and the Logic of Sortal Terms
(Malden, MA and Oxford: Wiley-Blackwell, 2009), Chapters 2 and 8. I am still persuaded that these
reasons are correct, as I shall make clear in Chapter 6 below. But I now think that it is crucial to
disentangle these considerations from the ones that – as I now believe – misled me concerning the
prerequisites of infant language mastery.
‘amorphous lump’ picture of reality, whereby reality is ‘carved’ into
discrete objects by us, entirely through our application of ultimately
language-based criteria of identity.36 This prevents him from countenan-
cing – indeed, constrains him to dismiss as naive – the view that human
infants are capable of enjoying perceptually based singular thoughts prior
to their acquisition of a mastery of the apparatus of singular reference
embodied in some culturally transmitted natural language, and hence
creates for him a spuriously difficult problem as to how that mastery is
achieved. But, for reasons that I shall speak more about in Chapter 6,
I consider that the ‘amorphous lump’ picture of reality is not merely
empirically unsupported but fundamentally incoherent.
10 the fregean thesis upheld

I conclude that the search for primitive Dummettian criteria of identity is
uncalled for, since there is no empirical or philosophical reason to suppose
that they are required in order to account for an infant’s mastery of the
linguistic apparatus of singular reference or, more generally, to underpin
our capacity to individuate objects. Now, admittedly, this is not quite the
end of the matter, since it might still be hypothesized – and, indeed,
conceivably be confirmed empirically – that an infant’s exercise of its innate
capacity to individuate objects in its perceptual environment is ultimately
grounded in certain cerebral information-processing procedures which utilize
a computational code whose syntax and semantics are those of a ‘feature-
placing’ language, rather than a subject–predicate one like that of first-order
quantificational logic. Hence some of the questions presented by the search
for primitive Dummettian criteria of identity, which I have avoided by appeal
to innate individuative capacities, might conceivably arise again at the level of
what some philosophers have – rather inaptly – called the ‘language of
thought’,37 in the form of questions concerning the nature of the algorithms
supposedly utilized by the brain in its realization of the exercise of those
capacities. However, disputes at this level will have no direct bearing on the
Fregean thesis as both Dummett and I have been interpreting it – namely as a
thesis concerning the semantics of the singular terms used by human beings in
their communications with each other by means of natural languages or by
means of humanly invented artificial languages, such as the formalized
languages of logic and mathematics. And, in any case, it seems that empirical
36
See Dummett, Frege: Philosophy of Language, p. 577.
37
See Jerry A. Fodor, The Language of Thought (New York: Crowell, 1975).
90 Identity
scientists are still a very long way off from possessing the requisite tools and
data to be able to address these new questions concerning the ‘format’ of
human cerebral information processing.
Fortunately, then, we do not need to go into the obscure matter of just
what primitive Dummettian criteria of identity are supposed to appeal to
and how they might be articulated. But having rejected such criteria as
otiose and spurious – at least in the context of the semantics of natural and
artificial languages – we may, I contend, nonetheless retain the claim that
the Fregean thesis holds with complete universality: that there must be a
criterion of identity associated with every significant singular term.38 We
can also still maintain that these criteria may always be formulated in
accordance with schema (B) – although how precise such a formulation
may be where ordinary language is concerned is a contentious matter, to
which I shall return shortly. We can similarly still maintain, of course, that
every such singular term incorporates in its sense the sense of some sortal
term, which duly determines the criterion of identity associated with that
singular term. We need not abandon, then – not, at least, for any reason
thus far under consideration – the view that a necessary condition of a
general term’s being a sortal term is that a criterion of identity should be a
component of its sense. We need not countenance the thesis that there are
and must be criterionless sortals, supposedly supplying concepts in terms of
which human infants ‘primitively’ individuate objects in their perceived
environment, because even an infant’s perceptual individuation of such
objects may be – indeed, I think, must be – seen as being governed by
criteria of identity, even if only by criteria that are merely implicitly grasped
by the subjects responsive to them. As I remarked in Chapter 2, something
like the concept of a relatively cohesive, discrete, and intactly mobile hunk
of matter seems to figure prominently in a human infant’s segmentation of
its perceived environment into distinct objects. But, as I also indicated
there, it would be wrong to think of this concept as not being a sortal, and
also wrong to think that its application is not governed by a distinctive
criterion of identity – wrong, in particular, to suppose that its application
is somehow secured just by an infant’s perceptual ‘tracking’ behaviour and
38
I am setting aside here, for present purposes, the considerations mentioned in an earlier footnote
that lead me to think that certain sortal terms and concepts, notably the concept of a person, are
criterionless. In their case, however, my claim is only that a properly non-circular criterion of identity
is not available, and I am happy to concede – indeed, to insist – that there are important constitutive
principles substantively constraining the identity conditions of individuals falling under such
concepts, so that these identity conditions are not just ‘brute’, in the sense that nothing
informative can be said about them.
associated motor responses, such as reaching and grasping, without the
benefit of any implicit intellective grasp of such an object’s distinctive
identity conditions.
Now, it is evidently possible that the sortal concepts featuring in an
infant’s most basic individuative practices are not precisely to be identified
with any of the sortal concepts still exercised by an adult equipped with a
full mastery of a natural language. To that extent, we should not expect to
be able to capture any one of these infantile sortal concepts exactly in terms
of any normal English expression, such as ‘hunk of matter’, since the latter
has for adults connotations which it would be unreasonable or implausible
to project upon infant subjects. This acknowledgement, however, does not
in principle undermine my earlier stance on the question of how infants
might at first be introduced to the use of singular terms in natural
language, even if it makes the matter a little more complicated. There is
no question of my having to go back on my repudiation of the ‘amorphous
lump’ picture of reality, only a question of a slight degree of mismatch
between infantile and adult individuative schemes.
11 type-(b) identity criteria for concrete

natural-language sortals
It only remains now to be shown that some sortal terms in natural
language describing kinds of concrete objects can indeed be seen as being
governed by criteria of identity which can be formulated in accordance
with schema (B) – for the only specific proposal to this effect so far
investigated has been Davidson’s, which was found to be defective.
I would suggest that a good many artefactual sortals may fairly readily be
shown to submit to such a treatment – sortals such as ‘ship’, ‘watch’, and
‘table’. These are certainly not criterionless sortals, supplying concepts of
the type allegedly exercised in an infant’s primitive acts of perceptual
individuation, even if we thought – mistakenly, in my view – that concepts
of that type existed. For a child must clearly learn to recognize that there
are objects such as ships which, to be adequately conceived of, need to be
thought of as being capable of persisting through certain kinds of change
in their component parts. Of course, spelling out explicitly an appropriate
criterion of identity in this case is by no means an easy matter, as the
notorious problem of the ship of Theseus makes clear.39 Nonetheless, at
39
I discuss this problem and some of its implications in my ‘On the Identity of Artifacts’, Journal of
Philosophy 80 (1983), pp. 220–32.
92 Identity
least to a first approximation we may suggest as a type-(B) criterion of
identity for ships something like the following principle: if x and y are
ships, then x is identical with y if and only if the collections of parts of
which x and y respectively are composed are such that one of these
collections has been generated from the other by a process of gradual
replacement of component parts broadly preservative of the structural
and functional relationships of those parts – where a total retention of
parts constitutes, of course, a ‘null’ replacement. This proposal needs to be
filled out, evidently, by a suitable description of the sorts of component
parts that are properly in question – not, for instance, subatomic par-
ticles, but characteristic ship-parts, such as masts or propellers – and their
requisite structural and functional relationships, which may differ, of
course, from one kind of ship to another (for instance, sailing ships
and steamships).
Now, the ultimate test of a satisfactory criterion of identity is that its
deliverances should nowhere come into conflict with the formal logical
laws of identity. But it is unreasonable to suppose that any such
precisely satisfactory criterion of identity actually governs the ordinary-
language use of a sortal such as ‘ship’. This, indeed, is why philosophers
can concoct puzzle cases which bring our linguistic intuitions into
conflict where the identity conditions of objects such as ships are
concerned. What does this imply, though? Not that the ordinary-
language use of ‘ship’ is not governed by any criterion of identity at
all, but only that actual usage may reflect the employment of a variety
of competing criteria, none of which may quite meet the exacting
standards demanded of a satisfactory criterion of identity for abstract
mathematical objects such as sets. Nor should we expect ordinary
unreflective language users to be able to articulate explicitly the criterion
governing their use of a sortal such as ‘ship’: their implicit employment
of a specific criterion may, rather, often be revealed only through their
ability to respond in principled ways to appropriate questions put to
them concerning the identity or persistence of ships subjected to various
kinds of change. Finally, it is not to be expected, nor would it be
commendable, for philosophers to rest content with having extracted
the vague and often conflicting criteria governing sortals in actual
everyday use: rather, it is part of their proper task to suggest such
revisions of that use as may best contribute to the elimination of this
vagueness and conflict, to the extent at least that this may be achieved
without depriving the reformed concepts of useful application.
12 conclusions
The main conclusions that I have arrived at in this chapter may now be
summarized as follows.
1 It is no defect in type-(B) criteria of identity that they involve quantifica-
tion over the very sorts of individuals for whose identity they provide a
criterion, although in framing such criteria every caution must be taken
not to presuppose already an account of the identity conditions of the
individuals concerned. On the other hand, type-(A) criteria of identity,
while they do – unnecessarily, as it turns out – avoid quantification over
the sorts of individuals for whose identity they supply a criterion, are for
that very reason severely limited in their scope, in a way in which type-
(B) criteria are not: which is why I favour schema (B) over schema (A) as
representing the general canonical form of a criterion of identity.
2 Contrary to what Dummett suggests, we need not suppose that a child’s
initiation into the use of singular terms in natural language must proceed
on the basis of its acquired grasp of identity criteria in principle expressible
in a language altogether lacking the resources for singular reference or
quantification over individuals – a ‘feature-placing’ language. For we may
plausibly contend instead that such initiation merely exploits the child’s
innate capacities for the perceptual individuation of objects – capacities
whose exercise plausibly involves, moreover, at least the implicit intellec-
tual grasp of certain basic criteria of identity, rather than just purely bodily
perceptual ‘tracking’ behaviour and correlated motor responses. Hence,
like Dummett, we may consistently endorse the Fregean thesis as holding
with complete universality, but without conceding to him that this
requires us to recognize a class of identity criteria which conform neither
to schema (A) nor to schema (B). Nor, of course, need we accede to
Dummett’s ‘amorphous lump’ picture of reality.
3 Consequently, we are at liberty to claim – at least so far as any consider-
ation arising in this chapter is concerned – that singular terms in natural
language are always governed by criteria of identity and that such criteria
may always be expected to be capable of expression in conformity with
schema (B). The only caveat required here is that we should not
overestimate the precision of the criteria governing actual ordinary-
language usage, which may well fall short in many cases of the logical
ideal. It remains a legitimate task for philosophers, however, to propose
revisions to and refinements of those actual-language criteria, in order to
bring them into closer proximity with that ideal.
chapter 6
Identity conditions and their grounds
As we saw in Chapter 5, a criterion of identity for entities of a kind K is a

principle specifying, in an informative and non-circular way, the condi-
tions that are logically necessary and sufficient for an entity x and an entity
y of kind K to be numerically the same K. If C is a criterion of identity for
Ks, then a full grasp of the concept of a K will include a grasp of the fact
that the identity of Ks is governed by C. In this sense, C is built into the
concept of a K. Let us call a concept that has a criterion of identity built
into it in this way an individuative concept. Then we can ask: What is the
source or foundation of individuative concepts and how are they related to
the reality that they purport to characterize? Can we freely invent such
concepts, subject only to the constraint of logical consistency, and then
deploy them in ‘carving up’ reality in any of many different equally
legitimate ways? Or are we necessarily constrained by reality itself in the
formation of our individuative concepts – in which case, how? If it is a
mind-independent fact that reality contains entities governed by some
possible identity criteria and does not contain entities governed by others,
how are we to determine which of them it does contain? And what is it
about the entities that reality does contain that is the source or foundation
of their identity conditions? In this chapter, I shall defend robustly realist
and essentialist answers to these questions. And in the course of defending
these answers, I shall argue that reality must contain some entities whose
identity is primitive, in the sense that no informative and non-circular
principle can capture their identity conditions.
1 identity, identity conditions,

and criteria of identity
Many philosophers have, for various reasons, expressed scepticism about
the very idea of a criterion of identity, conceived as a metaphysical – as
opposed to a merely epistemic or heuristic – principle. Some think that
94
Identity conditions and their grounds 95
any role that such a principle could be supposed to play is already played
by Leibniz’s law, so that nothing further needs to be said on the matter.
Others accuse advocates of the idea of supposing, mistakenly, that there is
anything substantive to be said about the relation of identity, whose nature
is in fact perfectly simple and straightforward. Yet others accuse them of
the even grosser error of imagining that identity must be differently defined
for different kinds of entities, and thus of failing to understand that all talk
of identity is perfectly univocal. However, in my view, all such criticisms
are wide of the mark, misconstruing what a criterion of identity is properly
conceived to be. As we saw in Chapters 2 and 5, a criterion of identity for
entities of kind K is simply a principle which specifies, in an informative
and non-circular way, the conditions that are logically necessary and
sufficient for an entity x and an entity y of kind K to be numerically the
same K. (Leibniz’s law, be it noted, is not in this manner a principle
restricted in its application to any single kind of entities, and so cannot
qualify as a criterion of identity in the sense at issue.) It is manifestly the
case that there are such criteria and that these criteria can differ for
different kinds K. For example, if x and y are sets, then x and y are the
same set if and only if x and y possess the same members. By contrast, if x
and y are ordered sets, then they are the same ordered set if and only if they
possess the same members in the same order. As a consequence, no set can
be identified with any ordered set. Why not? Well, consider the ordered
sets ha, bi and hb, ai, which possess the same members but in a different
order. (I assume, for the purposes of the example, that a and b are
themselves distinct.) By the criterion of identity for ordered sets, these
ordered sets are not identical with each other. However, if each of them
were a set, they would be the same set according to the criterion of identity
for sets, because they possess the same members. But then we would have
the intolerable consequence that these entities are both identical and
distinct – a contradiction.
To this objection, a so-called ‘relative’ identity theorist, of P. T. Geach’s
persuasion, might respond that these entities are the same set, but different
ordered sets – the supposed lesson being that entities should never be said to
be identical or distinct simpliciter, or absolutely, but only qua entities of this
or that kind, and that identical Ks may nonetheless be distinct Js.1 But of
1
See P. T. Geach, Reference and Generality, 3rd edn (Ithaca, NY: Cornell University Press, 1980), pp.
181 and 216. I criticize Geach’s view in my Kinds of Being: A Study of Individuation, Identity and the
Logic of Sortal Terms (Oxford: Blackwell, 1989), Chapter 4, and again in my More Kinds of Being:
A Further Study of Individuation, Identity and the Logic of Sortal Terms (Oxford and Malden, MA:
Wiley-Blackwell, 2009), Chapter 4.
96 Identity
what does the relative identity theorist purportedly speak here when he
says that they are at once the same set but different ordered sets? We must
take him to be speaking of the ordered sets ha, bi and hb, ai, for it was
these items that were introduced for purposes of illustration. But what
then is the it with which they are purportedly identical, qua sets? Presum-
ably, the set {a, b}. However, that set – being a set – does not possess its
members in any order, even though we refer to its members in a certain
order when we designate it by the expression ‘{a, b}’. But any ordered set,
such as the ordered set ha, bi, does possess its members in a certain order –
in this case, a comes before b in the order in which they are possessed.
Consequently, the set {a, b} cannot be the same ordered set as the ordered
set ha, bi, for it lacks an essential property of any ordered set, namely that
of possessing its members in a certain order.
Someone may worry at this point that I am neglecting the well-known
method of ‘reducing’ ordered sets to sets, proposed independently by
Wiener and Kuratowski.2 According to this proposal, the ordered set
ha, bi may be ‘reduced’ to the set {a, {a, b}}, while the ordered set hb, ai
may similarly be ‘reduced’ to the set {b, {b, a}}. This proposal preserves, as
it should, the non-identity of the ordered sets ha, bi and hb, ai, by
reducing them to different sets, and so provides no comfort for the relative
identity theorist. But ‘reduction’ is an unfortunate choice of term in this
context, for it is clear that all that the proposal really does is to provide, for
any ordered set, a set which can be regarded as its unique representative.
The ordered set ha, bi certainly cannot be identified with its representative,
the set {a, {a, b}}, because the former has only a and b as its members,
whereas the latter has as its members a and the pair {a, b}. If – as we may
readily presume for the sake of the example – neither a nor b is a set, then
it follows that ha, bi and {a, {a, b}} differ, for instance, in that the latter,
but not the former, possesses a set amongst its members.
To this it may be replied, perhaps, that ‘membership’ must be
understood in two different senses – the ordinary set-theoretical sense,
and another sense which applies only to ordered sets. Let us differentiate
between these putative senses of ‘membership’ by distinguishing between
being a memberS and being a memberO. Then it may be said that b is a
memberO – but not a memberS – of {a, {a, b}}, in virtue of the fact that
{a, b} is a memberS of {a, {a, b}}. Or, more exactly, that b is the second
memberO of {a, {a, b}}, in virtue of the fact that a and {a, b} are the only
2
See Patrick Suppes, Axiomatic Set Theory (New York: Dover, 1972), p. 32. The version that I discuss
here is Kuratowski’s.
two membersS of {a, {a, b}}. It might then be thought that nothing now
stops us from simply identifying the ordered set ha, bi with the set {a, {a,
b}}, because we are no longer entitled to insist that ha, bi does not have the
pair {a, b} as a member: true enough, it doesn’t have it as a memberO, but –
it may now be said – it does have it as a memberS. The fact is, however, that
nothing in the concept of the ordered set ha, bi connects it in any
significant way with the pair {a, b}. The only connection between these
entities was forged, entirely arbitrarily, by the Wiener–Kuratowski pro-
posal itself. It is not as though that proposal provides the only tenable way
of ‘reducing’ ordered sets to sets – indefinitely (indeed, infinitely) many
other ways would work just as well. In other words, it is no part of the
essence of the ordered set ha, bi that it bears any ontological relationship
with the pair {a, b}. By contrast, it is very much a part of the essence of the
set {a, {a, b}} that it bears such a relationship with that pair, for the very
identity of that set is determined by its members, one of which is this pair.
In short, the Wiener–Kuratowski proposal, construed as providing a
genuine reduction of ordered sets to sets, whereby the former may be
identified with sets of a certain kind, does not respect the essential differ-
ences between sets and ordered sets.
2 conceptions, concepts, and essences

I have dwelt on the foregoing example because it is simple and yet also
illustrates some important general points concerning objects and criteria of
identity. One lesson of the example is that the Wiener–Kuratowski proposal
does not provide an adequate conception of an ordered set. It does not, because
it does not accurately reflect the essence of such an entity. By the concept of a
K, I shall henceforth – in effect, by stipulation – mean a conception of a K that
is adequate, inasmuch as it accurately reflects the essence of such an entity.
Concepts and conceptions are ways of thinking of entities but, in the usage that
I am now adopting, conceptions may be more or less adequate ways of
thinking of the relevant entities, whereas concepts are to be understood
precisely as adequate conceptions of them – adequacy in a conception of a
K consisting in the conception’s accurately reflecting the essence of Ks. But
what is the ‘essence’ of a kind of entities, K ? Here I agree with John Locke,
who remarked that in the ‘proper original signification’ of the word essence it
denotes ‘the very being of any thing, whereby it is, what it is’.3 However, we
3
98 Identity
need to distinguish here between general and individual essences. The
general essence of a K is ‘what it is’ to be a K – and is therefore shared by
all particular Ks – whereas the individual essence of a particular K is ‘what
it is’ to be this particular K, as opposed to some other particular K. So, for
example, it is part of the general essence of an ordered set to possess some
entities in a certain order as its members, and it is part of the individual
essence of a particular ordered set to have its particular members in a
certain specific order – for instance, it is part of the individual essence of
the ordered set ha, bi to have just the entities a and b, in that order, as its
members. To the extent that we grasp such facts about the general and
individual essences of ordered sets, we possess the concept of that kind of
entity and of particular entities of that kind, because we possess adequate
conceptions of them, or adequate ways of thinking of them. Such a concept
I shall call an individuative concept, because different particulars may be
said to fall under it, these being individually distinguished from one
another in accordance with a criterion of identity that is built into the
concept and is grounded in the general essence of that kind of entity. An
inadequate conception of an ordered set would, for example, be to think of
such a thing merely as what is represented by (or is ‘reducible’ to) the
corresponding set, according to the Wiener–Kuratowski proposal. Such a
conception might be satisfactory for some purposes, but it would fail to
reflect the essence of an ordered set and in that sense fail to be adequate.
These observations can be extended to entities of other kinds, although in
many cases it will be much harder to determine whether or not a concep-
tion of a putative kind of entities can be deemed adequate.
But, it may be asked, how do we know that we do in fact possess an
adequate conception of entities of some putative kind – even, say, of
ordered sets? After all, the history of set theory notoriously reveals that early
set theorists mistakenly thought that they possessed an adequate conception
of sets. The inadequacy was revealed by Russell’s famous paradox, and
consisted in the fact that a set was conceived to be, in effect, the extension
of any possible predicate that is meaningful – a conception which rested on
the false presupposition that every such predicate has an extension. (The
predicate ‘is non-self-membered’ is meaningful, but cannot, on pain of
contradiction, be supposed to have an extension – the set of sets that are
not members of themselves – for there can be no such set.) A minimum
condition on the adequacy of a conception of a K is that it should at
least be possible – that is, really, or ‘metaphysically’, possible – that Ks,
thus conceived, should exist. Or so I suggest. I suppose it might be
objected by some philosophers that we can have adequate conceptions of
metaphysically impossible entities, such as, perhaps, round square cupolas.
However, this is in effect to abandon all constraints on adequacy of
conception, thereby undermining the very notion of such adequacy. An
adequate conception of Ks must be a coherent way of thinking of Ks – but
there is no coherent way of thinking of something whose existence is
metaphysically impossible. Yet, it may now be asked, how can we know
whether Ks, conceived in a certain way as entities of a certain kind, are
metaphysically possible? If what is being asked here is how we can know
this for certain, without any risk of error, then the answer, I think, is that
we cannot ever know this. However, I nevertheless think that we are very
often entitled to feel confident that we do in fact know that something of
this sort is the case, at least in certain instances. To suppose otherwise is to
abandon all hope that we can engage in rational thought and discourse – a
counsel of despair whose own coherence is dubious in the extreme. What
I am saying, in effect, is that a minimum condition of our own rationality
is that we can, at least sometimes, grasp the essences of at least some
possible kinds of entities. This is consistent with acknowledging that we
may be mistaken in thinking that we have grasped the essence of a certain
kind of entity, but it is also consistent with acknowledging that such errors
may be corrected through further thought and reflection on our own part.
We are, for example, entitled to suppose – but not to be absolutely certain
– that reflection on Russell’s paradox has now provided us with an
adequate conception of a set, where before our conception was inadequate.
Sets and ordered sets, it would certainly seem, are abstract rather than
concrete entities. For that very reason, some may be inclined to deny their
very existence, even if it is granted that they are entities of a kind whose
existence is possible. (Others will say that any abstract entity that could exist
must exist – that abstract entities can only be necessary, not contingent,
beings – but this seems highly questionable.) In any case, it should surely
be agreed that, if one can grasp essences at all, one may grasp the (general)
essence of a possible kind of entity without yet knowing whether or not
entities of that kind actually exist. In this sense, at least, ‘essence precedes
existence’, for many possible kinds of entities. That seems clearly to be the
case with many entities of kinds whose instances, if they exist, are neces-
sarily concrete and contingent beings. For example, we seem to have a
perfectly clear concept of an entity of the following kind: a particular
quantity or mass of homogeneous and infinitely divisible matter or stuff. But
whether there actually are any such entities in existence would appear to be
a question that could at best be settled only empirically. As it turns out, the
actual world, it would seem, contains no such quantities of matter, but
100 Identity
instead – what might be mistaken for them – aggregates of ‘fundamental’
particles that are either punctiform or else spatially extended but nonethe-
less non-composite. Such, at any rate, is what modern theoretical physics
suggests. Superficial observation of the natural world does not reveal to us
which, if either, of these views of physical reality is correct, but the results
of controlled experimentation do appear to favour the second hypothesis
over the first. The most important point, however, is that scientific
experimentation could not be undertaken with a view to supporting either
hypothesis in the absence of clear concepts, on the part of the scientists
concerned, of the kinds of entities whose existence is in dispute. Unless
one already grasps the essence of entities of a putative kind K, one cannot
appeal to empirical evidence in support of – or, indeed, against – the claim
that entities of that kind actually exist. For we need to know what it is
whose existence we are affirming or denying before we can appeal to
evidence in support of our claim.
3 concrete material objects and

their criteria of identity
For purposes of further illustration and discussion, let us focus on some
putative concrete entities of less arcane kinds than those postulated by
fundamental physics – such as a bronze statue and the lump of bronze of
which it is (as we say) ‘made’. What, in essence, are entities of these
putative kinds supposed to be? Well, to a first approximation, a lump of
bronze is supposed to be a maximal connected aggregate of bronze
particles – what Locke called a ‘parcel of matter’. By a connected aggregate
of certain particles I mean a whole consisting of those particles, with each
particle in the whole standing to any other in the ancestral of the adherence
relation, so that between any two such particles there is a chain of particles,
adjacent members of which adhere to one another. And by a maximal
connected aggregate of particles I mean a connected aggregate of particles
which is not a proper part of a larger such aggregate. The criterion of
identity for lumps of bronze is clearly something like this: if x and y are
lumps of bronze, then x and y are the same lump of bronze if and only if x
and y are maximal connected aggregates of exactly the same bronze
particles. For present purposes, we need not inquire too closely into the
nature of bronze particles: we may treat them as minimal parts of bronze;
that is, as parts of bronze which cannot be divided into entities that are
themselves parts of bronze. Of course, modern scientific chemistry will
give a more informative account, in terms of the atomic theory of matter.
However, I am not persuaded that only someone in possession of such an
account can have an adequate conception of what a lump of bronze is: to
insist upon this would be, it seems to me, to impose too stringent a
constraint on adequate conception. It is part of the essence of a lump of
bronze that it is composed of bronze particles, and it may well be part of
the essence of a bronze particle that it has a certain atomic constitution –
but it needn’t follow that it is part of the essence of a lump of bronze that it
is composed of particles with such a constitution. For being part of the
essence of need not necessarily be construed as being a transitive relation. In
any case, this much seems clear: just because a criterion of identity for
entities of a kind K makes reference to, or quantifies over, entities that may
be of other kinds – as, indeed, our proposed criterion of identity for lumps
of bronze quantifies over bronze particles – it doesn’t follow that the
criterion in question cannot be adequately stated without mentioning
criteria of identity for entities of those other kinds. For example: the
criterion of identity for sets, as we have seen, is that if x and y are sets,
then x is the same set as y if and only if x and y have the same members. But
entities of any kind whatever may be members of sets, many such kinds
having different criteria of identity. It doesn’t follow that the criterion of
identity for sets cannot be adequately stated without mention of those
other criteria, whatever they might be. Indeed, if this were a requirement
for an adequate statement of the criterion of identity for sets, no one could
ever adequately state such a criterion, because no one can know all the
criteria of identity for all the kinds of entities that do or could exist.
Now, it is evidently a consequence of the criterion of identity for lumps
of bronze just proposed that no lump of bronze can survive the loss or
replacement of any of the bronze particles composing it, since it cannot be
identical with the lump of bronze resulting from any such loss or replace-
ment. For if lump L1 is composed of particles p1, p2, . . ., pn, and particle pn
is lost or replaced by particle pm, where pn is distinct from pm, then the
resulting lump L2 is composed either of particles p1, p2, . . ., p(n!1) or else of
particles p1, p2, . . ., p(n!1), pm – and in either case L1 and L2 are maximal
connected aggregates of different bronze particles, and so are distinct lumps
of bronze according to the proposed criterion. Indeed, subsequent to such
a loss or replacement of particle pn, there is clearly no lump of bronze in
existence with which L1 can any longer be identified – neither L2 nor any
other lump – and so L1 must have ceased to exist. (A particle may be taken
to be lost by a lump of bronze when it no longer adheres to any of the
particles of which that lump is a maximal connected aggregate, whether
because the particle ceases to exist or because it merely becomes separated
102 Identity
from the other particles. A particle is replaced when, instead of it, another
particle adheres to the remaining particles in question. It may be thought
worrisome that adherence threatens to be a vague relation, in the sense that
there are, plausibly, borderline cases of adherence between particles. I am
not persuaded that this really presents any serious problem, but cannot go
into the matter in any detail at the moment. At any rate, if it is a problem,
it is one that I shall ignore for the time being.)
But what about a bronze statue? What kind of entity is that supposed to
be? Well, clearly, is it is supposed to be a material object made of bronze
and possessing a certain specific shape, fit to represent whatever it is
supposed to be a statue of. (Let us set aside, for present purposes, the sort
of ‘statues’ that might be produced by so-called abstract artists, which need
not represent anything.) Of course, a lump of bronze is made of bronze and
will inevitably possess, at any given time at which it exists, some shape or
other. But that in itself doesn’t suffice to show that a bronze statue may be
identified with a lump of bronze – the lump of bronze of which, at any
given time, it is made. And, indeed, reflection reveals that a bronze statue
is conceived to be an entity of such a kind that it can, in principle, survive
the loss or replacement of some of the bronze parts of which it is made. If,
for instance, some bronze particles from the interior of a statue were to be
lost or replaced – indeed, even if they were to be replaced by particles of a
different kind, such as lead – this would not be regarded as implying that
the statue itself had ceased to exist. And yet, as we know, the implication
would be that the lump of bronze of which the statue had formerly been
made had ceased to exist. This suffices to show that the criterion of identity
for bronze statues differs from that for lumps of bronze, and consequently
that no bronze statue can be identified with a lump of bronze. There really
is no reasonable way to avoid this conclusion, and with it the conclusion
that sometimes two distinct material objects – to wit, a bronze statue and
the lump of bronze of which, at any given time, it is made – exist in exactly
the same place at the same time. At least, there is no reasonable way to
avoid the latter conclusion short of denying the very existence of either
lumps of bronze or bronze statues. For, thus far, I have not been inquiring
into the reasons we may have for thinking that entities of either of these
kinds actually exist, only into their natures or essences.
As for the question of what the criterion of identity for bronze statues is,
given that it differs from that for lumps of bronze, that is not quite so easy
to answer, precisely because the criterion must allow for the possibility of
such a statue’s surviving a loss or replacement of some of the bronze
particles composing it at any given time, but not such a loss as to alter
significantly its outward shape. It would be a useful exercise to try to
formulate such a criterion in a satisfactory fashion, but that is a task that
I shall set aside for present purposes, as the desiderata that need to be met
are clear enough. Meanwhile, we can learn something from the fact that
such a task is not a trivial one. We learn, first, that we were right to insist
that criteria of identity are substantive metaphysical principles, by no
means rendered redundant by Leibniz’s law. We also learn that we may
have a practical grasp of such principles before we are capable of formulat-
ing them explicitly – and for many purposes the practical grasp is all that is
needed.
4 some misunderstandings of the identity-criteria

sceptics exposed
This is a convenient point at which to expose more fully some of the
misunderstandings of those who are sceptical about criteria of identity.
Consider, in particular, those who object to the foregoing line of argument –
based on an appeal to criteria of identity – for the conclusion that two
distinct material objects may exist in exactly the same place at the same
time, such as a bronze statue and the lump of bronze of which it is made.
Typically, they raise an objection of the following sort.4
Consider the bronze statue, S, and the lump of bronze, L, at a certain
moment of time, t. Ex hypothesi, S and L are composed at t of exactly the
same bronze particles, arranged in exactly the same way. But, surely – so
the objectors urge – all of S’s properties at t are determined by, or supervene
upon, the properties and relations of the bronze particles at t, and the same
applies to all of L’s properties at t. In that case, however, how can those
particles simultaneously compose two distinct objects, which differ in
certain of their properties? After all, we have been supposing that S and
L differ in their identity conditions and thus in their persistence conditions,
so that S, for example, has the property of being able to survive a change in
its component particles, while L does not. Again, L supposedly has the
property of being able to survive a significant change in its shape, but S
does not. But, the objector asks, what can possibly be the ontological
ground of these supposed differences in their properties at t, given that S
and L are, at t, indistinguishable in respect of their material composition?
The property of being able to survive a certain sort of change is a modal
4
See, for example, Eric T. Olson, ‘Material Coincidence and the Indiscernibility Problem’,
Philosophical Quarterly 51 (2001), pp. 337–55.
104 Identity
property, which is akin – it might be thought – to an object’s powers or
dispositions. But an object’s powers or dispositions are seemingly grounded
in properties and relations of its material constituents. For instance, if an
object is soluble, or brittle, or elastic, this will be because it is composed of
particles possessing certain properties or arranged in a certain way. Hence
two objects that were composed of exactly the same particles arranged in
exactly the same way couldn’t possibly differ in respect of the powers or
dispositions that they possessed. How, then, could S and L, which are
composed of exactly the same particles arranged in exactly the same way,
differ in that L can survive a significant change of shape but S cannot? Isn’t
this as absurd as supposing that L can dissolve, or break, or stretch – that is,
that L is soluble, or brittle, or elastic – but S cannot? What would make L
and S differ in this supposed way, given their identical material
composition?
Here it may be replied that L and S differ in this way precisely because L
is a lump of bronze, whereas S is instead a bronze statue – and consequently
they possess different persistence conditions. But the objector will respond
that this answer merely pushes back the problem one step. For now he will
ask how it is that S and L, despite having an identical material compos-
ition, come to differ in respect of what kind of thing they are – S having
the property of being a bronze statue, but not the property of being a lump
of bronze, while L has the property of being a lump of bronze, but not the
property of being a bronze statue. Even granted that such properties exist
and are had by objects composed by material particles arranged in a certain
way, how can two objects composed by exactly the same particles arranged
in exactly the same way differ in that only one of them has one of these
properties, while only the other has the other property? In virtue of what do
the two properties get assigned to different objects? How do we even have
two objects here, enabling the previous question to be so much as raised?
The first thing to be said about this supposed conundrum is that it is
patently mistaken to see any sort of parallel between the fact that a bronze
statue cannot survive a significant change of shape and the fact, say, that a
brittle object cannot be bent without breaking. Brittleness in an object is
indeed grounded in the nature of the object’s material constitution, in a
way that detailed scientific investigation could reveal. A statue made of
glass would be brittle for this reason. But the fact that such a statue could
not survive a significant change of shape has nothing, essentially, to do
with its brittleness. True enough, if a brittle statue were to break owing to
its brittleness, it could well undergo a significant change of shape and
thereby cease to exist. But the reason why a statue cannot survive a
significant change of shape and why, hence, a statue that broke in that way
would cease to exist, is simply that having (approximately) the shape it
does is part of the essence of any statue, and thus one of its essential
properties. What it is for something to be a statue is, in part, for it to have
a certain shape. But the essence of a material object is not conferred upon it
by its material constitution, in the way that its powers or dispositions may
be. On the contrary, it is part of the essence of any material object that it
should have a certain sort of material composition. For instance, it is part
of the essence of a bronze statue that, at any time at which it exists, it
should be composed of bronze particles – or, at least, of bronze and/or
suitably similar particles, if it has undergone repair – arranged in a certain
overall shape. Similarly, it is part of the essence of a living organism that, at
any time at which it exists, it should be composed of organic tissue capable
of sustaining the characteristic metabolic processes of life. Thus, far from
its being the case that a bronze statue’s essential properties are conferred
upon it by its material constitution, it is the statue’s essence that deter-
mines what sort of material composition it can have. To believe that
bronze statues exist is to believe that the world includes entities whose
(general) essence is that of a bronze statue, and therefore that bronze
particles sometimes compose such entities, with the persistence
conditions that are implied by that essence. Trying to derive the existence
of bronze statues and their persistence conditions from facts about how
bronze particles are sometimes arranged, without appeal to the essence of
entities of that kind, would be putting the cart before the horse. (Nor
would it help to include facts about the intentions of sculptors and the
like.) The same applies with regard to lumps of bronze and their persist-
ence conditions. And yet, providing precisely such derivations is what the
envisaged objector is, in effect, insisting that we should be able to do. It is
an entirely unreasonable demand to make, and that it cannot be met is
indicative of no failing in the position that the objector is trying to oppose.
Nor, of course, is there anything special about bronze statues and lumps
of bronze in the foregoing regard. The same applies to the bronze particles
whose existence and nature the objector takes to be unproblematic. He
must have some view as to what these essentially are, if he seriously thinks
that there is reason to believe in their existence. This view will have to
include some account of their persistence conditions, if he takes them to be
entities in whose nature it is to persist. One cannot seriously be said to
believe in the existence of entities of some putative kind K if one simply
has no conception of what the essence of such entities might be. For one
cannot believe that Ks exist if one has no conception of what it is to be a K.
106 Identity
What it is to be a K, however, is the general essence of Ks, which is
determinative of the identity conditions of Ks – and hence of their
persistence conditions, if they are by their nature persisting entities. (I
should emphasize that I am speaking here of individual persistence condi-
tions, rather than of sortal persistence conditions, in the sense of the latter
explained in Chapter 2. Individual persistence conditions are the condi-
tions necessary and sufficient for the continued existence of a given
individual, and are determined by that individual’s criterion of identity,
whereas sortal persistence conditions are the conditions necessary and
sufficient for an individual of a given sort’s continuing to exist as an
individual of that sort. In other words, individual persistence conditions
are just ‘diachronic’ identity conditions.) Now, in the case of entities like
bronze statues and lumps of bronze, the objector demands that we explain
how their persistence conditions are determined by, or supervene upon,
facts about things of other kinds, namely facts about the properties and
relations of the bronze particles composing them – and then takes us to
task when we find it impossible to meet the demand. But if this sort of
demand were made general, it would apply also with regard to these other
things – in this case, bronze particles. Clearly, however, the demand
cannot be made perfectly general, on pain of either circularity or an infinite
regress. Indeed, it is a misconceived demand, and should not have been
made even in respect of entities like bronze statues and lumps of bronze.
The objector is simply labouring under a misapprehension concerning the
relationship between facts about essence, persistence conditions, and
material constitution. Facts about essence explain facts about persistence
conditions, and impose constraints on facts about material constitution. The
objector mistakenly – indeed, incoherently – supposes that the proper
direction of explanation and determination is the reverse of this.
5 problems with ‘conceptualist’ accounts of essence

and identity conditions
However, seeing off the objector in this fashion does not leave us without
difficult questions on our hands. To be sure, it would be misguided to try
to explain the essence and persistence conditions of composite material
objects like bronze statues and lumps of bronze in terms of facts about
their material constitution – facts about the properties and relations of the
particles that compose them. Facts of the latter sort are simply not suited
to be the ground or source of facts about the essence and persistence
conditions of such objects. But what, then, is their ground or source? Here
we need to ask what, if anything, essences are – what their ontological
status is. One very common view would be that, to the extent that talk
about essences is legitimate at all, essences are conceptual in character. On
this view, the essence of a kind of entities K is simply constituted by ‘our’
concept of a K – or, if not by ‘our’ concept, then at least by some thinking
being’s concept. But what is a concept? As I indicated earlier, I myself
regard a concept as a way of thinking of something – but I have also
stipulated that a concept is an ‘adequate’ conception of something, which
accurately reflects the nature or essence of the entity or entities in question.
Clearly, then, it is not open to me to say that the essence of Ks is
constituted by our concept of a K, because this would leave no room for
the notion that the concept of a K is a conception of Ks that reflects the
essence of Ks. Essences, on my account, must be mind-independent, if the
question can sensibly be put as to whether or not a conception of Ks
adequately reflects the essence of Ks. But before I defend and explicate my
own view, what can be said of the rival view, that essences are always
constituted by concepts? Let us call this view conceptualism. First, concep-
tualism is a strongly anti-realist view. Second, it is a view that is doubtfully
coherent. Let us deal with the first point first.
In virtue of what, according to conceptualism, can it be truly said that
there exist entities that fall under, or satisfy, our concepts – including,
most centrally, our individuative concepts? That is to say, what does it take
for there to be Ks, on this view? This is simple enough, it may be said: there
must be entities that possess whatever features they are that we have built
into our concept of a K. So, for example, if K is lump of bronze, conceived
as a maximal connected aggregate of bronze particles, then there must be
just such things. This will be the case if, sometimes, some bronze particles
adhere to one another so as to form a maximal connected whole. Well and
good: but remember that conceptualism is the doctrine that all essences are
constituted by concepts. So, in particular, the doctrine must be taken to
extend to the essence of bronze particles – what it is to be a bronze particle.
(It must also extend to the essence of the relation of adherence, but I won’t
dwell on that equally important fact for the moment, for the concept of
adherence is not an individuative concept.) Bronze particles, on this view,
exist just in case there are some things that possess whatever features we
have built into our concept of a bronze particle. However, either the
concept of a bronze particle is relevantly similar to that of a lump of
bronze, in that it characterizes the nature of such an entity in terms of
properties and relations of entities of other kinds, or it is not. If it is, then
the next question is just pushed back one stage. If it is not – and this is the
108 Identity
next question – then what does it take for the world to contain entities
falling under the concept? What, in this case, must the world contribute to
the fact that entities of this putative kind exist? Since, according to
conceptualism, all essences are constituted by concepts – where concepts
are understood to be ‘ways of conceiving’ deployed by thinkers – the
conceptualist cannot suppose that how the world is, in respect of what
kinds of entities it contains, is something that is the case independently of
what concepts thinkers deploy. On this view, what it is for the world to
contain entities of a kind K just is for the concept of a K to have
application, or be applicable. Consequently, an adherent of this view
cannot cash out what it is for such a concept to have application in terms
of there being in the world entities answering to the concept. For, as I say,
on this view, there being in the world such entities just is a matter of the
concept’s ‘having application’. So a different understanding of ‘having
application’ must at least implicitly be in play.
What is this alternative understanding? I think that it can only be
something like this: the concept of a K ‘has application’ just in case
thinkers find it useful, or convenient, to conceive of the world as contain-
ing Ks. This may require the concept in question to be logically consistent –
thus ruling out, for example, the applicability of such concepts as ‘round
square cupola’ – but otherwise the constraints would seem to be purely
pragmatic. This, it seems clear, is a deeply anti-realist view. It is a view
according to which, in Hilary Putnam’s well-known words, there isn’t a
‘ready-made world’5 – or, if you like, there isn’t any truth about ‘what is
there anyway’, to use Bernard Williams’s equally familiar phrase.6 Or, yet
again, it is a view according to which, to employ Michael Dummett’s
somewhat less felicitous metaphor, reality is an ‘amorphous lump’ – one that
can be ‘cut up’ in indefinitely many different but equally legitimate ways,
depending on what ‘conceptual scheme’ we or other thinkers happen to
deploy.7 It may also be the view to which David Wiggins is committed,
willy-nilly, by the doctrine that he calls ‘conceptualist realism’ – committed in
virtue of the fact that the only notion of individuation that he admits is a
cognitive one, whereby individuation is a singling out of objects by thinkers.8
5
See Hilary Putnam, ‘Why There Isn’t a Ready-Made World’, in his Realism and Reason: Philosophical
Papers, Volume 3 (Cambridge University Press, 1983).
6
See Bernard Williams, Descartes: The Project of Pure Enquiry (Harmondsworth: Penguin, 1978), p.
64. Williams himself, of course, is not a conceptualist anti-realist, holding that what he calls an
‘absolute conception’ of reality is possible.
7
See Michael Dummett, Frege: Philosophy of Language, 2nd edn (London: Duckworth, 1981), p. 563.
8
See David Wiggins, Sameness and Substance Renewed (Cambridge University Press, 2001), p. 6.
Not only is this view deeply anti-realist; it is also, as I have said, doubtfully
coherent. For those who philosophize in these terms rarely stop to think
about how their doctrine is supposed to accommodate thinkers, their
thoughts, and the concepts that they deploy. For these, too, are putative kinds
of entities, whose essences, according to the conceptualist doctrine, must
like all others be constituted by ‘our’ concepts of them. It is at this point that
the conceptualist manifestly paints himself into a corner from which there is
no escape. There simply is no coherent position to be adopted according to
which all essences are constituted by concepts, because concepts themselves
are either something or else nothing – they either exist or they do not. If they
don’t, then conceptualism is out of business. But if they do, then they
themselves have an essence – what it is to be a concept. The conceptualist, to
be consistent, must say that the essence of concepts is constituted by our
concept of a concept. But what could this mean? And what could it mean,
according to conceptualism, to say that the concept of a concept ‘has
application’ – that there are concepts? I don’t believe that conceptualism
has any intelligible answer to such questions. The lesson, I take it, is that at
least some essences must be mind-independent, in a way that conceptualism
denies. The next task is to try to understand what this entails and how it can
be possible.
6 a realist view of essence and identity conditions

So what are essences, if they are not – or, at least, not always, ‘constituted
by concepts’? The temptation is to say that they are abstract or ‘Platonic’
entities that can be grasped only by the intellect. This temptation should
be resisted, in my view – though not for the reason that I am hostile to
abstract entities quite generally, for I am not. But it would be equally
disastrous, I think, to try to locate essences as entities to be found in the
world of concrete things existing in space and time. This, in effect, was the
error of the objector who challenged us to explain how the essence of a
material object of any kind is determined by, or supervenes upon, its
material constitution. Or, rather, it was the error of any essentialist who
was tempted to try to meet that challenge in the terms in which it was
posed. Paradoxical though it might superficially seem to be, the well-
advised essentialist will say that essences are not entities at all – are not to
be included amongst what there is in the world, whether abstract or
concrete. Rich though the world is in respect of the many kinds of entities
that it contains, the essences of entities of those kinds are not further entities
to be included in an inventory of what there is. A fortiori, essences are not
110 Identity
‘constituted’ by anything – neither by concepts nor by entities of other
types, such as properties or propositions. Part of the point here is that all
entities have essences, so that if we took essences themselves just to be
entities of a certain kind, then they too would have to have essences, and so
on ad infinitum. The essence of an entity of kind K can certainly involve
other entities of various kinds – as, for example, the essence of a bronze
statue involves bronze particles and a certain characteristic shape. But the
essence is not itself an entity, nor is it somehow ‘composed’ or ‘constituted’
by the entities, or kinds of entity, which it ‘involves’. Indeed, we cannot
really do much better than to repeat that the essence of a K is what it is to
be a K – without, of course, any implication that ‘what it is to be a K’ is
itself something, an entity of another kind. If there are, or even merely could
be, Ks, then there is a fact of the matter as to what it is to be a K, which
obtains – at least in many cases – quite independently of thinkers and their
concepts. (Some kinds of entities are genuinely mind-dependent in some
fashion, which is why I qualify the foregoing statement in the way that
I do.) It is such a fact that a thinker must grasp in order to be able to
possess, and evaluate, the belief that Ks exist. For, as I have said before, one
is in no position to be able to judge whether or not there is evidence
in favour of the existence of Ks unless one knows what Ks are supposed to
be – what their essence is. But facts about essence are not made true by the
existence of anything whatever. Essential truths are, in that sense, truths
without truthmakers. And any rational thinker must be able to grasp at least
some essential truths, simply in order to be able to know what it is that he
or she is thinking and reasoning about. Moreover, essences are the ultimate
grounds of all modal truths – at least, where the kind of modality that we
are concerned with is so-called metaphysical possibility and necessity. If
something is possibly or necessarily the case in this sense, it is so in virtue
of the essences or natures of the entities, or kinds of entities, that are
implicated in the modal fact in question. For example, that it is metaphys-
ically impossible for a statue to survive the loss of its shape is a fact that
obtains in virtue of the essence of a statue – in virtue of what it is to be a
statue. Similarly, that it is metaphysically possible for a bronze statue and a
lump of bronze to exist in exactly the same place at the same time is a fact
that obtains in virtue of the essences of a bronze statue and a lump of
bronze.
And so it is with all facts about what is metaphysically possible or
necessary. These facts are not grounded in truths about so-called ‘possible
worlds’, however we choose to conceive of the latter. But are there, in fact,
possible worlds? Before we can answer that existential question, we must
first determine what it is that we are supposed to be talking about – what
kind of entity a possible world putatively is, and so what its essence is. We
may be told, for instance, that a possible world is a maximal consistent set
of propositions, or that a possible world is a maximal spatiotemporally
connected sum of concrete objects. But are such entities possible? Could
entities of these kinds exist – in the metaphysical sense of ‘could’? Perhaps
so – although careful reflection on the matter may reveal that the case is
otherwise. For, in each case, the putative essence of possible worlds is
specified in terms of entities of other kinds – sets, propositions, sums, and
concrete objects. And it may turn out that, in virtue of the essences of
entities of these kinds, there could be no such thing as a ‘possible world’ in
either of the senses canvassed. Be that as it may. It doesn’t really matter for
our purposes, for it is already apparent that facts about possible worlds
could not be the ultimate ground of all metaphysical possibility and
necessity. For they can only constitute that ground if they themselves exist
and hence are possible, and whether this is so will depend upon the essences
of other entities of the kinds just mentioned. But then, even if possible
worlds do exist, they cannot constitute the ground of all metaphysical
possibility, because they cannot constitute the ground of the possibilities
on which their own existence depends. The lesson is that no deep insight
into modal metaphysics can be gained by reflection on possible worlds.
Essences alone are the ground of all modal truth. Possible worlds are, at
best, just one kind of entity amongst very many that do or could exist. And
if they do or could exist, some modal truths will obtain in virtue of their
essences, both general and individual. For example, such a modal truth
might be that nothing can exist in more than one possible world. But, even
so, many other modal truths will obtain in virtue of the essences of other
kinds of entity. Possible-worlds metaphysics is an attempt to get the
metaphysics of modality on the cheap. It is a snare and a delusion for
the unwary metaphysician. But this is a subject to which I shall return in
much more detail in Chapter 8, since it certainly deserves fuller
examination.
7 replies to some anticipated objections

I imagine that essentialism of the sort that I am recommending will be
attacked for a number of predictable reasons, one of which will be this. It
will be complained that such essentialism implies excessive liberality about
the kinds of entity that could possibly exist. For instance, it may not seem
to exclude the possibility of such (allegedly) entia non grata as something of
112 Identity
the following supposed kind: a material object consisting of a lump of
bronze and a live cat situated ten metres away from that lump. The
thought is that such an object would supposedly be one which, in virtue
of its essence, would come into existence whenever a live cat moved to a
position ten metres away from a lump of bronze and would go out of
existence whenever the cat moved to a different distance from the lump –
and that it is absurd to suppose that anything can come into or go out of
existence for this sort of reason. However, it would be rash to suppose,
quite generally, that no entity of any genuine kind can be such that it is part
of its essence that it comes into existence when two entities of other kinds
move to a certain distance from each other and goes out of existence when
they move further apart. After all, a hydrogen atom is just such an entity: it
comes into existence when a proton and an electron are close enough to
each other for the latter to be captured by the former and goes out of
existence when the electron is subsequently ejected from its orbital.
Why, however, do we say that a new entity – a hydrogen atom – is first
created and later destroyed in these circumstances? Because new causal
powers are brought into being when a proton and an electron stand in the
specified relation to one another – causal powers that are not possessed by
either free protons or free electrons individually. These powers must be
attributed to something as their bearer, and this thing is a hydrogen atom. It
is – or so I am strongly inclined to urge – part of the essence of any
material object to have certain causal powers that are not possessed by any
of its material constituents (if it has any) individually. That, indeed, is why
I am very doubtful whether a mere ‘mereological sum’ of material objects
should be considered to be a further material object in its own right. For all
that is supposedly required for the existence of a certain mereological sum
is the coexistence of the various objects of which it is said to be the sum.
I cannot, however, see how such a sum differs from the bare plurality of
those objects. In short, I don’t see how it can be supposed that they
together compose it – and hence that it exists – merely in virtue of their
coexistence. A genuine principle of composition for material objects of any
kind must, I believe, be one in virtue of which an object composed
according to that principle acquires causal powers additional to any that
are possessed by its material components individually.
Returning, then, to our example of the putative object composed by a
lump of bronze and a live cat situated ten metres away from it: if I could be
persuaded that new causal powers were brought into being by this quite
specific arrangement, analogous to those acquired by a hydrogen atom
upon its formation from a proton and an electron, then I would be happy
to recognize this as an object of a genuine kind. But I shall need a lot of
persuasion! Note, in this connection, that lumps of bronze do not fare badly
by this criterion, for in virtue of the mutual adherence of their constituent
particles, they do acquire causal powers that are not possessed by the
particles individually – powers such as a certain moment of inertia that is
different from any possessed by any of those particles individually. Note,
too, that what I have just said about ‘mereological sums’ of material objects
does not carry over to sets of material objects, because sets are abstract
objects and are not, in any case, composed by their members – the members
of a set are not parts of it (even if, following David Lewis’s proposal, we
could say that subsets of a set were parts of it).9 Finally, note that the
requirement that material objects should have causal powers over and
above those individually possessed by their component parts (if they have
any) does not present problems for such objects as bronze statues, which
are ‘made’ of, but not identical with, lumps of matter. For even if a bronze
statue has the same causal powers as the lump of bronze of which it is
made, it still has, therefore, causal powers over and above those individu-
ally possessed by the bronze particles composing both it and the lump – for
the lump itself, as I have just remarked, has such additional causal powers.
Moreover, it is at least arguable that the statue has some causal powers not
even possessed by the lump, such as the power to evoke an aesthetic
response in someone viewing it.
8 some additional observations and a summary

conclusion
As I have already indicated, it is one thing to establish that entities of a
putative kind K genuinely could exist, but another to establish that they
actually do exist. The former task is at least partially an a priori one, which
is completed once we succeed in grasping the essence of entities of kind K.
Consider again the putative kind lump of bronze. A lump of bronze, if there
can be such a thing, is in essence a maximal connected aggregate of bronze
particles and, supposedly, a material object. Provided that there is such a
species of matter as bronze, and provided that it is particulate in nature –
rather than, say, homogeneous and infinitely divisible – then, provided
also that bronze particles can adhere together in the way necessary for them
to compose such an aggregate, such an aggregate can exist. But there is a
further requirement to be met if such an aggregate is to qualify as a
9
See David Lewis, Parts of Classes (Oxford: Blackwell, 1991).
114 Identity
material object, namely that in virtue of its component bronze particles’
adhering together in the specified way, additional causal powers are gener-
ated which can be attributed to the aggregate as a whole but not individu-
ally to its component particles. Only empirical inquiry, it seems, can
ultimately reveal to us whether bronze is a distinctive species of matter
that is particulate in nature; whether its particles can adhere together; and
whether, in virtue of their adhering together, new causal powers are
generated. Then, however, the further empirical discovery that sometimes
bronze particles do adhere together in the required way will suffice to
establish that some lumps of bronze do actually exist. What is clear is that,
inasmuch as a grasp of essences is necessary in order to establish empirically
that entities of any putative kind actually exist, the task of establishing such
existential truths cannot be a wholly a posteriori one, but includes, rather,
ineliminable a priori elements.
We have also seen that, in the case of many kinds of entity, their essence
involves entities of other kinds – as, for example, the essence of entities of
the kind lump of bronze involves bronze particles, and the essence of entities
of the kind hydrogen atom involves protons and electrons. When the essence
of entities of a kind K involves entities of other kinds, Ks stand in a
relationship of essential dependence with entities of those other kinds. Thus,
for instance, hydrogen atoms are essentially dependent on the protons and
electrons that compose them. Again, sets are essentially dependent on their
members. However, it would seem that essential dependence is necessarily
an asymmetrical relationship.10 Thus protons and electrons are not essen-
tially dependent on hydrogen atoms, as is demonstrable by the fact that
they can exist without them. More importantly, the identity of a hydrogen
atom plausibly depends on the identities of its constituent proton and
electron, just as the identity of a set depends on the identities of its
members. (We may speak loosely of hydrogen atoms exchanging electrons,
but I think it would be preferable to say that hydrogen nuclei capture and
lose electrons.) By contrast, the identity of a proton or an electron does not
depend on the identity of any hydrogen atom of which it may happen to
be a component, nor does the identity of any entity depend on the identity
of the sets of which it is a member. But because essential dependence is an
asymmetrical relationship, it would seem that – barring an infinite regress
of an apparently unacceptable sort – there must be some fundamental
kinds of entities, whose essences do not involve entities of any other kind
10
See further my The Possibility of Metaphysics: Substance, Identity, and Time (Oxford: Clarendon
Press, 1998), Chapter 6.
and whose identities, therefore, do not depend on the identities of entities
of any other kind. Entities of such kinds will have primitive identity and,
accordingly, no informative and non-circular statement of their identity
conditions – no criterion of identity – will be forthcoming in their case.
This does not mean, however, that it will be impossible to determine
whether – and if so, when – persisting entities of such kinds come into
being and go out of existence. For if, as it seems they must, such entities
have distinctive causal powers, we may be able to tell, with some degree of
confidence, whether or not such an entity continues to exist by seeing
whether or not its distinctive causal powers continue to be exercised. This,
in effect, is how in practice we tell whether or not something like an
electron continues to exist or has been destroyed. The distinctive causal
powers of an electron are grounded in its essential properties, such as its
distinctive rest mass, unit negative charge and spin of one-half. When, for
example, an electron is ‘annihilated’ by interacting with a positron, their
opposing charges cancel each other and their combined mass is converted
into a burst of energy. From these effects we can judge that nothing
remains that possesses the distinctive rest mass and charge of an electron
and that, therefore, the original electron has ceased to exist. This is despite
the fact that an electron, being a simple or non-composite entity, is not
essentially dependent for its existence or identity on any other particular
entity of any kind.
I am now in a position to give a summary answer to the most pressing
question that has arisen in this chapter: How are identity conditions
grounded? The answer, I propose, is that the identity conditions of entities
of any kind K are grounded in the essence of Ks – what it is to be a K. The
essence of a kind K is not, however, a further entity of any kind, neither
abstract nor concrete. As rational beings, we must be able to grasp the
essences of at least some kinds of entities. Indeed, it is part of the essence of
a rational being that it has such a grasp. In grasping the essence of entities
of a kind K, we come to grasp their identity conditions and may – albeit,
very often, only with some effort – be able to formulate an explicit criterion
of identity for Ks, which specifies their identity conditions in an informa-
tive and non-circular way. Such a criterion will make reference to, or
quantify over, entities of other kinds involved in the essence of Ks, unless
K is a fundamental kind of entity – in which case, its essence will not
involve entities of any other kinds and no informative and non-circular
criterion of identity for Ks will be forthcoming. It is knowledge of essences
that grounds all of our modal knowledge of what is metaphysically possible
or necessary, including our knowledge of what sorts of changes an entity
116 Identity
can or cannot survive or persist through, if indeed it is an entity in whose
nature it is to persist. The persistence conditions of such an entity are not
grounded in its material constitution, if it is a material object, but in its
general essence. This is why it is a mistake to find it at all puzzling that two
distinct material objects, of different kinds and possessing different identity
conditions, may simultaneously be composed by the same material con-
stituents, as in the case of the bronze statue and the lump of bronze of
which it is ‘made’. Our individuative concepts – our concepts of those kinds
that have individual objects as their instances, distinguishable in principle
from one another – are ways of thinking of entities of those kinds that
accurately reflect their general essences, thus involving a grasp of their
identity conditions and thereby an understanding of what makes one
particular instance of such a kind distinct from another. The essences of
such kinds are not constituted by our individuative concepts, but are in
general entirely mind-independent. Conceptualism – the doctrine that all
such essences are constituted by our individuative concepts – leads inexor-
ably to an incoherent global anti-realism and must be rejected. However, it
is not enough, in general, to grasp the essence of entities of a kind K in
order to know whether or not entities of that kind actually exist. For that
purpose, we must appeal to empirical evidence – at least in the case of
concrete, as opposed to abstract, entities. At the same time, to the extent
that such empirical inquiry always depends upon a grasp of essences, there
is an unavoidable a priori element in our knowledge of the existence of
entities of any kind whatever.
part iii
Modality
chapter 7
Identity, vagueness, and modality
I have said a great deal about identity in previous chapters and now want to
develop in more depth the transition, already begun in Chapter 6, to an
explicit focus on modality – that is, the important family of notions that
includes those of necessity, possibility, and contingency. However, before
going into the heart of such notions, there are some important issues in
philosophical logic that I want to discuss concerning the relationship
between identity and modality. More specifically, I want to examine
certain controversial modal claims concerning identity that have been
much debated in the philosophical literature of the past forty years or so,
particularly the alleged necessity of identity – that is, the claim that truths of
identity are necessarily true. Paralleling this claim, as we shall shortly see, is
another often-made claim about identity: that truths of identity can never
be vague or indeterminate, other than merely as a consequence of impreci-
sion in our language or thought about the world. The world itself, it is often
maintained, cannot be vague or contain ‘vague objects’. Both of these
claims about identity – that it can never be contingent and never be
vague – are ones whose alleged proofs, I shall argue, are open to question
and indeed open to question for very similar reasons. The alleged proofs,
I shall try to show, are subtly question-begging and try to make logic
deliver answers to metaphysical questions that logic is inherently incapable
of answering. I shall start with the issue of vagueness and identity and,
along the way, show how this parallels the issue of necessity and identity.
1 vague objects and vague identity: evans’s argument

A good place to begin a discussion of vagueness and identity is with Gareth
Evans’s classic one-page article ‘Can There Be Vague Objects?’,1 whose
brevity belies its subtlety and importance. Despite the paper’s title, Evans’s
1
Gareth Evans, ‘Can There Be Vague Objects?’, Analysis 38 (1978), p. 208.
119
120 Modality
purpose was to demonstrate, by means of a reductio ad absurdum argu-
ment, that there cannot fail to be a fact of the matter as to whether an
object a is identical with an object b – so that his direct concern seems
really to be with vague identity rather than with vague objects.2 It will prove
instructive in due course to compare Evans’s argument with another
notorious ‘proof’ of a metaphysically contentious doctrine, the Barcan–
Kripke proof of the necessity of identity. More precisely, Evans’s argument
may fruitfully be compared with a closely related proof of the non-
contingency of identity. It seems not implausible, indeed, that Evans had
the Barcan–Kripke proof partly in mind as a model for his own argument,
given the obvious similarities between them and the notoriety of the
Barcan–Kripke proof at the time at which he was writing.
As has just been said, what is at stake in Evans’s paper is the possibility
of there failing to be a fact of the matter as to whether an object a is identical
with an object b. That this is so seems clear from his opening remark – ‘It
is sometimes said that the world might itself be vague’ – for he contrasts
vagueness of this supposed kind with ‘vagueness being a deficiency in our
mode of describing the world’, with which he clearly has no quarrel. In
other words, his target is what may be called ontic rather than semantic
vagueness – although whether ‘vagueness’ is really an apt word in the ontic
case is a moot point. As we have already observed, it is also a moot point
whether, in the light of its contents, the title of Evans’s paper is apt in
representing it as concerning the question whether there can be vague
objects. It seems that the real question is, rather, whether there can be
objects whose identities are ontically indeterminate: that is, once again,
whether there can ever fail to be a fact of the matter as to whether an object
a is identical with an object b.
Here, however, another preliminary observation is in order before we
turn to Evans’s argument itself. This concerns Evans’s curious remark that
the idea whose coherence he seeks to call into question is ‘the idea that the
world might contain certain objects about which it is a fact that they have
fuzzy boundaries’. This remark confirms that Evans’s concern is with ontic
rather than with semantic vagueness, but it is puzzling in its suggestion
that the idea of ontic indeterminacy of identity is necessarily connected
with the idea of the possession of ‘fuzzy boundaries’ – by which one
assumes is meant ‘fuzzy’ spatial or temporal boundaries. It is true that cases
2
A closely related argument was independently developed by Nathan Salmon – see his Reference and
Essence (Oxford: Blackwell, 1982), pp. 243–6. See also David Wiggins, Sameness and Substance
Renewed (Cambridge University Press, 2001), pp. 162–3. But I shall concentrate on Evans’s version.
Identity, vagueness, and modality 121
of semantic vagueness frequently concern the drawing of such boundaries.
For instance, it may be said that our use of the name ‘Mount Everest’ does
not determine a precise spatial boundary between terrain that is part of the
mountain so named and terrain that is not. In this case, there are many
different ways of drawing a precise spatial boundary, all of which are
equally consistent with our use of the name: the ‘fuzziness’ lies not in
any boundary that may be drawn, for each boundary that may be drawn is
a precise one – rather, it lies in our language, which does not determine
that any given one of these precise boundaries must be drawn in preference
to any other. But it is far from obvious that ontic indeterminacy of identity
would have to be grounded in a genuine ‘fuzziness’ in boundaries them-
selves, quite independent of language – as though boundaries could
somehow really be ‘smeared out’. It isn’t even clear what could be meant
by saying this. Fortunately for the advocate of ontic indeterminacy of
identity, however, making sense of such a notion is not crucial to the
position that he seeks to defend. As we shall see in section 5 below,
the most plausible cases of ontically indeterminate identity do not turn
on the issue of boundaries at all. However, Evans’s assumption – that ontic
indeterminacy of identity would have to have something to do with ‘really’
fuzzy boundaries – is widely shared and has done much to perpetuate
scepticism about the possibility of such indeterminacy.3
Now let us turn to Evans’s remarkably simple argument. His ‘proof’
contains just five lines. It begins with the following proposition, assumed
for reductio:
(1) ▽(a ¼ b)
Evans indicates that (1) is to be understood as expressing the assumption
that the sentence ‘a ¼ b’ is of indeterminate truth-value, with the idea of
indeterminacy being expressed by the sentential operator ‘▽’. So, it seems,
(1) may read as ‘It is indeterminate whether it is true that a ¼ b’, or, more
concisely, ‘It is indeterminate whether a ¼ b’. And as was implied earlier,
we may take this to be another way of saying ‘There is no fact of the matter
as to whether a ¼ b’. For the purposes of reductio, (1) is being assumed to
be true, so it is being assumed that there is a fact of the matter as to
whether there is no fact of the matter as to whether a ¼ b. We shall return
to this point later, since it bears on something that Evans says at the very
end of his paper.
3
See Rosanna Keefe, Theories of Vagueness (Cambridge University Press, 2000), p. 15.
122 Modality
To explain and justify the next step of his proof, Evans says that ‘(1)
reports a fact about b which we may express by ascribing to it the property
“λx[▽(x ¼ a)]”’(I use here, for clarity, the more familiar lambda symbol in
place of Evans’s circumflex). Because Evans takes (1) to report this (pur-
ported) fact and expresses the (purported) fact by
(2) λx[▽(x ¼ a)]b
he takes it that (2) follows from (1). I shall assume that (2) may be read as ‘b
has the property of being such that it is indeterminate whether it is
identical with a’, or equivalently as ‘b has the property of being such that
there is no fact of the matter as to whether it is identical with a’.
Next, Evans asserts as a premise this:
(3) ~▽(a ¼ a)
which we may read as asserting ‘It is not indeterminate whether a is
identical with a’, or equivalently as ‘There is a fact of the matter as to
whether a is identical with a’. Presumably, what justifies this premise is
that it is true, and so a fact, that a is identical with a, whatever object a
might be. For, surely, if it is indeed a fact that a is identical with a, then
there is a fact of the matter as to whether a is identical with a – the fact in
question being the fact that a is identical with a.
Evans then supposes that, just as (2) follows from (1), so the following
follows from (3):
(4) ~λx[▽(x ¼ a)]a
Modelling our reading of (4) on that of (2) above, (4) may be read as ‘It is
not the case that a has the property of being such that it is indeterminate
whether it is identical with a’, or equivalently as ‘It is not the case that a has
the property of being such that there is no fact of the matter as to whether
it is identical with a’.
Finally, Evans says that ‘by Leibniz’s law, we may derive from (2) and
(4)’ the conclusion of his proof:
(5) ~(a ¼ b)
Evans clearly has in mind here the version of Leibniz’s law which asserts
that if an object a is identical with an object b, then a has any property that
b has and vice versa. Contraposing, if a does not have some property that b
has, then a is not identical with b. Now, in lines (4) and (2) respectively it is
stated that a does not have a certain property – the property of being such
that it is indeterminate whether it is identical with a – and that b does have
this property. Consequently, it may be inferred from (2) and (4) by the
contrapositive of Leibniz’s law – as above interpreted – that a is not
identical with b, which is what (5) states.
(5) itself does not directly contradict (1), so we do not yet formally have a
reductio ad absurdum proof of the falsehood of (1). To make good this
seeming deficiency, Evans makes the following final remark, which has
given rise to some puzzlement and a great deal of discussion:
If ‘Indefinitely’ and its dual, ‘Definitely’ (‘△’) generate a modal logic as strong as
S5, (1)–(4) and, presumably, Leibniz’s law, may each be strengthened with a
‘Definitely’ prefix, enabling us to derive
(5*) ∆~(a ¼ b)
which is straightforwardly inconsistent with (1).
The first oddity about this remark is that we were initially prompted to
read the sentential operator ‘▽’ not as ‘indefinitely’, but as something like
‘it is indeterminate whether’. In fact, the nearest that Evans comes to
spelling out exactly how we are to read a formula like (1) is when he says,
by way of introducing (1) as an assumption for reductio, ‘Let “a” and “b” be
singular terms such that the sentence “a ¼ b” is of indeterminate truth-
value’. This actually suggests a reading of (1) as ‘The sentence “a ¼ b” is of
indeterminate truth-value’. However, this is a metalinguistic statement,
whereas Evans quite explicitly intended his symbol ‘▽’ to be a sentential
operator; that is to say, an expression that forms a sentence of a given
language when it is prefixed to another sentence of the same language. This
is why it seems natural to read (1) as was proposed earlier, namely as ‘It is
indeterminate whether (it is true that) a is identical with b’.
However, another possible reading would be something like ‘It is
indeterminately true that a is identical with b’, where this is seen as being
analogous to the modal statement ‘It is contingently true that a is identical
with b’.4 And, indeed, this analogy might superficially seem advantageous
if one wants, as was suggested earlier, to draw certain parallels between
Evans’s proof and the Barcan–Kripke proof of the necessity of identity.
But this reading requires us to understand (1) as expressing, so to speak, a
way in which it is (supposedly) true that a is identical with b – to wit,
‘indeterminately’, as opposed to ‘determinately’. It is not inconceivable
that Evans himself did have something like this in mind.5 And, indeed, a
4
Compare Terence Parsons, Indeterminate Identity: Metaphysics and Semantics (Oxford: Clarendon
Press, 2000), pp. 45 and 204 ff.
5
Compare again Parsons, Indeterminate Identity, pp. 204 ff.
124 Modality
reading like this might well be appropriate if semantic vagueness were at
issue, because a ‘supervaluational’ treatment of such vagueness would
supply a reading of ‘It is indeterminately true that a is identical with b’
as saying that the sentence ‘a is identical with b’ is true on some but not all
precisifications of the references of the names ‘a’ and ‘b’.6 After all, it may be
said, being true on some precisifications is, at least in some sense, a way of
being true. However, we are now taking it that semantic vagueness is not
what is at stake – and it is not easy to make clear sense of an ‘ontic’
analogue of such ‘indeterminate truth’. We are taking the interest of
Evans’s proof to lie in its apparent demonstration that there cannot fail
to be a fact of the matter as to whether or not an object a is identical with
an object b. And this is undoubtedly how most other philosophers have
viewed it too. So we shall carry on viewing it in this way.
But now the question is whether, if we view the proof in this way, we
can make sense of Evans’s final remark, quoted above. At first sight, at
least, it doesn’t look as though we can. For if ‘▽’ is read as ‘it is indeter-
minate whether (it is true that)’, or ‘there is no fact of the matter as to
whether’, how could this sentential operator be understood to have a dual,
‘△’, in the sense familiar in modal logic? The modal operators ‘◊’ and ‘□’
are ‘duals’ in this familiar sense, with each being definable in terms of the
other together with negation – so that ‘◊p’ is equivalent to ‘~□~p’ and ‘□p’
is equivalent to ‘~◊~ p’. Obviously, Evans’s remark, quoted above, that (5*)
is ‘straightforwardly inconsistent with (1)’ presumes an analogous equiva-
lence between ‘▽p’ and ‘~△~p’, because only if (1) is thus equivalent to
‘~△~(a ¼ b)’ does it contradict (5*). But if we read Evans’s other operator,
‘△’, as ‘it is not indeterminate whether (it is true that)’, or ‘there is a fact of
the matter as to whether’, do ‘▽’ and ‘△’ turn out to be suitably inter-
definable with the help of negation? Is it the case that ‘▽p’ is equivalent to
‘~△~p’ on this reading? That is to say, is ‘It is indeterminate whether p’
equivalent to ‘It is not not indeterminate whether not p’? (The double
negation here is required, of course, since we have elected to read ‘△’ as ‘it
is not indeterminate whether’.)
Now, ‘It is not not indeterminate whether not p’ is obviously equiva-
lent, by double negation elimination, to ‘It is indeterminate whether not
p’, so our question reduces to one of whether this is in turn equivalent to
‘It is indeterminate whether p’. But then, surprising though this might
have seemed prior to investigation, it turns out that our question does in
fact have a positive answer. For it seems clear that ‘It is indeterminate
6
Compare David Lewis, ‘Vague Identity: Evans Misunderstood’, Analysis 48 (1988), pp. 128–30.
whether p’ is true if and only if ‘It is indeterminate whether not p’ is true.
That is to say, it seems clear that, as we have proposed otherwise to express
it, ‘There is no fact of the matter as to whether p’ is true just in case ‘There
is no fact of the matter as to whether not p’ is true. For if there was a fact of
the matter as to whether not p, this would either be because it was a fact
that not p or because it was a fact that p – and, either way, it would
follow that there was likewise a fact of the matter as to whether p. We see,
then, that even if Evans’s operators ‘▽’ and ‘△’ are interpreted in the
fashion that we have proposed, they do turn out to be interdefinable with
the help of negation in a manner that exactly parallels the interdefinability
of the dual modal operators ‘◊’ and ‘□’.
However, this is by no means enough to confirm Evans’s speculation, in
his final remarks, that his two operators ‘generate a modal logic as strong as
S5’. So we are not in a position to endorse his attempt to turn his
derivation of (5) from (1) and (3) into an argument with a conclusion that
is ‘straightforwardly inconsistent with (1)’, namely (5*), by ‘strengthening’
(1) to (4) and Leibniz’s law with the prefix ‘△’. At the same time, it also
appears that nothing so ambitious as this is needed in order to turn Evans’s
derivation of (5) from (1) and (3) into a formal reductio ad absurdum proof,
given the interpretation of the operators ‘▽’ and ‘△’ now being proposed.
For it appears that on this interpretation we can simply extend the existing
derivation of (5) from (1) and (3) by going on to derive (5*) directly from
(5). Recall once more that, as we are now proposing to interpret it, ‘△’ may
be read as ‘it is not indeterminate whether (it is true that)’, or equivalently
as ‘there is a fact of the matter as to whether’. Now, for the purposes of
reductio, (1) is assumed be true. As Evans himself says, it supposedly
‘reports a fact’. And we may agree with Evans that premise (3) is true –
indeed, that it is logically true. But if the derivation of (5) from these is
valid, then it is truth-preserving, so that if (1) and (3) are true, so too is (5).
But if (5) is true, then it is true, and so a fact, that a is not identical with b,
in which case there is a fact of the matter as to whether a is not identical
with b: which is what (5*) says. So we may extend Evans’s original
argument by deriving (5*) directly from (5). To be sure, to call (5*) a
strengthening of (5), given our proposed reading of Evans’s sentential
operator ‘△’, would be highly misleading. For on this interpretation it is
not the case that (5*) entails but is not entailed by (5) and so (5*) is not in
this sense ‘stronger than’ (5). The question at issue now, however, is
whether the original derivation of (5) from (1) and (3) may legitimately
be turned into a derivation of (5*) from (1) and (3), given the proposed
interpretation of the operator ‘△’ – and it seems clear enough that it can.
126 Modality
And then all that is further needed in order to turn Evans’s original
argument into a formal reductio of (1), on this interpretation, is the
interdefinability of ‘▽’ and ‘△’ that we established earlier, for this allows
us to derive the negation of (1) from (5*).
Let us now briefly sum up our findings so far. It seems that Evans’s
sentential operator ‘▽’ can and should be interpreted as meaning ‘it is
indeterminate whether (it is true that)’, or equivalently as ‘there is no fact
of the matter as to whether’, and that on this interpretation it is, with the
help of negation, interdefinable with his other sentential operator, ‘△’, so
that ‘▽p’ is logically equivalent to ‘~△~p’. It also appears that, with ‘▽’
and ‘△’ thus interpreted, Evans has no problem in turning his original
argument from (1) and (3) to (5) into a formal reductio ad absurdum proof
of the impossibility of ontic indeterminacy of identity, subject only to the
following condition: that his original argument – which we shall hence-
forth refer to simply as ‘Evans’s argument’ – is itself valid. We shall shortly
see, however, that there is reason to suppose that Evans’s argument is not
in fact valid.
2 is evans’s argument question-begging?

There is reason to suspect, on closer inspection of Evans’s argument, that it
is subtly question-begging. By a ‘question-begging’ argument is meant,
roughly speaking, one which in some manner already assumes or presup-
poses something that it is supposed to establish. Now, of course, by no
means every question-begging argument can be convicted of containing an
invalid step. An argument for a conclusion p that had p as its only premise
would be blatantly question-begging, but it does not contain an invalid
step: for p certainly entails p. However, an argument can be more subtly
question-begging than this. It may be, for example, that the validity of a
step in the argument depends in some way upon something that the
argument is supposed to establish – and in this sort of case, its being
question-begging may well be indicative of its being invalid. Such may be
the case with Evans’s argument.
Why, then, might we consider that Evans’s argument is subtly question-
begging? For the following reason.7 The crux of Evans’s argument is his
7
See my ‘Vague Identity and Quantum Indeterminacy’, Analysis 54 (1994), pp. 110–14; and my The
Possibility of Metaphysics: Substance, Identity, and Time (Oxford: Clarendon Press, 1998), pp. 63 ff.
For further discussion, see Harold Noonan, ‘E. J. Lowe on Vague Identity and Quantum
Indeterminacy’, Analysis 55 (1995), pp. 14–19; and my ‘Reply to Noonan on Vague Identity’,
Analysis 57 (1997), pp. 88–91.
use of Leibniz’s law in an attempt to show that, on the supposition that (1)
is true, a and b differ in their properties and hence that a is not identical
with b. The property that b is supposed to possess but a to lack is
symbolized by Evans as ‘λx[▽(x ¼ a)]’. That b possesses this property is
asserted in (2), which is taken to follow from (1). But notice that if it is
valid to derive (2) from (1), then it is equally valid, by parity of reasoning,
to derive the following from (1):
(2*) λx[▽(x ¼ b)]a
(2*) asserts that a possesses the property of being such that it is indetermin-
ate whether it is identical with b. But now we may ask the following
question: is this property, λx[▽(x ¼ b)], which has just been attributed to
a, the same as or different from the property that was previously attributed
to b, λx[▽(x ¼ a)]? These ‘two’ properties ‘differ’ only by permutation of a
and b. So it would appear that, on the assumption that it is indeed
indeterminate whether a is identical with b, it is by the same token
indeterminate whether these properties themselves are identical – and
thereby equally indeterminate whether they are different. (Recall that ‘It
is indeterminate whether p’ is equivalent to ‘It is indeterminate whether
not p’.) But in that case it seems that the most that can be concluded is that
it is indeterminate whether a and b differ in their properties and hence not
that a is not identical with b, but only that it is indeterminate whether a is
identical with b – which is just what was originally assumed.
If this diagnosis is correct, Evans’s argument is question-begging in the
following way. The argument attempts to establish, through an application
of Leibniz’s law, the non-identity of a and b, by showing that b possesses a
property that a lacks. And it attempts to derive this conclusion from the
assumption that it is indeterminate whether a is identical with b. However,
given that assumption, the very property in respect of which b is supposed
to differ from a is one such that it is in fact indeterminate whether it is
different from a property that a must equally be supposed to possess.
Hence the alleged difference in the properties of a and b, required to
establish their non-identity, already presupposes their non-identity and
hence cannot be used to establish it. The problem arises, of course, from a
special feature of the properties concerned, namely their identity-involving
character. The properties in question are λx[▽(x ¼ a)] and λx[▽(x ¼ b)],
which are ‘identity-involving’ in that each of them involves the identity of
an object – a in the one case and b in the other. But since the properties ‘differ’
only by permutation of a and b, their own identity or distinctness turns
entirely on the identity or distinctness of a and b themselves. Hence, if the
128 Modality
latter is indeterminate, as has been assumed, so too is the identity or distinct-
ness of these properties indeterminate – the consequence being that Leibniz’s
law is powerless to distinguish the objects by means of such properties.
However, although this diagnosis calls into question the ability of
Evans’s argument to establish its intended conclusion, it does not yet show
where exactly the argument can be supposed to go wrong. But it will be
noticed that in offering this diagnosis we have said nothing about a crucial
step in the argument, namely the derivation of (4) from (3). (3) seems to be
a perfectly uncontentious logical truth, but (4) is the line in which a is
asserted not to possess the property attributed to b in line (2). Now, as we
have seen, given that the inference from (1) to (2*) is valid – which it must
be if the parallel inference from (1) to (2) is valid – (1) entails that a
possesses a property that is not determinately distinct from the property that
a is denied to possess in (4). The two claims (2*) and (4) are clearly in
tension with each other, because the first attributes to an object a property
that is not determinately distinct from a property that the second denies
that object to possess. But that a possesses the property attributed to it in
(2*) is not an inconsistent claim in itself and cannot be inconsistent with
the trivial logical truth (3). Hence the inference from (3) to (4) must
generate a tension between (2*) and (4) that did not exist between (2*)
and (3). And this implies that the claim made in (4) goes beyond anything
entailed by (3). We may conclude that if Evans’s argument is invalid, the
most plausible place to locate its invalidity lies in the inference from (3) to
(4). This suggestion is one that we shall return to shortly.
3 lessons from the parallel between evans’s

argument and the barcan–kripke proof of
the necessity of identity
As was remarked earlier, there is a seeming parallel between Evans’s
argument and the Barcan–Kripke proof of the necessity of identity8 – or,
more exactly, between Evans’s argument and a closely related modal proof
of the non-contingency of identity.9 To say that objects a and b are
8
For this see Saul A. Kripke, ‘Identity and Necessity’, in M. K. Munitz (ed.), Identity and
Individuation (New York University Press, 1971). Of course, one important formal difference is
that Kripke’s proof does not involve property abstraction. However, in section 4 below we shall be
looking at a ‘stripped-down’ version of Evans’s argument which likewise avoids property abstraction,
so this difference is not as important as might at first be imagined.
9
On these parallels, compare David Wiggins, Sameness and Substance Renewed, p. 163, and Rosanna
Keefe, ‘Contingent Identity and Vague Identity’, Analysis 55 (1995), pp. 183–90.
contingently identical is to say that they are identical but might have
been non-identical. This is a supposition that one might attempt to
reduce to absurdity by means of an argument formally paralleling Evans’s,
simply by reading his operator ‘▽’ as meaning ‘it is contingent that’, on
the understanding that ‘It is contingent that p’ is equivalent to ‘p and
possibly not p’. Thus reinterpreted, Evans’s argument may be para-
phrased as follows. Suppose that (1) it is contingent that a is identical
with b. Then it follows that (2) b possesses the property of being such
that it is contingently identical with a. However, (3) it is not contingent
that a is identical with a. And from this it follows that (4) a does not
possess the property of being such that it is contingently identical with a.
But from (2) and (4) it follows by Leibniz’s law that a is not identical
with b, which contradicts our initial assumption that a is contingently
identical with b (recalling here, once more, that ‘a is contingently
identical with b’ means ‘a is identical with b but a might not have been
identical with b’).
One thing to notice about this argument for the non-contingency of
identity (hereafter ‘NCI’) is that it does not need to be supplemented in
the way that Evans’s argument had to be in order to turn the latter into a
formal reductio ad absurdum proof, because when Evans’s operator ‘▽’ is
read as ‘it is contingent that’ the conclusion (5) directly contradicts the
assumption (1). Although the arguments are formally indistinguishable,
then, their status as formal proofs is not the same.
Notwithstanding this difference between Evans’s argument and the
argument for NCI, both may be charged with committing the same error
of formal inference. The error, if error it is, lies in the inference of (4) from
(3). And, indeed, the Barcan–Kripke proof of the necessity of identity may
also be charged with committing an exactly similar logical error.10 In the
latter case, the erroneous step, according to this line of objection, is the
inference of ‘a possesses the property of being such that it is necessarily
identical with a’ from ‘It is necessary that a is identical with a’. Since this
step – the Barcan–Kripke step, as I shall call it – is much more familiar
than, although formally exactly like, the step from (3) to (4) in Evans’s
proof and the argument for NCI, let us focus on it for the time being.
Now, of course, a general complaint may be raised against the Barcan–
Kripke step that it moves from a proposition ontologically committed
merely to the existence of a certain object, a, to one ontologically commit-
ted in addition to the existence of a certain property – and, indeed, to what
10
Compare my ‘On the Alleged Necessity of True Identity Statements’, Mind 91 (1982), pp. 579–84.
130 Modality
may appear to be a very strange kind of property. However, general
complaints of this sort, for what they are worth, need not at present
concern us, either with regard to the Barcan–Kripke proof or with regard
to Evans’s argument and the argument for NCI. We need have no hostility
towards properties in general and – while it must be acknowledged that we
cannot, on pain of paradox, suppose every meaningful predicate to express
a property – it would be tendentious to respond to the arguments now
under consideration by contending that the properties that they invoke
simply do not exist. Certainly, if one can find fault with the arguments
without needing to deny the existence of the properties, this will be a more
satisfactory method of rebuttal.
So what, exactly, might be thought to be wrong with the Barcan–Kripke
step? Just the following. Even if it is conceded that ‘It is necessary that a is
identical with a’ entails that a possesses some corresponding property, it
may be disputed what property this is – and, of course, there might be
more than one such property. One property that a might be thought to
possess in virtue of the necessary identity of a with a is the property of
being necessarily identical with itself or, more simply put, the property of
necessary self-identity. This, clearly, is a property that a could share with
many other things – plausibly, indeed, it is one which it does and must
share with every other thing. Obviously, this is a quite different property
from the property of being necessarily identical with a which, it seems
evident, a alone can possess. The question then is whether a may be said to
possess the latter property simply in virtue of the fact that it is necessary
that a is identical with a.
To answer this question, we need to think about the grounds of
necessary truths. Some necessary truths are grounded purely in the laws
of logic, which are themselves necessary truths.11 An instance of a logical
law need not itself qualify as a logical law, but it will inherit the necessity of
the law of which it is an instance. The law of the reflexivity of identity –
that everything is identical with itself – is a necessary truth. And an
instance of the law, such as the singular proposition that a is identical
with a, inherits that necessity. Hence it is necessary that a is identical with
a. Against this it may be objected that if a is a contingent being, then a
does not exist in every possible world, whence it cannot be true in every
possible world that a is identical with a. There are various ways to reply to
this objection – for instance, by championing a kind of ‘free’ logic that
allows a singular proposition to be true even if its singular terms are
11
Compare my The Possibility of Metaphysics, pp. 13 ff.
‘empty’, thus denying that it entails the corresponding existential propos-
ition. According to such an approach, that a is identical with a may be true
even in a possible world in which a does not exist, so that even if a is a
contingent being, it may nonetheless be affirmed that it is necessary that a
is identical with a. Another strategy is to say that, where a is a contingent
being, the proposition that a is identical with a is necessary in a restricted
sense, namely in the sense that it is true in every possible world in which a
exists. But whatever we say, it seems clear that we should say that some sort
of necessity attaches to the fact that a is identical with a and that the
ground of this necessity lies in the laws of logic.
What is by no means clear, however, is that the fact that a possesses the
property of being necessarily identical with a – supposing there to be such
a fact – is one whose ground could be held to lie solely in the laws of logic.
The problem is that it would, it seems, be a substantive metaphysical fact of
an essentialist character, whereas the laws of logic are properly conceived as
being metaphysically neutral. No similar concern attaches to the thought
that the laws of logic can ground the fact that a, like anything else,
possesses the property of being necessarily self-identical. The laws of logic
can ground facts about the properties of individuals, but only, it would
seem, facts involving properties that are perfectly general in this way. The
putative property of being necessarily identical with a is not, however, a
perfectly general property. On the contrary, it is a property which, if it
exists, a alone can possess. And the existence of such properties and their
attribution to individual objects are matters for metaphysics, not logic.
The problem with the Barcan–Kripke step, then, is that it purports to
extract a metaphysical fact from a purely logical one. Our proposed
objection to Evans’s argument and the argument for NCI is just the same:
that each of them tries to pull a metaphysical ‘rabbit’ out of a purely logical
‘hat’. This, then, is what seems objectionable about the inference from (3)
to (4) in each case.
Here it may be protested that there can be nothing logically suspect
about that inference because it simply exploits the formal device of so-
called property abstraction, which is equally at work in the inference from
(1) to (2). However, here we may pose a dilemma for the defendants of the
arguments. Either property abstraction is simply a notational reformu-
lation, so that ‘λx[Fx]a’ is just an elaborate way of rewriting ‘Fa’, or else the
property abstract ‘λx[Fx]’ is seriously intended to denote a property, in a
way in which the predicate in ‘Fa’ need not be supposed to do. It should be
borne in mind here, as always, that not every predicate can automatically
be taken to denote a property, on pain of contradiction. If so-called
132 Modality
property abstraction is not understood necessarily to involve the denotation
of a property, then it may indeed be no more than an elaborate rewriting
device with a highly misleading name. But in that case lines (2) and (4) of
Evans’s argument and the argument for NCI are simply superfluous and
we should evaluate the arguments in the form in which they would be left
without them. This we shall do in a moment. On the other hand, if
property abstraction is understood necessarily to involve the denotation of
a property, then neither the inference from (1) to (2) nor the inference
from (3) to (4) can be construed as a perfectly innocent logical step that
cannot be subject to the sort of objection that was raised earlier.
4 a stripped-down version of evans’s argument

Now we need to explore the possibility, just mentioned, of simply stripping
down Evans’s argument and the argument for NCI by removing lines (2)
and (4). The problem now, of course, is that the arguments are supposed
to involve an application of Leibniz’s law, construed as the principle that if
an object a is identical with an object b, then a has any property that b has
and vice versa. And this principle cannot be applied unless properties are
invoked in the arguments. The best that one could do instead is to invoke
the principle of the substitutivity of identity. But how could that possibly
work in the case of Evans’s argument? How are we supposed to derive (5)
directly from (1) and (3) by means of this principle? It might be supposed
that we could assert the following as an instance of the principle of the
substitutivity of identity:
(6) a ¼ b ! (~▽(a ¼ a) ! ~▽(a ¼ b))
and contrapose this to give
(7) (~▽(a ¼ a) & ▽(a ¼ b)) ! ~(a ¼ b)
Then, conjoining (1) and (3) and applying modus ponens to their conjunc-
tion and (7), we might suppose that we could detach the consequent of (7),
which is (5).
One apparent problem with this strategy is that we seem to be using
classical truth-functional operators and classical bivalent logic, when the
presence of the indeterminacy operator precludes us from doing that.12
Thus, for example, the contraposition of (6) to give (7) might be called
into question. However, interesting though this line of objection may be,
12
Compare Parsons, Indeterminate Identity, p. 47.
it has the drawback that it will appear question-begging to someone who
has yet to be persuaded that the notion of ontic indeterminacy of identity
is really intelligible. In any case, such an objection would obviously not be
appropriate when the operator ‘▽’ is interpreted as expressing contingency,
as in the argument for NCI, so let us consider whether it would be
legitimate to reformulate that argument in this stripped-down fashion.
And here we may again note that, in fact, the Barcan–Kripke argument in
its original Kripkean formulation did not make use of property abstraction
and proceeded along lines just like those now under consideration.13
The answer that we may give to this query recapitulates one that may be
given regarding Kripke’s original argument for the necessity of identity.14
In essence, it is this. The principle of the substitutivity of identity is in fact
a schema, of the form
(*) x ¼ y ! (Fx ! Fy)
where the predicate letter ‘F ’ may be uniformly replaced throughout by
any predicate and the variables be bound by universal quantifiers or
replaced by constants to give a logically true formula. In the case of the
argument for NCI, the predicate that would need to be substituted for ‘F ’
in (*) to deliver (6) as an instance of the principle is ‘~▽(a ¼ ξ)’, where ‘ξ’
marks an argument-place to be completed by the name of an object. (6) is
obtained when ‘x’ and ‘y’ are replaced by ‘a’ and ‘b’ respectively. However,
in order to derive (5) from (6), (1) and (3), we must discern this same
predicate as present in (3), on pain of falling foul of a fallacy of equivoca-
tion. Now, of course, it is an article of faith of Fregean semantics that a
proposition like (3), ‘~▽(a ¼ a)’, may be ‘carved up’ in different ways
without this implying that it involves any kind of ambiguity. Thus it is
assumed that (3) may equally well be characterized as saying of a that it is
not contingent that a is identical with it and as saying of a that it is not
contingent that it is identical with itself. However, that these really are just
two ways of saying exactly the same thing is not, perhaps, as uncontentious
as the Fregean orthodoxy assumes it to be.
Even if we set aside the question whether the predicates now at issue
denote properties, it is clear that these predicates have different meanings –
‘is not contingently self-identical’ and ‘is not contingently identical with a’
certainly do not mean the same and so it is at least questionable whether,
when they are predicates of the same subject term, ‘a’, the sentences thus
13
See Kripke, ‘Identity and Necessity’, p. 136.
14
See my ‘On the Alleged Necessity of True Identity Statements’.
134 Modality
formed have exactly the same meaning. When two expressions with
different meanings are each combined with another univocal expression,
to form in each case a meaningful sentence, it would seem surprising that
this could result in their forming sentences with exactly the same meaning.
It is certainly not obvious that ‘a is not contingently self-identical’ and ‘a is
not contingently identical with a’ are synonymous, but both of these
English sentences are supposed to be representable by the same symbolic
formula, ‘~▽(a ¼ a)’, which is assumed to be univocal. And the closest
English equivalent to this formula, ‘It is not contingent that a is identical
with a’, is assumed just to be another way of saying exactly the same thing.
But all of this is certainly open to debate. Indeed, returning to the business
about ‘property abstraction’, it seems that one way of construing this
technical device is precisely as a means of predicate disambiguation, rather
than a means of denoting properties. The idea would be that a formula like
‘▽(a ¼ a)’ is ambiguous, because it can be parsed as resulting from the
combination of the name ‘a’ with either of two different predicates, with
one parsing being read as ‘λx[▽(x ¼ x)]a’ and the other as ‘λx[▽(a ¼ x)]a’.
The whole point of avoiding ambiguity in formal logic is that in such logic
there should be a one-to-one correspondence between meaning and form,
so that valid inferences can be identified as such purely in virtue of their
form. The upshot of all this is that the ‘stripped-down’ version of the
argument for NCI, invoking the principle of the substitutivity of identity
in place of Leibniz’s law, may be accused of involving a fallacy of equivo-
cation which arises from an insufficiently perspicuous logical syntax.
We need to make it clear exactly what, according to this construal, is
objectionable about the ‘stripped-down’ versions of the arguments for NCI
and against the indeterminacy of identity. The objection to the argument
for NCI is that in order for the conclusion (5) to be derived from (1) and (3)
by means of the principle of the substitutivity of identity, the monadic
predicate chosen to replace the schematic letter ‘F ’ in that principle will
have to be ‘~▽(a ¼ ξ)’, rather than ‘~▽(ξ ¼ ξ)’. However, (3)’s status as a
purely logical truth is plausible only if it is parsed as the result of filling
both argument-places of the second of these predicates with the name ‘a’;
that is, as saying of a that it is not contingent that it is self-identical.
Indeed, if (3) is instead parsed as saying of a that it is not contingent that it
is identical with a – which it needs to be if the argument for NCI is not to
involve a fallacy of equivocation – then it appears that the argument turns
out to be question-begging in a perfectly straightforward way, because (3)
so parsed is effectively nothing less than an assertion of the non-
contingency of identity. Recall that a here is an arbitrarily chosen object.
And what (3) so parsed says of this object – and so, in effect, of any
object – is that it is not contingent that it is the very object that it is: in
other words, that it could not have been any other object. But this is
precisely what the doctrine of the non-contingency of identity amounts to.
The alternative parsing of (3) is quite different in its metaphysical import,
for on that parsing (3) merely says of any arbitrarily chosen object that it
could not have failed to be self-identical. And an exactly parallel objection
can be levelled at Evans’s argument, namely that his premise (3),
depending on how it is parsed, is either too weak to sustain his conclusion
that identity is never indeterminate or else implicitly presupposes it. On
the innocuous parsing, (3) says of an arbitrarily chosen object that it is not
indeterminate whether it is self-identical, whereas on the question-begging
parsing it says of an arbitrarily chosen object that it is not indeterminate
whether it is just that object. But precisely what it means to assert that an
object may have indeterminate identity is that an object may be such that
it is indeterminate whether it is just that object, as opposed to another.
5 a plausible example of ontically

indeterminate identity
It is one thing to query Evans’s argument and quite another to say that
there are genuine counterexamples to his conclusion. But there do seem to
be some,15 which are worth describing here, partly in order to illustrate the
point made earlier that ontic indeterminacy of identity need have nothing
to do with ‘fuzzy’ spatial or temporal boundaries and partly to provide
material for some remarks about the notion of ‘singular reference’. One
example involves the capture of a free electron by a helium ion, which thus
comes to have two orbital electrons, one of which is subsequently emitted.
Throughout this episode there exist two electrons, neither of which begins
or ceases to exist during the period of time involved. But, it may plausibly
be maintained, there is no fact of the matter as to whether the electron that
is emitted is identical with the electron that was captured. This is because,
during the period in which both electrons are orbiting the helium nucleus,
they are in a state of so-called quantum ‘superposition’ or ‘entanglement’.
During this time, there are certainly two electrons orbiting the nucleus,
each with a spin in a direction opposite to that of the other: but there is, it
seems, no fact of the matter as to which electron has the spin in one of the
15
Compare my ‘Vague Identity and Quantum Indeterminacy’; and my The Possibility of Metaphysics,
pp. 62 ff.
136 Modality
directions and which has the spin in the other direction. In fact, nothing
whatever differentiates one of the electrons from the other during this
time. Suppose, now, that we call the captured electron ‘a’ and the emitted
electron ‘b’. Then the claim is that there is no fact of the matter as to
whether a is identical with b.
One might imagine that there are in fact two alternative possible courses
of events in this scenario. The first is that the captured electron continues
to orbit the nucleus and the electron that was previously orbiting it is later
emitted. The second is that the captured electron is later emitted and the
electron that was previously orbiting the nucleus continues to do so. But
our claim is that no fact of the matter can distinguish between these
supposedly different courses of events. By this is meant not just that we
cannot possibly tell which course of events actually occurred, but that it is a
misconception to think that there really are these two distinct possibilities.
The facts of the matter just amount to this and not more than this: that
one electron was captured, two electrons orbited the nucleus for a while,
and then one electron was emitted. There is simply no further fact of the
matter as to the identity or distinctness of the captured electron and the
emitted electron. That this is the proper way to characterize the situation
seems to be not only perfectly intelligible but also almost certainly correct.
If Evans’s argument were correct, this could not be so. But now we have
good reason not only to reject Evans’s argument as fallacious, but also to
reject the thesis that it is supposed to prove – that ontic indeterminacy of
identity is impossible. It is not only possible, but also very plausibly
exemplified in the domain of sub-atomic particles.16
Here, however, the following complaint may be raised. It may be urged
that if one is to offer a genuine example of ontically indeterminate identity,
then it is important that the singular terms employed – in this case, the
names ‘a’ and ‘b’ – are not terms whose references are vague. They must be
‘precise’ designators, for if they are not, then it would appear that we are
merely dealing with a case of semantic vagueness, not genuine ontic
indeterminacy of identity. But is it not the case, in the foregoing example,
that the names ‘a’ and ‘b’, introduced as names of the captured electron
and the emitted electron respectively, are vague rather than precise desig-
nators? For isn’t it the case that the manner in which these terms have been
introduced leaves it indeterminate whether ‘a’ applies to the emitted
electron and ‘b’ applies to the captured electron? So isn’t it just this
16
See further Stephen French and Decio Krause, Identity in Physics: A Historical, Philosophical, and
Formal Analysis (Oxford: Clarendon Press, 2006).
indeterminacy of reference that leaves it indeterminate whether the
sentence ‘a is identical with b’ is true?
This line of objection would appear to be misplaced. Of course, it
would not be misplaced if it were correct to suppose that there really are
two distinct possible courses of events in the scenario, as outlined earlier.
For in that case we could quite properly say that the name ‘a’ has been
introduced in such a fashion that it is left undetermined whether it refers
(1) to an electron that is captured and thereafter continues to orbit the
nucleus, or (2) to an electron that is captured and is thereafter emitted, or
indeed (3) to an electron that is captured and to another electron that is
later emitted – and similarly with regard to the name ‘b’. But our claim is
that there simply are no distinct possibilities of the sort now being suggested.
To suppose that there are is precisely to suppose that the example under
discussion does not involve genuine ontic indeterminacy of identity – and
as such entirely begs the question at issue. In other words, only if it is
already assumed that the example does not really involve ontic indetermin-
acy of identity can it be classified as a case of semantic vagueness arising
from our failure to fix precisely the references of the names involved. If this
is right, we simply couldn’t fix the references of these names any more
‘precisely’, because the facts themselves don’t admit the distinctions that
would be required for this.
The lesson is that some singular terms may necessarily fail to make
determinately identifying reference. In our example, the name ‘a’ and the
definite description ‘the captured electron’ are such terms. But this is not
to say that they are ‘vague designators’ in the sense required by the
preceding line of objection, for a vague designator in that sense is a
singular term whose reference could be made determinate in principle, or
which, in other words, is capable of ‘precisification’. We might, of course,
still call them ‘vague designators’ in another sense – implying thereby
simply that statements containing them may be of indeterminate truth-
value, without any presumption that their references could be precisified
so as to eliminate such indeterminacy.17 It would be improper to com-
plain, then, that our proposed counterexample to Evans’s thesis defeats
itself by turning into a harmless case of semantic vagueness, because it
can only be seen in that light if it is already presumed that ontic
indeterminacy of identity is not involved in the case. And it would be
17
On these contrasting conceptions of a vague singular term, compare Keefe, Theories of Vagueness,
pp. 159–60.
138 Modality
similarly question-begging, of course, to raise a similar complaint in
defence of the argument for NCI, by invoking the distinction between
‘rigid’ and ‘non-rigid’ designators. Both complaints attempt to rebut a
metaphysical thesis by semantic sleight of hand. As such, they repeat the
original error of Evans’s argument and the parallel argument for NCI: the
error of trying to establish substantive metaphysical claims by means of
purely logical argument.
chapter 8
Necessity, essence, and possible worlds
The claim that I want to argue for in this chapter is one which, I acknowledge,
many analytic metaphysicians will consider to be at best highly provocative
and at worst extremely repugnant. It is that talk of possible worlds can
contribute nothing whatever of substance to our understanding of the nature
and ground of metaphysical modality – that is, metaphysical necessity and
possibility. If I am right, then a great many philosophers and logicians have
been wasting a good deal of their time in elaborating accounts of the putative
nature of possible worlds and their supposed relationships to one another.
I readily concede that their time has not been completely wasted, to the extent
that what they have been doing is engaging in a purely technical exercise of
modelling various different systems of modal logic set-theoretically, using
‘possible worlds’ as elements in the set-theoretical structures constructed for
these purposes. For these purposes, however, it really doesn’t matter what
‘possible worlds’ are supposed to be, much less whether or not they really exist.
My criticisms will be directed primarily at those amongst these theorists
whose attitude towards talk of possible worlds is ontologically serious,
although it will extend also to those who, while not being ontologically serious
about possible worlds, still think that the ‘fiction’ of possible worlds can reveal
something about the nature and ground of metaphysical modality – if indeed
there really are any theorists of the latter kind. In this chapter, then, I shall be
trying to complete the task, begun in Chapter 6, of showing that possible-
worlds metaphysics is an irretrievably flawed attempt to get the metaphysics of
modality ‘on the cheap’.
1 the language and ontology of possible worlds

Why did philosophers ever start talking about possible worlds? For some-
thing like the following reason, it seems, even if what I am about to say is,
in virtue of its brevity, somewhat of a caricature of a complicated story.
First of all, philosophers wanted to understand the nature of necessity and
139
140 Modality
possibility, as reflected in our modal vocabulary and patterns of modal
reasoning. We say things of the form ‘It is necessarily the case that p’ and
‘It is possibly the case that q’, and we reason, for example, from the former
as a premise to ‘It is actually the case that p’ as a conclusion, and from this
in turn to ‘It is possibly the case that p’. Again, we reason from ‘It is
possibly the case that p and q’ as a premise to ‘It is possibly the case that p
and it is possibly the case that q’ as a conclusion, but not vice versa. How
are we to codify this kind of reasoning? One way is to try to formulate
some axiomatic system of modal logic containing certain distinctively
modal axioms and rules of inference, in the hope that it will reflect the
patterns of modal reasoning that we are, intuitively, most strongly inclined
to regard as valid. But it turns out that many different systems of modal
logic can be constructed, no one of which indisputably satisfies this goal
better than any other. We want to decide which of these systems to
endorse. For that purpose, it seems, we need to inquire more closely into
what we mean when we make modal claims. What does it mean to say ‘It is
possibly the case that p’, for instance? It is also observed that modal logic is
non-truth-functional and, more generally, non-extensional. The truth or
falsehood of ‘It is possibly the case that p’ is not determined solely by the
truth or falsehood of p – for, although it is true if p is true, it need not be
false if p is false. So what does determine its truth or falsehood?
Then, however, the following felicitous observation is made: the relation
between ‘necessary’ and ‘possible’ is rather similar to that between ‘all’ and
‘some’, which is one that we already understand very well. If we introduce
quantification over a putative class of entities, ‘worlds’, by constructing
statements of the form ‘In some world, w, it is the case that p’ and ‘In every
world, w, it is the case that p’, and interpret ‘It is possibly the case that p’ as
meaning the former and ‘It is necessarily the case that p’ as meaning the latter,
then – lo and behold! – we find that we can account for many of the putative
entailment relations between modal statements that we find intuitively com-
pelling. Indeed, if we go further and postulate ‘accessibility relations’ between
‘worlds’, and impose different conditions on the character of these relations –
reflexivity, symmetry, transitivity, and the like – we find that we can recon-
struct, in these terms, various of the different rival axiomatic systems of modal
logic that we started with. We seem now to be making genuine headway with
our attempt to understand modal claims and modal reasoning, because we
have found a language in which to recast them – the language of possible
worlds – whose logical characteristics are entirely familiar to us, since it is
extensional and simply makes use of the familiar logical notions of truth-
functions and quantifiers.
Necessity, essence, and possible worlds 141
Is this really an advance in our understanding, though, or only the
semblance of that? Well, if understanding something is reducing it to, or
recasting it in terms of, something else that is already more familiar to us,
then, indeed, this is an advance in our understanding. But one is bound to
wonder whether it wasn’t just an accident of our intellectual history that
the more familiar thing was, in any given case, in fact more familiar to us.
One may certainly wonder about that in the present case. Just because
classical first-order predicate logic was already very familiar to us when the
attempt began to be made to codify systems of modal logic, it doesn’t
follow, of course, that things had to be that way around – difficult though
it may be, from our current perspective, to think of our intellectual history
as proceeding in a contrary direction. Be that as it may, let us grant that for
us, at least, with the intellectual history that we actually have, interpreting
modal statements and reasoning in terms of possible worlds constituted an
advance – at least of some sort – in our understanding of the former. Still,
the process could hardly have been left at the stage that I have so far
brought it to in my thumb-nail sketch. Someone was bound to ask: Well,
if ‘It is possible that p’ means, or is semantically equivalent to, ‘In some
(accessible) world, w, it is the case that p’, don’t we have to know what
‘worlds’ are – what sort of entities we are putatively quantifying over here –
if we are really to understand what we mean by making such a modal
claim? In point of fact, I think it might have been much better if this
question hadn’t been asked, although that can only be said with the benefit
of hindsight. At the time at which the question was raised, there was pretty
widespread endorsement of Quine’s notorious criterion of ontological
commitment – ‘to be is to be the value of a variable’.1 In that climate,
not to have been prepared to venture a view as to what sort of things
‘worlds’ might be would have smacked of intellectual dishonesty, if not
downright cheating. If we are going to take quantification over ‘worlds’
seriously, it would have been said, then we must venture to say what
worlds are supposed to be – and, indeed, offer an account of their identity
conditions, in view of Quine’s other famous dictum, ‘No entity without
identity’.2 But if we are not going to take it seriously, how can we suppose
that it really throws any light at all upon the meaning of modal claims?
What happened, thus, was that a good many philosophers set them-
selves to carry out this task: to construct theories of possible worlds, telling
1
See W. V. Quine, ‘Existence and Quantification’, in his Ontological Relativity and Other Essays
(New York: Columbia University Press, 1969).
2
See W. V. Quine, ‘Speaking of Objects’, in his Ontological Relativity and Other Essays.
142 Modality
us about the nature of these putative entities and explaining their supposed
identity conditions. What would count as a good theory of this kind? Well,
first of all, of course, the theory would have to represent ‘worlds’ as being
things of a kind such that the various quantified sentences whose variables
supposedly range over these things make sense to us, as putative transla-
tions of ordinary modal statements. For example, it obviously wouldn’t do
to suppose that ‘worlds’ are things like planets or galaxies, not least because
one of the worlds is supposed to be the actual world and that cannot with
any plausibility be identified with, say, the Earth or the Milky Way. We
might perhaps suppose, however, that the actual world is the entire universe,
in the cosmologist’s sense – the very big thing that supposedly originated
with the Big Bang and is now some 13 billion or so years old. David Lewis,
of course, took that sort of route in his theory of possible worlds.3
Accordingly, he took other possible worlds to be other very big things of
a similar sort: in effect, parallel universes. Other theorists – fearing, perhaps,
the charge that they were crudely intruding into the domain of empirical
science – took a more Platonic line, proposing that worlds, including the
actual world, are abstract entities of an extremely complicated sort, such as
maximal consistent sets of propositions.4 Thus the actual world, on this
view, turns out to be the set of all actually true propositions – assuming, of
course, that there is such a set. Let us call these two broad schools of
thought the concretists and the abstractionists respectively. The concretists
complain, inter alia, that the abstractionists rely upon modal notions in
their account of what worlds are because, for instance, the notion of
consistency is a modal notion – two propositions being mutually consistent
just in case they could both be true together. The abstractionists reply,
perhaps, that their aim is not to reduce modal notions to non-modal ones,
but just to ‘explicate’ them in terms of possible worlds as they conceive of
the latter. Perhaps, too, they attempt a tu quoque response, contending that
the concretists also must, at least implicitly, assume their worlds to be
possible universes – not, for instance, ones in which some contradictions
are true. A vigorous debate thus proceeds between the abstractionists and
the concretists, each denying that the other has postulated the right kind of
thing to be the value of a ‘world’ variable in the various possible-worlds
translations of modal statements. This debate, understandably, develops
into something of a stalemate, so a tie-breaker is needed. Appeal is perhaps
3
See David Lewis, On the Plurality of Worlds (Oxford: Blackwell, 1986).
4
See, for example, Alvin Plantinga, The Nature of Necessity (Oxford: Clarendon Press, 1974), though
he invokes (what he calls) states of affairs rather than propositions.
made, at this point, to familiar principles of theoretical economy and
simplicity and a cost-benefit sheet is drawn up for each theory – rather
differently, of course, by its advocates and by its opponents. The method
of inference to the best explanation is perhaps invoked, as though the rival
theorists are somewhat in the position of theoretical scientists with con-
trary hypotheses concerning the underlying nature of some empirically
observable phenomenon, such as the motion of the planets or the spectra
of chemical substances.
Other theorists then emerge upon the scene, seeing themselves as stand-
ing to the concretists and abstractionists rather as instrumentalists stand to
theoretical scientists who advocate different ‘realist’ hypotheses in explan-
ation of certain observable phenomena. These – the modal ‘fictionalists’ –
are the philosophers who, in Russell’s memorable phrase, hope to reap all
the advantages of theft over honest toil.5 Granted, they say, that concretism
can be seen to be more economical and comprehensive than abstractionism,
once the cost-benefit analysis has been done properly, it is surely even more
economical to repudiate its ontological commitments and interpret modal
claims as involving no more than a pretence that concrete worlds – parallel
universes – exist. That way, we can still appeal to concretism to explain
what we mean in making modal claims and how modal logic works,
without having to commit ourselves seriously to the claim that the con-
cretist’s putative worlds are real. But matters are not quite so straightfor-
ward, their realist opponents protest. It proves more difficult than might
have been supposed to throw away the bathwater of concrete possible
worlds without also throwing away the baby that the fictionalist wanted
to save. And so the debate goes on – and on, and on! But should we, in all
seriousness, be engaged in this seemingly endless and irresolvable three-way
debate between concretism, abstractionism, and fictionalism? Standing
back from it all, won’t a level-headed and unprejudiced philosopher want
to say at this point: a plague on all your houses? It is not as though there are
no other options on the table. There is, for instance, so-called combinator-
ialism – though what, exactly, that really amounts to is far from perfectly
clear, and its adequacy as a theory of modality may certainly be called into
question.6 And, of course, there are always plain old-fashioned anti-realism,
eliminativism, and conventionalism, for those whose taste in matters of
modality is even more deflationary than the fictionalist’s – deflationary to
5
See, especially, Gideon Rosen, ‘Modal Fictionalism’, Mind 99 (1990), pp. 327–54.
6
See, especially, D. M. Armstrong, A Combinatorial Theory of Possibility (Cambridge University Press,
1989).
144 Modality
the point of rejecting the notion of modal truth altogether, or representing
it as being a matter of semantic stipulation. However, these other options
are apt to seem worse than the ones they seek to replace, at least for any
philosopher who has sympathy for the thought that there are mind-
independent modal truths. What to do, then? Such a philosopher seems
left with two equally unappealing alternatives: either to re-enter the debate
about possible worlds and try to work out a clearly superior theory con-
cerning them, or else to resort to modal primitivism – the view that nothing
illuminating can be said about the meaning and ground of modal claims,
which must accordingly be accepted as ‘brutely’ either true or false, or as
corresponding or failing to correspond to irreducible modal facts.
2 serious essentialism: its claims and merits

Fortunately, this is a false dilemma. There is another option and this is
essentialism, suitably understood. I say ‘suitably understood’ because, of
course, many possible-worlds theorists will happily describe themselves as
essentialists and propose and defend what they call essentialist claims,
formulated in terms of the language of possible worlds. They will say,
for instance, that an essential property of an object is one that that object
possesses in every possible world in which it exists, or, alternatively, that is
possessed by the ‘counterpart(s)’ of that object in every possible world in
which that object has a ‘counterpart’. And they will claim that some, but
not all, of an object’s actual properties are essential to it in this sense. But
this is not serious essentialism, as I understand the latter. It is at best ersatz
essentialism. So what is serious essentialism? To begin to answer this, we
need to ask what essences are. But this question is potentially misleading,
for it invites the reply that essences are entities of some special sort. As we
shall see, however, it is incoherent to suppose that essences are entities.
According to serious essentialism, as I understand it, all entities have
essences, but their essences are not further entities related to them in some
special way. So, what do we, or rather what should we, mean by the
‘essence’ of a thing – where by ‘thing’, in this context, I mean any sort of
entity whatever? As I remarked in Chapter 6, we can, I think, do no better
than to begin with John Locke’s words on the matter, which go right to its
heart. To recall: essence, Locke said, in the ‘proper original signification’ of
the word, is ‘the very being of any thing, whereby it is, what it is’.7 In short,
7
the essence of something, X, is what X is, or what it is to be X. In another
locution, X ’s essence is the very identity of X. But here it is appropriate and
important to draw a distinction between general and individual essences.
Any individual thing, X, must be a thing of some general kind – because,
not least, it must belong to some ontological category. Remember that by
‘thing’ here I just mean ‘entity’. So, for example, X might be a material
object, or a person, or a property, or a set, or a number, or a proposition,
or . . . the list goes on, in a manner that depends on what ontological
categories one thinks should be included in it.8 If X is something of kind K,
then, X ’s general essence is what it is to be a K, while X ’s individual essence
is what it is to be the individual of kind K that X is, as opposed to any other
individual of that kind. So suppose, for example, that X is a particular cat.
Then X ’s general essence is what it is to be a cat and X ’s individual essence
is what it is to be this particular cat, X.
But why suppose that things must have ‘essences’ in this sense? First of
all, because otherwise it surely makes no sense to suppose that we can really
talk or think comprehendingly about things at all. If we do not know what
a thing is, how can we talk or think comprehendingly about it? How, for
instance, can I talk or think comprehendingly about Tom, a particular cat,
if I don’t know what cats are and which cat, in particular, Tom is? Of
course, I am not saying that I must know everything about Tom in order to
be able to talk or think comprehendingly about him. But I must surely
know enough to distinguish the kind of thing that Tom is from other
kinds of things, and enough to distinguish Tom in particular from other
individual things of Tom’s kind. Otherwise, it would seem that my talk
and thought – assuming it to be comprehending, rather than a mere
parroting of other people’s opinions – cannot really fasten upon Tom, as
opposed to something else.9 However, denying the reality of essences
doesn’t only create an epistemological or semantic problem, it also creates
an ontological problem. Unless Tom has an ‘identity’ – whether or not
anyone is acquainted with it – there is nothing to make Tom the particular
thing that he is, as opposed to any other thing. Anti-essentialism commits
8
For my own account of ontological categories, see my The Four-Category Ontology: A Metaphysical
Foundation for Natural Science (Oxford: Clarendon Press, 2006).
9
I say much more about this in Chapter 2, to which I refer the reader for a fuller and in some ways
more nuanced account. It may perhaps be allowed that, by parroting another person’s opinion,
I vicariously manage to ‘talk’ and ‘think’ about what they talk and think about, but there must
evidently be a terminus to a regress of this kind if there is to be comprehending talk or thought about
anything at all. Someone, somewhere, must know what they are thinking about if anyone, anywhere,
is to do so.
146 Modality
us to anti-realism, and indeed to an anti-realism so global that it is surely
incoherent. It will not do, for instance, to try to restrict one’s
anti-essentialism to ‘the external world’, somehow privileging us and our
language and thought. How could it be that there is a fact of the matter as
to our identities, and the identities of our words and thoughts, but not as to
the identities of the mind-independent entities that we try to represent in
language and thought? On the other hand, how could there not be any fact
of the matter as to our identities and the identities of our words and
thoughts? Everything is, in Joseph Butler’s memorable phrase, what it is
and not another thing. That has sounded to many people like a mere truism
without significant content, as though it were just an affirmation of the
reflexivity of the identity relation. But, in fact, Butler’s dictum does not
merely concern the identity relation but also identity in the sense of essence.
It implies that there is a fact of the matter as to what any particular thing is,
its ‘very being’, in Locke’s phrase. Its very being – its identity – is what
makes it the thing that it is and thereby distinct from any other thing.
Essences are apt to seem very elusive and mysterious, especially if talked
about in a highly generalized fashion, as I have been doing so far. Really,
I suggest, they are quite familiar to us. First, we need to appreciate that in
very many cases a thing’s essence involves other things, to which it stands in
relations of essential dependence. Consider the following thing, for
instance: the set of planets whose orbits lie within that of Jupiter. What
kind of thing is that? Well, of course, it is a set, and as such an abstract
entity that depends essentially for its existence and identity on the things
that are its members – namely, Mercury, Venus, Earth, and Mars. Part of
what it is to be a set is to be something that depends in these ways upon
certain other things – the things that are its members. Someone who did
not grasp that fact would not understand what a set is. Furthermore,
someone who did not know which things are this set’s members, or at least
what determined which things are its members, would not know which
particular set this set is. So, someone who knew that its members are the
planets just mentioned would know which set it is, as would someone who
knew what it is to be a planet whose orbit lies within that of Jupiter. This is
a simple example, but it serves to illustrate a general point. In many cases,
we know what a thing is – both what kind of thing it is and which
particular thing of that kind it is – only by knowing that it is related in
certain ways to other things. In such cases, the thing in question depends
essentially on these other things for its existence and/or its identity. To say
that X depends essentially on Y for its existence and identity is just to say
that it is part of the essence of X that X exists only if Y exists and part of the
essence of X that X stands in some unique relation to Y.10 Knowing a thing’s
essence, in many cases, is accordingly simply a matter of understanding the
relations of essential dependence in which it stands to other things whose
essences we in turn know.
I said earlier that it is wrong to think of essences as themselves being
entities of any kind to which the things having them stand in some special
kind of relation. Locke himself unfortunately made this mistake, holding
as he did that the ‘real essence’ of a material substance just is its ‘particular
internal constitution’ – or, as we would now describe it, its atomic or
molecular structure.11 This is a mistake that has been perpetuated in the
modern doctrine, made popular by the work of Saul Kripke and Hilary
Putnam, that the essence of water consists in its molecular make-up, H2O,
and that the essence of a living organism consists in its DNA – the
suggestion being that we discover these ‘essences’ simply by careful scien-
tific investigation of the things in question.12 Now, as we saw earlier, it
may well be part of the essence of a thing that it stands in a certain relation
to some other thing, or kind of things. But the essence itself – the very being
of the thing, whereby it is, what it is – is not and could not be some further
entity. So, for instance, it might perhaps be acceptable to say that it is part
of the essence of water that it is composed of H2O molecules (an issue that
I shall return to shortly). But the essence of water could not simply be
H2O – molecules of that very kind. For one thing, if the essence of an entity
were just some further entity, then it in turn would have to have an essence
of its own and we would be faced with an infinite regress that, at worst,
would be vicious and, at best, would make all knowledge of essence
impossible for finite minds like ours. To know something’s essence is
not to be acquainted with some further thing of a special kind, but simply
to understand what exactly that thing is. This, indeed, is why knowledge of
essence is possible, for it is a product simply of understanding, not of some
mysterious kind of quasi-perceptual acquaintance with esoteric entities of
any sort. And, on pain of incoherence, we cannot deny that we understand
what at least some things are, and thereby know their essences.
10
See further my The Possibility of Metaphysics: Substance, Identity, and Time (Oxford: Clarendon
Press, 1998), Chapter 6.
11
Thus at one point Locke remarks: ‘we come to have the Ideas of particular sorts of Substances, by
collecting such Combinations of simple Ideas, as are by Experience . . . taken notice of to exist
together, and are therefore supposed to flow from the particular internal Constitution, or unknown
Essence of that Substance’ (Essay, ii, xxiii, 3).
12
See, especially, Saul Kripke, Naming and Necessity (Oxford: Blackwell, 1980); and Hilary Putnam,
‘The Meaning of “Meaning”’, in his Mind, Language and Reality: Philosophical Papers Volume 2
(Cambridge University Press, 1975).
148 Modality
Here it may be objected that it is inconsistent of me to deny that
essences are entities and yet go on, as I apparently do, to refer to and even
quantify over essences. Someone who voices this objection probably has in
mind, once more, Quine’s infamous criterion of ontological commitment,
encapsulated in his slogan ‘to be is to be the value of a variable’. I reply, in
the first place, that I could probably say all that I want to about my version
of essentialism while avoiding all locutions involving the appearance of
reference to and quantification over essences, by paraphrasing them in
terms of locutions involving only sentential operators of the form ‘it is part
of the essence of X that’ (where ‘the essence of X’ is not taken to make an
independent contribution to the meaning of the operator, which might be
represented symbolically by, say, ‘EX’ in a sentential formula of the form
‘EX(p)’). The latter is a kind of locution that I certainly do want to use and
find very useful. However, I think that effort spent on working out such
paraphrases in all cases would be effort wasted. If a paraphrase is logically
equivalent to what it is supposed to paraphrase – as it had better be, if it is
to be any good – then it surely carries the same ‘ontological commitments’
as whatever it is supposed to paraphrase, so that constructing paraphrases
cannot be a way of relieving ourselves of ontological commitments. We
cannot discover those commitments simply by examining the syntax and
semantics of our language, for syntax and semantics are very uncertain
guides to ontology. In other words, I see no reason to place any confidence
in Quine’s notorious criterion.
Another crucial point about essence is this: in general, essence precedes
existence. That is to say, we can in general know the essence of
something X prior to knowing whether or not X exists. Otherwise, we
could never find out that something exists. For how could we find out
that something, X, exists before knowing what X is – before knowing,
that is, what it is whose existence we have supposedly discovered?
Consequently, we know the essences of many things which, as it turns
out, do not exist. For we know what these things would be, if they
existed, and we retain this knowledge when we discover that, in fact,
they do not exist. Conceivably, there are exceptions. Perhaps it really is
true in the case of God, for instance, that essence does not precede
existence. But this could not quite generally be the case. However,
saying this is perfectly consistent with acknowledging that, sometimes,
we may only come to know the essence of something after we have
discovered the existence of certain other kinds of things. This is what
goes on in many fields of theoretical science. Scientists trying to
discover the transuranic elements knew before they found them what
it was that they were trying to find, but only because they knew that
what they were trying to find were elements whose atomic nuclei were
composed of protons and neutrons in certain hitherto undiscovered
combinations. They could hardly have known what they were trying
to find, however, prior to the discovery of the existence of protons and
neutrons – for only after these sub-atomic particles were discovered and
investigated did the structure of atomic nuclei become sufficiently well
understood for scientists to be able to anticipate which combinations of
nucleons would give rise to reasonably stable nuclei.
Here it may be objected that Kripke and Putnam have taught us that
the essences of many familiar natural kinds – such as the kind cat and the
kind water – have been revealed to us only a posteriori and consequently
that in cases such as these, at least, it cannot be true to say that ‘essence
precedes existence’, whatever may be said in the case of the transuranic
elements. The presupposition here, of course, is that Kripke and Putnam
are correct in identifying the essence of water, for example, with its
molecular make-up, H2O. Now, I have already explained why I think
that such identifications are mistaken, to the extent that they involve the
illicit reification of essences. But it may still be urged against me that even
if, more cautiously, we say only that it is part of the essence of water that
it is composed of H2O molecules, it still follows that the essence of water
has only been revealed to us – or, at least, has only been fully revealed to
us – a posteriori. In point of fact, however, the Kripke–Putnam doctrine
is even more obscure and questionable than I have so far represented it as
being. Very often, it is characterized in terms of the supposed modal and
epistemic status of identity-statements involving natural kind terms, such
as ‘Water is H2O’, which are said to express truths that are at once
necessary and a posteriori. In such a statement, however, the term ‘H2O’
is not functioning in exactly the same way as it does in the expression
‘H2O molecule’. The latter expression, it seems clear, means ‘molecule
composed of two hydrogen atoms and one oxygen atom’. But in ‘Water
is H2O’, understood as an identity-statement concerning kinds, we must
take ‘H2O’ either to be elliptical for the definite description ‘the stuff
composed of H2O molecules’ or else simply as being a proper name, in
which case we cannot read into it any significant semantic structure. On
the latter interpretation, ‘Water is H2O’ is exactly analogous to ‘Hesperus
is Phosphorus’ and its necessary truth reveals nothing of substance to us
concerning the composition of water. If we are inclined to think other-
wise, this is because we slide illicitly from construing ‘H2O’ as a proper
name to construing it as elliptical for the definite description ‘the stuff
composed of H2O molecules’. Now, when ‘Water is H2O’ is understood
on the model of ‘Hesperus is Phosphorus’, its necessary a posteriori truth
150 Modality
may in principle be established in a like manner – namely by appeal to
the familiar logical proof of the necessity of identity,13 together with the a
posteriori discovery of the co-reference of the proper names involved – but not
so when it is construed as meaning ‘Water is the stuff composed of
H2O molecules’, for the latter involves a definite description. Thus far, then,
we have been given no reason to suppose that ‘Water is H2O’ expresses an a
posteriori necessary truth that reveals to us something concerning the essence
of water. The appearance that we have been given such a reason is the result of
mere sleight of hand. It might be thought that ‘Water is the stuff composed of
H2O molecules’ follows unproblematically from the supposed empirical truth
‘Water is H2O’ (construed as an identity-statement involving two proper
names) and the seemingly trivial, because analytic, truth ‘H2O is the stuff
composed of H2O molecules’. But the latter, when the first occurrence of
‘H2O’ in it is interpreted as a proper name, is no more trivial than ‘Water is the
stuff composed of H2O molecules’ – and this is how it must be interpreted for
the inference to go through.
There is another important consideration that we should bear in mind
when reflecting on the frequently invoked analogy between ‘Water is H2O’
and ‘Hesperus is Phosphorus’. It is all very well to point out that the discovery
that Hesperus is Phosphorus was an empirical one. But it was not purely
empirical, for the following reason. The identity was established because
astronomers discovered that Hesperus and Phosphorus coincide in their orbits:
wherever Hesperus is located at any given time, there too is Phosphorus
located. However, spatiotemporal coincidence only implies identity for things
of appropriate kinds. It is only because Hesperus and Phosphorus are taken to
be planets and thereby material objects of the same kind that their spatiotem-
poral coincidence can be taken to imply their identity. (I shall return to this
sort of issue in a little more detail shortly.) But the principle that distinct
material objects of the same kind cannot coincide spatiotemporally is not an
empirical one: it is an a priori one implied by what it is to be a material object
of any kind – in other words, it is a truth grounded in essence. It is only because
we know that it is part of the essence of a planet not to coincide spatiotempo-
rally with another planet that we can infer the identity of Hesperus with
Phosphorus from the fact that they coincide in their orbits. Thus one must
already know what a planet is – know its essence – in order to be able to
establish by a posteriori means that one planet is identical with another. By the
same token, then, one must already know what a kind of stuff is – know its
13
Of course, I challenge this proof in Chapter 7, but let us set aside those doubts now, for the sake of
argument.
essence – in order to be able to establish by a posteriori means that one kind
of stuff is identical with another. It can hardly be the case, then, that we can
discover the essence of a kind of stuff simply by establishing a posteriori the
truth of an identity-statement concerning kinds of stuff – any more than we
can be supposed to have discovered the essence of a particular planet by
establishing a posteriori the truth of an identity-statement concerning that
planet. So, even granting that ‘Water is H2O’ is a true identity-statement that
is both necessarily true and known a posteriori, it does not at all follow that it
can be taken to reveal to us the essence of the kind of stuff that we call ‘water’.
Be all this as it may, however, we still have to address the question
whether, in fact, we ought to say that it is part of the essence of water that
it is composed of H2O molecules. So far, we have seen only that the
Kripke–Putnam semantics for natural kind terms have given us no reason
to suppose that we ought to. I am inclined to answer as follows. If we are
using the term ‘water’ to talk about a certain chemical compound whose
nature is understood by theoretical chemists, then indeed we should say
that it is part of the essence of this compound that it consists of
H2O molecules. But, at the same time, it should be acknowledged that
the existence of this compound is a relatively recent discovery, which could
not have been made before the nature of hydrogen and oxygen atoms and
their ability to form molecules were understood. Consequently, when we
use the term ‘water’ in everyday conversation and when our forebears used
it before the advent of modern chemistry, we are and they were not using it
to talk about a chemical compound whose nature is now understood by
theoretical chemists. We are and they were using it to talk about a certain
kind of liquid, distinguishable from other kinds of liquid by certain easily
detectable features, such as its transparency, colourlessness, and tastelessness.
We are right, I assume, in thinking that a liquid of this kind actually exists,
but not that it is part of its essence that it is composed of H2O molecules. At
the same time, however, we should certainly acknowledge that empirical
scientific inquiry reveals that, indeed, the chemical compound H2O is very
largely what bodies of this liquid are made up of. In fact, the natural laws
governing this and other chemical compounds make it overwhelmingly
unlikely that this kind of liquid could have a different chemical composition
in different parts of our universe. But the ‘could’ here is expressive of mere
physical or natural possibility, not metaphysical possibility.14 Only an illicit
14
For extended discussion of the need to distinguish between these two species of possibility, see my
The Four-Category Ontology, Chapter 9 and Chapter 10.
152 Modality
conflation of these two species of possibility could reinstate the claim that
water is essentially composed of H2O molecules.15
So far, I have urged that the following two principles must be endorsed
by the serious essentialist: that essences are not entities and that, in general,
essence precedes existence. But by far the most important principle to
recognize concerning essences, for the purposes of the present chapter, is
that essences are the ground of all metaphysical necessity and possibility.16 One
reason, thus, why it can be the case that X is necessarily F is that it is part of
the essence of X that X is F. For example, any material object is necessarily
spatially extended because it is part of the essence of any material object
that it is spatially extended – in other words, part of what it is to be a
material object is to be something that is spatially extended. But this is not
the only possible reason why something may be necessarily F. X may be
necessarily F on account of the essence of something else to which X is
suitably related. For example, Socrates is necessarily the subject of the
following event: the death of Socrates. This is because it is part of the
essence of that event that Socrates is its subject, even though it is not part
of Socrates’s essence that he is the subject of that event. It is not on
account of what Socrates is that he is necessarily the subject of that event
but, rather, on account of what that event is. This is not to say that Socrates
could not have died a different death, only that no one but Socrates could
have died the death that he in fact died. And what goes for necessity goes
likewise, mutatis mutandis, for possibility. I venture to affirm that all facts
about what is necessary or possible, in the metaphysical sense, are
grounded in facts concerning the essences of things – not only of existing
things, but also of non-existing things. But, I repeat, facts concerning the
essences of things are not facts concerning entities of a special kind, they are
just facts concerning what things are – their very beings or identities. And
these are facts that we can therefore grasp simply in virtue of understand-
ing what things are, which we must in at least some cases be able to do, on
pain of being incapable of thought altogether. Consequently, all know-
ledge of metaphysical necessity and possibility is ultimately a product of
the understanding, not of any sort of quasi-perceptual acquaintance, much
less of ordinary empirical observation.
15
I say much more about the case of water in my ‘Locke on Real Essence and Water as a Natural Kind:
A Qualified Defence’, Proceedings of the Aristotelian Society, Supplementary Volume 85 (2011),
pp. 1–19. There I discuss, inter alia, the notorious ‘Twin Earth’ thought experiments.
16
Compare Kit Fine, ‘Essence and Modality’, in James E. Tomberlin (ed.), Philosophical Perspectives,
8: Logic and Language (Atascadero, CA: Ridgeview, 1994).
How, for example, do we know that two distinct things, such as a
bronze statue and the lump of bronze composing it at any given time,
can – unlike two planets – exist in the same place at the same time?
Certainly not by looking very hard at what there is in that place at that
time. Just by looking, we shall not see that two distinct things occupy that
place. We know this, rather, because we know what a bronze statue is and
what a lump of bronze is. We thereby know that these are different things
and that a thing of the first sort must, at any given time, be composed by a
thing of the second sort, since it is part of the essence of a bronze statue to
be composed of bronze. We know that they are different things because, in
knowing what they are, we know their identity conditions, and thereby
know that one of them can persist through changes through which the
other cannot persist – that, for instance, a lump of bronze can persist
through a radical change in its shape whereas a bronze statue cannot.
These facts about their identity conditions are not matters that we can
discover purely empirically, by examining bronze statues and lumps of
bronze very closely, as we might in order to discover whether, say, they
conduct electricity or dissolve in sulphuric acid. Rather, they are facts
about them that we must grasp prior to being able to embark upon any
such empirical inquiry concerning them, for we can only inquire empiric-
ally into something’s properties if we already know what it is that we are
examining.
3 the errors of conceptualism

At this point I need to counter a rival view of essence that is attractive to
many philosophers but is, I think, ultimately incoherent. I call this view
conceptualism. It is the view that what I have been calling facts about
essences are really, in the end, just facts about certain of our concepts – for
example, our concept of a bronze statue and our concept of a lump of bronze.
This would reduce all modal truths to conceptual truths or, if the term is
preferred, analytic truths. Now, I have no objection to the notion of
conceptual truth as such. Perhaps, as is often alleged, ‘Bachelors are
unmarried’ indeed expresses such a truth. Let us concede that it is true
in virtue of our concept of a bachelor, or in virtue of what we take the
word ‘bachelor’ to mean. But notice that ‘Bachelors are unmarried’ has a
quite different modal status from an essential truth such as ‘Statues are
composed of matter’. In calling the former a ‘necessary’ truth, we cannot
mean to imply that bachelors cannot marry, only that they cannot marry
and go on rightly being called ‘bachelors’. The impossibility in question is
154 Modality
only one concerning the proper application of a word. But in calling
‘Statues are composed of matter’ a necessary truth, we certainly can’t be
taken to mean merely that statues cannot fail to be composed of matter
and go on rightly being called ‘statues’ – as though the very same thing which,
when composed of matter, was properly called a ‘statue’ might exist as
something immaterial. No, we must be taken to mean that statues cannot
fail to be composed of matter full stop. Statues are things such that, if they
exist at all, they must be composed of matter. That is because it is part of
the essence of a statue to be so composed. In contrast, it is not part of the
essence of any bachelor to be unmarried, for a bachelor is just an adult
male human being who happens to be unmarried, and any such human
being undoubtedly can marry. So, ‘Statues are composed of matter’ is
certainly not a mere conceptual truth, and the same goes for other truths
that are genuinely essential truths – truths concerning the essences of
things. They have, in general, nothing to do with our concepts or our
words, but with the natures of the things in question. Of course, since
concepts and words are themselves things of certain sorts, there can be
truths concerning their essences. Indeed, what we could say about ‘Bach-
elors are unmarried’ is that it is, or is grounded in, a truth concerning the
essence of the concept bachelor, or of the word ‘bachelor’. We could say,
thus, that it is part of the essence of the concept bachelor that only
unmarried males fall under it, and part of the essence of the word
‘bachelor’ that it applies only to unmarried males.
But I said that conceptualism is ultimately incoherent. Indeed, I think it
is. For one thing, as we have just seen, the proper thing to say about
‘conceptual’ truths is, very plausibly, that they are grounded in the essences of
concepts. That being so, the conceptualist cannot maintain, as he or she
does, that all putative facts about essence are really just facts concerning
concepts. For this is to imply that putative facts about the essences of
concepts are really just facts concerning concepts of concepts – and we have
set out on a vicious infinite regress. The conceptualist will object, no
doubt, that this complaint is question-begging. However, even setting
the complaint aside, we can see that conceptualism is untenable. For the
conceptualist is at least committed to affirming that concepts – or, in
another version, words – exist and indeed that concept-users do, to wit,
ourselves. These, at least, are things that the conceptualist must acknow-
ledge to have identities, independently of how we conceive of them, on
pain of incoherence in his or her position. The conceptualist must at least
purport to understand what a concept or a word is, and indeed what he or
she is, and thus grasp the essences of at least some things: and if of these
things, why not of other kinds of things? Once knowledge of essences
is conceded, the game is up for the conceptualist. And it must be con-
ceded, even by the conceptualist, on pain of denying that he or she knows
what anything is, including the very concepts that lie at the heart of his or
her account. For recall, all that I mean by the ‘essence’ of something is
what it is.
So, why is anyone ever tempted by conceptualism? I am afraid that it is
the legacy of scepticism, particularly scepticism concerning ‘the external
world’. The sceptic feels at home with himself and with his words and
concepts, but expresses doubt that we can ever really know whether those
words and concepts properly or adequately characterize things in the
external world. He thinks that we can know nothing about how or what
those things are ‘in themselves’, or indeed even whether they are many or
one. According to the sceptic, all that we can really know is how we
conceive of the world, or describe it in language, not how it is. But by what
special dispensation does the sceptic exclude our concepts and our words
from the scope of his doubt? For are they not, too, things that exist? There
is, in truth, no intelligible division that can be drawn between the external
world, on the one hand, and ourselves and our concepts or our language on
the other. Here it may be protested: but how, then, can we advance to
knowledge of what and how things are ‘in themselves’, even granted that
the sceptic is mistaken in claiming a special dispensation with regard to the
epistemic status of our concepts and our words? However, the fundamen-
tal mistake is to suppose, with the sceptic, that such an ‘advance’ would
have to proceed from a basis in our knowledge of our concepts and words –
that is, from a knowledge of how we conceive of and describe the world to a
knowledge of that world ‘as it is in itself’, independently of our conceptual
schemes and languages. This ‘inside-out’ account of how knowledge of
mind-independent reality is to be acquired already makes such knowledge
impossible and must therefore be rejected as incoherent.
But what alternative is there? Again, knowledge of essence comes to the
rescue. Because, in general, essence precedes existence, we can at least
sometimes know what it is to be a K – for example, what it is to be a
material object of a certain kind – and thereby know, at least in part, what
is or is not possible with regard to Ks, in advance of knowing whether, or
even having good reason to believe that, any such thing as a K actually
exists. Knowing already, however, what it is whose existence is in question
and that its existence is at least possible, we can intelligibly and justifiably
appeal to empirical evidence to confirm or cast doubt upon existence
claims concerning such things. By ‘empirical evidence’ here, be it noted,
156 Modality
I emphatically do not mean evidence constituted purely by the contents of
our own perceptual states at any given time, as though all that we have to
go on is how the world in our vicinity looks or otherwise appears to be.
That, certainly, is not the conception of ‘empirical evidence’ that is
operative in scientific practice, which appeals rather to the results of
controlled experiments and observations, all of which are reported in terms
of properties and relations of mind-independent objects, such as scientific
instruments and laboratory specimens. The growth of objective knowledge
consists, then, in a constant interplay between an a priori element –
knowledge of essence – and an a posteriori element, the empirical testing
of existential hypotheses whose possibility has already been anticipated a
priori. This process does not have a foundational ‘starting point’ and it is
constantly subject to critical reappraisal, both with regard to its a priori
ingredients and with regard to its empirical contributions. Here we do not
have a hopeless ‘inside-out’ account of objective knowledge, since our own
subjective states as objective inquirers – our perceptions and our concep-
tions – are accorded no special role in the genesis of such knowledge.
Those subjective states are merely some amongst the many possible objects
of knowledge, rather than objects of a special kind of knowledge which
supposedly grounds the knowledge of all other things. But, to repeat, it is
crucial to this account that knowledge of essences is not itself knowledge of
objects or entities of any kind, nor grounded in any such knowledge – such
as knowledge of our own concepts.
4 the redundancy of possible worlds

It is now high time that I returned to my original theme: the language of
possible worlds and its bearing upon the nature and ground of metaphys-
ical modality. I have already made it clear that, in my opinion, all modal
facts concerning what is metaphysically necessary or possible are ultimately
grounded in the essences of things – and hence not in facts concerning
entities of any sort, since essences are not entities. But – it may perhaps be
urged – this in itself does not necessarily prevent the language of possible
worlds from casting at least some light on the nature and ground of
metaphysical modality. Well, let us see. First, let us consider non-
fictionalist construals of the language of possible worlds, according to
which possible-worlds variables in that language range over a domain of
existing entities of some kind, such as Lewisian parallel universes or max-
imal consistent sets of propositions. According to possible-worlds theorists
adopting this approach, any modal statement in which the modal
terminology involved is expressive of metaphysical modality is semantically
equivalent to one quantifying over existing entities of the favoured kind –
as it might be, parallel universes or maximal consistent sets of propositions.
Moreover, according to such an approach, the truth or falsehood of the
modal statement in question is grounded in facts concerning those entities.
For example, the truth or falsehood of the statement ‘Possibly, there are
talking donkeys’ is, supposedly, grounded in facts concerning the inhabit-
ants of certain parallel universes or facts concerning the membership of
certain maximal consistent sets of propositions. But, I suggest, it should
strike one as being obviously problematic to suppose that – where the
metaphysical modalities are concerned – modal facts are grounded in facts
concerning existing entities of any kind. The salient point, once again, is
that essence precedes existence. An existing entity must at the very least be a
possible entity – that is to say, something whose essence does not preclude
its existence. And what is true of an entity will likewise depend at least in
part on what it is – its essence. It can only be the case, for example, that
some parallel universe does in fact contain amongst its inhabitants such a
thing as a talking donkey if there could be such things as parallel universes
and such things as talking donkeys inhabiting them. The very facts that are
being proposed as the grounds of modal truths already presuppose modal
truths, simply because they are, supposedly, facts concerning existing
entities of certain putative kinds.
The upshot is this. Suppose we grant that there could be such things as
Lewisian parallel universes or maximal consistent sets of propositions
because, understanding what these entities are – knowing their essences –
we know that their essences do not preclude their existence. Let us go
further and suppose that such things do in fact exist. Even so, non-modal
facts concerning such entities could not constitute the ground of all modal
truths. Why not? Because, first and foremost, such facts could not consti-
tute the ground of modal truths concerning those entities themselves. If these
entities exist, then there must indeed be modal truths concerning them,
since there are modal truths concerning any existing entity. So, for
example, if parallel universes exist, it must either be true, concerning them,
that infinitely many of them could exist, or else be true, concerning them,
that only finitely many of them could exist. Similarly, it must either be
true, concerning them, that two or more of them could be qualitatively
indiscernible, or else be true, concerning them, that any two of them must
be qualitatively distinct. And so on. Quite evidently, however, the con-
cretist – as we have elected to call him – cannot contend that, for example,
‘Possibly, there are infinitely many possible worlds’ is true or false for the
158 Modality
same sort of reason that he contends that ‘Possibly, there are infinitely
many electrons’ is true or false. For the latter is true, he maintains, just in
case there is a possible world – a parallel universe – in which there are
infinitely many electrons (or electron ‘counterparts’). But he cannot main-
tain that the former is true just in case there is a possible world in which
there are infinitely many possible worlds. For, knowing what a ‘possible
world’ is supposed to be according to the concretist – to wit, a ‘parallel
universe’, akin to our cosmos – we know already that it is not the sort of
thing that could have another such thing amongst its inhabitants, let alone
infinitely many other such things. The implication is that, far from its
being the case that non-modal facts concerning possible worlds – whatever
the latter are conceived to be – are the ground of all modal facts, there
must be modal facts which are not grounded in the existence of entities of
any kind, including possible worlds. And if this must be so for some modal
facts, why not for all, as serious essentialism contends?
In reply, some concretists will no doubt urge, as Lewis himself did, that
there are some modal facts which cannot even be expressed without using
the language of possible worlds, because its expressive power is demon-
strably greater than that of a language which utilizes only modal operators,
such as ‘possibly’ and ‘necessarily’. However, this cannot be represented as
a non-question-begging reason to endorse the language of possible worlds
and its ontological implications, because opponents are certainly entitled
to challenge the assumption that the additional expressive power of such a
language serves to convey genuine modal truths which are not otherwise
expressible. Since the alleged additional modal truths are, on the
concretist’s own insistence, only expressible in the language of possible
worlds, the credentials of that language must be established before the
alleged truths can be accepted as genuine, so they cannot be advanced as
evidence in favour of it.
Now the abstractionist may protest at this point that he, at least – unlike
the concretist – never intended to suggest that modal truths could be
reduced, without remainder, to non-modal truths concerning possible
worlds and that this exempts him from the foregoing strictures. The
abstractionist openly acknowledges, for example, that he appeals to an
unreduced notion of consistency in explaining what he takes a ‘possible
world’ to be – to wit, a maximal consistent set of propositions, or some-
thing like that. This might be an acceptable response if the only modal
notion being relied upon by the abstractionist were that of consistency –
the notion, that is, of the possible joint truth of two or more propositions.
But my complaint does not focus on this well-known feature of
abstractionism and its consequent repudiation of any aspiration to offer a
reductive account of modality. Rather, my complaint focuses on the fact
that abstractionism, just like concretism, appeals to existing entities of
certain putative kinds in presenting its account of the semantics of modal
statements. In this case, the entities in question are abstract objects such as
propositions and sets thereof. But propositions and sets, if they exist, are just
further entities, concerning which various modal truths must hold. For
example, it must either be true, concerning sets, that they could have
contained different members, or else it must be true, concerning sets, that
they could not have contained different members. Suppose it is true.
Suppose, that is, that the following modal statement is true, where S is
any given set whose actual members are certain objects: ‘Possibly, S has
members that are different from its actual members’. What is this supposed
to mean, according to the abstractionist? Clearly, something like this:
‘Some maximal consistent set of propositions contains the proposition
that S has members that are different from its actual members’. But S
was supposed to be any set we like. So what happens if we try to let S be the
maximal consistent set of propositions whose actual members are all and
only the propositions that are actually true – in other words, if we try to let
S be the maximal consistent set of propositions that the abstractionist
identifies as the actual world, Wɑ? In that case, the abstractionist translates
the putative modal truth ‘Possibly, Wɑ has members that are different from
its actual members’ as meaning ‘Some maximal consistent set of propos-
itions contains the proposition that Wɑ has members that are different
from its actual members’ – or, in the language of possible worlds, ‘In some
possible world, the actual world is different from how it actually is’. But it
is very hard to see how the abstractionist could allow this to be true. The
implication is that his semantics for modal statements compels him to
deny, after all, that any set whatever could have contained different
members. Now, I am not quarrelling with that verdict as such, since
I consider that it is part of the essence of any set that it has the members
that it does – that their identities determine its identity. However, it is
plainly not a verdict that should be forced upon one merely by the
machinery that one invokes to articulate the semantics of modal state-
ments: rather, it is one that should emerge from an adequate understand-
ing of what sets are – an understanding that carries modal implications and
one which the abstractionist himself must possess prior to constructing his
preferred machinery for modal semantics. That abstractionism runs into
this and similar problems is just a symptom of the fact that abstractionism,
like other possible-worlds accounts of metaphysical modality, has simply
160 Modality
mislocated the meaning and grounds of modal truths, by trying to find
them in facts concerning a special class of entities of an esoteric kind – in
this case, maximal consistent sets of propositions.
What, finally, of fictionalism? Well, that approach can be dismissed
without more ado, I think, because in seeking to reap the advantages of
theft over honest toil, it relies on the toil in question at least being effective.
If the toil was wasted effort, no advantages can be got from it. But we have
seen that both concretism and abstractionism fail on their own terms,
whence there is no profit to be had in a theory which rests on a pretence
that either of them is true. This would be like stealing the harvest of a
farmer whose crops had failed. I conclude that the language of possible
worlds, whether or not it is interpreted in an ontologically serious manner
and whatever possible worlds are taken to be, can throw no real light at all
on the nature and ground of metaphysical modality. If possible worlds,
whatever they are taken to be, exist at all, that is a fact which may hold
some interest for the ontologist – who is, after all, concerned to provide as
full and accurate an inventory of what there is as is humanly possible – but
it is not one that can usefully be recruited for the purposes of modal
metaphysics. For that, I suggest, we have no viable option but to turn to
serious essentialism.
part iv
Conditionality
chapter 9
The truth about counterfactuals
Do counterfactual conditionals have truth-values? If so, what are the truth-

conditions of counterfactuals? Do such conditionals constitute a logically
distinct sub-class of conditionals in general and, if so, how do they differ
logically from other conditionals? That is to say, how, if at all, does the logic
of counterfactuals differ from the logic of other conditionals, in respect of the
patterns of valid inference that counterfactuals sustain? Is it correct to think of
counterfactuals as belonging to a wider class of ‘subjunctive’ conditionals, whose
truth-conditions and logic differ fundamentally from those of so-called ‘indica-
tive’ conditionals – or is this grammatical distinction of mood either inapplic-
able to conditionals or, at best, something of merely pragmatic significance? Is
the conditional connective ‘if . . ., then –’ really ambiguous? All of these
questions need to be answered if we are to arrive at the truth about
counterfactuals.
Let me begin by laying my cards on the table. First, I do not think that ‘if’ is
ambiguous. Nor do I think that the subjunctive/indicative distinction, as
applied to conditionals, is really semantically or logically very important.
I believe that there is a single, unified logic of conditionals. And I believe that
conditionals do, in general, have truth-values (although there may be truth-
value gaps where conditionals are concerned). At the same time, I do believe
that we can distinguish between certain classes of conditionals on semantic
grounds, in a way which does not impugn the underlying univocality of ‘if’.
Also, I believe that many conditionals, and especially many counterfactuals,
are highly context-sensitive – in the sense that what proposition is expressed
by a counterfactual conditional sentence typically depends upon the context
of utterance. All of this I hope to explain and justify in due course.
1 subjunctive and indicative conditionals
Consider first the supposed subjunctive/indicative distinction, as applied
to conditionals. Counterfactuals are commonly said to be a sub-class of
subjunctive conditionals. Of course, this cannot strictly be correct, if we
163
164 Conditionality
define a counterfactual conditional simply as a conditional whose antece-
dent is false, or presumed to be false by the speaker. For an ‘indicative’
conditional might have either of these properties. Even so, it may be urged
that where an indicative conditional is known or believed to have a false
antecedent, there can be little or no point in asserting it, so that in practice
the only assertible counterfactual conditionals are subjunctive ones. (I shall
discuss later the apparent exception of ‘Dutchman’ conditionals, such as ‘If
that’s a Ming vase, then I’m a Dutchman’.) And then it may be pointed
out that, even so, there are other subjunctive conditionals which are
asserted without any presumption on the part of the speaker that the
antecedent is false – so that counterfactuals (or all assertible ones) form a
proper subset of subjunctive conditionals. Opponents (such as Vic Dud-
man) object that the grammatical distinction between subjunctive and
indicative mood is not in fact one that has genuine application in English –
however it may be with languages such as Latin and Italian – so that those
who classify counterfactuals as ‘subjunctive’ conditionals are labouring
under the illusion of a false grammatical theory – the ‘Latin prose theory’.1
I am not as hostile as this to the subjunctive/indicative distinction.
I think that it does have application in English, even if it is not marked
so prominently by distinct verbal forms in English as in some other
languages. But I do not believe that the distinction reflects any underlying
distinction of logic or truth-conditions.2 That such an underlying distinc-
tion is indeed involved is often argued for by appeal to the notorious
Oswald/Kennedy examples. ‘If Oswald did not kill Kennedy, then some-
one else did’ is indicative and plausibly true, but ‘If Oswald had not killed
Kennedy, then someone else would have’ is subjunctive and plausibly false.
Yet these two conditionals differ only in respect of mood, it is said, whence
it follows that their difference in mood reflects a difference in truth-
conditions.3 My answer to this is that it is easy to find another indicative
conditional which differs only in mood from the subjunctive conditional
in question, and yet which plausibly does not differ from it in truth-
conditions – namely, ‘If Oswald hasn’t killed Kennedy, then someone else
1
See V. H. Dudman, ‘Conditional Interpretations of “If”-Sentences’, Australian Journal of Linguistics
4 (1984), pp. 143–204; ‘Indicative and Subjunctive’, Analysis 48 (1988), pp. 113–22; and ‘On
Conditionals’, Journal of Philosophy 91 (1994), pp. 113–28. See also Jonathan Bennett, ‘Farewell to
the Phlogiston Theory of Conditionals’, Mind 97 (1988), pp. 509–27.
2
See my ‘Jackson on Classifying Conditionals’, Analysis 51 (1991), pp. 126–30, for criticisms of one
important attempt to justify such a logical distinction.
3
See Ernest Adams, ‘Subjunctive and Indicative Conditionals’, Foundations of Language 6 (1970),
pp. 89–94.
The truth about counterfactuals 165
will have’.4 This is what might be asserted, prior to confirmation of the
assassination’s occurrence, by a speaker who would subsequently be pre-
pared to assert the counterfactual, ‘If Oswald hadn’t killed Kennedy, then
someone else would have’. So what the examples show is that we can have
two different indicative conditionals, one of which coincides in its truth-
conditions with a certain counterfactual and the other of which does not.
Hence, although we must indeed acknowledge that there are two distinct
classes of conditionals involved, we should not conclude that this distinc-
tion is one which corresponds to any distinction in grammatical mood. My
own view, to be explained more fully later, is that the real difference turns
on a difference between the modal notions in terms of which the truth-
conditions of the two classes of conditionals are to be stated. As I shall put
it, we have a distinction here between ‘epistemic’ and ‘alethic’ conditionals,
corresponding to a distinction between ‘epistemic’ and ‘alethic’ modality.
But underlying that distinction we can still recognize a commonality of
logical form, which permits us to deny that there is any ambiguity in the
sense of ‘if’.
2 rival theories of the indicative conditional

If there is no distinction of logical importance marked by the difference
between subjunctive and indicative conditionals, several currently popular
theories of the indicative conditional are doomed to failure. One such
theory is the Grice–Jackson theory that indicative conditionals have the
truth-conditions of the so-called material conditional, and are further
distinguished only in respect of their assertibility-conditions.5 Since
Jackson concedes, of course, that counterfactuals cannot have these
truth-conditions (on pain of rendering all counterfactuals vacuously true),
it is crucial to his position that counterfactuals differ in their truth-
conditions from indicative conditionals. But I have just argued that this
is not, in general, so. One can find, for any given counterfactual, an
indicative conditional which plausibly has the same truth-conditions,
differing from it only in what it suggests with respect to the speaker’s
presumptions concerning the truth-value of the antecedent.6 But since
the counterfactual in question cannot be supposed to have the
4
See my ‘Indicative and Counterfactual Conditionals’, Analysis 39 (1979), pp. 139–41; and Christopher
Bryant, ‘Conditional Murderers’, Analysis 41 (1981), pp. 209–15.
5
See H. P. Grice, ‘Indicative Conditionals’, in his Studies in the Way of Words (Cambridge, MA:
Harvard University Press, 1989); and Frank Jackson, Conditionals (Oxford: Blackwell, 1987).
6
See also Bryant, ‘Conditional Murderers’.
166 Conditionality
truth-conditions of the material conditional, neither can the corresponding
indicative conditional – whence it follows that the Grice–Jackson theory
cannot be regarded as an adequate theory of the truth-conditions of
indicative conditionals quite generally. This still leaves open the possibility
that the Grice–Jackson theory adequately explicates a certain sub-class of
indicative conditionals – although only on pain of conceding that indica-
tive conditionals are systematically ambiguous in respect of the meaning of
the connective ‘if . . ., then –’ that features in them. It is preferable,
I believe, to follow my own route and hold on to the univocality of ‘if’,
while accounting for the distinctions between such indicative conditionals
in terms of differences between the types of modality to be invoked in their
analysis.
Another popular theory which must go by the board, if I am right to
dismiss the subjunctive/indicative distinction as being of no logical import-
ance, is the theory that indicative conditionals lack truth-values altogether
and merely have assertibility-conditions – such a conditional being held to
be highly assertible to the extent that there is deemed to be a high
conditional probability of the consequent relative to, or given, the antece-
dent.7 On this view, an indicative conditional does not express a propos-
ition, and consequently does not have a probability of truth (that is, of
being true). The argument for this is that if such a conditional did have a
probability of truth, its probability would have to be equal to the condi-
tional probability of its consequent given its antecedent, and this can be
shown to lead to absurdity (the proof of which is due to David Lewis).8
Now, once again, the people who support this theory may concede that
subjunctive conditionals are different, and do indeed have truth-values.9
But then, as before, if I am right to reject the subjunctive/indicative
distinction as being of no logical significance, it follows that the theory
in question will not do, at least for many indicative conditionals (and if not
for these, then why for any?).
7
See, for example, Dorothy Edgington, ‘Do Conditionals Have Truth-Conditions?’, Critica 18 (1986),
pp. 3–30, reprinted in Frank Jackson (ed.), Conditionals (Oxford University Press, 1991). I shall
discuss this view much more fully in Chapter 10 below, but some preliminary discussion of it here is
appropriate.
8
See David Lewis, ‘Probabilities of Conditionals and Conditional Probabilities’, Philosophical Review
85 (1976), pp. 297–315, reprinted in Jackson (ed.), Conditionals. Lewis himself does not accept the
conclusion of this argument, because he rejects the premise. Like Jackson, Lewis proposes that the
indicative conditional has the truth-conditions of the material conditional, and that while
the assertibility of the indicative conditional is measured by the conditional probability of its
consequent given its antecedent, the probability of the conditional differs from this.
9
Dorothy Edgington does not concede this: see her ‘Do Conditionals Have Truth-Conditions?’,
p. 178. But I suspect that few other theorists would be prepared to go to this extreme.
Even so, it is perhaps incumbent on me to explain how I think that
indicative conditionals can have truth-values, given the argument alluded
to above. The answer is that I do not accept that the assertibility of an
indicative conditional is measured by the conditional probability of its
consequent given its antecedent, and accordingly do not consider that the
probability of such a conditional’s truth must be equal to that conditional
probability, with the absurd consequences that follow. Suppose I have a
fair die, and am about to throw it. Should I regard as highly assertible the
conditional ‘If I throw this die, it will not land with the six uppermost’,
simply because the conditional probability of its landing not-six given that
I throw it is high? I don’t think so – and I still don’t think so even if we
make the die a thousand-faced one. On considering whether to buy one
lottery ticket out of a thousand on sale, I don’t assert ‘If I buy this ticket,
then I won’t win the prize’, even though I fully realize that the conditional
probability of my not winning, given that I buy the ticket, is very high.
Indeed, I’m pretty sure that that conditional is false – for if I didn’t think
so, I would be foolish to enter the lottery. This isn’t to say that I am pretty
sure that the following conditional is true: ‘If I buy this ticket, then I shall
win the prize’. For the negation of the first conditional is not equivalent to
this second conditional. Rather, the negation of the first is equivalent to
the much weaker conditional ‘If I buy this ticket, then I may win the prize’.
That this is so will emerge more clearly when I come to state the general
form of the truth-conditions of conditionals, which will be seen to involve
the explicit employment of modal notions.
3 some methodological considerations

At this point, let us step back a little to consider some methodological
issues. How should we set about devising and defending a theory of
conditionals – a theory which at once gives an account of their meaning
and systematizes their logic? First, what is such a theory supposed to be a
theory of ? Presumably, it is supposed at least to be a theory of ‘If . . .,
then –’ sentences as they feature in ordinary natural language. But this way
of stating the aims of such a theory makes it sound a rather parochial affair.
Conditional statements do not have to be made using the word ‘if’, even if
we restrict ourselves to English. And why, in any case, should we suppose
that ‘ordinary usage’ reflects any very determinate underlying meaning or
principles of inference, as regards ‘if’-sentences and related constructions?
As philosophers, we are surely not merely in the business of lexicography.
Nor should we be engaged in some sort of psychological inquiry into the
168 Conditionality
mental processes generating people’s utterances of ‘if’-sentences. Psycholo-
gists – such as those working on the notorious Wason selection task – tend
to take a rather dim view of the capacities of ordinary folk to reason
correctly with conditionals, and if they are right we ought not to feel too
concerned to respect the ‘intuitions’ of ordinary speakers in framing a
semantic theory of conditionals.10 Our task is more properly conceived of
as a normative one, of trying to explain what we should mean by uttering
conditional statements – and while it would be unsatisfactory to end up
with a theory which conflicted with all of our pre-theoretical intuitions
about the use of ‘if’-sentences, we should be prepared to correct some of
those intuitions in the light of theoretical insight. Certain fundamental
principles we must undoubtly hold on to – such as that modus ponens is
without exception a valid rule of inference for the conditional (modus
tollens, on the other hand, might be something that we should be prepared
to compromise over).11 In framing a general theory of conditionals, includ-
ing counterfactuals, we should, other things being equal, prefer to repre-
sent ‘if’ as being univocal rather than as being systematically ambiguous,
and prefer to represent conditional logic as being reducible to something
more basic rather than as being sui generis. These are general constraints of
simplicity and economy which govern all systematic theory construction.
The theory with which we end up should be capable of explaining ordinary
usage where that can be deemed not to be confused, but should also be
able to explain why deviance should arise in cases in which, according to
the theory, ordinary usage is confused. In explaining these features of
usage, the theory should respect the distinction between semantics and
pragmatics, but not invoke it in an ad hoc way to save the theory from
possible embarrassments.
In implementing these general methodological considerations, we
should, I think, approach the question of the meaning of conditionals
from two ends, hoping to meet in the middle. At one end, we can
provisionally lay down what we take to be certain patterns of inference
involving conditionals which the theory ought to endorse as valid (for
example, patterns invoking the rule of modus ponens). At the other end, we
can tentatively propose an account of the truth-conditions of conditionals
which seems to capture our core understanding of what we intend in
10
I myself am rather sceptical about some of the psychologists’ claims in this regard: see my
‘Rationality, Deduction and Mental Models’, in K. Manktelow and D. E. Over (eds.), Rationality
(London: Routledge, 1993).
11
The validity of modus ponens has been challenged by some theorists, though I find such challenges
unconvincing: see my ‘Not a Counterexample to Modus Ponens’, Analysis 47 (1987), pp. 44–7.
asserting ‘if’-sentences.12 In order to make these ends meet, we may have to
make further adjustments to them, though we should try to keep these to a
minimum. The final result should be an account of the truth-conditions of
conditionals which generates a reasonably comprehensive system of condi-
tional logic and meets the methodological constraints mentioned earlier. (In
effect, what we are aiming at is a state of ‘reflective equilibrium’ in respect of
our intuitions and principles concerning conditionals.) Thus we cannot
dogmatically declare in advance that certain contentious patterns of inference
involving conditionals are ‘fallacious’, even if we can find apparent counter-
examples to them. That our account of the truth-conditions of conditionals
does not sanction such patterns cannot be taken as clear evidence in support of
that account without reference to wider methodological considerations. If
another theory does sanction those inference-patterns but can explain why
they should appear deviant, and at the same time offers a simpler and more
economical overall account of conditionals and their logic, then this other
theory may be more deserving of our allegiance.
4 a sketch of a general theory of conditionals

Let us represent the conditional connective (hoping it to be univocal) by
the box-arrow, ‘□!’.13 (I use the ordinary arrow, ‘!’, for material impli-
cation.) Then I take the following inference-patterns to be at least highly
plausible:
(P1) p □! q ‘ p ! q
(P2) p □! q ‘ ~(p □! ~q)
(P3) p □! q, p □! r ‘ p □! (q & r)
(P4) p □! q, q □! r ‘ p □! r
In connection with (P2), recall my earlier contention (in section 2) that the
negation of ‘If I buy this ticket, then I won’t win the prize’ is equivalent to
12
Geoffrey Hunter, in ‘The Meaning of “If” in Conditional Propositions’, Philosophical Quarterly 43
(1993), pp. 279–97, appears to think that only the first of these strategies is available to us, on the
grounds that there is ‘nothing more to “if” than its logical powers’ (p. 290) and consequently that
‘there is no semantics for conditionals other than the logical powers as expressed in [an appropriate]
axiom-system’ (p. 292). But this claim is very much open to question, for few would dispute that the
logic of truth-functions, say, or modal logics, can be studied both from a syntactic (or proof-
theoretic) and from a semantic (or model-theoretic) point of view – so why should matters be any
different as regards the logic of conditionals?
13
Formal properties of the theory presented in this section are discussed in more detail in my
‘A Simplification of the Logic of Conditionals’, Notre Dame Journal of Formal Logic 24 (1983),
pp. 357–66.
170 Conditionality
‘If I buy this ticket, then I may win the prize’.14 (P2) thus licences the
inference from ‘If I buy this ticket, then I shall win the prize’ to ‘If I buy
this ticket, then I may win the prize’. (It might be imagined that ‘If I buy
this ticket, then I may win the prize’ has the logical form ‘p □! ◊q’, where
the diamond, ‘◊’, expresses possibility. However, ‘p □! ◊q’ clearly does
not contradict ‘p □! ~q’, whereas ‘If I buy this ticket, then I may win the
prize’ clearly does contradict ‘If I buy this ticket, then I won’t win the
prize’. What ‘p □! ◊q’ symbolizes in these circumstances is, rather,
something like ‘If I buy this ticket, then it will be possible for me to win
the prize’.15 Definition (D1), stated at the end of this section, has the
implication that ‘~(p □! ~q)’ entails, but is not entailed by, ‘p □! ◊q’, at
least in a modal logic as strong as S5.)
Note that (P1), taken in conjunction with the fact that modus ponens is
valid for material implication, implies that modus ponens is likewise valid
for the box-arrow:
(P5) p □! q, p ‘ q
Clearly, however, ‘p □! q’ must be stronger than (entail but not be
entailed by) ‘p ! q’ if not only (P1) but also (P2) is correct, because when
the box-arrows in (P2) are replaced by arrows an invalid principle results.16
(P3) seems uncontentious, but (P4) is more controversial, implying as it
does that the box-arrow is a transitive connective. I shall discuss this matter
more fully later.
Having proposed some intuitively plausible principles of inference for
conditionals (which we should be ready to revise if need be), let us now look
at our task from the other end, and try to formulate a plausible account of the
truth-conditions of conditionals, designed to reflect what we intuitively
intend to convey by making a conditional statement. One plausible idea is
14
See further my ‘A Simplification of the Logic of Conditionals’.
15
The use of a future-tense construction here – ‘will be possible’ – should not seem mysterious,
because all of the conditionals considered in this paragraph are future-tense ones, in the sense that
their antecedents and consequents all concern events or states of affairs lying in the speaker’s future.
It is a peculiarity of English grammar that it does not use a future-tense verb, ‘will buy’, in the
antecedent of such a conditional.
16
Hunter does not accept (P2), contending, ‘It is perfectly possible, even with A contingent, for
A □!B and A □! ~B both to be true’ (Hunter, ‘The Meaning of “If” in Conditional Propositions’,
p. 284, with my notation substituted for his). As an example, he invites us to consider the
contrapositives of the two if-sentences ‘If he did not catch the 2.15, he did not catch the boat’
and ‘If he did catch the 2.15, he did not catch the boat’, both of which could be true. But, first, I do
not accept the validity of contraposition (see section 6 below); and, second, it seems that the second
of Hunter’s two if-sentences should be understood as one in which ‘if’ really means ‘even if’ – and
Hunter himself excludes such if-sentences from his purview (p. 280).
that to affirm something of the form ‘If p, then q’ is to imply that there is some
sort of ‘necessary connection’ between p and q, in the sense that we deem it to
be no accident that p and the negation of q are not jointly true. This suggests
that we treat the box-arrow as expressing so-called ‘strict’ implication, so that
‘p □! q’ is equivalent to something of the form ‘□(p ! q)’, with the box ‘□’
suitably interpreted according to context. However, against this it can be
pointed out that, on this interpretation, ‘p □! q’ comes out as vacuously true
whenever the antecedent p is impossible (invoking the appropriate notion of
possibility). This would imply, for instance, that if it was impossible for me to
buy a lottery ticket, I should have to affirm as true the statement ‘If I had
bought a ticket, then I would have won’. It is more plausible to urge that the
truth of the latter presupposes that it was indeed possible for me to buy a
ticket. However, it would apparently be unsatisfactory to conclude that
‘p □! q’ is simply equivalent to the conjunction of ‘□(p ! q)’ and ‘◊p’, for
this would have the opposite defect of making all conditionals with impossible
antecedents automatically false, thus defeating the purpose of a conditional
such as ‘If N were the greatest natural number, then there would be a natural
number greater than N ’. Of course, in the case of a conditional such as the
latter, the consequent expresses a necessary truth. So perhaps what we need to
say is that when ‘p □! q’ is true, then either p is possible or else q is necessary.
In the case of ‘If I had bought a ticket, then I would have won’ – assuming this
to be true – the first disjunct holds (it was possible for me to buy a ticket),
whereas in the case of ‘If N were the greatest natural number, then there
would be a natural number greater than N’ the second disjunct holds (it is
necessary that there is a natural number greater than N, for any natural
number N ). But if it is neither the case that p is possible nor the case that q
is necessary, then I think we should judge ‘p □! q’ to be false. The proposal,
then, is to define ‘p □! q’ as follows:
(D1) p □! q ¼df □(p ! q) & (◊p _ □q)
5 some applications of the theory

I should stress that (D1) is only intended to define the logical form of the
conditional: it still leaves ample scope for variation in interpretation
through various interpretations of the modal operators ‘□’ and ‘◊’. For
instance, the very weakest interpretation of these operators is one on which
they are simply redundant, making ‘p □! q’ come out as equivalent to
‘(p ! q) & (p _ q)’, which is in turn equivalent simply to ‘q’. And,
interestingly enough, there is a class of conditionals for which this does
172 Conditionality
indeed seem to be the right interpretation, namely ‘Austinian’ conditionals
such as ‘There are biscuits on the sideboard, if you want some’.17 The latter
plausibly both entails and is entailed by ‘There are biscuits on the sideboard’.
Alternatively, the modal operators can be interpreted as expressing some more
substantive type of modality, whether epistemic, causal, logical, or deontic,
and as incorporating in many cases a sensitivity to contextual parameters.
We can illustrate this flexibility of interpretation by returning again to
the example of the Oswald/Kennedy conditionals. I urged earlier that the
counterfactual conditional ‘If Oswald had not killed Kennedy, then some-
one else would have’ and the indicative conditional ‘If Oswald has not
killed Kennedy, then someone else will have’ should be seen as equivalent,
differing only in what they pragmatically imply as regards the speaker’s
knowledge of the truth-value of the antecedent. And I also urged that the
other, more familiar indicative conditional, ‘If Oswald did not kill Ken-
nedy, then someone else did’, should be seen as differing from the first
indicative conditional in respect of the modality involved in its interpret-
ation. Specifically, I suggest that the ‘did’-indicative involves an epistemic
modality, whereas the ‘will have’-indicative and the counterfactual both
involve an alethic modality. (An alethic modality is one which, like causal
or logical necessity and possibility, concerns how things are as opposed to
what we know of them.) Thus the ‘will have’-indicative and the counter-
factual both convey the thought that the possible state of affairs of
Oswald’s failing to kill Kennedy is one which is somehow bound to lead
to an alternative assassin’s taking over Oswald’s role, whether by design or
by fate. By contrast, the ‘did’-indicative is used to convey the thought that,
given what the speaker purports to know, he knows for sure that someone –
perhaps Oswald and perhaps someone else – did indeed kill Kennedy. In
both cases we can represent the conditional as having the logical form
‘□(p ! q) & (◊p _ □q)’, but take ‘□A’ to mean, in the one case, something
like ‘it is inevitable that A’, and in the other case something like ‘it is
certain that A’ (with correspondingly different interpretations for ‘◊A’).
Thus the ‘did’-indicative is plausibly true, on this analysis, because it is
certain (known for sure) that it is not both the case that Oswald did not
kill Kennedy and that no one else did either – that is, it is certain that
someone killed Kennedy – and at the same time it is not certain that Oswald
17
See J. L. Austin, ‘Ifs and Cans’, reprinted in his Philosophical Papers, 2nd edn (Oxford University
Press, 1970), p. 210. I concede, however, that there may be something to be said for following
Hunter and denying that if-sentences like this are really ‘conditionals’ at all: see his ‘The Meaning of
“If” in Conditional Propositions’, pp. 279–80.
killed Kennedy. And the other two conditionals plausibly come out as
false, because although it was not inevitable for Oswald to kill Kennedy,
neither was it inevitable for his failure to be attended by the success of
another attempt (that is, it was quite possible for his failure not to be
attended by the success of another attempt).
A problem for my approach may seem to be posed by the case of
‘Dutchman’ conditionals, such as ‘If that’s a Ming vase, then I’m a
Dutchman’. It has been suggested that such a conditional is simply a so-
called ‘material’ conditional, asserted as true only because its antecedent
and consequent are both assumed to be false.18 But no interpretation of the
modal operators in definition (D1) will permit ‘p □! q’ to reduce in any
instance to ‘p ! q’ (in contrast with the way in which it was earlier shown
to reduce to ‘q’ for ‘Austinian’ conditionals). However, it may be ques-
tioned whether the ‘Dutchman’ conditional can really be taken at its face
value, as a conditional with the antecedent and consequent that it appears
to have. As I urged previously, a minimal requirement upon conditionals is
that they be subject to the rule of modus ponens – that someone affirming
something of the form ‘If p, then q’ should be prepared to infer ‘q’, given
the additional information that ‘p’ is true. But, clearly, someone who
affirms ‘If that’s a Ming vase, then I’m a Dutchman’ would not be prepared
to infer ‘I’m a Dutchman’ upon discovering – to his surprise – that it is a
Ming vase. Quite why the idiom takes the form that it does is somewhat
perplexing, but I can see no good reason for treating it purely at its face
value. (A possible explanation is that the idiom exploits the fact that no
conditional with a false consequent can be true if its antecedent is true, so
that to assert what appears to be a conditional with a blatantly false
consequent serves to emphasize the speaker’s rejection of the antecedent.
On the other hand, the idiom is clearly related to another rhetorical device
which does not exploit the conditional form: if someone asserts ‘That’s a
Ming vase’, another speaker may express his contempt for that judgement
by asserting, sarcastically, ‘Oh, yes, and I’m a Dutchman!’)
6 counterfactual ‘fallacies’
At this point it is appropriate for me to reveal – perhaps to no one’s great
surprise – that definition (D1), taken in conjunction with standard prin-
ciples of modal logic, makes all of the principles of conditional logic floated
18
See, for example, J. L. Mackie, Truth, Probability and Paradox (Oxford: Clarendon Press, 1973), pp.
71–2.
174 Conditionality
earlier – (P1) to (P5) – turn out to be valid. We have, it seems, managed to
‘make ends meet’. But not all of those principles will be happily accepted
by all theorists of conditionals – least of all, perhaps, principle (P4), which
represents conditionals as being transitive. Writers on counterfactuals are
apt to say that the ‘fallacy of transitivity’ is one of three major fallacies
involving such conditionals,19 the other two being the ‘fallacy of
strengthening the antecedent’ and the ‘fallacy of contraposition’ – these
last two being inferences of the following forms respectively:
(F1) p □! q ‘ (p & r) □! q
(F2) p □! q ‘ ~q □! ~p
(Many of the same writers hold that in the case of indicative conditionals all
three forms of inference are valid – though, if I am right in contending that the
indicative/subjunctive distinction is of no logical significance, this divided
stance on the alleged fallacies cannot be a consistent one.) Let me remark,
right away, that definition (D1) does not sanction the validity of either (F1) or
(F2), but only the validity of the following significantly weakened principles:
(P6) p □! q, ~(p □! ~r) ‘ (p & r) □! q
(P7) p □! q, ◊~q ‘ ~q □! ~p
It certainly is true that (D1) sanctions the validity of the transitivity principle,
(P4), but – I would urge – this principle has in any case much greater intuitive
appeal than either of the others (in their unqualified forms).20 Nonetheless,
apparent counterexamples have been devised, which we need to examine.21
One of the best known is Robert Stalnaker’s Hoover counterexample,22
19
Notably, David Lewis: see his Counterfactuals (Oxford: Blackwell, 1973), pp. 31 ff.
20
Hunter defends the validity of contraposition while denying that of both strengthening the antecedent
and transitivity: see his ‘The Meaning of “If” in Conditional Propositions’, p. 285. But even he concedes
that the validity of contraposition is still open to debate. Someone who believes that all three principles
are valid even for subjunctive conditionals is Peter Urbach: see his ‘What Is a Law of Nature? A Humean
Answer’, British Journal for the Philosophy of Science 39 (1988), pp. 193–210.
21
Hunter offers what seems to me to be a particularly weak ‘counterexample’ to transitivity involving
indicative conditionals, asserting baldly that ‘from “If you strike that match it will light, and if it will
light then it is not wet” you may not validly infer “If you strike that match then it is not wet”’: see
his ‘The Meaning of “If” in Conditional Propositions’, p. 285. Clearly, if the latter conditional is
deemed false, the following should be deemed true: ‘If you strike that match, it may be wet’. But no
one deeming this to be true ought to assert the two conditionals of Hunter’s premise. Hunter’s
conclusion only sounds at all strange if one reads into it some implication that striking a match can
cause it not to be wet: but although conditionals are often used to intimate the existence of a causal
connection, they don’t have to be understood in this way.
22
See Robert Stalnaker, ‘A Theory of Conditionals’, in Studies in Logical Theory, American
Philosophical Quarterly Monograph 2 (Oxford: Blackwell, 1968), reprinted in Jackson (ed.),
Conditionals.
J. Edgar Hoover being the head of the FBI at the time that Stalnaker was
writing: ‘If J. Edgar Hoover were today a communist, he would be a traitor’
(true), ‘If J. Edgar Hoover had been born in the Soviet Union, he would today
be a communist’ (true), therefore (?) ‘If J. Edgar Hoover had been born in the
Soviet Union, he would be a traitor’ (false, surely). It is worth noting that the
alleged counterexample is more persuasive when the premises are presented in
this order, rather than in the order dictated by principle (P4). This in itself
suggests that some pragmatic effect is in operation. My own view is that what
the alleged counterexample demonstrates is not the invalidity of principle
(P4), but rather the context-sensitivity of counterfactuals like these. Before
I demonstrate how this explanation of the Hoover example works, I need to
show in just what sense counterfactuals are context-sensitive.
7 the context-sensitivity of counterfactuals

Quite generally, a sentence (type) is ‘context-sensitive’ just in case the
proposition expressed by (a token of) it is partly determined not merely by
the standard meanings of its constituent words but also by its circum-
stances of utterance. Thus, for instance, the proposition expressed by (a
token of) the sentence (type) ‘I am hungry’ is partly determined by the
identity of the speaker and the time of utterance. Consequently, an
assertion of the sentence ‘I am hungry’ can be consistent with a denial of
that same sentence, provided the circumstances of the assertion and the
denial are appropriately different. This is obvious enough with sentences
containing indexical expressions such as ‘I’, ‘here’, and ‘now’. But I think it
is also true of many counterfactual conditionals, quite independently of
any indexical expressions which they may contain. That is, even when the
references of any such indexical expressions have been fixed by the circum-
stances of utterance, there may still be a residual context-sensitivity as
regards what proposition a counterfactual sentence should be taken to
express. This additional sensitivity is a sensitivity to certain conversational
intentions of the speaker, as the following example serves to show.23
Suppose we are together in a room which we both know to contain a
considerable amount of highly flammable gas owing to a gas leak, and we
both observe the presence there of a third person, Brown, concerning whom
we know the following facts: first, that he has in his hand a box of dry and
perfectly sound matches; and second, that he is an extremely cautious
23
See also my ‘Jackson on Classifying Conditionals’, where this and other examples are discussed in
more detail.
176 Conditionality
individual who is exceptionally sensitive to the presence of gas and strongly
averse to risking its ignition by a naked flame. You may assert, with very
good reason, ‘If Brown had struck one of those matches just now, then there
would have been an explosion’. Your reason is that naked flames in the
presence of flammable gas cause explosions. Plausibly, what you say – the
proposition that you express – is true. Even so, I may assert, with equally
good reason, ‘If Brown had struck one of those matches just now, then there
would not have been an explosion’. My reason is that Brown, being hyper-
cautious, only strikes matches when he is sure that flammable gas is not
present, and consequently that he would only have struck a match just now
if he had first ensured the removal of all such gas. Plausibly, what I say – the
proposition that I express – is also true. Yet how can that be so, since you and
I appear to contradict each other? What you say is of the form ‘p □! q’ and
what I say is of the form ‘p □! ~q’, and yet according to principle (P2), from
‘p □! q’ one may infer ‘~(p □! ~q)’, so that from what you say you should
infer, it seems, the negation of what I say.
We appear to disagree with each other. But in fact there need no more be a
genuine disagreement between us here than there would be if I were to say ‘I
am hungry’ and you were to say ‘I am not hungry’. The difference between us
is not a difference concerning the truth-value of any proposition that either of
us expresses, but a difference concerning what propositions we take our
utterances to express in these circumstances. You are using the
counterfactual sentence ‘If Brown had struck one of those matches just
now, then there would have been an explosion’ to convey something about
the causal relation between naked flames and explosions in the presence of
flammable gas. I am using the counterfactual sentence ‘If Brown had struck
one of those matches just now, then there would not have been an explosion’
to convey something quite different but compatible with what you say –
something about Brown’s cautiousness in the presence of flammable gas.
8 possible-worlds interpretations of counterfactuals

One way to render perspicuous the context-sensitivity of counterfactuals is
to utilize the language of ‘possible worlds’ in stating their truth-
conditions.24 (This approach is familiar anyway through the work of
24
Hunter is very scathing about possible-worlds analyses of conditionals: see his ‘The Meaning of “If”
in Conditional Propositions’, pp. 289–90. He objects that we can often determine the truth-value of
a conditional even though ‘[a]bsolutely nothing about . . . possible worlds enters our heads’ (p. 289).
But a possible-worlds semantics for conditionals purports to state their truth-conditions, not to
describe the psychological processes that we undergo in assessing their truth-values. I myself have no
Stalnaker and Lewis, although I disagree with the details of their analyses.)
Earlier I put forward (D1) as defining the logical form of all conditional
sentences, including counterfactual conditionals, but stressed that the
interpretation of any particular conditional sentence would depend upon
an appropriate reading of the modal operators involved. Now, as is
standard, we can give a generic reading of the necessity operator in terms
of possible worlds by taking ‘□A’ to say that ‘A’ is true in every possible
world. Likewise we can take ‘◊A’ to say that ‘A’ is true in some possible
world. But what worlds we take ourselves to be quantifying over in any
particular case will depend, first of all, upon what type of modality we have
in mind – logical, causal, or deontic, for instance. Thus we may restrict our
domain of quantification to all physically possible worlds (worlds in which
our actual laws of nature still hold true), or we may restrict it to all
deontically possible worlds (worlds in which no actual moral imperative
is infringed). Further than this, though, we may impose certain constraints
on the similarity between the worlds over which we quantify and the actual
world. Indeed, it seems that we shall have to do this in interpreting
counterfactual conditionals, for the simple reason that in many cases not
to do so would result either in vacuity or in trivial falsehood.
Consider, for instance, the counterfactual conditional ‘If Brown had
struck one of those matches just now, then there would have been an
explosion’. According to (D1), this has the following possible-worlds
interpretation: in every possible world it is not both the case that Brown
strikes one of those matches and there is not an explosion, and either in
some possible world Brown strikes one of those matches or else in every
possible world there is an explosion. Now, even if we restrict the domain of
quantification to physically possible worlds, this interpretation will inevit-
ably make an assertion of this counterfactual false unless further constraints
are imposed on the range of worlds to be taken into account. For there will
of course be some physically possible worlds in which Brown strikes one of
the matches and there is not an explosion – for instance, worlds in which a
super-efficient sprinkler system puts out the lighted match quickly enough
to prevent the gas igniting. So we have to specify that the range of worlds
to be taken into account must be similar to the actual world in certain
relevant respects – for instance, they must not contain such sprinkler
systems.
sympathy for the metaphysics of possible worlds, as I make clear in Chapter 8, but am happy to
exploit them for present purposes purely as a heuristic device with no serious ontological
implications.
178 Conditionality
But in which respects should similarity be deemed relevant in any given
case? That, I contend, depends upon what the speaker intends to convey
by asserting a given counterfactual in the circumstances in which he does.
For instance, if his intention in asserting the counterfactual just cited is to
convey something about the causal relation between naked flames and
explosions in the presence of flammable gas, then we need to interpret
what he says as restricting the range of possible worlds taken into account
to ones which are at least similar to the actual world in having such gas
present in the room. However, such worlds will only contain the event of
Brown’s striking a match if they also differ considerably from the actual
world with regard to Brown’s psychology. Hence if, conversely, it is
appropriate to a speaker’s intentions to suppose that the range of relevant
worlds is restricted to ones which are similar to the actual world with
regard to Brown’s psychology, we should not expect a speaker to be
prepared to assert the counterfactual ‘If Brown had struck one of those
matches just now, then there would have been an explosion’. The import-
ant point to appreciate is that we cannot accommodate similarity to the
actual world both in respect of the presence of gas in the room and in
respect of Brown’s psychology, without restricting ourselves to worlds in
which Brown does not strike a match – under which restriction any counter-
factual beginning ‘If Brown had struck one of those matches just now, . . .’
turns out to have, in the relevant sense of ‘impossible’, an impossible
antecedent. And yet, without reference to the speaker’s intentions, there
is no principled way of deciding which of these aspects of similarity should
be accommodated and which disregarded.
The crucial fact to emerge from all this is that a speaker’s intentions in
asserting a counterfactual can help to determine the propositional content
of his assertion by fixing an appropriate measure of similarity across
possible worlds for the proper evaluation of the truth or falsity of what
he asserts. This should not be seen as making the truth of a counterfactual
assertion dependent upon the whims of the speaker: what depends on the
speaker’s intentions is what measure of similarity is to be deemed appro-
priate for the evaluation of what he asserts, but whether or not his assertion
is to be evaluated as true by that standard is not subject to his whim.
9 the question of transitivity

Now we can return to the alleged counterexamples to transitivity – that is,
the apparent violations of principle (P4) – such as Stalnaker’s Hoover
example. My contention is that we are seduced into conceding that this is a
counterexample by sleight of hand, rendered possible by the context-
sensitivity of counterfactuals.25 Upon being presented with a counterfac-
tual sentence, we naturally endeavour, if possible, to construe it according
to a similarity-measure which will make it come out as true, provided that
we can think of such a measure which is not unduly strange. This is just an
application of a principle of charity in our practice of interpretation: other
things being equal, try if you can so to interpret an assertion that it
plausibly comes out as expressing a true belief. Now, presented with the
first premise of the Hoover example, this principle is easy to honour: we
simply adopt a similarity-measure which makes worlds in which Hoover is
still head of an FBI serving a capitalist US government relevantly similar –
for in all such worlds in which Hoover is a communist, he is also a traitor.
We can also easily honour the principle of charity in respect of the second
premise of the Hoover example, by adopting a similarity-measure which
makes worlds in which Hoover adheres to the ideology of his native
country relevantly similar – for in all such worlds in which Hoover is born
in a communist-run state, he is a dutiful communist. However, when we
come to the conclusion of the Hoover example it is no longer so easy to
think of a similarity-measure which will, without an undue effort of the
imagination, make the counterfactual in question have an interpretation
under which it is to be evaluated as true. We are therefore apt to adopt the
same similarity-measure as we did in interpreting the second premise – not
least because both counterfactuals have the same antecedent – and
according to this measure the conclusion comes out as false.
But in order to see this sequence of counterfactuals as constituting a
genuine counterexample to transitivity, it is obviously imperative that they
should all be interpreted according to the same similarity-measure. Other-
wise, all that we have on our hands is a fallacy of equivocation. And here
we should observe that if we now go back to the first premise and apply the
same similarity-measure as was used to evaluate the second premise as true
and the conclusion as false, it turns out that according to this similarity-
measure the first premise is also false – for, of the worlds in which Hoover
adheres to the ideology of his native country, there are plainly many in
which he is a communist and yet is not a traitor, such as worlds in which
he is born in the Soviet Union and remains there all his life. On the other
hand, if, in evaluating all three counterfactuals, we adopt the similarity-
measure which we originally used to interpret the first premise, then it is
25
See my ‘Conditionals, Context and Transitivity’, Analysis 50 (1990), pp. 80–7, for an earlier version
of the explanation which follows. The present version is, I think, rather better.
180 Conditionality
the second premise which comes out as false, because if the only relevantly
similar worlds are ones in which Hoover remains head of an FBI serving a
capitalist-run US government, there can plainly be many such worlds in
which Hoover is born in the Soviet Union and yet is not a communist –
for instance, worlds in which he is a dissident emigrant to the USA.
Finally, if we adopt a similarity-measure which makes relevantly similar
those worlds in which Hoover both adheres to the ideology of his native
country and remains head of an FBI serving a capitalist US government,
then it turns out that the conclusion of the Hoover example should, after
all, be evaluated as true. For in all such worlds in which Hoover is born in
the Soviet Union, he is a committed communist who has come to work in
the service of a capitalist US government – work which requires him to
have adopted American citizenship – with the consequence that he is
indeed a traitor to his adopted country.
Thus we see that however we achieve consistency in applying the same
similarity-measure to evaluate all three counterfactuals, we cannot end up
with an evaluation which makes both premises true and the conclusion
false. We are only lulled into thinking that we have a counterexample to
transitivity on our hands because we are tempted to adopt a new similarity-
measure to interpret the second premise, since we instinctively look for a
measure which will make the assertion come out as true, if this can
reasonably be done. Then we forget to review our evaluation of the first
premise in the light of this change. But if the premises are presented in the
reverse order, the impression that we have a counterexample to transitivity
gives way to an impression that both premises cannot plausibly be evalu-
ated as true without equivocation (equivocation over the choice of
similarity-measure to be adopted in evaluating them).
10 conclusion
Conditionals in general present an extremely perplexing set of linguistic
phenomena which often seem to defy a simple, uniform treatment of them
for logical purposes. In this chapter I have tried to show how one can
defend a relatively simple core theory of the logic of conditionals while
respecting the many subtle differences of their interpretation which the
complexities of usage demand, especially in the case of counterfactuals.
(And here I should stress that the logic of conditionals generated by
definition (D1) is, of course, entirely reducible to standard monadic modal
logic, whereas this is not so for the systems of Stalnaker and Lewis, in
which the counterfactual conditional connective is irreducibly dyadic.)
Counterfactuals are indeed highly context-sensitive, but provided that this
is properly recognized they can still be seen to obey a relatively straightfor-
ward underlying logic and to possess determinate truth-values in a wide
range of cases. This context-sensitivity is not something that we should
deplore, but an inevitable feature of the indispensable communicative role
which counterfactuals play in rational discourse.
chapter 10
Conditionals and conditional probability
As I made clear in the previous chapter, my own view – which is contrary

to mainstream opinion amongst philosophical logicians – is that, while
there is a genuine distinction to be drawn between indicative and counter-
factual (or, more generally, subjunctive) conditionals, even if the nomen-
clature for characterizing this distinction may be called into question, the
distinction is not one that arises at the level of the logical form of condi-
tionals of the two kinds. In my view, conditionals of both kinds submit to
the same logical analysis, both being reducible to propositions involving
only truth-functional connectives and (monadic) modal operators. Thus,
my contention is that any conditional of the form ‘If p, then q’
is analysable as being logically equivalent to a proposition of the form
‘□(p ! q) & (◊p ˅ □q)’ – see definition (D1) of Chapter 9 – although in
the case of counterfactual conditionals the kind of modality involved is
‘alethic’, whereas in the case of indicative conditionals it is ‘epistemic’. As
I say, this is far from being the most widely accepted view, which is that
indicative conditionals in natural language are logically equivalent to so-
called material conditionals – so that an indicative conditional of the form
‘If p, then q’ is logically equivalent to a proposition of the form ‘p ! q’,
where the arrow ‘!’ symbolizes the truth-functional connective com-
monly known as material implication – whereas counterfactual condition-
als require a quite different possible-worlds analysis, according to which,
roughly speaking, ‘If p were the case, then q would be the case’ means ‘In
all the closest possible worlds in which p is the case, q is also the case’. It is
important to note that a proposition of the latter form is not reducible to
one involving only truth-functional connectives and (monadic) modal
operators. Thus, according to mainstream philosophical opinion, the word
‘if’, as it functions in natural language, is ambiguous in a very deep sense –
something which seems to me to be inherently extremely implausible. But
this mainstream view is not the only rival to my own. Another is the view
that indicative conditionals, at least, are to be analysed in terms of the
182
Conditionals and conditional probability 183
notion of conditional probability, as this is understood in the mathematical
theory of probability. This view seems to me to be preferable to the
mainstream view, since there is some prospect of extending it to embrace
also subjunctive conditionals and thereby preserving the univocality of ‘if’.
However, despite this virtue, I think that the view also has insuperable
defects which require us to reject it, leaving – I hope – my own view as the
most plausible one, all things considered.
Now, much philosophical ink has been spilt on the relationship between
conditional statements and conditional probability, including a famous proof
by David Lewis that the probability of a conditional’s being true cannot on
pain of triviality be identified with a conditional probability, and a defence by
Dorothy Edgington of the thesis that, for this very reason, conditional
statements do not express propositions and so do not have truth-conditions.1
It is Edgington’s position that I particularly wish to examine in this chapter,
although I also have more general points to make both about conditionals and
about the notion of conditional probability. Above all, I wish to defend the
view that the notion of conditionality is conceptually prior to that of conditional
probability, so that the latter can be explicated in terms of the former, but not
vice versa. My own view of conditionals can, of course, happily accept this, but
Edgington’s view equally obviously cannot. Thus, by attacking Edgington’s
view – which, I should emphasize, I hold in the highest regard and for that
reason take to be all the more worthy of critique – I hope indirectly to garner
support for my own, or at least to remove a serious threat to it. But I shall say
no more about what I earlier called the mainstream view, which in my opinion
is seriously flawed as an account of the semantics of indicative conditionals and
far from satisfactory as an account of the semantics of counterfactuals, not to
speak of its highly implausible implication that ‘if’ is ambiguous. I should also
make it clear that in the present chapter, in contrast to the previous one, which
concentrated on counterfactual conditionals, the focus will be on indicative
conditionals, since it is this kind of conditional that the probabilistic approach
seems best suited to deal with, even if it is in principle extensible to subjunctive
conditionals as well. Thus this and the previous chapter, between them, are
intended to convey my current views about the semantics of conditionals quite
generally, both indicative and subjunctive.
1
David Lewis’s results are to be found in his ‘Probabilities of Conditionals and Conditional
Probabilities’, Philosophical Review 85 (1976), pp. 297–315, reprinted in F. Jackson (ed.),
Conditionals (Oxford University Press, 1991). For more on such triviality proofs, see E. Eells and
B. Skyrms (eds.), Probability and Conditionals (Cambridge University Press, 1994). For Edgington’s
views, see especially her ‘On Conditionals’, Mind 104 (1995), pp. 235–329, which provides the main
target for most of what I have to say here.
184 Conditionality
1 edgington on conditionals
According to Edgington, probability may be interpreted as degree of belief –
in the sense that degrees of (rational) belief are held to bear to one another
relationships captured by the axioms of the standard probability calculus.2
Clearly, it cannot plausibly be claimed that it is psychologically impossible
for a person actually to possess degrees of belief, or ‘partial’ beliefs, which
do not stand in such relationships – though it can be argued that such a
person’s belief distribution would not be coherent, in the sense that a
‘Dutch book’ could be made against him or her. Hence we are dealing
with the beliefs of a putatively ‘rational’ subject, at least in this fairly
minimal sense of ‘rational’.
Now, standardly, in the mathematical theory of probability (which
Edgington has no wish to query on this score), the conditional
probability of B given A (where A and B are propositions) is defined as
the ratio of the probability of the conjunction of A and B to the probability
of A (provided that p(A) > 0), as follows:3
(1) p(B|A) ¼ p(A & B)/p(A)
Accordingly, Edgington proposes to endorse the corresponding equiva-
lence, with ‘degree of belief’ standing in for ‘probability’ (and with the
corresponding proviso that b(A) > 0):
(2) b(B|A) ¼ b(A & B)/b(A)
This is interpreted as meaning: one’s degree of belief in B given A equals
the ratio of one’s degree of belief in the conjunction of A and B to one’s
degree of belief in A. But what is a ‘degree of belief in B given A’?
According to Edgington, it is just a degree of belief in the conditional ‘If
A, then B’.4 (She holds this because of her acceptance of what is sometimes
called ‘Adams’s thesis’ – a thesis which I shall examine and criticize later in
this chapter.) Then, in the light of Lewis’s triviality result, Edgington
2
Compare F. P. Ramsey, ‘Truth and Probability’, in his The Foundations of Mathematics and Other
Logical Essays (London: Kegan Paul, 1931).
3
Note, however, that Bruno de Finetti – with Ramsey held to be one of the founders of the
‘subjective’ theory of probability – denied that (1) should be seen as a definition of conditional
probability, holding it to be, rather, ‘a theorem derived from the requirement of coherence’: see
J. von Plato, Creating Modern Probability (Cambridge University Press, 1994), p. 270. The proviso
that p(A) 6¼ 0 is required, of course, because division by zero is mathematically impossible.
4
For present purposes, we may take this to be an indicative conditional, although Edgington believes
that her approach may be extended to include subjunctive or counterfactual conditionals. For more
on the distinction between indicative and subjunctive conditionals, see Chapter 9.
contends that ‘If A, then B’ does not express a proposition, and so does not
have truth-conditions nor, consequently, a truth-value. According to her,
to assert ‘If A, then B’ is not to assert the truth of any proposition: rather, it
is conditionally to assert the consequent, B, under the supposition that, or
‘given’, the antecedent, A.
2 some problems for edgington’s view

Edgington’s doctrine, although superficially attractive, appears – to me, at
least – utterly mysterious upon deeper inspection. First of all, we always
supposed that belief is a propositional attitude: that what we believe is
always that such-and-such, where ‘such-and-such’ is something propos-
itional in nature. Not so, it seems, on Edgington’s theory – or, rather, not
always so. Sometimes we believe (that) such-and-such: for instance, we
may believe – to some degree, on Edgington’s account – (that) A or (that)
A and B. But at other times we believe, to some degree, B given A, or if A
then B, which (allegedly) is not a proposition. Or perhaps not: for in more
considered moments Edgington seems to want to say that we have condi-
tional beliefs (as opposed to belief in conditionals), so that it is not that ‘B
given A’ is somehow the object of our belief, but rather that we have a
degree of belief in B conditional upon, or given, A.5 On this interpretation,
the contents of belief are indeed always propositional, but it is just that in
addition to straightforward beliefs, we have ‘conditional beliefs’, which are
no more beliefs than conditional assertions are assertions.
Either way, though, difficulties loom. Suppose we take the first inter-
pretation, according to which ‘b(B|A)’ expresses a degree of belief in ‘B
given A’, or (equivalently, for Edgington), a degree of belief in the condi-
tional ‘If A, then B’. And remember how ‘b(B|A)’ is defined – as a ratio of
two degrees of belief; that is, as the ratio between b(A & B) and b(A). Now,
is it not odd to say that a ratio between two degrees of belief is itself a
degree of belief ? Surely, this is just as odd as saying that a ratio between two
degrees of temperature is itself a degree of temperature. If I say that the
ratio between the average winter temperature and the average summer
temperature in a certain place is one half, what I am saying is that it is
twice as hot in summer as in winter in that place, on average. I am not
saying anything at all about what the temperature is at that place at any
time. (I shall return to this analogy later.)
5
See Edgington, ‘On Conditionals’, p. 263.
186 Conditionality
The confusion here – if that is what it is – may be traced back to the
original ‘definition’ of ‘conditional probability’:
(1) p(B|A) ¼ p(A & B)/p(A)
In particular, we may be misled by the appearance of the symbol ‘p’ on
both sides of the equation. What we need to appreciate is that the
significance of ‘p’ as it appears on the left-hand side of the equation has
nothing whatever to do with its significance as it appears on the right-hand
side. There is orthographic sleight of hand going on here. For, on the left-
hand side, ‘p’ – or, more exactly, ‘p( )’ – does not really appear at all, as a
well-formed symbol, in the way that it does (twice) on the right-hand side.
Rather, what significantly appears on the left-hand side is ‘p( | )’ – which is
a quite different symbol from ‘p( )’. There is no good reason at all for
including ‘p’ in both of these symbols. ‘p( )’ no more occurs significantly in
‘p( | )’ than ‘rat’ occurs significantly in ‘Socrates’. In short: conditional
probability, so-called, is not a kind of probability – not, at least, if (1) is
supposed to be taken as defining what it is. The very term ‘conditional
probability’, it now seems, is an oxymoron.
But – it may be asked – why can’t the same move be made as was
made earlier regarding the notions of conditional assertion and condi-
tional belief? That is, why not just say that ‘conditional probability’ is the
probability of one proposition, B, given another proposition, A? Accord-
ingly, we could write it as ‘p(B)/A’, rather than as ‘p(B|A)’. I imagine that
Edgington would, in fact, have no objection to this notation, since she
has no objection to something very similar to it, namely ‘pA(B)’.6 But let
us be clear about what this manoeuvre can achieve, which I think is very
little. We should not be under any illusion that, by making this move, we
have ensured that in talking of ‘conditional probability’ we really are
talking about probability. For, just as we should not be under any illusion
that ‘p’ in ‘p( | )’ means the same as ‘p’ in ‘p( )’, so too we should not be
under any illusion that ‘p’ in ‘p( )/’ means the same as ‘p’ in ‘p( )’ either.
Indeed, ‘p’ has no independent meaning in any of these symbols. So it is
just a cheat to assume that we are talking about the same sort of thing –
‘probability’ – in both cases. The fact is that the only relation that so-
called ‘conditional probability’, as defined by (1), has to any kind of
probability – where the latter is understood as a notion explicated by the
axioms of the standard probability calculus (Kolmogorov’s axioms) – is
that the former is defined as a ratio of probabilities. And again I make the
6
See Edgington, ‘On Conditionals’, p. 263.
point that a ratio of two degrees of some quantity is not, in general, itself
a degree of that – or indeed any – quantity. This is simply an application
of ‘dimensional analysis’ – the technique which tells us, for instance, that
a quantity which is the product of velocity and time has the dimension of
length.
Perhaps it may be urged against me here that degrees of probability, and
likewise of belief, are ‘dimensionless’, unlike degrees of temperature, so
that this point does not apply to the former. Alternatively, it might be
urged that even with degrees of temperature, such a ratio does determine a
‘temperature’, but one measured according to a different scale. However,
all that these objections really appeal to is the fact that a ratio between
numerical values is always itself a numerical value, provided that the
denominator is not zero (pure numbers are, of course, ‘dimensionless’).
They do not address my central complaint that the ‘p’ in ‘p( | )’ and its
variants bears a merely orthographic relationship to the ‘p’ in ‘p( )’: in
short, that the meaning of ‘conditional probability’ has been left entirely
undetermined by (1), even if ‘its’ numerical value has been fixed – whatever
‘it’ is. At the very least, then, even if I have not proved that the ratio of
probabilities specified in (1) is not a probability, I hope it is clear that those
who think otherwise have some explaining to do: they need to explain in
what sense such a ratio is a probability.
These last remarks, it will be seen, provide the basis of an objection to
the alternative way, canvassed above, of understanding the notion of
(degree of) conditional belief, according to which ‘b(B|A)’ is understood
as denoting the degree of belief in B, conditional upon, or given, A (rather
than as the degree of belief in ‘B given A’). My point would again be that
we deceive ourselves if we think that we have articulated here a genuine
notion of belief, which bears more than a nominal relationship to the
standard notion, as expressed by the verb ‘believe’. The ‘b’ in ‘b(B|A)’ bears
a merely orthographic similarity to the ‘b’ in ‘b(A)’, where the latter is
taken to denote degree of belief in a proposition A. In short, theorists like
Edgington have done nothing whatever to show that they are entitled to
use the expressions ‘degree of belief in A’ and ‘degree of belief in B,
conditional upon A’ (or ‘degree of conditional belief in B, given A’) with
any expectation that the words ‘degree of belief’, as they figure in those
expressions, bear any significant semantic relationship to one another
whatsoever.
There is, however, an obvious response which the devotees of condi-
tional probability can make here. This is to say that these conditional
notions are the semantically prior ones, and that the notions of ‘absolute’
188 Conditionality
probability and ‘absolute’ belief are to be understood as special cases of the
general notions. Thus, on this view, the ‘absolute’ probability of A, p(A), is
just the conditional probability of A given the tautology, T:7
(3) p(A) ¼ p(A|T)
Likewise, then, devotees of the notion of conditional belief could perhaps
say that a degree of belief in a proposition A, b(A), is ‘really’ just the degree
of conditional belief in A given T, b(A|T). This approach may work,
perhaps, for probability (although one might argue that even here it is
merely cosmetic, because ‘p( |T)’ is effectively just a stylistic variation on
‘p( )’, and still represents a function of just one argument). But it is not so
easy to say this about belief, as I shall now explain.
Remember, we canvassed two different construals of the notion of
‘conditional belief in B given A’, which carry across to the special case
where A ¼ T. According to one, to have a degree of belief in A given T is to
believe, to some degree, ‘A given T ’, or ‘If T, then A’ – where this is not to
have some degree of belief in any proposition. So it turns out, on this
construal, that we were wrong to suppose that belief is a propositional
attitude at all. No object of belief has truth-conditions! On the other
construal, to have a degree of belief in A given T is to believe, to some
degree, A, given, or conditional upon, T. On this construal, the object of
belief is always something propositional, with truth-conditions and (poten-
tially, at least) a truth-value. Even so, the suggestion now is that we can
only understand belief in a proposition as an irreducibly conditional
notion: to have a high degree of belief that, say, it will rain tomorrow, is
‘really’ to have a high degree of conditional belief that it will rain, given
some arbitrary tautology – say, given that it will either rain or not rain
tomorrow. (Some might want to go even further than this, and urge that
all belief is ‘really’ conditional belief given, or relative to, some body of
‘background knowledge’, K: but such a doctrine has all the disadvantages
of the view now being examined, and more besides.) This suggestion is
frankly incredible. Moreover, the implication is that whenever we make
what appears to be an unconditional assertion – for instance, ‘It will rain
tomorrow’ – we are ‘really’ making a conditional assertion, in this instance
one such as ‘If it will either rain or not rain tomorrow, then it will rain
tomorrow’. And, on Edgington’s theory, this conditional assertion cannot
7
Compare Richard Swinburne, An Introduction to Confirmation Theory (London: Methuen, 1973),
pp. 34 ff.
be something with truth-conditions or a truth-value. Remarkably, it turns
out that nothing that we assert is ever really true.8
3 alternative notions of conditional probability

Of course, we can say that there is such a thing as the probability that B
will/would have if A is/were true – and we could quite intelligibly call this
‘conditional probability’.9 But, transparently, this notion of conditional
probability cannot be used to explicate conditionals (which is what
Edgington wants to do), since it relies on a conditional for its very
expression – for instance, if we adopt the following definition:
(4) pA(B) ¼ x ¼df If A, then p(B) ¼ x
(‘The conditional probability of B given A is x’ means ‘If A (is true), then
the probability of B is x’.) I might add that, very arguably, we understand
‘if’ better than we do ‘given’, so that this approach is preferable to the
reverse approach of Edgington, who attempts – in my view – to explain the
obscure by the more obscure.
Edgington, however, explicitly denies that one can read ‘It is probable
that B given A’ as ‘If A, then it is probable that B’: she thinks this is a
‘howler’.10 Her argument is that this would have the absurd consequence
that all probabilities are 1 or 0. Why? Because p(A|A) ¼ 1 and p(A|~A) ¼ 0:
but if we read these as ‘If A, then p(A) ¼ 1’ and ‘If ~A, then p(A) ¼ 0’,
then, given the logical truth of ‘A ˅ ~A’, we can derive ‘p(A) ¼ 1 ˅ p(A) ¼
0’. She adds: ‘“I’m sure that A if A” does not have the consequence that if
A (is true), then I’m sure that A (is true)’.11 However, this argument
assumes that ‘p(A|A)’ is correctly defined as ‘p(A & A)/p(A)’, and conse-
quently does ¼ 1. We are perfectly at liberty to define ‘The conditional
probability of A given A ¼ x’ as meaning ‘If A (is true), then the
probability of A ¼ x’, and simply deny that x in this case must ¼ 1.
Edgington’s argument is clearly question-begging. What it does show (at
best) is that the two different definitions of conditional probability are
incompatible – it doesn’t show which, if either of them, is ‘right’.
In any case, we could offer yet another definition of conditional prob-
ability which would be altogether immune to this argument of
8
I note that Edgington does allow that a conditional with a true antecedent may be said to be true/
false according as the consequent is true/false – but this is ‘truth’ only by courtesy, and not
fundamental to her theory of conditionals: see her ‘On Conditionals’, pp. 290–1.
9
Compare Donald Nute, Topics in Conditional Logic (Dordrecht: Reidel, 1980), p. 109.
10 11
See Edgington, ‘On Conditionals’, p. 269. Edgington, ‘On Conditionals’, p. 269.
190 Conditionality
Edgington’s. According to this definition, ‘The conditional probability of
B given A ¼ x’ means ‘If the probability of A ¼ 1, then the probability of
B ¼ x’:
(5) pA(B) ¼ x ¼df If p(A) ¼ 1, then p(B) ¼ x
According to this definition, pA(A) ¼ 1 just in case: If p(A) ¼ 1, then
p(A) ¼ 1. And since the latter conditional is certainly true – even by
Edgington’s theory, since it is of the form ‘If P, then P’ – it follows that,
indeed, pA(A) ¼ 1. As for p~A(A), this ¼ 0 just in case: If p(~A) ¼ 1, then
p(A) ¼ 0. And the latter is also certainly true, by the axioms of the
probability calculus. However, Edgington can now no longer appeal to
the logical truth of ‘A ˅ ~A’ to derive the supposedly absurd conclusion
‘p(A) ¼ 1 ˅ p(A) ¼ 0’. She would need to appeal, instead, to ‘p(A) ¼ 1 ˅
p(~A) ¼ 1’, which is just equivalent to the very conclusion that she wants
to reject. As for her other remark, quoted earlier, the corresponding thing
to say now would be this: ‘I’m sure that A if A’ does not have the
consequence that if I am sure that A, then I’m sure that A. But, of course,
it precisely does have this consequence.
Another way of writing the definition of conditional probability just
proposed is this:
(6) pA(B) ¼df the probability x such that if p(A) ¼ 1, then p(B) ¼ x
In words: the conditional probability of B given A is the probability which
B has/would have if the probability of A is/were 1. It should be observed
that this definition results in the same value for the conditional probability
of B given A as does the standard ratio-based definition of conditional
probability, (1), when that is taken in conjunction with the usual ‘Bayesian’
assumption that we should update our subjective probabilities according to
the principle of ‘conditionalization’. According to the latter principle, the
value that p(B) should have when A is discovered to be true is the value
that p(B|A) had beforehand.12 The rationale for this is that upon learning
that A – that is, when p(A) ¼ 1 – p(B|A) ¼ p(B), and so, provided that the
value of p(B|A) does not change upon learning that A, it follows that the
value that p(B|A) has before learning this equals the value that p(B) has or
would have after learning this. However, this is just what the newly
proposed definition, (6), implies with regard to pA(B) – namely that its
value is equal to the value that p(B) has or would have if p(A) ¼ 1.
12
See Colin Howson and Peter Urbach, Scientific Reasoning: The Bayesian Approach (La Salle, IL:
Open Court, 1989), pp. 284 ff.
But now shouldn’t we say that the definition just proposed is, in fact,
much more intuitive than the standard one which is stated in terms of a
ratio of ‘absolute’ probabilities? (A defender of the latter definition cannot,
as we have just seen, complain that the new definition has different and
therefore unacceptable implications.) But if that is so, then, precisely
because the new definition uses a conditional in its statement, one cannot
appeal to the notion of conditional probability thus defined in order to
explicate conditionals: the notion of conditionality is the conceptually
prior notion. My charge, thus, is that an intuitively plausible explication
of the concept of conditional probability presupposes the very concept of
conditionality which Edgington aspires to explicate in terms of conditional
probability. (Here it may be asked why an explication of ‘if’ which uses ‘if’
should be any more problematic than the Tarskian truth-condition for ‘A
and B’ which uses ‘and’, namely: ‘A and B’ is true iff A is true and B is true.
My answer is that the latter is unproblematic precisely because, as one of
the base clauses of an inductive truth-definition, it does not pretend to
provide an explication of the concept of conjunction.)
Edgington may object, however, that when probability is interpreted as
degree of belief, it is not always the case that the conditional probability of
B given A, as determined by the standard ratio, is equal to the probability
that B would have if the probability of A were 1. For example,13 she may say
that I have a high degree of conditional belief that, given that the CIA are
bugging my office, I won’t know about it – and yet, if I were to know for
sure (have a degree of belief ¼ 1) that the CIA were bugging my office,
I would have a zero degree of belief that I didn’t know about it. However,
if this really is so, then one might say, for this very reason, that degrees of
belief cannot be interpreted in terms of the probability calculus – that their
formal properties do not conform to those of any kind of probability: and
that would compromise Edgington’s entire approach to conditionals and
conditional beliefs. (One might say this, in particular, if one held that the
‘Bayesian’ principle of ‘conditionalization’ is central to the project of
interpreting degrees of belief in terms of the probability calculus: for,
exactly to the extent that the CIA case looks like a counterexample to
my proposed definition of conditional probability, (6), it looks like a
counterexample to the principle of conditionalization as well.) On the
other hand, however, the supposed counterexample can itself be chal-
lenged, on the grounds that it depends, illicitly, upon an epistemically
irrelevant indexical (first-person) characterization of the belief in question.
13
Compare Edgington, ‘On Conditionals’, p. 269.
192 Conditionality
Clearly, I have a high degree of conditional belief that, given that the CIA
are bugging my office, I won’t know about it, simply because I have a high
degree of conditional belief that, given that the CIA are bugging any
ordinary citizen’s office, that citizen won’t know about it – for I don’t
consider myself to be any exception in this regard. However, if I were to
know for sure that the CIA were bugging some ordinary citizen’s office,
I would indeed have a very high degree of belief that that citizen wouldn’t
know about it – even if that citizen happened to be myself. So it seems that
the case in question, when properly unpacked, does not really constitute a
counterexample to the claim at issue.14
Let me sum up my conclusions so far. I have proposed an intuitively
appealing definition of ‘conditional probability’, according to which the
conditional probability of B given A is the probability that B would have if
A had a probability of 1. (At the end of the present chapter, however, I shall
propose a small modification to this definition, which renders superfluous
my response to the problem case discussed in the preceding paragraph.)
Now, clearly, it would be circular to use the notion of conditional prob-
ability, thus defined, in order to explicate the notion of a conditional
statement or a conditional belief, because the definition itself employs a
conditional. Furthermore, while it is true that the value of the conditional
probability of B given A, as determined by this definition, will equal the
value of that probability as determined by the standard definition of
conditional probability (the ratio-based definition) – assuming, at least,
that we update probabilities according to the principle of
‘conditionalization’ – nonetheless, there are compelling reasons for think-
ing that the standard ‘definition’ cannot properly qualify as a definition
(that is, explication of the meaning) of the term ‘conditional probability’,
as opposed merely to providing a correct measure of the value of a condi-
tional probability. Nor is it plausible to suppose that the notion of
conditional probability is just primitive, standing in need of no definition
or explication.15 I conclude that we do need to define conditional probabil-
ity and have no real option but to define it in a way which already takes the
notion of a conditional statement for granted – and hence that Edgington’s
attempt to explicate conditional statements and beliefs in terms of condi-
tional probability puts the cart before the horse.
14
For another way of handling such cases, see D. H. Mellor, ‘How to Believe a Conditional’, Journal
of Philosophy, 90 (1993), pp. 233–48, 243.
15
Pace Edgington, ‘On Conditionals’, p. 270.
4 adams’s thesis
However, whether or not one can explicate conditionals in terms of
conditional probability, it is a further question whether or not Adams’s
thesis is correct, this being the thesis that the probability of ‘If A, then B’
equals the conditional probability of B given A – where probability is
construed as degree of (rational) belief.16 Stated using Edgington’s sym-
bolism, we have:17
Adams: b(If A, then B) ¼ b(B|A) ¼ b(A & B)/b(A), provided that b(A) > 0
Of course, given Lewis’s proof, Adams’s thesis can’t easily be true unless
conditionals lack truth-conditions; that is, do not express propositions –
although one can get around this in various arcane ways, such as by
interpreting conditional sentences as being strongly context-dependent.18
But that is a consequence which Edgington is happy to embrace. However,
it seems plain enough from examples that Adams’s thesis is at least highly
questionable. The lottery example is perhaps the clearest way of making
the point.19 The conditional probability that I shall not win the lottery,
given that I buy just one ticket out of a hundred thousand on sale, is
plainly very high – that is (in my terms), if the probability of my buying
just one ticket were 1, the probability of my not winning would be very
high. Does it follow that I should assign a high probability (degree of
belief) to ‘If I buy just one ticket, I shall not win the lottery’? Remember,
we can make the odds as high as we like, without altering the nature of the
example. If Edgington is right, there must be odds which would justify me
in believing this conditional with as high a degree of belief as I assign to
pretty well any of my very firmest beliefs, such as that the sun will rise
tomorrow. But even at these odds, I suggest, someone purchasing just one
lottery ticket would in fact deny that he believed that he would lose if he
bought just one ticket. A very strong belief that one will lose if one
buys a ticket ought to be sufficient to deter one from buying a ticket.
After all, a very strong belief that one will win if one buys a ticket
16
See Ernest Adams, The Logic of Conditionals (Dordrecht: Reidel, 1975), p. 3.
17
See Edgington, ‘On Conditionals’, p. 263, where she calls this simply ‘The Thesis’.
18
See Edgington, ‘On Conditionals’, pp. 305 ff. In Chapter 9, I urge that conditionals are context-
dependent, but not in a way that would automatically give comfort to those who favoured an appeal
to their context-dependency in order to evade the implications of Adams’s thesis.
19
In my ‘The Truth about Counterfactuals’, Philosophical Quarterly 45 (1995), pp. 41–59 (see also
Chapter 9 of the present book), I used this example to make a point about assertibility, which
enabled Edgington to parry it in her ‘On Conditionals’, p. 287 n. But really the example works
equally well with regard to degree of belief.
194 Conditionality
should suffice to induce one to buy a ticket – and so, by the same token,
an equally strong belief that one will lose if one buys a ticket should
suffice to induce one not to buy one. On Edgington’s theory, a
purchaser of just one ticket ought to have the latter belief, and so
ought not to buy a ticket – and yet people do buy lottery tickets, and so
either we must charge them with irrationality or else we must deny that
they ought to have the belief which Edgington’s theory says that they
ought to have.
Against me, it may be said that what is to be gained by winning is
much greater than what is to be lost by losing, so that the two cases are
not parallel. However, we could adjust the example and compare the
original lottery case with a gamble in which there is a very high condi-
tional probability of winning a small amount – say, equal to the price of
the stake – given that one takes the gamble. In this case, according to
Edgington, one has a very high degree of belief that one will win a small
amount if one plays, and this, it would seem, should suffice to induce
one to play. By the same token, then, if one has an equally strong belief
that one will lose a similarly small amount if one plays, this should
suffice to induce one not to play (and this, by Edgington’s account, is
what obtains in the original lottery case). Yet lottery ticket buyers don’t
refrain from playing, and so must either be convicted of irrationality
or else cannot be credited with a very high degree of belief that they
will lose if they play. (What they do believe, of course, and with
complete conviction, is that if they were to play, then the probability
of their losing would be very high: but that, in my view, is quite another
matter.)
Note here that it will not help Edgington to point out that, in the
original lottery case, one knows that there is a very small but finite chance
of winning a great deal of money if one buys a ticket. Of course, I can
acknowledge that fact and use it to explain why it can be rational to buy a
ticket. But ‘If I buy a ticket, there will be a small but non-zero chance of
my winning a large sum’ entails ‘If I buy a ticket, I may win’ – and this is
incompatible with ‘If I buy a ticket, I shall not win’. Edgington, however,
says that I ought to believe the latter very strongly, so she cannot allow that
I should also believe the former very strongly. In confirmation of my claim
that these two conditionals are incompatible, consider the following
imaginary conversation between X and Y:
X: If you buy a ticket, you will not win.
Y: Yes I may – somebody has to win and it could be me.
Clearly, Y here is disagreeing with X. Of course, ‘Not A’ is not, in general,
incompatible with ‘Possibly A’: but this just goes to show that ‘If you buy a
ticket, you will not win’ and ‘If you buy a ticket, you may win’ – which are
incompatible, as the disagreement between X and Y indicates – cannot be
construed as a pair of conditionals which differ only in that one has a
consequent of the form ‘Not A’ and the other a consequent of the form
‘Possibly A’.20
Besides the lottery case, there are other apparent counterexamples to
Adams’s thesis, such as the following one concerning coin-spinning.21
Suppose I assert, with complete and justifiable conviction, ‘If you spin
this coin, it will land heads’. That would be an appropriate thing to assert if
I knew for sure that the coin was a two-headed one, or otherwise heavily
biased to land heads. Suppose then that I knew, to the contrary, that the
coin was perfectly fair: in that case, I surely ought to deny what I previously
asserted, and with just as much conviction. However, according to Edging-
ton’s account, I ought instead to be indifferent in the latter case between
the conditional ‘If you spin this coin, it will land heads’ and ‘If you spin
this coin, it will land tails’ – I should (by Adams’s thesis) have the same
degree of belief (neither high nor low) in each of them, because for a fair
coin the conditional probability of its landing heads given that it is spun
equals the conditional probability of its landing tails given that it is spun (a
probability of 50%, or 0.5, in each case). But if a high degree of belief in ‘If
this coin is spun, it will land heads’ expresses a strong conviction that the
coin is heavily biased to land heads, then a middling degree of belief in that
same conditional should apparently express a moderate conviction that the
coin is heavily biased to land heads – whereas in fact what one has in the
hypothesized circumstances is a strong conviction that the coin is unbiased.
We can restate this argument more formally as follows, letting B be the
proposition ‘This coin is heavily biased to land heads’, S the proposition
‘This coin is spun’ and H the proposition ‘This coin will land heads’. Now,
it is plausible to claim that one’s degree of belief in ‘If S, then H’ should be
proportionate to one’s degree of belief in B, being high when the latter is
high and low when the latter is low. (Indeed, some philosophers would
make the even stronger claim that the disposition statement B is analytic-
ally equivalent to the conditional ‘If S, then H’, in which case one’s degree
of belief in each should certainly be the same.) However, according to
20
See my ‘The Truth about Counterfactuals’, p. 47 (and also Chapter 9 of the present book).
21
Compare A. Hájek and N. Hall, ‘The Hypothesis of the Conditional Construal of Conditional
Probability’, in Eells and Skyrms (eds.), Probability and Conditionals.
196 Conditionality
Adams’s thesis, one’s degree of belief in ‘If S, then H’ ought to be (approxi-
mately) equal to one’s degree of belief in ‘If S, then not-H’, when one has a
high degree of belief that the coin is unbiased and in consequence has a low
degree of belief in B. Moreover, these equal degrees of belief must sum to
unity, according to Adams’s thesis, and so cannot both be low. Therefore,
according to Adams’s thesis, when one has in such circumstances a low degree
of belief in B, one should not have a low degree of belief in ‘If S, then H’,
contrary to the plausible claim made previously. Hence Adams’s thesis must
be mistaken if that claim is correct. In confirmation of its correctness and in
further disconfirmation of Adams’s thesis, observe that it is in fact perfectly
possible for a rational subject to have low degrees of belief in both ‘If S, then H’
and ‘If S, then not-H’ – to wit, when the subject strongly believes the coin to
be unbiased. In such circumstances, the subject will rightly reject both of the
conditionals, ‘If this coin is spun, it will land heads’ and ‘If this coin is spun, it
will not land heads’, in favour of the conditional ‘If this coin is spun, it may or
may not land heads’.
5 edgington’s response
Earlier in this chapter,22 I raised some questions concerning the standard
ratio-based definition of conditional probability. Recall that, according to
that definition, the conditional probability of B given A, written ‘p(B|A)’,
or alternatively ‘pA(B)’, is defined as follows:23
(7) p(B|A) ¼df p(A & B)/p(A), provided that p(A) > 0
One question that I raised was the following: what entitles us to suppose
that the expression ‘p( | )’, as defined by (7), really signifies any kind of
probability? Of course, it is easily shown that the value of p(B|A) must lie
between 0 and 1 and thus within the numerical value range of a probabil-
ity. But so, too, may the values of many other functions lie within this
range. An absolute probability function, signified by an expression such as
‘p( )’, is a function of just one argument – that argument being a propos-
ition – with a numerical value between 0 and 1. In (7), however, we are
purportedly introduced to a different kind of probability function, which
is a function of two arguments – both of them propositions – with a
22
And in the original paper on which those parts of this chapter are based, my ‘Conditional
Probability and Conditional Beliefs’, Mind 105 (1996), pp. 603–15.
23
(7) differs from (1) above only in that I have made it explicit that it is supposed to be a definition and
have included in it the proviso that p(A) > 0.
numerical value between 0 and 1. What is such a probability supposed to
be a probability of ? Not the probability of a proposition, clearly, since the
function takes not a single proposition but a pair of propositions as its
arguments. The answer will be offered that such a probability is simply the
conditional probability of one proposition, B, given another proposition, A.
Definition (7), however, throws no light at all on what is meant by saying
this, beyond telling us that it is a way of talking about the ratio between
the (absolute) probabilities of two propositions, (A & B) and A. Why call
this ratio a ‘conditional’ probability?
Dorothy Edgington thinks that she has an answer to this question. This
is what she says:
The question Lowe raises . . . is a fair one, and it has a straight answer. Let us use
the notation ‘pA(B)’ for ‘the probability of B given A’. Take any law of probability,
any consequence of the axioms of probability theory, e.g.: p(~B) ¼ 1–p(B); p(B ˅
C) ¼ p(B) þ p(C) – p(B & C). Add the standard definition of conditional
probability, and we can prove a parallel law: pA(~B) ¼ 1–pA(B); pA(B ˅ C) ¼
pA(B) þ pA(C) – pA(B & C), etc. The probability of B given A, on the standard
definition, deserves a name which contains the word ‘probability’ because, if p( ) is
a probability function in which p(A) 6¼ 0, then pA( ) [or: p( |A)] is a probability
function, according to the axioms.24
Now, as we have already observed, p( | ), as defined by (7), is a function of
two arguments, whereas the ‘probability functions’ with which the stand-
ard axioms of probability are concerned are functions of just one argument.
Consequently, the mere fact that, when definition (7) is added to those
axioms, one can prove various laws governing p( | ) which ‘parallel’
(Edgington’s expression) the laws governing standard probability
functions can by no means imply that p( | ) itself is such a function, for
the very simple reason that p( | ) is not a function of just one argument.
But, given that p( | ) cannot qualify as a probability function in the
standard sense, why should a mere parallelism between the laws of p( | )
and the laws of probability suffice to warrant our calling p( | ) a ‘probabil-
ity’ function in any sense? I shall return to this question in the next section,
where I shall try to show that, in fact, such a parallelism provides a very
weak reason indeed to think of ‘p( | )’ as signifying any kind of probability.
Before proceeding, however, it is important to appreciate that Edging-
ton’s claim in the passage just quoted is not that p( | ) itself is a probability
function, but rather that ‘if p( ) is a probability function in which
24
Dorothy Edgington, ‘Lowe on Conditional Probability’, Mind 105 (1996), pp. 617–30: see p. 620.
I have adapted her logical symbolism very slightly to bring it into conformity with my own.
198 Conditionality
p(A) 6¼ 0, then pA( ) [or: p( |A)] is a probability function, according to the
axioms [of probability theory]’. (Note that ‘[or: p( |A)]’ here is Edgington’s
own wording, not my interpolation.) What exactly are we to make of this?
Presumably, Edgington considers that ‘pA( )’, or ‘p( |A)’, like ‘p( )’,
expresses a probability function of just one argument – as it were, the
‘given A probability’ of a proposition. To see the situation in the clearest
possible light, let us recast (7) using Edgington’s own preferred subscript
notation, ‘pA( )’, as follows:
(7*) pA(B) ¼df p(A & B)/p(A), provided that p(A) > 0
We could then summarize matters as follows. The standard definition,
(7*), treats the ‘A’ in ‘pA(B)’ as occupying one of two argument-places in a
two-place functional expression. But it seems that, in the quoted passage,
Edgington implicitly treats the ‘A’ in ‘pA(B)’ as being, rather, an index
which distinguishes the putative one-argument probability function, pA( ),
from another such probability function, p( ) – the latter being a familiar
absolute probability function. Consequently, it may be objected that the
standard definition, (7*), does not in fact serve to define what Edgington
takes herself to mean by the expression ‘pA(B)’.
Here, however, it may be pointed out that, quite generally, if f(x, y) is a
function of two arguments, x and y, then we may define in terms of it a
related function of just one argument, fa(y), by letting x have a constant
value, x ¼ a. And this, it may be said, is all that Edgington is implicitly
doing, quite innocuously. Thus it may be conceded that, strictly speaking,
(7*) doesn’t itself serve to define ‘pA( )’, as Edgington uses this expression,
but it may also be urged that it is a simple matter to construct such a
definition from (7*) in the foregoing manner. However, in that case, it
would seem that pA( )’s entitlement to be called a ‘probability’ function
rests squarely on p( | )’s own entitlement to be so called, since we are now
understanding the former to be defined in terms of the latter. So now, by a
somewhat circuitous route, we are led back to our earlier question whether
a mere parallelism between the laws of p( | ) and the laws of probability
suffices to warrant our calling p( | ) a ‘probability’ function in any sense.
6 conditional probability and conditional

truth-value
At this point, I think it is instructive to explore an analogy between
probability and truth-value. For this will enable us to see, amongst other
things, why a mere parallelism between the laws of probability and those of
‘conditional probability’, as defined by (7) or (7*), does not in fact suffice
to warrant a description of the latter as being a kind of probability.
Let t( ) be a function which takes a single proposition as argument and
has one of two numerical values, namely, the value 1 just in case the
proposition in question is true and the value 0 just in case the proposition
in question is false. We may read the functional expression ‘t(A)’ as ‘the
truth-value of A’. Now, by analogy with definition (7*), let us formulate
the following putative definition of what we may be tempted to call the
conditional truth-value of B given A:
(8) tA(B) ¼df t(A & B)/t(A), provided that t(A) 6¼ 0
We see that, according to (8), the ‘conditional truth-value of B given A’ is,
quite simply, the ratio of the truth-value of (A & B) and the truth-value of
A, provided that the latter is not 0. Of course, if t(A) is not 0, then it must
be 1. Accordingly, we see that tA(B) is a function which takes two propos-
itions, A and B, as arguments and has as its value the truth-value of the
conjunction (A & B) if A is true, and is otherwise undefined (that is, if A is
false). So, where it is defined, tA(B) takes the same value as t(B).
Indeed, we can take the analogy between ‘conditional truth-value’ and
‘conditional probability’ even further. As we saw earlier, Edgington herself,
following Ernest Adams,25 has proposed that the probability of a condi-
tional, ‘B if A’, just is the conditional probability of B given A (provided
that p(A) 6¼ 0):26
(9) p(B if A) ¼ pA(B)
As we noted earlier, David Lewis has famously shown that, on pain of
triviality, there is no proposition that ‘B if A’ can be taken to express, if (9) is
correct.27 Edgington happily embraces this consequence of (9), and con-
tends that conditionals have neither truth-conditions nor, hence, truth-
values (with certain exceptions to be mentioned in a moment). But
suppose that one were to propose, by analogy with (9), the following:
(10) t(B if A) ¼ tA(B)
This states that the truth-value of a conditional, ‘B if A’, is simply the
conditional truth-value of the consequent, B, ‘given’ the antecedent, A.
25
See again Adams, The Logic of Conditionals, p. 3.
26
See Edgington, ‘On Conditionals’, p. 263. See further section 4 of the present chapter, on Adams’s
thesis.
27
See again Lewis, ‘Probabilities of Conditionals and Conditional Probabilities’.
200 Conditionality
Now, paralleling Lewis’s triviality proof, it is easy to show that there is no
proposition that ‘B if A’ can be taken to express, if (10) is correct. This is
because, for any proposition X, t(X) must either take the value 1 or else the
value 0: but ‘t(B if A)’, according to (8) and (10), denotes a value only
provided that t(A) is not 0, and is otherwise undefined. Hence t(B if A)
cannot be the truth-value of any proposition. In fact, ironically enough,
I imagine that Edgington herself ought to feel quite sympathetically
towards (10), since she herself allows that the conditional ‘B if A’ has the
truth-value of B provided that A is true, but otherwise lacks a truth-value.28
And, indeed, there is a quite long-standing tradition which sees the indica-
tive ‘if’ as having what is sometimes called a ‘defective’ truth-table,
whereby it takes the truth-value of the consequent when the antecedent
is true, but otherwise lacks a truth-value.
Now, all of this may be independently quite interesting, but we need to
bring matters back to the point. The point was that Edgington suggests
that conditional probability, as defined by (7) or (7*), deserves the name
‘probability’ because the laws of conditional probability parallel those of
ordinary, ‘absolute’ probability. Now, however, it is easy enough to see
that, by the same token, the laws of ‘conditional truth-value’, where the
latter is taken as being defined by (8), parallel the laws of ordinary,
‘absolute’ truth-value. For example, one ‘absolute’ law is that if t(B) ¼ 1
then t(~B) ¼ 0, and another is that t(B ˅ C) ¼ 1 iff either t(B) ¼ 1 or
t(C) ¼ 1, and paralleling these we have the ‘conditional’ laws that if tA(B) ¼ 1,
then tA(~B) ¼ 0, and that tA(B ˅ C) ¼ 1 iff either tA(B) ¼ 1 or tA(C) ¼ 1.
What, then, are we entitled to say about the notion of ‘conditional truth-
value’, as purportedly defined by (8)? If Edgington’s line of argument (as
quoted earlier) is correct, the conclusion should be that we are entitled to
think of ‘conditional truth-value’ as a kind of truth-value, the truth-value of
one proposition ‘given’ another proposition. But this, I submit, is mani-
festly absurd. There is surely not, in addition to ordinary, ‘absolute’
truth-value – truth or falsehood simpliciter – a peculiar kind of ‘relative’
truth-value. ‘tA(B)’, as defined by (8), cannot be taken to denote a new
kind of truth-value. If the expression ‘tA(B)’ is to be espoused at all, it is
better explained, rather, as denoting the (ordinary) truth-value that B
has if A is true. This, however, is to explain the notion of ‘conditional
truth-value’ in conditional terms. And my contention is that, in like
manner, we can only make clear sense of the notion of ‘conditional
probability’ if we attempt to explain it, too, in conditional terms – not,
28
See Edgington, ‘On Conditionals’, pp. 290–1.
that is, as a new kind of probability, but rather as the (ordinary) probability
that a proposition has if certain conditions obtain. In short: talk
about conditional probability is properly construed not as talk about a
conditional kind of probability, but rather as talk of a conditional kind
about probability.
7 a new conditional definition of conditional

probability
Edgington, it must be pointed out, quite explicitly denies that the notion
of conditional probability can be adequately explicated or defined in
conditional terms. (Understandably so, for to concede this would be to
concede that the notion of a conditional judgement is fully intelligible
independently of the notion of conditional probability.) The definition
that I myself proposed earlier was, in essence, this:29
(11) For any proposition B, pA(B) is the probability that B has if a
probability of 1 is assigned to A
Concerning this definition, Edgington maintains, in the light of various
putative difficulties which she raises for it, that it is ‘inadequate, and
beyond repair’.30 However, it is of considerable interest to note how,
according to Edgington herself, the probability function pA( ) relates to
the probability function p( ), namely as follows:31
Start with p( ). Now assign zero probability to all the possibilities in which ~A is
true – assign probability 1 to A; keep the relative probabilities of the possibilities in
which A is true the same as before; and you have pA( ).
But what is said here appears to be fully captured by the following
conditional statement, which I now propose as a definition replacing (11):
(12) For any proposition B, pA(B) is the probability that B has if a
probability of 1 is assigned to A and, for any propositions C and D
which entail A, the ratio p(C)/p(D) is left unaltered in value
Of course, the propositions (A & B) and A are themselves propositions
which entail A, so that it is implicit in the antecedent of (12) that the ratio
p(A & B)/p(A) is to be left unaltered in value. This enables us to see why the
standard definition of pA(B), (7*), assigns it precisely the same value that
29 30
See (6) in section 3 above. See Edgington, ‘Lowe on Conditional Probability’, p. 625.
31
Edgington, ‘Lowe on Conditional Probability’, p. 620.
202 Conditionality
(12) assigns to pA(B) – that is, why the standard definition is extensionally
correct. Why? Because if the antecedent of (12) is realized – so that A is
assigned a probability of 1 – B and (A & B) will then both have the same
probability; and the probability that (A & B) then has will equal the ratio
between that probability and the probability that A then has, since this last
probability will just be 1; but, ex hypothesi, the ratio in question will be left
unaltered in value from its original value; consequently, if the antecedent
of (12) is realized, B will then have a probability equal to the value of the
ratio of the probabilities originally assigned to (A & B) and A – which is
precisely the value assigned to pA(B) by (7*).
All that remains to be asked is what grounds Edgington could possibly
have for refusing to allow that (12) constitutes precisely the ‘repair’ which
she claimed (11) to be ‘beyond’. Plainly, she cannot object that (12) assigns
a different value to pA(B) from that assigned to it by (7*), which she
accepts, since I have just proved otherwise. Moreover, as we have just seen,
(12) has the virtue of explaining precisely why (7*) assigns the correct value
to pA(B).
8 concluding remarks on conditionality

and other logical notions
In this and the previous chapter I have tried to set out and justify my
current views concerning the semantics and logic of conditionals, both
indicative and subjunctive, partly offering direct defences of my own chief
claims and partly presenting arguments against the main rival views. My
view has what I take to be the attractive feature of providing a unitary logic
of conditionals, both indicative and subjunctive, in virtue of assigning the
same fundamental logic form to conditionals of both kinds, while at the
same time allowing there to be a genuine semantic distinction between
them, explicable by appeal to the different types of modality in terms of
which propositions of this logical form may be interpreted – ‘alethic’
modality in the case of subjunctive or counterfactual conditionals, and
‘epistemic’ modality in the case of indicative conditionals. Thus, ‘if’ is not
fundamentally ambiguous, according to my view. I take it to be an
additional virtue of my view that it provides a way of analysing the notion
of conditionality in terms of the notion of modality, thus reducing the
number of fundamental logical notions that we need to recognize. In other
words, it turns out, if I am correct, that not all of the five core logical
notions examined in this book – reference, predication, identity, modality,
and conditionality – are fundamental. Of course, I have also proposed, in
Chapter 8, an account of metaphysical modality which holds that truths
regarding what is metaphysically necessary or possible are grounded in
truths about essence, where the notion of essence is understood in a
distinctively Aristotelian way. However, I would not want to represent
this view as maintaining that the notion of metaphysical modality is
reducible to that of essence and, in any case, I have already made it clear
that I want to acknowledge types of modality other than metaphysical – or,
more generally, alethic – modality, notably what I have hitherto called
epistemic modality. The family of modal notions is, then, much wider, in
my view, than the ambit of the notion of essence, which serves as a ground
only for modal truths of one specific variety, the metaphysical. Hence I am
happy to acknowledge the general notion of modality as being a funda-
mental logical notion, unlike the notion of conditionality.
The notions of reference and predication look likewise to be fundamen-
tal. In particular, the idea that predication can be reduced to identity – in
other words, that every predicative proposition is really an identity prop-
osition, affirming the identity of ‘one’ thing with ‘another’, even if only
their ‘partial’ identity, whatever that can exactly be taken to mean – seems
to me hopeless, although some philosophers have espoused it. According
to this view, to affirm, for instance, that Mars is red is to affirm a –
presumably only ‘partial’ – identity between Mars and redness, a certain
particular and a certain universal. But I do not believe that such a view can
hope to capture all the distinctions between types of predication that were
discussed in Chapters 3 and 4. Rather, I see predications of identity as
being just one of these types of predication amongst others. This is not to
imply, however, that the notion of identity is, conversely, reducible to that
of predication. As I see it, identity is an irreducible and indefinable formal
relation, unique amongst relations in being truly predicable of anything
whatever and that thing itself and never truly predicable of any one thing
and anything else. As can be seen from this very characterization of identity,
we have to presuppose the notion of identity even to say what it is. For the
very words ‘itself’ and ‘else’ imply identity and difference, or non-identity.
Equally hopeless, I should say, would be the idea that the notion of
reference is reducible to one or more other logical notions, such as the
notions of predication and identity. Some philosophers might suppose that
the statement ‘Mars is red’, in which reference is seemingly made to the
planet Mars, is somehow reducible to a statement affirming that redness,
the universal, is identical with one of a set of universals, namely all those
universals that we would ordinarily think of as being exemplified by Mars
and, collectively, only by Mars. This would be to endorse the so-called
204 Conditionality
‘bundle’ theory of particulars, identifying each particular with a ‘bundle’ of
universals. But this theory is unsatisfactory for reasons too well known to
be discussed here. In any case, this line of thought, even if successful in the
case of reference to particulars, such as Mars, would not provide us with a
way of eliminating or reducing reference to universals. And the reverse
strategy of doing away with universals in favour of particulars, as nominal-
ism proposes, would conversely still leave reference to particulars on our
hands. A more promising strategy might seem to be to try to eliminate
reference in favour of predication and quantification, by construing ‘Mars
is red’ as meaning ‘There is exactly one thing that is Mars and it is red’, in
line with Russell’s theory of descriptions. However, the predicate ‘– is
Mars’ ostensibly means ‘– is identical with Mars’, in which reference is
again apparently made to Mars. This could be avoided by resorting once
more to the bundle theory, but we have already seen that this still leaves
reference to universals on our hands.
My tentative conclusion, then, is that of the five logical notions that
have been the central topics of this book, those of reference, predication,
identity, and modality are truly irreducible and fundamental in character.
To these I would add the notions of negation, existence, and truth, men-
tioned in the Preface, and also the notion of generality, as expressed by the
quantifiers ‘all’ and ‘some’. But these are subjects for another occasion.
Bibliography
Adams, E. W. The Logic of Conditionals (Dordrecht: Reidel, 1975).

‘Subjunctive and Indicative Conditionals’, Foundations of Language 6 (1970),
pp. 89–94.
Aristotle, Categories and De Interpretatione, trans. J. L. Ackrill (Oxford: Clarendon
Press, 1963).
Armstrong, D. M. A Combinatorial Theory of Possibility (Cambridge University
Press, 1989).
What Is a Law of Nature? (Cambridge University Press, 1983).
Austin, J. L. ‘Ifs and Cans’, reprinted in his Philosophical Papers, 2nd edn (Oxford
University Press, 1970).
Bennett, J. ‘Farewell to the Phlogiston Theory of Conditionals’, Mind 97 (1988),
pp. 509–27.
Bower, T. G. Development in Infancy (San Francisco: Freeman, 1974).
A Primer of Infant Development (San Francisco: Freeman, 1977).
Brody, B. Identity and Essence (Princeton, NJ: Princeton University Press, 1980).
Bryant, C. ‘Conditional Murderers’, Analysis 41 (1981), pp. 209–15.
Davidson, D. ‘The Individuation of Events’, in his Essays on Actions and Events
Dudman, V. H. ‘Conditional Interpretations of “If”-Sentences’, Australian
Journal of Linguistics 4 (1984), pp. 143–204.
‘Indicative and Subjunctive’, Analysis 48 (1988), pp. 113–22.
‘On Conditionals’, Journal of Philosophy 91 (1994), pp. 113–28.
Dummett, M. Frege: Philosophy of Language, 2nd edn (London: Duckworth, 1981).
The Interpretation of Frege’s Philosophy (London: Duckworth, 1981).
Edgington, D. ‘Do Conditionals Have Truth-Conditions?’, Critica 18 (1986),
pp. 3–30, reprinted in Jackson (ed.), Conditionals.
‘Lowe on Conditional Probability’, Mind 105 (1996), pp. 617–30.
‘On Conditionals’, Mind 104 (1995), pp. 235–329.
Eells, E. and Skyrms, B. (eds.), Probability and Conditionals (Cambridge
University Press, 1994).
Evans, G., ‘Can There Be Vague Objects?’, Analysis 38 (1978), p. 208.
Fine, K. ‘Essence and Modality’, in James E. Tomberlin (ed.), Philosophical
Perspectives, 8: Logic and Language (Atascadero, CA: Ridgeview, 1994).
Fodor, J. A. The Language of Thought (New York: Crowell, 1975).
205
206 Bibliography
Frege, G. ‘Function and Concept’, in Translations from the Philosophical Writings
of Gottlob Frege.
Die Grundlagen der Arithmetik [1884], translated as The Foundations of
Arithmetic by J. L. Austin (Oxford: Blackwell, 1953).
‘On Concept and Object’, in Translations from the Philosophical Writings of
Gottlob Frege.
Translations from the Philosophical Writings of Gottlob Frege, 2nd edn, ed. and
trans. P. T. Geach and M. Black (Oxford: Blackwell, 1960).
French, S. and Krause, D. Identity in Physics: A Historical, Philosophical, and
Formal Analysis (Oxford: Clarendon Press, 2006).
Geach, P. T. Reference and Generality, 3rd edn (Ithaca, NY: Cornell University
Press, 1980).
Grice, H. P. ‘Indicative Conditionals’, in his Studies in the Way of Words
(Cambridge, MA: Harvard University Press, 1989).
Hájek, A. and Hall, N. ‘The Hypothesis of the Conditional Construal of
Conditional Probability’, in Eells and Skyrms (eds.), Probability and
Conditionals.
Howson, C. Logic with Trees (London: Routledge, 1997).
Howson, C. and Urbach, P. Scientific Reasoning: The Bayesian Approach (La Salle,
IL: Open Court, 1989).
Hunter, G. ‘The Meaning of “If” in Conditional Propositions’, Philosophical
Quarterly 43 (1993), pp. 279–97.
Jackson, F. Conditionals (Oxford: Blackwell, 1987).
(ed.), Conditionals (Oxford University Press, 1991).
Keefe, R. ‘Contingent Identity and Vague Identity’, Analysis 55 (1995), pp. 183–90.
Theories of Vagueness (Cambridge University Press, 2000).
Kripke, S. A. ‘Identity and Necessity’, in M. K. Munitz (ed.), Identity and
Individuation (New York University Press, 1971).
Naming and Necessity (Oxford: Blackwell, 1980).
LePore, E. and McLaughlin, B. (eds.), Actions and Events: Perspectives on the
Philosophy of Donald Davidson (Oxford: Blackwell, 1985).
Lewis, D. K. Counterfactuals (Oxford: Blackwell, 1973).
On the Plurality of Worlds (Oxford: Blackwell, 1986).
Parts of Classes (Oxford: Blackwell, 1991).
‘Probabilities of Conditionals and Conditional Probabilities’, Philosophical
Review 85 (1976), pp. 297–315, reprinted in Jackson (ed.), Conditionals.
‘Vague Identity: Evans Misunderstood’, Analysis 48 (1988), pp. 128–30.
Locke, J. An Essay Concerning Human Understanding [1690], ed. P. H. Nidditch
Lowe, E. J. ‘Conditional Probability and Conditional Beliefs’, Mind 105 (1996),
pp. 603–15.
‘Conditionals, Context and Transitivity’, Analysis 50 (1990), pp. 80–7.
The Four-Category Ontology: A Metaphysical Foundation for Natural Science
‘Identity, Individuality and Unity’, Philosophy 78 (2003), pp. 321–36.
Bibliography 207
‘Impredicative Identity Criteria and Davidson’s Criterion of Event Identity’,
Analysis 49 (1989), pp. 178–81.
‘Indicative and Counterfactual Conditionals’, Analysis 39 (1979), pp. 139–41.
‘Individuation’, in M. J. Loux and D. W. Zimmerman (eds.), The Oxford
Handbook of Metaphysics (Oxford University Press, 2003).
‘Jackson on Classifying Conditionals’, Analysis 51 (1991), pp. 126–30.
Kinds of Being: A Study of Individuation, Identity and the Logic of Sortal Terms
(Oxford: Blackwell, 1989).
‘Locke on Real Essence and Water as a Natural Kind: A Qualified Defence’,
Proceedings of the Aristotelian Society, Supplementary Volume 85 (2011),
pp. 1–19.
‘The Metaphysics of Abstract Objects’, Journal of Philosophy 92 (1995), pp. 509–
24.
More Kinds of Being: A Further Study of Individuation, Identity and the Logic of
Sortal Terms (Malden, MA and Oxford: Wiley-Blackwell, 2009).
‘Not a Counterexample to Modus Ponens’, Analysis 47 (1987), pp. 44–7.
‘Noun Phrases, Quantifiers and Generic Names’, Philosophical Quarterly 41
(1991), pp. 287–300.
‘On the Alleged Necessity of True Identity Statements’, Mind 91 (1982),
pp. 579–84.
‘On the Identity of Artifacts’, Journal of Philosophy 80 (1983), pp. 220–32.
‘One-Level versus Two-Level Identity Criteria’, Analysis 51 (1991), pp. 192–4.
The Possibility of Metaphysics: Substance, Identity, and Time (Oxford: Clarendon
Press, 1998).
‘Rationality, Deduction and Mental Models’, in K. Manktelow and D. E. Over
(eds.), Rationality (London: Routledge, 1993).
‘Reply to Noonan on Vague Identity’, Analysis 57 (1997), pp. 88–91.
‘A Simplification of the Logic of Conditionals’, Notre Dame Journal of Formal
Logic 24 (1983), pp. 357–66.
‘Sortals and the Individuation of Objects’, Mind and Language 22 (2007),
pp. 514–33.
Subjects of Experience (Cambridge University Press, 1996).
A Survey of Metaphysics (Oxford University Press, 2002).
‘The Truth about Counterfactuals’, Philosophical Quarterly 45 (1995), pp. 41–59.
‘Two Notions of Being: Entity and Essence’, in Robin Le Poidevin (ed.), Being:
Developments in Contemporary Metaphysics (Cambridge University Press,
2008).
‘Vague Identity and Quantum Indeterminacy’, Analysis 54 (1994), pp. 110–14.
‘What Is a Criterion of Identity?’, Philosophical Quarterly 39 (1989), pp. 1–21.
Mackie, J. L. Truth, Probability and Paradox (Oxford: Clarendon Press, 1973).
Mellor, D. H. ‘How to Believe a Conditional’, Journal of Philosophy, 90 (1993), pp.
233–48.
Noonan, H. ‘E. J. Lowe on Vague Identity and Quantum Indeterminacy’,
Analysis 55 (1995), pp. 14–19.
Nute, D. Topics in Conditional Logic (Dordrecht: Reidel, 1980).
208 Bibliography
Olson, E. T. ‘Material Coincidence and the Indiscernibility Problem’,
Philosophical Quarterly 51 (2001), pp. 337–55.
Parsons, T. Indeterminate Identity: Metaphysics and Semantics (Oxford: Clarendon
Press, 2000).
Plantinga, A. The Nature of Necessity (Oxford: Clarendon Press, 1974).
Putnam, H. ‘The Meaning of “Meaning”’, in his Mind, Language and Reality:
Philosophical Papers, Volume 2 (Cambridge University Press, 1975).
‘Why There Isn’t a Ready-Made World’, in his Realism and Reason:
Philosophical Papers, Volume 3 (Cambridge University Press, 1983).
Quine, W. V. ‘Existence and Quantification’, in his Ontological Relativity and
Other Essays (New York: Columbia University Press, 1969).
‘On What There Is’, in his From a Logical Point of View, 2nd edn (Cambridge,
MA: Harvard University Press, 1961).
‘Speaking of Objects’, in his Ontological Relativity and Other Essays.
Ramsey, F. P. The Foundations of Mathematics and Other Logical Essays (London:
Kegan Paul, 1931).
‘Truth and Probability’, in his The Foundations of Mathematics and Other
Logical Essays.
‘Universals’, in his The Foundations of Mathematics and Other Logical Essays.
Rosen, G. ‘Modal Fictionalism’, Mind 99 (1990), pp. 327–54.
Russell, B. Our Knowledge of the External World (London: George Allen and
Unwin, 1922).
‘Propositional Functions’, in his Introduction to Mathematical Philosophy
(London: George Allen and Unwin, 1919).
Salmon, N. Reference and Essence (Oxford: Blackwell, 1982).
Smith, B. ‘Against Fantology’, in M. E. Reicher and J. C. Marek (eds.), Experience
and Analysis (Vienna: HPT & ÖBV, 2005).
‘Of Substances, Accidents and Universals: In Defence of a Constituent
Ontology’, Philosophical Papers 26 (1997), pp. 105–27.
Stalnaker, R. C. ‘A Theory of Conditionals’, in Studies in Logical Theory, American
Philosophical Quarterly Monograph 2 (Oxford: Blackwell, 1968), reprinted in
Jackson (ed.), Conditionals.
Strawson, P. F. ‘Entity and Identity’, in H. D. Lewis (ed.), Contemporary British
Philosophy, Fourth Series (London: George Allen and Unwin, 1976),
reprinted in his Entity and Identity and Other Essays (Oxford: Clarendon
Press, 1997).
Individuals: An Essay in Descriptive Metaphysics (London: Methuen, 1959).
Suppes, P. Axiomatic Set Theory (New York: Dover, 1972).
Swinburne, R. An Introduction to Confirmation Theory (London: Methuen, 1973).
Tiles, J. E. Things That Happen (Aberdeen University Press, 1981).
Urbach, P. ‘What Is a Law of Nature? A Humean Answer’, British Journal for the
Philosophy of Science 39 (1988), pp. 193–210.
Van Fraassen, B. C. Laws and Symmetry (Oxford: Clarendon Press, 1989).
Von Plato, J. Creating Modern Probability (Cambridge University Press, 1994).
Bibliography 209
Wiggins, D. Sameness and Substance Renewed (Cambridge University Press, 2001).
Williams, B. Descartes: The Project of Pure Enquiry (Harmondsworth: Penguin,
1978).
Williamson, T. ‘Criteria of Identity and the Axiom of Choice’, Journal of
Philosophy 83 (1986), pp. 380–94.
Wittgenstein, L. Philosophical Investigations, trans. G. E. M. Anscombe (Oxford:
Blackwell, 1958).
Index
abstract objects, 15, 21, 54, 74, 92, 99, 109, 113, chemical compounds, 151
142, 146 conceptions, 97
activities, 32 adequate, 97–8, 101, 107
actual world, the, 142, 159, 177–8 concepts, 97–8, 107–9, 153–6
Adams, E. W., 199 conceptual truths, 153–4
Adams’s thesis, 184, 193, 195–6 conceptualism, 107–9, 116, 153–4
adherence, 102, 107, 113 concrete objects, 15, 75, 91, 109
aggregates, 100–1, 113 conditional connective, 169
amorphous lump, 89, 91, 93, 108 conditional logic, 7, 168–9, 180, 202
animals, 16–18, 23, 25, 29 conditional probability, 7, 166–7, 182–4, 187,
anti-essentialism, 145 189–90, 192–4, 198, 201
anti-realism, 108, 116, 146 ratio-based definition of, 184, 186, 191–2, 196
Aristotle, 5–6, 55–7 conditional truth-value, 199–200
Armstrong, D. M., 40, 44 conditionalization, 190–2
aspect, 31 conditionals, 6, 167–8, 182
assertibility, 167 Austinian, 172–3
assertibility-conditions, 165–6 context-sensitivity of, 163, 175–6, 179, 181
attention, 22–3 counterfactual, 163–4, 172, 175–8, 180
attributes, 32–3, 56–8, 60–1 Dutchman, 164, 173
axiom of extensionality, 14, 74 with impossible antecedents, 171
axiom of regularity, 76 indicative, 163–5, 167, 172, 183
logical form of, 171
Barcan–Kripke proof, 120, 123, 128–30 material, 165, 173, 182
Barcan–Kripke step, 129–31 possible-worlds interpretations of, 176–7
belief, 185, 188 subjunctive, 25, 163–4, 166, 183
degrees of, 184–5, 187–8, 191, 193, 195 transitivity of, 174, 178–80
Bower, T. G., 86, 88 consistency, 142, 158
box-arrow symbol, 169 constants, 34, 51
brittleness, 104 count nouns, 12
Brody, B., 73 counterfactual fallacies, 173
Butler, J., 146 counting, 12–13
criteria of application, 12
categorial concepts, 11, 19, 21, 23, 25, 27–8 criteria of identity, 2, 4, 11–20, 25, 27,
categorial ontology, 3, 31, 39, 55, 62 69–71, 73–4, 79, 84, 87, 89, 91, 94–5, 98,
categorial terms, 13–14, 16, 27 100–2, 115
categorial uniqueness, problem of, 39 impredicative, 69
categorialism, 2, 11–12, 19, 21–4, 26–30 primitive, 84, 87, 89–90
perceptual, 21 type-(A), 80–1, 93
category mistakes, 57, 61 type-(B), 75, 77–8, 80–2, 84, 92–3
causal powers, 112–15 criterial relations, 14, 16, 19, 77, 81–2
characterization, 31, 33, 35 cross-categorial terms, 34
210
Index 211
Davidson, D., 76–7, 81, 91 H2O, 147, 149–52
de Finetti, B., 184 hierarchies of subsumption, 13
definite descriptions, 28, 149–50, 204 hunks of matter, 17–18, 24–5, 29, 90
definitions, 80 Hunter, G., 169–70, 174, 176
demonstratives, 80, 85–7 hydrogen atoms, 112, 114
directions, 71–2, 75, 78, 80
dispositions, 32, 41, 104–5, 195 identity, 4–5, 14, 35, 37–8, 51, 60, 92, 95, 119, 146,
accidental, 43–4 150, 203–4
essential, 43 indeterminate, 134–7
idiosyncratic, 41–2 of indiscernibles, 73
Dretske, F., 40 necessity of, 119–20, 123, 128–9, 133, 150
Dudman, V. H., 164 non-contingency of, 128–34, 138
Dummett, M., 71, 84–8, 93, 108 personal, 72
primitive, 115
Edgington, D., 7, 183–9, 191–4, 197–202 relative, 95–6
electrons, 38, 112, 114–15, 135–6 substitutivity of, 132–4
empirical evidence, 155 vague, 120
empty set, 76 identity conditions, 5, 81, 92, 94, 103, 106, 115–16,
essence precedes existence, 99, 148–9, 152, 155, 157 141, 153
essences, 5–6, 56–7, 97, 99–102, 105–15, 144–5, indeterminacy, 121
147–8, 150–4, 156–7, 203 indexical expressions, 175
general, 97, 145 individual accidents, 56
individual, 98, 145 individual substances, 33, 39
real, 147 individuation, 11, 19, 52, 86–90, 93, 108
essential dependence, 114–15, 146 principles of, 19–20
essential truths, 153–4 individuative concepts, 94, 98, 107, 116
essentialism, 144 individuators, 20
serious, 144, 158, 160 inference problem, 40
Evans, G., 5, 119–24, 126–9, 131–2, 135–7 inference to the best explanation, 143
events, 65, 76, 81, 152 infinity, 62
exemplification, 31, 36–7 inherence, 56, 58, 60
dispositional, 36–7 innateness, 86, 88–9, 93
occurrent, 36 instantiation, 31, 33, 35, 40
existence, 4, 51, 54, 60, 64, 66, 116, 204 islands, 18
existential dependence
strong, 37–8 Jackson, F., 165
weak, 37–8 Jansen, L., 41
external world, the, 155
Kant, I., 4
facts, 121–2, 125, 157 kinds, 12, 14, 27, 32–3, 38, 40–1, 75, 104, 111
Fantology, 51, 54–5, 57 natural, 19, 149
feature-placing language, 86, 89, 93 Kripke, S. A., 5–6, 133, 147, 149
Fine, K., 6 Kuratowski, K., 96–8
forms of thought, 1
four-category ontology, the, 31–5, 37, 39, 45 lambda symbol, 122
Frege, G., 3–4, 51–5, 57, 70, 72–5, 78, 80–2 language of thought, 89
Fregean thesis, the, 70–1, 73, 78–9, 82–4, 87, Leibniz, G. W., 73
89–90, 93 Leibniz’s law, 95, 103, 122, 125, 127–8, 132, 134
functional expressions, 71–2 Lewis, D. K., 6, 113, 142, 158, 166, 177, 180, 183–4,
functions, 54 193, 199
fuzzy boundaries, 120–1, 135 life, 16, 18
sameness of, 16, 19
Geach, P. T., 74, 95 lines, 71–2
generality, 204 living organisms, 105
God, 148 Locke, J., 5, 12, 16–17, 70, 97, 100, 144, 146–7
212 Index
logic, 34, 119 perception, 21–2
formal, 55, 63, 134 persistence conditions, 104–6, 116
free, 130 individual, 106
laws of, 130–1 sortal, 11, 13–14, 18–19, 106
second-order, 57, 62 person-stages, 72
lotteries, 167, 193–4 phase changes, 19, 42
lumps, 100–7, 110, 113–14, 153 philosophical logic, 1
planets, 150
manifestations, 32 possible worlds, 6, 54, 110–11, 139–41, 144, 156, 158
Marcus, R. B., 5 abstractionism concerning, 142–3, 158–60
material composition, 103–5 accessibility relations between, 140
material constitution, 105–6, 109, 116 concretism concerning, 142–3, 157–8, 160
material implication, 182 fictionalism concerning, 143, 160
material objects, 16 similarity between, 177–9
mereological sums, 112–13 precisifications, 124, 137
metamorphosis, 14, 16, 18 predicables, 56–7, 63
metaphysical possibility, 98, 110–11, 115, 139, 151–2 predicate logic, 34, 50–1, 57, 141
modal combinatorialism, 143 predicates, 52–3, 98, 131, 133
modal logic, 6–7, 124, 139–40, 180 predications, 56, 58, 63–4, 203
modal operators, 171, 182 categorial, 3, 50, 63–5
modal primitivism, 144 dispositional, 3, 31
modality, 6, 119, 139, 156–7, 160, 166, 177, 203–4 formal, 2–3
alethic, 165, 172, 182, 202 material, 2
epistemic, 165, 172, 182, 202 occurrent, 3, 31
modes, 32–3, 37, 40, 56, 58, 60–1 primary substances, 52, 55–6, 58, 60–1, 72
modus ponens, 168, 170, 173 principle of charity, 179
mountains, 15, 17–18, 20, 121 principle of instantiation, the, 44–5
principles of composition, 112
naming, 29–30 probability calculus, 184, 186, 190–1
natural kind terms, 26–30, 149 probability functions, 196–8, 201
natural laws, 32, 40–2 proper names, 26–30, 51, 72, 79, 149
conditional, 42, 44–5 properties, 34, 51, 53–4, 57, 62, 122, 127, 130–3
natural necessitation, 40 property abstraction, 131, 133–4
NCI. See identity:non-contingency of propositional attitudes, 185, 188
necessary self-identity, 130–1 propositions, 142, 157, 159, 175–6, 183, 185, 193,
necessary truths, 130 197, 200
Nef, F., 43 atomic, 51, 60
negation, 204 protons, 112, 114
numbers, 79 Putnam, H., 108, 147, 149
objects, 32–3, 39, 51–4 quantification, 140–1, 204

Ockham, William of, 39 quantifiers, 4, 28, 51, 60
Oderberg, D. S., 41 restricted, 60
ontological categories, 2–3, 34, 37, 39, 50, 62–3, unrestricted, 65
145 quantities of matter, 100
ontological commitment, 51, 141, 148 quantum entanglement, 135
ontological dependency, 37, 39, 56, 72, 81 Quine, W. V., 4, 51–3, 141, 148
ontological square, the, 34, 36–9
ontology, 51 Ramsey, F. P., 39
pure, 65 rationality, 99, 115, 184
reference, singular, 2, 12, 26–8, 70, 80–5,
parallelism, 71–2 88–9, 203
paraphrase, 148 vague, 136
particulars, 31–2, 34, 57 Russell, B., 1, 3–4, 28, 51, 55, 143, 204
bundle theory of, 204 Russell’s paradox, 52, 98–9
Index 213
scepticism, 155 tense, 31
secondary substances, 56–8, 60–1 thoughts, singular, 2, 19–20, 23–30, 89
semantic values, 52–4 Tooley, M., 40
set membership, 96 tracking, perceptual, 22–5, 87–8, 90, 93
set theory, 76 transcategorial terms, 13, 16–17, 27
sets, 15, 20, 53, 74–6, 95, 98, 101, 113–14, transuranic elements, 148
146, 159 tropes, 32, 39, 57
ordered, 95–6, 98 truth, 204
ship of Theseus, 91 truthmakers, 50, 65, 110
ships, 92
singular terms, 70–3, 78–80, 82–4, 87, 89–91, universal generalizations, 40, 61
93, 136–7 universals, 2–4, 31–4, 37, 40, 45, 57, 203
Smith, B., 50
sortal concepts, 11 vague objects, 119–20
sortal logic, 40, 42 vagueness, 5, 15–17, 102, 119
sortal terms, 12–13, 15, 42, 74, 79–80, 82–3, 85, 87, ontic, 120
90, 92 semantic, 120, 124, 136
artefactual, 91 van Fraassen, B. C., 40–1
criterionless, 83–4, 87–8, 90–1 variables, 28, 34, 40, 60–1, 65
sortalism, 2, 11
species, 56–7, 59, 61, 63 Wason selection task, 168
Stalnaker, R. C., 174, 177–8, 180 Wasserman, R., 41
states of affairs water, 41–2, 147, 149–51
dispositional, 40–1 Wiener, N., 96
occurrent, 40 Wiggins, D., 108
statues, 100, 102–4, 106, 110, 113, 153–4 Williams, B., 108
Strawson, P. F., 86 Williamson, T., 72
subjects, 34, 57, 61 Wittgenstein, L., 65, 70, 85, 87

Forms of Thought

Uploaded by

Copyright:

Available Formats

Forms of Thought

Uploaded by

Document Information

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

Forms of Thought

Uploaded by

Copyright:

Available Formats

FORMS OF THOUGHT

Forms of thought are involved whenever we name, describe, or

e . j . l o w e is Professor of Philosophy at the University of Durham.

This publication is in copyright. Subject to statutory exception

First published 2013

Library of Congress Cataloguing in Publication data

Cambridge University Press has no responsibility for the persistence or

List of ﬁgures page vii

part i reference and predication

part iii modality

Figure 3.1 The ontological square, version i page 34

Table 3.1 Logical symbolism for the four-category ontology page 36

I have given this book the subtitle A Study in Philosophical Logic in

The chapters of this book draw in some measure on previously published

Some kind of knowledge of logical forms, though with most people it

1 reference and predication

Individuation, reference, and sortal terms

1 sortal, categorial, and transcategorial terms

2 criteria of identity and sortal persistence

3 two notions of individuation

4 object perception and singular thought

5 categorialism and linguistic reference

Two styles of predication – dispositional

In the preceding chapter, I devoted a considerable amount of discussion to

1 the four-category ontology and its logic

instantiated by exemplified by instantiated by

Objects characterized by Modes

Figure 3.1. The ontological square, version i

Figure 3.2. The ontological square, version ii

explicitly only a single formal ontological relation, identity. In a logic that is

Objects Kinds Attributes Modes

Instantiation: a/β, f/G

Figure 3.3. The ontological square, version iii

2 cross-categorial ontological dependencies

Figure 3.4. The ontological square, version iv

3 laws of nature and the inference problem

4 conditional laws and the problem of

5 conditional laws and the principle

$x~(x/β ® $φ(φF & x/φ))

$x(x/β & ~$φ(φF & x/φ))

a/β & ~$φ(φF & a/φ)

"φ~(φF & a/φ)

DE[a, G ] & DE[a, (G ® F )]

$φ(φG & a/φ)

~$φ(φF & a/φ)

β(G ® F) & a/β

$φ(φF & a/φ)

$X(O[a, X ] & ~DE[a, X ] & $φ(φ(X ® F ) & a/φ)) (i)

O[a, G ] & ~DE[a, G] & $φ(φ(G ® F ) & a/φ)

DA[a, G ] DE[a, G ] (ii)

(DA[a, G ] & $φ(φ(G ® F ) & a/φ)) ® G = F

~(DA[a, G ] & $φ(φ(G ®F ) & a/φ)) G=F

$φ(φ(F ® F ) & a/φ)

O[a, F ] & ~DE[a, F ]

$φ(φ(F ® F ) & a/φ)

O[a, F ] & ~DE[a, F ] & $φ(φ(F ® F ) & a/φ)

$X (O[a, X ] & ~DE[a, X ] & $φ(φ(X ® F ) & a/φ))

Ontological categories and categorial predication

I have said a good deal in previous chapters about the notion of an

1 fantology; or, ‘ontology lite’