MacBride - Identity and Modality - Oxford University Press, USA. 0199285748

Identity and Modality
MIND ASSOCIATION OCCASIONAL SERIES
This series consists of occasional volumes of original

papers on predefined themes. The Mind Association
nominates an editor or editors for each collection, and
may cooperate with other bodies in promoting
conferences or other scholarly activities in connection
with the preparation of particular volumes.
Publications Officer: M. A. Stewart
Secretary: R. D. Hopkins
Identity and Modality
Fraser MacBride
CLARENDON PRESS · OXFORD

1
Great Clarendon Street, Oxford ox2 6dp
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© the several contributors 2006
The moral rights of the authors have been asserted
Database right Oxford University Press (maker)
First published 2006
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Identity and modality / [edited by] Fraser MacBride.
p. cm.—(Mind Association occasional series)
Includes bibliographical references and index.
ISBN-13: 978–0–19–928574–7 (alk. paper)
ISBN-10: 0–19–928574–8 (alk. paper)
1. Identity (Philosophical concept) 2. Modality (Theory of knowledge) I. MacBride,
Fraser. II. Series.
BD236.I4155 2006
111 .82—dc22 2006009859
Typeset by Laserwords Private Limited, Chennai, India
Printed in Great Britain
on acid-free paper by
Biddles Ltd, King’s Lynn, Norfolk
ISBN 0–19–928574–8 978–0–19–928574–7
1 3 5 7 9 10 8 6 4 2
CONTENTS
Acknowledgements vii
Notes on contributors ix
Introduction 1
Fraser MacBride
Part I: Modality 11
1. The Limits of Contingency 13

Gideon Rosen
2. Modal Infallibilism and Basic Truth 40

Scott Sturgeon
3. The Modal Fictionalist Predicament 57

John Divers and Jason Hagen
4. On Realism about Chance 74

Philip Percival
Part II: Identity and Individuation 107
5. Structure and Identity 109

Stewart Shapiro
6. The Identity Problem for Realist Structuralism II:

A Reply to Shapiro 146
Jukka Keränen
7. The Governance of Identity 164

Stewart Shapiro
vi / Contents
8. The Julius Caesar Objection: More Problematic than Ever 174

Fraser MacBride
9. Sortals and the Binding Problem 203

John Campbell
Part III: Personal Identity 219
10. Vagueness and Personal Identity 221

Keith Hossack
11. Is There a Bodily Criterion of Personal Identity? 242

Eric T. Olson
Index 261
ACKNOWLEDGEMENTS
This volume grew out of a conference held at the University of St Andrews

in July 2000, the first of an ongoing series of conferences held under the
auspices of Arché (now the AHRC Research Centre for the Philosophy of
Logic, Language, Mathematics and the Mind).
The conference was funded by the British Academy, the British Society for
the Philosophy of Science, the John Wright Trust, the Mind Association, the
Royal Society of Edinburgh, the Scots Philosophical Club, and the School of
Philosophical and Anthropological Studies at the University of St Andrews. The
Analysis Trust and the Stirling–St Andrews Graduate Programme performed
the invaluable service of subsidising graduate attendance at the conference.
I gratefully acknowledge the support of the Arts and Humanities Research
Council who funded a period of leave during which the volume was prepared
for publication.
Fraser MacBride
Birkbeck College London
This page intentionally left blank
NOTES ON CONTRIBUTORS
John Campbell isWillisS. andMarion Slusser Professor of Philosophy, University

of California at Berkeley. He is the author of Past, Space and Self (1994) and
Reference and Consciousness (2002).
John Divers is a Professor of Philosophy at Sheffield University. He is the author
of Possible Worlds (Routledge 2002).
Jason Hagen is a graduate student in the Philosophy Department at Purdue
University.
Keith Hossack is Lecturer in Philosophy at King’s College London. He is the
author of The Metaphysics of Knowledge (Oxford University Press, forthcoming).
Jukka Keränen is a visiting Professor in the Philosophy Department at North
western University.
Fraser MacBride is Reader in Philosophy at Birkbeck College London. He is the
author of several papers in metaphysics and the philosophy of mathematics.
Eric T. Olson is Reader in Philosophy at Sheffield University. He is the
author of The Human Animal: Personal Identity Without Psychology (Oxford University
Press, 1997).
Philip Percival is Reader in Philosophy at Glasgow University. His publications
include papers on metaphysics and philosophy of science.
Gideon Rosen is Professor of Philosophy at Princeton University. He is the
author (with John P. Burgess) of A Subject With No Object (Oxford University
Press, 1997).
Stewart Shapiro is O’Donnell Professor of Philosophy at the Ohio State
University and Professorial Fellow in Arché: the AHRC Research Centre for
the Philosophy of Logic, Language, Mathematics & Mind at the University
of St Andrews. His is the author of Foundations Without Foundationalism (Oxford
x / Notes on Contributors
University Press, 1991), Philosophy of Mathematics: Structure and Ontology (Oxford
University Press, 1997), and Thinking about Mathematics: Philosophy of Mathematics
(Oxford University Press, 2000).
Scott Sturgeon is Reader in Philosophy at Birkbeck College London. He is
the author of Matters of Mind (Routledge, 2000).
Introduction
Fraser MacBride
The papers in this volume constellate about fundamental philosophical issues

concerning modality and identity: How are we to understand the concepts of
metaphysical necessity and possibility? Is chance a basic ingredient of reality?
How are we to make sense of claims about personal identity? Do numbers
require distinctive identity criteria? Does the capacity to identify an object
presuppose an ability to bring it under a sortal concept? In order to provide a
guide to the reader I will provide a brief overview of the content of the papers
collected here and some of the interrelations that obtain between them.
Part I: Modality
In ‘The Limits of Contingency’ Gideon Rosen sets out to examine the
modal status of metaphysical and mathematical propositions. Typically such
propositions—that, for example, universals or aggregates or sets exist—are
claimed to be metaphysically necessary. But such claims of metaphysical
necessity, Rosen maintains, are inherently deficient. This is because the kinds
of elucidation philosophers typically offer of the concept of metaphysical
necessity fail to pin down a unique concept of necessity; in fact no conception
exactly fits the elucidations given, and at least two conceptions—which
Rosen dubs ‘Standard’ and ‘Non-Standard’—fit the elucidations equally well.
According to the Standard Conception, the synthetic apriori truths of basic
2 / Introduction
ontology are always necessary. By contrast, according to the Non-Standard
Conception, such truths are sometimes contingent. Consider, for example,
Armstrong’s claim that qualitative similarity between particulars is secured
by the recurrence of immanent universals. By the lights of the Standard
Conception this claim, if it is true, is metaphysically necessary. For whilst
it is not a logical or a conceptual necessity—there is no reason think
it’s denial self-contradictory or otherwise inconceivable—it is not aposteriori
either. But, by the lights of the Non-Standard Conception, Armstrong’s claim
is contingent. For other metaphysical accounts that eschew universals—in
favour, for example, of duplicate tropes—are also compatible with the nature
of the similarity relation. So, if it is true, Armstrong’s claim tells us only about
how similarity happens to be secured in the actual world; in other possible worlds
similarity is secured differently. Since philosophical elucidations of the concept
of metaphysical necessity favour neither the Standard nor the Non-Standard
Conception Rosen concludes that philosophical discourse about metaphysical
necessity is shot through with ambiguity, an ambiguity that we ignore at
our peril.
In ‘Modal Infallibilism and Basic Truth’ Scott Sturgeon investigates
further the relationship between metaphysical possibility and intelligibility.
Most philosophers agree that apriori reflection provides at best a fallible
guide to genuine possibility. The schema (L) that says: if a proposition is
intelligible then it is genuinely possible, is generally recognized not to be
valid. Nevertheless, Sturgeon argues, philosophers have frequently failed to
practise what they preach. They have been led by (L) to advance contradictory
claims about the fundamental structure of reality. Sturgeon provides as a
representative example of the capacity of (L) to mislead, a battery of six
basic metaphysical claims about change and identity that Lewis has advanced
but together generate contradiction. They generate contradiction because,
Sturgeon maintains, Lewis accepts at least one instance of (L), inferring from
the intelligibility of objects that endure identically through time—Sturgeon
calls these ‘enduring runabouts’—that such objects are genuinely possible.
The contradiction Sturgeon uncovers suggests that (L), if it is true at all, must
be significantly qualified. But, Sturgeon argues, there is no restricted reading of
(L) that is valid either. The first restriction Sturgeon considers qualifies (L) to
accommodate Kripke’s insight that there are intelligible propositions that
fail to mark genuine possibility because the negations of these propositions
are aposteriori necessary. Sturgeon rejects (L) so qualified because there is at
Fraser MacBride / 3
least one intelligible proposition P whose equally intelligible negation ¬P
fails to be a posteriori necessary but nevertheless it cannot be the case that
P and ¬P are both genuinely possible. Sturgeon provides as an example
of such a P the Lewisian proposition that concrete possible worlds are the
truth-makers for claims of genuine possibility. After considering yet further
unsatisfactory qualifications to (L) Sturgeon concludes that philosophers
have been mislead by the ‘ep-&-met tendency’, the human tendency to
fuse epistemic and metaphysical matters; what is required is to recognize
where this tendency misleads us whilst—and this is where the task becomes
almost insuperably difficult—continuing to respect the fact that it is a
cornerstone of our modal practice that intelligibility defeasibly marks genuine
possibility.
In ‘The Modal Fictionalist Predicament’, John Divers and Jason Hagen
turn to consider the metaphysics of modality itself. According to ‘genuine
modal realism’, the metaphysical status of modal statements is rendered
perspicuous by translating claims about what is possible into (counterpart-
theoretic) claims about possible worlds. But the doctrine that there really are
such outlandish entities as possible worlds encounters familiar metaphysical
and epistemological difficulties. However the adherents of ‘modal fictionalism’
maintain that the benefits of possible worlds discourse may be secured without
these associated costs. They attempt to achieve this by conceiving of possible
worlds discourse as itself just an immensely useful fiction that does not
commit us to the existence of possible worlds. Part of what makes modal
fictionalism plausible is what Divers has called a ‘safety result’: the result that
translating our ordinary modal claims in and out of the fictional discourse
of possible worlds will never lead us astray. However Divers and Hagen
question whether the modal fictionalist is in a position to take advantage
of this result. Two objections to modal fictionalism have arisen over the
decade since the doctrine was first advanced. According to the first objection,
modal fictionalism, despite surface appearances, is committed to the existence
of a plurality of possible worlds. According to the second objection modal
fictionalism is not even consistent; its acceptance results in modal collapse,
so that for any modal claim X, both X and ¬X are true. Divers and Hagen
argue that each objection may be avoided by deft handling of the doctrine.
But what, they maintain, modal fictionalists cannot do is to avoid one or other
of these objections whilst maintaining a right to the safety result that makes
modal fictionalism plausible in the first place. Divers and Hagen conclude that
4 / Introduction
modal fictionalism is in a serious predicament. Modal fictionalism must be
rescued from this predicament if it is to be considered a genuine competitor to
genuine modal realism.
Philip Percival’s ‘On Realism about Chance’ considers the metaphysical
status of another modal notion, namely chance. Chance, as Percival conceives
it, is a single-case (applying to individual events), temporally relative (liable
to change over time), objective probability (existing independently of what
anyone thinks about it) towards which our cognitive attitudes are normatively
constrained. Percival construes the question of whether chance exists as the
question of whether there are objectively true statements of the form ‘the
chance at time t of event E is r’. Famously, Lewis has advanced realism about
chance but Percival takes issue with this assessment, arguing for scepticism
about the kinds of reason one might give for realism about chance. One
common reason for affirming realism about chance is that chance may be used
toexplain statistical phenomena or the warrantednessof certain credences. But,
Percival argues, the notion of chance cannot perform this kind of explanatory
role. Consequently, an inference to the best explanation of (e.g.) statistical
phenomena cannot be employed to ground realism about chance. Another
reason commonly offered for affirming realism about chance is that chance
may be analysed in terms of non-chance. If chance is analysable then either
chance supervenes (relatively) locally upon non-chance or chance supervenes
globally upon non-chance. Buthowever chance supervenes, Percival argues, no
extant analysis—including Lewis’s ‘best-system’ analysis—succeeds. Percival
concludes upon the sceptical reflection that there is little prospect of a correct
analysis of chance being forthcoming in the future that vindicates realism
about chance.
Part II: Identity and Individuation

The next three papers reflect upon the identity and individuation of math-
ematical objects. In his ‘Structure and Identity’ Stewart Shapiro reflects
upon the doctrine (advanced in his Philosophy of Mathematics: Structure and Ontology
(Oxford: OUP, 1997)) that mathematical objects are places in structures where
the latter are conceived as ante rem universals. This doctrine—that Shapiro
dubs ‘ante rem structuralism’—suggests that there is no more to a mathem-
atical object than the (structural) relations it bears to the other objects within
Fraser MacBride / 5
the structure to which it belongs. However, as Shapiro recognizes, when
conceived in this way ante rem structuralism is open to a variety of criticisms.
This is because there appears to be more to a mathematical object than the
relations it bears to other objects within its parent structure. Mathematical
objects enjoy relations to (i) items outside the mathematical realm (e.g. the
concrete objects they are used to measure or count) and (ii) objects that belong
to other structures inside the mathematical realm. Moreover, (iii) there are
mathematical objects (e.g. points in a Euclidean plane) that are indiscernible
with respect to their (structural) relations but nevertheless distinct. This
makes it appear that ante rem structuralism is committed to the absurdity of
identifying these objects. Shapiro seeks to overcome these difficulties by a
series of interlocking manoeuvres. First, he seeks to overturn the metaphysical
tradition about numbers, suggesting that it may be contingent whether a
given mathematical object is abstract or concrete. Second, Shapiro questions
whether mathematical discourse is semantically determinate. Finally, Shapiro
rejects the requirement that ante rem structuralism provide for the non-trivial
individuation of mathematical objects.
In ‘The Identity Problem for Realist Stucturalism II: A Reply to
Shapiro’ Jukka Keränen argues that Shapiro nevertheless fails to provide an
adequate account of the identity of numbers conceived as places in structures.
According to Keränen, it is an adequacy constraint upon the introduction
of a type of object that some account be given of the kinds of fact that
metaphysically underwrite the sameness and difference of objects of this type.
More specifically, Keränen favours the view that facts about the sameness and
difference of objects must be underwritten by facts about the properties they
possess or relations they stand in. He holds up set theory as an exemplar of a
theory that meets this adequacy constraint, grounding the identity of sets—via
the Axiom of Extensionality—in facts about their members. Keränen doubts,
however, whether ante rem structuralism can meet this adequacy constraint
because there are no structural properties or relations that can be used to
distinguish between (e.g.) the structurally indiscernible points in a Euclidean
plane. Of course, the structuralist can meet the constraint by force majeure,
positing a supply of haecceitistic properties to distinguish between structurally
indiscernible objects. But, as Keränen reflects, the positing of haecceities opens
up the possibility of indiscernible structures that differ only haecceitistically.
Since mathematical discourse lacks the descriptive resources to distinguish
between these structures, this manoeuvre on the part of the structuralist
6 / Introduction
threatens to render reference to mathematical objects deeply inscrutable.
Keränen concludes that the particular difficulties encountered by ante rem
structuralism in particular reflect deep difficulties for ontological realism in
general.
‘The Governance of Identity’ is Shapiro’s response to Keränen. Shapiro
first concedes, for the sake of argument, the adequacy constraint on the
introduction of a type of object Keränen imposes. Shapiro then argues that
indiscernible objects within a structure S may be distinguished by embedding
S within a larger structure S∗ whose positions are discernible. Later, lifting the
concession, Shapiro questions whether it is necessary to supply non-trivial
identity conditions for a type of object introduced. He concludes rather that
identity must be taken as primitive.
In ‘The Julius Caesar Objection: More Problematic than Ever’ Fraser
MacBride further explores issues surrounding the identity and individuation
of numbers from a Fregean point of view. According to Frege it is a requirement
upon the introduction of a range of objects into discourse that identity criteria
are supplied for them—criteria that determine whether it is appropriate
to label and then relabel an object on a different occasion as the same
again. In order to introduce cardinal numbers into discourse Frege therefore
proposed the following principle—Hume’s Principle that specifies necessary
and sufficient conditions for the identity of cardinal numbers: the number of
Fs = the number of G’s iff there is a 1-1 correspondence between the Fs and
the Gs. Famously, however, Frege became dissatisfied with Hume’s Principle as
a criterion of identity, maintaining that it failed even to settle whether (e.g.)
the number two was identical or distinct to an object of an apparently quite
different sort (e.g.) the man Caesar. MacBride subjects this difficulty—the
so-called ‘Julius Caesar Objection’—to critical examination, arguing that
beneath the superficial simplicity of the problem that bedevilled Frege there
lies a welter of distinct difficulties. These may be arranged along three
different dimensions. (A) Epistemology: does the identity criterion supplied
for introducing numbers into discourse provide warrant for the familiar
piece of common-sense knowledge that numbers are distinct from persons?
(B) Metaphysics: doesthe identity criterion given determine whether the things
that are numbers might also be such objects as Caesar? (C) Meaning: does the
identity criterion supplied bestow upon the expressions that purport to denote
numbers the distinctive significance of singular terms? It is because, MacBride
argues, these different problems and the interrelations between them often
Fraser MacBride / 7
fail to be disentangled that (in part) the different (purported) solutions to the
Julius Caesar—neo-Fregean and supervaluationist solutions—fail. MacBride
concludes by suggesting that Frege may have been too strict in imposing
the requirement that objects introduced into discourse have identity criteria,
noting that not even sets have identity criteria in the strict sense Frege
required.
John Campbell’s ‘Sortals and the Binding Problem’ sets out to question
the related doctrine that singular reference to an object depends upon a
knowledge of the sort of object (whether a number or a man) to which one is
referring. Part of what makes this doctrine plausible is the fact that, as Quine
emphasized, our pointing to something remains ambiguous until the sort of
thing that we are pointing is made evident. For example, I can point towards
the river and variously be taken to refer to the river itself which continues
downstream, a temporal part of the river that exists contemporaneously with
my pointing gesture, the collection of water molecules that occupies the river
when I point, and so on. But if I specify the sort of object to which I wish to
draw your attention then it becomes determinate what I am pointing to. These
kinds of consideration have led philosophers to adopt what Campbell calls
‘The Delineation Thesis’: Conscious attention to an object has to be focused
by the use of a sortal concept that delineates the boundaries of the object
to which you are attending. Campbell argues however that the delineating
thesis is false. Instead, Campbell proposes, attention to an object arises from
the way in which the visual system binds together the information it receives
in various processing streams. Roughly speaking, the visual system does so by
exploiting the location of an object together with the Gestalt organization of
characteristics found at that location. Since this integration may be achieved
without the use of a semantic classification of an object as of a certain sort it
appears that we can single out an object without the use of a sortal concept.
Philosophers have nevertheless been mislead into supposing the Delineation
Thesis because, Campbell maintains, of the typical use that is made of sortal
concepts in demonstrative constructions (‘that mountain’) and our readiness
to withdraw these constructions when it transpires that these sortal concepts
are misapplied (when, for example, it turns out that our attention is being
drawn to what is merely a hill). Campbell argues nonetheless that sortal
concepts employed in demonstrative construction serve merely to orientate
our attention to an object without necessarily contributing to the content of
what is said by the use of these constructions.
8 / Introduction
Part III: Personal Identity

The remaining two papers of the collection turn to a consideration of issues
related to the identity of persons. In ‘Vagueness and Personal Identity’
Keith Hossack considers the influential ‘Bafflement Argument’ put forward by
Bernard Williams, an argument that threatens to undermine the materialistic
conception of the self. The well-known thought experiments about personal
identity suggest that there are possible situations—where, for example, a
subject undergoes fission—in which it is indefinite whether the subject
survives. If materialism is true it appears that this indefiniteness must be
objective. For it appears that there are no sharp boundaries to the biological
processes or physical mechanisms that sustain human life. By contrast, if
dualism is true it appears that this indefiniteness can only be a matter of
ignorance. For the kinds of issues in ethics and philosophy of religion that give
rise to dualism suggest that the boundaries between souls must be sharp.
Williams’s Bafflement Argument suggests however that we cannot make sense
of objectively indefinite identity in the case of persons, and so materialism must
be abandoned. This is because we cannot make sense—we are baffled by—the
suggestion that it is objectively indefinite whether I (or you) will continue
to exist tomorrow. Hossack seeks to defend materialism by showing that the
Bafflement Argument owes its persuasive force to a skewed conception of the
self that fails to recognize that the correct way to understand the ‘I’ concept is
as the intersection of subjective and objective ways of thinking about the self.
What is wrong with the Cartesian conception of the self is that it fails to give
due weight to the location of the self in the objective worldly order. But what
is wrong with the bodily conception of the self—a conception advanced, for
example, by Strawson and Evans—is that by identifying the self with the body
it fails to sufficiently stress the subjective aspect of the ‘I’ concept. The mistake
that underlies the Bafflement Argument, Hossack maintains, is a misguided
solipsistic conception of the self that arises from focusing exclusively upon the
subjective aspect of the ‘I’ concept. Once this mistake is corrected by giving
proper weight to the place of persons in the objective order—without falling
over into the corresponding failings of the bodily conception of the self—the
Bafflement Argument need no longer pose a threat to materialism.
Eric T. Olson’s ‘Is There a Bodily Criterion of Personal Identity?’
continues the theme of questioning how we relate to our bodies. One of the
perennial debates about personal identity concerns whether we should adopt
Fraser MacBride / 9
a bodily criterion of personal identity as opposed, say, to a psychologistic
criterion. But this debate only makes sense if there is such a thing as a bodily
criterion of personal identity; about the existence of such a criterion Olson
expresses scepticism. The bodily criterion is supposed to offer an account
according to which we are our bodies or, at least, that our identity over time
consists in the identity of our bodies. So the bodily criterion is supposed to
be a non-trivial thesis about our bodies and how we are related to them that
determines that we go where our bodies go. But, Olson argues, we cannot
specify the bodily criterion in such a way as to ensure that it does what it is
supposed to do. Olson’s argument for this conclusion proceeds by elimination,
considering in turn a variety of different purported specifications of the
bodily criterion. Either these criteria imply too little or they imply too much:
either (1) they say nothing about, or leave it open that we may survive, the
destruction of our bodies or (2) they imply that you could never be a foetus or
a corpse. It may be suggested that the difficulties identified are a consequence
of the surreptitious assumption of a Cartesian account of body ownership. But
Olson dismisses this suggestion, arguing that the accounts of body ownership
proposed by Shoemaker and Tye imply that the bodily criterion is not the
substantial thesis debate assumes but a trivial consequence of materialism.
How did such a depth of misunderstanding arise? Olson ventures a diagnosis.
We are misled by the superficial grammar of such expressions as ‘Wilma’s
body’; in this case, an expression that appears to be the name of an object with
which Wilma enjoys an especially intimate relationship. But, Olson argues, we
should no more believe that ‘Wilma’s body’ names a special object than we
should believe that the expression ‘Wilma’s mind’ names another object with
which she enjoys a different, but not less intimate, relationship.
Part I
Modality
1
The Limits of Contingency
Gideon Rosen
1. What is Metaphysical Necessity?

There are two ways to understand the question. We might imagine it asked by
an up-to-date philosopher who grasps the concept well enough but wants to
know more about what it is for a proposition to hold of metaphysical necessity.
Alternatively, we might imagine it asked by a neophyte who’s never heard the
phrase before and simply wants to know what philosophers have in mind by it.
My main interest in this paper is the first sort of question. But for several
reasons it will help to begin with the second. Suppose it were your job to
explain the concept of metaphysical necessity to a beginner. What might
you say?
The task is not straightforward. The concepts of metaphysical possibility
and necessity are technical concepts of philosophy. Not only is the phrase
‘metaphysical necessity’ a bit of jargon. No ordinary word or phrase means
exactly what the technical phrase is supposed to mean. So you cannot say,
‘Ah, it’s really very simple: What we call ‘‘metaphysical necessity’’ is what you
call. . .’. If you’re going to explain the technical idiom to the neophyte you’re
going to have to introduce him to a novel concept. You’re going to have to
teach him how to make distinctions that he does not already know how to
An earlier version of this paper was presented at the Arché conference in St Andrews in June 2000.
I am grateful to Scott Sturgeon for his exemplary comments on that occasion and for extensive
correspondence. I have been unable to take many of his important suggestions into account.
14 / The Limits of Contingency
make. And so the question comes down to this: How is this novel concept to
be explained?
The best sort of explanation would be an informative definition: an explicit
specification, framed in ordinary terms, of what it means to say that P
is metaphysically necessary. Unfortunately, no such definition is readily
available. But we know in advance that it must be possible to get along without
one. For the fact is that no one comes to master the concept of metaphysical
necessity in this way. To the contrary, the project of definition and analysis in
modal metaphysics invariably presupposes aprior grasp of the technical modal
concepts. For up-to-date philosophers it always works like this: First, we learn
what it means to say that P is metaphysically necessary. Then we look for an
account of what this comes to in other terms. We may or may not find one.
It would not be surprising if the basic concepts of modal metaphysics were
absolutely fundamental.1 But even if we do, our capacity to recognize it as
correct will depend on prior grasp of the metaphysical modal idiom. And it is
this prior grasp that we are attempting to inculcate in our neophyte.
If we do not begin with a definition, we must offer some sort of informal
elucidation. We all know roughly how this works in other parts of philosophy.
The neophyte is presented with a battery of paradigms and foils, ordinary
language paraphrases (with commentary), and bits and pieces of the inferential
role of the target notion, and then somehow as a result of this barrage he cottons
on. To ask how the concept of metaphysical necessity might be explained to the
neophyte is to ask how this informal elucidation ought to go in the modal case.
No doubt, there is more than one way to proceed. But here is one possibility.
2. An Informal Elucidation
The first thing to say is that metaphysical necessity is a kind of necessity. To
say that P is metaphysically necessary is to say that P must be the case, that it
has to be the case, that it could not fail to be the case, and so on. If the ordinary
modal idioms were univocal, this would be enough. But it clearly isn’t enough.
When I drop an apple there is a sense in which it cannot fail to fall. When I
promise to meet you there is a sense in which I have to keep the date. But
these claims involve two very different modal notions, and neither is a claim
1
For an argument to this effect, see Kit Fine (2002).
Gideon Rosen / 15
of metaphysical necessity. So granted, the metaphysical ‘must’ is a kind of
‘must’. The challenge remains to distinguish it from the many others.
Some distinctions are easy. Thus, unlike the various practical and ethical
‘musts’, the metaphysical ‘must’ is alethic. If P is metaphysically necessary, then
it’s true. And unlike the various epistemic and doxastic musts (e.g. the ‘must’
in ‘She must be home by now. She left an hour ago.’) claims of metaphysical
necessity are not in general claims about what is known or believed.
Other distinctions are less straightforward. Thus, some philosophers believe
in something called logical necessity, and some believe in something called analytic
or conceptual necessity, where a truth is logically necessary when some sentence
that expresses it is true in all models of the language (or some such thing) and
conceptually necessary when it is true in virtue of the concepts it contains
(or some such thing). It is controversial whether these are genuine species of
necessity. After all, it is one thing to say that P is necessary in some generic
sense because it is a truth of logic. It is something else to say that P therefore enjoys
a special sort of necessity. But if there is a distinctively logical or conceptual
species of necessity, then it is (presumably) both alethic and non-epistemic,
and in that case we must say something to distinguish metaphysical necessity
from such notions.
At this point the usual procedure is to invoke an epistemological distinction
along with certain crucial paradigms. One says: ‘Unlike the various logical
and semantic species of necessity, metaphysically necessary propositions are
sometimes synthetic and aposteriori. To a first approximation, the logicoconceptual
necessities are accessible to ‘Humean reflection’. To suppose the falsity of a
logical or a conceptual truth is to involve oneself in the sort of self-
contradiction or incoherence that a sufficiently reflective thinker might detect
in the armchair simply through the exercise of his logical and semantic
capacities. By contrast, some metaphysically necessary truths can be rejected
without such incoherence. The most famous examples are the Kripkean
necessities: true identities flanked by rigid terms; truths about the essences
of individuals, kinds, and stuffs. But one might also mention the claims of
pure mathematics in this connection. Mathematical truths are among the
paradigms of metaphysical necessity. But logicism not withstanding, it is not
self-contradictory to reject mathematical objects across the board, or to deny
selected existential principles such as the axiom of infinity. So if substantive
truths of these sorts can be necessary in the metaphysical sense, metaphysical
necessity differs from logical or conceptual necessity. Indeed, the natural thing
to say is that metaphysical necessity is a strictly weaker notion, in the sense that
some metaphysically necessities are neither logical nor conceptual necessities,
but not vice versa.
Let’s call any modality that is alethic, non-epistemic, and sometimes
substantive or synthetic a real modality. So far we have it that metaphysical
modality is a real modality. But this is still not enough to pin the notion
down. For the same might be said of the various causal or nomic modalities:
physical necessity, historical inevitability, technical impossibility (as in: ‘It’s
impossible to fabricate an artificial liver’), and so on. And here the natural
thing to say is that among the real modalities, the metaphysical modalities
are absolute or unrestricted. Metaphysical necessity is the strictest real necessity
and metaphysical possibility is the least restrictive sort of real possibility in the
following sense: If P is metaphysically necessary, it is necessary in every real
sense: If P is really possible in any sense, then it’s possible in the metaphysical
sense. So if you can’t square the circle because it’s metaphysically impossible
to square the circle, then it’s certainly not physically or biologically possible
for you to do so. But if you can’t move faster than the speed of light because
to do so would be to violate a law of nature, then it does not follow that
superluminal velocities are metaphysically impossible. One has the palpable
sense—though philosophy might correct it—that some of the laws of nature
might have been otherwise. To say this is not just to say that these laws are not
logical or conceptual truths. That is too obvious to be worth saying. And it’s
certainly not to say that the laws of nature amount to physical contingencies.
No, to entertain the philosophical suggestion that the laws might have been
otherwise is to presuppose that there exists a genuine species of contingency
‘intermediate’ between physical contingency on the one hand and conceptual
contingency on the other. Focus on this sense of contingency, it might be said,
and you are well on your way to knowing what ‘metaphysical’ modality is
supposed to be. It is the sort of modality relative to which it is an interesting
question whether the laws of nature are necessary or contingent.
3. A Question about the Informal Elucidation

Informal explanations of this sort are the indispensable starting point for modal
metaphysics. In the end we may hope for more: an account of what it is for a
proposition to be metaphysically necessary; an account of what in reality makes it the case that P
Gideon Rosen / 17
is metaphysically necessary when it is. But before we can ask these profound questions,
we must identify our topic. We must distinguish metaphysical possibility and
necessity from the various other species of possibility and necessity. And it is
natural to suppose that for this restricted purpose something like the informal
explanation sketched above should be sufficient.
In fact, this is almost universally supposed. A small handful of philosophers
reject the notions of metaphysical possibility and necessity altogether.2 But
among those who accept them, it is universally assumed that a question
about the metaphysical modal status of any given proposition is clear and
unambiguous, at least as regards the predicate. We may not be able to say what
metaphysical necessity really is in its inner most nature. But thanks to the
informal elucidation sketched above or something like it, we know enough
about it to ask unambiguous questions about its nature and its extension. Our
questions may be hard to answer. In some cases we may not even know where
to begin. But even so, it is perfectly clear what is being asked when we ask
whether P holds as a matter of a metaphysical necessity.
One of my aims in this paper is to reconsider this supposition. I shall suggest
that the informal explanation sketched above is consistent with two distinct
conceptions of necessity and possibility; or better, since no single conception is
fully consistent with the sketch, that two relatively natural conceptions fit
the elucidation equally well. If this is right then our working conception
of metaphysical necessity is confused in the sense in which the Newtonian
conception of mass is supposed to have been confused.3 Questions about
metaphysical necessity are ambiguous, and where divergent resolutions of
the ambiguity yield different answers, the modal question as we normally
understand it has no answer. Indeed, I shall suggest that this is just what we
should think about an interesting (though largely neglected) class of questions.
4. The Standard Conception and the Differential Class

If there are two conceptions of metaphysical necessity, they must overlap
considerably in extension. The informal explanation functions as a constraint
2
Dummett (1993: 453) calls it ‘misbegotten’, though he is elsewhere moderately sympathetic
to a closely related notion of ‘ontic necessity’ (cf. Dummett 1973: 117; 1981: 30). Field (1989: esp.
235 ff.) expresses general skepticism about the notion.
3
Field (1974).
on both, and it includes a list of paradigms, a significant number of which
must count as ‘metaphysically necessary’ on any modal conception worth the
name. On the account I propose to consider, the (non-indexical) logical truths
and the conceptual truths more generally will count as necessary on both
conceptions, as will the uncontroversial Kripkean necessities: (propositions
expressed by) true identities flanked by rigid terms and essential predications:
propositions of the form Fa, where a is essentially F. The two conceptions will
diverge in application to certain claims of fundamental ontology, which do
not slot easily into any of these categories.
For an example of the sort of claim I have in mind, consider the axioms
of standard set theory. At least one is plausibly analytic—the axiom of
extensionality, according to which sets are identical if and only if they have
just the same members. Insofar as this axiom is uncontroversial, it does not
entail that sets exist. It says that if sets exist, they are extensional collections.
And since this is presumably part of what it is to be a set—or if you prefer,
part of what the word ‘set’ means—the axiom of extensionality will count as
metaphysically necessary on any reasonable conception of the notion.4
The same cannot be said for the remaining axioms. Consider the simplest:
the pair set axiom:
(Pairing) For any things x and y, there exists a set containing just x and y.
In conjunction with Extensionality, Pairing entails that given a single non-set,
infinitely many sets exist. The truth of Pairing is not guaranteed by what it is
to be a set, or by what the word ‘set’ means. It may lie in the nature of the
sets to satisfy the principle, but only in the sense that if there are any sets,
then it lies in their collective nature to conform to pairing. (That is part of
what makes them the sets, it might be said.) It may be that no relation deserves
the name ‘∈’ unless it satisfies the pairing axiom, just as nothing deserves the
name ‘bachelor’ unless it is male. But it is not in the nature of bachelorhood to
be instantiated; and likewise, it is not in the nature of the epsilon relation
that something should bear it to something else. You can know full well
what set membership is supposed to be—what it is to be a set, what the word
‘set’ means—without knowing whether any sets exist, and hence without
knowing whether Pairing is true.
4
But see Frankel, Bar Hillel, and Levy (1973: 27–8). For discussion of the analyticity of
extensionality, see Maddy (1997: 39).
Gideon Rosen / 19
What is the modal status of the Pairing axiom? Suppose it’s true. Is it a
metaphysically necessary truth or a contingent one? As we have said, it is tradi-
tional to regard the truths of pure mathematics as paradigms of metaphysical
necessity. On this view, while there may be room for dispute about whether
sets exist, and if so, which principles they satisfy, there is no room for dispute
about the modal status of those principles. If sets exist and satisfy Pairing, then
Pairing holds of necessity. If sets do not exist (or if they do exist and somehow
fail to satisfy the principle) then not only is Pairing false; it could not possibly
have been true.
What I call the Standard Conception of metaphysical necessity extends
this familiar thought to a range of synthetic claims in metaphysics. As another
example, consider classical mereology. Once again, some of the axioms are
plausibly analytic. ‘A is part of A’; ‘If A is part of B and B of C, then A is part of
C’. But the ‘analytic core’ of the theory does not entail that composite things
exist, or that they must exist given the existence of at least objects. It says (in
effect) that a relation counts as the mereological part-whole relation only if it
is transitive and reflexive. But it does not say whether two things ever manage
to stand in this relation.
By way of contrast, consider the axiom that gives the theory its teeth.
UMC: Whenever there are some things, there is something that they
compose (where the Fs compose X iff every F is part of X and
every part of X overlaps an F ).
UMC is not a conceptual truth. Given anodyne input it delivers an entity
composed of my head and your body, Cleopatra’s arms and Nixon’s legs.
And whatever one thinks of such scattered monstrosities, it is not a sign of
logicolinguistic confusion to reject them.5 Nor is it true in virtue of the nature
of the part–whole relation. Once again, a conditional version of the principle
might be accorded such a status, viz.: If there are mereological aggregates, then whenever
there are some things, there is something they compose. But you can know perfectly well
what a mereological aggregate is supposed to be (as the opponents of classical
mereology clearly do) without being in a position to assert the unconditional
version of UMC.
UMC and pairing have at least this much in common. (a) They are
substantive principles. They can be rejected without self-contradiction or
5
For more on the epistemological status of UMC, see Dorr and Rosen (2001).
absurdity. (b) They entail the existence of a distinctive sort of object (perhaps
conditionally on the existence of things of some other, more basic, sort).
(c) Their epistemological status is uncertain, but they are palpably more apriori
than aposteriori. If they are empirical truths they are empirical truths of a
peculiar sort, since it is hard to imagine a course of experience that would
bear differentially upon their acceptability. (d) They concern matters of basic
ontology. Unlike the principles of ornithology, for example, they are not
concerned with what exists hereabouts. To put the point somewhat grandly,
they concern the structure of the world, not just its inventory. (e) They
are standardly regarded as metaphysically non-contingent. Philosophers have
questioned the existence of sets and mereological aggregates. But hardly
anyone has suggested that the basic principles governing such things might
have been other than they are.
These are some central features of what I shall call the Differential Class:
the class of claims with respect to which the two conceptions of metaphysical
necessity will diverge. On the Standard Conception, the synthetic apriori truths
of basic ontology are always necessary. On the Non-Standard Conception,
as I shall call it, they are sometimes contingent.
This characterization of the Differential Class leaves much to be desired.
Matters will improve somewhat as we proceed. But for now it may help to list
some further examples.
Existence claims elsewhere in mathematics, e.g. the existential principles

of arithmetic and analysis.
Neo-Fregean abstraction principles of the form ‘F(a) = F(b) iff a and
b are equivalent in some respect’, e.g. ‘The temperature of a = the
temperature of b iff a and b are in thermal equilibrium.’
Meinongian abstraction principles to the effect that for any (suitably
restricted) class of properties, there exists an abstract entity (arbitrary
object, subsistent entity) that possesses just those properties.
Accounts of the ontological underpinnings of genuine similarity; e.g. the
neo-Aristotelian claim that whenever a and b are genuinely similar, they
have an immanent universal part in common.
Accounts of the ontological underpinnings of persistence through time,
e.g. the claim that whenever a persisting object exists at a time it has a
momentary part that exists wholly at that time.
Gideon Rosen / 21
In each of these domains we are concerned with synthetic, seemingly non-
empirical facts of metaphysics. The Standard Conception does not say which
claims are true in these areas. But it does say that the truth, whatever it is, could
not be otherwise. If Peter van Inwagen (1990) is right that a plurality of material
things constitute a single thing only whether their activity constitutes the life
of an organism, then the Standard Conception says this is so of necessity. If
Hartry Field (1989) is right that abstract objects do not exist, then according to
the Standard Conception, this sort of nominalism is a necessary truth.6
5. The Non-Standard Conception

The Standard Conception is familiar. Insofar as you have any use for the concept
of metaphysical necessity, it is probably your conception. The Differential Class
is a class of metaphysical principles par excellence, and we normally take it for
grantedthatmetaphysicshasa metaphysically non-contingentsubjectmatter.7
That’s what we think. But consider the Others: a tribe of outwardly compet-
ent philosophers whose contact with the mainstream has been intermittent
over the past (say) thirty years. The Others share our tradition and they are
concerned with many of the same problems. In particular, they take them-
selves to have absorbed the main lessons of the modal revolution of the 1960s.
Metaphysical modality is the modality that mainly interests them, and they
do not confuse it with analyticity and the other semantico-epistemological
modalities. When they introduce the notion to their students their informal
gloss is much like ours. In particular, they agree that the Kripkean ‘aposteriori’
necessities are paradigm cases of metaphysical necessity, along with the truths
of logic and the analytic truths more generally.
You’ve been looking in on the Others, reading their journals, attending
their conferences; and so far as you can tell they might as well be some of
6
See also Field (1993). As noted above, Field himself rejects the notion of metaphysical necessity.
For him, modal questions about abstract objects can only be questions about the conceptual
modalities. Since Field himself regards both nominalism and its negation as non-self-contradictory
in the relevant sense, he regards the existence of mathematical objects as a contingent matter.
7
With some exceptions. It is widely acknowledged, for example, that the debate over
materialism (or physicalism) concerns a contingent proposition. (See Lewis 1983.) The suggestion
to follow is that much of what passes under the name ‘ontology’ might be understood in a similar
spirit.
Us. But now you see something that makes you wonder. In the philosophy
of mathematics seminar, Professors P and N disagree about whether sets exist.
According to P, the utility of set-theoretic mathematics gives us reason to
believe the standard axioms. N agrees that set theory is useful, but points out
that it is just as useful in worlds without sets as it is in worlds that have them,
and so maintains, on grounds of economy, that set theory is best regarded as
a useful fiction. You’ve heard most of this before, but you are struck by the
suggestion that sets might exist in some worlds but not in others.8 You know
that we sometimes indulge in loose talk of this sort amongst ourselves. But you
want to know whether N takes the idea seriously. So you ask, and she answers:
‘I meant exactly what I said. Platonism may be profligate, but it is not
incoherent or self-contradictory. I can conceive a world in which sets exist.
I can conceive a world in which they don’t. Each view thus corresponds to
a metaphysical possibility. There might have been sets, but then again, there
might not have been. The only question is which sort of world we inhabit.
That’s what my colleague and I disagree about.’
You are flabbergasted—not simply by the suggestion that the truths of
mathematics might be contingent, but by the blithe transition from a claim of
conceivability to a claim of metaphysical possibility. You point out that we’ve
known for years that the inference from conceivability to possibility is no good.
‘There is no incoherence in the supposition that water is an element,’ you say.
‘But even so, we know that water could not possibly have been an element.
You agree about this. So how can you be so blasé about the corresponding
inference in the case of sets?’
‘Ah, but the cases are very different,’ says N. ‘The ancients could see no
incoherence in the supposition that water is an element. Indeed, insofar as
they had reason to believe that water was an element, they had reason to
believe that there was no such incoherence. Perhaps this gives a sense in which
it was conceivable for the ancients that water should have been an element. And if
so, we agree: that sort of conceivability does not entail possibility. But when
I say that a world containing sets is conceivable, I have in mind a somewhat
different sort of conceivability. I’m talking about what we call informed or
correct conceivability. Here’s the idea:’
‘If the ancients could conceive a world in which water is an element, this is
only because they were ignorant of certain facts about the natures of things.
8
See van Fraassen (1977).
Gideon Rosen / 23
In particular, it is because they did not know what it is to be water. They did
not know that to be water just is to be a certain compound of hydrogen
and oxygen—that to be a sample of water just is to be a quantity of matter
predominantly composed of molecules of H2 O. This is not to say that they
did not understand their word for water. But it’s one thing to understand a
word, another to know the nature of its referent. The ancients could see no
contradiction in the supposition that water is an element because they did not
know that water is a compound by its very nature. But we know this; and given
that we do, we can see that to suppose a world in which water is an element is
to suppose a world in which a substance that is by nature a compound is not a
compound. And that’s absurd.’
‘In one sense of the phrase, P is conceivable for X if and only if that X can see
no absurdity or incoherence in the supposition of a world in which P is true.
Correct conceivability begins life as an idealization of this notion of relative
conceivability. To a first approximation, P is correctly conceivable iff it would
be conceivable for a logically omniscient being who was fully informed about
the natures of the things. The mind boggles at this sort of counterfactual,
to be sure. But once we see what it amounts to, we can see that it is merely
heuristic. If it’s true that an ideally informed conceiver would see no absurdity
in the supposition of a P-world, this is because there is no such absurdity
to be seen. The ideally informed conceiver is simply an infallible detector of
latent absurdity. And once we see this we can drop the reference to the ideal
conceiver altogether.’
‘As we understand the notion, metaphysical possibility is, as it were, the
default status for propositions. When the question arises, ‘‘Is P metaphysically
possible?’’ the first question we ask is ‘‘Why shouldn’t it be possible?’’ According
to us, P is metaphysically possible unless there is some reason why it should not
be—unless there is, as we say, some sort of obstacle to its possibility. Moreover,
the only such obstacle we recognize is latent absurdity or contradiction.9 If the
9
What is an ‘absurdity’ in the relevant sense? For present purposes, it will suffice to take an
absurdity to be a formal contradiction: a proposition of the form P & not-P, or a = a. [This assumes
that propositions, as distinct from the sentences that express them, may be said to have a ‘form’.]
A complication arises from the fact that not everyone agrees that contradictions are absurd in the
relevant sense. Dialethists maintain that some contradictions are not manifestly absurd. Nearly
everyone else disagrees. This proponent of the Non-Standard Conception may remain neutral on
this point. His fundamental contention is that a proposition is metaphysically impossible when
it entails a manifest absurdity or impossibility. For the purposes of exposition, I assume that this
question arises, ‘‘Why shouldn’t there by a world at which P is true?’’ the only
answer cogent response is a demonstration that the supposition that there is
such a world involves a contradiction or some other manifest absurdity. (This
is tantamount to a principle of plenitude. It has the effect that the space of
possible worlds is as large as it can coherently be said to be.) Now, whether
P harbors an absurdity is not in general an apriori matter. To say that it does
is to say that P logically entails an absurdity given a full specification of the
natures of the items it concerns. And since these natures are often available
only aposteriori, it is often an aposteriori matter whether P is correctly conceivable.
That’s why it is often an aposteriori matter whether P is metaphysically possible.’10
The Others have adopted the Non-Standard Conception of the metaphysical
modalities. According to this conception, correct conceivability—logical
consistency with propositions that express the natures of things—is both
necessary and sufficient for metaphysical possibility. This need not be construed
as a reductive analysis. It may be that the full account of correct conceivability
must make use of metaphysical modal notions.11 But even if the equivalence
is not reductive, it may nonetheless be true. And if it is then it would appear
to yield a series of deviant verdict about the Differential truths.
Consider the Pairing axiom once again. The axiom and its denial are
both logically consistent. Moreover, it is plausible that both are correctly
conceivable. To be sure, we have no adequate conception of what it is to be
a set. But even in the absence of a fully explicit such conception, we can
amounts to entailing a contradiction. But this is not strictly speaking a commitment of the view.
A complete account of the Non-Standard Conception would involve an account of the more
fundamental notion.
10
The Non-Standard Conception presented here is inspired by some remarks of Kit Fine.
Fine (1994) defines metaphysical necessity as truth in virtue of the natures of things. However,
Fine would not agree that the account is revisionary in the ways I have suggested. In particular,
he would not agree that the account entails that the existential truths of mathematics and
metaphysics are uniformly contingent. The question would seem to come down to whether
natures are to be construed as ‘conditional’ or ‘Anti-Anselmian’ (see below): whether it can lie in
the nature of some thing that it exist, or whether it can lie in the nature of some kind that it have
instances whenever some more basic kind has instances. I am grateful to Fine for conversation
on these questions and for his eye-opening seminar at Princeton in 1999. But he would certainly
resist my abuse of his ideas in the present context.
11
The account of correct conceivability involves three ingredients: the notion of a proposition,
the notion of logical entailment among propositions, and the notion of an absurdity or contradiction. It may well
be that a correct account of some or all of these notions presupposes the notion of metaphysical
necessity.
Gideon Rosen / 25
consider the alternatives. It is hardly plausible that it lies in the nature of the
set-membership relation to violate the Pairing axiom.12 So either the nature of
the relation is silent on whether Pairing is true, or it lies in the nature of the
relation to satisfy the principle. In the former case it is automatic that both
the axiom and its negation are correctly conceivable. So the relevant case is
the latter. Here is what the Others have to say about it.
‘If it lies in the nature of the sets (or the relation of set-membership) to
conform to Pairing, then it is indeed incoherent to suppose a world in which
sets exist and Pairing is false. But that is not the negation of Pairing. The negation
of the principle amounts to the claim that either there are no sets, or sets exist and some
things X and Y lack a pair set. Our claim that the negation of the axiom is correctly
conceivable depends on the thought that no contradiction follows from the
supposition that sets do not exist. Because we can see no such absurdity, and
we can’t see how more information about the natures of the items in question
could make a difference, we conclude that the Pairing axiom and its negation
both correspond to genuine possibilities.’
This little speech brings out an important feature of the Others’ talk of
natures and essences. For the Others, all natures are conditional or Kantian or
perhaps Anti-Anselmian. To say that it lies in the nature of the Fs to be G is
to articulate a condition that a thing must satisfy if it is be an F. It is to
give a (partial) account of that in virtue of which the Fs are F. It is not obviously
incompatible with this interpretation that existence (or existence given the
existence of things of some more basic kind) should be part of the nature of
a thing or kind. But even when it is, it will not be incoherent to deny the
existence of that thing or kind (or to deny it when the alleged condition has
been satisfied). As Kant says in a somewhat different context:
If, in an identical proposition, I reject the predicate while retaining the subject,
contradiction results . . . But if we reject subject and predicate alike, there is no
contradiction; for nothing is then left that can be contradicted. To posit a triangle
and yet to reject its three angles is contradictory; but there is no contradiction
in rejecting the triangle together with its three angles. (Critique of Pure Reason,
A 595/B623)
12
Properly formulated. If it lies in the nature of set-membership that if sets exist then von
Neumann-style proper classes exist as well, then the unrestricted version of Pairing given in the
text will be ruled out by the nature of the membership relation.
In a similar spirit the Others say: If it lies in the nature of God to exist (or to
exist necessarily), then to posit God and yet to reject his (necessary) existence
is absurd. But there is no contradiction in rejecting God altogether. And
similarly, if it lies in the nature of the sets to satisfy Pairing, then to posit a
system of sets and yet to reject Pairing is absurd; but there is no contradiction in
rejecting the sets along with Pairing. So even if Pairing is somehow constitutive
of what it is to be a set, its negation is nonetheless correctly conceivable and
therefore possible.
Let’s consider one more application of the Non-Standard Conception, this
time to a thesis about the constitution of ordinary particulars. D. M. Armstrong
has long maintained that whenever two particulars resemble one another, this
is because they share an immanent universal as a common part (Armstrong
1978). Let us grant the coherence of the very idea of an immanent universal,
wholly located in distinct particulars. In fact, let us grant that in the actual world
similarity works as Armstrong says it does. The question will then be whether
it is absurd to suppose a world in which qualitative similarity is secured by
some other mechanism: e.g. a world in which similar particulars are similar
because they contain exactly resembling tropes, or because they instantiate
one or another primitive similarity relation. For the sake of argument, we
may suppose that these alternative theories are not conceptually confused or
self-contradictory.13 On the Non-Standard Conception, the suggestion that
they are nonetheless impossible must then amount to the claim that they are
incompatible with the nature of qualitative similarity or some other item. But
is that plausible?
We have assumed that in the actual world qualitative similarity works
as Armstrong says it does. And in light of this, someone might say, ‘So
that is what qualitative similarity turns out to be. This is not an analytic
matter; and it is not exactly an empirical matter either. But it is nonetheless
the case that for two particulars to be similar just is for them to share an
13
Once again, it is hard to know whether this is the case. The alternatives have not been
developed in sufficient detail. However, the arguments typically brought against these and other
proposals do not purport to show that the accounts are straightforwardly contradictory or
incoherent. They purport to show that they are uneconomical, or implausible, or less explanatory
than the alternatives, and so on. It is just barely possible that in the theory of universals there
is in the end exactly one coherent (non-self-contradictory) position. If so, then the Standard
Conception and the Non-Standard Conception will concur in calling it necessary. If not, then
the two conceptions will diverge. The true account will be necessary in the Standard sense but
contingent in the Non-Standard sense.
Gideon Rosen / 27
immanent universal as a common part.’ If this were correct, then in this case
the Non-Standard Conception would support the orthodox verdict that the
correct metaphysical account of similarity in the actual world amounts to a
metaphysically necessary truth.
We cannot rule this out without further investigation. But it is implausible
on its face. Note that nothing in the story rules out worlds in which something
like the trope theory or the primitive resemblance theory is correct: worlds in
which there are no immanent universals wholly present in their instances, but
in which particulars stand in relations of (let us say) quasi-similarity by virtue of
satisfying one of these alternative theories. These quasi-similar particulars may
look (quasi-)similar to observers. They may behave in (quasi-)similar ways in
response to stimuli. They may be subject to (quasi-)similar laws. The proposal
under consideration nonetheless entails that they are not really similar: that
quasi-similarity stands to genuine similarity as fool’s gold stands to gold, or as
Putnam’s XYZ stands to water. But on reflection this seems preposterous. If it
walks like similarity and quacks like similarity then it is (a form of) similarity.
If you were deposited in such a world (or if you could view it through your
Julesvernoscope) and were fully informed both about its structure and about
the structure of the actual world, would you be at all tempted to conclude
that over there nothing resembles anything else? Surely not.
Suppose that’s right. Then the various metaphysical accounts are all
compatible with the nature of the similarity relation. The true theory (namely,
Armstrong’s) tells us how similarity happens to be grounded. It describes the
mechanism by which similarity is secured in the actual world, much as the atomic
theory of fluids describes how fluidity happens to be realized in this world.
But it goes well beyond a specification of the underlying nature of similarity.
And if that’s right—if the nature of similarity is in this sense thin—then
the alternatives may be correctly conceivable, in which case they represent
genuine possibilities according to the Others.
We should pause to note a peculiar consequence of the Non-Standard
Conception. The view suggests that many of the synthetic propositions of
fundamental metaphysics are metaphysically contingent. But it does not say
that these propositions are unknowable, or that they can only be known
empirically. To the contrary, nothing in the view is incompatible with
the thought that the powerful methods of analytic metaphysics supply an
altogether reasonable canon for fixing opinion on such matters. Now analytic
methodology is for the most part an apriori matter. If the doctrine of immanent
universals is to be preferred as an account of qualitative similarity, this is
because it is elegant, intrinsically plausible, philosophically fruitful, immune
to compelling counterexample, and so on. All of these features are presumably
available to apriori philosophical reflection insofar as they are available at all.
The view therefore yields a new species of the so-called ‘contingent apriori’. One
need not appeal to claims involving indexicals (‘I am here now’) or stipulative
reference fixing (‘Julius invented the zip’). According to the Others, the claims
of basic ontology (including the existential claims of mathematics), are both
contingent and apriori (insofar as they are knowable); but in this case the
mechanism has nothing to do with indexicality.14
6. The Two Conceptions and the Informal Explanation

Let us suppose—just for a moment—that the Non-Standard Conception is
tolerably clear, in the sense that there might a coherent practice in which pro-
positions are classified as ‘necessary’ or not depending on whether their nega-
tions are correctly conceivable. One might object that a notion of this sort, how-
ever interesting, does not deserve the name ‘metaphysical necessity’. After all,
the main controls on this notion are supplied by the informal elucidation with
which we began. A modal notion deserves to be called ‘metaphysical’ only to the
extentthatitconformstothisaccount. Andthe Non-Standard Conception falls
short in one obvious respect. We explain what we mean by ‘metaphysical neces-
sity’ in part by holding up the truths of mathematics and fundamental ontology
and saying, ‘You want to know what metaphysical necessity is supposed to be?
It’s the sort of necessity that attaches to claims like that.’ Since the Non-Standard
Conception threatens to classify many of these paradigms as ‘contingent’, this
counts against regarding it as a conception of metaphysical necessity.
14
Note that if these mathematical and metaphysical truths are indeed both apriori and
contingent, then the warrant for them (whatever it comes to) will presumably be available even
in worlds where they are false. Apriori warrant is therefore fallible: an interesting result, but not a
problem. Compare the force of considerations of simplicity in the empirical case. We are supposed
to have reason to believe the simplest theory simply in virtue of its simplicity; but there are
deceptive worlds in which the simplest empirically adequate theory is wildly false. This does not
show that simplicity is not a reason for empirical belief; it just shows that in deceptive worlds a
belief can be both false and justified. The present picture supports a similar conception of (one
sort of) apriori warrant. Thanks to a referee for Oxford University Press on this point.
Gideon Rosen / 29
The charge is one of terminological impropriety, and as such it is ultimately
inconsequential. But it seems to me that the Others have a telling response
nonetheless. They may say, ‘Tu quoque. Our notion may not fit your informal
explanation to the letter. But neither does yours. We think we know what
you mean by ‘‘metaphysical necessity’’. At any rate, we can construct a modal
notion much like yours, relative to which the Differential truths are clearly
necessary. But it is a restricted necessity, on a par with physical necessity. As
we normally think, the laws of physics are metaphysically contingent: true in
some genuinely possible worlds, false in others. But they are also necessary in a
sense: true in each of a distinguished subclass of worlds. By our lights, what you
call ‘‘metaphysical necessity’’ has a similar status. It does not amount to truth
in every genuinely possible world, but rather to truth in each of a distinguished
subclass of worlds: the worlds compatible with the basic facts—or perhaps one
should say laws—of metaphysics: the most fundamental facts about ‘‘what
there is and how it hangs together’’. This hypothesis squares brilliantly with
your taxonomic practice. But it is at odds with the idea that the metaphysical
modalities differ from the physical modalities in being unrestricted.’
Theinformalelucidationincludestheclaimthatthemetaphysicalmodalities
are absolute among the real modalities. The Non-Standard Conception
appears to satisfy the condition. It is certainly less restrictive than the Standard
Conception, and it is hard to think of a natural modal conception of the
relevant sort that is less restrictive.15 So if the Non-Standard Conception sins
against the informal elucidation by reclassifying some of the paradigms, the
Standard Conception sins against the absoluteness clause. This is the basis for
my suggestion that while neither conception fits the informal explanation to
the letter, both conceptions fit it well enough, and so bear roughly equal title
to the name ‘metaphysical modality’.
7. Is the Non-Standard Conception Coherent?

All of this assumes, of course, that the two conception are genuinely tenable.
There are questions on both sides. Let’s begin with objections to the Non-
Standard Conception.
15
It is easy to construct gerrymandered up real modalities that are less restrictive.
The Others claim that apart from its heterodox classification of claims in
the Differential Class, the Non-Standard Conception amounts to a recogniz-
able conception of metaphysical modality. But there are reasons to doubt
this, some of which are quite familiar. Consider the following exemplary
challenge.
Let God be Anselm’s God—a necessarily existing perfect spirit—and

consider the proposition that God exists. It is not incoherent to suppose
there is a God; and pace Anselm, it is not incoherent to suppose there
is not. The Non-Standard Conception therefore entails that Anselm’s
God is a contingent being. But that’s absurd. If Anselm’s God exists at
some world, He exists at all worlds by His very nature. So the Non-Standard
Conception is incoherent. It entails that God’s existence is both necessary
and contingent.
There are several ways to approach the problem, some of which would
require substantial modification in the Non-Standard Conception. These
modifications may be independently motivated. But it seems to me that the
view has the resources to evade this particular problem as it stands.
Let’s begin with a question. Anselm’s God is supposed to be a necessary
being. But necessary in what sense? If he is supposed to be necessary in the
Standard sense, there is no problem. It might well be a contingent matter in
the Non-Standard sense whether the basic laws of metaphysics require the
existence of a perfect spirit, just as it may be metaphysically contingent in
the Standard sense whether the laws of physics require the existence of (say)
gravitons.
But it’s not very Anselmian to suppose that God’s perfection involves
only Standard necessary existence. Surely, ’tis greater to exist in every genuinely
possible world than merely to exist in every world that resembles actuality in
basic respects. So if we admit the Non-Standard Conception, it will be natural
to suppose that God’s existence is supposed to be Non-standardly necessary. But
in that case we can afford to be less ecumenical. What would a necessary being
in the Non-Standard sense have to be like? It would have to be a being whose
non-existence is not correctly conceivable, which is to say: a being whose
non-existence together with a complete specification of the (conditional,
Kantian, anti-Anselmian) natures of things logically entails a contradiction or
some similar absurdity. But upon reflection it seems clear that there can be no
Gideon Rosen / 31
such thing. The Anti-Anselmian natures of things are given by formulae of
the form:
To be an F is to be . . .
To be A is to be . . .
But it seems clear that no collection of such formulae can yield a contradiction
when conjoined with a negative existential proposition of the form
There are no Fs, or.
A does not exist
The proposition that a Non-Standard necessary being sense exists is thus
incoherent; it is not correctly conceivable. The proponent of the Non-Standard
Conception may therefore resist the objection.
The same response applies to non-theological versions of the objection. It
is sometimes said, for example, that the idea of Number includes the idea
of necessary existence, so that nothing counts as a number unless it exists
necessarily. (Of course, the textbook definitions tend to omit this condition,
just as they omit to mention that numbers do not exist in space and time.
But still it might be said that our ‘full conception’ of the natural numbers
entails that numbers exist necessarily if they exist at all.)16 The worry is
that the Non-Standard theorist will be forced to concede that is coherent to
suppose that numbers so-conceived exist, and also that it is coherent to deny
their existence, in which case it will follow, absurdly, that numbers are both
necessary and contingent.
The response is to distinguish two senses in which numbers might be said
to be necessary. If the claim is that numbers, if they exist, must be necessary
in the Standard sense, then once again there is no problem. It might be
contingent in the Non-Standard sense whether some Standardly Necessary
Being exists. On the other hand, if the claim is that numbers must be necessary
in the Non-Standard sense, then we may conclude straight away that numbers
so-conceived are impossible, since it is not correctly conceivable to suppose
that they exist.
As a final example, consider the claim that there exists an actual golden
mountain. Since there is no golden mountain in the actual world, we know that
this proposition is not possibly true. But is the proponent of the Non-Standard
Conception entitled to this verdict? Is the supposition of a world in which
16
Balaguer (1998).
there exists an actually existing golden mountain logically incompatible with
the natures of things? Couldn’t you know all there was to know about what
it is to be gold, what it is to be a mountain, and what it is to be actual without
being in a position to rule out the existence of an actual golden mountain?
No. For there to be an actual golden mountain is for there to be a golden
mountain in the actual world. And in the relevant sense, the actual world
has its complete intrinsic nature essentially. To be the actual world is to
be a world such that P, Q, . . . where these are all the contingently true
propositions. Propositions of the form ‘Actually P’ are singular propositions
about this world and will thus be true (or false) in virtue of the nature
of the actual world. It follows that for propositions of this sort, the Non-
Standard Conception agrees with the Standard one. All such propositions are
metaphysically non-contingent.
8. Objections to the Standard Conception

There is much more to say about whether the Non-Standard Conception
represents a tenable conception of the metaphysical modalities.17 But if we
suppose that it does, then our critical focus naturally shifts to the Standard
Conception. For once we have the Non-Standard Conception clearly in focus,
it is no longer obvious that the Standard Conception represents a genuine
alternative. A skeptic might suggest that it was just thoughtless acquiescence
in tradition that led us to regard the substantive principles of fundamental
ontology as metaphysically necessary according to our usual understanding
of the notion. After all, if there really is no obstacle to the possibility of a world
in which (say) mereological aggregates do not exist, is it really so obvious that
such worlds should be deemed impossible? Presumably, we have never faced
17
In his very useful comments on an earlier version of this paper, Scott Sturgeon objected
to the Non-Standard Conception on the ground that David Lewis’s theory of possibility—his
version of modal realism—and its negation are both correctly conceivable, whereas it is absurd to
suppose that a modal account of this sort might be a contingent truth. In response, I am inclined
to say that Lewis’s metaphysics of many worlds, shorn of its modal gloss, is indeed contingent in
the Non-Standard sense, and that no contradiction follows from this concession. On the other
hand, Lewis’s package includes account of what it is for a truth to be necessary, and that account is
either compatible with the nature of necessity (in which case the negation of Lewis’s theory is an
impossibility) or incompatible with it (in which case Lewis’s theory itself is an impossibility).
Gideon Rosen / 33
the question directly. And it is tempting to suppose that when we do, our
reaction should be not to reaffirm the Standard verdict, but rather to conclude
that what I have been calling the Non-Standard conception really is our own
conception and that we have been systematically misapplying it in such cases.
Tobe sure, even given the tenability of the Non-StandardConception, we still
know how to classify truths as necessary or contingent in the Standard sense.
We still know how to identify the truths (or putative truths) of fundamental
ontology, along with the uncontroversial metaphysical necessities. That is, we
know how to apply the Standard Conception in practice. So never mind what
we would say if we were to confront the question sketched above. Is there any
reason to doubt that the Standard Conception as I have described it tracks a
perfectly genuine modal distinction (even if it is not the only such distinction
in the neighborhood?)
Let’s not deny that it tracks a distinction. The question is whether that
distinction amounts to a distinction in modal status. Let me explain.
As we have seen, from the standpoint of the Non-Standard Conception,
Standard metaphysical necessity is best seen as a restricted modality. To be
necessary in the Standard sense is to hold, not in every genuinely possible
world, but rather in every world that meets certain conditions. Now it is
sometimes supposed that restricted modalities are cheap. After all, given any
proposition, φ we can always introduce a ‘restricted necessity operator’ by
means of a formula of the form
φ (P) =df (φ → P).
And in that case, there can be no objection to the Standard Conception. The
trouble is that most such ‘restricted necessity operators’ do not correspond to
genuine species of necessity. Let NJ be the complete intrinsic truth about the
State of New Jersey, and say that P is NJ-necessary just in case NJ strictly implies
P. It will then be NJ-necessary that Rosen is in Princeton, but NJ-contingent
that Blair is in London. But of course we know full well that there is no sense
whatsoever in which I have my location of necessity while Blair has his only
contingently. So NJ-necessity is not a species of necessity.
The moral is that one cannot in general infer, from the fact that a certain
consequence (φ → P) holds of necessity, that there is any sense in which the
consequent (P) holds of necessity. (If there were then every proposition would be
necessary in a sense, even the contradictions.)
Now, metaphysical necessity on the Standard Conception is supposed to be
a restriction of Non-Standard metaphysical necessity for which the restricting
proposition φ is the conjunction of what we have been calling the ‘laws’
of metaphysics. The challenge is thus to show that Standard necessity so
conceived amounts to a genuine species of necessity—that it is more like
physical necessity than it is like NJ-necessity.
It is unclear what it would take to meet this challenge. There is some
temptation to say that φ-necessity amounts to a genuine species of necessity
only when the restricting proposition φ has independent modal force—only when
there is already some sense in which it must be true. But what could this mean?
Consider the Mill–Ramsey–Lewis (MRL) account of the laws of nature,
according to which a generalization L is a law just in case L is a theorem of
every true account of the actual world that achieves the best overall balance
of simplicity and strength (Lewis 1973). Let us grant that this standard picks
out a tolerably well-defined class of truths. Still, one might ask, ‘Why should
propositions incompatible with the laws so conceived be called impossible?’
Consider a related class of truths: those propositions that would figure in every
true account of the State of New Jersey that achieves the best overall balance
of simplicity and strength. If the Encyclopedia Britannica is any guide, one such
truth is the proposition that New Jersey is a haven for organized crime. But
one needs a dark view of things to suppose that this proposition is in some
sense necessary. It certainly doesn’t follow from the fact that it is important
enough to be worth mentioning in a brief account of New Jersey that it enjoys
a distinctive modal status. So why is it than when the MRL-theory in question
is a theory about the entire world, we are inclined to credit its general theorems
with some sort of necessity?
One way with this sort of question is a sort of nominalism. There no
objective constraints on which restricted necessities we recognize. We take
an interest in some but not in others. We hold their associated restricting
propositions fixed in counterfactual reasoning for certain purposes. And in
these cases we dignify the operator in question with a modal name. But our
purposes might have been otherwise, and if they had been then we might have
singled out a different set of operators. On this sort of view there can be no
principled objection to the Standard Conception. The worst one can say is that
the restricted necessity upon which it fastens is not particularly interesting or
useful. But one cannot say that it fails to mark a genuine modal distinction,
Gideon Rosen / 35
for on the view in question any modal distinction we see fit to mark as such is
ipso facto genuine.
If we set this sort of nominalism to one side, then one natural thing to say is
that a putative restricted necessity counts as genuine only when the boundary
it draws between the necessary and the contingent is non-arbitrary or non-
ad hoc from a metaphysical point of view. (Note that this is at best a necessary
condition.) The truths about NJ are not a natural class from the standpoint of
general metaphysics; nor are the Mill–Ramsey–Lewis generalizations about
New Jersey. On the other hand, the most important general facts about nature
as a whole may well be thought to constitute a metaphysically significant class
of facts. And if so, there would be no objection on this score to the idea that
physical necessity defined in Lewis’s way amounts to a genuine species of necessity.
The Standard Conception of metaphysical necessity conditionalizes upon
what we have been calling the basic laws or facts of fundamental ontology. Just
as the Mill–Ramsey–Lewis laws of nature are supposed to represent the goal
of one sort of natural science, the metaphysical laws are supposed to represent
the goal of one sort of metaphysics: nuts and bolts systematic ontology.
Clearly, there is no worry that these truths might constitute an arbitrary class
from the standpoint of metaphysics. But it might still be wondered whether
anything substantial can be said about what unifies them, and in particular,
about what fits them to serve in the specification of a restricted modality.
I have a conjecture (and some rhetoric) to offer on this point. Consider the
true propositions in the Differential class: the truths in the theory of universals
and the metaphysics of material constitution; the truths about how abstract
entities of various sorts are ‘generated’ from concrete things and from one
another. To know these truths would not be to know which particulars there
are or how they happen to be disposed in space and time. But it would be to
know what might be called the form of the world: the principles governing how
objects in general are put together. If the world is a text then these principles
constitute its syntax. They specify the categories of basic constituents and
the rules for their combination. They determine how non-basic entities are
generated from or ‘grounded in’ the basic array. Worlds that agree with the
actual world in these respects, though they may differ widely in their ‘matter’,
are nonetheless palpably of a piece. They are constructed according to the
same rules, albeit in different ways, and perhaps even from different ultimate
ingredients. In this sense, they are like sentences in a single language. The
metaphysically necessary truths on the Standard Conception may not be
absolutely necessary. But they hold in any world that shares the form of the
actual world in this sense.
Combinatorial theories of possibility typically take it for granted that
the combinatorial principles characterize absolutely every possibility: that
possible worlds in general share a syntax, as it were, differing only in the
constituents from which they are generated or in the particular manner or
their arrangement. The Non-StandardConception isnot strictly combinatorial
in this sense, since it allows that the fundamental principles of composition—the
syntax—may vary massively from world to world. The actual grammar is
not privileged. Any coherent grammar will do. But the Standard Conception
carves out an inner sphere within this larger domain: the sphere of worlds that
share the combinatorial essence of actuality. As I have stressed, it is unclear
what it takes to show that a class of truths is sufficiently distinguished to count
as a legitimate basis for a restricted modality. Nonetheless, the foregoing may
be taken to suggest that if any restricted modality is to be reckoned genuine,
the restricted modality marked out by Standard Conception should be so
reckoned.
9. Physical Necessity Reconsidered

This way of thinking raises a question about the boundary between physical
necessity and Standard metaphysical necessity. Some physical necessities
will presumably be Standardly contingent. Suppose the laws of nature
involve particular numerical constants that determine the strengths of the
fundamental forces or the charges or masses of the fundamental particles.
It will then be natural to suppose that the precise values of these constants
are not aspects of the general combinatorial structure of the world and that
they are therefore contingent in the Standard sense. But other claims that
might feature in the Mill–Ramsey–Lewis theory of the natural world might
be candidates for metaphysical necessity in the standard sense: that the laws
of nature all assume a certain mathematical form (e.g. that they are quantum
mechanical); that the space–time manifold has certain geometrical features,
e.g.: that it has only one ‘time’ dimension; that the ultimate particles are
excitation states of one-dimensional strings; and so on. It is not inconceivable
that such physical features should be sufficiently basic to count as aspects of
the underlying form or structure of the world: that any world in which such
Gideon Rosen / 37
physical features failed to be manifest, would fail to share a syntax with the
actual world. And insofar as this is so, these physical truths should be reckoned
metaphysically necessary on the Standard conception for the same reason
that the facts of fundamental ontology are to be reckoned necessary on that
conception.
The point I wish to stress, however, is that on the present conception it is
to be expected that the border between Standard metaphysical necessity and
physical necessity should be vague—not simply because the notion of physical
necessity (or a law of nature) is vague, but also because it is vague when a truth
is ‘fundamental’ or ‘structural’ enough to count as part of the combinatorial
essence of the world. This is not the prevailing view on this matter. Most
writers take it for granted that the question whether a certain law of nature
is also metaphysically necessary is a well-defined question whose answer is in
no way up for stipulation. On the present conception, that is unlikely to be
the case. If the question is whether some given law of nature is a Non-Standard
necessity, then indeed, for all we have said, it may be sharp. However hard it
may be to find the answer, the question then is whether the negation of the
law is ruled out by the natures of the properties and relations it concerns, and
we have seen no reason to believe that this question is a vague one. (There
may be such reasons, but we have not seen them.)18 On the other hand, if
the question is whether the law is a Standard metaphysical necessity, then we
should expect that in some cases it will have no answer, since the boundary
between structural or formal truths and mere ‘material’ truths has only been
vaguely specified.
10. Conclusion
We have distinguished two conceptions of metaphysical necessity, both of
which cohere well enough with the usual informal explications to deserve
the name. According to the Non-Standard Conception, P is metaphysically
necessary when its negation is logically incompatible with the natures of things.
According to the Standard Conception, P is metaphysically necessary when
18
For example, it might turn out to be a vague matter whether P holds in virtue of the nature
of things. This is immensely plausible when P is a proposition about a particular organism or a
biological species.
it holds in every (Non-Standard) possible world in which the actual laws of
metaphysics also hold, where the basic laws of metaphysics are the truths about
the form or structure of the actual world. Neither conception has received a
fully adequate explanation. But if both are tenable, then our discourse about
necessity is shot through with ambiguity. The ambiguity only matters when
we are discussing the modal status of metaphysical propositions—or perhaps
the modal status of certain laws of nature. But when it does matter, we ignore
it at our peril. We are inclined to believe that questions about the modal status
of the claims of mathematics and metaphysics are unambiguous. But if I’m
right, that is not so. In particular, it may be metaphysically necessary in one
sense that sets or universals or mereological aggregates exist, while in another
sense existence is always a contingent matter.
References
Armstrong, D. M. (1978), Universals and Scientific Realism, 2 vols. (Cambridge: CUP).
Balaguer, Mark (1998), Platonism and Anti-Platonism in Mathematics (Oxford: OUP)
Dorr, Cian, and Rosen, Gideon (2001), ‘Composition as a Fiction’, in R. Gale (ed.),
Blackwell’s Guide to Metaphysics (Oxford: Blackwell), 151–74.
Dummett, Michael (1973), Frege: Philosophy of Language (2nd edn., Cambridge, Mass.:
Harvard University Press).
(1981), The Interpretation of Frege’s Philosophy (Cambridge, Mass.: Harvard University
Press).
(1993), ‘Wittgenstein on Necessity: Some Reflections’, repr. in his The Seas of
Language (Oxford: OUP), 446–61.
Field, Hartry (1974), ‘Quine and the Correspondence Theory’, Philosophical Review 83:
200–28.
(1989), Realism, Mathematics and Modality (Oxford: Blackwell).
(1993), ‘The Conceptual Contingency of Mathematical Objects’, Mind, 102:
285–99.
Fine, Kit (1994), ‘Essence and Modality’, in J. E. Tomberlin (ed.), Philosophical Perspectives
8: Logic and Language (Atascadero, Calif.: Ridgeview), 1–16.
(2002), ‘The Varieties of Necessity’, in J. Hawthorne and T. Gendler (eds.),
Conceivability and Possibility (Oxford: OUP), 253–81.
Frankel, Abraham, Bar Hillel Yehoshua, and Levy, Azriel (1973), Foundations of Set
Theory (2nd edn., Amsterdam: North Holland).
Lewis, David (1973), Counterfactuals (Cambridge, Mass.: Harvard University Press).
Gideon Rosen / 39
(1983), ‘New Work for a Theory of Universals’, Australasian Journal of Philosophy, 61:
343–77.
Maddy, Penelope (1997), Naturalism in Mathematics (Oxford: OUP).
van Fraassen, Bas (1977), ‘Platonism’s Pyrrhic Victory’, in Marcus et al. (eds.), The
Logical Enterprise (New Haven: Yale University Press), 39–50.
van Inwagen, Peter (1990), Material Beings (Ithaca, NY: Cornell University Press).
2
Modal Infallibilism and Basic Truth
Scott Sturgeon
1. Introduction
This paper concerns intelligibility and possibility. It argues no linking principle
of the form
(L) Int(ø) ⇒ ø
is valid. The notion of intelligibility will vary throughout the discussion.
Details will be given as needed. The notion of possibility will remain constant.
Let me explain it now.
Modal operators will stand for genuine modality: ø will mean ø is genu-
inely possible; ø will mean ø is genuinely necessary. This will be our
analogue of metaphysical modality. The latter is oft equated with logical mod-
ality, when logic is ‘broadly construed’. But that suggests there is something
intrinsically apriori about metaphysical modality. I must blanche the discussion
of just that bias. So I use new terms. They leave open whether genuine modality
This paper is dedicated to David Lewis. It grew from a conference response to Gideon Rosen’s
‘The Limits of Contingency’. I thank Fraser MacBride for inviting that response, Rosen for his
thought-provoking paper, and MacBride again for inviting the expansion of my thoughts on the
day. My views of apriority and modality have evolved considerably since the conference. They can
be found in Epistemic Norms (Oxford: OUP, forthcoming). Finally: thanks to Dorothy Edgington,
Dominic Gregory, John Hawthorne, Jen Hornsby, Barry Lee, Stephan Leuenberger, Jonathan
Lowe, Fraser MacBride, Mike Martin, Gideon Rosen, Nick Shea, Maja Spener, Alan Weir, Tim
Williamson, Crispin Wright, and Dean Zimmerman for helpful feedback.
Scott Sturgeon / 41
plays a role in anything canonically apriori. By such modality I mean simply
this: the most absolute realistic modal space. The idea, of course, is twofold. Its
components can be glossed via possibility or necessity. Consider each in turn.
When ø is genuinely possible, it is a mind- and language-independent fact
that ø can happen.1 That fact does not spring from how we think or talk
(even in the rational ideal). Genuine possibility is like genuine actuality. It
does not depend on us for its existence. It does not depend on us for its nature.
It is a realistic domain. Realism is a component of genuine possibility. And so
is weakness: whenever there is a realistic sense in which ø can happen, ø is
genuinely possible. Such possibility is the weakest kind of realistic possibility. It
is entailed by every kind of realistic possibility. The ‘diamond face’ of genuine
modality is a two-part affair. It is the most inclusive realistic space of possibility.
Similarly: when ø is genuinely necessary, it is a mind- and language-
independent fact that ø must happen. That fact does not spring from how we
think or talk (even in the rational ideal). Genuine necessity is like genuine
actuality. It is a realistic domain. Realism is a component of genuine necessity.
And so is strength: when ø is genuinely necessary, it is necessary in any realistic
sense. Genuine necessity is the strongest kind of realistic necessity. It genuinely
entails every kind of realistic necessity. The ‘box face’ of genuine modality is a
two-part affair. It is the least inclusive realistic space of necessity.
Our topic is whether intelligibility marks genuine possibility. We will work
it in several stages. By way of motivation, Section 2 presents six claims of
Basic Metaphysics. The very last is got by detachment from an instance
of (L). Section 3 derives contradictions from the six. That prompts scrutiny of
(L). Section 4 considers and rejects several readings. Section 5 diagnoses their
defects. That leads to the best reading of (L). Section 6 contains its critique.
Section 7 closes with remarks on reason and modality.
2. Six Claims of Basic Metaphysics

(i) Instantaneous things last for an instant. Persisting things last longer. And
notionally at least, they might do so three ways. They might last solely
by composition from instantaneous things. They might last by no such
composition. They might combine parts drawn from each category. The first
1
Unless ø concerns mind or language, of course. I leave the point tacit.
42 / Modal Infallibilism and Basic Truth
method of persistence is perdurance. The second is endurance. Mixed persistence
has no common name.
Suppose O1 , . . . , On manifest condition C. Suppose their so doing has
nought to do with time or other things. It has just to do with them. In the
event, C is an intrinsic condition. It is an intrinsic property when n equals 1. It
is an intrinsic relation when n is larger. Our first claim of Basic Metaphysics
concerns intrinsics. Specifically, it concerns their shift. The key idea is simple.
Since intrinsics have only to do with what satisfies them, they cannot be
shifted. That would involve satisfying yet not satisfying a condition (which
cannot be). Intrinsics are thus for life. They cannot be temporary. Appearances
to the contrary mislead.
Yet appearances are to the contrary, as we will see. Intrinsics do seem to shift.
So how is the magic done? Through perdurance: seeming shift in intrinsics
is none but intrinsically varying temporal parts. Such parts do not shift their
intrinsics. But persisters into which they build may seem to; and they may do so
because their temporal parts intrinsically vary. Intrinsic change is thing-level
sleight of hand. It is done by change of intrinsically varying temporal parts.
This is a matter of Basic Metaphysics. Our first claim is thus
(1) Necessarily, intrinsic change comes via perdurance.
(ii) When intrinsic change is discussed, shape is the pat example. Lewis founds
orthodoxy thus:
Persisting things change their intrinsic properties. For instance shape: when I sit, I have
a bent shape; when I stand, I have a straightened shape. Both shapes are temporary
intrinsic properties; I have them only some of the time.2
And he polishes it thus:

When I change my shape, that is not a matter of my changing relationship to other
things, or my relationship to other changing things. I do the changing, all by myself.3
Shape is intrinsic. This is a matter of Basic Metaphysics. Our second claim is thus
(2) Necessarily, shape is intrinsic.
(iii) Intrinsics come in greater-than-one adicity. As Lewis notes when extending
orthodoxy:
2
On the Plurality of Worlds (Oxford: Blackwell, 1986), 204–5.
3
‘Rearrangement of Particles: Reply to Lowe’, in his Papers in Epistemology and Metaphysics
(Cambridge: CUP, 1999), 187.
Scott Sturgeon / 43
Exactly as some properties are just a matter of how the thing itself is, without regard
to any relationship to any second thing, so some relations are just a matter of how two
things are vis-à-vis one another, without regard to any relationship to any third thing.
The relation is intrinsic to the pair of relata. The ever-changing distances of particles
from one another seem to be temporary intrinsic relations.4
Distance between things is intrinsic. Our third claim of Basic Metaphysics

is thus
(3) Necessarily, distance is intrinsic.
(iv) Some things are none but the sum of their parts. And any collection seems
to bring one of them with it: namely, the sum of its members. The sum is
the least inclusive thing with each member as part. It is none but the sum
of its parts. The view that sums perforce spring from collections is mereological
universalism. And it too is a matter of Basic Metaphysics. Our fourth claim is thus
(4) Necessarily, for any O1 , . . . , On there is a sum
= [O1 + · · · + On ].
(v) If 1 and 2 have strictly identical parts, they are but one sum. After all,
sums are none but the sum of their parts. Strict identity of part ensures that
of sum. Our fifth claim of Basic Metaphysics is thus
(5) Necessarily, if 1 and 2 have strictly identical parts, 1 = 2 .
(vi) So far so good. (1)–(5) are popular claims. They yield conflict, though,
when married to this line from Lewis:
There might be things that endure identically through time or space, and trace out
loci that cut across all lines of [Humean] continuity. It is not, alas, unintelligible that
there might be suchlike rubbish. Some worlds have it.5
These are enduring runabouts. Lewis infers their possibility from their intelligib-
ility. And by that, of course, he means more than their understandability.
Impossibilities may be understandable enough. By intelligibility Lewis means
something like ‘free of conceptual difficulty after serious reflection’. We will
have more to say about that from Section 4 onward.
4
Ibid. 192–3.
5
Philosophical Papers (Oxford: OUP, 1986), vol. ii, p. x; see also ‘Rearrangement of Parti-
cles’, 195.
The possibility of enduring runabouts is a matter of Basic Metaphysics. Our
sixth claim is thus
(6) Enduring runabouts are possible.
3. Two Problems
Now we are in trouble. (6) interacts badly with the combination of (1), (2),
(4), and (5); and it interacts badly with that of (1) and (3). In the first case we
get the problem of shapeshifting. In the second we get that of placeshifting.
Consider each in turn.
Shapeshifting: (6) says endurers can run about. (4) guarantees that at each
moment they do so there is a sum composed just of them existing at
that moment. (5) guarantees it is one sum. Hence the sum endures. Yet its
parts runabout. The sum thereby changes shape. (1) and (2) thus ensure it
perdures. No good! Nothing can perdure and endure. The former is done
solely by composition from instantaneous things. The latter is done without
such composition. The combination of (1), (2), (4), (5), and (6) conflict. They
cannot all be true. This is the puzzle of shapeshifting.
One might deny, of course, that enduring runabouts can shapeshift their
sum. One might say they can only move so as to preserve sum shape. That
would dissolve the puzzle. But it would do so by fiat. It would be too ad hoc
for systematic philosophy. We should insist: if endurers can be at all, they can
run about so as to shapeshift their sum.
To reinforce the worry, let be a sum of enduring runabouts. Suppose it
changes shape. Now echo Lewis:
When changes its shape, that is not a matter of its changing relationship
to other things, or its relationship to other changing things. does the
changing, all by itself.
At first blush this looks incredible. After all, ’s changing shape is a matter of
its runabout parts. They change its shape. The shift is precisely a matter of ’s
relationship to changing things. It is a matter of ’s relationship to runabout
parts. Yet none of those parts is identical to . There is a good sense, then, in
which ’s shapeshift is due to its relation to other changing things.
At second blush one can make out Lewis’s line. For as he puts it: ‘the ‘‘are’’
of composition is just the plural of the ‘‘is’’ of identity.’6 So in a good sense
6
‘Rearrangement of Particles’, 195.
Scott Sturgeon / 45
parts are not otherly relative to their sum. They are their sum. Their sum is
them. ’s shapeshift is just the running about of its parts. It is an intrinsic
change in an enduring thing (which can not be). Hence we face a genuine
puzzle. (1), (2), (4), and (5) look good. They cannot all be true. The possibility
of enduring runabouts yields conflict in Basic Metaphysics. That possibility
was got from intelligibility. The conflict grows, at least in part, from the view
that intelligibility yields genuine possibility. It grows from an instance of
(L) Int(ø) ⇒ ø.
That motivates questioning the validity of (L).
Placeshifting: (6) says endurers can run about. Distance between them can
shift. (3) guarantees that is intrinsic change. (1) then ensures they perdure.
No good! Nothing can perdure and endure. The former is done solely by
composition from instantaneous things. The latter is done without such
composition. (1), (3), and (6) conflict. They cannot all be true. This is the
puzzle of placeshifting. The genuine possibility of enduring runabouts yields
conflict in Basic Metaphysics. That possibility is got from intelligibility. Here
too conflict grows, at least in part, from the view that intelligibility yields
genuine possibility. Here too it grows from an instance of (L).7
The schema’s validity is our main topic. It is one of some urgency. For
as we have just seen, free movement from intelligibility to possibility breeds
conflict in Basic Metaphysics. So we must ask: what kind of intelligibility, if
any, marks genuine possibility? Are there readings of ‘Int(—)’ which make
(L) valid?
4. Links
Let us say ø is prima facie apriori coherent iff it is coherent after a bit of apriori
reflection. And let us write pfac(ø) to express that. The naive apriori
infallibilist says this is valid
7
It might be thought the shapeshifting and placeshifting puzzles show enduring runabouts are
not intelligible. After all, the puzzles clearly show enduring runabouts are not free of conceptual
difficulty after serious reflection (to borrow a phrase used earlier). But it is important to remember
this: the difficulty attaches, in the first instance, to the conjunction of (1) thru (6). To blame (6) for
the bother ignores the fact that apriori warrant, like its aposteriori cousin, is holistic. Echoing
Quine: apriori intelligibility faces the rub of contradiction as a corporate body.
(p) pfac(ø) ⇒ ø.8
She says belief in ø based on pfac(ø) is infallible. But she is obviously wrong.
After all, apriori reflection can self-correct. When we see that not all clear
concepts yield extension sets, or that not all infinite sets are equinumerous, for
instance, apriori reflection self-corrects. (p) is invalid: genuine impossibilities
can be prima facie apriori coherent. Inferring ø from pfac(ø) is at best a
fallible affair.
So let us idealize. Let us say ø is limit coherent iff it is coherent at the limit
of apriori reflection. The idea, roughly, is that ø is coherent even when faced
with all that is rational after ideal apriori reflection. Let us write lim(ø) to
express that. The less-naive apriori infallibilist says this is valid
(I) lim(ø) ⇒ ø.
She says belief in ø based on lim(ø) is infallible. It is not obvious she is wrong.
After all, refuting her was a key move in Kripke’s war to segregate apriority
and modality. Yet win the war he did. For Kripke showed limit-case apriori
reflection does not mark genuine possibility. He showed non-modal fact can
rub out limit coherence. Something coherent at the limit of apriori reflection
can be genuinely impossible; and non-modal fact can show it to be so.
For instance: let be the proposition that David Lewis is Bruce LeCatt. Both
and ¬ are limit coherent. No amount of reflection can show whether
Lewis is LeCatt. The matter is not apriori. In fact they are one person. But you
cannot tell by reflection, not even ideal reflection. You have to look. Since they
are one, however, they cannot help but be so. There is no way to pull them
apart. Not only is true, it is genuinely necessary. Hence ¬ is genuinely
impossible. Despite its limit coherence, ¬ fails to be genuinely possible. It is
a counter-instance to (l). That schema too is invalid.
In refuting (l), Kripke’s strategy is clear. He locates non-modal claims
such that
(a) both they and their negation are limit coherent;
and
(b) it is apriori obliged, in the limit, to think the matter they raise is
non-contingent.
8
I shall understand the validity of schemata to consist in their lack of counter-instance which
is logically simple or the negation of such. That keeps discussion where it belongs, on base-case
bother. Everything to follow could be recast, mutatis mutandis, in more general terms.
Scott Sturgeon / 47
So it is with , the proposition that Lewis is LeCatt. Not only is it aposteriori
whether is true, it is apriori the matter is non-contingent. It is apriori that
if is true, then it is necessary; and it is apriori that if ¬ is true, then
it is necessary. However things turn out it could not have been otherwise.
That is the view one should have in the limit of apriori reflection. Since the
first-order matter is not apriori, though, limit-case apriori reflection does not
mark genuine possibility. That was Kripke’s insight. How should a less-naive
apriori infallibilist react?
Good question.
Let us say Kripke claims are non-modal claims like and ¬. And let us
say true Kripke claims are Kripke truths. Such truths are non-modal. They and
their negations are limit coherent. But it is apriori the matter they raise is
non-contingent. Then we define: ø is Kripke coherent iff it is limit coherent in
light of Kripke truths. And we write kc(ø) to express that. We set
kc(ø) = lim(ø/K),
with K the set of Kripke truths. A conservative reaction to Kripke admits (l)
is invalid but insists Kripke coherence marks genuine possibility. It replaces
(l) with
(k) kc(ø) ⇒ ø.
This is mild aposteriori infallibilism. It says belief in ø based on kc(ø) is
infallible. The view is at work, I believe, in a good deal of post-Naming-and-
Necessity philosophy. It retains the naive and less-naive apriorist’s infallibilism.
It gently resists their apriorism. The view makes use of aposteriori matters.
But it does so only by appeal to Kripke truths. What should we make of it? In
particular, should we believe Kripke coherence marks genuine possibility?
No. There are Kripke coherent claims that are not genuinely possible. There
are counter-instances to (k). For example, consider the view that Lewisian
worlds truthmake claims of genuine modality. The idea, basically, is that
points in the space of genuine possibility are Lewisian worlds. Such worlds
are maximal sums of spatiotemporally related parts. Lewis equates these sums
with maximal genuine possibilities. So his idea is twofold: it postulates maximal
sums of a certain sort; and it identifies them with maximal genuine possibilities.
Call the view ‘L’. Both L and ¬L are prima facie coherent. And they look
limit coherent too. So a question is presently sharp: are there non-modal
truths to rub out their apriori coherence? Are there truths expressible in
non-modal idioms that render L or ¬L incoherent?
I doubt it. Both L and ¬L look Kripke coherent:
(7) kc(L)
(8) kc (¬L).
In English: it is Kripke coherent that genuine possibilities are Lewisian; and it
is Kripke coherent they are not. Take apriori reflection to the limit, throw in
Kripke truth, both L and ¬L stay coherent. 9
This spells trouble for (k). After all, (7), (8), and (k) jointly yield
(9) L
and
(10) ¬L.
But consider the view that Lewisian modal metaphysics is true but genuinely
might not have been:
(11) L & ¬L.
This is genuinely impossible. Its first conjunct entails its second is true only if
there is a Lewisian truthmaker at which there is no such truthmaker. (11)’s
conjuncts cannot both be true. So we have
(12) (L ⊃ ¬¬L).
Yet this and (9) yield
(13) L,
which conflicts with (10). So (9) rules out (10). Similarly, consider the view
that Lewisian modal metaphysics is false but genuinely might have been true:
(14) ¬ L & L.
This too is genuinely impossible. For its possibility implies that of its right-hand
side. Yet that leads to (9). As we have just seen, though, (9) leads via (12) to (13);
and (13) conflicts with (14)’s left-hand side.10 The moral is clear: whichever
9
We know at least one Lewis world exists: the maximal spatiotemporal sum of which we are
parts. Suppose no other Lewis world exists. Does this show L is false? No. To get that result one
must also suppose that things could genuinely be other than they are. Yet that is a modal claim
par excellence. By stipulation neither it nor its negation is a Kripke truth.
10
This line of thought leans on S5 for genuine modality. That is contentious, of course. See
Graham Forbes, The Metaphysics of Modality (Oxford: OUP, 1985); Nathan Salmon, ‘The Logic of What
Might Have Been’, Philosophical Review, 98 (1989), 3–34 and references therein. But notice: L is an
identity claim. The conclusion of the argument can be got by appeal to the necessity of identity
Scott Sturgeon / 49
of L and ¬L turns out true, it is genuinely necessary as well. (9) and (10)
cannot both be true. At least one Kripke coherence is genuinely impossible.
(k) is invalid. Not surprisingly, the schema breaks down on the metaphysics of
genuine modality. There are Kripke coherent claims about that metaphysics
which genuinely cannot be true. Mild aposteriori infallibilism breeds conflict
in modal metaphysics.
It might be replied that L or ¬L is a Kripke truth. After all, the idea behind L
is that claims like
(l) (Donkeys talk) iff there is a Lewisian world at which donkeys talk
reduce their modal left-hand to their non-modal right-hand sides. And if that
is right, of course, those claims and their right-hand sides are in K. They
are both Kripke truths. This would be good and bad news for (k). It would
reveal putative counter-instances to be no such thing. But it would render the
schema trivial. For if (l) and its ilk count as non-modal, that notion covers not
only claims about actuality expressed in non-modal terms, but those about
non-actuality so expressed. (k) is trivially valid. Genuine possibility trivial
follows from non-modal facts so understood. On this view, Kripke coherence
does imply genuine possibility. But that is not because bona fide intelligibility has
been shown to mark genuine modality. Nor is it because a mild non-modal
weakening of that notion has been shown to do so. It is because ‘intelligibility’
has been stretched by fiat to cover genuine modality.
5. Diagnosis
Consider and L, the propositions that Lewis is LeCatt and that genuine
possibilities are Lewisian worlds. They bear the marks of Kripke truth:
(a) both they and their negations are limit coherent;
and
(b) it is apriori obliged, in the limit, to think the matters they raise are
non-contingent.11
and distinctness. That appeal entails nothing about the logic of genuine modality. I present the
S5-argument because I think it best edifies L’s modal status. I accept its underlying logic. But that
logic is inessential to the point at hand.
11
Note satisfies (a) because it is aposteriori. It is to be gleaned from experience. L is not like
that. It satisfies (a) because apriori concerns underdetermine its truth-value. That is why Lewis is right
When a claim satisfies these principles, it is an apriori red flag. It cannot be ruled
out apriori. It cannot be ruled in apriori. The claim’s subject matter can be seen
apriori to be non-contingent. And the same holds of its negation. Hence we
can see, on purely apriori grounds, that apriori reflection must be insensitive
to impossibility somewhere. Either a given red flag or its negation is the rub.
Either the first cannot be true despite remaining in play at the limit of apriori
reflection, or its negation cannot be true despite remaining in play at that
limit. Impossibility slips through the net of apriori reflection.12
This prompts a strategy. Let us say ø is apriori open—or open for short—iff
two things are true:
(i) both ø and ¬ø are limit coherent;
but
(ii) it is not apriori obliged the matter they raise is non-contingent.
And let us write o(ø) to express that. When ø is open, both ø and ¬ø are
coherent at the limit of apriori reflection; but it is not settled there that ø is
non-contingent. The thought is to replace (k) with
(o) o(ø) ⇒ ø.
This is seasoned apriori infallibilism. It says openness marks genuine possibility.
If a claim and its negation are limit coherent, and it is not apriori they are
non-contingent, then they are genuinely possible.
What should we make of the idea?
Well: neither Kripke-style claims like , nor Lewis-style claims like L,
are counter-instances to (o). For none are open. None satisfy the schema’s
antecedent. It is apriori the matter they raise is non-contingent. That is why
they can play spoiler in apriori discussion of (l) and (k). But this fact about
them precludes their cutting against (o).
Further, we can see apriori that there can be no base-case apriori counter-
instance to (o). For such a claim would be a logically simple (or negated
to insist L-commitment should rest on L’s role in systematic philosophy. See On the Plurality of
Worlds, 3–5.
12
This means there will be further systematic slip-ups by apriority. For instance, let R be any
red flag. Neither R nor its negation can be ruled out apriori. But it is apriori that R is necessary or
impossible. So consider any contingent claim C which is apriori independent of R and its negation.
If R is necessary, then (C&¬R) is limit coherent yet impossible. If R is impossible, then (C&R)
is limit coherent yet impossible. False red flags logically embed to generate systematic apriori
blindspots. I ignore this in what follows, focusing on the root source of bother.
Scott Sturgeon / 51
logically simple) open apriori impossibility. Yet no such claim can be open and
apriori impossible. The latter precludes the former. If such a claim is apriori
impossible, it is apriori the matter it raises is non-contingent. But if it is apriori
the matter it raises is non-contingent, it is not open. Hence no base-case claim
can be open and apriori impossible. None can be an apriori counter-instance
to (o).
This looks good. (o) side-steps problems which infect (p), (l), and (k). And
it does so in an explanatory way. Base-case apriori counter-instances to the
latter two schemata have two things in common: they satisfy (a) and (b). They
and their negation are limit coherent; and it is apriori the matter they raise is
non-contingent. Openness rules out the last feature. For this reason, no base
claim can apriori satisfy (o)’s antecedent yet fail to satisfy it is consequent.
Having said that, there is an obvious shortcoming with (o). The principle
is too restricted. The definition of openness guarantees it is closed under
negation. ø is open iff ¬ø is open. (o) is valid, therefore, only if openness marks
contingency. (o)’s validity guarantees that of
(o)* o(ø) ⇒ ø is contingent.
Yet there are non-contingent possibilities marked by apriori reflection.
The fact that everything is itself, for instance. Surely we mark this pos-
sibility with apriori reflection. (o) misses it altogether. The schema is
incomplete. It is suited to contingent possibilities (if any). It skips non-
contingent ones.
To plug the gap we need a condition designed for apriori necessities. So let
us say ø is apriori forced—or forced for short—iff three things are true:
(iii) ø is limit coherent,
(iv) ¬ø is not limit coherent,
but
(v) it is apriori obliged that the matter they raise is non-contingent.
And let us write f(ø) to express that. Then we can say ø is apriori apt—or apt
for short—iff ø is open or forced. And we can write apt(ø) to express that.
Then we swap (o) for
(a) apt(ø) ⇒ ø.
This is best-shot apriori infallibilism. It amounts to endorsing
(o) o(ø) ⇒ ø
and
(f) f(ø) ⇒ ø.
Best-shot apriori infallibilism says aptness marks genuine possibility. It builds
its mark out of two conditions. One is designed for apriori necessity. The other
is designed for such contingency. What should we make of it?
Well, we have seen no base claim can be apriori open and impossible. There
is such an apriori counter-instance to (a), therefore, only if a base claim can
be apriori forced and impossible. Yet these too are incompatible. If you show
something apriori to be impossible, it is limit coherent it is false. That means
it is not forced. It fails condition (iv). Just as there can be no basic apriori
counter-instance to (o), there can be none to (a). It too side-steps direct
refutation.
In my view, (a) grounds the best modal infallibilism. Just as (l) is motivated
by weakness in (p), and (k) is motivated by weakness in (l), (a) is motivated
by cracks in all three. It is designed to avoid their pitfalls; and it does so
demonstrably. If there is a valid base-case link from apriority to genuine
possibility, (a) schematizes it.
6. Problems
I reject (a). I have no direct refutation, of course. The schema does not permit
one. But I do have two indirect worries. The first generalizes a point made
with force by Gideon Rosen in ‘The Limits of Contingency’ (Chapter 1 in this
volume). The second springs from my take on the fit between epistemology
and metaphysics.
[A] Many claims of Basic Metaphysics are open. Apriori reflection does not
determine a truth-value for them, nor does it say they are non-contingent.
Stripped of its modal operator, for instance, (1) is like that:
(1)− Intrinsic change comes via perdurance.
This claim is Basic. When true it is not made so by anything else. Yet (1)− is
open. For recall: intrinsic change is strictly impossible. Whenever it seems to
occur, sleight of hand is in play. (1)− says it comes by perdurance. That claim
is limit coherent. But so is the view that the trick comes by enduring things
shifting properties which look to be, but are not, intrinsic. Were the adicity of
Scott Sturgeon / 53
them properly subtle—for instance, were they to involve a difficult-to-spot
relation to time—intrinsic change would look to happen; but it would come
by endurance rather than perdurance. It would come by property- rather than
thing-level sleight of hand. This too is limit coherent. Yet limit-case reflection
does not secure the matter is non-contingent. It leaves it unsettled. Hence
(1)− is open. It is an open Basic claim.
As Rosen notes, though, Joe Metaphysician assumes Basic truths are non-
contingent. He embraces the necessity of (1)− after deciding it is the best
explanation of intrinsic change. When looking to explain such change, he
seeks the best amongst coherent options. He concludes the best is necessary.
By letting b(ø) mean ø is Basic, we can schematize his practice with
(b) [b(ø) & ø] ⇒ ø.
Joe assumes Basic truths are necessary. He accepts
(b)* b(ø) ⇒ ø is non-contingent.
And that makes for trouble. After all: (a) guarantees openness marks contin-
gency, and (b) guarantees Basicness marks non-contingency. No claim can
be contingent and non-contingent. Hence none can be Basic and open. We
have just seen, though, that Basic claims can be open. So there is potential
conflict in practice. Either (a) or (b) is invalid. Either aptness does not ensure
possibility, or Basic truths need not be necessary. Joe must inflate his space of
possibilities or reject openness as one of its marks. The first option pays for (a)
with (b). The second does the reverse. One of them must go. Joe cannot have it
both ways. If aptness marks genuine possibility, open Basics of Metaphysics are
contingent. If such Basics are not contingent, aptness does not mark genuine
possibility.
Rosen’s point spreads, of course, to any area of inquiry driven by (b)*. It
covers all traditional areas of philosophy (ethics, logic, mind, and so forth).
They all harbour debate about Basics. They all presume Basic truths are
necessary. Yet the timeless nature of debate within them strongly suggests
that incompatible claims about Basics are apt. In turn that means one of two
things: either such Basics are contingent, or aptness does not mark necessity.
[B]. Humans tend to fuse epistemic and metaphysical matters: credence with
chance, certainty with truth, certainty with causal determination. I take it we
see this in our students and ourselves. Call it the ep-&-met tendency. It explains,
I think, why it is so natural to say apriori reflection infallibly depicts genuine
possibility. But here as elsewhere nature should be resisted. For genuine
modality is mind-and language-independent. It is a non-epistemic domain.
This should ensure apriori reflection is at best a fallible guide to genuine
possibility. That is how good epistemology and metaphysics fit together. When
doing the former on mind- and language-independent fact, the result is
humble pie. It is epistemology cleansed of error-free capacities to interrogate
reality. It is magic-free epistemology. It is fallibilism.
Most philosophers agree. When it comes to modality, though, they do not
apply the lesson. They write as if they accept an under-revised version of the
infallibilist model induced by the ep-&-met tendency. They write as if they
replace (l) with something like
(k) Apriori reflection infallibly depicts genuine possibility unless corrected
by Kripke truth.
But (k) looks like magic. After all, the space of genuine possibility is fully
realistic. Neither its contours nor contents spring from apriori practice. Nor
do they spring from such practice tutored by Kripke truth. By my lights, then,
(k) is no better than
Credence about the future infallibly depicts its chance unless corrected by
present categorical fact;
or
Certainty about the past infallibly depicts it unless corrected by present
fact.
These views only tempt when blinded by the ep-&-met tendency. Once it is
thrown off they do not. So it should be with (k). The schema trades in magic
epistemology. To think otherwise is to fall prey to the ep-&-met tendency. Or
so it seems to me.
We have seen directly, moreover, that (k) is invalid. We have found Kripke
coherent impossibilities. Reflection on them led to
(a) Apriori reflection infallibly depicts genuine possibility unless when
admitting a claim and its negation it also admits their non-contingency.
But that looks even more magical. After all, the idea is that apriori reflection
infallibly marks genuine possibility unless it self-corrects. Yet the space of such
possibility is fully realistic. It would be little short of miraculous, then, if our
Scott Sturgeon / 55
apriori practice had in-built corrections to hand whenever needed. Would
that it were so! By my lights, (a) is no better than
Non-degenerate credence about future outcome infallibly depicts its
chance unless it admits the chance of such outcome is degenerate;
or
Perceptual belief about one’s surroundings is infallible unless one knows one
suffers perceptual delusion.
These are crazy views. They only tempt when blinded by the ep-&-met
tendency. Once it is thrown off they do not. And so it should be with (a). The
schema trades in magic epistemology. To think otherwise is to fall prey to the
ep-&-met tendency. Or so it seems to me.
7. Coda
Our conclusion is simple. No reading of
(L) Int(ø) ⇒ ø
is valid. But we should ask: what difference does that make?
Well, some discussion of physicalism requires (L) to have a valid reading. It
maintains one can work out apriori whether physicalism is true simply on the
basis of facts aside from whether physicalism is true. It says one need only work
out whether fundamental physicals imply the rest. If they do, physicalism
is true. If they do not, it is false. If (L) has no valid reading, however, the
question is moot. Physicalism does not require everything to follow apriori
from fundamental physicals.13
Set aside such discussion of physicalism. Ask yourself this: does anything
else turn on (L)’s validity? I cannot see that it does. After all, intelligibility is
obviously some kind of guide to genuine possibility. The question is what kind.
(L) says it is an infallible one. But the truth is more modest. Intelligibility is a
fallible guide to genuine possibility.
13
See David Chalmers’s The Conscious Mind (Oxford: OUP, 1996) and Frank Jackson’s From Meta-
physics to Ethics (Oxford: OUP, 1998). For discussion see my Matters of Mind (London: Routledge, 2001).
Let Int(ø) ø mean ø’s intelligibility is defeasible reason for its genuine
possibility. We have several good readings of
(L)* Int(ø) ø.
Indeed, we have six ready to hand:
(p)* pfac(ø) ø,
(l)* lim(ø) ø,
(k)* kc(ø) ø,
(o)* o(ø) ø,
(f)* f(ø) ø,
(a)* apt(ø) ø.
Prima facie, limit and Kripke coherence are reason to think a claim is genuinely
possible. So are openness, forcedness, and aptness. In each case an apriori (or
quasi-apriori)14 condition points to genuine possibility. And in each case it does
so defeasibly. The range of defeaters will vary from case to case. But the moral
will not: intelligibility defeasibly marks genuine possibility.
It is a good question why that is so. It is not a good question whether
it is. There is no question but that intelligibility defeasibly marks genuine
possibility. That fact is a cornerstone of our modal practice. Without it no
practice remains.15
14
Kripke coherence is idealized apriority tutored by Kripke truth. Such truth is aposteriori.
Kripke coherence is thus quasi-apriori.
15
For discussion of the link between sensuous imagination and genuine modality, see chapter 5
of Matters of Mind and references therein. And for a perspective at odds with the final section of this
effort, see my Epistemic Norms.
3
The Modal Fictionalist Predicament
John Divers and Jason Hagen
1. Introduction
A central aim of genuine modal realism (GMR) is to provide a translation
schema:
(R) A iff A*
in which the instances of ‘A’ are modal claims and the instances of ‘A*’ are
their possible-world (counterpart-theoretic) translations. Correspondingly,
the modal fictionalist (locus classicus, Rosen 1990) proposes that we translate
modal claims via the weaker schema:
(F) A iff According to GMR, A*.
Thus, for instance, with regard to:
(1) It is possible that there are talking donkeys
the realist holds:
(R1) It is possible that there are talking donkeys iff
there is a world at which there are talking donkeys
whilst the fictionalist holds:
Thanks to the following for various helpful comments and suggestions: Rod Bertolet, Jan
Cover, Andrew Evenson, Fraser MacBride, Joseph Melia, Daniel Nolan, Scott Shalkowski, Richard
Woodward and an anonymous referee.
58 / The Modal Fictionalist Predicament
(F1) It is possible that there are talking donkeys iff
According to GMR, there is a world at which there are talking
donkeys.
One who affirms (F1) and its left-side is committed to affirming its right-side
but, it is natural to think, one is not thereby committed to the existence
of any world other than our own. Further, the fictionalist will interpret
unqualified possible world claims—e.g. ‘There is a world at which there are
talking donkeys’ or ‘There is a world at which there are blue swans’—as being
elliptical for the qualified thoughts expressed by right-side instances of (F).
Broadly construed, then, the fictionalist proposal is that we can use realistic
sounding possible-world talk, retain at least some of the philosophical benefits
of doing so, and yet avoid commitment to the existence of a plurality of worlds.
The modal fictionalist proposal has attracted a great deal of discussion in
the decade since its inception. In this paper we will argue that the upshot is
that modal fictionalism is ensnared in a serious predicament.
In Section 2, we summarize the two potentially lethal objections to fic-
tionalism that have emerged and the only result which suggests that modal
fictionalism is capable of delivering any of the specific theoretical benefits
associated with realistic quantification over possible worlds. The fictionalist
predicament, we argue, is that she cannot avoid both objections while main-
taining her claim on the result. We proceed by considering two different
strategies that the fictionalist may adopt in order to translate modal claims
about the plurality of worlds. In relation to the first strategy we argue (in
Section 3) that the Brock–Rosen objection is avoided but the Hale objection
is avoided only by resorting to measures that are desperate (Hale 1995b) but
which also deprive the fictionalist of her result. In Section 4 we turn to
the second strategy and consider two relevant sub-strategies. On the direct
sub-strategy, we show that the Brock–Rosen objection is avoided if and only
if the Hale objection is not avoided. On the indirect sub-strategy, we show that
the Brock–Rosen objection is not avoided and that absurdity follows from
the fictionalist schema.
2. Two Objections and One Benefit

We start with an objection developed independently by both Stuart Brock
(1993) and Gideon Rosen (1993). Brock and Rosen argue that the fictionalist is
John Divers and Jason Hagen / 59
covertly committed to the proposition:
(P) There is a plurality of worlds.
The basic form of their argument may be represented as follows:
1 (B1) According to GMR, at every world, P Hyp
2 (B2) Necessarily P iff According to GMR, at Inst of (F)
every world, P
1,2 (B3) Necessarily P 1,2 iff elim
1,2 (B4) P 3T
Clearly, if this ontological commitment can be pinned on the fictionalist,
then fictionalism fails in its primary aim. For regardless of whatever benefits
fictionalism may deliver, they come at the cost of a commitment to possible
worlds.1
The second objection is a dilemma developed by Bob Hale (1995a). Hale
argues that the fictionalist who holds P false is in trouble if she maintains that
the falsehood of P is contingent or if she maintains that it is necessary. If the
fictionalist holds that P is contingently false, the argument proceeds, then the
instance of (F) that deals with the modal claim ‘Possibly P’—i.e. it is possible
that there is a plurality of worlds—cannot adequately capture the content
of that claim. We believe that the contingency horn of Hale’s dilemma can
be blunted (see Divers 1999a). So all that the fictionalist must do in order to
resist Hale’s dilemma is establish her right to assert that P is contingently
false. That can be done by defending the view that P, though false, is possible.
The fictionalist’s right to assert that P is possible is established, we presume,
if she can plausibly deny that P is impossible. Since we confront the latter
issue at the first premise of the impossibility horn of Hale’s dilemma, we will
only develop that horn of the dilemma. The objection associated with the
impossibility horn is that the impossibility of P commits the fictionalist to the
truth of any arbitrary modal claim ‘X’. The argument to this effect runs as
follows:
1
An extended version of the argument, as developed by Rosen (1993) in particular, produces
(B1) as a consequence of prior premises. Starting with the premise that a certain modal claim is
true—e.g. the claim that it is necessarily contingent that there are kangaroos—it deduces (B1)
from the truth of the fictionalist translation of that claim. In this way, the right to reject (B1) will
depend on the truth of this starting premise. We will qualify our main points as required to take
this more extensive variant of the argument into account, but the variant need not, and will not,
figure prominently in our argument.
1 (H1) Not (Possibly P) Hyp

2 (H2) GMR strictly implies P Hyp
1,2 (H3) Not (Possibly GMR) 1,2 strict
implication
1,2 (H4) GMR strictly implies (X*) 3 Paradox strict
implication
5 (H5) If GMR strictly implies Hyp
(X*) then (According to
GMR, X*)
1,2,5 (H6) According to GMR, X* 4,5 modus
ponens
7 (H7) X iff (According to GMR, X*) Inst of (F)
1,2,5,7 (H8) X 6,7 iff elim
If the above objection stands, then modal fictionalism is not so much as
consistent since it delivers the conclusion that, for any modal claim X, both X
and its negation (also a modal claim) are true.2
The benefit claimed on behalf of the fictionalist by Divers (1999b) is that
she has the resources to prove a certain safety result, thereby providing
justification for the practice of doing modal logic by proxy in first-order logic.
The practice of doing modal logic by proxy is the familiar one in which we set
about evaluating a modal inference as follows: start with the modal premises;
‘translate’ modal premises into first-order claims about worlds; use first-order
logic to draw a worldly conclusion from the worldly premises; move back
from the worldly conclusion to its modal translation; and finally, conclude
that the modal conclusion follows from the modal premises. The safety result
that has to be established is that, in the following sense, this practice of doing
modal logic by proxy never leads us astray:
2
The Hale (1995a and b) objection and Rosen (1995) reply do focus on a counterfactual
treatment of the fictionalist prefix. But the counterfactual conditional functions in that context
as a representative of non-material conditionals in general Thus consider Hale’s comment on his
own dilemma : ‘We know two things about [the fictionalist prefix]—(1) it has to be non-factive
and (2) if my dilemma is to be avoided it cannot be phrased as a strong (strict or counterfactual)
conditional or entailment’ (Hale 1995b: 75–6, our emphasis). What is important for our purposes
is: (a) that the fictionalist result discussed below depends on treating the prefix in terms of strict
implication and (b) that the Hale objection applies when the prefix is so understood (which is not
in doubt).
(SR) Necessarily, if worldly claim B* is a logical consequence of worldly
claim A*, then modal claim B is a logical consequence of modal
claim A.3
The fictionalist can demonstrate this result by invoking the fictionalist schema
(F), if she proceeds as follows:
(Step 1) Incorporate a modal characterization of logical consequence to give a
modal version of the safety result as follows:
(MSR) (A* strictly implies B*) strictly implies (A strictly implies B).4
(Step 2) Assume that the fictional operator, ‘According to GMR’ is closed
under strict implication, thus:
(CSI) ( (According to GMR, X) and (X strictly implies Y) )
strictly implies (According to GMR, Y).
(Step 3) Derive (in classical S4 modal propositional logic) the modal version of
the safety result (MSR) from the necessitated fictionalist schema (NF):
(NF) Necessarily, (A iff According to GMR, A*).
3
The very natural thought that the fictionalist cannot deliver this result by means of a
semantic theory that involves quantification over possible worlds is defended in Divers (1999b).
4
An anonymous referee makes a good point in suggesting that perhaps what the fictionalist
should be aiming to show in this regard is not (MSR) but a result about logical consequence
characterized other than in modal terms. We make two points in reply. First, the move from
(SR) to (MSR) is made on behalf of the fictionalist in order to secure the safety result (SR). If the
move is wrong or unjustified, the effect is to enhance our present anti-fictionalist dialectic. If a
proof of (MSR) ought not to be construed as a justification of (SR), then so much the better for
our present case and so much the worse for the fictionalist. For the position would then be that
the fictionalist has no proper claim on the safety result (SR) and—a fortiori—no ground for
claiming that she can maintain the safety result while meeting the objections.
Secondly, perhaps the kind of result that the fictionalist ought to seek to establish is:
(SR*) Necessarily ( (If B* is derivable from A*) then Necessarily (If A then B) ).
Fortunately, the revised result is immediately obtained from the original strict implication
result (MSR):
(MSR) Necessarily (If Necessarily (If A* then B*) then Necessarily (If A then B) )
given the necessity of the (first-order) derivability relation:
(ND) Necessarily ( (If B* is derivable from A*) then Necessarily (If A* then B*) ).
All we need is the transitivity of strict implication. Moreover the same point holds for the related
result concerning first-order semantic consequence in place of first-order derivability. So there is
a sense in which (MSR) may be a stronger safety result than (SR).
(Step 4) Show that the modal version of the safety result follows from the
weaker schema (F) by establishing (again in classical S4 modal propositional
logic) the lemma:
(Lem) (F) strictly implies (NF).
Here we should note two important points. First, the capacity to underwrite the
practice of doing modal logic by proxy is the only substantial benefit for which
the fictionalist has a supporting argument. It has been argued that fictionalism
cannot deliver a possible worlds semantics for modal languages (Divers 1995)
and no one has argued that fictionalism gives—say—an account of the truth-
makers for modal claims or that it articulates an analysis of modal concepts in
nonmodal terms. This is what we mean when we claim that the result is the
only result which suggests that modal fictionalism is capable of delivering any
of the specific theoretical benefits associated with realistic quantification over
possible worlds.5 Second, it may be possible to establish the safety result by some
other means, but in the absence of any alternative demonstration, the strategy
outlined above presently offers the fictionalist her one and only guarantee of
the result. Moreover, if our argument is correct, then, in abandoning the prin-
ciples that she must abandon in order to meet the objections, the fictionalist
must weaken significantly the logic that she has at her disposal for showing
what follows from the fictionalist schema. In light of these considerations, we
think it crucial to the credibility of fictionalism that the fictionalist be able to
show that she is entitled to implement the strategy outlined.
Overall, then, our challenge to the fictionalist is to develop her original
proposal in a way that proves capable of implementing the above strategy
for securing the safety result and of avoiding both the Brock–Rosen and
5
Thus, in our terms it is not a ‘benefit’ of modal fictionalism, as may be alleged, that it offers
an ontologically economical or naturalistic account of modality. Our concern is with the content
of the intimated ‘account’. Which kinds of substantial explanations—semantic, metaphysical,
conceptual—if any, can fictionalism deliver? If modal fictionalism can deliver any benefits then,
perhaps it may have the advantage over rival theories that deliver similar benefits (genuine
realism, actualist realism) that it does so naturalistically and economically. But that is a secondary
matter, and fictionalism must deliver the explanatory goods (‘benefits’) in order to earn attention
for its ‘theoretical virtues’. It ought to be noted in this regard that Rosen (1990) does suggest
that fictionalism, unlike realism, delivers the epistemological benefit of accounting for the role of
imagination in the formation of our modal beliefs. While we cannot engage fully here with this
claim on behalf of fictionalism, we register the objection that the claim will be justified only if the
fictionalist is prepared to endorse the view that the truthmaking facts for modal claims are facts
about a story (GMR).
Hale objections. We will argue that neither of the two main developments of
fictionalism can meet this challenge.
3. The First Development

Following the observations of Harold Noonan (1994), Rosen (1995) responds
to the Brock–Rosen objection by specifying a version of the realist theory
GMR that incorporates the Lewis (1968) principles for translating the formulas
of quantified modal logic into formulas of counterpart theory. On those
principles, modal claims are translated into claims about the parts of worlds:
what is necessarily the case is a matter of what is true of the parts of all worlds,
what is possibly the case is a matter of what is true of the parts of some worlds,
and what is non-modally the case is a matter of what is true of the parts of
the world of which we are parts. So, for instance, the initial hypothesis of the
Brock–Rosen argument:
(B1) According to GMR, at every world, P
is false. For within the resources of Lewis (1968), the claim that
(2) At every world, P
can only mean
(R2) ∀x(Wx → ∃y∃z(Wy & Wz & ∼(y=z) & Iyx & Izx) ).6
But (R2)—the claim that every world has a plurality of worlds as parts
—does not hold according to GMR. It is inconsistent with postulate (P2)
of Lewis (1968: 27). However, by accepting this solution to the Brock-
Rosen objection, Rosen knows that the fictionalist commits herself to
grasping the ‘impossibility’ horn of Hale’s dilemma. On the Lewis (1968)
approach,
(3) It is possible that P
is translated as
(R3) ∃x(Wx & ∃y∃z(Wy & Wz & ∼(y=z) & Iyx & Izx) ).
6
Here we use ‘Ixy’ to stand for the part-of relation, while making it explicit when the relata
are worlds. This diverges from how Lewis (1968) originally interpreted the symbol as a primitive
in-a-world relation, but still consistent with the relation that he says he had ‘foremost in mind’
in that 1968 article (1983: 39).
(R3)—the claim that some world has a plurality of worlds as parts—does
not hold according to GMR. Indeed, to underscore the point, the Lewis (1968)
schema translates its negation, ‘Not Possibly P’, as a claim which is true
according to GMR—i.e. that no world has a plurality of worlds as parts. Thus,
given (F), the fictionalist is committed to denying (3) and to affirming the
initial hypothesis of the Hale argument:
(H1) Not (Possibly P).
Rosen (1995) then pinpoints three aspects of Hale’s argument that might be
resisted in order to ‘fix’ fictionalism. First, one might reject the hypothesis,
(H5), which partly articulates the thesis that the fictive prefix can be adequately
‘glossed’ by some strict conditional.7 Second, there is the move from (H3)
to (H4) which depends on the standard ‘paradoxical’ treatment of strict
implication on which any proposition whatever is strictly implied by any
impossibility. Third, the fictionalist might seek to avoid commitment to (H1)
by adducing (non ad hoc) grounds on which to claim that (H1) lacks a truth-
value. Hale (1995b) characterizes all of these attempts to protect the fictionalist
proposal as ‘desperate fixes’, and there is certainly scope for regarding each
as unsatisfactory in its own right. But here, we will only point out the
implications that each move has for the preservation of the safety result.
Concerning the first fix, it is provable that (H5) is a consequence of the
closure of the fictional operator under strict implication (CSI) given the
further premise that According to GMR, GMR:
1 (1) ( (According to GMR, GMR) & Hyp (Inst CSI)
GMR strictly implies X*) )
strictly implies (According to
GMR, X*)
2 (2) According to GMR, GMR Hyp
3 (3) GMR strictly implies X* Hyp
2,3 (4) (According to GMR, GMR) & 2,3 & Intro
GMR strictly implies X*
1,2,3 (5) According to GMR, X* 1,4 strict imp
1,2 (6) If GMR strictly implies X*, then 3,5 CP
According to GMR, X*
7
It is worth noting that (H5) is far weaker than any claim of meaning equivalence between the
fictional prefix construct, ‘According to GMR, A’ and the strict conditional ‘GMR strictly implies
A’. (H5) is only a material conditional, and that is all the Hale argument requires.
We assume that ‘According to’ must be understood as reflexive so that there
is no question of rejecting premise (2). Thus, rejection of the conclusion (H5)
compels rejection of the closure principle (CSI). Since the assumption of (CSI)
constitutes Step 2 in the argument for the safety result, rejection of (H5) leaves
the fictionalist with no legitimate claim on the result.8
Concerning the second fix, we note that rejection of the principle
(H4)—that anything is strictly implied by an impossibility—marks a retreat
to a modal propositional logic that is much weaker than S4. But the proof of
the safety result relies on the soundness of S4 at Step 3 and Step 4.
The thought behind the third fix is that one can avoid commitment to (H1)
by claiming that the sentence ‘Not (Possibly P)’ lacks a truth-value. More fully,
if we consider a version of GMR, say GMR*, in which the definition of ‘world’
proceeds via the unanalysable, fictive, and theoretical term ‘worldmate’, then
we treat ‘world’, in effect, as a fictional term—one like ‘hobbit’, ‘gimbles’,
etc.—that earns its meaning solely by virtue of the role that it plays in
the fiction. We might then choose to treat such terms in a broadly Fregean
manner so that some of the declarative sentences in which they occur, and
notably those that do not fall within the syntactic scope of an appropriate
fictional operator, are deemed to lack a truth-value. So the sentence that
expresses P—viz. ‘There is a plurality of worlds’—would be taken to lack
a truth-value and so, consequently, would modal complexes such as ‘Not
Possibly P’ and ‘Possibly P’ in which it occurs. There are many points at which
this sort of rationale might be questioned, but we will only consider the
impact that accepting such truth-value gaps has on the argument for the
safety result.
One such gap would appear, for example, at Step 1 of the argument with
the statement of the modal version of the safety result:
(MSR) (A* strictly implies B*) strictly implies (A strictly implies B)
(MSR) will lack a truth-value since its antecedent lacks one; the instances of
‘A*’ and ‘B*’ will be unprefixed formulas of counterpart-theory which feature
talk of ‘worlds’. So while lack of truth-value for (H1) would be enough to
8
Reflecting on the point made in n. 4, above, perhaps the fictionalist should not endorse the
thesis that her prefix is closed under strict implication, but only that it is closed under logical
consequence otherwise understood, e.g. as derivability or as semantic consequence. That being
so, perhaps the present objection would fall. But the result would be placed in jeopardy since we
have no guarantee that it follows from the alternative, weaker, closure principles.
render the Hale argument unsound, the lack of truth-value of (MSR) would
also be enough to render unsound the proof of the fictionalist result.
We conclude that none of Rosen’s fixes will sustain the only available
argument for the only beneficial result that has been claimed on behalf
of fictionalism. Moreover, any of these fixes will significantly restrict the
logic of fictionalism, so that the capacity to deduce theoretically beneficial
consequences from the fictionalist translation scheme is, in general, dimin-
ished. But before leaving the version of fictionalism that incorporates the Lewis
(1968) translation principles, it is important to emphasize a further and broader
difficulty that fictionalism inherits from them.
The Lewis (1968) translation principles fail to provide truth-preserving
translations for an important range of modal claims. The claims in question,
which might be termed ‘advanced’ modal claims (Divers 1999a), involve
modalizing about entities, or collections of entities, that are not (all) parts of
any single spatiotemporally unified whole. Such claims include modal claims
about collections of Lewisian worlds, understood as spatiotemporally disjoined
individuals. However, amongst such claims we also find, arguably—and
certainly according to Lewis’s genuine modal realism—modal claims about
properties, numbers, propositions, sets, and states of affairs. The Lewis (1968)
translation principles for quantified modal logic are intended to deal only with
modal formulas that express modal claims about ordinary, spatiotemporally
unified individuals. Accordingly, advanced modal claims such as
(4) It is possible that there are natural properties
(5) For any proposition, it is contingent whether it is thought
(6) 4 is necessarily even
are not intended by Lewis to be within the expressive range of the formulas
of (first-order) quantified modal logic; counterpart-theoretic translations are
not intended to apply to these cases. This is, at least, the obvious and charitable
interpretation. The application of Lewis (1968) principles to such claims simply
distorts truth-values as when, for example, (4) comes to express:
(R4) ∃x(Wx & ∃y(Iyx & Ny) ).
According to GMR, it is not the case that some world has a natural prop-
erty—for Lewis, a transworld set—as a part.9 So the fictionalist would be
9
For more on why the 1968 principles fail to properly account for advanced modal claims, and
on why other variations on the 1968 approach are inadequate, see Divers (1999a).
committed to the falsehood of (4) and to other such errors. Incorporating
the Lewis (1968) translation principles into the reading of (F), we get:
(F4) It is possible that there are natural properties iff
According to GMR, ∃x(Wx & ∃y(Iyx & Ny) )
whereby the falsehood of the right-side—according to GMR, some world has
a natural property as a part—visits falsehood on the left.
How significant is all of this? Well, if the best version of genuine modal
realism on offer is one that must restrict its translations to ordinary modal
claims, that is certainly a significant boundary on the realist programme. But
the fictionalist, who aims only to keep up with the realist, is no worse off than
that. Of course, if a version of genuine realism is available which is capable of
dealing just as adequately with the translation of advanced modal claims as
it does with the translation of ordinary modal claims, then the fictionalist is
obliged to keep up with that version of realism.
This last observation paves the way for a second development of the original
fictionalist proposal that the fictionalist might explore in the effort to avoid
the objections and secure the result.
4. The Second Development

Divers (1999b) argued that the fictionalist can solve all of her problems by
incorporating the genuine modal realist approach to advanced modal claims
presented in (Divers 1999a). This approach unfolds as follows.10
The genuine realistregardsour non-modal claimsaboutordinary individuals
such as donkeys, swans, planets, etc., as implicitly world-restricted claims. For
example:
(7) There are donkeys
is to be interpreted—by default, in ordinary contexts of use—as the (true)
claim
(R7) ∃x(Ixα & Dx)
10
The need for the genuine modal realist to consider a special treatment for modal claims about
the plurality of worlds has been widely recognized in the modal fictionalist literature (e.g. Menzies
and Pettit 1994; Noonan 1994). The suggestions in those papers and in this as to the form that the
treatment should take are developments of a hint given by Lewis (1986: 6) in remarks on restrictive
modifiers.
—that the actual world (α) has donkeys as parts. However, the genuine realist
cannot regard true non-modal existential claims about properties, numbers,
propositions—e.g.
(P) There is a plurality of worlds
(8) Natural properties exist
—as claims in which the quantifiers are world-restricted. If read that way,
they express falsehoods from the genuine realist’s standpoint—viz.:
(RP) ∃x∃y(Ixα & Iyα & Wx & Wy & ∼ (y = x) )
(R8) ∃x(Ixα & Nx).
If (P) and (8) are to express truths, as the genuine realist requires, the quantifiers
they contain should be interpreted as unrestricted, ranging over all of logical
space and not just the actual part of it—thus:
(RP*) ∃x ∃y(Wx & Wy & −(y = x) )
(R8*) ∃x(Nx).
Ordinary modal claims are interpreted by the realist as existential and uni-
versal quantifications into places held in corresponding non-modal claims by
world-restricting terms. So we have the translations:
(R9) There are donkeys iff ∃x(Ixα & Dx)
(R10) It is possible that there are donkeys iff ∃y(Wy & ∃x(Iyx & Dx) )
(R11) It is necessary that there are donkeys iff ∀y(Wy → ∃x(Ixy & Dx) ).
But sentences taken to express true advanced modal claims cannot be re-
garded as expressing generalizations of previously world-restricted contents.
In sentences taken to express true advanced modal claims—e.g.
(3) It is possible that P
(4) It is possible that there are natural properties
—the non-modal sentences that the modal items modify—viz.
(P) There is a plurality of worlds
(8) Natural properties exist
never expressed world-restricted contents in the first place. (Thus, see
(RP*) and (R8*) above.) There is no world-restricting element afoot in
the content of (P) or (8) to sustain subsequent existential or universal genera-
lization.
As an alternative, the genuine realist can say that in advanced modal
claims, the modal modifiers are redundant. When A is read as such a
claim, the genuine realist should impose the advanced possibility translation
schema:
(RP-A) It is possible that A iff A.
With regard to the typical advanced possibility claims that we have considered,
this schema yields:
(RP-A3) It is possible that there is a plurality of worlds iff
∃x ∃y(Wx & Wy & ∼(y=x) )
(RP-A4) It is possible that natural properties exist iff ∃x (Nx).
By invoking the classical definitions of necessity and contingency in terms of
possibility, the realist can derive from (RP-A) further schemas to deal with
advanced necessity and contingency claims respectively:
(RN-A) It is necessary that A iff A
(RC-A) It is contingent that A iff A and not-A.11
Intuitively, what holds of logical space unrestrictedly is not a contingent
matter, and in non-contingent matters what is possible is also necessary. Such
intuitions are captured by these realist schemas.
In formulating translation schemas for advanced modal claims, there are
two ways that the fictionalist might proceed. The most direct approach is to
apply the general fictionalist schema (F) to the genuine realist translations of
advanced modal claims that are provided by (RP-A), (RN-A), and (RC-A). This
approach produces the following:
(FP-A) It is possible that A iff According to GMR, A
(FN-A) It is necessary that A iff According to GMR, A
(FC-A) It is contingent that A iff According to GMR, A & not-A.
An indirect approach is to accept the directly obtained fictionalist translation
schema for one type of advanced modal claim (e.g. advanced possibility claims),
and then use the classical definitions of the other modal operators in terms
of that basic case to derive schemas for the other types of advanced modal
claims. The two approaches—direct and indirect—are not equivalent. Using
11
In (RC-A), this is contingency in the factive sense: it is contingent that A iff A and possibly
not-A. Hence, (RC-A) from (RP-A). Contrast contingency in the non-factive sense: it is contingent
whether A iff it is not necessary that A and it is not impossible that A.
the schema for advanced possibility as the basic case, the approaches provide
the same schema for advanced necessity but different ones for advanced
contingency. Beginning with the target sentence,
(9) It is necessary that A,
translation into its classical possibility equivalent gives us:
(10) It is not the case that it is possible that not A.
Applying (FP-A) to 10, the sentence obtained is equivalent to the right-side of
(FN-A):
(11) It is not the case that (According to GMR, not A).
If it is not the case that not A is true according to GMR, then A is true
according to GMR. Thus, the derived schema for advanced necessity would
be equivalent to (FN-A):
(DFN-A) It is necessary that A iff According to GMR, A.
Turning to contingency, we begin with:
(12) It is contingent that A,
and translate it into its classical possibility equivalent:
(13) A and possibly not A.
Applying (FP-A) to (13), the derived advanced contingency schema is:
(DFC-A) It is contingent that A iff (A and According to GMR, not A).
The major difference between this derived schema and (FC-A) is that, given
the consistency of GMR, (FC-A) renders all advanced contingency claims false
whilst on the indirect approach some advanced contingency claims must be,
from the fictionalist’s perspective, true. For example, the fictionalist must
affirm the crucial claim that it is contingent that not-P when that claim is
translated as a right-side instance of (DFC-A):
(14) It is not the case that P and According to GMR, P.12
However, we will now show that it does not matter whether the fictionalist
opts for the direct or indirect approach to the translation of advanced modal
12
Divers (1999b) had tried to earn for the fictionalist the right to assert the contingent falsehood
of P by appealing to (14), but we will now show why that move is unsatisfactory.
claims. Either way, she cannot escape her predicament. If the direct approach
is to be available to the fictionalist, then she must accept all three schemas
(FP-A), (FN-A), and (FC-A). But given the acceptance of (FP-A) and (FN-A),
the fictionalist will be committed, as the realist is, to affirming the equivalence
of advanced possibility and necessity:
(15) It is possible that A iff it is necessary that A.
This is disastrous. For with respect to the instance
(16) It is possible that P iff it is necessary that P,
the fictionalist needs to assert the left-side in order to avoid Hale’s objection
but needs to deny the right-side in order to avoid a commitment (via the
Brock–Rosen argument) to P. So the direct approach will not work.
If the indirect approach is adopted, then the fictionalist has to accept at least
one directly parasitic schema from which the others are derived. But problems
arise in connection with accepting any of them.
The fictionalist must reject the contingency schema (FC-A) in order to avoid
Hale’s objection; one who accepts (FC-A) is constrained, given the consistency
of GMR, to deny the contingent falsehood of P.
The fictionalist must reject the necessity schema (FN-A) in order to avoid
what might be considered a generalized version of the Brock–Rosen objection.
Where ‘Trans (Necessarily P)’ stands for whatever the appropriate fictionalist
translation of ‘Necessarily P’ is, a generalized form of the Brock–Rosen
argument is this:
1 (1) Trans (Necessarily P) Hyp
2 (2) (Necessarily P) iff Trans (Necessarily P) Hyp
1,2 (3) Necessarily P 1,2 iff
1,2 (4) P 3, T
Given (FN-A), the fictionalist is committed to the soundness of an instance of
this argument-form since she will be committed to the truth of its first two
premises:
(1*) According to GMR, P
(2*) Necessarily P iff According to GMR, P.
Finally, the fictionalist must reject the possibility schema since it commits her
to an absurd consequence—that according to GMR there is no plurality of
worlds—in conjunction with her claim that P is false:
1 (1) Not-P Hyp

1 (2) Possibly (Not-P) Possibility Introduction
3 (3) Possibly (Not-P) iff According to Instance of (FP-A)
GMR (Not-P)
1,3 (4) According to GMR, not-P 2,3 iff13
So the indirect approach is not an option either.14 The second stra-
tegic development of fictionalism, like the first, fails to vindicate fiction-
alism.
5. Conclusion
The challenge that we put to the fictionalist remains: show that fictionalism
can be developed in a way that justifies doing modal logic by proxy (or
delivers some other benefit) whilst avoiding the disjunction of modal collapse
and commitment to possible worlds. Since no development of the original
proposal has thus far proved capable of meeting that challenge, we conclude
that fictionalism is in a serious predicament from which it must be rescued if
it is to remain a viable option in modal philosophy.
13
Of course, the fictionalist could block this argument against (FP-A) by rejecting the unres-
tricted validity of possibility introduction just as one could block the (generalized) Brock–Rosen
argument by rejecting the validity of necessity elimination. But we take the validity of these rules
to be so obviously constitutive of our alethic modal concepts that to abandon them would be to
change the subject of translation.
14
There is another reason, besides the one given by the above argument by elimination, for
rejecting the indirect approach to translating advanced modal claims. Once the fictionalist rejects
the direct approach, she must retreat to a restricted version of the fictionalist schema (F) since she
no longer endorses all instances of (F), certain advanced contingency claims being the exceptions.
Yet the proof of the safety result depends on the necessitated fictionalist schema (NF) at Step
3 and the equivalence of that schema to (F) at Step 4. Given that only a restricted version of
the schema is now available to the fictionalist, the safety result, and so her justification of doing
modal logic by proxy, does not cover all modal claims. The result may still be significant even
if it covers inferences involving only ordinary modal claims and it is plausible that that was all
that the fictionalist originally envisaged, but the restriction promises to put the fictionalist at a
disadvantage to the realist. As for the realist, there is no obvious reason why she should not be
able to extend the first-order, proxy treatment of modal inferences to cases involving advanced
modal claims.
References
Brock, Stuart (1993), ‘Modal Fictionalism: A Reply to Rosen’, Mind, 102: 147–50.
Divers, John (1995), ‘Modal Fictionalism Cannot Deliver Possible World Semantics’,
Analysis, 55: 81–9.
(1999a), ‘A Genuine Modal Realist Theory of Advanced Modalizing’, Mind, 108:
217–39.
(1999b), ‘A Modal Fictionalist Result’, Noûs, 33: 317–46.
Hale, Bob (1995a), ‘Modal Fictionalism—A Simple Dilemma’, Analysis, 55: 63–7.
(1995b), ‘A Desperate Fix’, Analysis, 55: 74–81.
Lewis, David (1968), ‘Counterpart Theory and Quantified Modal Logic’, Journal of
Philosophy, 65: 113–26.
(1983), ‘Postscript to ‘‘Counterpart Theory and Quantified Modal Logic’’ ’,
Philosophical Papers, (Oxford: OUP), 39–46.
(1986), On the Plurality of Worlds (Oxford: Basil Blackwell).
Menzies, Peter, and Pettit, Philip (1994), ‘In Defence of Fictionalism about Possible
Worlds’, Analysis, 54: 27–36.
Noonan, Harold (1994), ‘In Defence of the Letter of Fictionalism’, Analysis, 54: 133–9.
Rosen, Gideon (1990), ‘Modal Fictionalism’, Mind, 99: 327–54.
(1993), ‘A Problem for Fictionalism about Possible Worlds’, Analysis, 53: 71–81.
(1995), ‘Modal Fictionalism Fixed’, Analysis, 55: 67–73.
4
On Realism about Chance
Philip Percival
‘Chance’ is a single-case, temporally relative, objective probability for which

a normative principle of direct inference holds. It is single case in so far as
it applies to particular events, such as the explosion of the Challenger Space
Shuttle (or, better, to propositions asserting the occurrence of such events); it
is temporally relative in so far as it changes over time, as when, seemingly, the
chance of the Challenger exploding increased as the temperature dropped; it
is objective in so far as claims of the form ‘the chance at t of E is r’ purport to be
true independently of what anyone thinks; and a normative principle of direct
inference holds for it in so far as the chance of an event, or at any rate a cognitive
attitude, such as belief, or credence (i.e. partial belief), towards that chance,
normatively constrains cognitive attitudes towards the event itself, in the
manner, e.g. of a requirement that an agent who believes there to be a one half
chance of heads on a certain coin-toss ought, in the absence of direct evidence
regarding the outcome, to distribute his credence in heads and tails equally.
These stipulations are not arbitrary: at worst, they constitute well-motivated
sharpenings of one use of the word ‘chance’. But although features of a concept
I am grateful to Jim Edwards, Barry Loewer, and Peter Milne for helpful conversations, as I am
to the Arts and Humanities Research Board for funding the research leave during which the first
draft of this paper was written. I am unusually and especially heavily indebted to Christopher
Hitchcock, who corrected several misunderstandings on my part. Without his careful and patient
explanations of certain aspects of non-standard analysis and their bearing on the problem of fit,
my discussion of this problem in Section 2.3 and the Appendix would have contained several
errors it now avoids. Responsibility for any errors that remain is entirely mine.
Philip Percival / 75
can be stipulated, the existence of something falling under the concept cannot
be. It remains to be settled whether chance exists.
To say that chance exists is to say that some statement of the form ‘the
chance at t of E is r’ is objectively true. I shall focus on the question as to whether
‘non-trivial’ chance—that is, chance having values other than 0 or 1—exists.
I call the view that it does ‘realism’ about chance.1 Thus defined, realism about
chance is neutral regarding whether chance is a (natural) property, tendency,
disposition, or ‘display’ of any such.2 It is also neutral as to whether chance
is analysable, Humean supervenient, or supervenient on non-chance.3 And it
is neutral as to whether chance is constrained by our capacities to ascertain
its values. It might not be neutral with respect to determinism: many authors
maintain, albeit controversially, that the existence of non-trivial chance
requires the falsity of determinism.4 But I shall not address the objection
that realism about chance is false because determinism is true. To sideline
it, I shall assume—as quantum mechanics leads many to suspect—that
determinism is false.5
1
There is no established terminology in this area. Black’s (1998) use of ‘realism’ about chance
lacks the neutrality of mine.
2
See Giere (1973), Mellor (1969, 1971), Popper (1959), and Sapire (1991).
3
Non-chance at a world w comprises everything at w other than chance and what presupposes
chance. It includes all of w’s particular events, like deaths and the breaking of chemical bonds,
and the instantiations at it of properties and relations like negative charge and being the father of.
But other features of non-chance are controversial. For example, whether it includes causes at
indeterministic worlds depends on controversial features of indeterministic causation. McDermott
(1999: sect. 2.6) argues that even if realism about chance is correct, counterfactuality does not
supervene on non-modal matters (including chance). Since he offers a counterfactual analysis of
causation, in my terms this commits him to holding that non-chance includes indeterministic
causes. But indeterministic causes are excluded from non-chance by those such as Lewis (1986:
175–84) and Noordhof (1999) who advance analyses of indeterministic causation in terms of
chance. Either way, the thesis that chance supervenes on non-chance is neither entailed by, nor
entails, the thesis that chance is ‘Humean supervenient’ in Lewis’s (1986: pp. ix–xvii) sense, since
(i) it does not impose stringent requirements (such as localization to spacetime points) on the
‘non-chance’ properties in the subvenient base; (ii) it is not committed to the assumption that
all combinatorially possible distributions of properties in the base are metaphysically possible,
and (iii) it is not consistent with the supposition that chance is an unanalysable natural property
localized at spacetime points and subject to the combinatorial hijinks described in (ii). (Cf.
Sturgeon (1998).)
4
See Mellor (1971), Giere (1973: 475), and Lewis (1986: 117–21). Mellor (1982, 1995) subsequently
retracts his earlier view, and argues that non-trivial chance is compatible with determinism.
5
Cf. Lewis (1986: 58–9).
76 / On Realism about Chance
In recent years, the terms of the debate over realism about chance have
been set by a series of studies in which David Lewis (1980, 1986, 1994) develops
a stunning exposition and defence of (one version of ) the doctrine. Ironically,
however, his ambivalence as to whether chance supervenes on non-chance
embodies a dilemma that threatens realism’s coherence. At the outset, Lewis
(1980, 1986) felt forced to eschew the supervenience of chance on account
of so-called ‘undermining’: if chance were supervenient, the truth of the
proposition that captures the way the subvenient base actually is would
determine chance in such a way as to give some chance of that proposition
being false, and, hence, of the chances not being what they in fact are.6 By the
end, however, Lewis (1994: 484–5) embraces with evident relief a ‘best-systems’
analysis that ensures chance’s supervenience: if chance were not supervenient,
rational credence would not be something it must be—namely, supervenient
on symmetries and relative frequencies.
Of course, this volte face does not entangle Lewis (1994) in double-think:
he reckons the undermining difficulties he had raised earlier to have
been resolved. Nevertheless, his unwavering advocacy of realism about
chance notwithstanding, the dilemma still has bite: amongst realists about
chance, each horn continues to attract passionate advocates.7 I shall not
explore the possibility of developing the dilemma into a full-blown argument
for the incoherence of the concept of chance, however. Although I shall argue
that the best-systems analysis Lewis came finally to espouse is untenable, I
6
Undermining in this sense is barely credible in itself, but Lewis (1980: 109–13; 1986: pp. 121–31,
xiv–xvii) is more concerned with its apparent inconsistency with the ‘Principal Principle’—a
normative principle connecting credence about a proposition’s chance with credence in the
proposition itself—that he takes to be constitutive of the notion of chance. (This principle is
explained in n. 24, below.) Lewis’s worry has met with several responses. See Vranas (2002) for
the retort that the apparent inconsistency rests on the assumption that the supervenience of
chance on non-chance is necessary (an assumption questioned by Lewis’s claim that Humean
supervenience is only contingently true). And see Thau (1994) and Hall (1994) for the retort that
the Principal Principle is false, but not constitutive of the notion of chance. Lewis (1994) himself
favours a third response. He holds that although the Principal Principle is constitutive of the
notion of chance but only approximately true, this much is consistent with the existence of
chance: he argues that chance is that feature of the world that best approximates the principles that
are constitutive of chance. (See too Strevens (1995). Hall (2004) has further references.)
7
Regarding the first horn, Bigelow, Collins, and Pargetter (1993) argue that quite apart from its
impact on the Principal Principle, undermining is absurd in itself. (See n. 6, above.) Regarding the
second horn, compare Hall’s (2004: 94) remark that ‘it is widely assumed that . . . it is unacceptably
mysterious’ why, if chance is not Humean supervenient, rational credence should be constrained
by chance in the way that it is.
shall not do so with a view to evaluating the dilemma. In particular, I shall
have nothing to say about the undermining difficulties raised for this analysis
in its first horn. My aim is merely to question whether, on the assumption
that realism about chance is coherent, there is any reason to believe it.
Some take an affirmative answer to this question to be relatively straightfor-
ward: contemporary science seems to attribute chances to various events, and,
following Quine, this fact alone, they say, warrants belief in the existence of
chance.8 I think this line of thought superficial, however. That science appears
to employ chance in describing the world, and formulates laws concerning it, is
no doubt a prima facie reason for realism about chance. But the instrumentalist,
constructivist, and anti-realist traditions in the philosophy of science have
enough going for them for there to be no guarantee that this appearance will
withstand critical scrutiny. Science is good, but it is not that good.
More substantial groundings might be attempted for belief in chance, but I
shall argue that they too are inadequate. One way to demonstrate the existence
of chance would be to show that currently accepted facts about non-chance
necessitate it, and some analyses of chance might purport to do just that. The
least worse of those that are currently available—Lewis’s—fails, however, on
account of a defect that has passed unnoticed in the furore over undermining
(Section 2). In the absence of a successful analysis, the realist must search for
an inductive reason for believing in chance. The most promising course would
be to try to identify an explanatory role chance best fulfils, and then appeal
to inference to the best explanation. There are two candidates for such a role,
but neither gives the realist what he needs (Section 1).
1. Chance and Inference to the Best Explanation

Chance has been invoked with a view to explaining statistical phenomena,9
and, more idiosyncratically, the warrantedness of certain credences.10 I shall
argue that neither consideration supports an inference to the best explanation
from non-chance to chance: chance is explanatorily vacuous with respect
8
Cf. Lewis’s (1986: 178) remark: ‘You may not like single-case chances—I don’t either—but I
cannot see how to make sense of certain well-established scientific theories without them.’
9
See Mellor (2000: 20) and Hall (2004: 110). Hall seems to maintain that it is only if chance does
not supervene on non-chance that it can explain indeterministic statistical phenomena.
10
See Mellor (1969, 1971, and 1982).
to indeterministic statistical phenomena (Section I.1), while attempts to
introduce it as a ground of warrant beg the question (Section I.2).
1.1 Statistical Phenomena

Let ‘statistical phenomena’ include both statistical regularities concerning the
relative frequencies that possible outcomes of certain processes tend to in
large-scale trials, together with these regularities’ instances and the individual
events that make them up. A statistical regularity takes some such form as: for
all large-scale trials on a set-up S, the relative frequency with which outcome
O is obtained tends to (approximately) r. An instance of the regularity is then
some such trial in which the relative frequency of the outcome O tends to
(approximately) r. For example, a statistical regularity involving a fair coin
might consist in the fact that there are three-hundred large-scale trials in
which the coin is tossed, in each of which the relative frequency of heads tends
to (approximately) 1/2. The fact that the relative frequency of heads tended
to 1/2 on the seventeenth of these trials is an instance of the regularity. The
individual events that comprise this instance are its successive tosses and their
outcomes.
Indeterminism must be defended if chances are to be advocated on the
grounds that they play a vital role in the best explanation of statistical
phenomena. This is a point regarding explanation that is independent of the
metaphysical issue alluded to earlier regarding whether or not the existence
of chances other than 0 and 1 is compatible with determinism: it remains valid
even if non-trivial chance and determinism are compatible.11 If determinism
is true, all actual phenomena not involving chance are inevitable at times
prior to their occurrence: propositions describing the events that constitute
them are entailed by propositions that describe prior (‘initial’) conditions
and physical laws without mention of chance. Since entailments of this kind
are characteristic of paradigmatic ‘deductive nomological’ explanations, both
deterministic statistical regularities, and events comprising their instances, are
explicable without invoking chance.12 No room remains for a claim of the
11
See p. 75, above.
12
One might object that this attitude to the explanation of deterministic statistical regularities
is insufficiently sensitive to the distinction between an instance of a statistical regularity and
the regularity itself, and that though the fact that e.g. 100 successive deterministic tosses of a
coin on some occasion yielded 47 heads might be explained in the manner envisaged, the fact
form: such events and regularities are best explained by invoking chance. If
determinism is true, there are ideal explanations of them involving no appeal
to chance. Although we might not know these explanations individually, or,
even, have any means of knowing them, belief that determinism is true obliges
belief that they exist.
Nevertheless, I am happy to concede that belief in indeterminism is
reasonable: quantum mechanics might well provide grounds for it. But
what exactly is the crucial explanatory role chance is supposed to play with
respect to indeterministic statistical phenomena? It is natural to focus first on
statistical regularities. Certainly, it is no easier to accept ‘brute’ indeterministic
regularities of this kind than it is to accept the brute non-statistical regularities
that have been the primary focus of the post-Humean controversy over
fundamental deterministic laws. Indeed, whereas the latter have prompted
some anti-Humeans to hypothesize a contingent relation of necessitation
between universals in an attempt to reduce their apparent arbitrariness,13 the
attempt to explain statistical regularities via chances seems more compelling. A
natural explanation of why this should be so is suggested by claimed analogies
between the theoretical roles played by chance and, e.g., mass and force.14 It
views the difference between appeals to a necessitation relation, and appeals to
chance, as the difference between pseudo-science and the real thing. Appealing
to a relation of necessitation between universals so as to explain fundamental
non-statistical regularities is pseudo-science: so doing tries to get something
on the cheap when science stops. In contrast, appealing to chance to explain
indeterministic statistical regularities is just science: talk of chance is integral
to physics.
One should be wary both of this explanation, and of the apparent asymmetry
for which it purports to account, however. Even if talk of chance is integral
to physics, it does not follow that science gives chance an explanatory role to
play with respect to statistical (or any other) phenomena. Talk of numbers is
integral to physics too, but it can hardly be supposed that physics exploits the
explanatory power, e.g. of the number 49. Moreover, the simple model that
underlies naı̈ve enthusiasm for the explanatory pretensions of chance is quite
that it regularly happens that when the coin is tossed deterministically a large number of times
approximately 50% heads results cannot be. The most this consideration could show, however, is
that ideal explanations of deterministic statistical regularities must advert to statistical regularities
amongst initial conditions.
13 14
See Dretske (1977) and Armstrong (1983). Giere (1973: 473–4, 486).
untenable. It takes chance to play a direct role in the explanation of statistical
regularities as follows:
Explanans Each G has chance r of being F.
Explanandum The relative frequency of F’s in each ‘trial’ comprising
large numbers of G’s tends to (approximately) r.
That explanations of this kind have an immediate appeal is undeniable.15 But
their appeal is illusory. It stems from an entailment that acts as an intermediary
between the explanans and the explanandum, and once this entailment is made
explicit, the explanation is exposed as a sham. The intermediary entailment
is the proposition that the chance of the relative frequency of F’s among
large numbers of G’s tending to (approximately) r is high. When this is made
explicit, the model splits into two:
Explanans1 Each G has chance r of being F.
Explanandum1 / The chance of the relative frequency of F’s among large
Explanans2 numbers of G’s tending to r is high.
Explanandum2 The relative frequency of F’s in trials comprising large
numbers of G’s tends to (approximately) r.16
Thefirstsub-explanationisimpeccable,sinceexplanans1 entailsexplanandum1 .
But it is this fact alone that accounts for the original explanation’s appeal.
For the second sub-explanation is spurious. Even if—and, perhaps, especially
if—the chances involved do not supervene on non-chance, how is the fact
that a proposition had a ‘chance’ that was ‘high’ supposed to explain its
truth? Why, as must be the case if belief in the existence of chance is to
be grounded in this simple model, is the explanation of an indetermin-
istic event by its chance the better the higher that chance? Since there are
neither immediate nor mediate deductive connections between the propos-
ition asserting of an event that it has a high chance, and the proposition
asserting the event’s occurrence, what is the connection between these two
propositions such that the one explains the other? The only specifiable
15
Mellor (2000: 20) writes that ‘[T]he chance ch(H) = p of heads on a coin toss explains the
fact (when it is fact) that the frequency f(H) of heads on many such tosses is close to p.’
16
The reason Mellor (2000: 20) gives for his claim that the chance of heads on each toss of a
coin explains the fact that a relative frequency of heads in the long run which approximates that
chance results is that ‘(given the laws of large numbers) the chance of this fact is close to 1’.
connection is epistemic: the truth of the former warrants high credence
in the latter. But in that case a dilemma arises. If the ‘explains’ relation is
never epistemic, chance is impotent: the high chance of an event cannot
explain it. If, however, the ‘explains’ relation is sometimes epistemic, chance
is superfluous: we already possess epistemic relations in abundance—namely,
evidential relations—in terms of which to explain indeterministic events
without invoking chance.17
The force of the first horn is clear enough, but let us consider an example
that serves to illustrate the force of the second. Suppose that the relative
frequency of heads when some coin is tossed a large number of times tends
to 2/3. Realists themselves often emphasize that postulating a chance of heads
on each toss equal to 2/3 to explain this fact only has bite in so far as this
chance can be linked in a law-like way to other properties of the coin and the
tossing device.18 In the case at hand, call these properties, ‘bias’ and ‘fairness’
respectively. Ex hypothesi, bias and fairness together comprise evidence, of certain
strengths, that the outcome of a given toss will be heads, and that the relative
frequency of heads in large-scale trials will tend to 2/3. So if epistemic relations
can suffice for explanatoriness, chance is an idle intermediary: bias and fairness
possess the requisite explanatory power already.
In part because of the failure of the simple model, some authors maintain
that even if chance exists, indeterministic matters—indeterministic statistical
regularities, their instances, and the events that constitute them—are inex-
plicable, and that those who maintain the contrary confuse the explanation
of the chance of an event with an explanation of the event itself.19 On the
other hand, their opponents take scepticism of this ilk to wrongly attribute
the defects of the simple model to what is modelled, and turn to models of
indeterministic explanation that are more refined.
Though these refinements are to be applauded in some respects, they too
fail to underpin realism about chance. This much is obvious in the case
17
Although Mellor (1976: 235–6) argues that an explanandum is better explained the higher its
chances according to the explanans, even he does not insist that only chance provides probabilistic
explanations. On his account, explanans and explanandum must be linked by a high epistemic
probability. But while Mellor holds that chance can underpin this probability, he declines to rule
out the possibility of explanations in which the epistemic probability is not so underpinned. He
writes ‘Perhaps all explanatory probabilities are chances, perhaps not.’
18 19
Mellor (1969: 26). Woodward (1989: sect. 3).
of Railton’s (1978) deductive nomological model. Let E be an indeterministic
event. According to Railton, the core of an explanation of E comprises a
deductive argument from chance laws and initial conditions to the conclusion
that, immediately prior to its occurrence, E had a chance r of occurring. The
explanation is completed by conjoining this argument with the statement
that E in fact occurred. The salient feature of this model is that how good
the explanation is does not depend on the chance accorded the event to be
explained: provided the chance is ascribed correctly, the explanation is equally
good whatever the chance ascribed E. Clearly, belief in the existence of chance
cannot possibly be grounded by an inference to the best explanation if the
model is correct. For the model presupposes chance. There is no merit in the
thought that it is reasonable to believe that chance exists because certain
explanations that conform to Railton’s model are best.
For the realist’s purposes, alternative models that exploit the thought that
indeterministic events can have causal histories are more promising. Once
this thought is granted, indeterministic events can be given both ‘plain’
explanations, and, pace Lewis (1986: 229–31), ‘contrastive’ ones too: for one can
explain why an indeterministic event occurred by citing one of its causes, and
one can explain why it occurred rather than some other event that might
have occurred instead by citing a cause that meets certain conditions.20
Models of this kind suggest a less direct strategy for introducing chance
via its role in the explanation of indeterministic events. In full generality,
the strategy is this. What makes something an explanans of an explanandum
is the existence of a certain kind of link between them. This link must be
objective. In the deterministic case, logic itself provides the link. But logic
cannot be the link in the indeterministic case, and whatever one calls it the
link can only arise from chance. On this strategy, chances must be invoked not
because they themselves explain indeterministic events, but because they make
the explanation of one indeterministic event by another possible (in a way
Railton’s model fails to capture). An obvious, and not unattractive instance
of this strategy, is to take the link between explanans and explanandum to
be causation, the idea being that indeterministic events are only explicable if
20
A ‘plain’ explanation of (the occurrence of ) an event E answers a ‘plain’ why-question of
the form ‘Why did E occur?’ A ‘contrastive’ explanation of (the occurrence of ) an event E answers
a ‘contrastive’ why-question of the form ‘Why did E occur rather than E*?’ See Hitchcock (1999)
and Percival (2000) for defences of the claim that contrastive explanations of indeterministic events
are possible.
they have causes, and that chances must be postulated in order to account for
indeterministic causation.
I doubt that this strategy can succeed, however, for two reasons. First,
whether indeterministic causation presupposes chance is controversial. Con-
sider the following example. Suppose R-atoms form a large population,
and that 0.001% of them decay within the hour when left to their own
devices, whereas 10% of them decay immediately upon being struck by a
q-particle. Suppose decay is indeterministic. Consider an R-atom that is
struck by a q-particle and then decays. Did the q-particle cause the atom’s
decay? For some, an affirmative answer turns on issues about chance—for
example, on such matters as what the chance of the atom’s decay was
when it was struck, what the chance of decay would have been had it
not been struck, and whether the first of these chances is greater than
the second, etc. Much of the rationale for this diagnosis, however, stems
from the conviction that indeterminism prevents counterfactuals like ‘had
the q-particle not struck the atom, the atom would not have decayed
when it did’ from being true. But if, as others maintain, this conviction
is misplaced, the way is clear for a counterfactual analysis of causation
to be retained in the shift from determinism to indeterminism. On this
approach, the q-particle’s impact can be supposed to have caused the
atom’s decay without having to assume that chance exists.21 Admittedly,
on this approach, counterfactual truth can’t supervene on actuality: on the
assumption of indeterminism, nothing actual—not even the entire course of
history—suffices to make it true that if the atom hadn’t been struck by the q-
particle, it wouldn’t have decayed. To my mind, however, some realists about
chance—namely, those who deny chance supervenes on non-chance—are
in no position to dismiss those who deny that counterfactuality supervenes on
actuality.
A second objection to the strategy of arguing for the existence of chance
indirectly, via the claim that only the existence of chance renders any
explanation of indeterministic events possible, addresses a wider audience. It
too focuses upon certain counterfactuals. Unlike the first, however, it does
not invoke non-supervenient counterfactuals. It concedes that indeterminism
prevents counterfactuals like ‘had the atom not been struck by the q-particle,
it would not have decayed when it did’ from being true. But it exploits
21
See McDermott (1999).
counterfactuals about rational credence of the form ‘had the atom not been
struck by the q-particle, rational credence in the atom’s decaying when it
did would have been r’. Many counterfactuals of this ilk remain true under
indeterminism. Of course, as Mellor is fond of observing, credences, rational
or otherwise, cannot themselves explain the worldly events at which they are
directed. This in itself does not prevent sufficient conditions for the explanation
of an indeterministic event being given in terms of rational credence, however.
Suppose process P is indeterministic, that it had outcome E but might have had
outcome E*, and that in large scale trials that token P repeatedly, the relative
frequency of E tends to some r close to zero, while that of E* tends to some
r* close to one. Let C be an event which ‘interferes’ with P, and that amongst
tokens of P attended by C, the relative frequencies of E and E* are reversed.
Relative to statistical evidence, then, rational credence in outcome E is close to
one unless P is attended by C, in which case it is close to zero. Now consider an
agent who, believing a token of P not to have been attended by C, is surprised
when its outcome is E. His question ‘Why E rather than E*?’ might surely be
answered to his subsequent satisfaction if he is told ‘Because appearances to the
contrary, C attended P’. Under indeterminism, counterfactual dependence of
rational credence can still provide the objective link between explanans and
explanandum that explanation requires.
1.2 Temporally Relative Warrant

Mellor (1971: 58–62) writes:
I propose to account for chance in terms of a feature of the world, ascertainable by
the methods of science, that warrants adopting some partial beliefs rather than others
. . . On the present analysis, ascribing chances to single trials expresses the fact that
their function is to warrant certain partial beliefs on the possible outcomes of such
a trial.
Mellor’s basic idea seems to be that, primarily, the explanatory role of chance
is to ground the warrantedness of certain credences. This idea generates an
inference to the best explanation as follows:
(1) There are times t and events E such that, at t, one’s credence in E
ought to be r (0<r<1), i.e. so that at t (only) credence in E to degree
r is warranted.
(2) The best explanation of why certain credences are warranted in this
sense is that there is something in the world that warrants them.
So
(3) There is a feature of the world—‘chance’ (of E to degree r)—that
grounds these credences’ warrant.
This inference reverses the customary normative transition from chance
to credence: normatively, the direction is from (credence about) chance to
credence, whereas the inference extracts the existence of chance from norm-
atively constrained credence. In this, Mellor’s strategy bears some similarities
to Lewis’s (1980).22 But it is different in two important respects. First, Lewis seeks
merely to read off the formal properties of chance from those of credence:
in contrast, in trying to establish the existence of chance via (warranted)
credence, Mellor attempts, so to speak, for the matter of chance what Lewis
attempts for its form.23 Secondly, whereas the normative principle governing
chance to which Lewis appeals—the Principal Principle—is internalist in so
far as it constrains rational credence in an event by the agent’s credence in
the event’s chance,24 Mellor relies on a principle that is externalist: he takes
a certain credence at t in an event to be warranted by the chance at t of the
event itself.
Mellor’s externalism about the normative power of chance is embodied in
premise (2), and constitutes one respect in which this premise is controversial:
some internalists will object that the idea of a feature of the world having
the power to demand a certain credence is no less absurd than the moral
realist’s idea of a feature of the world—moral wrongness—having the power
to demand a certain action.25 But (2) is controversial in a second respect that
is even more fundamental. For it presupposes that epistemic norms enjoy an
22
Lewis (1980: acknowledgement) writes ‘I am also much indebted to Mellor (1971), which
presents a view very close to mine; exactly how close I am not prepared to say.’
23
In particular, Lewis tries to show that chance is a probability, that it is temporally relative,
and that the past has a chance that is equal to 1. (In contrast, I have accorded chance the first two
of these features by definition.)
24
Let X be the proposition that the chance at t of p is x, and E be any proposition compatible
with X that is ‘admissible’ at t. Then the Principal Principle states that an initial credence function
is reasonable only if credence in p conditional on X&E, is equal to x. (Here, admissible propositions
‘are the sort of information whose impact on credence about outcomes comes entirely by way of
credence about the chances of those outcomes’. See Lewis (1980: 86–96).)
25
Cf. Black’s (1998: 384) remark that ‘credence makers [that are not Humean-supervenient]
look metaphysically ‘‘queer’’ in rather the way in which John Mackie thought objective values
queer—they build a strange bridge between what is ‘‘out there’’ in the world . . . and something
from a different conceptual realm (reasonableness of confidence in an occurrence)’.
objectivity that is often disputed. After all, if there is no fact of the matter
as to whether the credences in question are warranted—as there isn’t if
sentences of the form ‘S’s credence to degree r in p at t is warranted’ have
meanings of such a kind that their proper ‘assertoric’ use is restricted to the
expression of certain pro-attitudes, rather than beliefs that are objectively true
or false—then there is nothing relevant to be explained.
Fortunately, there is no need to become embroiled in these controversies.
The inference can be seen to fail even if (2) is conceded. Let us call the warranted
credences adverted to in (1)–(3) ‘M-warranted’. Jointly, (1) and (2) yield:
(M) For all r, p and t: credence r in p at t is M-warranted iff the chance

of p at t is r.
Given (M), M-warrant must share chance’s temporally relativity (and objectiv-
ity). It differs markedly, therefore, from the concepts of warrant that are
employed in everyday epistemic appraisals. Centrally, these concepts invoke
available evidence, and cognitive abilities. As such, they pertain to an agent’s
state of information, and his perceptual and intellectual faculties, but not to
time. With respect to them, Smith’s credence r at t in p might be rational, or
warranted, even though Jones’s credence r at t in p is irrational, or unwar-
ranted: for Smith might possess evidence for p, or a faculty like vision, that
Jones lacks. In contrast, M-warrant is relative to time but not to anything
else: if any agent’s credence r at t in p is M-warranted, so too is every
agent’s.
It follows that (1) is very far from being a datum. On the contrary, it begs
the question as to whether there is any reason to believe that some credences
enjoy M-warrant. Of course, an independent reason for believing in chance
might well provide, via chance’s normativity, an indirect reason for believing
in M-warrantedness. But this is irrelevant: only a direct reason will do, since
the whole point of the inference under discussion is to introduce chance via
M-warrant.
Upon reflection, it is clear that there cannot be a direct reason for believing
that some credences have M-warrant. The problem is not M-warrant’s
temporal relativization as such, since temporally relativized notions of warrant
are readily definable in terms of evidential warrant. Let Tw be the proposition
that captures a possible world w’s course of history up to t. Then a temporally
relativized notion of T-warrant is defined by:
(T) For all w, r, p, and t: at w, credence r in p at t is T-warranted iff total
evidence Tw uniquely evidentially warrants credence r in p.26
T-warrant is not M-warrant, however. At t, credence r in p is M-warranted iff
at t, r is the chance of p. But the credences that are T-warranted at t do not
have the same values as the chances at t. If t is very shortly after the Big Bang,
the course of history up until t won’t constitute enough evidence to ensure
that for all p, rational credence in p based on this evidence alone coincides
with the chance of p at t.
Other temporally relative notions of warrant are definable in terms of
evidential warrant, however. To help motivate their definitions, and to see
why none of them can be identified with a notion of M-warrant that serves
the realist strategy under discussion, consider a paradigm case of empirical
judgement regarding the chance of an event. The judgement is retrospective,
and it is grounded in statistical information about the relative frequency of an
event of that type in an appropriate reference class. For example, imagine a
newly constructed downward-sloping pin table. There are eight slots in which
a ball, first projected by a rigid lever that has just one setting, and then winding
its way down through the net of pins, might settle. Suppose one knows
beforehand that, at unit-intervals, the table is to be tested a thousand times,
beginning at t + 1. Let pm,n be the proposition that on the mth trial, the ball
will settle in slot n. An observer S who at t distributes his credence over the p1,n
is only rational if he distributes it in accordance with evidence ES,t he possesses
at t, i.e. so that cS,t (p1,n ) = c(p1,n |ES,t ), where c is a credence function that
properly respects evidence. Different rational observers will distribute their
credences differently, in accordance with their possession of different evidence.
If any of these credences are the same as the chances, that is just coincidence.
Initially, then, an observer’s credences are appraised according to whether
they respect his evidence. The appraisal is not in terms of M-warrant.
Now consider the situation at t + 1001. The relative frequency of slot n after
m trials is seen to be fm,n and, as m grows, fm,n is seen to appear to stabilize
around rn . In the light of this new information, the various observers’ initial
credences in the result of the first trial can be reappraised. It can now be
seen that, irrespective of the evidence then available to the agent, credences
26
Of course, such a definition idealizes customary evidential norms. At best, these norms
constrain credences by evidence. Unlike (T), they do not require a unique distribution of credence
conditional upon evidence.
ct (p1,n ) = rn have certain virtues that went unrecognized initially.27 Let these
virtues be acknowledged, retrospectively, by calling them ‘Rm -warranted’.
The argument just given for the claim that an agent’s credences are subject
to a notion of Rm -warrant that is independent of the evidence available to the
agent, is the closest one can get to a direct argument for M-warrantedness. It
falls short of what is required, however. Rm -warrant is not the requisite notion
of M-warrant, even if it is temporally relative, in the sense that for all agents S
and S*, cS,t (p1,n ) = r is Rm -warranted iff cS∗,t (p1,n ) = r is Rm -warranted. It is
only the case that for some m, Rm -warrant is M-warrant, if there is a guarantee
that the values of the credences entertained at t that are Rm -warranted are the
chances at t. But there can be such a guarantee only if the chances at t are defin-
able in terms of the statistical evidence in terms of which Rm -warrant is defined.
So a dilemma arises that is fatal to any attempt to introduce chance on the back
of an independently introduced notion of Rm -warrant. If chance is not defin-
able in terms of the statistical evidence in terms of which Rm -warrant is defined,
the attempt fails. But if chance is so definable, the attempt is superfluous.
Unless there is some m such that chance is definable in terms of Rm -warrant,
for no m is Rm -warrant the same as M-warrant. The moral is that Mellor’s
strategy fails. There can be no direct, prior evidential ground for introducing
M-warrant. To have reason to believe that some credences are M-warranted,
one must first have reason to believe that chance exists.
2. Analyses of Chance
Attempts to ground realism about chance by an inference to the best
explanation of some feature of non-chance do not succeed. The realist might
seek to circumvent the need for such an inference, however, by trying to
analyse chance in terms of non-chance.28
I shall argue that this strategy is to no avail. If chance can be analysed in terms
of non-chance, it supervenes on non-chance. Analyses of chance in terms of
non-chance can be classified according to the kind of supervenience thesis they
27
Exactly what these virtues are, and how highly they should be valued, is debatable, but
irrelevant.
28
In saying this, I do not presume that chance lacks explanatory powers if it is analysable. The
point is just that even if analysable chance explains some part of non-chance, the gap between
non-chance and chance can be negotiated without inference to the best explanation.
entail: ‘molecular’ analyses entail a (relatively) local supervenience of chance
on non-chance, while ‘holistic’ analyses entail one that is (relatively) global.29
I shall argue that extent molecular (Section 2.1) and holistic (Section 2.2–2.4)
analyses of chance fail.
2.1 Frequentist Analyses

Frequentist analyses of chance purport to define the chance of an event as the
relative frequency with which a certain outcome is obtained in a reference class
of ‘trials’.30 The naive manifestation of this hope defines the chance an event
E has at a time t as the relative frequency of an event of the same kind as E in
a reference class comprising actual instances of some process that would have
to take place were E to occur. Often, however, as in the case of a coin that is
tossed once and then destroyed, the reference class that seems most relevant is
too small for the relative frequencies of outcomes in it to be plausibly thought
a strictly reliable guide to chance. More sophisticated proposals attempt a
definition of chance in terms of relative frequencies of outcomes in reference
classes that are merely hypothetical. For example, irrespective of whether a
coin is ever actually tossed, one might try defining the chance of its landing
heads if it is tossed on some specific occasion as the relative frequency (or
limiting relative frequency) of heads that would obtain were it subjected to
some ideal sequence of tosses.
Unfortunately, although these proposals are in some respects an improve-
ment on the naive one, they succumb to similar difficulties. Whether there
are facts about the relative frequencies with which the possible outcomes of
an indeterministic process would be obtained were it tokened repeatedly is
controversial: since a fair but never tossed coin might have landed heads on
one-thousand successive occasions had it been tossed a thousand times, can it
be a fact that it would have landed otherwise?31 To be sure, McDermott’s (1999)
29
The supervenience of chance on non-chance is strictly ‘global’ iff for no proposition p is
there a proper part of a world w on which the chance of p at t at w supervenes. If the supervenience
of chance on non-chance is not strictly global, it is less global (equivalently, more ‘local’) the
smaller are the various proper parts of non-chance at w on which the chances at w supervene.
30
Frequentist analyses of chance are not species of ‘frequentism’ in the classical sense. Classical
frequentists, such as von Mises (1957: 17–18), simply deny that chance in my sense exists: they
hold that although objective probability exists, it is not single-case.
31
Cf. Lewis (1986: 90, 176). Lewis (1986: 58–65) does maintain that {¬A, A →C, A →
p(¬C) > 0} is consistent, but he holds that these propositions can only all be true together in the
special case in which, had A occurred, the occurrence of ¬C would have been ‘quasi-miraculous’.
‘realism’ about counterfactuals paves the way for commonplace facts about
what the outcome of an indeterministic process would have been had the
process been instigated. But even if there is a fact about the relative frequency
with which a possible outcome would have obtained had some indeterministic
process been tokened n times, identifying this relative frequency with the
chance (beforehand) of obtaining an outcome of that kind on the one occasion
the process is actually tokened does violence to the concept of chance: there is
always a positive chance that the two differ. Nor does resorting to the infinite
case resolve these difficulties. It is even more controversial to suppose that
there are facts about what the limiting frequency of an outcome would have
been if the relevant process had been tokened infinitely often. But even if
there are such facts, it is hard to see how such a sequence of tosses could
avoid physical changes in the trial set-up that affect the chance distributions
over outcomes on individual trials. And even if such changes were somehow
avoided in a fact of the form ‘were an infinite series of trials conducted on
the trial set-up T, the limiting relative frequency with which outcome O
would be obtained is r,’ the concept of chance still leaves room for a gap
between the chance of O on any of those trials and the limiting relative
frequency of its occurrence over all of them. Pace Mellor (2000: 20), such a
gap remains a metaphysical possibility: there is no inconsistency, for example,
in the idea of a fair coin being fairly tossed infinitely often without ever
landing heads.32
In any case, all frequentist analyses of chance face the problem of the
reference class. Processes fall under not one ‘kind’ but many. The naive
suggestion that, e.g., the chance of heads with a certain coin is the frequency of
heads in an actual reference class invites the question: ‘Which reference class?’.
32
Cf. Giere (1973: 477–8) and Lewis (1986: 90). Following Mellor (1995: ch. 3), call the idea that
events with chance 1 cannot but occur the ‘necessity condition’ on chance, and the idea that,
necessarily, an infinite series of independent trials each having a chance r of outcome O has a
limiting relative frequency r of O, the ‘frequency condition’ on chance. By the Bernoulli theorem,
the sequence of chances, on n independent trials on each of which the chance of O is r, of the
relative frequency of O diverging by a small amount from r, tends to the limit zero as n increases.
To reject the frequency condition on chance, as I have done, one must therefore take one of
three (non-exclusive) options: (1) reject the necessity condition on chance; (2) deny that chance
is defined over an infinite sequence of independent trials; or (3) opt for a non-standard probability
calculus in which probabilities can take infinitesimal values. In the hope of retaining the necessity
condition on chance while keeping the domain over which chance is defined broad, Lewis (1986:
61, 89–90, 333) embraces (3).
If the coin is a 1998 penny tossed left handed to a maximum height of three
metres on to a stone floor, should the reference class comprise tosses with
these features? If so, how fine-grained should the characterization of the toss
be? Resorting to hypothetical relative frequencies does not avoid the problem:
what would count as repeated tokening of a process is as dependent on the
selection of one of the many kinds under which the process falls as what does
count as a tokening of it.
Attempts at frequentist analyses of chance fail. But perhaps they fail merely
because they are molecular.33 Let us consider whether a holistic analysis fares
any better.
2.2 Lewis’s ‘Best System’ Analysis of Chance

Lewis’s (1994) fundamental idea is that a logical link between chance and
non-chance arises provided (local) matters of particular fact are taken in their
entirety. On his analysis, the chance of an event at a world is determined by the
entire course of history at that world. The chance (beforehand), e.g., of getting
heads with this coin on that toss is determined not directly, by the local patterns
of heads and tails on tosses of this coin, or on tosses of those kinds of coins
of that type, or, even, on all tosses of all coins, but indirectly, by the pattern,
literally, of all events. The problems facing attempts to analyse the chance of
an event in terms of relative frequencies of that kind of event are removed
at one stroke. There is no danger of there being many equally deserving
reference classes, or of the preferred reference class being empty: so to speak,
with respect to every chance event the ‘reference class’ comprises all events.
The analysis is therefore holistic, and the supervenience thesis it entails global.
Thus characterized, however, Lewis’s ‘analysis’ is merely structural, or
programmatic. The totality of events at a possible world is not literally a
reference class. It is all very well to deem molecularism to be at fault. But the
devil is in the detail. How is a more holistic analysis of chance going to work?
Lewis’s answer to this question builds upon his ‘best system’ analysis of non-
probabilistic laws of nature under the supposition of determinism. Regularity
33
The supervenience theses entailed by analyses in terms of actual relative frequencies are
clearly local. But so too are those entailed by an analyses that rely on counterfactuals about what
the outcomes of indeterministic processes would have been had they been instigated. Because
such counterfactuals don’t supervene on actuality (as a condition of the analysis being successful),
they constitute a small part of non-chance.
theorists take laws of nature to be regularities. Since some regularities are
clearly not laws—such as the fact (assuming it to be one) that all lumps of solid
gold have a mass of less than a billion kilos—those that are laws must be hived
off. Lewis (1973: 73–5; 1986: 121–4) effects the separation as follows. He holds
that, for all worlds w, the laws at w are whatever regularities are expressed
by theorems or axioms of whatever deductive system S best systematizes w’s
entire course of history in the following sense: S does best with respect to (an
appropriate weighting of) the competing criteria of strength and simplicity in
a competition between deductive systems that are true at w and which are
concerned only with ‘natural’ properties.34 Some regularities will not make it
into the best system, and on the analysis these are distinguished from laws and
classified as merely accidental.
If chances are taken to be primitive natural properties, extending this analysis
to the indeterministic case, and hence to chance laws, is almost trivial.35 If there
are chance laws, and chance is a natural property, the deductive systems that
do best in competition with one another will claim, truly, that there are certain
regularities involving chance (just as, under the supposition of determinism,
they make true claims concerning non-chancy matters involving mass and
charge). Lewis suggests but one emendation. Regularities that have merely
chanced to come about should not be accounted laws: they too are merely acci-
dental. So deductive systems that entail regularities of this ilk must be excluded
from the competition. In Lewis’s (1986: 126) own words, the result is as follows:
Previously, we held a competition between all true systems. Instead, let us admit to
the competition only those systems that are true not by chance; that is, those that not
only are true, but also have never had any chance of being false. The field of eligible
competitors is thus cut down. But then the competition works as before. The best
system is the one that achieves as much simplicity as is possible without excessive loss
of strength, and as much strength as is possible without excessive loss of simplicity. A
law is a regularity that is included, as an axiom or as a theorem, in the best system.
This proposal is an analysis of chance law, not an analysis of chance itself: it

adverts to competing deductive systems the axioms and theorems of which
34
This account is unfaithful in one respect, since Lewis tends to identify laws not with
regularities expressed by axioms or theorems of the best system, but with those axioms or
theorems themselves. There is a familiar issue here, but it is not one which is relevant to the main
concerns of this paper.
35
Lewis holds that determinism is incompatible with the existence of (non-trivial) chance.
(See n. 4, above.)
embody single-case probability functions that are interpreted uncritically as
chance functions. In the modification by means of which Lewis aspires to ana-
lyse chance, such functions are left uninterpreted. So the competition is held
between deductive systems that are only partially interpreted, and the notion of
the best such system invoked with a view to defining not just chance ‘laws’, but
the uninterpreted probability functions too, and, hence, ‘chance’ itself. That
is, what the chances of particular events are (at a world), and hence what regu-
larities involving chances there are (at that world), and hence what the chance
laws are (at that world), are matters to be defined en masse in terms of which of
the various deductive systems that articulate regularities involving as yet unin-
terpreted single-case probabilities does best (at that world) in the competition.
Two conditions must be fulfilled for this strategy to have any hope of
success. Firstly, since chance is temporally relative—the chance of an event
happening during some temporal interval changing over time depending (at
least in part) on which events chance to happen beforehand—for all worlds
w, the deductive system that does best with respect to w must determine, in
combination with (non-chance at) w, for each time t, a probability function
Pt defined over all the events that, at w, are chancy at t.36 Secondly, the
criteria of ‘strength’ and ‘simplicity’, that sufficed to pick out the best system
in the analysis of law alone, must be supplemented. When the project was
merely to define ‘law’ (at a possible world), the best system analysis needed
to look at the world in question to ascertain which deductive systems are
true at that world. It could then select from among the deductive systems
that are true the one that does best with respect to the further criteria of
simplicity and strength. But once the more ambitious project is pursued of
defining ‘chance’ too, the requirement of truth cannot be employed to select
the candidate systems on which the criteria of simplicity and strength are to
operate: the probability ‘claims’ the systems embody are as yet uninterpreted,
and so cannot be true or false. Without some further criterion to do duty for
the previous requirement of truth, maximal strength and simplicity would
be all too easily achievable independently of the world in question, and the
laws of chance would turn out to be necessary (and absurd). To avoid this
consequence, Lewis supplements the criteria of strength and simplicity with
36
In his earlier treatments, Lewis (1980: 91; 1986: 131–2) explicitly avoids the assumption
that for all times, chance at a time is defined over all propositions. He subsequently makes
the assumption implicitly (Lewis 1994: 483). As Hoefer (1997) observes, much turns on which
propositions chance at a time is defined over.
an additional criterion of ‘fit’ between a deductive system and a world. In
effect, this criterion does for the best system analysis of ‘chance’ (and ‘law’)
what the requirement of truth does for the best system analysis of ‘law’ alone.
It is not enough for one of the competing deductive systems to be maximally
simple and strong. The uninterpreted ‘claims’ it makes must also fit the world
in question. More accurately, in Lewis’s (1994: 480) own words:
The virtues of simplicity, strength, and fit trade off. The best system is the system that
gets the best balance of all three. As before, the laws are those regularities that are
theorems of the best system. But now some of the laws are probabilistic. So now we
can analyse chance: the chances are what the probabilistic laws of the best system say
they are.37
Although this analysis has attracted much critical attention, most of it

concerns the ‘undermining’ difficulties Lewis (1980: 111–12) once deemed
fatal.38 None of it focuses on the difficulty I shall raise.39 Because this difficulty is
fundamental, it besets modifications and restrictions of the analysis that have
been proposed as resolutions of the undermining difficulties.40 I call it ‘the
problem of fit’.
2.3 The Problem of Fit

Lewis’s elucidatory remarks concerning ‘fit’ are disconcertingly sketchy. In the
paper in which he eventually defends the best system analysis, he merely says:
[S]ome [systems] will fit the actual course of history better than others. That is, the
chance of that course of history will be higher according to some systems than
according to others. (Though it may well turn out that no otherwise satisfactory
system makes the chance of the actual course of history very high; for this chance will
come out as a product of chances for astronomically many chance events.)41
37
This statement is misleading. It invites a reading on which ‘the chances’ means ‘the
(temporal) chances of particular events’. But of course the laws of chance no more state the
chance (at some earlier time) of, e.g., the Challenger space shuttle exploding than do deterministic
laws state when, e.g., a solar eclipse will occur. In either case, such matters of particular fact
depend not just on laws, but on earlier matters of particular fact.
38
See n. 6, above.
39
This statement was true until the final drafting. I only came to know of Elga’s (2004)
identification and treatment of the difficulty at the last minute.
40
See the proposals in Hoefer (1997) and Halpin (1998).
41
Lewis (1994: 480). This statement is also misleading. It appears to define the measure of a
deductive system’s ‘fit’ as what that system takes to be ‘the chance of [the actual] course of history’,
There are both conceptual and formal difficulties with this gloss. The suppos-
ition that candidate deductive systems will specify their own measures of fit
by specifying the chance of entire courses of history is in tension with the fact
that chance is temporal. The chances of an event of a certain kind occurring
in a temporal interval <t, t + n > vary beforehand according, at least in part,
to what chances to happen earlier. For example, although the chance of one
hundred successive heads with a fair coin would have been very tiny at the
outset, it would have reached one-half just before the hundredth toss had a
run of 99 heads occurred in the meantime. Similarly, if, as Lewis supposes, the
chances at t are defined over entire futures Ft , chance at t will be defined over
entire courses of history Ht Ft too. (‘Ht ’ is an entire history up to an including t;
‘Ft ’ is an entire future after t. Since there is no chance of the past not occurring,
at t the chance of Ht is 1.) Somehow, then, the best deductive systems must
combine with facts about the non-chances at w to give each nomologically
possible course of history Ht Ft different chances at different times—say,
by entailing conditionals Ht → [Pt (Ht Ft ) = r]. Yet nothing in this requires
that such systems specify non-temporal chances for the various Ht Ft , and
there seems much in the concept of chance to prevent them from so doing.
‘Propensity’ theories of the ontology of chance, whereby chance is a measure
of the ‘tendency’ of a cause to produce an effect, or the display of a dispositional
property,42 should deem the notion of non-temporal chance confused.
In presupposing that for all worlds w at which chance exists, the deductive
system that does best at w does best, in effect, as a result of specifying the
non-temporal chance (probability) at w of w’s course of history, Lewis seems
merely to assume, counterintuitively and without warrant, that, at w, there
is some such chance to be specified. Still, the conceptual difficulty of this
assumption does not refute the spirit of his approach. Perhaps he might
concede that chance is essentially temporal by merely declining to interpret
whereas the best systems analysis has been seen to define chance in terms of the fit of deductive
systems employing an uninterpreted probability function. However, the threat of circularity
is mere appearance. It is dissipated by the fact that both interpreted and partially interpreted
deductive systems can be said to fit a world. If a deductive system D employs an interpreted
probability function P that accords a chance r to world w’s course of history, this is because the
truth-condition of P’s claims is specifiable in terms of which among the deductive systems that
employ an uninterpreted probability function best fits w. These systems will include a deductive
system that is just like D save for the fact that it employs an uninterpreted probability function
formally similar to D’s chance function P.
42
Respectively, as in Giere (1973) and Mellor (1971).
the non-temporal probability of a world’s course of history according to the
deductive system which does best at that world as the world’s ‘chance’.43 So
doing would invite the objection that if there are no non-temporal chances of
courses of history at worlds, the competition between deductive systems ought
to include systems that, like theories consisting in a set of history-to-chance
conditionals of the form Ht → [Pt (A) = r], say nothing about non-temporal
probabilities, the upshot being that some measure of a system’s fit to a world
other than a non-temporal probability it ascribes the world’s course of history
must be found. But perhaps this objection would beg the question by refusing
to take talk of ‘system’ in Lewis’s analysis seriously: perhaps the measure of
fit need not apply to systems that don’t ascribe non-temporal probabilities to
worlds because such systems would do so badly by the criterion of simplicity
that they can be excluded from serious consideration. Indeed, the best system
analysis might try to turn the tables at this juncture by using this consideration
against the intuition that chance is inherently temporal: one’s pre-theoretical
intuitions notwithstanding, if the laws of chance are determined by which
system does best, and any aspiring system that eschews a non-temporal
probability function defined over entire courses of history will be hopeless
with respect to the criterion of simplicity, chance must be non-temporal in
part after all.44
Be this as it may, let us turn to the formal difficulty in Lewis’s gloss on fit, for
it is this that I wish to emphasize. While Lewis (1994: 480) says that the chance
of a world according to a competing deductive system will be a ‘product of
chances for astronomically many chance events’, he does not say what the
factors in this product are. For illumination, we must turn to a hint Lewis
(1986: 128) gave earlier:
[A] system fits a world to the extent that the history of that world is a comparatively
probable history according to that system . . . If the histories permitted by a system
formed a tree with finitely many branch points and finitely many alternatives at each
43
Cf. Hoefer (1997). Hoefer argues that if chance supervenes on non-chance in a Humean way,
there are independent reasons for denying that chance is defined over propositions so ‘big’ as to
be undermining. As he in effect observes (e.g. 1997: 328), supposing chance restricted to ‘small’
propositions is compatible with holding it to be expressed by a restriction of a probability function
that is defined over ‘big’ propositions too.
44
There is another alternative. Perhaps Lewis assumes that for every set S comprising, for all t,
and A, a conditional Ht → [Pt (A) = r], S will entail a proposition that specifies a non-temporal
probability for each course of history.
point, and the system specified chances for each alternative at each branch point, then
the fit between the system and a branch would be the product of these chances along
that branch; and likewise, somehow, for the general infinite case.
Lewis (1986) says nothing more about the notion of fit,45 and his use of the
expression ‘somehow’ is well justified. There is no tenable infinite analogue of
this ‘product’ method for determining the probability of a world according to
a system (and hence the system’s measure of fit). Moreover, serious problems
arise for attempts to construct an alternative.
To see this, let us first elaborate the finite case Lewis considers. Each deductive
system in the competition holds finitely many courses of history nomologically
possible. These courses of history can be represented by the branches of a
finite tree. The nodes of the tree are ordered into levels doing duty for times.
At each level t, each node ti at that level represents a nomologically possible
momentary state of the world at time t. The branchings from ti represent the
possible futures that the system says have some probability of coming about at
t if (a) the momentary state of the world at t is as the node ti represents it as
being, and (b) the momentary states of the world prior to t are as the nodes on
the branch beneath ti represent them as being. Importantly, time is discrete:
each node has immediate successor nodes that represent possible states that
have some probability of coming about next. Each node ti is associated with
a probability function Pti , that specifies, for each of ti ’s immediate successor
nodes ti+1,j , a probability at ti of ti+1,j . (So the values for the various j’s of
Pti (ti+1,j ) sum to 1.) The probability (chance) of a branch of the tree is the
product of the probabilities (chances), at each of the nodes on the branch, of
the immediate successor nodes.
Of course, this finite case is unrealistic: chance does not determine the
topology of time, and it is more realistic to suppose that time is dense
or continuous. On this supposition, however, no time has an immediate
successor, and a strict analogue of the finite case is impossible: there can be no
probabilities at the nodes on a branch of the next node on the branch, and so
45
It is unsurprising that in this earlier discussion Lewis (1986: 128) treats the notion of
fit offhandedly, since there he rejects the best system analysis of chance on account of the
undermining difficulties. It is perfectly legitimate for him to bring his discussion to an end by
remarking ‘never mind the details, if, as I think, the [best system analysis] won’t work anyway’. In
contrast, when Lewis (1994) eventually takes the undermining difficulties to have been resolved,
he ought perhaps to have reflected further on the trickiness of defining fit before embracing the
analysis.
no product of them. Let us consider, therefore, the prospects for an inexact
analogue. Perhaps a node Bt on a branch B might contribute some other
probability value to a product that is B’s overall probability. In particular, for
all branches B and nodes Bt , if a probability B Pt is defined at Bt over each higher
branch-segment Bt St+k that is directly above Bt , the contribution Bt makes to
a product equal to the probability of the branch B might be the limit of the
sequence of probabilities B Pt (Bt St+k ) as k tends to zero.46
Unfortunately, however, thisproposal solvesone problem only toencounter
another: although we now have probabilities for each of the nodes on a branch
that can be multiplied together so as to get the probability of the whole branch,
we have too many. Standardly, the product of these values would be useless as
a measure of the probability of the branch even in the case where time is dense:
for a denumerable product of values in the interval [0,1) is zero. Resorting
to non-standard methods on which the value of such a product can be
infinitesimal might sidestep this problem.47 Yet so doing would be futile if time
is not just dense but continuous. Standardly, continuous time is constituted by
continuum many times. But if selecting at each node on a branch some unique
probability value generates continuum many values, the product method of
obtaining a value for the whole branch remains untenable: non-denumerable
products are defined neither standardly nor non-standardly. The only way
left by which to try to uphold such a method would be to embrace the ideas of
non-standard analysis exclusively. From within the conceptual scheme these
ideas provide, entire courses of history taking place over continuous time
will be represented by branches constituted not by continuum many, but by
hyperfinitely many nodes.48 Continuous time would then no longer be a bar
to the method, since products of hyperfinitely many reals or infinitesimals
are defined in non-standard analysis. This manoeuvre is desperate, however. If
the cardinality of continuous time is only contingently hyperfinite, adopting
the non-standard conceptual scheme exclusively will leave the competing
deductive systems ‘blind’ to some temporally continuous worlds at which
chance exists (if it exists at any worlds). Biting the bullet and maintaining that
chance does not exist at worlds at which there are continuum many times
46
A branch segment Bt St+k directly above Bt is the segment between Bt and a higher node
Bt+k . It includes Bt+k , but not Bt itself.
47
Lewis advocates a non-standard calculus on which chances can take infinitesimal values on
independent grounds. (See n. 32, above.)
48
In non-standard analysis, hyperfinite numbers succeed the positive integers.
would be too hard a course. Maintaining that the cardinality of continuous
time is non-contingently hyperfinite would be harder still. I conclude that
Lewis is wrong to suggest that a measure of fit in realistic cases can be modelled
on a product method that is appropriate to the finite case.
While Lewis’s own suggestion as to how to elucidate a measure of fit
between competing systems and worlds does not work, the question arises
as to whether some alternative will. In particular, the question arises as to
whether the classical treatment of stochastic processes might serve the best
system analysis’s needs. I do not think it can. On this treatment, the various
deductive systems in the competition are taken to specify an atemporal
probability function P, defined over entire courses of history, that is basic.
Temporal probabilities Pt () are then defined as conditionalizations P(-|Ht ) of
this function on the course of history up to and including t. Various conceptual
and formal difficulties beset this proposal, however, even if recourse is made
to non-standard probabilities. Each version of it turns out to be at least as
unattractive as the version that is committed to the following: (i) Lewis’s
favoured necessity condition on chance is false—nomologically possible
events can have zero chance—even though (ii) it is apriori that there is no
possible world at which an event occurs that had zero chance of occurring;
moreover (iii) it is apriori that chance at a time cannot depend on matters of
particular fact at later times and, hence, on some views (including Lewis’s)
(iv) it is apriori that backwards causation cannot occur.49
2.4 ‘Fit’ as a Primitive?

At this point, a proponent of the best system analysis might think to take the
degree to which a deductive system fits a possible world to be a primitive notion.
To my mind, however, this option is untenable. To adopt an analogy in the
philosophy of mind, it should come as no surprise that there is no published
49
I say more about the difficulties that beset attempts to glean a measure of fit from the classical
treatment of stochastic processes in the Appendix. Elga (2004) agrees that allowing the competing
deductive systems to employ non-standard probability functions won’t solve the problem of fit.
His grounds differ from mine, however. They are that with respect to simple possible ‘worlds’,
amongst the ‘constructable’ non-standard probability functions, those that comply with the
necessity condition on chance, and that in a certain sense approximate standard probability
functions, give counterintuitive results: there are worlds such that a non-standard probability
function that approximates the standard probability function that, intuitively, fits the world best,
has worse fit than does a non-standard probability function that approximates some standard
probability function that, intuitively, fits the world badly.
attempt to solve the difficulties confronting classical analytic behaviourism
by resorting to a holistic doctrine according to which, for each mental state
M, a subject S is in M iff S’s being in M best fits S’s behaviour, where the
notion of a mental state’s ‘fitting’ a pattern of behaviour is left primitive.
Of course, the term ‘fit’ does invoke an epistemological notion on which
we have an independent grip: there is a well-established practice of basing
judgements of chance on, e.g., statistical evidence. But it would be quite wrong
to protest that best system analyses merely adapt to logical and semantical
ends an epistemological notion to which Lewis’s opponents—in particular,
realists about chance who deny chance supervenes on non-chance—are
committed. Denying that chance supervenes on non-chance does not commit
one to a notion of fit having the formal properties best system analyses of
chance require. That realism about non-supervenient chance must offer an
epistemology detailing ways in which evidence can support hypotheses about
chances falls short of saying that it presupposes facts about how a course of
history in its entirety bears evidentially on ‘complete’ chance hypotheses, i.e.
regarding all matters that are chancy.
3. Conclusion
My conclusion is no less modest than the aim with which I began. I have
given grounds for scepticism regarding the strength of reasons one might give
for realism about chance. Chance has no explanatory role with respect to
non-chance that would ground realism about it via an inference to the best
explanation, and while an analysis of chance in terms of non-chance might
excuse the realist about chance of a need to make any such inference, extent
analyses of chance fail.
I would like to end, however, by speculating that there is little prospect that,
at some time in the future, a correct analysis of chance will yield a warrant for
belief that chance exists. If chance is analysable, Lewis’s best-systems analysis is
surely along the right lines: only global facts about non-chance could possibly
determine an event’s chance. But weakening Lewis’s presumption that the
system that best fits a world is the system that gives the world’s entire course
of history highest probability (chance) won’t yield a warrant for believing
in chance, even if so doing results in an analysis of chance that is correct.
For a fundamental obstacle remains: even if the measure of a system’s fit is
determined by the probability it gives some much smaller part of non-chance,
there will no reason to believe that any system has best fit independent of a prior
reason for believing that chance exists.
Elga’s (2004: 72) proposed account of fit serves to illustrate the point.50 Elga
suggests that one deductive system X fits a world better than another system
Y iff ‘the chances X assigns to the test propositions are predominantly greater
than the corresponding chances that Y assigns’. It is clear that even if this
criterion can be employed in a version of the best system analysis that is
correct,51 no warrant for belief in the existence of chance would result. In
the absence of an independent reason for realism about chance, there is simply
no reason whatsoever to believe that one system assigns chances to the ‘test’
propositions that are predominantly greater than those that are assigned by
all the other systems.
Appendix
In effect, Lewis’s product method constructs the non-temporal probability a
deductive system accords an entire course of history out of certain temporal
probabilities that, given events prior to time t, the system says hold at t.52
That this method cannot be extended from discrete to continuous time
need not undermine his more basic idea that the measure of a system’s fit
to a world is a (non-temporal) probability it accords the world: some other
relationship might hold between this probability and the values the system
50
As I remarked in n. 39, above, I only became aware of Elga’s paper just before going to press.
51
Elga leaves what the ‘test’ propositions are implicit in all but the case of the very simplest of
idealized ‘worlds’. He also leaves the meaning of ‘predominantly greater’ open, and it is difficult
to see how the argument he gives for the claim that in this case the criterion yields the correct
results might be extended to the case of possible worlds proper. Moreover, in objecting to the
suggestion that a system’s fit to a world should be identified with the value for the world’s course
of history as given by a non-standard probability function the system employs, he presupposes the
necessity condition on chance. Yet his own criterion flaunts this condition. These all too hasty
remarks notwithstanding, Elga’s treatment of the problem of fit is a significant advance.
52
Actually, this isn’t quite true even in the case Lewis considers of an indeterministic finite
world with a first moment and discrete time. The product of probabilities along a branch Lewis
calls ‘the’ probability (chance) the system accords the entire branch is really temporal: it is a
probability (chance) at the first moment. Moreover, it is not the probability (chance) of an entire course
of history either. It is the probability (chance) at the first moment of the rest of history, excluding
the ‘initial conditions’ that obtain at that moment and which determine that probability (chance).
gives to temporal probabilities (chances) in the light of the world’s course of
history. In particular, the classical treatment of stochastic processes suggests
that temporal probabilities might be viewed as conditionalizations of a non-
temporal probability function P. A deductive system in the competition for
‘best’ would specify a non-temporal probability function of this kind defined
over each course of history Ht Ft the system deems nomologically possible.
The temporal probabilities Pt () that the system determines at a world w in
combination with w’s course of history (or, more generally, w’s non-chances)
are just the conditional probabilities P(-| Ht,w ). From this perspective, the
complete theory of chance at a world is given by a single probability function
defined over entire courses of history (and hence over sets of them). The
chances at a world at a time are given by conditionalizing this function on that
world’s course of history up to and including that time.
This method also encounters formal and conceptual problems, however.
Chance no more precludes chance processes the outcome spaces of which are
continuous than it precludes continuous time. Looked at standardly, then, the
set of courses of history over which the probability function characterized by a
competing deductive system is defined will have (at least) the cardinality of the
continuum. Standardly, only a denumerable subset of these courses of history
can have a positive value: the rest must have probability zero. So the necessity
condition on chance would be violated if this unconditional probability
expresses non-temporal chance.53 Moreover, even if it doesn’t express this, the
necessity condition on chance will still have to go: if a course of history has
zero unconditional probability, it will have zero probability conditional on any
history Ht , while ex hypothesi the conditional probability P(-| Ht ) is a temporal
probability Pt () that expresses temporal chance. So we must choose between
the standard perspective and the necessity condition on chance.
Suppose then that we try to keep the necessity condition on chance by going
non-standard, and giving each nomologically possible course of history at
least infinitesimal probability. Giving only infinitesimal values to continuum
many courses of history would bring disaster: it would be consistent to raise
these values by a finite factor, and, in effect, no system would fit better
than any other. This result might be averted by adopting non-standard ideas
exclusively, so that the cardinality of the courses of history over which a
53
According to the necessity condition on chance, an event which has zero chance is not
nomologically possible. See n. 32, above.
system’s probability function is defined is hyperfinite. The infinitesimal values
of the function will not then be flexible in the same way, since raising them
by any finite factor will result in their no longer summing to one. Yet
exclusively adopting a non-standard perspective faces the same dilemma that
arose when so doing was proposed as the saviour of Lewis’s product method.
If the cardinality of the set of nomologically possible courses of history is
contingently hyperfinite, competing deductive systems that are wedded to
the non-standard conceptual scheme will be blind to some possible worlds.
But biting the bullet and supposing that chance does not exist at worlds at
which, e.g., continuum many courses of history are nomologically possible
is too hard a course, while it is harder still to suppose that the cardinality of
the set of nomologically possible courses of history at worlds at which chance
exists is non-contingently hyperfinite. So the non-standard perspective is more
trouble than its worth.
Alternatively, then, we should try to avoid the disastrous consequences
of assuming that systems give only infinitesimal probability to each course
of history they deem nomologically possible, not by requiring the set of
these courses of history to have a hyperfinite cardinality, but by allowing
some systems to give finite probabilities to some entire courses of events they
deem nomologically possible. Yet this manoeuvre proves too slight. While the
possibility is now admitted that some system will do best at some world in
virtue of giving that world a finite probability, it will remain impossible for
any system to do best at a world by giving that world a higher infinitesimal
probability than does any other similarly strong and simple system. Yet ruling
out this much builds too much into the concept of chance. If the laws
of chance at a world determine non-temporal chances at that world for
entire courses of history, and some of those chances are infinitesimal, why
shouldn’t there be a world at which the chance of its own course of history
is infinitesimal? Whereas infinitesimals were brought in to accommodate the
intuition that some events have a positive chance that must be smaller than
any positive real, we now find ourselves asked to accept a definition of chance
on which there is an a priori guarantee that events with infinitesimal chance
don’t happen.
It remains to consider the possibility that these problems all stem from
burdening the ‘conditional probability’ method of handling the interplay
between non-temporal and temporal chance with the necessity condition on
chance. The attractions of this condition notwithstanding, giving it up might
be an acceptable price to pay for an otherwise satisfactory analysis. Yet even the
retreat to zero chances would leave some of the problems already discussed
intact. In particular, the guarantee that no course of history, and hence no
event, with infinitesimal chance ever occurs, will simply go over into the
guarantee that no course of history, and hence no event, will transpire at a
world at which it has zero chance.
Finally, there is a further conceptual difficulty that sacrificing the necessity
condition on chance won’t solve. Adopting this method would rule out
the temporally backwards dependence of chances at a time on events that
chance to happen after that time (and, hence, on some analyses of causation,
backwards causation too). Of course, backwards causation has been held to
be conceptually incoherent. But Lewis himself doesn’t hold it so, and it is
in any case unsatisfactory that it should be precluded by the exigencies of
a solution to a formal problem that is generated by a purported analysis of
chance. There is no way to modify the conditional probability method for
establishing the desired interplay between temporal and non-temporal chance
so as to accommodate backwards dependence of chance on later events,
however. Only a temporal reference in the proposition conditioned on can
make the conditional probability P(−|A) a temporal probability Pt (). But if
there is a world at which a temporal probability Pt () is backwards dependent,
conditionally, not just on w’s history up until t but on a later event A, so that
the temporal reference in A is some t* later than t, there is no way back from
t* so as to allow the conditional probability P(−| At∗ ) to be equated with the
temporal probability Pt ().
References
Armstrong, D. (1983), What is a Law of Nature? (Cambridge: CUP).
Bigelow, J., Collins, J. and Pargetter, R. (1993), ‘The Big Bad Bug: What are the
Humean’s Chances?’, British Journal of the Philosophy of Science, 44: 443–62.
Black, R. (1998), ‘Chance, Credence, and the Principal Principle’, British Journal of the
Philosophy of Science, 49: 371–85.
Dretske, F. (1977), ‘Laws of Nature’, Philosophy of Science, 44: 248–68.
Elga, A. (2004), ‘Infinitesimal Chances and the Laws of Nature’, Australasian Journal of
Philosophy, 82: 67–76.
Giere, R. (1973), ‘Objective Single Case Probabilities and the Foundations of Statistics’,
in P. Suppes et al. (eds.), Logic, Methodology and the Philosophy of Science IV (Amsterdam:
North Holland), 467–83.
Hall, N. (1994), ‘Correcting the Guide to Objective Chance’, Mind, 103: 505–17.
(2004), ‘Two Mistakes about Credence and Chance’, Australasian Journal of Philosophy,
82: 93–111.
Halpin, J. (1998), ‘Lewis, Thau, and Hall on Chance and the Best-System Account of
Law’, Philosophy of Science, 65: 349–60.
Hitchcock, C. (1999), ‘Contrastive Explanation and the Demons of Determinism’,
British Journal of the Philosophy of Science, 50: 585–612.
Hoefer, C. (1997), ‘On Lewis’s Objective Chance: ‘‘Humean Supervenience
Debugged’’ ’, Mind, 106: 321–34.
Lewis, D. (1973), Counterfactuals (Oxford: Blackwell).
(1980), ‘A Subjectivist’s Guide to Objective Chance’, in R. C. Jeffrey (ed.), Studies
in Inductive Logic and Probability, ii (Berkeley and Los Angeles: University of California
Press). (Page references are to the reprinting as ch. 19 of Lewis 1986.)
(1986), Philosophical Papers, ii (Oxford: OUP).
(1994), ‘Humean Supervenience Debugged’, Mind, 103: 473–90.
McDermott, M. (1999), ‘Counterfactuals and Access Points’, Mind, 108: 291–334.
Mellor, D. H. (1969), ‘Chance’, Proceedings of the Aristotelian Society, suppl. vol. 43: 11–36.
(1971), The Matter of Chance (Cambridge: CUP).
(1976), ‘Probable Explanation’, Australasian Journal of Philosophy, 54: 231–41.
(1982), ‘Chance and Degrees of Belief ’, in Mellor (1991: ch. 14).
(1991), Matters of Metaphysics (Cambridge: CUP).
(1995), The Facts of Causation (London: Routledge).
(2000), ‘Possibility, Chance and Necessity’, Australasian Journal of Philosophy, 78:
16–27.
Noordhof, P. (1999), ‘Probabilistic Causation, Preemption and Counterfactuals’, Mind,
108: 95–125.
Percival, P. (2000), ‘Lewis’s Dilemma of Explanation under Indeterminism Exposed
and Resolved’, Mind, 109: 39–66.
Popper, K. (1959), ‘The Propensity Interpretation of Probability’, British Journal for the
Railton, P. (1978), ‘A Deductive-Nomological Model of Probabilistic Explanation’,
Sapire, D. (1991), ‘General Causation’, Synthese, 86: 321–47.
Strevens, M. (1995), ‘A Closer Look at the ‘‘New’’ Principle’, British Journal of the Philosophy
of Science, 46: 545–61.
Sturgeon, S. (1998), ‘Humean Chance: Five Questions for David Lewis’, Erkenntnis, 49:
321–35.
Thau, M. (1994), ‘Undermining and Admissibility’, Mind, 103: 491–503.
von Mises, R. (1957), Probability, Statistics and Truth (New York: Dover).
Vranas, P. (2002), ‘Who’s Afraid of Undermining?’, Erkenntnis, 57: 321–35.
Woodward, J. (1989), ‘The Causal Mechanical Model of Explanation’, in P. Kitcher and
W. Salmon (eds.), Scientific Explanation (Minneapolis: University of Minnesota Press),
357–83.
Part II
Identity and Individuation
5
Structure and Identity
Stewart Shapiro
The purpose of this paper is to further articulate my preferred version

of mathematical structuralism in light of some criticisms of my Philosophy
of Mathematics: Structure and Ontology (1997) and Michael Resnik’s Mathematics as a
Science of Patterns (1997). Some of these criticisms are shortcomings in the original
presentation, and so I take this opportunity to correct aspects of the view.
The view in question is what I call ante rem structuralism, the thesis that
mathematical structures exist prior to, and independent of, any exemplifica-
tions they may have in the non-mathematical world. The contrast is with an
approach that either adopts an Aristotelian in re view that a given structure exists
only in the systems that exemplify it or the more common eliminative thesis
that structures do not exist at all—talk of structures is to be paraphrased away.
In his classic ‘What Numbers Could Not Be’, Paul Benacerraf (1965) proposed
an eliminative view, and an intriguing variation on that theme is articulated
in Geoffrey Hellman’s Mathematics Without Numbers (1989). I call Hellman’s view
‘modal eliminative structuralism’, since it avoids commitment to the (actual)
existence of models of the various mathematical theories by careful placement
of modal operators. Instead of speaking of systems of objects, Hellman speaks
of possible systems. The official argument in my Structure and Ontology is that when
I am indebted to Jonathan Kastin and Jukka Keränen, as well as Geoffrey Hellman, Michael Resnik,
Crispin Wright, Katherine Hawley, and especially Fraser MacBride for extensive and fruitful
exchanges.
110 / Structure and Identity
fully articulated, the various versions of structuralism roughly share their
virtues and shortcomings, but I thought—and think—that the realist, ante
rem approach is the most perspicuous of the lot.
Before going further, I would like to acknowledge an orientation. As I see
it, the goal of philosophy of mathematics is to interpret mathematics, and
articulate its place in the overall intellectual enterprise. One desideratum is to
have an interpretation that takes as much as possible of what mathematicians
say about their subject as literally true, understood at or near face value. Call
this the faithfulness constraint.
According to ante rem structuralism, natural numbers are places in
structures, and places in structures are bona fide objects. This accords with
faithfulness. Grammatically, numerals seem to function as singular terms, and
according to ante rem structuralism, numerals are singular terms. Eliminative
structuralism, on the other hand, interprets statements about mathematical
objects as implicit generalizations over all systems of a certain sort, or all
possible systems of a certain sort. In other words, on the eliminative program,
numerals are understood to be bound variables, perhaps under a modal
operator. This does not accord with faithfulness.
A second, and weaker, desideratum is to develop an interpretation that
does not go too much beyond what mathematicians say about their subject.
Surely the philosopher is going to say some things that the mathematician
does not say. Mathematicians, as such, do not usually address philosophical
issues about their subject. For example, they do not say much about what
the natural numbers are, nor how we obtain mathematical knowledge, nor
how mathematics applies to the physical world. Presumably, philosophical
questions about mathematics are not to be answered solely in mathematical
terms. The second desideratum is to not attribute mathematical properties to
mathematical objects unless those attributions are explicit or at least implicit
in mathematics itself. Call this the minimalism constraint.
Richard Dedekind’s philosophical methodology was to define a system of
objects, and then abstract the structure of the system (e.g. Dedekind 1872,
1888). For example, he constructed the system of cuts in rationals, and then
abstracted the real numbers from the cuts. According to Dedekind, the
abstracted items—the real numbers—are not part of the system abstracted
from, but are instead something ‘new’ which the mind freely ‘creates’
(see Shapiro 1997: ch. 5, §4). Dedekind’s friend Heinrich Weber suggested
instead that real numbers be identified with cuts. Dedekind replied that there are
Stewart Shapiro / 111
many properties that cuts have which would sound very odd if applied to the
corresponding real numbers (Dedekind 1932: iii. 489–90). For example, cuts
have members. Do real numbers have members? Dedekind’s Benacerraf-type
point is consonant with the minimalism constraint. Weber’s proposal assigns
mathematical properties to real numbers that the mathematician does not. In
the preface to his later treatise on the natural numbers, Dedekind (1888) agreed
that Cantor and Weierstrass both gave ‘perfectly rigorous’, and so presumably
acceptable, accounts of the real numbers, even though they were different
from his own and from each other. But Dedekind wrote that his approach is
‘somewhat simpler, I might say quieter’ than the others. I presume he meant
that his own account of the real numbers has less excess baggage than the
others. Again, the minimalism constraint.
Neither the faithfulness constraint nor the minimalism constraint are
absolute. They are defeasible criteria, among others. Hellman, for example,
competently argues that his modal reinterpretation of mathematics avoids
some sticky philosophical problems that literalists like Resnik and myself must
face (see also Hellman 2001). This, he argues, justifies a non-literal reading of
mathematical language.
In a relatively recent note, W. V. O. Quine (1992) advocated a breathtaking
extension of structuralism beyond mathematics to the entire web of belief.
But he stopped short of adopting a realist view of structures (either ante rem
or in re), even for mathematics:
My global structuralism should not . . . be seen as a structuralist ontology. To see it thus
would be to rise above naturalism and revert to the sin of transcendental metaphysics.
My tentative ontology continues to consist of quarks and their compounds, also
classes of such things, classes of such classes, and so on . . .
Quine’s argument here seems to be that since scientists (and mathematicians)

do not themselves say that ante rem structures exist, and they do say
that classes, quarks, and natural numbers exist, then we should accept
classes, quarks, and numbers, but not ante rem structures. I submit that
here Quine takes the minimalism constraint too far, or is misled by it.
The proposal in my book, and presumably in Resnik’s too, is to interpret
the mathematician’s own numbers as places in a structure. This does not
strike me as pernicious metaphysics. The faithfulness constraint is to avoid
contradicting standard mathematics and the minimalism constraint is to
avoid attributing more mathematical properties to mathematical objects than
standard mathematics does. I submit that ante rem structuralism does neither
of those.
I have discussed, and defended, the faithfulness constraint in Shapiro (1994)
and Chapter 1 of Shapiro (1997). I do not have any arguments in favor of the
minimalism constraint, nor do I know how to (or even care to) articulate
it further, to determine its limits for example. The minimalism constraint is
more a matter of taste, in the sense that, other things equal, I would find a
‘quieter’ philosophy more pleasing. Of course, my personal preferences carry
no weight in a sober philosophical discussion. Philosophy must go beyond
statements of personal taste. I mention the criteria here to provide some
perspective, and to note that while I am still attracted to the minimalism
constraint, I concede that it led me astray (see below).
The first issue to be dealt with here concerns the very characterization of ante
rem structuralism. Its slogans either seem to be incoherent or point to outright
falsehoods. In one way or another, the other issues presented here concern the
identity relation on mathematical objects, i.e. the identity relation on places
in structures. One issue concerns whether places from different structures can
be, or should be, identified. Is the natural number 2 the same or different from
the second von Neumann ordinal? Is the natural number 2 identical to the real
number 2? Are there determinate facts answering these questions? Another
issue concerns the identity relation within the same structure. What are we
to make of structures that have different places with the same structural, or
relational properties? This comes up whenever a structure has a non-trivial
automorphism. One reading of the slogans of ante rem structuralism suggests
that I am committed to identifying such places as the complex numbers i and
−i since they have the same relational properties—despite the easy theorem
that i and −i are distinct. Indeed, every point of Euclidean space has the
same relations to the rest of space as every other point. Space is completely
homogeneous. Do I have to say—absurdly—that there is only one point?
Much of this paper concerns some criticisms of ante rem structuralism that
appear in various notes and reviews, and I have encountered versions of them
in conversation and correspondence with friends, colleagues, supporters, and
opponents. The most articulate versions of the arguments are in a pair of
papers, one by Jonathan Kastin, and the other by Jukka Keränen (presented
at the Pacific Division of the American Philosophical Association, 1998, and
his subsequent (2001) ). I will reformulate the criticisms myself, and take
responsibility if I do not come up with the most compelling versions of them.
1. What Is (Ante Rem) Structuralism?

The main thesis of ante rem structuralism is that a natural number, say, is
a place in the natural number structure, and the natural number structure
exists independent of any concrete system that may exemplify it. So there
are two ontological categories: structures and places. Shapiro (1997) focused
on the structures, and perhaps did not say enough, and did not say the right
things, about places. What is a natural number? Or, in other words, what is a
place in the natural number structure?
The problems turn on the slogans associated with this version of structur-
alism. Benacerraf (1965) himself initiated some of the rhetoric:
[I]n giving the properties . . . of numbers you merely characterize an abstract structure . . .
[T]he ‘elements’ of the structure have no properties other than those relating them
to other ‘elements’ of the same structure . . . To be the number 3 is no more and no
less than to be preceded by 2, 1, and possibly 0, to be followed by 4, 5, and so forth . . .
What is peculiar to 3 is that it defines that role . . . by representing the relation that
any third member of a progression bears to the rest of the progression. (Benacerraf
1965: 291)
Famously, Benacerraf argues from this that numbers are not objects. If
numbers are sets, then they must be particular sets, and so natural numbers
would have membership relations with other sets. Thus numbers will have
set-theoretic properties which go beyond the relations of natural numbers to
other natural numbers, contradicting the slogan. Benacerraf ’s point here is
consonant with the above minimalism constraint. It invites the simple retort
that natural numbers are not sets, but the argument is general. The underlying
idea seems to be that any bona fide object has some properties that do not
concern its relations to natural numbers. So natural numbers are not bona
fide objects.
Russell (1903: §242) makes a related point against Dedekind’s abstraction
(see Hellman 2001: 194 n):
. . . it is impossible that the ordinals should be, as Dedekind suggests, nothing but the
terms of such relations as constitute a progression. If they are to be anything at all, they
must be intrinsically something; they must differ from other entities as points from
instants, or colours from sounds. What Dedekind intended to indicate was probably a
definition by means of the principle of abstraction . . . But a definition so made always
indicates some class of entities having . . . a genuine nature of their own . . .
One burden of my book was to dispute this: ‘the issue raised by Benacerraf . . .
concerns the extent to which a place in a structure is an object’ (1997: 3). This is
the matter before us now.
Dummett (1991: ch. 23) dubs views like mine ‘mystical’ structuralism.
According to that view, according to Dummett, ‘mathematics relates to abstract
structures, distinguished by the fact that their elements have no non-structural
properties’. Thus, for example, the zero place of the natural number structure
‘has no other properties than those which follow from its being the zero’ of
that structure. Resnik’s early (1981) has similar passages:
In mathematics, I claim, we do not have objects with an ‘internal’ composition arranged
in structures, we have only structures. The objects of mathematics, that is, the entities
which our mathematical constants and quantifiers denote, are structureless points or
positions in structures. As positions in structures, they have no identity or features
outside a structure.
Kastin’s ‘Shapiro’s mathematical structuralism’ identifies passages in my

book where I endorse the Benacerraf–Dummett–Resnik slogans about math-
ematical objects. In the introduction (1997: 5–6), I wrote that there ‘is no more
to being the natural number 2 than being the successor of the successor of 0,
the predecessor of 3, the first prime, and so on’. Later in the book, I am only a
little more careful:
The essence of a natural number is its relations to other natural numbers . . . The
number 2, for example, is no more and no less than the second position in the natural
number structure; 6 is the sixth position . . . The essence of 2 is to be the successor of
the successor of 0, the first prime, and so on. (1997: 72)
The extra care here, if one can call it that, is the mention of essence. Kastin
follows Benacerraf in pointing out that slogans like this seem to conflict with
the thesis that numbers exist as objects.
The slogans are incoherent, or misleading at best (see also Hellman 1999:
924). Start with Dummett’s ‘mystical’ version that mathematical objects ‘have
no non-structural properties’. This is obviously false, on any realist reading.
The number 3 has the property of being the number of my children (pictures
on request), and 17 has the property of being Seymore’s favorite number.
Clearly, ‘being the number of Stewart Shapiro’s children’ and ‘being Seymore’s
favorite number’ are non-structural properties, no matter how liberally we
manage to construe the notion of a structural property.
These examples undermine the slogan by invoking contingent properties.
One response would be to reformulate it to disallow these counterexamples:
individual natural numbers have no non-structural necessary properties; the
number 0 has no other necessary properties than those which follow from
its being the zero of the natural number structure. In other words, for any
property P and natural number n, if Pn then P is a property concerning how
the natural numbers relate to each other.
This does not work either. Traditional Judaism has it that, necessarily, the
number 1 has the property of being the number of deities. Some forms of
dualism have it that, necessarily, 2 is the number of parts of each human.
Being the number of deities and being the number of parts of a human are not
structural properties. Of course, I am not saying that traditional Judaism and
this form of dualism are correct, but we don’t want ante rem structuralism to
rule them out.
Next retort: individual natural numbers have no non-structural mathematical
properties; the number 0 has no other mathematical properties than those
which follow from its being the zero of the natural number structure. For
any mathematical property P and natural number n, if Pn then P is a property
concerning how the natural numbers relate to each other.
To pursue this suggestion, we would have to say what makes a property
mathematical (as opposed to physical, theological, metaphysical). We can
neatly beg the question by defining a mathematical property of a number n to
be a property that concerns how n relates to other numbers. But this fiat badly
fails to capture the intuitive sense of ‘mathematical’. The number 0 has the
property of being the number of real square roots of −1, the number 4 has the
property of being the maximum number of colors needed for a map. These
properties are mathematical, par excellence, and yet both go beyond the relations
of the natural numbers to each other, and neither of the facts follows from the
characterization of the natural number structure alone. Other mathematical
objects and structures are involved.
As noted above, the ‘official’ view in my book is that individual natural
numbers have no non-structural essential properties. Shamefully, I did not say
anything about what an essential property is. Let us see if we can do any better
here. Simon Blackburn’s Oxford Dictionary of Philosophy (1994) defines essence to be
the ‘basic or primary element in the being of a thing; the thing’s nature, or
that without which it could not be what it is’. Blackburn adds that a ‘thing
cannot lose its essence without ceasing to exist’.
In these terms, the supposedly careful version of my stated thesis is that the
basic or primary element in the being of the natural number 2—the nature
of the natural number 2—consists in its relations to other natural numbers.
The number 2 could not be what it is without its relations to other natural
numbers. This sounds all right, as far as it goes, but I admit it does not go
very far. The last clause in Blackburn’s definition is particularly useless here,
since presumably there is no sense to a number losing its relations to other
numbers, and thereby ceasing to exist. It seems that the notion of ‘essential
property’ has a more natural home with contingent objects.
Let us suppose that this is resolved, and that we know what we are talking
about with the essence/accident distinction concerning mathematical objects.
Kastin and Hellman point out that, even so, numbers seem to have some non-
structural essential properties. For example, the number 2 has the property
of being an abstract object, the property of being non-spatio-temporal, and
the property of not entering into causal relations with physical objects. At
least intuitively, these seem essential to the natural numbers, if the natural
numbers exist. To paraphrase the Blackburn definition, being abstract, and
the like, seem to be among the basic, or primary elements in the being of the
natural numbers. If they were not abstract, they would not exist. Abstractness
is certainly not an accidental property of a number—or is it? We will come back
to this shortly.
To continue the sequence of Lakatos-style monster-barring, perhaps we can
try a pair of theses: (i) a given natural number, like 2, is uniquely characterized by
its relations to other natural numbers, and (ii) other essential properties of
the number 2 flow from, or are consequences of, this characterization. When
we say that there is no more to being the number 2 than being the successor
of 1, the predecessor of 3, the first prime, etc., we thereby also say that 2 is
abstract, causally inert, etc. Anything which is no more than the successor of 1,
the predecessor of 3, etc. is abstract, causally inert, etc.
I do not know how to make this way out more illuminating and less
problematic. Starting with (ii), it is not at all clear what notion of ‘consequence’
is in play. It is not a logical consequence of ‘2 is the successor of 1’, ‘2 is the
predecessor of 3’, and the like, that 2 is abstract. Indeed, the abstractness of 2
is not a logical consequence of the sum total of the mathematical properties
that 2 enjoys. Abstractness is not itself a mathematical property at all. Perhaps
we can say that the abstractness of 2 follows from its characterization as a
place in the natural number structure plus some conceptual and metaphysical
truths about structures, but these metaphysical principles would have to be
articulated and defended somehow.
One might say instead that the essential properties of 2 supervene on the fact
that 2 is a particular place in the natural number structure, but this requires
a developed notion of supervenience. Supervenience is itself a modal notion,
and it is not clear how to wield it in a non-trivial way when dealing with
objects that supposedly exist of necessity (if they exist at all). I suppose we can
say that in any possible world in which the natural number structure exists,
its second place is abstract, causally inert, etc. But how does this help?
Clause (i) of the retrenched thesis is that a given natural number is uniquely
characterized by its relations to other natural numbers. The problem is that this
does not manage to distinguish structuralism from any other view that takes
numbers to be objects. Surely any realist and, for that matter, any intuitionist
will agree that 2 is uniquely characterized by being the predecessor of 3, without
even mentioning any of its other structural properties, like being the successor
of 1 or being the first odd prime. According to any view that accepts the
existence of numbers, 2 is the predecessor of 3, and nothing else is. Moreover,
on any reasonable characterization of supervenience, the Platonist will hold
that all of the necessary properties of 2 supervene on this characterization.
In the end, I think that ante rem structuralism is a version of traditional
Platonism, or what I call realism in ontology, just because it takes numbers
to exist as bona fide objects. The problem is that we have now retrenched the
slogan to the point where it does not say anything distinctively structuralist.
Putting the monster-barring aside, there are at least two further ways out of
thisdilemma. The firstrunsagainstacceptedwisdom concerning mathematical
objects. The idea is to hold that abstractness, non-spatio-temporality, and the
like, are indeed accidental properties of (some) mathematical structures and
objects. Hear me out. Recall that on my view, structures are like traditional
universals, in that they can have instances—systems of objects exemplify
structures. Suppose that one takes an Aristotelian in re view of structures
which are exemplified by systems of concrete objects, and a more Platonic view
of structures which have no concrete exemplifications. On such a view, an
exemplified structure is located where the objects exemplifying it are located,
and it inherits the causal efficacy of those objects. Suppose that an announcer
says: ‘the runner was thrown out by the shortstop.’ The natural, intended
reading of this would be that the runner was thrown out by the person playing
the of role of shortstop. This is what I call the ‘places are offices perspective’
(1997: 10). On the present proposal, we can also say that the position threw out
the runner, since the position has the causal efficacy of its occupant that day.
The proposal is a structuralist version of Penelope Maddy’s (1990) account of
sets of physical objects. For Maddy, sets of physical objects (and sets of those sets,
etc.) are located where the physical objects are located and enjoy the physical
properties of those objects. So for Maddy the singleton of the person playing
shortstop threw out the runner (and the shortstop also threw out the singleton
of the runner, but fortunately for the offense, only one out was recorded).
For Maddy, pure sets, such as the empty set and ω, are not located anywhere.
Here, physically unexemplified structures are not located anywhere.
The only place I know of where a similar view is broached is in David
Armstrong (1989: 76), which argues against a corresponding ‘mixed’ orientation
toward universals:
Once you have uninstantiated universals you need somewhere special to put them, a
‘Platonic heaven’, as philosophers often say. They are not to be found in the ordinary
world of space and time. And since it seems that any instantiated universal might have
been uninstantiated—for example, there might have been nothing past, present, or
future that had that property—then if uninstantiated universals are in a Platonic
heaven, it will be natural to place all universals in that heaven.
Strictly speaking, abstracta are not ‘located’ anywhere. That is part of what it
is to be abstract. So on the view in question here, an unexemplified structure is
not located anywhere. Once instantiated, the structure is located wherever the
system is located. I have no objection to a metaphorical extension of the notion
of ‘location’ to include uninstantiated universals, and if we want, we can use the
colorful phrase ‘Platonic heaven’ for these ‘locations’. It seems to me, however,
that Armstrong’s argument turns on taking this metaphor too seriously.
Speaking strictly and literally, we have no need to postulate ‘somewhere
special to put’ the unexemplified structures (see also MacBride (1998: esp.
211) ), and so there is no point to ‘locating’ all structures in this special ‘place’.
If we take the language less literally, perhaps Armstrong’s point is that since
at least some universals—the uninstantiated ones—are abstract, then we
might as well declare that all universals are abstract. Other things equal, this
declaration does make for a simpler theory in one respect, but in the case of
ante rem structuralism, perhaps other things are not equal.
The proposed view of structures and the abstract-concrete dichotomy takes
some of the ante out of ante rem structure. Perhaps. The view does have it that
a structure exists independent of any instances it has. It is still ‘before the
things’, to translate the Latin. But once a structure is exemplified, it takes on
the physical and causal properties of its instantiations.
For what it is worth, the proposal goes well with the vague minimalism
constraint broached above. Speaking strictly, the thesis that mathematical
structures are abstract, acausal, etc. is not part of mathematics. In a sense,
mathematics leaves this bit of metaphysics open. To be sure, mathematicians do
not attribute causal or other physical properties to mathematical objects—to
places in structures—but on the view in question, one would not expect them
to. A given mathematical structure might not have any physical properties, and
even if it does have physical properties, a mathematician as such does not care
about them. As noted in my and Resnik’s books, the mathematician studies
a structure independent of any physical exemplifications it may have. On the
present view, whether a given structure is abstract depends on how the physical
world is, and pure mathematics does not concern itself with such matters.
In Physics B, Aristotle delimits what he takes to be a distinctive feature of
mathematical methodology:
The next point to consider is how the mathematician differs from the physicist.
Obviously physical bodies contain surfaces, volumes, lines, and points, and these are
the subject matter of mathematics . . . Now the mathematician, though he too treats
of these things (viz. surfaces, volumes, lengths, and points), does not treat them as
(qua) the limits of a physical body; nor does he consider the attributes indicated as
the attributes of such bodies. This is why he separates them, for in thought they are
separable from motion, and it makes no difference nor does any falsity result if they
are separated . . . [G]eometry investigates physical lengths, but not as physical . . .
(193b–194a)
See also Metaphysics M (1077b–1078a). Aristotle’sidea seemstobe thatmathemat-
icsconcernsitself with physical objects, butthe mathematician isnotconcerned
with their physical properties. In thought, the geometer separates surfaces,
lines, and points from the physical objects that contain them, and focuses
on their geometric properties. Here, we have the mathematician separating
structures from any systems they may be structures of. The physical exempli-
fications are of no concern to the mathematician, qua mathematician. But with
Aristotle, it does not follow that mathematical structures have no physical
properties. According to the present proposal, some do and some don’t.
For Aristotle, Plato’s mistake was to conclude that geometrical objects
are metaphysically separate from their physical instantiations, just because
mathematicians manage to ignore certain physical aspects of their subject
matter. But unlike Plato, Aristotle left no room for mathematical objects that
have no physical instantiations. Here we leave it genuinely open whether a
given structure has a physical exemplification.
On the view in question, I would retract my offhand comment that
‘mathematical objects—places in structures—are abstract and causally inert’
(1997: 112). The closest I can come to this traditional wisdom is the Aristotelian
comment that the practice of (pure) mathematics is independent of any
physical properties a given structure may have.
The context of my offhand comment was to note that structuralism is not
compatible with crude versions of the causal theory of knowledge. That much
is still correct. Since a given structure might not be physical, the epistemology
of mathematics had better not be entirely causal. We may obtain knowledge
of small structures via causal contact with their instances, as outlined in my
book (1997: 112–16), but we do not learn about larger structures that way.
These larger structures are not instantiated at all (in the physical realm).
Strange or not, this is one way out. To summarize, on the resolution in
question, we maintain that the essential properties of a natural number consist
of its relations to other natural numbers, and then claim that the purported
non-mathematical counterexamples, like abstractness, are not essential to the
natural numbers.
Adopting this proposal does not leave us completely home and dry. What
of the property of being possibly abstract? Surely it is essential to the number
512, or the natural number structure itself, or the set-theoretic hierarchy,
that it might be abstract, in the sense that there don’t have to be physical
instantiations of this structure. Clearly, the property of being possibly abstract
is not a structural property of that number or those structures. So we seem
to have a counterexample to the proposal. To pursue the strategy, we would
have to deny that modal locutions like ‘being possibly abstract’ correspond to
properties, or else insist that the essence-accident distinction is out of place
with them, or else we can restrict the slogan to non-modal properties, if this
last act of fiat does not look too desperate.
The other resolution of the problem concerning the slogans is perhaps
disappointing, since it does not invoke an interesting metaphysical position.
The ante rem structuralist can simply give up the slogan that all essential
properties of natural numbers concern the relations of such numbers to other
natural numbers. I would take back statements like there ‘is no more to
being the natural number 2 than being the successor of the successor of 0,
the predecessor of 3, the first prime, and so on’ (1997: 5–6) and the ‘essence
of a natural number is its relations to other natural numbers’ (1997: 72).
Recall that according to the view that Dummett (1991: ch. 23) calls ‘mystical’
structuralism ‘mathematics relates to abstract structures, distinguished by the fact
that their elements have no non-structural properties’. And Resnik’s ‘[t]he
objects of mathematics . . . are structureless points or positions in structures.
As positions in structures, they have no identity or features outside a structure.’
The present option is to concede the rejection of this ‘mystical structuralism’,
since there are no ‘abstract structures’ in this sense.
The concession does not affect the underlying philosophy of ante rem
structuralism. The revisions are limited to which slogans we can use and take
seriously, in order to postpone a detailed articulation of the view. The official
stance is (still) that a natural number is a place in the natural number structure.
The latter is construed after the fashion of an ante rem universal in that it exists
independent of any exemplifications it may have in the non-mathematical
realm. Then we rely on the account of ante rem structures articulated in my
book, or if you like it better, the account of patterns in Resnik’s book. The
underlying view that numbers and other mathematical objects, so construed,
are bona fide objects is unaffected.
The number 2, so construed, has lots of properties. Some of these properties,
like being the number of doors in my dining room, are contingent. Some of
the properties of the number 2, like being the number of parents of a normal
human, may be biologically necessary. Other properties of the number 2
are metaphysical and still others are mathematical. Of the mathematical
properties, some, like being the first prime, are internal to the natural number
structure. Others, like being the number of complex square roots of −1, are
external to the natural number structure. It does not matter. On the view in
question, every property that 2 enjoys comes in virtue of its being that place
in the natural number structure. This is because that is what 2 is. It is only the
slogan that must be given up.
2. Cross-Structural Identity
We have it that each mathematical object is a place in a structure. Now we
turn to the identity relation on mathematical objects. The next section focuses
on the identity relation on places within the same structure; here our topic
is places from different structures. As with much philosophy of mathematics,
the issue begins with the Benacerraf (1965). According to von Neumann’s
interpretation of arithmetic, each natural number is the corresponding finite
ordinal, so that 2 is {φ, {φ} }; according to Zermelo’s interpretation, 2 is
{ {φ} }. So which is 2?
The structuralist answers to this question are that both are and that neither
is. When we say that the ordinal {φ, {φ} } is 2, the ‘is’ is not identity. Rather, we
say that the ordinal occupies a certain place in a given system that exemplifies
the natural number structure—the system of finite ordinals. The ‘is’ in
‘{φ, {φ} } is 2’ is something like the ‘is’ of predication—I call it the ‘is’ of
office-occupancy (1997: 83). The same goes for the statement that { {φ} } is
two. We could say instead that { {φ} } is a 2, or, more fully, that { {φ} } occupies
the two-role in the Zermelo system. It is similar to saying that ‘Rabin was
Prime Minister’ and ‘Begin was Prime Minister’, and then wondering which of
them was the real Prime Minister. Both were and neither was. Both occupied
the office; neither is the office.
So far so good, but what of the ‘is’ of identity in the context of structuralism?
What are we to make of statements of identity or difference between places in
different structures? This is a situation where the faithfulness constraint and
the minimalism constraint led me astray. Resnik says that, in such cases, there
is no fact of the matter, or at least there is no fact unless the two structures
are taken to be part of a larger one (Evans (1978) notwithstanding). I more or
less agreed, but added a bizarre addendum. I take the liberty of quoting myself
at length. I began by endorsing indeterminacy, suggesting a notion of relative
identity (1997: 79–81):
I would think that a good philosophy of mathematics need not answer questions like
‘Is Julius Caesar = 2?’ and [‘Is {φ, {φ}} = 2?’]. Rather, a philosophy of mathematics
should show why these questions need no answers . . . It is not that we just do not care
about the answers; we want to see why there is no answer to be discovered—even for
a realist in ontology . . . One can form coherent and determinate statements about the
identity of two numbers: 1 = 1 and 1 = 4 . . . But it makes no sense to pursue the iden-
tity between a place in the natural number structure and some other object, expecting
there to be a fact of the matter. Identity between natural numbers is determinate;
identity between numbers and other sorts of objects is not, and neither is identity
between numbers and the positions of other structures . . . The Frege–Benacerraf
questions do not have determinate answers, and they do not need them . . .
We point toward a relativity of ontology, at least in mathematics. Roughly, math-
ematical objects are tied to the structures that constitute them. Benacerraf [1965, §
III.A] himself espoused a related view, at least temporarily: . . . ‘Identity statements
make sense only in contexts where there exist possible individuating conditions . . .
[Q]uestions of identity contain the presupposition that the ‘entities’ inquired about
both belong to some general category’ . . . In mathematics, at least, the notions of
‘object’ and ‘identity’ are unequivocal but thoroughly relative . . .
But the faithfulness constraint led me to balk at the clean relativity (1997:
81–2):
I do not wish to go as far as Benacerraf in holding that identifying positions in different
structures (or positions in a structure with other objects) [is] always meaningless. On
the contrary, mathematicians sometimes find it convenient, and even compelling,
to identify the positions of different structures. This occurs, for example, when set
theorists settle on the finite von Neumann ordinals as the natural numbers. They
stipulate that 2 is {φ, {φ} } and so it follows that 2 = { {φ} }. For a more straightforward
example, it is surely wise to identify the positions in the natural number structure
with their counterparts in the integer, rational, real, and complex number structures.
Accordingly, the natural number 2 is identical to the integer 2, the rational number 2,
the real number 2, and the complex number 2 (i.e. 2 + 0i). Hardly anything could be
more straightforward. For an intermediate case, mathematicians occasionally look for
the ‘natural settings’ in which a structure is best studied. An example is the embedding
of the complex numbers in the Euclidean plane, which illuminates both structures.
It is not an exaggeration to state that some structures grow and thrive in certain
environments . . . The point here is that cross-identifications like these are matters of
decision, based on convenience, not matters of discovery.
Charles Parsons (1990: 334) also questions the extreme relativity broached just
above:
. . . one should be cautious in making such assertions as that identity statements
involving objects of different structures are meaningless or indeterminate. There is
an obvious sense in which identity of natural numbers and sets is indeterminate, in
that different interpretations of number theory and set theory are possible which
give different answers about the truth of identities of numbers and sets. In a lot
of ordinary, mathematical discourse, where different structures are involved, the
question of identity or non-identity of elements of one with elements of another just
does not arise (even to be rejected). But of course some discourse about numbers
and sets makes identity statements between them meaningful, and some of that . . .
makes commitments as to the truth value of such identities. Thus it would be quite
out of order to say (without reference to context) that identities of numbers and sets
are meaningless or that they lack truth-values.
My stated view at the time was that there are no cross-structural-identity

facts that we start with, but identification-decisions create such facts.
Vann McGee (1997: 39) allows a role for this sort of stipulation in mathem-
atics:
What we do when we ‘identify’ the natural numbers with the von Neumann numbers
is to adopt a convention assigning truth values to sentences like ‘5 ∈ 7’, to which
[previous] usage, together with the mathematical facts, assigned no values. We are
free to do this, in the same way that we are permitted, it if suits our purposes, to
assign truth values to borderline attributions of vague terms by adopting conventions
making the vague terms more precise. Identifying the numbers with the sets is not
a matter of pruning our ontology, but rather a matter of refining our usage. What
there is is not a question of notational convenience, but how we use our symbols is,
in large measure, a matter of convenience.
This passage comes on the heels of McGee’s use of Quine–Putnam considera-

tions to reject ‘referential realism’, the view that mathematical language has a
determinate reference relation. The purpose of McGee (1997) is to defend what
I call ‘realism in truth-value’, the view that properly declarative mathematical
sentences have objective, determinate truth-values, without this referential
realism. In contrast, Shapiro (1997) proposes a referential realism. Accord-
ingly, the singular terms of arithmetic have determinate reference, to places
in the natural number structure. The singular terms of set theory also have
determinate reference, to places in the set-theoretic hierarchy structure. The
question here is whether some of the set-theoretic terms and some of the
arithmetic terms refer to the same objects.
Kastin correctly pointed out that if mathematical objects exist, as objects
not of our making, then identities between them cannot be stipulated. Among
existing objects, identities are discovered, not made. So the allowed stipulations
are inconsistent with the realism in ontology (see also Hellman 1999: 924;
2001: 192).
Nevertheless, some cross-structural identities are at least prima facie natural,
and perhaps even inevitable. Suppose that during a lecture, a mathematician
writes a closed integral on a blackboard, and asks the class to determine
whether the number it denotes is a natural number. A philosophy student,
fresh from a study of structuralism—having read part of Resnik’s book and/or
part of mine—quickly gets the floor while the other students are busy trying to
resolve the integral, and proudly announces the solution: there is no fact of the
matter. He says that the integral denotes a place in the real number structure,
and so it is indeterminate whether places from the real number structure
are the same as or distinct from places in the natural number structure.
Clearly, something has gone wrong; the philosophy student is confused. The
mathematician asked a sensible, substantive mathematical question. The job
of the students was to answer it, and our job as philosophers is to interpret it.
Following the faithfulness constraint, I wanted the mathematician’s language
to be understood literally, at face value. My solution, if you want to call it that,
was that the mathematical community had stipulated that the non-negative
whole numbers of the real number structure are indeed the natural numbers.
This makes the mathematician’s question a proper mathematical one.
Recall that Frege (1884: e.g. § 66) dropped his first attempt to characterize
the natural numbers since it provided no way to determine the truth-value of
a statement of identity between a number and an object not given as a number.
This became known as the ‘Caesar problem’. As above, I suggested that a clean
resolution of the Caesar problem is a motivation for structuralism. The issue
at hand here suggests that in the end, the ante rem structuralist does have a
version of Frege’s Caesar problem, and has not resolved it.
Resnik suggested that statements of identity between items from different
structures are determinate only in the context of a theory, or structure, that
encompasses both. If the structures are kept separate, there is no fact of the
matter concerning whether their places are identical or distinct from each
other. Let us use the word ‘rubric’ for a theory or structure that encompasses
ranges of mathematical objects.
Let us return to the imaginary mathematician, introduced above, who
asked whether a given closed integral denotes a natural number. According
to Resnik, since this mathematician is speaking of the natural numbers and
the real numbers under a single rubric, there would be a determinate identity
relation on the whole of her domain of discourse in that context. Obviously,
this is what she intended. I am not sure that Resnik’s position is stable, however,
since it invites a third-man type of regress. Suppose that a mathematician M1
is studying the natural numbers, and nothing else. So he is limited to a single
structure. Mathematician M2 is the mathematician introduced above, who is
talking about the natural numbers and the real numbers at the same time.
Suppose we ask if the natural numbers of M1 are the same or different from
the natural numbers of M2 . Resnik tells us that in such cases, there is no fact
of the matter. We have identity-facts only if there is a combined rubric for the
natural numbers of M1 and the natural numbers of M2 , but by hypothesis,
there isn’t. Our character M1 studies the natural numbers and nothing else. So
we envision a third mathematician M3 who studies the natural numbers of M1
and those of M2 . Are her natural numbers the same as those of M1 and/or M2 ?
No fact of the matter unless there is a rubric combining the natural numbers
of M1 , M2 , and M3 . And on it goes.
Recall that M2 studies the natural numbers and the real numbers. Suppose
that by using the resources of the full theory, she proves something about
the natural numbers. That is, M2 establishes a theorem in the language of
arithmetic, but the proof invokes her extra resources, the real numbers. This is
a common situation in mathematics, pretty close to the norm. Although the
proof is not available to M1 , is the sentence nevertheless true of the natural
numbers of M1 —or the natural numbers of any mathematician whether
or not he looks beyond the natural number structure? On pain of massive
revisions of mathematical practice (cf. the faithfulness constraint), we had
better say is true of the natural number structure. Our first mathematician
M1 may not have the resources to know (until he meets M2 ), but is true
in that context nonetheless. On Resnik’s view, however, we would be forced
to say that there is no fact of the matter whether holds of the natural
numbers of M1 .
A straightforward reply might go like this:
(*) Of course holds of the natural numbers of M1 . No matter how we
settle the nice question of identity, it is clear that the natural numbers
of M1 are isomorphic to those of M2 , and isomorphic structures are
equivalent. So from the fact that holds of the natural numbers of
M2 , it follows that it holds of the natural numbers of M1 . Q.E.D.
The problem isthatin order tosaythis, andspeakof an isomorphism between the
natural numbers of M1 and those of M2 , we have to bring the natural numbers
of M1 under the same rubric as the natural numbers of M2 . Isomorphisms
relate structures to each other. We can speak of an isomorphism between S and
T only if we are speaking of S and T at the same time—under the same rubric.
Suppose that we ourselves here and now utter (*). Are we talking about, or
including, the natural numbers of M1 ? According to Resnik, there is no fact of
the matter. And so we still do not know that holds of the natural numbers
of M1 . In short, the dilemma is that in order to invoke the isomorphism, we
have to bring the natural numbers of M1 under the same rubric as those of
someone else. On Resnik’s view, as soon as we do that, there is no fact of the
matter as to whether we are talking about the natural numbers of M1 .
Perhaps instead we avoid talk of isomorphism and try the following:
(**) There is a meta-theorem that the sentence holds of all ω-
sequences. Since the natural numbers of M1 constitute an ω-
sequence, we have that holds of those natural numbers.
This is no better. The problem now is that in order to say (**), we must envision
a rubric that consists of all ω-sequences, which, of course, would include the
natural number structure. But does this encompassing rubric include the
natural numbers of M1 , the mathematician who is just studying the natural
numbers and not a universe containing all ω-sequences? One would think
the new rubric includes the domain of M1 , since it speaks of all ω-sequences.
But Resnik’s view does not sanction this manifest intuition. The rubrics are
different and so there is no fact of the matter concerning identities between
them. Is there then a fact of the matter as to whether the original natural
numbers of M1 are the same as their counterparts in this new rubric, the one
that speaks of all ω-sequences? On Resnik’s view, it seems, there is no such fact
of the matter, and so we cannot apply this meta-theorem.
Let’s review some options on the issue of identity between the places of
different structures. One might insist on indeterminacy, following Resnik and
my former self (minus the stipulation stuff). On this view, there is no fact of
the matter whether the natural number 2 is identical to the real number 2 or
the set {φ, {φ}}. On the option in question here, the locutions ‘the natural
number 2’ and ‘the real number 2’ each have determinate reference—to a
place in the natural number structure and a place in the real number structure
respectively. The view is that there is no fact of the matter concerning whether
these referents are identical or distinct. The indeterminacy is charged to the
(mathematical) world, not to mathematical language. It is an ontological or
metaphysical indeterminacy.
This option runs against the Quinean dictum ‘no entity without identity’.
Quine’s thesis is that within a given theory, language, or framework, there
should be a definite criteria identity relation on its objects. Against this, the
present option answers these Caesar-type questions of identity or distinctness
with: ‘there is no fact of the matter.’
Of course, the Quinean dictum is not set in stone, and the ante rem
structuralist is free to demur from it, but let us explore some further proposals
that meet it. A second option would be to hold that there is a determinant
identity relation on the places of different structures, but that in at least some
cases, the identities are inscrutable. For example, there is indeed a fact of the
matter whether the natural number 2 is identical with or distinct from the
real number 2 and the set {φ, {φ} }, but we may have no way of knowing
this. It is an epistemic indeterminacy. Given that we have no clue as to how
at least some of these identities can be discovered, we may be left with some
unknowable mathematical facts. So in this case, the indeterminacy is charged
to our paltry epistemic situation.
A third option, the one I propose here, is to insist that places from dif-
ferent structures are distinct. This fits with at least one of my repeated
remarks/slogans: ‘mathematical objects are tied to the structures that consti-
tute them’. On the this third option, the natural number 2, the real number
2, and the set {φ, {φ} } are tied to different structures and so they are distinct
objects. Also, as indicated in the long quote above, the present option would fit
in well with the slogan, discussed in the previous section, that the essence of a
mathematical object is its relations to other objects in the same structure. The
natural number 2 and the real number 2 and the set {φ, {φ} } enjoy different
relations to different objects. It is part of the essence of the real number 2 that
it has a square root, it is greater than this square root, and it is less than π .
The natural number 2 does not have those properties. Similarly, it is part of the
essence of the real number 3 that there is a number that is exactly half of it.
The natural number 3 does not have that property.
To quote myself again: ‘Julius Caesar and {φ, {φ} } have essential properties
other than those relating to other places in the natural number structure’
(Shapiro 1997: 80). At the time, I said that this misses the point, but I guess it
does not.
On this view, the lingering indeterminacy is charged to the language. The
phrase ‘the number 2’ is systematically ambiguous, denoting a place in the
natural number structure, a place in the integer structure, a place in the
real number structure, a place in the ordinal structure, etc. Even the phrase
‘the natural numbers’ is systematically ambiguous, denoting both the natural
number structure itself and various substructures of other structures. For
example, ‘the natural numbers’ denotes a substructure of the real numbers
and another substructure of the complex numbers. So the phrase ‘the natural
number 2’ is also ambiguous. In addition to denoting a place in the natural
number structure, the term denotes the corresponding place in the real
number structure. Our third option, then, is a sort of semantic indeterminacy.
Damage control: I propose that even with this sharp and (admittedly) crude
resolution to the ‘Caesar’ issue, we can still come reasonably close to the
faithfulness and minimalism constraints. The crucial mathematical fact here
is that isomorphic systems are equivalent. If B is isomorphic to B , then any
sentence in the relevant language true of one is true of the other. This fact is
what drives McGee’s (1997) aforementioned defense of realism in truth value
without referential realism (and what I call ‘realism in ontology’). Here, the
point is that it simply does not matter whether a mathematician is thinking of a
given structure or a system that exemplifies it—perhaps a system constructed
from the places of another structure. So the mathematician can go back and
forth between the structure itself and any of the systems that exemplify it.
So when I said in the book that the mathematical community stipulates that
the finite ordinals are to be the natural numbers or that natural numbers
are also real numbers, I should have said instead that they (permanently or
temporarily) agreed to study a particular system that exemplifies the natural
number structure. In doing so, they invoke what I call the ‘places-are-offices’
perspective. But they still shed light on the natural number structure, since
the given system exemplifies that structure.
This sort of stipulation does not undermine the ontological realism, or
even the referential realism. The mathematicians’ decision to study the Von
Neumann finite ordinals is not to say that these ordinals are the natural
numbers. A study of these ordinals sheds light on the natural numbers
structure because the ordinals exemplify the structure. The fact that the finite
ordinals exemplify the natural number structure, of course, is not a matter of
stipulation.
Mathematicians make such agreements/stipulations for a variety of reasons.
In the case of the von Neumann ordinals, the move is made for purposes of
uniformity. The set-theorist needs ‘numbers’ for studying cardinality and it
is convenient to avoid new sorts of variables. Since the given ω-sequence of
sets will serve as natural numbers, he uses them for that purpose. The finite
ordinals serve as natural numbers because they have the right structure.
Other situations, like the embedding of the natural numbers in the reals,
or the complex numbers in the Euclidean plane, have powerful ramifications
within mathematics. That is, the decision to focus on a particular system sheds
new light on the underlying structure. It sometimes happens that we only
grasp a given structure when it is embedded in a richer one. In present terms,
we learn more about a given structure by studying a system that exemplifies
the structure, this system consisting of places from another, richer structure.
To quote Kreisel (1967: 166):
. . . very often the mathematical properties of a domain D become only graspable when
one embeds D in a larger domain D . Examples: (1) D integers, D complex plane; use of
analytic number theory. (2) D integers, D p-adic numbers; use of p-adic analysis. (3) D
surface of a sphere, D three-dimensional space; use of three-dimensional geometry.
Non-standard analysis [also applies] here . . .
Some of the embeddings are extremely natural, and have come to be part of
the framework for mathematics—to the point where names are shared. Once
again, cases in point are the mutual embeddings of the natural numbers, the
integers, the real numbers, etc. So, to repeat the resolution, the numeral ‘2’
denotes a natural number, a real number, a complex number, a finite ordinal,
etc., and the phrase ‘natural numbers’ itself ambiguously refers to the natural
number structure itself, a particular substructure of the integers, a particular
substructure of the real numbers, etc. Similarly, ‘π ’ denotes both a real and a
complex number.
One result of this interpretation is that it is often left indeterminate just
which structure and place we are talking about. However, I suggest that in
the practice of mathematics—whether pure or applied—this indeterminacy
does not interfere with communication or understanding. In virtually all
cases, even if it is not clear from the context which structure, and thus which
place, we are talking about, it does not matter since all of the relevant systems
are isomorphic to each other. So we can talk about the structure and the
systems exemplifying it at once.
A shift in reference can take place silently and smoothly, and on the fly.
One can use a numeral to refer to a natural number and immediately discuss
the properties of the corresponding real number, and transfer some of that
information back to the natural number.
So let us return to the aforementioned imaginary mathematician who
wrote a closed integral and asked whether it denotes a natural number. On the
interpretation suggested here, we can take his question literally, as referring to
the natural number substructure of the real number structure. If one does not
like this resolution of the ambiguity, we can render the query like this: ‘does
this term denote the image of a natural number in the obvious and natural
embedding?’ Not quite literal, but pretty close.
3. Identity and Indiscernibility

This final section concerns an issue that was raised in conversation by a few
prominent philosophers and logicians. It is broached in Burgess (1999: 287–8)
and Hellman (2001: 192–6), and is articulated in some detail in Keränen’s
‘The Individuation Problem: Realist Structuralism is Dead’, and a follow up
‘The Identity Problem for Realist Structuralism’ (2001) (written in response
to an ancestor of this paper). Keränen’s ‘The Identity Problem for Realist
Structuralism II’, Chapter 6 in this volume, is a reply to the present paper, and
my brief ‘The Governance of Identity’, Chapter 7 in this volume is a brief reply
to that.
The problem concerns structures that have what we may call ‘indiscernible’
places. Let me illustrate with a few examples. In my chapter 4 (1997: 115–16),
devoted to epistemology, I introduced the finite ordinal structures. The ordinal-
three-structure, for example, has three places with a linear order. It is
exemplified by the system consisting of my children, taken in birth order. In the
structure, each place can be uniquely characterized. One place comes first, and
it alone satisfies the formula ∀y(¬y < x). One place comes second, satisfying
∃y∃z(y < x & x < z); and one place comes third, satisfying ∀y(¬x < y). I also
consider finite cardinal structures, which have no relations. They are about as simple
as structures can get. The three-structure is exemplified by my children, the
five-structure by the starting players on a basketball team, etc. Each finite
cardinal structure is completely homogeneous in the sense that there is no
formula in the relevant language that distinguishes the places from each other.
Indeed, since the structures have no relations, then the relevant language has
no non-logical terminology. Any formula, with one free variable, that holds
of any place of the three-structure, holds of the other two places.
The Euclidean plane is another homogenous structure. Any formula (x),
with only x free, in the language of Euclidean geometry, that holds of a
given point holds of every point. The complex numbers, of course, are not
homogeneous, but they provide a more subtle example of the phenomenon
in question. Any formula (x), with only x free, in the language of complex
analysis, that holds of the complex number i also holds of −i. Generally,
any formula in the language of complex analysis that holds of the complex
number a + bi also holds of a − bi. So within the language of complex analysis,
a + bi is indiscernible from a − bi.
The situation under discussion arises whenever a structure has a non-trivial
automorphism, a one-to-one function from the domain to itself (other than
the identity mapping) that preserves the relations of the structure. Any
permutation of a finite cardinal structure is an automorphism, for the simple
reason that there are no relations to preserve. Any rigid translation or rotation
of the Euclidean plane is an automorphism, and with the complex numbers,
the function that takes a + bi to a − bi is an automorphism. Recall that
isomorphic systems are equivalent. So if f is an automorphism of a given
structure and fa = b, then for any formula (x), with only x free, we have
(a) iff (b). So a is indiscernible from b, even if a = b. Any structure with
a non-trivial automorphism violates the identity of indiscernibles—if the
only ‘discernings’ we allow are formulas, with only one free variable, in the
language of the structure.
Is this a problem for ante rem structuralism? Keränen argues that it is. His
bold conclusion is that the very idea of extracting the structure of a system with
a non-trivial automorphism, and treating that structure as an entity in its own
right, is incoherent. The problem, he says, is that the ante rem structuralist is
committed to the identity of indiscernibles, and this forces the structuralist to
identify all of the places of each finite cardinal structure with each other, so that
each structure has only one place after all. The ante rem structuralist must
also identify the complex numbers i and −i, which contradicts the theorem
that each number (other than 0) has two distinct square roots.
So why is the structuralist committed to the identity of indiscernibles?
Keränen proposes a general metaphysical thesis that anyone who proposes a
scientific or philosophical theory of a type of object must provide an account
of how those objects are to be ‘individuated’. Let L be a linguistic practice
and let a be an object in the purview of L. As Keränen puts it, an ‘account
of individuation’ for L would specify ‘the fact of the matter that makes a the
object it is, distinct from any other object’ in the purview of L, by ‘providing a
unique characterization thereof ’.
Keränen suggests that the burden on any advocate of any theory of objects
is to fill in the blank in the following:
(IND) ∀x(x = a ≡ ).
He states that ‘the ontology of a practice must, by that theory’s lights, consist
of items that can be referred to without ambiguity—that is, of properly
individuated items.’
There is a question of what resources one is allowed to use in order to
discharge this ‘individuation task’. Some philosophers have posited that each
object has an ‘haecceity’, which is a property enjoyed by it and it alone. For
those philosophers, the individuation task is trivially discharged, by filling in
the blank in (IND) with a term for the haecceity for a:
∀x(x = a ≡ Ha (x) ),
or perhaps just
∀x(x = a ≡ x = a).
Keränen concedes that if one does accept haecceities, then this unilluminating
trick does the individuation job. However, philosophers who eschew haec-
ceities—or philosophers who must eschew haecceities—have to resolve the
individuation task some other way.
Keränen argues that when it comes to structures with non-trivial auto-
morphisms, the ante rem structuralist leaves himself no room to resolve the
individuation task. According to the slogans discussed in Section 1, above,
the essential properties of each place of an ante rem structure consist of the
relations between that place and other places in the same structure. So one
would think that the sum total of the intra-structural relations would have
to resolve the individuation task—if anything can. Unfortunately, I said as
much myself. In defining the notion of a freestanding structure, I wrote:
‘Every office is characterized completely in terms of how its occupant relates
to the occupants of the other offices in the structure’ (1997: 100). This, it seems,
was a mistake, if ‘characterize completely’ entails ‘can be identified uniquely’,
as seems plausible. In Keränen’s terminology, it follows that intra-structural
relations resolve the identity task. Presumably, a structuralist cannot accept
haecceities for places, since a haecceity seems to be a non-structural property.
Keränen argues that we ante rem structuralists are committed to the
following thesis: for any structure S and any place a of S,
(STR) ∀x(x = a ≡ ∀( ∈ Sa ≡ (x) )),
where the leftmost quantifier ranges over the places of S, and Sa is the set
of ‘relational predicates’ in the language of the structure S that satisfy the
given place.
In this requirement, the ‘relational predicates’ in Sa are those that specify
relations of a place in the structure to other places in the structure. For example,
in arithmetic, ‘being the fourteenth prime’ is a relational predicate. According
to Keränen, the main formal criterion is that Sa contains no predicate whose
description essentially involves a singular term denoting either a place of the
structure or an object in a system exemplifying the structure.
Keränen then shows that (STR) entails the identity of indiscernibles, in
the strong form indicated above: every object is uniquely discerned by a
formula with one free variable. This, together with the aforementioned facts
about isomorphism, entails that no ante rem structure can have non-trivial
automorphisms, and so a structuralist account of Euclidean geometry and
complex analysis is refuted.
Is the individuation task a reasonable demand on a philosophical or scientific
theory? This raises the age old problem of the identity of indiscernibles. I do
not see how there can be a non-trivial resolution of the individuation task for
any view that holds that the usual array of mathematical objects exist. The
reason is that there are too many objects and not enough formulas.
Recall the form for resolving the individuation task:
(IND) ∀x(x = a ≡ ).
For the task to be accomplished in a non-trivial way, the blank would be filled
with a formula that does not have a singular term denoting a.
Consider, first, real analysis. There are uncountably-many real numbers,
but any language used by the philosopher to characterize the real numbers
has only countably many formulas. And so only countably many reals can be
‘individuated’ by formulas in the form (IND). Most real numbers cannot be so
individuated.
To be sure, for any pair a,b of distinct real numbers, there is a formula that
applies to a and not to b. Suppose, without loss of generality, that a < b. Then
there is a rational number r such that a < r < b. Since the rational number r
is a quotient of integers, it can be identified uniquely, and so there is a formula
equivalent to x < r. This formula is true of a but not true of b. However,
the individuation task is not to merely distinguish any pair of distinct objects
from each other, but to individuate each object. As Keränen puts it, the job is
to specify for each object a, ‘the fact of the matter that makes a the object
it is, distinct from any other object’, by ‘providing a unique characterization
thereof ’.
We can solve the individuation problem for real analysis if we are allowed to
fill the blank in (IND) with a set of formulas. Let a be a real number and let be
the collection of all formulas equivalent to x < s, where s is a rational number
greater than a, together with the collection of all formulas equivalent to t < x,
where t is a rational number less than a. Then, a satisfies every member of
and, by the completeness property, no other real number does.
The scope of this trick is limited. A language with a finite or count-
able alphabet has only continuum-many sets of formulas, and so at most
continuum-many objects can be individuated by sets of formulas in that lan-
guage. If a mathematical theory countenances more than continuum-many
objects, then the individuation task cannot be resolved in this way. Most of
the objects of functional analysis and set theory are still left unindividuated.
Against this, one might think that the principle of extensionality resolves
the individuation task for set theory. Two sets are identical if they have the
same members. So for any given set a, the blank in (IND) would be filled with a
statement that a has the members it does. But first we need to individuate the
members of a, so that we can refer to them uniquely. How is this accomplished?
Starting with the empty set, each hereditarily finite set can be individuated with
a single formula in the language of set theory. Sets of hereditarily finite sets
can then be individuated with sets of formulas, as above. So by the foregoing
trick, we can individuate the members of Vω+1 . But this is as far as it goes. A
bit after that, we run out of sets of formulas.
I suppose that the philosopher has the option of introducing a language
with an uncountable alphabet, and then using formulas (or sets of formulas)
from that language to fill the blank in (IND). For example, we might include
a name for each real number, or a name for each real-valued function, or
a name for each set. Then we can satisfy (IND) in a trivial manner. But
this just raises a question about how the characters in the language get
individuated.
In any case, someone who opts for realism in ontology and accepts
the individuation task must either introduce haecceities or adopt some
equally trivial resolution. Presumably, this route is not available to the ante
rem structuralist, due to the slogans about how places-in-structures are
characterized.
Recall that an in re realist still speaks of the natural number structure,
the complex number structure, etc. This Aristotelian philosopher thinks of
the structures as existing, but only in the systems that exemplify them. I
submit that if the ante rem structuralist has an unsurmountable individuation
problem, then so does this in re realist. By hypothesis, any system exemplifying
the complex number structure will consist of properly individuated items. In
the complex plane, for example, the two imaginary unit places are occupied by
the ordered pairs 0, 1 , 0, −1 . Presumably, these two pairs of real numbers
are themselves nicely individuated. Is there still a fact of the matter as to which
place is occupied by which pair? Is it determinate that 0, 1 occupies the i place
and not the −i place? We cannot stipulate that 0, 1 occupies the i place until
we know which place that is. So, if we take the individuation task seriously, one
cannot adopt an Aristotelian view of structures either. The only option left
is to reject the existence of structures altogether, a thoroughgoing eliminative
view, like those of Benacerraf and Hellman.
So what of the eliminative structuralist? Keränen seems to assume that
such a philosopher does not have an unresolvable individuation problem. He
says that for the eliminativist, the objects that fill the places in structures are
themselves individuated independently (of structuralism, presumably). For
example, any system exemplifying the complex number structure will have
two different objects in the i and −i places. Those objects are individuated
somehow in the background theory.
To assess whether this is sound, we need to enquire after the background
theory. As I noted in the book several times, the background ontology for
an eliminative program must be quite large, as large as the ontology of
mathematics. So by the foregoing considerations, only a trivial resolution
of the individuation task is available. Keränen’s point, confirmed in his
contribution to this volume in Chapter 6, is that the eliminative structuralist
is free to invoke haecceities, and thus a trivial resolution, while the ante rem
structuralist, like Resnik or myself, does not have this option. We are saddled
with the impossible task of a non-trivial resolution.
For what it is worth, I have never been moved by the identity of indiscernibles.
There are two ways that this thesis about identity can be understood, and
both are contentious (at best). First, the identity of indiscernibles might be a
metaphysical doctrine, stating that the universe of objects has a certain feature,
namely that each object—each grain of sand, each piece of dandruff, each
electron, each space–time point, each set—can be uniquely characterized,
either by a formula of some language or by a property, propositional function,
etc. If the individuation is supposed to be done with a formula, then the
thesis is that our linguistic or semantic resources are somehow up to this task
of individuation. But why think this? Why think that our human languages
are capable of uniquely picking out each and every object in the universe,
including the abstract objects?
If the individuation is to be done with properties or propositional functions,
regarded as objective and independent of language, then why think that
the intensional realm of properties and propositional functions matches up
to the realm of objects in the relevant way? What reason is there to think
that the realm of properties and propositional functions is up to the task of
individuating each and every object? Unless, of course, there are haecceities,
in which case the identity of indiscernibles is trivially true, and not very
interesting. If there are not enough properties to individuate each object, then
there will be distinct objects that share all of their properties.
These approaches to the identity of indiscernibles seem to presuppose
that ‘object’ is some sort of metaphysical primitive. Our philosophy aims at
determining the facts about these objects, and may succeed or fail in this task.
The other orientation toward the identity of indiscernibles comes from a more
Quinean approach to ontology (see Kraut 1980). It is an inversion of sorts.
The idea is that the universe does not come pre-packaged into objects. Which
objects there are is a function of our overall conceptual scheme. If, by using
the totality of our conceptual resources, we cannot distinguish a from b—if
everything true of a is true of b—then we should identify a and b. The proposal
is that if a and b are truly indistinguishable, then we have a single object before
us, not two. This seems to be a principle for restructuring the ship of Neurath,
the primary job left to philosophy from this perspective. This Quine–Kraut
proposal is of a piece with the criteria of simplicity and Occam’s razor.
Notice that this requirement, as stated, is not as strong as the identity
of indiscernibles. It does not say that (in a properly regimented language)
we should be able to uniquely characterize each object. The Quine–Kraut
proposal only says that we can distinguish any pair of distinct objects. As above,
this much can be accomplished for the real numbers. However, we would
still have to reject the places in the finite cardinal structures, since we
have left ourselves no resources to distinguish the places in this structure.
Perhaps this is an acceptable loss. Any work that those structures do can be
performed by the corresponding finite ordinal structures, and, as above, the
places of finite ordinal structures are nicely individuated internally. When
talking about cardinality, we would just ignore the ordering in the ordinal
structures.
Nevertheless, I submit that the Quine–Kraut requirement is not warranted
for mathematics. The fact is that we have powerful and eminently useful
theories which violate the distinguishability requirement. Euclidean space,
complex analysis, and set theory are cases in point. Consider geometry. As
we have seen, there is no non-trivial way to individuate or distinguish the
points, and yet it is a theorem of Euclidean geometry that there are lots of
different points. Since it is not an option to identify every point in Euclidean
space—so that there is just one point after all—the only way to satisfy the
Quine–Kraut requirement would be to add resources to Euclidean geometry to
distinguish the points (by breaking the non-trivial automorphisms). We could,
for example, drop the theory altogether and replace it with analytic geometry,
defined over n-tuples of real numbers. The talk of synthetic Euclidean geometry
would consist of simply ignoring the analytic parts of the structure. Similarly,
we could do without the complex number structure altogether, and make do
with the system of ordered pairs of real numbers instead.
Notice, however, that this maneuver to salvage the Quine–Kraut require-
ment demands revisions in established mathematical and scientific theories,
a rather unQuinean conclusion. As we have seen, the case of set theory
is even worse. Although there are no non-trivial automorphisms of the
set-theoretic hierarchy, cardinality considerations show that the relevant
distinctions cannot be made, let alone the relevant individuations.
To say the least, invoking haecceities would also be a very unQuinean move.
We might allow distinctions involving resources other than formulas with one
free variable. With complex analysis, we can distinguish i from −i as follows:
the pair i, −i satisfies the formula x + y = 0, and the pair i, i does not. If
i were identical to −i, then the ordered pair i, −i would be the same as the
pair i, i , and so these pairs would satisfy the same formulas. Of course, we
still have no way of telling, among i and −i, which is which, but maybe we do
not need that ability. In Euclidean geometry, if a and b are distinct points, then
the pair a, b satisfies a formula that a and b determine only one line, while
the pair a, a does not satisfy this formula.
We can apply the technique generally, if we can countenance minor
revisions. If S is any system of objects, let S be the same system with another
binary relation ≺, which is a linear order (or a well-order). Then if a and b are
distinct, then the pair a, b satisfies the formula (x ≺ y ∨ y ≺ x) and the pair
a, a does not. We add the new relation symbol just to discharge our task of
distinguishing the items in our ontology.
Keränen raises the possibility of solving the individuation problem using
formulas with more than one free variable, and he shows that this does
not work (due to the fact that isomorphic systems are equivalent). Here
we use formulas with two free variables to distinguish two objects a, b, by
showing that a, a is distinct from a, b . The isomorphism theorem does not
undermine this.
If the identity sign is a primitive of the language, and we have singular
terms denoting each object, then we can trivially individuate each object. In
complex analysis, i satisfies x = i and nothing else does. If this trivial resolution
is allowed, then the Quine–Kraut requirement is no requirement at all. Every
theory meets it.
In set theory, if a and b are distinct sets, then we have that ∃x( (x ∈ a & x ∈ /
b) ∨ (x ∈ / b & x ∈ a) ). Does this manage to distinguish a and b? Alternately, we
can let c be a name of such an x, and note that the sets a, b are distinguished by
(c ∈ a & c ∈/ b) ∨ (c ∈/ b & c ∈ a). If the first disjunct holds, then a satisfies c ∈ x
and b does not; if the second disjunct holds then a satisfies c ∈ / a and b does not.
This last assumes that we can use names for sets in making the distinctions.
The rejection (or trivialization) of the identity of indiscernibles is quite
general. The foregoing arguments are independent of the theses concerning
structuralism. Anything I said that suggests that I am committed to a
non-trivial resolution of the individuation task must be withdrawn.
As noted above, Keränen correctly derives the identity of indiscernibles from
the above thesis (STR). Thus I agree that (STR) entails that there is a non-trivial
resolution of the individuation task. However, if ante rem structuralism is to
be a full philosophy of mathematics, then there should be ante rem structures
of Euclidean space, the complex numbers, and the set-theoretic hierarchy. So
for us ante rem structuralists, there can be no non-trivial resolution of the
individuation task. Thus, we must reject (STR). But why think that Resnik
and I are committed to it?
One way around (STR) would be to introduce haecceities for places in
structures. So the two square roots of −1 have different haecceities, and each
point of Euclidean space has its own haecceity. Notice that this move does
not resolve the issue of the previous section. On the present plan, there is
a haecceity for the natural number 2 and there is a haecceity for the real
number 2. But this, by itself, does not say (one way or another) whether
those two haecceities, and thus those two places, are identical or distinct from
each other.
Once again, however, haecceities run against the slogans like ‘the essence of
each mathematical object consists of its relations to other places in the same
structure’. If a given mathematical object has a haecceity, that haecceity is
essential to that object, and said haecceity does not concern its relations to
other objects in the same structure.
In Section 1, I broached the possibility of dropping the slogans, in favor of a
more detailed articulation of the theses of ante rem structuralism. But let us
maintain the slogans for the sake of this argument, and drop haecceities. Let
a be a complex number b + ci, where b and c are real numbers and c = 0. It
would follow that the set Sa of relational properties enjoyed by a includes all
of its essential properties. Let a be b − ci. Then a satisfies every member of Sa .
Indeed, according to the slogans, the essential/relational properties of a are
the same as those of a. Yet we know that a is distinct from a.
So it seems that if we maintain the slogan about essences, we have two
distinct objects having all of their essential properties in common. This does
not strike me as problematic. If we do not invoke haecceities, then why should
we think that distinct objects always have distinct essences? Why think that
two distinct objects cannot have all of their essential properties in common?
Thus, I do not see how the ante rem structuralist is committed to a crucial
premise for the identity of indiscernibles, and for Keränen’s principle (STR).
There is, however, at least one embarrassing remark in my book on this
matter: ‘Quine’s thesis is that within a given theory, language, or framework,
there should be definite criteria for identity among its objects. There is no
reason for structuralism to be the single exception to this’ (1997: 92). The
context of that remark was Resnik’s claim that there is sometimes no fact
of the matter concerning when two structures are the same or different. What
I meant—or think I meant, or should have meant—was that if we are to
develop a theory of structures, then there must be a determinate identity
relation between structures. Given Quine’s slogan, which I endorsed, there
is no room for a view that takes structures seriously as objects and leaves
the identity relation between structures indeterminate. Surely the same goes
for places within a given structure (independently of the issues from the
previous section). When it comes to mathematical objects—places within a
given structure—identity must be determinate. Given two places in the same
structure, it is determinate whether they are identical or distinct. Either a = b
or a = b. However, I regret the use of the phrase ‘criteria for identity’ in the
remark, since it seems to suggest a commitment to a non-trivial resolution of
Keränen’s individuation task, and with the slogans of ante rem structuralism,
to (STR) and the problems that engenders. If I indeed meant that structuralism
requires a criterion for individuation, then I hereby take the remark back.
Hellman (2001: 193–4) raises an extension of the foregoing complaint against
ante rem structuralism. He suggests that the places of an ante rem structure
must be determined solely by the relations of the structure, and this, he
suggests, is incoherent:
. . . on Shapiro’s view . . . numerals denote the ‘places’ in a unique, archetypical
structure answering to what all progressions have in common . . . But the ‘places’
of the ante rem archetype (call it N, ϕ, 1 ) are entirely determined by the successor
function ϕ, and derivative from it in the sense of being identified merely as the terms
of the ordering induced by ϕ. Now in the case of an in re structure, we understand
a particular successor relation in the ordinary way as arising from the given relata,
reflecting, e.g. an arrangement of some sort . . . But if the relata are not already given
but depend for their very identity upon a given ordering, what content is there to talk
of ‘the ordering’? What can ‘succession’ mean, if we are abstracting from all in re cases and
if we can’t even speak of relata without making reference to the relation intended? . . .
This, I submit, is a vicious circularity: in a nutshell, to understand the relata, we must
be given the relation, but to understand the relation, we must already have access to
the relata.
Hellman paraphrases the objection as follows: ‘a purely structuralist view
of objects serving as relata cannot be sustained without departing substantially
from a structuralist view of structural relations, in effect, of structures themselves’
(2001: 195 n).
I do not know what to make of the metaphors of the successor relation
of the natural number structure being ‘given’, and our ‘having access’ to the
places of the structure. Presumably, Hellman does not mean to saddle the ante
rem structuralist with a Platonist epistemology that posits some sort of direct
access to structures and their places and relations. From the more literal parts
of the passage, Hellman means that the ante rem structuralist is committed to
a thesis that the places of a structure S are somehow ‘determined’ or ‘identified’
by the relations of S. Metaphysically, the relations are prior to the places, and
the relations fix the places.
This cannot be correct. Consider, once again, the finite cardinal structures,
which are degenerate cases of ante rem structures. The four places of the
cardinal four structure, for example, cannot be ‘determined’ by the relations
of the structure, for the simple reason that the structure has no relations
(other than identity). I did not mean to give the impression that the relations
of a structure are prior to its places. To be frank, I had not considered this bit
of metaphysics before, but the view must be that a structure is determined by
its places and its relations. Neither is prior to the other.
In the introduction, I did write that ‘Structures are prior to places in the
same sense that any organization is prior to the offices that constitute it. The
natural number structure is prior to ‘‘6’’, just as ‘‘baseball defense’’ is prior to
‘‘shortstop’’ or ‘‘U.S. Government’’ is prior to ‘‘Vice President’’ ’ (1997: 9). Later
in the book, I used the word ‘objects’ instead of ‘places’: ‘The structure is prior
to the mathematical objects it contains, just as any organization is prior to the
offices that constitute it’ (1997: 78). With these passages, I said (or meant to say)
that a given structure is ontologically or metaphysically prior to its places. This
is the central ontological thesis of ante rem structuralism. I did not intend the
passage to imply that the relations of each structure are prior to its places (nor
that its places are prior to its relations).
After broaching this objection, Hellman (2001: 195) asks ‘In what sense
can ante rem structuralism claim to be distinguishable from straightforward
objects platonism? Such platonism, after all, can readily speak of structures
and isomorphisms all it likes, once given its abstract archetypes.’ As above,
ante rem structuralism is an instance of the view I call ‘realism-in-ontology’.
Admittedly, it might be a variant of traditional platonism, at least as far as
ontology is concerned (see Shapiro 1997: 72–4). Perhaps it is just a difference
of emphasis.
Hellman (2001: 195–6) argues that even if the structuralist manages to
introduce a ‘privileged successor-type relation on an ante rem structure’, he is
then faced with a Benacerraf-type problem of multiple instantiations:
For if we suppose, for the sake of argument, that an ante rem progression, N, ϕ, 1
is somehow attained, we immediately see that indefinitely many others, explicitly
definable in terms of this one, qualify equally well as candidates to serve as the referents
of our numerals. We need only permute . . . the first two elements [of N], defining
ϕ ∗ from ϕ in the obvious way . . . [T]he structure N, ϕ ∗ , 1∗ is . . . able to serve as
ante rem archetypical progression every bit as well as N, ϕ, 1 [n.b., 1∗ is 2]. Indeed, on
what conceivable ground are we able to say that one and not the other is ‘the result
of Dedekind abstraction’? It makes no sense to say that 1 is ‘really first’ for ‘first’ has
no meaning except relative to a successor-type function. Relative to ϕ ∗ , 1∗ [i.e. 2],
not 1, is ‘first’ . . . Shapiro . . . does not avoid the multiplicity leading to Benacerraf ’s
argument, now directed at ante rem structures. That any number of such structures
are equally well qualified in all respects to count as the result of Dedekind abstraction
seems to undermine that very notion.
I do not know what to make of this objection. In the terminology of my

book, N, ϕ ∗ , 1∗ is not a structure. Rather, it is a system consisting of the
places of the natural number structure N, ϕ, 1 under the defined relations.
This system exemplifies the natural number structure, as do countless other
systems. I cannot give a knock-down argument against the claim that the
numerals of ordinary arithmetic refer to the system N, ϕ ∗ , 1∗ , rather than
the (original) structure. Indeed, I have no argument that the numerals of
ordinary arithmetic do not refer to the Von Neumann finite ordinals or
the Zermelo numerals. The underlying theme of the project (and indeed of
Hellman’s own modal eliminative structuralism) is the now commonplace
observation that the language of arithmetic only determines its ontology up
to isomorphism. My proposal is that we interpret the first-order variables of
arithmetic as ranging over the places of the ante rem structure. In Hellman’s
terms, I propose that we take the ante rem structure as the result of Dedekind
abstraction. Given that isomorphic systems are equivalent, I cannot rule out
alternate interpretations.
I also cannot rule out the hypothesis that there is more than one ante
rem structure isomorphic to the natural numbers (i.e., N, ϕ, 1 ). Hellman
is correct that an ontological realist cannot simply stipulate that there is at
most one structure for each isomorphism type. I also cannot rule out the
possibility that different structures share their elements (contra the proposal
toward the end of the previous section). So perhaps the realm of ante rem
structures contains both the postulated N, ϕ, 1 and N, ϕ ∗ , 1∗ . Unless
simplicity or Ockham’s razor can be invoked, but I do not want to broach that
issue right now.
It is best to remain modest. In responding to the above objections, I do
not claim to have made plain, once and for all, the notion of an ante rem
structure. Parsons (1990: 335) wrote that the
. . . absence of notions whose non-formal properties really matter . . . makes math-
ematical objects on the structuralist view continue to seem elusive, and encourages
the belief that there is some scandal to human reason in the idea that there are
such objects. My claim is that something close to the conception of objects of this
kind, already encouraged by the modern development of arithmetic, geometry, and
algebra, is forced on us by higher set theory.
I only hope that I have made ante rem structuralism sufficiently clear and
interesting for the discussion to continue.
References
Armstrong, D. M. (1989), Universals: An Opinionated Introduction (Boulder, Colo.: Westview
Press).
Benacerraf, P. (1965), ‘What Numbers Could Not Be’, Philosophical Review, 74: 47–73.
Blackburn, S. (1994), The Oxford Dictionary of Philosophy (Oxford: OUP).
Burgess, J. (1999), Review of Shapiro (1997), Notre Dame Journal of Formal Logic, 40: 283–91.
Dedekind, R. (1872), Stetigkeit und irrationale Zahlen, (Brunswick, Vieweg); trans. as
Continuity and Irrational Numbers, in Essays on the Theory of Numbers, ed. W. W. Beman (New
York: Dover Press, 1963), 1–27.
(1888), Was Sind und was Sollen die Zahlen? (Brunswick, Vieweg); trans. as The Nature
and Meaning of Numbers, in Essays on the Theory of Numbers, ed. W. W. Beman (New York,
Dover Press, 1963), 31–115.
(1932), Gesammelte Mathematische Werke, iii, ed. R. Fricke, E. Noether, and O. Ore
(Brunswick: Vieweg).
Dummett, M. (1991), Frege: Philosophy of Mathematics (Cambridge, Mass.: Harvard Uni-
versity Press).
Evans, G. (1978), ‘Can there be vague objects’, Analysis 38, 208.
Frege, G. (1884), Die Grundlagen der Arithmetik (Breslau: Koebner); trans. J. Austin as The
Foundations of Arithmetic (2nd edn., New York: Harper, 1960).
Hellman, G. (1989), Mathematics Without Numbers (Oxford: OUP).
(1999), Review of Shapiro (1997), Journal of Symbolic Logic, 64: 923–6.
(2001), ‘Three Varieties of Mathematical Structuralism’, Philosophia Mathematica,
9/3: 184–211.
Kastin, J. (1998), ‘Shapiro’s Mathematical Structuralism’ (manuscript).
Ker änen, J. (1999), ‘The Identity Problem: Realist Structuralism is Dead’, read at the
Pacific Meeting of the American Philosophical Association, Mar. 1999.
(2001), ‘The Identity Problem for Realist Structuralism’, Philosophia Mathematica,
9/3: 308–30.
Kraut, R. (1980), ‘Indiscernibility and Ontology’, Synthese, 44: 113–35.
Kreisel, G. (1967), ‘Informal Rigour and Completeness Proofs’, Problems in the Philosophy
of Mathematics, ed. I. Lakatos (Amsterdam: North Holland), 138–86.
MacBride, F. (1998), ‘Where are Particulars and Individuals?’, Dialectica, 52: 203–27.
McGee, Vann (1997), ‘How We Learn Mathematical Language’, Philosophical Review, 106:
35–68.
Maddy, P. (1990), Realism in Mathematics (Oxford: OUP).
Parsons, C. (1990), ‘The Structuralist View of Mathematical Objects’, Synthese, 84:
303–46.
Quine, W. V. O. (1992), ‘Structure and Nature’, Journal of Philosophy, 89: 5–9.
Resnik, M. (1981), ‘Mathematics as a Science of Patterns: Ontology and Reference’,
Noûs, 15: 529–50.
(1997), Mathematics as a Science of Patterns (Oxford: OUP).
Russell, B. (1903), The Principles of Mathematics (London: Allen & Unwin).
Shapiro, S. (1994), ‘Mathematics and Philosophy of Mathematics’, Philosophia Mathem-
atica, 2/3: 148–60.
(1997), Philosophy of Mathematics: Structure and Ontology (New York: OUP).
6
The Identity Problem for Realist
Structuralism II: A Reply to Shapiro
Jukka Keränen
In his Philosophy of Mathematics: Structure and Ontology, Stewart Shapiro gave what
is perhaps the most comprehensive presentation of realist structuralism to
date. In my ‘The Identity Problem for Realist Structuralism’,1 I argued that
any theory with the same general shape as that of Shapiro’s will either fail
to present a tenable account of mathematical ontology or else must betray
the very motivations from which it stems. While the argument in that paper
was intended to criticize realist structuralism in general, Shapiro is right to
identify his theory as a primary target and in ‘Structure and Identity’, Chapter
5 in this volume, he comes to its defense. Shapiro’s paper has two parts,
the first being a reply to certain criticisms by Jonathan Kastin, the second
being a reply to me. In this paper, I will give a detailed reaction to the
latter. I will argue that Shapiro has not yet provided a satisfactory solution
to the ‘identity problem’ for places in structures. In fact, I will conclude by
suggesting that any form of realism is either open to the identity problem or
else faces the well-known Benacerrafian problem of ‘multiple reductions’ in a
new guise.
Special thanks are due to Ken Manders and John McDowell.

1
Keränen (2001).
Jukka Keränen / 147
1. Identity and Individuation

The argument in ‘The Identity Problem’ had two main components. On the
one hand, I showed that mathematical structures typically contain places that
are indiscernible if individuated solely by the relations they have to one another.
On the other hand, I argued that any account of place-identity available
to the realist structuralist entails that relationally indiscernible places are
identical. Since she maintains that mathematical singular terms denote places
in structures, the realist structuralist is therefore committed to saying that,
for example, 1 = −1 in the group of integers (Z, +). I called this predicament
the identity problem.
The claim that constituted the first component of my argument is, I should
think, generally agreed to be true; it is the second component that is now
under dispute. The core of the second component consisted of establishing
two claims. First, I argued that any theory of ontology must be able to furnish
an account of identity for the objects of which it is a theory. I proposed that the
following is a natural way of furnishing such an account. Given a language L
and given two quantifiers ‘∀x’ and ‘∀y’ ranging over the domain of discourse
of L, just complete the blank in the following identity schema:
(IS) ∀x∀y(x = y ⇔ ).
When suitably completed, the result is an account that specifies the circum-
stances under which two singular terms of L denote the same object and,
more generally, specifies what governs the identity of objects in the domain of
discourse of L.
The second claim of the second component, perhaps the most contentious
aspect of my argument overall, was that the realist structuralist can only
employ intra-structural relational properties in completing the blank in (IS).
Let φ(x) indicate that x has the (possibly complex) property φ. Given any
system S, and given two quantifiers ‘∀x’ and ‘∀y’ ranging over the places of
the structure S of S, I argued that the realist structuralist should complete the
blank in (IS) as follows:
(STR) ∀x∀y(x = y ⇔ ∀φ(φ ∈ ⇒ (φ(x) ⇔ φ(y) ) ) ),
where ‘∀φ’ is taken to be ranging over the relational properties in S. I further
argued that the set should only contain relations that can be specified
without making explicit reference to particular elements in S. In this way, each
148 / The Identity Problem: Reply to Shapiro
place of S is individuated by the ‘purely structural’ relational properties its
occupants have to the occupants of the other places of S.2 Thus, whenever two
distinct elements in S that have the same intra-systemic relational properties,
according to (STR) they occupy the same place of S. While this conclusion
should seem prima facie plausible, in many cases it leads into absurdities. It
entails, for example, that in the additive group of integers (Z, +), the 1-place
is the −1-place. Since the realist structuralist takes mathematical singular
terms to denote places in structures, she would now have to say that 1 = −1.
The same problem arises for any structure whose instantiating systems have
non-trivial automorphisms.
I should note that there are two minor differences between the published
version of my argument outlined above and the earlier version Shapiro is
addressing in his ‘Structure and Identity’. First, in the earlier version, I began
by proposing that an account of identity could be furnished by completing
the blank in the following schema:
(IS*) ∀x(x = a ⇔ ).
I derived the crucial principle (STR) from this schema and pointed out that it
is a form of the ‘Leibniz Principle’ of the Identity of Indiscernibles for places
in structures. In the later, published version, I began by proposing that an
account of identity could be furnished by completing the blank in the schema
(IS) above. I showed that (STR) is simply a particular way of completing this
schema and, again, pointed out that it is a form of the Leibniz Principle. I
should emphasize that in either case, the Leibniz Principle itself does not carry
any direct argumentative weight. I will return to this issue in Section 3, below.
Let us briefly clarify the relationship between (IS) and (IS*). Given two
quantifiers ‘∀x’ and ‘∀y’ ranging over the domain of discourse of L, it is easy to
show that
∀x∀y(x = y ⇔ ∀φ(φ ∈ ⇒ (φ(x) ⇔ φ(y) ) ) ),
and
For all singular terms ‘a’ in L, ∀x(x = a ⇔ ∀φ(φ ∈ ⇒ (φ(x) ⇔
φ(a) ) ) ),
2
I framed my argument in this way in order to avoid the question of the individuation of
the relations themselves. In any case, the outcome resembles very closely what Shapiro tells us
about places and their occupants, and as long as satisfies my restriction, there is no problem in
viewing the individuating relations as relations in S also.
are interderivable in ordinary first-order logic, provided every element in the
domain of discourse of L has a singular term denoting it. However, most
mathematical languages do not satisfy this proviso and hence, I ended up
choosing (IS) as my starting point. After all, we would like an account of
identity for L to apply to all objects in the domain of L, not just to the ones
with singular terms denoting them.
Second, in the earlier version, I spoke of predicates individuating objects. I
am indebted to Shapiro for pointing out the dubiousness of this approach;
accordingly, the later, published version of the argument is formulated in
terms of properties. Still, as we shall presently see, his concerns here are different
from mine. I will return to this issue in Section 2, below.
Shapiro structures his response to my argument around the distinction
between what he calls a ‘trivial’ and ‘non-trivial’ way of completing the blank
in the identity schema.3 The trivial way of completing the blank amounts to
saying that each object is individuated by its ‘primitive thisness’, or haecceity, and
that there is simply nothing further to say. The non-trivial way of completing
the blank requires us to specify which general properties4 (or predicates)
individuate objects. With this distinction in view, Shapiro’s response to my
argument has four principal components. First, he argues that to furnish a
non-trivial way of completing the blank is too much to ask of any theory of
(mathematical) ontology: our (mathematical) languages simply do not have
the resources for uniquely characterizing every object in every domain of
(mathematical) objects. I take it that this is his main argument; henceforth,
I shall call it the ‘trivializing’ objection. Second, Shapiro points out that if
realist structuralism falls prey to the identity problem, then so does its chief
rival, eliminative structuralism. Third, he distinguishes two ways of construing
the Leibniz Principle and argues that, ultimately, neither one can serve my
argumentative purposes. Finally, he attempts to ‘analyze away’ the apparent
commitment he made in Philosophy of Mathematics to furnishing a non-trivial
account of identity for places in structures. I will address each of these
components in turn.
3
In the earlier version of my argument, I called the challenge of completing the blank in (IS)
the ‘individuation task’; Shapiro follows me in doing so. I shall avoid using this phrase here since,
as I shall explain below, there is a sense in which it is misleading.
4
A general property is one that can be had by several distinct objects, as opposed to a property
that can be had by only one object. For example, being yellow is a general property, whereas being
identical to a is not.
2. The Trivializing Objection

Shapiro tells us that he does not see ‘how there can be a non-trivial resolution
of the individuation task for any view that holds that the usual array of
mathematical objects exists’. The reason for this inability is that ‘there are too
many objects and not enough formulas’. For example, ‘there are uncountably-
many real numbers, but any language used by the philosopher to characterize
the real numbers has only countably many formulas’. There is a certain ‘trick’
that would allow us to individuate the real numbers but, as Shapiro notes,
‘the scope of this trick is limited’. In the end, ‘most of the objects of functional
analysis and set theory are left unindividuated’ (p. 135, above). Shapiro
concludes that my challenge for realist structuralism is one that no theory of
mathematical ontology could meet except trivially, by adopting Haecceitism.
It seems to me that Shapiro’s argument conflates two distinct theoretical
tasks. In order to see this, we must get clear on what an ‘account of identity’ is,
and what it is not supposed to accomplish. In ‘The Identity Problem’, I said that:
There are two basic requirements. First, given what we have just said, an account
of identity for the domain of discourse of L should specify the circumstances under
which ‘a = b’ is true in L by specifying what sort of facts govern the identity of
the objects referred to in L. In other words, an account of identity should specify
the circumstances under which ‘a’ and ‘b’ denote the same object. Second, although
the first requirement is phrased in terms of singular terms, an account of identity for
the domain of discourse of L should obviously apply to all objects in that domain, not
just to the ones with singular terms denoting them. (2001: 312)
Officially, then, I did not want to claim that in order to furnish an account
of identity for a given domain one must be able to explicitly specify a unique
characterization of each and every object in it. All I wanted to claim was that
one must complete the identity schema in a way that indicates schematically
what governs the identity of objects in that domain. For example, according
to Zermelo–Frankel set theory, the identity of any set is governed by which
sets are its members. According to Haecceitism, the identity of any object is
governed by its haecceity.
Introducing a bit of terminology may help make the point clearer. First, an
account of identity for a language L is simply a universally quantified formula that
indicates what governs the identity of objects in the domain of discourse of
L. Second, a condition of identity for a given object a in that domain is an instance
of some account of identity for L. Finally, sometimes we can explicitly display
what the condition of identity for a given object a amounts to. For example,
first, the account of identity for the language of Zermelo–Frankel set theory
is simply the Axiom of Extensionality:
∀x∀y(x = y ⇔ ∀z(z ∈ x ⇔ z ∈ y) ).
Second, the condition of identity for the set {∅} in Zermelo–Frankel set
theory is
∀x(x = {∅} ⇔ ∀z(z ∈ x ⇔ z ∈ {∅}) ).
Finally, third, for small sets such as {∅} we can explicitly display what its
condition of identity amounts to:
∀x(x = {∅} ⇔ (∅ ∈ x ∧ ∀z(z = ∅ ⇒ z ∈ / x) ) ).
This should make it clear that in order to furnish an account of identity for L, it
is certainly not required that we be able to explicitly display what the condition
of identity amounts to for every object in its domain. Neither is it required
that we be able to write down the condition of identity for every object. Note,
indeed, that if this was required, we could not furnish even the trivial account
of identity for the real numbers, contra Shapiro. As Shapiro himself emphasizes,
our mathematical languages do not have uncountably many singular terms
and hence, there are not enough instances of the formula Hx , ‘the haecceity
of x’, for all real numbers.
Shapiro’s prose suggests that he might now raise the following objection.
Our inability to write down the condition of identity for an object means that
that object is somehow ‘left unindividuated’ (p. 135, above). However, I am
unable to make sense of this phrase in the present context. For the purposes
of my argument in ‘The Individuation Problem’, I assumed that realism about
mathematical ontology is true. Given this assumption, there seems to be little sense in
saying that some objects in the domain of discourse of L are ‘not individuated’.
Such objects exist quite independently of our cognitive activities; therefore, of
course they are individuated regardless of our cognitive activities. Each one is the
object it is, distinct from any other, and (so I argued) there is some property that
makes it so, whether or not we are able to specify what that property is. This
is the reason why I now feel that the phrase ‘individuation task’ is misleading
in the realist context. After all, it is not as if we are doing the ‘individuating’ for
the objects in a given domain generally, let alone for each one individually.
That we are unable to specify explicitly the property that individuates a given
object does not mean that that object is somehow ‘left unindividuated’.
I should note that the foregoing line of thought relies on a conception of
ontology that is of a piece with what Shapiro calls the ‘metaphysical’ way of
construing the Leibniz Principle. The idea is that the universe comes ‘pre-
packaged’ into objects, all of which are uniquely individuated by some (possibly
complex) property. This conception of ontology is to be contrasted with what
Shapiro calls the ‘Quinean’ conception according to which the universe does
not come pre-packaged into objects. ‘Which objects there are is a function of our
overall conceptual scheme.’ Thus, ‘if, by using the totality of our conceptual
resources, we cannot distinguish a from b [. . .] then we should identify a and
b’ (p. 137, above). It is not entirely clear to me which conception of ontology
Shapiro favors, although his tentative admission that the identity of places
might be governed by haecceities would suggest he is not a Quinean in the sense
relevant here. Whatever the case, it seems to me that Shapiro’s ‘trivializing’
objection is equally faulty under the Quinean conception. For, surely even the
Quinean will agree that the fact that we do not have enough singular terms
to pick out each real number individually is compatible with the idea that real
numbers are properly ‘individuated’.5 Even if Shapiro were to go Quinean,
he would still have to recognize that in order to have a domain of properly
individuated objects in view, it is not necessary to be able to write down the
condition of identity for each one. Having an account of identity will suffice.
I submit, then, that my argument is not open to Shapiro’s ‘trivializing’
objection. It is perfectly compatible with saying, for example, that the identity
of objects in Zermelo–Frankel set theory is governed by the Axiom of
Extensionality and thus, that the Axiom of Extensionality constitutes a non-
trivial account of identity. Even though we cannot write down the condition
of identity for each and every set, the Axiom of Extensionality in no way
commits us to denying that sets we ordinarily take to be distinct really are
distinct. In contrast, the realist structuralist’s account of identity (STR) commits
her to saying that some objects we (suppose we) know to be distinct are in fact
identical. Again, this has nothing to do with our inability to write down the
condition of identity for every place in every structure. Rather, (STR) entails that
in some cases the conditions of identity, were we to write them down, would
be identical even though we (suppose we) know that the objects in question
are distinct. The realist structuralist is in a peculiarly uncomfortable situation.
5
Indeed, as Shapiro notes, otherwise the Quinean conception of ontology is actually refuted.
See p. 138, above.
3. The Leibniz Principle

I will now make six brief remarks concerning the skepticism Shapiro expresses
about the Leibniz Principle. As I noted above, Shapiro distinguishes two ways of
construing this principle, the ‘metaphysical’ and the ‘Quinean’. I will consider
each one in turn.
(1) Against the metaphysical conception, Shapiro considers the status of the
Leibniz principle relative to two possibilities. The first possibility is that it is our
‘linguistic or semantic resources’ that govern the identity of objects. But, ‘why
think that our human languages are capable of uniquely picking out each and
every object in the universe, including abstract objects?’ (p. 137, above). Two
remarks. First, again, this query seems to conflate the task of furnishing and
account of identity for a domain on the one hand, and the task of writing down
a condition of identity for each object in a domain on the other. Second, I do
not see how our ‘linguistic or semantic resources’ could govern the identity
of objects under the metaphysical conception of ontology since according to
that conception, the universe comes ‘pre-packaged’ into objects.
The second possibility is that it is ‘properties or propositional functions,
regarded as objective and independent of language’ that govern the identity
of objects. But, ‘why think that the intensional realm of properties and
propositional functions matches up to the realm of objects in the relevant way?
What reason is there to think that the realm of properties and propositional
functions is up to the task of individuating each and every object?’ (p. 137,
above). Three remarks. First, although I do not know how to argue the point
in general, it seems to me that in the case of pure sets it is obvious that there are
enough properties to individuate them. Given any set s at any level in the
hierarchy, there is a well-defined collection of sets that belong to s, and only to
s, by the Axiom of Extensionality. It seems entirely natural to say that for any
set r, there is a unique property r belongs to that is enjoyed by all and only those
sets to which r belongs. Hence, given any set s, there is a well-defined collection
of properties that s, and only s, has. This shows that there are enough properties
to individuate every pure set and hence, enough properties to individuate
enough objects to serve as the domain of any mathematical theory. If there is
a problem here, it is not one of cardinality.
Second, no matter what one makes of the strength of the foregoing argu-
ment, Shapiro would still have to produce some reason against thinking that the
two realms match up in the relevant way. After all, the majority of traditional
metaphysics supposes that the identity of objects is governed by properties.
Third, as Shapiro acknowledges, there does not seem to be any quick, prima
facie objection to the ‘trivial’ possibility that each object at least has the unique
property being identical to itself. Immediately after the query quoted above, he
adds: ‘Unless, of course, one believes in haecceities, in which case the identity
of indiscernibles is trivially true, and not very interesting’ (p. 137, above).
All I really need to say here is that as far as the argument in ‘The Identity
Problem’ is concerned, it is beside the point whether the Leibniz principle
is interesting; the only relevant issue is whether it is true. According to the
haecceity account, it is. In sum, there are two possibilities here. One, if the
‘two realms’ match up in the ‘non-trivial’ way, I can run my argument in ‘The
Identity Problem’ unchanged. Two, if the ‘two realms’ match up only in the
‘trivial’ way, each object being individuated by its haecceity, I will argue in
Section 5 that the realist structuralism theory of ontology is confronted with
a new problem.
(2) Against the Quinean construal of the Leibniz Principle, Shapiro points
out that the ontologies of many mathematical theories fly in its face. For
example, the domain of functional analysis has many more objects than we
could ever hope to individuate, or even pair-wise distinguish, by using our
conceptual resources. One remark. Again, in order to furnish an account of
identity for a given domain, it is not required that we be able to ‘uniquely pick
out’ each object in that domain. If the Quinean takes it, as I think she should,
that an account of identity is all we really need here,6 this easy refutation of
the Quinean picture is not available.
I submit, then, that Shapiro has not yet given any convincing reason to
think that the Leibniz principle should be rejected in the case of mathematical
objects. My own attitude about the Leibniz principle is this. I believe that
given any domain of objects, there is some fact that metaphysically underwrites
the distinctness of any two distinct objects in that domain.7 I will argue in
Section 4, below, that that feature must be analyzable in terms of properties
of individual objects. Hence, I believe that given any domain of objects, some
formula of the form
∀x∀y(x = y ⇔ ∀φ(φ ∈ ⇒ (φ(x) ⇔ φ(y) ) ) ),
6
I think that Quine himself would find this perfectly acceptable. See e.g. Quine (1992: 23 ff.).
7
See Keränen (2001: 312–13).
is true of that domain—a form of the Leibniz principle. I would like to
emphasize that to maintain this is not so metaphysically contentious as one
might suppose. The point is that the old disputes over the Leibniz principle
concern the status of robust forms thereof. It is indeed contentious whether
there is always some true formula of this form if it is required that the set
contains general properties. However, the ‘trivial’ accounts of identity do not
require this. They merely require that each object has the unique property
being identical to itself, and this much virtually any opponent of the Leibniz
principle would allow—including, as we have seen, Shapiro himself. Since I
am placing no apriori constraints on the contents of the set , my position
should seem acceptable to them.
4. The ‘Trivial’ Accounts of Identity

I will now comment on Shapiro’s efforts to ‘analyze away’ the apparent
commitment he made in Philosophy of Mathematics to furnishing a non-trivial
account of identity for places in structures. My overall strategy will be to
argue, first, that Shapiro has no choice but to adopt Haecceitism and, second,
that doing so will vitiate realist structuralism.
Shapiro’s principal motivation for wanting to analyze away the apparent
commitment to furnishing a non-trivial account of identity is that he thinks
such an account cannot be furnished for domains of mathematical theories.
However, as I have tried to show in the foregoing two sections, Shapiro’s
argumentrestson a conflation of twovery differenttheoretical tasks: furnishing
an account of identity on the one hand, and specifying a condition of identity
on the other hand. I maintain, accordingly, that his adoption of a trivial
account of identity for places in structures has no serious motivation.
Shapiro could nevertheless stand his ground and maintain that even if his
‘trivializing’ objection is set aside, he can avoid the Identity Problem by showing
that the realist structuralist does not have to furnish a non-trivial account of
identity. I will now consider the two ways in which Shapiro suggests this could
be accomplished. The first, of course, is the adoption of Haecceitism. The
second is the idea that distinct objects can have all their essential properties in
common. As to the first, I have always conceded that there is indeed no logical
obstacle to adopting Haecceitism about places in structures. I will argue in
Section 5, below, however, that doing so would compromise one of the central
motivations of structuralism. If my argument there is accepted, Shapiro would
then have to go with his second proposal. I will now argue that, in fact, the
second proposal commits him to Haecceitism as well.
Recall that essential properties of an object are the ones it has necessarily.
Thus, if φE (x) indicates that x has the property φ essentially, Shapiro’s second
proposal amounts to endorsing
(a) ∀x∀y(x = y ⇒ ∀φ(φE (x) ⇔ φE (y) ) )
while rejecting
(b) ∀x∀y(x = y ⇐ ∀φ(φE (x) ⇔ φE (y) ) ).
Thus, Shapiro concludes, he ‘does not see how the ante rem structuralist is
committed to [the] crucial premise for the identity of indiscernibles, and for
Keränen’s principle (STR)’ (p. 140, above).
I maintain that, strictly speaking, there are no distinct but essentially
indiscernible objects. Again, given any domain of objects, I believe that there
must be some fact that metaphysically underwrites the distinctness of any two
distinct objects in that domain. Suppose that you think that the objects a and
b are essentially indiscernible and yet distinct; you must still think that there
is something about the world that is responsible for the objects being two and
not one. Suppose that you think that the non-essential properties of a and b
are not up to the task;8 you must still think that there is something about the
world that is. Suppose that you think that it is a primitive, ‘brute’ fact that a
and b are two objects there rather than one; surely you must still think that
there is something about each one that makes it the case that it is the object it
is, and not the other. For, surely you want to be able to make sense of each
object being identical to itself, and distinct from all others, no matter what
other objects there are.9 Thus, you must think that each object at least has
the property being identical to itself. Note that we cannot say that it is the very
same property of ‘being identical to itself’ for then we would again have no
metaphysical foundation for saying that the two objects really are two and not
one. Thus, we must understand the description of this property ‘indexically’,
8
I take it to be obvious that individuating properties must be essential (although, perhaps, not
vice versa). We do not want to say that, for example, the identity of number 1 is governed in part by
1 having the (non-essential) property of being the number of stars in our solar system.
9
I take it that a bona fide structuralist is precisely committed to denying this. However, at this
point I am assuming that Shapiro has rejected the idea that mathematical objects are individuated
by their intra-structural relational properties.
so that each object has a unique property being identical to itself. But, surely the
property being identical to itself is an essential property of an object if anything is.
It follows trivially that
(b) ∀x∀y(x = y ⇐ ∀φ(φE (x) ⇔ φE (y) ) ).
Thus, if you reject (b), you are committed to adopting (b), a contradiction.
Of course, all the foregoing argument really shows is that if you reject (b) in
general, you must still allow the following: if H is the set of properties of the
form being identical to x, then
∀x∀y(x = y ⇔ ∀φ(φ ∈ H ⇒ (φ(x) ⇔ φ(y) ) )
is true of the domain in question—the most common construal of Haec-
ceitism. Shapiro is free to maintain the possibility of essentially indiscernible
objects as long as he excludes haecceities from the scope of this claim, admitting
that the identity of such objects is governed by their haecceities.
I submit, then, that Shapiro has no choice but adopt Haecceitism. Recall that
in Philosophy of Mathematics, Shapiro set out to provide a theory of mathematical
ontology. One of the central claims of ‘The Identity Problem’ was that in
order to provide such a theory one must furnish some account of identity for
the objects in the domains under consideration—a claim that Shapiro does
not dispute. Simply announcing that there can be essentially indiscernible
objects does not amount to discharging this task. But, as I have tried to show,
once Shapiro adopts the possibility of essentially indiscernible objects, the only
account of identity available to him is Haecceitism. It is my impression that
Shapiro would not regard this possibility as repugnant to the motivations of
structuralism; I will next argue that, alas, it is.
5. Realist Structuralism Reconsidered

In Section 1 of ‘Structure and Identity’, Shapiro considers four possible ways
of construing the fundamental structuralist thesis that ‘the essence’ of a
mathematical object is its relations to other mathematical objects:
(a) All necessary properties of mathematical objects are intra-structural;

(b) All mathematical properties of mathematical objects are intra-
structural;
(c) All essential properties of mathematical objects are intra-structural;
(d) (1) Each mathematical object is uniquely characterized by its relations
to other objects in the same structure, and (2) its other essential
properties are consequences of this being the case.
After discussing problems with each of these four possibilities, Shapiro regroups
by returning to (c). The problem with (c) was that some non-structural proper-
ties of mathematical objects are very plausibly essential; being non-spatio-temporal
might be an example (see p. 117, above). Shapiro proposes two ways in which
the realist structuralist could neutralize this problem. First, she could insist
that the putative non-structural essential properties are not essential after all
(p. 120, above). In some cases, however, doing so would be rather implausible.
Thus, second, the realist structuralist could simply reject the slogan that all
essential properties of mathematical objects are intra-structural. In more
detail, Shapiro’s final proposal is this:
(1) While some extra-structural properties are essential, (2) each property
a mathematical object has ‘comes in virtue of’ its being the place it is in
the structure to which it belongs.
Shapiro ‘would like to take back statements like ‘‘the essence of a natural num-
ber is its relations to other natural numbers’’ ’ (p. 120, above). All this, however, is
supposed ‘not [to] affect the underlying philosophy of structuralism’. Rather,
‘the revisions to ante rem structuralism are limited to which slogans we can
use and take seriously’. ‘The official stance is (still) that a natural number is no
more and no less than a place in the natural number structure’ (p. 121, above).
I want first to clarify the status of Shapiro’s final proposal. To say that
mathematical objects are places in structures does not amount to giving a theory
of mathematical ontology. To genuinely illuminate the nature of mathematical
ontology, the structuralist must also provide a theory of structures and places
in structures. I applaud Shapiro for doing the former, but do not see that he
has ever even attempted the latter. At the very least, he should endeavor to
say what makes a given place the place it is. In other words, he should furnish
an account of identity for places in a given structure.10 And, as I argued in
10
As far as ontology goes, the view Shapiro proposes is essentially classical Platonism. What
remains structuralist is his emphasis on the insight that mathematical objects are arranged in
structures and, presumably, that epistemology is to be built on this insight. The only genuinely
structuralist component of Shapiro’s final proposal is the claim that every property that a
mathematical object enjoys ‘comes in virtue of’ its being the place it is in the structure to which it
the preceding section, given his other commitments, Haecceitism is the only
account of identity available to Shapiro.
Suppose, then, that the realist structuralist adopts Haecceitism, maintaining
that the identity of places in structures in governed by haecceities, their
‘primitive thisness’. The problem with this move is that the resulting theory
of mathematical ontology will be unable to solve the ontological problem
that motivated it in the first place. Recall that the core motivation of the
ontological project of (any kind of) structuralism is the observation that, given
any mathematical theory, there are indefinitely many systems of objects that
could in principle serve as the domain of that theory. This is problematic
to a serious realist because she would like to say that, given any categorical
mathematical theory, any singular term in the language of that theory denotes
a unique object. Since we have no means of singling out any one system as
the domain of any given theory, the realist ambition is apparently defeated.
Further, if one accepts Benacerraf ’s well-known argument, it follows that
mathematical objects are not objects at all, given that we have no reason to
take, say, the number 2 to be any particular one of the indefinitely many
candidates. The realist structuralist solves this problem, quite plausibly, by
proposing that the domain of a given theory is simply the structure all the
systems exemplifying it have in common.
The kicker, then, is this. If the realist structuralist adopts Haecceitism, the
foregoing problem re-emerges on the level of structures. For, if the identity
of places in mathematical structures is governed by haecceities, there will be
indefinitely many distinct copies of the same structure. The point of realist structuralism
was precisely that while there are many systems that exemplify any given
structure, the structure itself is unique because the relations satisfied by the
multiple occupants of each place individuate the place itself uniquely. But, if
the identity of places is governed by haecceities, there is no reason to suppose
that there is just one unique structure of any given isomorphism type. At the
very least, the realist structuralist is forced to admit that there is no reason to
suppose that each mathematical singular term has a unique referent. Finally,
again, if one accepts Benacerraf’s argument,11 it follows that mathematical
objects are not objects at all.
belongs (p. 121, above). However, in discussing his proposal (d) a few pages earlier, Shapiro himself
points out that it is very unclear how one should construe a phrase such as ‘comes in virtue of’.
11
As, I take it, the average realist structuralist does.
6. Conclusion: The Identity Problem for Realism

I shall now step away from realist structuralism and consider the prospect for
a realist theory of mathematical ontology more generally. For, it seems to me
that the foregoing argument against realist structuralism is easily recast as an
argument against ontological realism in general.12
A realist about mathematical ontology maintains that mathematical objects
exist independently of our cognitive activities. Accordingly, she maintains that
each mathematical object is the object it is, identical only to itself and distinct
from any other, independently of our cognitive activities. Moreover, she
maintains that any categorical mathematical theory T has a unique domain of
discourse, and that the singular terms of T denote objects in that domain and
the quantifiers of T range over that domain. I maintain that given any such
theory, the ontological realist should furnish an account of identity for the
domain of discourse of the language L of T. There are four basic options. The
identity of mathematical objects is governed by either:
(a) intra-structural relational properties,
(b) extra-structural relational properties,
(c) non-relational properties, or
(d) haecceities.
It seems to me that adopting any one of these four options leaves the realist in
a difficult predicament. I will discuss each one in turn.
(a) Suppose that the identity of mathematical objects is governed by their
intra-structural relational properties. But, as I argued in ‘The Identity Problem’,
this leads into absurdities.
(b) Suppose that the identity of mathematical objects is governed by extra-
structural relational properties. There are two basic options here. First, one
could suppose that the identity of mathematical objects is governed by their
relations to physical objects. This is extremely implausible, however, partic-
ularly in the case of very large domains. Second, one could suppose that the
identity of mathematical objects is governed by their relations to non-physical
objects outside their ‘native’ structure. The most plausible way of fleshing out
this proposal is to suppose that each mathematical object is individuated by
12
I will only consider what Shapiro would call ‘metaphysical’ realism, setting aside the
‘Quinean’ variant. I intend to examine the possibility of Quinean realism at length elsewhere.
the relations in some rigid structure (one without automorphisms), avoiding
the identity problem. For example, we could suppose that each object in the
group of integers (a non-rigid structure) is individuated by the relations in the
ring of integers (a rigid structure). The problem with this proposal is that it
seems extremely unlikely that every non-rigid structure has a rigid structure
associated to it. What, for example, would be the rigid structure associated
with the Euclidean plane? Or, what of abstract groups in general?13
(c) Suppose that the identity of mathematical objects is governed by
their non-relational properties. The problem with this supposition is that
mathematical objects (famously) do not have any non-relational properties, save for the
trivial one of being identical to themselves.14
(d) Suppose that the identity of mathematical objects is governed by their
haecceities. The problem with this supposition is, again, that there is no reason
to think that each mathematical singular term has a unique, well-defined
referent. Further, we must again confront the old Benacerraf problem of
multiple reductions: mathematical objects are not objects at all, given that
there is no reason to take, say, the number 2 to be any particular one of the
indefinitely many candidates. Finally, since places in structures are individuated
by haecceities, the realist structuralist solution to the problem fails.
The ontological realist can of course concede these points and adopt a
form of eliminative structuralism. That is, she can concede that there is no
specifically mathematical ontology: a mathematical theory is about any system of
objects that satisfies its axioms. Mathematical statements are not to be taken
at face value: apparent singular terms, such as numerals, are disguised bound
variables ranging over elements in systems, and the identity of such elements
is governed by haecceities. This is not the place to discuss the merits of such a
view; I only want to emphasize that eliminative structuralism does not count
as ontological realism proper to the extent that by its lights, (categorical)
mathematical theories do not have unique domains of discourse.15
13
Recall that all abstract groups of order greater than 2 are non-rigid. Now, there is the obvious
thought that perhaps mathematical objects are simply sets in the pure hierarchy. Recall, then,
that it was precisely this thought that gave rise to the problem of multiple reductions and hence,
structuralism in the first place.
14
Perhaps mathematical objects have some general properties that can be considered non-
relational; being its own square root might be an example. It is clear, however, that there are not enough
such properties to individuate mathematical objects.
15
I take this occasion to record my attitude about the status of eliminative structuralism
vis-à-vis the identity problem. As Shapiro surmises (p. 136, above), I think that the eliminative
Finally, I suppose the realist could insist that the identity of mathematical
objects is primitive. However, as I explained in Section 4, above, I do not regard
this position to be genuinely distinct from Haecceitism. In sum, I suggest here,
and will argue at greater length elsewhere, that the ontological realist simply
cannot furnish a plausible account of identity for mathematical objects. If
we take the metaphysics of mathematical ontology seriously, we will have to
take individuation seriously. And, if we take individuation seriously, we must
take the identity problem and the Benacerraf problem seriously. And, finally,
if we take these two problems seriously, we will see that there is something
fundamentally wrong with ontological realism about mathematics.
Given what I have said thus far, the reader might expect me to now come
out in favor of nominalism about mathematical objects. For what it is worth,
then, I will not.16 While it is not possible to discuss my own views at length
here, I shall conclude this paper by giving a brief indication of the direction in
which I think the resolution is to be found.
I believe that there are mathematical objects, and I believe that (most)
mathematical singular terms have unique referents. Thus, I believe that
ontological realism about mathematics is true. However, I also believe that we
must re-examine what we can and should mean when we say that there are
mathematical objects. Indeed, we must re-examine what we mean when we
say that there are physical objects. For, I believe that the problems with realism
about mathematical objects result from transporting our naive realism about
physical objects into our theory of mathematical objects.
I believe with Carnap that ontological questions can only be raised and
answered in the context of a specific conceptual enterprise. In particular, I
propose that all object-employing conceptual practices contain two interre-
lated sets of norms that may be called a standard of identity and a standard of existence.
These sets of norms govern the way in which singular terms, quantifiers,
and predicate expressions function. My claim, then, is that all we can and
need to mean by ‘there are mathematical objects’ is that our mathematical
conceptual practices have consistent standards of identity and existence. That
structuralist is in principle free to adopt Haecceitism about the elements in the various systems. By
so doing, she can avoid both the identity problem, as well as the problem of multiple reductions
(unlike her realist counterpart). Assessing whether eliminative structuralism is ultimately tenable
is, of course, beyond the scope of the present paper.
16
For example, I feel that there is much to be said for Burgess’s call to resist (and perhaps even
combat) revisionism about the surface grammar of mathematical language.
is, the existence and identity claims that we commit ourselves to by adopting
these standards do not lead into contradictions. The picture is that in the case
of physical objects, the standards of existence and identity place requirements,
among other things, on the physical world, whereas in the mathematical case,
the requirements are ones pertaining only to consistency and, perhaps, the
quality of understanding. Yet, I maintain, mathematical and physical objects
are objects in the same sense: items that our singular terms denote, that are
or are not identical, that have properties, and so on. What is metaphysically
central about physical objects is not that their conditions of existence involve
spatially localized and temporally persisting hunks of matter but rather, that
we have consistent and unambiguous standards of existence and identity for
them in the first place. It is in this regard that non-physical objects are no
different from physical objects.
Given this framework of conceptual practices and norms, my proposal is
simply this: it is our conceptual resources generally that govern the identity
of mathematical objects. Consider, for example, the problematic case of non-
rigid structures. Even though there will be objects in such structures that
have the same intra-structural relational properties, we can simply decree how
many distinct copies of each given place-type there are. This does not mean
that what we can and cannot decree is arbitrary, depending as it does on our
previous commitments. In particular, the standards of identity and existence
impose various constraints on our ontological decisions, notably consistency.
Beyond such complex conceptual interrelations, however, numbers are and
can be whatever we want them to be.
References
Kastin, J. (1998), ‘Shapiro’s Mathematical Structuralism’ (manuscript).
Ker änen, J. (2001), ‘The Identity Problem for Realist Structuralism’, Philosophia Math-
ematica, 93: 308–30.
Quine, W. V. (1992), Pursuit of Truth (Cambridge, Mass.: Harvard University Press).
Shapiro, S. (1997), Philosophy of Mathematics: Structure and Ontology (New York: OUP).
7
The Governance of Identity
Stewart Shapiro
My exchange with Jukka Keränen exemplifies the long tradition of using the
thoughtful objections from opponents as an opportunity to clarify, further
articulate, and correct philosophical views. ‘Structure and Identity’, Chapter 5
in this volume, changes and extends aspects of the ante rem structuralism
developed in my (1997). Keränen’s ‘The Identity Problem for Realist Structur-
alism’ (2001) and ‘The Identity Problem for Realist Structuralism II: A Reply to
Shapiro’ Chapter 6 in this volume, constitute an advance in the articulation
of his metaphysical views from their public origin in his 1999 APA paper. At
some point, he and I may realize that we are talking at cross-purposes, or we
may end up with a clash of brute intuitions, and will agree to disagree. But we
are not there yet.
Let us review the basics. Keränen claims that any theorist who proposes an
ontology owes an ‘account of identity’. We discharge the obligation by filling
in the blank in the following scheme:
(IS) ∀x∀y(x = y ≡ ).
Presumably, we are not allowed to use the identity sign in discharging the
obligation. It won’t do to just put ‘x = y’ in the blank, resulting in the tautology
∀x∀y(x = y ≡ x = y). The account (IS) specifies ‘what governs the identity
of objects in the domain of discourse’. The crucial issue seems to be to
determine what it is to be ‘governed’, and why identity needs to be governed
by something.
We are told that in set theory, the axiom of extensionality does the job:
∀x∀y(x = y ≡ ∀z(z ∈ x ≡ z ∈ y)).
The extensionality principle shows how identity is governed by facts about
membership. The issue Keränen raises concerns the account of identity
for places of structures. From the slogans in Shapiro (1997) and certain
metaphysical principles, Keränen argues that the ante rem structuralist is
committed to
(STR) ∀x∀y(x = y ≡ ∀φ(φ ∈ → (φ(x) ≡ φ(y)))).
The variable φ ranges over the relational properties of a given structure S,
and the set ‘should only contain relations that can be specified without
making explicit reference to particular elements in’ the structure S. Notice
that (STR) requires that indiscernible places be identified. So (STR) entails that
ante rem structuralism fails for ‘non-rigid’ structures—those with at least one
non-trivial automorphism. In the complex plane, for example, (STR) entails
that i = −i, which is absurd.
In ‘Structure and Identity’, I proposed a modus tollens at this point, rejecting
(STR), and, in particular, rejecting the provisos on the set . To do this, I
have to either take back or reformulate some of the slogans and other remarks
in Shapiro (1997). Keränen maintains that the ante rem structuralist is still
committed to (STR), based on metaphysical principles concerning identity.
I’d be inclined to reject those metaphysical views, which would just amount
to a clash of intuitions. But I am not satisfied to leave it like that, at least not
yet.
Let us take a look at the metaphysics. In response to a point about our ability
to make distinctions with formulas in a fixed language, Keränen states that
the problem he raises against structuralism is not one of uniquely describing
each object. It is not a matter of whether the structuralist, or anyone else,
has the resources to pick out each object. This is not possible for any theory
(structuralist or otherwise) with a sufficiently large ontology, assuming that
the background language—the framework in which we give an account of
identity—is restricted to some fixed number of terms. We are told instead that
the project is to give a criterion of identity. A theorist who proposes an ontology
has to say, in general or schematic terms, what makes each object the object
it is, distinct from any other. We are also told that the individuation should
be done with properties (and relations), and not with the predicates available in
this or that language.
166 / The Governance of Identity
There is always the following, second-order logical truth:
∀x∀y(x=y ≡ ∀X(Xx ≡ Xy) )
(see Shapiro (1991, 63)). Will this work? If so, we would wrap up the ‘account
of identity’ for every second-order theory at once. I suppose that this depends
on what the second-order variables range over.
It would not be very deep or interesting to require that individuation be
done with sets (which is how the above sentence is shown to be a logical
truth). If the background theory is extensional, and allows the singleton
construction, then there is no real problem with identity. Singletons are the
extensional counterpart of haecceities, without being controversial. Clearly,
Keränen requires that individuation be done with intensional properties, and
the issue turns on what properties, or what kinds of properties, the theorist is
allowed to use when giving the criterion of identity.
The gloss on (STR) in ‘Identity Problem II’ says that the set ‘should only
contain relations that can be specified without making explicit reference to
particular elements in’ the given structure. I presume that this is not to be
taken literally. If the relations that individuate the places of a structure have
to be specified in some manner or other, then it is, after all, a matter of what we
can say in a given language—the language of the background metaphysics.
And of course we cannot do very many specifications with a finite or countable
language. The cardinality problem returns.
Keränen’s remarks are consonant with a tradition that takes properties
and relations to have a structure analogous to formulas in language. There
are conjunctive properties, disjunctive properties, etc. Keränen speaks of the
‘complexity’ of properties. For example, he mentions ‘relational’ properties.
These are monadic properties derived from relations. A typical example is the
property of being a child, which derives from the parent–child relation. Let
us say that the property of being a child invokes the parent–child relation.
Similarly, some properties invoke individual objects. For example, the property
of being Seymore’s child invokes Seymore. The property of being larger than
the White House invokes the While House.
So perhaps Keränen means that the members of the set in (STR) should
not invoke the places of the structure in question, in this sense. This replaces
the earlier talk of the elements of not containing singular terms denoting
places. But then what can the members of invoke? If it is just the relations
specified in the definition of the structure, then cardinality grounds alone
preclude (STR). Most structures have only a few relations, not enough to
generate uncountably many individuations.
Perhaps Keränen means that we should be able to specify the members of
schematically. We are required to specify the kinds of properties in by using
the relations of the structure. Keränen says that set theory has no identity
problem, since it is rigid (i.e. has no non-trivial automorphisms). To paraphrase
Keränen, let r be a set. Then there is a property Mr of ‘having r as a member’: for
any set t, the property Mr holds of t if and only if r ∈ t. Of course, if we cannot
characterize r uniquely, then we cannot ‘specify’ the property Mr . But this is
not relevant. Keränen holds that for each set r, there is such a property. The
axiom of extensionality entails that these membership properties suffice to
individuate each set t: it has all and only the properties Mr where r ∈ t. So the
membership properties tell us what makes t the set it is, distinct from any other.
I am not sure if Keränen thinks that the set-theoretic hierarchy meets (STR).
It depends on whether the properties Mr meet the provisos that he hoists
on the structuralist for set . The properties Mr are clearly relational, since
they invoke the membership relation, and this relation can surely be specified
without making explicit reference to particular elements in the set-theoretic
hierarchy. The relation is just ‘∈’. The property Mr is a set-theoretic analogue
of the ‘son of Seymore’ property. However, it is also clear that Mr invokes the
set r. I am not sure if the ‘r’ in Mr counts as an explicit ‘reference’ to a particular
element of the set-theoretic hierarchy. It sure looks like it does.
Consider the cardinal three structure. It has three places and no relations.
It is exemplified by my children. This is case of a non-rigid structure which
supposedly fails its identity problem. Suppose that we postulate ‘distinction
properties’: For each place a in the cardinal structure, let Da be the property
of being distinct from a. So for any place x, Da x if and only if a = x. The
distinction properties are analogous to the membership properties above.
They are relational properties, assuming that distinctness is a bona fide rela-
tion. If distinction properties are allowed in the set , then the cardinal three
structure satisfies (STR). Presumably, the problem here is that distinction
properties invoke the distinctness relation, which, in turn, invokes identity.
We might as well include haecceities.
Recall that a structure is rigid if it has no non-trivial automorphisms.
Keränen suggests that for the ante rem structuralist, any non-rigid rem
structure has an unsolvable identity problem. I submit that this is mistaken,
even by Keränen’s own lights.
Consider, for example, a structure S with a single relation, <. The axioms
of S are that < is a linear order, there is no first or last element, and for any
two elements x, y, there are only finitely-many z, such that x < z < y. Every
model of this theory is isomorphic to the integers, under ‘less-than’. So it is
categorical. But the structure S is not rigid. Let n be any integer. The function
that takes x to x + n is an automorphism. The only relation in the structure is
<, and we have that x < y if and only if x + n < y + n.
If Keränen is correct that the principle of extensionality provides an account
of identity for set theory, then the following
(IL) ∀x∀y(x = y ≡ ∀z(z < x ≡ z < y) )
provides an account of identity for S. Indeed, (IL) holds in S and it has exactly
the same form as extensionality. The <-relation governs identity in S in
exactly the same way that membership supposedly governs identity in the
set-theoretic hierarchy. Sets are individuated by their members; the places of
S are individuated by the places that are ‘smaller’ in the order.
We can also mimic Keränen’s membership properties. For each place r in
S, let Lr be the property of being larger than r in the given ordering. That is,
for any place s of S, Lr s if and only if r < s. If membership properties are a
legitimate means of individuating sets, then these ‘smallness properties’ suffice
to individuate the places of S. Each place t has all and only the smallness
properties Lr , where r < s. So if the set-theoretic hierarchy satisfies (STR), then
so does S, even though the latter is not rigid.
What of Keränen’s formal argument that non-rigid structures fail (STR)? This
is explicitly tied to the fact that isomorphic systems are equivalent—any sen-
tence in the language of the theory that is true of one is true of the other.
Suppose that f is an automorphism, and s, t are distinct places in the structure
such that fs = t. Let A(x) be a formula in the language of the structure, in which
the variable x is free. Then A(s) if and only if A(t). So s is indiscernible from t.
The problem here is that the isomorphism theorem only shows that s and
t cannot be distinguished (nor individuated) with a formula in the language
of the structure. However, as above, ‘Identity Problem II’ is explicit that the
account of identity is not a matter of finding a formula that picks out each
object uniquely. Keränen says instead that the individuation is to be done with
properties. The theorem that isomorphic structures are equivalent does not entail
that if an automorphism f takes s to t, then for every property P, we have Ps if
and only if Pt. The theorem only entails this for properties that are definable
in the language. In particular, automorphisms do not preserve the smallness
properties that individuate the places of S.
Of course, this is not an argument that every structure satisfies (STR). To
adjudicate this matter, we would need a fuller account (and justification) of
the properties that are allowed in the set . Let me concede, for the sake
of argument, that complex analysis, Euclidean space, and the finite cardinal
structures all fail (STR). Incidentally, another example is set theory with
urelements. Any permutation of the urelements gives rise to an automorphism
of the set-theoretic structure. As above, I remain unconvinced the ante rem
structuralist is committed to (STR). But again, for the sake of argument, let
us suppose for a while that (STR) is a genuine requirement. So there is no
ante rem structure for complex analysis, Euclidean space, the finite cardinal
structures, and set theory with urelements.
Under these circumstances, the best option for the ante rem structuralist
is to embed each of these structures in one that does meet (STR), following a
suggestion in my other contribution to this volume. For example, the cardinal
three structure would be replaced with the ordinal three structure. That
structure has three places, and a linear order. Any work, or any application, of
the cardinal structure can be done with the corresponding ordinal structure.
We can always ignore the order. Similarly, the complex number structure
would be replaced with R2 , with complex addition and multiplication defined
in the usual way. In effect, this allows us to use the linear order on the real
numbers to distinguish i (0,1 ) from −i (0,−1 ): since the real number 1
can be distinguished from −1, the pair 0,1 is distinguished from 0,−1 .
Keränen wonders what rigid structure might be associated with the Euclidean
plane. The structure of R2 works here too, if we include the operations of
complex analysis. This is the insight behind analytic geometry. We could still
pursue synthetic, Euclidean geometry by ignoring the analytic elements of
the structure. Those are the theorems that are invariant under the choice of a
reference frame and metric.
Let T be any structure. Suppose that we add a linear order (or perhaps a
well-order) to the places of T, or to be precise, we embed T into another
structure that has a linear order (or a well-order). Then (STR) holds for the
richer structure, following the trick we used with the structure S. Identity on
the richer structure is governed by the order.
I do not know if every structure the ante rem structuralist might want in
his repertoire can be accommodated. That is, I do not know if every structure
can be embedded in one that satisfies (STR). This depends on a version of
global choice. But surely the usual structures—the ones actually studied in
mathematics—can be so embedded.
Let us now cancel the concessions to (STR), made for the sake of argument.
Once again, in Keränen (2001) and ‘Identity Problem II’ we are told that the
principle of extensionality
∀x∀y(x = y ≡ ∀z(z ∈ x ≡ z ∈ y))
is in the form (IS) and gives an account of identity for set theory. The axiom
shows how identity is governed by the facts about membership. If I wanted to
be ornery, I might demand an ‘account of membership’ for set theory. This
would be to fill in the blank in the following scheme:
(MS) ∀x∀y(x ∈ y ≡ ).
To paraphrase Keränen, this is to ask for the facts that govern the membership
relation. If x ∈ y then there must be some fact that underwrites this connection.
For any sets x, y, someone who thinks that x ∈ y must think that there is
something about the world that is responsible for these objects to be related this
way. My ornery counterpart demands an articulation of this fact. I demand
an ‘account of membership’, just as Keränen demands an account of identity.
And, of course, we cannot use the membership relation in whatever fills the
blank in (MS), just as we cannot use identity in (MS). If the set theorist takes
the bait, and somehow fills in (MS), the mischief could continue. Someone
might demand an ‘account’ of the items used in the blank of (MS). Where
does it end?
The reply, I presume, is that membership is a primitive of set theory. This
means that we cannot—and certainly do not have to—give an ‘account’
of it. But if membership is such a primitive, then why isn’t identity another
primitive? Why do we have to give an ‘account’ of identity?
To be sure, there is an important asymmetry between identity and member-
ship. One of the epistemological strategies of ante rem structuralism is that the
axioms of a coherent, categorical theory constitute an implicit definition of the
primitive terms of the theory. The axioms of Euclidean geometry, for example,
implicitly define ‘point’, ‘line’, and ‘plane’ (see Shapiro 1997: 129–36; 1997:
157–70 for historical background). The axioms of second-order ZFC are not
categorical, but they do serve as at least part of an implicit definition of member-
ship. Although it is not in the form (IS), the axiomatization itself is (part of) the
account of membership in the only sense that one can demand an ‘account’.
In some cases, the axioms of a theory include the identity relation. But
we do not think of them as giving an implicit definition of identity. On the
contrary, identity is presupposed in giving the implicit definition. For example,
the second-order axioms of Peano arithmetic provide an implicit definition
of ‘natural number’, ‘successor’, ‘addition’, and ‘multiplication’. This is done
in terms of identity. It would not be correct to say instead that the axioms of
Peano arithmetic implicitly define ‘natural number’, ‘successor’, ‘addition’,
‘multiplication’, and ‘identity’ (and the other logical terminology ‘&’, ‘¬’, ‘∀’).
On the contrary, identity (and the other logical terminology) is presupposed
in giving the implicit definition of the non-logical terms.
As noted above, one can give an explicit definition of identity in second-
order logic, but perhaps that begs the present questions. If one uses only
first-order resources (or Henkin semantics for second-order languages), it
is not possible to give an implicit definition of identity. Let I be a binary
relation symbol, and let be any first-order axiomatization that is a purported
implicit definition of I as identity. Of course, does not include identity as
a logical symbol. Any model of has models in which I is not identity. The
usual axioms for identity force I to be a congruence, but not the identity
relation.
For what it is worth, my view is that we understand identity (and the other
logical terminology) well enough, and can use it (and them) to give implicit
definitions of various mathematical structures. For example, we declare that
the cardinal three structure has three distinct places. Nothing in the language
of the theory distinguishes those places, but they are three, and not one,
nevertheless. I remain unconvinced that a theory of a structure somehow
needs an ‘account of identity’ in Keränen’s sense. We know what identity is.
If we think of identity as one of the relations of a structure, then haecceities
are indeed relational properties of it. But I do not wish to engage this bit of
metaphysics.
Keränen and I may be at cross-purposes after all. Toward the end of
‘Identity Problem II’, he points out that his metaphysical views undermine any
ontological realism toward non-rigid theories. So the ante rem structuralist
is not the only one who would be inclined to reject the metaphysical views
in question. Anyone who holds that Euclidean space and complex analysis,
for example, describe collections of objects that exist independent of the
mind, must find a way to reject the metaphysical principles broached in
Keränen’s papers.
The very end of ‘Identity Problem II’ consists of a sketch of Keränen’s own
views on ontology. On the negative side, he writes that if we take the meta-
physics of mathematical ontology seriously, then we have to take the problem
of how they are individuated seriously, and if we do that we face the iden-
tity problem (and the Benacerraf problem of multiple realizations). This, he
claims, undermines ‘ontological realism’, which I call ‘realism in ontology’. But
Keränen goes on to reject nominalism—the view that mathematical objects
do not exist. He says that there are mathematical objects, and that most of
the singular terms of mathematical theories have unique referents. Moreover,
mathematical and physical objects are objects ‘in the same sense’. The conclu-
sion that ontological realism is true seems to be a direct contradiction of the
negative part.
To resolve this, Keränen suggests that we need a new account of math-
ematical existence. We are told that our conceptual resources ‘govern’ the
identity of mathematical objects. Little more than consistency is required for
us to speak of mathematical objects. So what about non-rigid structures, such
as Euclidean geometry and complex analysis? Keränen says that even though
there are objects in these structures that have the same intra-structural
properties, we can simply ‘decree’ how many distinct copies of each given
place-type there are.
Well, this is exactly what the ante rem structuralist does in defining struc-
tures. So at least so far, Keränen’s remarks are of a piece with my own
characterization of ante rem structuralism—the very view he attacks in the
original talk (1999), in (2002), and in ‘Identity Problem II’. Indeed, he seems to
adopt my otherwise controversial principle that any coherent implicit defin-
ition characterizes at least one structure. To take a simple example, consider
the cardinal three structure. As I define it, the structure has three places, and
no relations (other than identity). It is about as non-rigid as structures get.
Since there are no relations, any bijection of the places is an automorphism.
The definition of this structure amounts to a ‘decree’ that there are three
distinct ‘copies’ of each place-type, a decree that Keränen explicitly allows.
Keränen admits that the brief remarks at the end of ‘Identity Problem II’ are
only a start on proposals for ontology. For now, it looks like those proposals
amount to a rejection of the metaphysical principles that underlie the identity
problem. This may constitute a new metaphysics, which stands opposed to
the principles invoked in the attack on ante rem structuralism. I eagerly look
forward to the promised further developments.
References
Ker änen, J. (1999), ‘The Identity Problem: Realist Structuralism is Dead’, read at the
Pacific Meeting of the American Philosophical Association, Mar. 1999.
(2001), ‘The Identity Problem for Realist Structuralism’, Philosophia Mathematica,
9/3: 308–30.
Shapiro, S. (1991), Foundations without Foundationalism: A Case for Second-Order Logic (Oxford:
OUP).
(1997), Philosophy of Mathematics: Structure and Ontology (New York: OUP).
8
The Julius Caesar Objection: More
Problematic than Ever
Fraser MacBride
1. Introduction
Is it possible that Julius Caesar was not only a person but also a number? Might
the conqueror of Gaul—beneath his material guise—have been the bearer of
numerical properties? Common sense appears to inform us otherwise. Caesar
was a man and no doubt added up sums. But he could hardly have been the
result of a subtraction. And what appears to be true of Caesar also appears
to be true of other concrete items. They enjoy the benefits of being spatially
related and causally interacting with their neighbours. By contrast, numbers
appear quite different kinds of thing. They stand in equations and figure in
the ancestral of the successor relation to zero. It seems that there is a Moorean
fact, basic to our ordinary ways of thinking, any adequate philosophy of
mathematics must accommodate. The fact that must be respected is the fact
that no concrete item is a number. A theory encounters the ‘Caesar’ problem
when it fails to meet this adequacy constraint.
Thanks to audiences at the Universities of Bristol, Düsseldorf, and St Andrews for their helpful
comments. I would also like to thank Peter Clark, Bill Demopoulos, Katherine Hawley, Alex
Oliver, Michael Potter, Graham Priest, Stephanie Schlitt, Stewart Shapiro, Crispin Wright, and
an anonymous reader for Oxford University Press. I gratefully acknowledge the support of the
Leverhulme Trust whose award of a Philip Leverhulme Prize made possible the writing of this
paper.
Fraser MacBride / 175
These remarks serve only to introduce an aspect of the Caesar problem.
The Caesar problem demands our attention alongside other fundamental
issues that ask for elucidation and justification of the basic structure of our
conceptual scheme—for example, the problem of universals or the problem
of change. And, like them, the Caesar problem does not engage a single subject
matter easily isolated. It engages a variety of logical, semantical, and metaphys-
ical questions and their interrelations. The failure to recognize the rich and
varied nature of the Caesar problem has resulted in a failure to appreciate the
sort of difficulties that confront the provision of a satisfactory resolution. It
may also have resulted in the relative neglect of the Caesar problem, a problem
that deserves a rightful place alongside other more venerable, indeed ancient,
philosophical concerns.
2. What is the ‘Caesar Problem’?

2.1 Frege
Frege brought the ‘Caesar problem’ to the attention of analytic philosophy in
the course of his attempt to introduce abstract objects (directions, numbers)
into ordinary discourse. Frege began his discussion by forging the now familiar
connection between the notions of object and identity. Self-standing objects
(persons, mountains) may be identified and then re-identified on different
occasions and from different perspectives. So expressions that are used to refer
to objects must have associated with them a class of statements (Frege calls
them ‘recognition statements’) that settle identity criteria for these objects:
If for us the symbol a is to denote an object, then we must have a criterion which
determines in every case whether b is the same as a, even if it is not always in our power
to apply this criterion. (Frege 1950: §62).
Recognition statements determine when it is appropriate to label and then

relabel an object on a different occasion with the same expression. If abstract
objects are to be introduced into ordinary discourse then a class of recognition
statements must also be supplied for them.
Frege attempted to meet this constraint by stipulating identity criteria for
the abstract objects he proposed to introduce (1950: §§62–5). Numbers are
stipulated to be identical just if their associated concepts are 1-1 correspondent:
(HP) Nx:Fx = Nx:Gx iff F 1-1 G
176 / The Julius Caesar Objection
Directions are stipulated to be identical just if their associated lines are parallel:
(D =) Dir (a) = Dir (b) iff a is parallel to b.
However, Frege soon came to doubt whether these definitions supply genuine
recognition statements. He first concentrated his attention upon the case
of (D =):
In the proposition ‘the direction of a is identical with the direction of b’ the direction
of a plays the part of an object, and our definition affords us a means of recognizing
this object as the same again, in case it should happen to crop up in some other guise,
as the direction of b. But this means does not provide for all cases. It will not, for
instance, decide for us whether England is the same as the direction of the Earth’s
axis—if I may be forgiven an example that looks nonsensical. Naturally, no one is
going to confuse England with the direction of the Earth’s axis—but that is no thanks
to our definition of direction. (1950: §66)
Frege then went on to question (HP) ‘for the same reasons’ (1950: §68).
Frege appears to have deliberated in the following manner. Genuine
recognition statements distinguish an object of a given kind K from all other
objects. This means that they must determine the truth-values of two sorts
of identity statement—pure and impure. A pure identity statement says that
an object of kind K is identical to another object also explicitly stated to be of
kind K (‘Kx = Ky’). By contrast, an impure identity statement says that an
object of kind K is identical to another object that may not be described in
K terms (‘Kx = q’). (D =) and (HP) notably determine truth-values for pure
identity statements concerning, respectively, directions and numbers. They
each provide a rule for determining the truth-value of identity statements in
which, respectively, dual occurrences of direction and number terms flank the
identity sign (‘Dir (a) = Dir (b)’, ‘Nx:Fx = Nx:Gx’) by appeal to the obtaining
of a familiar equivalence relation (‘a is parallel to b’, ‘F 1-1 G’). But (D =) and
(HP) fail to settle truth-values for impure identity statements (‘Dir (a) = the
Earth’s axis’, ‘Nx:Fx = Caesar’) in which there are only single occurrences of
direction and number terms. They simply say nothing about statements of this
form. As a consequence (D =) and (HP) fail to supply identity criteria for the
objects they are designed to introduce.
2.2 Dimensions of the Caesar Problem

These rudimentary reflections do not, however, reveal the depth and breadth
of issues the Caesar problem raises. Beneath the superficial simplicity of Frege’s
reasoning lies a welter of distinct worries. These concerns may be arranged
along three distinct dimensions:
(A) Epistemology: do the identity criteria supplied for directions and num-
bers provide any creditable warrant for a familiar piece of knowledge,
namely that directions and numbers are distinct from such objects
as Caesar?
(B) Metaphysics: dothe identity criteria suppliedfor directionsandnumbers
determine whether the things that are directions or numbers might
also be such objects as Caesar?
(C) Meaning: do the identity criteria supplied for the novel expressions
putatively denoting directions and numbers bestow upon them the
distinctive significance of singular terms?
Whether these differentdimensionsof concern may ultimately be distinguished
will depend upon a variety of prior theoretical choices. For example, in the
context of a robust realism—say, aposteriori realism—that draws a clear
separation between ontological and linguistic issues, the metaphysical and
meaning dimensions are held apart. In the context of a minimalist approach to
ontology, these different dimensions of the Caesar problem will, by contrast,
be intimately related. In advance of settling anterior realist or anti-realist
choices the Caesar problem is likely to resist canonical specification. The
Caesar problem also raises different issues in the cases Frege considers (number
and direction). Here attention will be given primarily to the numerical case.1
2.3 The Epistemological Caesar Problem

According to one epistemological version of the Caesar problem suggested by
Frege’s remarks, we already know that Caesar is not a number. So adequate
identity criteria for numbers should settle that Caesar is not a number. But
(HP) leaves it open whether Caesar is or isn’t a number. Therefore, (HP) fails
to supply adequate identity criteria for the entities they respectively purport
to introduce.
Whilst this epistemological concern possesses considerable prima facie force.
a number of questionable assumptions are made that require (in this context)
to be legitimated:
(i) Common Sense: we ordinarily know that Caesar is not a number.
1
See Hale and Wright (2001: 352–66) for an extended discussion of the direction case.
(ii) Discrimination: adequate identity criteria for numbers should enable
a subject who grasps them to discriminate numbers from all other
kinds of object (however presented).
First, let it be granted that we do ordinarily know that Caesar is not a number
(Common Sense). According to the version of the Caesar problem delineated,
putative identity criteria for numbers that fail to confirm that ordinary item of
knowledge are inadequate. But the question needs to be asked: in what sense
inadequate? The guiding thought must be that identity criteria that fail to meet
this constraint fail to isolate the subject matter—the numbers—with which
ordinary discourse deals. In other words, this version of the Caesar problem
assumes that if (HP) is to be used to talk about an intended range of familiar
numbers these identity criteria must provide for all the discriminations that
we would ordinarily make between these and other objects (Discrimination).
However, Discrimination appears too strong an assumption to make. It may
reasonably be supposed that the ability to refer to a given kind of object
presupposes some capacity to recognize items of that kind. But it would also be
unreasonable to claim that a novice who only possesses limited recognitional
skills could never light upon the same subject matter as a thinker with more
refined discriminatory capacities. It therefore remains open that (HP) provides a
rudimentary but nevertheless genuine basis for identifying and re-identifying
numbers. They allow us to identify and distinguish between, respectively,
numbers (that are given as such) even though they fail to underwrite the
sort of discriminatory abilities required to identify and distinguish between
numbers and others objects differently presented.
These reflections invite a refinement of the epistemological concern. It is
certainly true that there is a distinction to be drawn between succeeding—on
however a rudimentary basis—in directing thought and talk upon a given
range of objects, and being able to draw sophisticated, fine-grained distinctions
amongst the objects in question. But the possibility of directing thought and
talk upon a subject matter presupposes a basic grasp of the sort of objects under
consideration. But someone who has only grasped (HP) has no grasp of the
sorts of objects that are under consideration. They don’t even know whether
persons are numbers. So they can hardly claim to be able to think or talk about
numbers.
However, this response is also contestable. Imagine a child otherwise
ignorant of arithmetic who is taught (HP). For all that has been said so far
this child may—within the arithmetical language game—go on to reliably
distinguish between different numbers (presented as such). According to the
result called Frege’s Theorem, (HP) entails the fundamental truths of arithmetic
(the Peano Postulates) (Wright 1983: 158–69; Boolos 1987). So, continuing the
fantasy, we can also imagine that the child goes on to develop basic arithmetical
skills (addition, multiplication, etc.). This means that the child may mingle
at school with peers who are taught arithmetic in the ordinary fashion and
come home with good test results in maths. Do we really want to say that this
computationally competent child does not succeed in talking about numbers?
Does the fact that he expresses incomprehension when asked whether Caesar is a
number immediately settle that his test results are nothing more than a sham?
Would it not be more appropriate to say that this child trains his thought
upon numbers well enough but lacks an additional piece of metaphysical
knowledge (that Caesar is not a number)?
Another assumption made by the epistemological version of the Caesar
problem also merits scrutiny. According to Common Sense we already know
Caesar is not a number. But do we know this? If we do, it is certainly not in
virtue of grasping Frege’s—let’s face it, pretty decent—guesses at identity
criteria for numbers and directions. So how else might we know this? It
may seem that there is no pressing need to answer this question. For there
is a tendency in contemporary philosophy to assign to common sense an
epistemologically and theoretically innocent nature. As a result the verdicts
of common sense are simply taken for granted. However, it is important
to bear in mind the possibility that common sense may itself be corrupt,
nothing other than the consequence of significant—albeit prolonged and
low level—theoretical labour. As Russell once remarked upon the common
sense understanding of such notions as ‘thing’ and ‘object’:
the thing was invented by prehistoric metaphysicians to whom common-sense is due.
(Russell 1911: 148)
It may indeed be that our common sense understanding delivers the verdict
that Caesar is no object. But it remains open that the endorsement of Common
Sense may incur significant epistemic costs.
There appear to be two possibilities concerning the putative knowledge that
Caesar is no number. Either this knowledge is immediate or it is derived. It
is important to realize that the former option is far from plausible. It is part
of the beguiling nature of the Caesar problem that when we try to form a
clear and distinct idea of Caesar we do not find it explicitly represented there
that he cannot be a number. Nor when we try to form a clear and distinct
idea of a number (say, zero) do we find it explicitly represented that zero
cannot be a person. Rather we encounter a modest silence on these matters.
Caesar has personal properties. Zero has numerical properties. But it is neither
explicitly ruled in nor ruled out that Caesar might be zero. So if we really
do know that Caesar is not a number then there must be some argument
implicit in our ordinary understanding that shows this to be the case. Since
the metaphysical version of the Caesar problem concerns the availability of
just such an argument let us turn our attention there.
2.4 The Metaphysical Caesar Problem

According to this development of Frege’s reasoning, it is impossible for
radically different kinds of object to overlap. (HP) fails to preclude the
possibility that the same objects fall under radically different kinds (persons,
numbers). This is because it leaves open whether a range of identity statements
concerning objects drawn from disparate kinds are true or false (for example,
‘Nx:Fx = Caesar’). Therefore, (HP) fails to provide adequate identity criteria
for the objects it is intended to introduce.
This version of the Caesar problem rests upon the following apparently sane
and sensible assumption
(iii) Sortal Exclusion: such radically contrasting kinds of objects as numbers
and persons cannot overlap.
But part of what makes the metaphysical Caesar problem so problematic is
that it is far from clear what legitimate grounds for Sortal Exclusion there might
be. It is frequently asserted that it is simply ‘absurd’ to suppose otherwise
(Parsons 1990: 308–9). However, brute intuition has proved a notoriously
unreliable guide in theory construction. So an argument for Sortal Exclusion is
wanted. One tempting strategy is to argue that numbers are abstract whereas
persons are concrete and thereby obtain Sortal Exclusion as a conclusion. Waive
the usual concerns about whether the abstract–concrete distinction is in good
enough shape to distinguish between mutually exclusive classes of abstract
and concrete items (Burgess and Rosen 1997: 12–25). Just suppose for current
purposes that concrete objects are located and capable of entering into causal
interaction whereas abstract objects are not. It still does not follow that the
kinds in question cannot overlap unless it is also presupposed that numbers
are abstract and persons are concrete.
Arguments for this last claim may appear readily forthcoming. It may be
thought that there are more numbers than concrete objects. Perhaps there
are infinitely many of the former and only finitely many of the latter. So,
it may be concluded, numbers cannot be concrete objects. Since persons are
concrete it follows that persons aren’t numbers. But this argument is too
quick. It neglects to rule out the possibility that some numbers (finitely many
of them) are concrete objects.
Another line of argument appeals to the necessary truth of a wide range of
mathematical claims. Necessary truths require necessary existences to serve
as their immutable subject matter. Since concrete objects are contingent it
follows that the objects picked out by mathematical truths cannot be concrete.
But this argument is also too quick. It assumes that the sentences that express
necessary truths must refer to necessary existences. This assumption may be
questioned. The necessary truth of a sentence may be sustained by virtue of its
constituent terms picking out different objects at different possible worlds. In
other words, the argument assumes that numerical terms are rigid designators.
However, if numerical terms are non-rigid then—for all that has so far been
established—they may at a given world pick out concrete objects (Caesar
amongst them).
A variation on this argument appeals to the role of numerical terms in
contingent counterfactual claims of applied arithmetic. Consider a range of
counterfactual circumstances in which the number of Fs remains the same
even though concrete non-Fs pass either in or out of existence from one
circumstance to the next. It follows that the expression ‘the number of Fs’
cannot refer to any concrete non-F. For example, the number of moons of Mars
is two. We can entertain counterfactual circumstances in which the number
of moons of Mars remains two even though Caesar had never existed. So the
number of moons of Mars cannot be Caesar. (Of course, it may take further
discussion to show that the numerical terms that figure in the statements of
pure arithmetic cannot pick out concrete items either.)
Here it is assumed that the numbers applied to concepts in counterfactual
circumstances actually exist there. Then since the concrete objects at issue
do not exist in those circumstances the desired consequence follows that the
numbers in question are none of the concrete things. But this assumption is
far from obligatory. The application of numbers to concepts in counterfactual
circumstances may not rest upon an ability to identify the numbers that exist
in those circumstances. Rather it may rest upon an ability to identify and count
with numbers in the actual world and then use these numbers to count the
objects falling under concepts in counterfactual circumstances from here. It is
therefore left open whether numbers exist or not in any given counterfactual
circumstance. Alternatively, the relation expressed by ‘is the same number
as’ may express an equivalence relation (a trans-world relation), weaker than
identity, between the shifting referents of numerical terms. If so, the fact that
the number of Fs remains the same across counterfactual circumstances fails
to determine that the number of Fs is not concrete in some world.
It may also be argued that concrete objects are contingent whereas numbers
are necessary and so no person can be a number. These claims may appear
beyond question. Surely it is ‘manifest’ that Caesar is no necessary existent
(Hale and Wright 2001: 366)? Surely it is beyond doubt that numbers cannot
be contingent? But these assumptions have been questioned. Field has argued
that numbers are contingent existences; after all, he claims, there is no logical
incoherence in the suggestion that numbers might fail at a given world to
exist.2 As part of a defence of a simple form of quantified modal logic (including,
crucially, the Barcan formula: ♦∃xφ → ∃x♦φ) Linsky and Zalta have argued
that every object exists necessarily.3 However, the point is not only that these
assumptions have been questioned. More significantly, the point is that a
failure to appreciate that these assumptions may be questioned constitutes a
failure to appreciate the problematic character of the (metaphysical) Caesar
problem itself.
A common first reaction to the Caesar problem is to take it as manifest that
Caesar is no number. So Sortal Exclusion is simply taken for granted. But we are
able to see our way past this initial response when it is appreciated that it is
no straightforward matter to settle whether different kinds of objects—that
are apparently as unlike as objects can be—are really distinct. The matter is
difficult to settle because (in part) it is not explicitly written into the nature
of persons that they are not numbers (or vice versa). But nor is it explicitly
written into the nature of persons or numbers that they are contingent or
necessary. So to assume on manifest grounds that Caesar cannot be necessary
or that numbers cannot be contingent is simply to ignore the problematic
character of the Sortal Exclusion assumption. What is wanted here is just an
2
See Field (1993). Field’s proposal is contested by Hale and Wright (1992, 1994). Field responds
in his (1993). For further discussion see MacBride (1999: 443–7).
3
See Linsky and Zalta (1994, 1996). Williamson defends a similar position in his (1998).
instance of what is wanted generally: a principled account of why objects of
one kind cannot possess features (necessary existence, abstractness) usually
associated with different kinds.
In response it may be claimed that it is plainly constitutive of being a person
to be contingent. After all, persons come to be and pass away. They begin and
cease to exist. They might not have existed. They can hardly be necessary
existents! Of course, this train of thought will hardly settle that Caesar is not a
number unless it is also shown that numbers cannot be contingent. But, more
significantly, ask yourself the question: is that an accurate statement of what
we know to be true of persons? Might it not be more accurate to say that persons
take on and then throw off a material guise and it is left open—a matter upon
which speculation may never cease—what, if anything, happened before,
next or whether they might never have existed? I do not mean to suggest that
persons continue to exist without bodies or as bare abstract entities. The point
is rather that to legitimate the Sortal Exclusion assumption that underwrites the
metaphysical version of the Caesar problem it must be demonstrated that no
person is a number. And if the resolution is to be theoretically satisfying this
fact must somehow be guaranteed by the underlying nature of persons and
numbers. But since there is no immediate incompatibility between being a
number and being a person the intriguing difficulty we have to confront is
that we have apparently no idea of how Sortal Exclusion might be legitimated.
In any case appeal to different principles of modal existence is far too
coarse-grained a basis upon which to ground Sortal Exclusion. For suppose that
one were to become convinced that persons exist necessarily. Would one then
feel any more comfortable with the suggestion that Caesar is a number? Or,
alternatively, suppose that one already believed God, or some other plausibly
non-numerical item, exists necessarily. Would it then be legitimate to suggest
God is a number? The intuitive response—not to mention the theological
one—is likely to be that it is not. And until it is established that there cannot
be different (non-overlapping) kinds of necessary existent this response cannot
be rejected out of hand.
So what gives rise to this intuitive response? It does not appear—as one
might initially have thought—to arise from any overt incompatibility between
the different kinds in question. Instead it appears to result from the fact that
we have no intellectual stomach for irresolvable metaphysical inscrutability.
For if Caesar is a number then this identity is simply brute. The different
ranges of properties associated with being a person and being a number are
so distinct in kind that there is nothing that might be said to render this
identity transparent to the understanding. It is entirely opaque how a single
object could be the subject of such diverse properties. Consequently even an
exhaustive investigation (at the limit of enquiry) of the personal properties
of Caesar will not enable us to decide whether Caesar is a number, and if so,
which one. Similarly, no amount of investigation of the numerical properties
of 4 will determine whether it is also a person.
The issues surrounding the Caesar problem encroach here upon traditional,
metaphysical concerns about the nature of substance, about what makes an
object a unified, integrated whole. Consider the following remarks from
Leibniz:
I also maintain that substances (material or immaterial) cannot be conceived in their
bare essence, devoid of activity; that activity is of the essence of subject in general . . .
it must be borne in mind above all that the modifications which can occur to a single
subject naturally and without miracles must arise from limitations and variations in
a real genus, i.e. of a constant and absolute inherent nature. . . . Whenever we find
some quality in a subject, we ought to believe that if we understood the nature of both
the subject and the predicate we would conceive how the quality could arise from it.
So within the order of nature (miracles apart) it is not at God’s arbitrary discretion
to attach this or that quality haphazardly to substances. He will never give them
any which are not natural to them, that is, which cannot arise from their nature as
explicable modifications . . . what is natural must be such as could become distinctly
conceivable by anyone admitted into the secrets of things. (Leibniz 1981: 65–6)
Two relevant thoughts may be distinguished here. First, it is claimed that

substances cannot be conceived as bare particulars. Second, it is stated that the
exhibition of properties by a substance must somehow be rendered intelligible
by the underlying ‘real kind’ of the substance. Both thoughts plausibly militate
against the identification of Caesar with a number. For if Caesar is a number
then the subject that underlies the relevant personal properties and the
subject that underlies the relevant numerical properties can be no more than
barely identical. Moreover, if it is Caesar’s nature to be human then he cannot
also be a number. For the possession of numerical properties is rendered
not one whit intelligible by an underlying human nature. Obviously these
reflections present no decisive case. It is arguable that (in certain limit cases)
bare identities may be properly admitted. Moreover, if one is willing to admit
such conjunctive kinds as being a person and a number then Caesar’s underlying
nature will render his possession of numerical qualities intelligible after all. Of
course, this raises the question of whether conjunctive kinds are themselves
intelligible.
More significantly, Leibniz’s views on substance flow from his endorsement
of the principle of sufficient reason—a principle that demands the intrinsic
intelligibility of the universe. But since we have jettisoned the principle it is
difficult to see how it can be maintained that the world ought to be intrinsically
intelligible. The demand that the world should conform to the patterns of
our thoughts about it, that it should be transparent to even our idealized
understanding, appears no more than conceit. Should we therefore continue
to maintain Sortal Exclusion or should we be prepared to simply leave it open
that—for all that we know—Caesar is a number?
2.5 The Meaning-Theoretic Caesar Problem

This version of the Caesar problem questions whether (HP) succeeds in
conferring contenton the natural number expressionsitpurportsto introduce.
Frege initially sought to introduce numerical terms (‘Nx:Fx’) contextually by
fixing the content of identity sentences in which they occur. However, (HP)
fails to settle the content of all the identity contexts—specifically contexts of
the form ‘Nx:Fx = q’—in which the introduced expressions feature. Recall:
(HP) simply fails to say anything about the significance of contexts that
feature a singleton occurrence of the numerical operator. Therefore, (HP)
fails to bestow the significance of singular terms upon the expression it
introduces.
In fact, this version of the Caesar problem is multiply ambiguous. It all
depends upon what ‘content’ is taken to mean. If ‘content’ means sense then
the complaint comes down to this. (HP) fails to determine whether identity
contexts of the form ‘Nx:Fx = q’ have any sense. This calls into question
whether (HP) fixes a sense even for sentences of the superficially tractable
form ‘Nx:Fx = Nx:Gx’. When it is articulated in terms of sense the meaning-
theoretic version of the Caesar problem may be developed in the following
manner.
Frege sought to introduce numerical terms by fixing the truth-conditions
of the identity sentences in which they occur. However, if this method of
fixing truth-conditions is to result in the introduction of genuine numerical
terms—singular terms that purport to stand for objects—then the sentences
of the form ‘Nx:Fx = Nx:Gx’ whose truth-conditions are fixed must be
genuinely logically complex. Sentences of this form must be understood as
saying of Nx:Fx that it satisfies the predicate ‘. . . = Nx:Gx’. Consequently,
(HP) will only succeed in conferring individual significance on the component
numerical terms of the sentences whose truth conditions it fixes if it also
determines a meaning for such predicates as ‘. . . = Nx:Gx’ and ‘Nx:Fx = . . .’.
To achieve this (HP) must also fix truth-conditions for all the sentences in which
these predicates occur. So (HP) must also fix truth-conditions for sentences of
the form ‘Nx:Fx = q’ (where ‘q’ is any singular term whatsoever). But (HP)
only fixes truth-conditions for sentences of the form ‘Nx:Fx = Nx:Gx’. It does
not provide truth-conditions for sentences of any other form. Therefore, (HP)
fails to provide a basis for supposing ‘Nx:Fx = Nx:Gx’ is logically complex and
that the expressions it contains are genuine singular terms.
However, if ‘content’ means reference then the complaint is quite different.
The failure of (HP) to settle truth-values for the identities ‘Nx:Fx = q’ is
interpreted as a failure to determine a definite referent for each of the
numerical expressions introduced. It is then doubted whether expressions
of the form ‘Nx:Fx’ are referential in the first place. Benacerraf famously
propounded a related argument (see his 1965 and Kitcher 1975). In this
particular case set-theoretic terms are taken as values of ‘q’. According to
Zermelo’s set-theoretic definition of natural number, 0 is the empty set and
the successor function takes x to the unit set of x. Von Neumann defined
the natural numbers a different way: 0 is the empty set but the successor
function takes x to the union of x and the singleton of x. Benacerraf argued
that the use of arithmetical vocabulary fails to settle whether ordinary
numerals refer to the Zermelo numbers or the von Neumann numbers
(whether ‘2 = {{Ø}}’ or ‘2 = {Ø, {Ø}}’ is true). The problem is that each
of these set-theoretic progressions serves as an equally effective model of
the number theory embodied in ordinary usage. He concluded that the
semantic function of ordinary arithmetical expressions must be other than
referential.
The problem that Benacerraf isolates for ordinary numerical terms is often
assimilated to the Caesar problem itself (see, for example, Shapiro (1997:
78–81) ). However, this would be a mistake for several reasons. As we have
seen, the Caesar problem has epistemological and metaphysical dimensions
that Benacerraf’s argument fails to capture. But even if attention is focused
solely upon the meaning-theoretic Caesar problem there are other reasons to
resist the assimilation. To begin with, the Caesar problem Frege confronted
concerned the significance of expressions (‘Nx:x = x’, ‘Ny:[y = Nx:x = x]’ . . .)
artificially introduced by means of a definition (HP), whereas Benacerraf’s
problem concerns the significance of ordinary terms (‘0’, ‘1’ . . .) that may
or may not have been introduced this way. Putting this issue aside, the
‘sense’ and ‘reference’ versions of the meaning-theoretic Caesar problem
need to be kept separated. The latter species of argument does not deny the
significance of contexts of the form ‘Nx:Fx = q’ but moves from the existence
of distinct eligible referents for the same numerical terms to the conclusion
that the semantic function of these expressions (‘Nx:Fx’) cannot be referential.
By contrast, the former sort of argument moves from the failure of (HP)
to address the status of ‘Nx:Fx = q’ to doubt whether such contexts are
significant at all.
Despite the important differences that obtain between these divergent
versions of the meaning-theoretic Caesar problem they are subject to a generic
doubt. The problematic character of the relevant class of Caesar problems is
revealed in the high semantic threshold they each impose on the introduction
of genuine singular terms:
(iv) Semantic Threshold: in order to confer significance on the terms and
predicates purportedly introduced by (HP) their significance should
be everywhere fixed.
The obvious doubt to entertain here is whether these differing versions of
the Caesar problem impose too high a threshold. More or less extreme forms
of this doubt may be entertained. For example, it may be claimed that no
context of the form ‘Nx:Fx = q’ requires to have its significance fixed in order
to introduce singular numerical terms. A related view is evidenced in Carnap’s
contention that such ‘mixed’ contexts that feature both mathematical and
non-mathematical terms are actually nonsense:
2. ‘Caesar is a prime number’ . . . (2) is meaningless. ‘Prime number’ is a predicate

of numbers; it can neither be affirmed or denied of a person. . . . The fact that the
rules of grammatical syntax are not violated easily seduces one at first glance into the
erroneous opinion that one has still to do with a statement, albeit a false one. But ‘a is
a prime number’ is false iff a is divisible by a natural number different from a and from
1; evidently it is illicit to put here ‘Caesar’ for ‘a’. This example has been chosen that
the nonsense is easily detectable. . . . (Carnap 1932: 67–8)
If such mixed contexts are nonsense then it can hardly be an adequacy
constraint on the introduction of numerical terms that a meaning is fixed
for these contexts.4 A less extreme doubt will discriminate between different
sentences of the form ‘Nx:Fx = q’ where ‘q’ takes different sorts of values.
It may be that there is no need for (HP) to fix the significance of some
of these contexts in order to effect the introduction of genuine singular
terms. It may also be that genuinely significant sentences that take different
values for ‘q’ generate different obstacles for the introduction of numerical
singular terms.
Examination of (an admittedly) provisional schedule of the different values
‘q’ may take reveals the range of distinct issues involved. First, ‘q’ may take
values that in advance of a consideration of the Caesar problem we might have
taken to denote paradigmatic extra-mathematical objects (‘Nx:Fx = Caesar’).
A distinction may also be drawn between the extra-mathematical cases that
feature reference to contingent as opposed to necessary non-mathematical
entities (‘Nx:Fx = the Earth’s axis’, ‘Nx:Fx = the True’). Alternatively, ‘q’ may
take values that denote objects characteristic of the mathematical domain.
Some of these cases will feature reference to elements of progressions that
plausibly might have been taken to be mathematical, but not distinctively
numerical, progressions ( (‘Nx:Fx = {Ø}’) (Benacerraf’s examples may be
located here). Others will involve reference to elements of numerical, but not
distinctively arithmetical, series (‘Nx:Fx = 2real ’). Finally, there is the special
case where the identities in question concern the relation between the objects
denoted by the putatively arithmetical terms (HP) introduces and the objects
denoted by the numerals of ordinary arithmetic (‘Nx:Fx = 2natural ’). The
ability to settle the significance of one of these different forms may not
result in an ability to settle the significance of another. For example, we may
be able to determine that ‘Nx:Fx’ refers to a mathematical rather than a
non-mathematical object. But then we may be unable to determine whether
it refers to an item drawn from one rather than another mathematical
progression.
These are not the only cases a consideration of which may be expected
to shed light upon the significance of numerical terms. There are particular
concerns about reference generated by the special character of series that
exhibit non-trivial automorphisms (Brandom 1996). For example, in complex
4
Benacerraf makes the related suggestion that ‘identity statements make sense only in contexts
where there exist possible individuating conditions’ (see his 1965: 285–9). Resnik offers a further
alternative: that impure identities ‘should be banned from a proper language for science’ (Resnik
1997: 244).
number theory −1 has two square roots (i and −i). But there is no way to settle
within the theory which square root our use of the signs ‘i’ or ‘−i’ denotes. It
seems that we cannot settle the reference of these terms. This is because every
predicate (not containing ‘i’ or ‘−i’) that is true of one of them is true of the
other. There are also general concerns about reference that apply irrespective
of the character of the series in question (Hodes 1984: 134–5). For the sake of
argument, suppose that the series of natural numbers N have been singled
out as the referents of the ordinary numerals. Then an alternative progression
may be formed from N that serves just as well as a source of eligible referents
for ordinary numerals. To see this we need merely permute a finite number of
elements of N and make compensating adjustments to the successor function.
For example, we might let ‘4’ designate 5 and ‘5’ designate 4 and employ
‘successor’ to stand for the function that differs from the successor function
only in assigning 3 to 5, 5 to 4, and 4 to 6. Since there are indefinitely many
ways of so permuting the elements of N there is no telling which number is
referred to by a given numeral.
It is sometimes thought that all these different concerns are expressive
of the same problem, the Caesar problem. This would be a mistake. The
Caesar problem originally arose as a result of the inability of (HP) to settle
the significance of identity claims of the form ‘Nx:Fx = q’. However, the
concerns that have just been raised about the reference of numerical terms
are not occasioned by any doubt about whether some identity is meaningful
or true. In the former case they turn upon the mathematical character of
the complex number series. In the latter case, the difficulty raised does not
concern agreement or disagreement about some object language sentence.
So, for example, it is accepted that ‘4 = 4’ is true whereas ‘4 = 5’ is false.
The difficulty confronted stems from another source. It stems—figuratively
speaking—from stepping back from our own language once the truth-values
of all the sentences have been settled and then considering the myriad different
ways in which it may be reinterpreted. It follows that a happy solution to
the Caesar problem that generates truth conditions and values for sentences
of the form ‘Nx:Fx = q’ cannot be expected to help (directly) in resolving
more recherché concerns of this sort. Correlatively, the inability of a given
solution of the Caesar problem to settle issues that arise once truth-conditions
and values have been settled need cast no doubt upon the credentials of the
resolution in question qua provider of truth-conditions and values.
3. Two Solutions
A first encounter with the Caesar problem often occasions a denial, a denial
that there is any significant problem to be addressed. We are so convinced that
there is something amiss with the identification of Caesar and a number that it
often takes a good deal of theoretical orientation before it is even appreciated
that (HP) fails to secure the result that Caesar is no number. The preceding
discussion sketched in a preliminary way some of the different issues that
underlie the Caesar problem. It is to be hoped that sufficient structure has been
imposed to enable us to question the character of our pre-theoretic conviction
and see that many distinct epistemological, metaphysical, and meaning-
theoretic forces may be at work inducing the belief that Caesar is no number.
A failure to appreciate or effectively treat of its many different dimensions
undermines several proposed solutions to the Caesar problem. Two such
solutions—supervaluationism and neo-Fregeanism—will be examined here.
3.1 The Supervaluationist Solution

In its most familiar guise, supervaluationism provides a method for dealing
with the semantic phenomenon of vague predicates (Fine 1975). There
are (apparently) no sharp boundaries between the objects to which vague
predicates apply and those objects to which they do not apply. Vague
predicates have borderline cases where they neither clearly apply nor fail to
apply. Nevertheless, these predicates may be ‘precisified’: a sharp boundary
may be fixed for their application. But there are many different ways of
precisfying a vague predicate and it would be arbitrary to choose one to express
the ‘real meaning’ of the predicate. So the supervaluationist attempts to
account for the phenomenon of borderline cases by taking into account all the
possible precisifications of a vague predicate. According to the supervaluationist
account, a sentence is true iff it is true on all precisifications of its constituent
vague expressions, false iff it is false on all precisifications, and neither true nor
false iff it is true on some but not other precisifications. The supervaluationist
interprets vagueness as a species of semantic indeterminacy. Predicates do not
turn out to be vague because they apply to vague objects. They turn out to
be vague because language users have not chosen between different possible
precisifications of them.
The inability of (HP) to settle whether Caesar is a number may likewise
be interpreted as a consequence of semantic indeterminacy (cf. Field 1974;
McGee 1997; Shapiro: Chapters 5 and 7, in this volume). (HP) determines
that the numerical terms it introduces refer to numbers (so presented) but
fails to determine whether objects otherwise depicted are so picked out. This
is because (HP) is a semantically indeterminate principle. There are many
different possible precisifications of it and (HP) does not select between them.
According to some precisifications of (HP), Caesar is a number; according to
others, Caesar is not. Consequently, it is neither true nor false that Caesar is a
number. Of course, it is open to us to constrain the admissible precisifications
of (HP). We may choose to unite (HP) with the additional principle that no
Nx:Fx is a person. Then there will be no precisification upon which Caesar is
picked out by a numerical term. It will be false that Caesar is a number.
Viewed from the supervaluationist perspective Frege overreacted to the
Caesar problem. Frege interpreted the inability of (HP) to settle whether
Caesar was a number to be a symptom of an underlying malady, the failure
of (HP) to introduce referring expressions. But really what was signalled by
the inability of (HP) to settle a definite reference for number words was the
indeterminacy of the terms it introduced. To use the terminology of the
previous section, Frege simply set the semantic threshold for introducing
referential expressions too high. He required that referring expressions must
refer determinately and therefore failed to recognize that the terms introduced
by (HP) referred indeterminately.
In order to appreciate some of the problems that attend a supervaluationist
account it is useful to consider a more simple and direct approach to the
Caesar problem. When reflecting upon the inability of (HP) to determine a
truth-value for the sentence ‘Julius Caesar is the number of planets’, Dummett
once suggested that the difficulty could be swept aside with ease. He wrote: ‘it
would be straightforward to provide by direct stipulation for the falsity of such
sentences’ (see his 1967: 111).5 There are two relevant difficulties associated
with this proposal.
First, it is unclear how a direct stipulation that sufficed for the falsity of all
such sentences might be constructed. Of course, it is true that no numerical
thing is a non-numerical thing. But this doesn’t need a stipulation to make it
5
It should be noted that Dummett’s views have evolved since 1967. Nevertheless, he appears to
remain at least willing to countenance the possibility that the Caesar problem is the consequence
of semantic indeterminacy. In the course of a discussion of Benacerraf he writes: ‘the received
senses of numerical terms do not impose any one specific identification of the natural numbers’
(Dummett 1990: 179).
so. Moreover, it must not be forgotten that what is at issue is the reference
of numerical terms and the application conditions of numerical predicates.
So this truth—whose sentential expression makes play with the relevant
class of problematic vocabulary the reference and application of which are in
question—provides no guide to whether a given sentence is true or false as a
consequence of it. This suggests that it might be better to proceed piecemeal,
providing a range of stipulations to distinguish numbers from different sorts
of non-numerical objects. But, as we have seen, it is no easy matter to settle
whether numbers do or do not possess a feature—for example, contingency
or concreteness—characteristic of a given sort (Section 2.4, above).
This first difficulty is a clue to the second. Suppose that Caesar leads a
double life. Suppose that in addition to leading his material existence Caesar is
also a number. In that case the stipulation that sentences that say Caesar is a
number are all false cannot succeed. For some of these sentences will be true
and true sentences cannot be stipulated to be false. So Dummett’s strategy of
directly stipulating the falsity of the relevant range of sentences presupposes
that Caesar is no number. Stipulation cannot suffice as a basis for determining
that Caesar is no number.
A similar difficulty afflicts the more sophisticated supervaluationist strategy.
According to the supervalutionist, the inability of (HP) to settle the truth-value
of such sentences as ‘Julius Caesar is the number of planets’ is a consequence of
semantic indeterminacy. In other words, (HP) may be precisified in a number
of arbitrary ways and some of these precisifications make such sentences true,
others make them false. But this assumes that it is legitimate to precisify
the numerical vocabulary introduced in such a way as make it false that
Caesar is a number. But suppose again that Caesar is a number. Suppose
that facts about his identity and distinctness from every other number are
determined by facts about 1-1 correspondences between concepts. Then there
are no precisifications upon which Caesar is not a number. Any attempt to
precisify (HP) in this way will conflict with underlying metaphysical facts.
Consequently, it will be inappropriate to apply the semantic machinery of
supervaluationism to explicate the apparent indeterminacy of the sentences
in question. It will not be the case that there are some precisifications upon
which these sentences are true and some precisifications upon which they are
false. The same difficulty will attend the attempt to precisify (HP) by adjoining
further stipulations (for example, that no number is contingent).
What this reveals is that both the direct stipulation strategy and super-
valuationism rest upon a common assumption, the assumption that (HP) is
semantically indeterminate. They are motivated by the idea that the vocab-
ulary introduced by (HP) is semantically neutral in the sense that it enjoins
no commitments that might conflict with antecedent facts. It is because, they
presume, the introduced vocabulary is free of such commitments that it is
possible to stipulate or precisify its use without there being any risk of offending
against any antecedent fact. If, however, the vocabulary introduced fails to
be neutral in this regard—if the use stipulated for it may conflict with ante-
cedent facts—then the mechanisms employed to resolve the indeterminacy
will misfire. It follows that we can have no assurance that direct stipulation or
supervaluationism succeed in resolving the Caesar problem in the absence of
an argument that (HP) is semantically indeterminate. But neither purported
solution shows this. They suppose (HP) is semantically indeterminate and
then seek to accommodate that happenstance.
The point deserves emphasis. (HP) fails to explicitly address the significance
of identity contexts of the form ‘Nx:Fx = q’. The direct stipulation and super-
valuationist strategies construe this silence to be a symptom of the semantic
indeterminacy of the terms (HP) introduces. But the silence of (HP) may be inter-
preted differently. It may be taken to reflect the epistemological inscrutability
of impure identities, the fact that we can just never know whether Caesar is
a number. Alternatively—and these do not exhaust the alternatives—the
silence may betoken the meaninglessness of impure identities, the failure of (HP)
to fix any significance for identities of this form. In the former case, the terms
introduced have a definite (albeit unknown) reference. In the latter case, the
terms have no sense and do not refer—not even indefinitely. In either case
the imposition of a supervaluationist semantics upon contexts of the form
‘Nx:Fx = q’ will fail to remedy the underlying malady.
One of the most significant tasks facing any resolution of the Caesar
problem is that of determining the character of the problem itself. Call this
the circumscription problem, the problem of circumscribing the character of the
problem that demands resolution. The direct stipulation and supervaluationist
strategies fail to address this problem. They assume rather than show that the
Caesar problem has a certain character (semantic indeterminacy). As a result
the direct stipulation and supervaluationist strategies fail to resolve the Caesar
problem.
3.2 The Neo-Fregean Solution
By contrast to the supervaluationist, the neo-Fregean proposes a solution to
the Caesar problem that relies upon a distinctive ‘philosophical ontology’
(Hale and Wright 2001: 385–96).6 The solution is framed in the context of a
theory of categories. A category is a collection of objects that share a common
criterion of identity. Different categories are distinguished by the different
criteria of identity associated with them. Now consider the possibility that
an object belonging to one category is identical to an object drawn from
another. The neo-Fregean claims that such trans-sortal identifications possess
a distinctive epistemological feature: ‘there is simply no provision for or
against such identities’; there are no encompassing identity criteria available
that would allow us to settle whether objects drawn from distinct categories
are the same or different (Hale and Wright 2001: 394).
On the basis of the claim that trans-categorical are evidence transcendent
in this way the neo-Fregean presents a dilemma. Either it is granted that there
are true trans-categorical identities or it is not. If there are such identities then
Frege may be convicted of overestimating the gravity of the Caesar problem. For
this so-called ‘problem’ arises from the inability of (HP) to settle a statement
concerning the identity of objects drawn from different categories (persons
and numbers). But if it is a general truth that such identities cannot be settled
then it can signal no defect in (HP) that it fails to settle a truth-value for
impure identity claims. Alternatively, it may be denied that there are any true
trans-categorical identities. But then Frege may be convicted of underestimating
the capacity of (HP) to solve the Caesar problem. For the criteria of identity
that (HP) stipulates for numbers are distinct from the characteristic criteria
of personal identity (whatever package of psychological and bodily conditions
that might be). Therefore persons and numbers belong to different categories
and this fact suffices for their numerical difference. So either the Caesar
problem dissolves or it is solved. Either way, Frege failed to show that (HP)
provided a defective mechanism for introducing numbers.
6
An earlier neo-Fregean treatment proposed to solve the Caesar problem by demonstrating
that—despite Frege’s proclamation to the contrary—(HP) does suffice for the falsity of impure
identity contexts. This result was to be achieved by appealing to differences between the types of
consideration that serve as canonical grounds for determining facts of personal and numerical
identity (Wright 1983: 107–17). In response to criticisms by Rosen (1993: 171–4) and Sullivan
and Potter (1997), the neo-Fregean adopted the position described above (Hale and Wright 2001:
367–85). See MacBride (2003: 128–35) for an overview and assessment of the debate.
Thedetailsoftheneo-Fregeanviewareclearlyopentoquestion.Consider,for
example, the neo-Fregean claim that trans-categorical identities are evidence
transcendent. This claim is motivated by the reflection that if it is legitimate
to countenance the identity of Caesar with a number then it is equally
legitimate to countenance the identity of, say, Frege with a Roman statue (an
object drawn from another distinct category). They both constitute cases of
bare, imponderable identities (Hale and Wright 2001: 394). But if persons and
artefacts turn out to fall under a common category—the category of physical
object—then it remains opens that the facts of identity and distinctness
amongst persons and artefacts may be settled by verifiable considerations (for
example, spatio-temporal duration and location). Clearly, the relevant notion
of category requires greater development before the neo-Fregean proposal can
be properly assessed.
The claim that all trans-categorical identities share an evidential status may
be questioned for another reason. Suppose Frege has mass m and the Roman
statue mass n > m. Then it is natural to reason in the following way: nothing
can have mass of both n and n > m values; so Frege and the statue cannot
be identical. If we are to countenance the possibility that an arbitrary object
(Frege) is really identical to an object located at another place (at the same
time) with different intrinsic properties then this line of reasoning will have
to be shown to be somehow at fault. I have described this elsewhere as the
‘problem of spatial intrinsics’ (by analogy with the more familiar problem of
temporary intrinsics; see MacBride (1998: 223–7) for further details). To solve
this problem we must give up some ordinary assumptions about intrinsic
property possession. We will have to give up the assumption that intrinsic
properties (mass, shape, etc.) are possessed simpliciter, that is, independently of
spatial location. Instead we will have to think something of the following sort:
intrinsic properties are possessed relative to spaces (and perhaps times too). And
then there will be no incompatibility generated by Frege possessing mass and
shape relative to the slice of space–time carved out by his life, and, another
mass and shape relative to the duration and location of the Roman statue.
What this suggests is a surprising result. Far less damage is done to our ordinary
ways of thinking by countenancing the possibility of an identity between
numbers and persons than by seriously entertaining the idea that different
sorts of physical objects might be identical. By contrast to the latter, the
identification of numbers and persons does not force any revision or particular
view of the way in which intrinsic properties are possessed. This reflects once
again a beguiling aspect of the Caesar problem noted earlier: the fact that there
is no overt incompatibility between being a number and being a person.
Of course, these are considerations of detail that may very well be addressed in
the context of a fuller development of the neo-Fregean approach. Nevertheless,
it is worthy of note that there are such details to be negotiated. For the neo-
Fregean intends their philosophical ontology to be placed at the service of a
logicist philosophy of mathematics. The greater the metaphysical depths the
neo-Fregean must fathom to make good their claims the less likely it appears
that the relevant theory of categories should draw on merely logical concepts
and techniques for its expression.
Independently of such considerations how well does the neo-Fregean
solution fare with respect to the critical task of negotiating the various
different aspects of the Caesar problem? How does the neo-Fregean solution
fare with respect to the circumscription problem? Unlike the supervaluationist
the neo-Fregeans do not assume that contexts of the form ‘Nx:Fx = q’ are
semantically indeterminate. But they do make the contrasting assumption
that such contexts are meaningful and determinate. Having made that
assumption the neo-Fregean then sets about demonstrating that either the
truth-values of impure contexts are imponderable or settled by category-
theoretic consideration. Recall, however, the version of the Caesar problem
that denied (HP) supplied impure identities with any meaning whatsoever
(Section 2.5, above). In that case, contrary to the neo-Fregean solution, impure
identities cannot have any sort of truth-value, unfathomable or otherwise.
Unfortunately, the neo-Fregean fails to address the circumscription problem. As a
result the neo-Fregean fails to provide a solution to the Caesar problem.
However, the neo-Fregean does offer two arguments to undermine the
contention that one might rest content with the situation that Carnap
was willing to tolerate—the situation where impure identities lack a truth-
condition or value (Hale and Wright 2001: 340–5).7 One argument operates
at the level of understanding and appeals to Evans’s Generality Constraint (Evans
1982: 100–5). Construed as a linguistic principle this constraint exercises a
control on the understanding of sentences. To understand an expression is
7
It is Heck (1997), rather than Carnap, that provides the immediate target of Hale and Wright’s
arguments. Heck argues that a Fregean-style proof on the infinity of the number series may
be carried out in the context of a many-sorted logic. He refrains, however, from endorsing the
view that the kinds of Caesar problem under consideration here may be resolved by adopting a
many-sorted logic.
to understand the contribution it makes to the meaning of all the significant
sentential contexts in which it occurs. So a subject may only grasp a particular
sentence if he or she grasps the range of significant sentences that result
from the permutation of understood constituents. Consider the possibility
currently at issue: that a subject may understand a range of pure numerical
identities (‘Nx:Fx = Nx:Gx’) and pure personal identities (‘Caesar = Julius’)
but fail to comprehend the significance of mixed identities (‘Nx:Fx = Caesar’).
The Generality Constraint appears to rule this possibility out. For if the subject
‘fully understands’ the pure identities then they must also understand the
sentences that result from the permutation of the constituent terms ‘Nx:Fx’,
‘Nx:Gx’, ‘Caesar’, ‘Julius’, and ‘. . . = . . .’. But impure identities occur amongst
the results of such a permutation. So it cannot be the case that (HP) serves to
fix the significance of pure numerical contexts whilst neglecting entirely the
significance of impure contexts.
This argument is open to question. First, maintaining the discussion at the
level of understanding, it may be granted that having a full understanding
of pure identities requires that a subject must be able to understand all the
significant permutations of them—impure identities included. But suppose
that (HP) provides only a partial understanding of pure numerical identities.
In that case there need be no conflict with the Generality Constraint. For a
subject whose understanding of pure numerical identities is mediated by (HP)
and yet fails to comprehend the significance of impure identities need be a
subject with only a partial understanding of the significance of pure identities.
Second, the argument stands in need of qualification. The Generality Constraint
does not require that a subject understand the range of grammatical sentences
that result from the permutation of understood constituents. It requires only
that a subject understand the resulting range of significant sentences. Suppose
impure identities even though grammatical are meaningless. It follows that
the Generality Constraint fails to rule out the possibility of understanding pure
but not impure identities.
The neo-Fregean argues however that impure identities are meaningful
and so this possibility does come into conflict with the Generality Constraint.
Hale and Wright declare ‘the thought dies hard that identity is categorically
appropriate simply to any object’ and offer two considerations in favour of
the contended significance of impure identities (2001: 344, 350–1). The first
consideration incorporates an appeal to the contrapositive of Leibniz’s Law
(‘the diversity of the dissimilar’). Suppose that Caesar and Nx:x = x belong
to mutually exclusive categories (persons and numbers). Then Caesar is a
person whereas Nx:x = x is not. Since Caesar has a property (being a person)
that Nx:x = x lacks it follows that they are distinct. So the relevant impure
identity (‘Caesar = Nx:x = x’) is false rather than meaningless. However,
this argument presupposes that claims of trans-categorical distinctness are
themselves meaningful and capable of receiving a truth-value. But if trans-
categorical identity statements are meaningless then so are trans-categorical
distinctness statements. So the argument from the diversity of the dissimilar
begs the question against the view that impure identities are meaningless.
The neo-Fregean therefore offers a second consideration. Hale and Wright
argue that it is ‘utterly unclear’ how a case may be made for the claim
that impure identities lack a sense. An appeal to intuitions of significance is
likely to be unsatisfactory because our intuitions do not speak in unison. An
appeal to stipulation is also out of order. They conclude: ‘what is needed
is a principled reason for denying that this configuration of individually
significant words adds up to an expression which, taken as a whole, expresses
something true or false’ (Hale and Wright 2001: 351). But this argument also
appears to beg the question. The meaning-theoretic version of the Caesar
problem arises from the fact that (HP) fails to settle any significance for
impure identities—it leaves a semantic gap there. This provides a principled
reasoning for doubting whether impure identities do express something. Of
course, if the default assumption is made that identity contexts exhibit a free
wheeling compositionality—so every grammatical permutation of them is
meaningful—then the difficulties encountered in settling the significance of
impure identities provide no grounds for doubting that they have a meaning.
They will have a meaning regardless. But since the Caesar problem raises the
question of whether every grammatical identity has a sense the assumption of
free wheeling compositionality can hardly be relied upon to bolster a solution
to the problem itself.
The neo-Fregean also supplies a metaphysical argument to undermine the
contention that impure identities are meaningless. Appeal is made to Frege’s
avowed ‘Platonism’: the contention that numbers belong to an inclusive
domain of objects.8 The neo-Fregean reasons that if numbers are to belong to
8
A further argument the neo-Fregean might plausibly employ here appeals to the fact that
we appear (at least) to be able to simultaneously count objects drawn from distinct categories.
For example, it seems intelligible to ask: how many natural numbers less than ten and planets
such a domain then there must be a fact of the matter about which objects the
numbers are. In other words, there must be a determinate truth about whether
a number is identical or distinct to any other object (however presented). The
neo-Fregean concludes that if Platonism is to be a legitimate position then
it cannot be the case the impure identities are meaningless. But if there is
any legitimacy to the concern that—for all (HP) settles—impure contexts
lack a sense then this argument simply places a question mark over whether
Platonism is a legitimate position. This in turn raises a doubt concerning
the effectiveness of any purported solution to the Caesar problem—the
neo-Fregean solution included—that presupposes Platonism.
4. Conclusion
LetusreturntoFrege’soriginalformulationoftheCaesarproblem(Section 2.1).
Frege linked the notions of object and identity and then claimed that if a symbol
a is to be used to denote an object then we must have available a cri-
terion that determines in every case whether b is the same as a. We have
seen that considerable difficulties confront any attempt to supply such a
criterion. This suggests that it may be time to reconsider whether Frege
was right to tie together the notions of object and identity in the manner he
proposed.
It is true that objects may be identified and re-identified and seen from
different perspectives. No doubt it is the capacity of objects to be identified and
re-identified in this way that is responsible for our treating discourse about
objects in a realist fashion. But there does not appear to be anything in our
ordinary interaction with objects that determines objects must be capable of
being identified and re-identified from every point of view. Rather it appears
that we ‘track’ objects across a range of relevant situations and perspectives. It
would be an imposition to suppose that the ‘tracking conditions’ with which
we habitually operate are identity criteria that tacitly determine the presence
or absence of an object across all situations. One may therefore wonder whether
of the solar system are there? This question appears to presuppose that there are facts of the
matter concerning the identity and distinctness of numbers and planets. But appearances may be
deceptive. Perhaps the question should be interpreted as asking for the result of the addition of the
number of numbers less than ten and the number of the planets. Then no facts of trans-categorical
identity will be called upon to settle the answer to the question.
Frege made any legitimate demand when he required identity criteria for the
objects he planned to introduce.
The doctrine that the notions of object and identity cannot be separated
has, however, become deeply entrenched. It is therefore unlikely that this
suggestion will be readily received. But there is something perplexing about
this persistent adherence to Fregean doctrine. For, despite the conviction that
every object has identity criteria, proponents of the view have been beggared
to provide any. This goes for all kinds of objects, all the way up from quantum
particles to persons. Even sets—often presented as the paradigm of objects
with clear and distinct identity criteria—fail to meet the prevailing standards.
For the Axiom of Extensionality that purportedly provides criteria of identity
for sets neglects to make any mention of times or possible worlds. As a
consequence Extensionality fails to determine whether sets are identical or
distinct at different times or different worlds. So even someone who submits
that there are well-defined identity criteria for sets must overcome a version
of the Caesar problem. Serious engagement with the Caesar problem and its
presuppositions may well assist in identifying and assessing the conflicting
intellectual forces that give rise to this perplexing situation.
References
Benacerraf, P. (1965), ‘What Numbers Could Not Be’, Philosophical Review, 74: 47–73;
repr. in P. Benacerraf and H. Putnam (eds.), Philosophy of Mathematics: Selected Readings,
(2nd edn., Cambridge: CUP, 1983), 272–94.
Boolos, G. (1987), ‘The Consistency of Frege’s Foundations of Arithmetic’ in
J. J. Thomson (ed.), On Being and Saying: Essays for Richard Cartwright (Cambridge,
Mass.: MIT Press), 3–20.
Brandom, R. (1996), ‘The Significance of Complex Numbers for Frege’s Philosophy of
Mathematics’, Proceedings of the Aristotelian Society, 96: 293–325.
Burgess, J., and Rosen, G. (1997), A Subject with No Object: Strategies for Nominalistic
Interpretation of Mathematics (Oxford: OUP).
Carnap, R. (1932), ‘Überwindung der Metaphysik durch logische Amalyse der
Sprache’, Erkenntnis, 2: 219–41; repr. as ‘The Elimination of Metaphysics through
Logical Analysis of Language’, in A. J. Ayer (ed.), Logical Positivism (London: Allen &
Unwin, 1959), 60–81.
Dummett, M. (1967), ‘Frege’s Philosophy’, in P. Edwards (ed.), The Encyclopaedia of
Philosophy (New York: MacMillan), 225–37; repr. in Dummett, Truth & Other Enigmas
(London: Duckworth, 1978), 87–115.
(1990), Frege: Philosophy of Mathematics (London: Duckworth).
Evans, G. (1982), The Varieties of Reference (Oxford: OUP).
Field, H. (1974), ‘Quine and the Correspondence Theory’, Philosophical Review, 83:
200–28.
(1980), Science without Numbers (Oxford: Blackwell).
(1993), ‘The Conceptual Contingency of Mathematical Objects’, Mind, 102:
285–99.
Fine, K. (1975), ‘Vagueness, Truth and Logic’, Synthese, 30: 265–300.
Frege, G. (1950), The Foundations of Arithmetic, trans. J. L. Austin (Oxford: Blackwell).
(1964), The Basic Laws of Arithmetic, trans. and ed. M. Furth (Berkeley and Los
Angeles: University of California Press).
Hale, B., and Wright, C. (1992), ‘Nominalism and the Contingency of Abstract
Objects’, Journal of Philosophy, 89: 111–35.
(1994), ‘A reductio absurdum: Field on the Contingency of Mathematical
Objects’, Mind, 103: 169–84.
(2001), ‘To Bury Caesar . . . ’ in their The Reason’s Proper Study (Oxford: OUP, 2001),
335–96.
Heck, R. (1997), ‘The Julius Caesar Objection’, in Heck (ed.), Language, Thought and Logic
(Oxford: OUP, 1997), 273–308.
Hodes, H. (1984), ‘Logicism and the Ontological Commitments of Arithmetic’, Journal
of Philosophy, 81: 123–49.
Kitcher, P. (1975), ‘The Plight of the Platonist’, Noûs, 12: 119–36.
Leibniz, G. W. (1981), New Essays on Human Understanding, trans. and ed. P. Remnant and
J. Bennett (Cambridge: CUP).
Linsky, B., and Zalta, E. (1994), ‘In Defence of the Simplest Quantified Logic’,
Philosophical Perspectives, 8: 432–58.
(1996), ‘In Defence of the Contingently Nonconcrete’, Philosophical Studies, 84:
283–94.
MacBride, F. (1998), ‘Where are Particulars and Universals?’, Dialectica, 52: 203–27.
(1999), ‘Listening to Fictions: A Study of Fieldian Nominalism’, British Journal for
the Philosophy of Science, 50: 431–55.
(2000), ‘On Finite Hume’, Philosophia Mathematica, 8/2: 150–9.
(2002), ‘Could Nothing Matter?’, Analysis, 62: 125–35.
(2003), ‘Speaking with Shadows: A Survey of Neo-Logicisim’, British Journal for the
(2005), ‘The Julio Cesar Problem’, Dialectica, 59: 223–36.
McGee, V. (1997), ‘How We Learn Mathematical Language’, Philosophical Review, 106:
35–68.
Parsons, C. (1964), ‘Frege’s Theory of Number’, in M. Black (ed.), Philosophy in America
(London: Allen & Unwin), 180–203; repr. in W. Demopoulos (ed.), Frege’s Philosophy
of Mathematics (Cambridge, Mass.: Harvard University Press, 1995), 182–210.
(1990), ‘The Structuralist View of Mathematical Objects’, Synthese, 84: 303–46.
Resnik, M. (1997), Mathematics as a Science of Patterns (Oxford: OUP).
Rosen, G. (1993), ‘The Refutation of Nominalism (?)’, Philosophical Topics, 21: 149–86.
Russell, B. (1911), ‘On the Relation of Sense-Data to Physics’, in his Mysticism and Logic
(New York: Longmans, Green & Co), 139–70.
Shapiro, S. (1997), Philosophy of Mathematics: Structure and Ontology (Oxford: OUP).
Sullivan, P., and Potter, M. (1997), ‘Hale on Caesar’, Philosophia Mathematica, 5: 135–53.
Williamson, T. (1998), ‘Bare Possibilia’, Erkenntnis, 48: 257–73.
Wright, C. (1983), Frege’s Conception of Numbers as Objects (Aberdeen: Aberdeen University
Press).
9
Sortals and the Binding Problem
John Campbell
1. The Delineation Thesis

Philosophers have sometimes maintained very strong theses of sortal depend-
ence; theses to the effect that singular reference depends on knowledge of the
sort of the object to which one is referring. They have held that the ability to
single out in experience a tree, or a mountain, or a lion depends on grasp of
such sortal concepts as ‘tree’, ‘mountain’, or ‘lion’. Their basic point is that
there can be different things of different kinds in the same place at the same
time. Suppose we consider a bend in the river. What object is there, at that
point? There is, of course, the river, which continues on downstream. There
is also a stage of a river: that is, the river as it is at this particular moment, for
example. A ‘river-stage’ so defined cannot last longer than a moment, unlike
the river. There is also a collection of water molecules at that point. This
collection will have been diffused in the sea by tomorrow, unlike the river itself.
There is also a stage of that collection of water molecules, at the same place:
a momentary stage in the life of that collection of water molecules. Quine
(1953) gave the basic argument here when he said:
Thanks to Rae Langton for a most helpful reply when an earlier version of this paper was presented
at the Arche meeting, and to the participants for discussion. Thanks also to David Wiggins, and
to the participants at an Oriel group meeting at which I presented another early draft. I have also
benefited from remarks by an anonymous referee. This paper is excerpted from the discussion of
these topics in my Reference and Consciousness (Oxford: OUP, 2002).
204 / Sortals and the Binding Problem
Pointing is of itself ambiguous as to the temporal spread of the indicated object. Even
given that the indicated object is to be a process with considerable temporal spread,
and hence a summation of momentary objects, still pointing does not tell us which
summation of momentary objects is intended, beyond the fact that the momentary
object at hand is to be in the desired summation. Pointing . . . could be interpreted
either as referring to the river . . . or as referring to the [collection of water-molecules]
. . . , or as referring to any one of an unlimited number of less natural summations
. . . such ambiguity is commonly resolved by accompanying the pointing with such
words as ‘this river’, thus appealing to a prior concept of a river as one distinctive type
of time-consuming process. (Quine 1953: 67)
The argument here is explicitly about pointing, which is a way of orienting

attention—it is one of the control mechanisms of attention—rather than
conscious attention itself. And you might argue—indeed, in this chapter, I
will argue—that Quine’s argument should be read as an argument about
the control of attention, rather than about what is involved in consciously
singling out an object. But that is not the natural reading of the significance
of Quine’s comment. On the natural reading, Quine’s comments imply that
if you are to single out an object in experience, you have to be employing a
relevant sortal concept, to delineate the boundaries of the object singled out.
The idea is that that sortal concepts play, as it were, a top-down role in making
it possible to pick out objects in experience. We could put the idea in terms of
what I will call the Delineation Thesis:
The Delineation Thesis: Conscious attention to an object has to be focused by
the use of a sortal concept which delineates the boundaries of the object
to which you are attending.
It is, indeed, very often assumed that if you are to attend to Fs, you must already
have the concept of an F (cf., e.g., Fodor 1998). But straight off, that is hard to
believe. Of course, you cannot be intentionally attending to Fs without having
the concept of an F; but for all that, you can be attending to an F without
knowing that it is an F and without having that concept. It is a fundamental
point that demonstrative reference to an object can succeed even though you
do not know what sort of thing you are dealing with. In a garden centre I
exclaimed at the beauty of a particular plant, only to be told that it was plastic.
There was no relevant ambiguity about which thing I was talking about. I
had plainly referred to the plastic plant. But a plastic plant is not a plant, any
more than a fake Rembrandt is a Rembrandt. You can succeed in singling out
John Campbell / 205
an object despite a mistaken belief about what sort of thing it is. A still more
radical case is the case in which you succeed in singling out the object even
though you have not the slightest idea what sort of thing it is. Suppose that
by some chance, an ordinary teacup from today survives for a thousand years.
And suppose that when it is discovered by our descendants then, they and their
circumstances have changed so dramatically that they could have considerable
conceptual difficulty in understanding what a teacup is. To grasp the concept
they would need to have a lot of background filled in of which they know
nothing as yet. But the discovered teacup prompts a lot of discussion. There
are learned conferences speculating as to its nature. A strong body of opinion
leans to the view that it ‘probably had some religious significance’. It is kept
in a glass case in a museum. Given the intense discussion it receives, it would
be absurd to say that our descendants have not managed to ‘single it out’.
They can certainly make demonstrative reference to it; their experience of
the object is sufficient to specify uniquely which thing is in question. In both
these cases, the plant and the teacup, what is happening is that your visual
system is managing to bind together the information from a single thing, and
you are consequently able to attend consciously to it, even though you have
not managed to apply the right sortal concept to it. Application of the correct
sortal concept seems therefore to be a more sophisticated phenomenon than
conscious attention to the object; the Delineation Thesis is simply false.
In this essay I want to go further into the possibility that the principles
used by the visual system to bind together all the information derived from a
single object may, in effect, play something of the role traditionally given to
sortal concepts. These principles achieve that effect, I will argue, by making it
possible to attend visually to an object even in the absence of knowledge as to
what sort of thing it is. It is the principles used by the visual system to bind
together information as all true of a single object that allow the subject to
delineate the object in experience. And it is your capacity for conscious visual
attention to the object that provides you with knowledge of the reference of a
demonstrative referring to that thing.
2. The Binding Problem

According to Quine, the involvement of the sortal concept is needed for there
to be a determinate answer to the question: ‘to which object is the subject
consciously attending?’. If we do not appeal to the subject’s grasp of a sortal
concept, how could we say what the difference is between attending to a river
and attending to a collection of water molecules, for example?
Different styles of conscious attention will be used in attending to different
sorts of object. For example, if you are consciously attending to a person over
a period of time, the way in which you keep track of that person will be quite
different to the way in which you keep track of a valley to which you are
attending over a period of time. In consciously attending to the person you
will, for example, keep track of the movements of that person; but in the case
of the valley you will make no such allowance for movement (certainly none
for movement with respect to the flanking mountains).
I want to propose that what underpins these differences in style of visual
attention is not a difference in which sortal concept the subject is using, but a
difference in the style of binding that the visual system is using. The binding
problem is the problem the visual system has of putting together different
features as features of a single object. The visual system processes different
characteristics of objects in different specialized processing streams. Shape,
size, colour, and motion, for example, each have processing streams devoted
to them. Since visual information-processing involves computation of the
object’s various characteristics in separate processing streams, the visual
system has the problem of putting together all the information, in various
processing streams, that relates to one and the same object. Although different
areas in the visual cortex are used to process different features of objects, such
as motion and depth, colour and shape, or location, there is no brain area
where the firing of neurons corresponds to the objects we see. So how are
different features put together as features of a single object?
Here is one way it could happen. Suppose that in each processing stream, not
only is there information relating to the kind of feature in question—shape
or motion or whatever—but also, information about where the feature is
located. If all the processing streams carry location information as well as
information relating to the specific property being processed, this gives the
beginning of a solution to the binding problem: features found at the same
location can be put together as features of one and the same object. The
evidence both from cognitive studies, and from physiology, is that something
like this solution is indeed what our visual system uses (cf. Treisman 1996;
Zeki 1993). Of course location could not be the only criterion used: if you
think of a hand grasping a glass, for example, the parts of the hand are
John Campbell / 207
no closer to one another than they are to the glass, but the hand and the
glass are visibly different things. This means that location cannot be the
only criterion used, and the natural suggestion is that the appeal to location
must be supplemented with an appeal to something like Gestalt grouping
principles in putting together the features of the hand as features of a single
object, and the features of the glass as features of a single object. A keen
sortalist might say that the visual system has to use processes which we can
as theorists characterize using sortal concepts: that the visual system uses
location plus the classification ‘hand’, or location plus the classification ‘glass’,
in solving the binding problem. But there are two problems with this. First,
such empirical evidence as there is does seem to suggest that it is general
Gestalt principles, rather than specific sortal classifications, that are used in
binding (Palmer and Rock 1994; Prinzmetal 1995). Secondly, as we shall see
in a moment, the evidence from neuropsychology is that sortal classification
is a more sophisticated phenomenon than visual binding; visual binding can
take place without sortal classification, but sortal classification in the absence
of binding seems impossible. In any case, an appeal to sortal concepts as part
of an empirical hypothesis about the way in which the visual system works
is not what philosophers have traditionally had in mind when they appealed
to sortal concepts; the suggestion was rather that an explicit grasp of sortal
concepts by the subject was essential to the delineation of perceived objects.
In summary, we can think of visual binding as exploiting the location
of the object together with something like the Gestalt organization of the
characteristics found at that location. And, course, the visual system can bind
together features over time, as when we keep track of a moving object, and
across sensory modality, as when we assign heard speech to the person seen
before us. This is a more primitive phenomenon, involving a more primitive
level of information-processing content, than the application of the subject’s
grasp of sortal concepts.
I would argue, furthermore, that we should think of the complex binding
parameter used in vision as playing a role also in the subject’s conscious
attention to the object. When the subject consciously attends to an object with
the aim of finding out more about it, information about the object has to be
selected from the various specialized processing streams in the visual system;
as I said, there seems to be no point at which all and only the information
relating to a single object is put together in the visual system. The particular
value of the binding parameters—location together with the relevant Gestalt
principles—in effect provides a way of identifying the object which can be used
in recruiting information about that object in various processing streams, or to
single out the object for the purpose of acting on the object. So the singling-out
of an object in experience need not involve the application of sortal concepts;
only the mechanisms of binding. Whether you are consciously attending to
a river or a mass of molecules, for example, will show up in how your visual
system binds together information from the thing over time. If you have to
keep moving downstream to keep track of the object of your attention, then
you are attending to a collection of water molecules rather than a river. If, on
the other hand, you are binding together information from any point in the
course of the river, as all relating to a single object, then you are attending to
the river itself. The distinction between consciously attending to a collection of
water molecules, and consciously attending to a river, is not particularly hard
to draw, even without appealing to any grasp of sortal concepts by the subject.
These differences in style of attention amount to differences in what I called
the complex binding parameter used by the visual system in putting together
the information true of the object. The binding parameter for perception of
a person will have to allow for the possibility of movement by the person;
the binding parameter for a valley will not have to allow for any possibility
of movement by the valley. As I said, the complex binding parameter in effect
provides an address for the object, by which it can be identified, at the level
of conscious attention, in a way that can be used in recruiting information
from various processing streams to allow verification of propositions about
the object, and action on the object. So the style of conscious attention to the
object that is appropriate will depend on what sort of object is in question.
The use of one style of conscious attention rather than another—that is,
the use of one type of complex binding parameter rather than another—is
nonetheless a more primitive phenomenon than the ability to use sortal
concepts to classify the objects to which you can attend. Animals other than
humans plainly have a repertoire of binding strategies available to them: a cat
keeping track of a mouse is plainly using different binding strategies than a
cat keeping track of its home. It seems evident, also, that you could be using
various styles of conscious attention in keeping track of various of the things
around you in a new environment, for example, without yet having learned
what sorts of things any of them are.
Philosophers interested in the fact that we can make demonstrative refer-
ence to various quite different sorts of physical object have frequently proposed
John Campbell / 209
that our understanding of sortal concepts has a foundational role to play in
making it possible for us to refer to objects of these various sorts. What I am
proposing in this essay, though, is that grasp of sortal concepts is a more soph-
isticated matter than is the mere capacity for demonstrative reference. What
is important, in our capacity for demonstrative reference to different sorts of
object, is our capacity to engage in different styles of conscious attention. And
indeed, that capacity to engage in different styles of conscious attention is what
plays the roles that have so often been assigned to our grasp of sortal concepts.
Another way to put the point is in terms of the classical distinction between
‘associative’ and ‘apperceptive’ visual agnosias (Lissauer 1890; for discussion
see Farah 2000). On the one hand, according to the classical distinction, there
are ‘associative’ agnosias in which the patient sees the object, but does not
recognize which object he is seeing—so the patient may be able to give quite a
full description of the volumetric properties of, for example, a glove, but be at
a loss to say what such a thing might be used for (cf. Humphreys and Riddoch
1987 on ‘semantic access agnosia’). A patient of this type seems perfectly capable
of using and understanding perceptual demonstratives. Such patients seems
to be plain counterexamples to the Delineation Thesis; there is no question but
that they have singled out the object in experience, though they are unable to
give a semantic classification of it. The patient is even, in this kind of case, able
to find what sort of thing it is that he is looking at; his understanding of the
demonstrative means that he has the resources to understand a demonstration
that the object is of this or that sort; it may be only the capacity for specifically
visual recognition of the sort of the object that is impaired. What makes it so
compelling that the patient is able to understand visual demonstratives is that
his situation is, after all, not so very different from that of an ordinary subject
given an unfamiliar view of an object of a familiar type, or a view of an object of
a sort he has never seen before. In those cases vision alone does not allow you to
classify the object. But it is nonetheless compelling that in those cases you are
still able to identify the object demonstratively, and to formulate hypotheses
about that thing, or plans to act on it thus and so. Consider, in contrast, the
so-called ‘apperceptive’ agnosics. An agnosic of this type may be able to copy
complex figures, even though he insists that he has no idea what he is drawing.
The patient will have no idea which objects there are to be found in the scene
he has successfully copied, even though that is evident to an ordinary subject
looking at the copy. Humphreys and Riddoch (1987) suggest that what is wrong
here is, in effect, a problem with binding (cf. their discussion of ‘integrative
agnosia’). The patient is able to see the simple components of the scene and
where they are; it is just that there is a difficulty with visual integration. When
such an agnosic views a scene, is he in a position to understand demonstrative
reference to the object before him? It seems evident that he is not. He cannot,
on the basis of vision, verify propositions about the object he is seeing; he is in
no position to act on the object, and most fundamentally, his experience does
not provide him with knowledge of which object is in question. It thus seems
that for experience of an object to provide you with knowledge of the reference
of a demonstrative referring to that object, there must be sufficient integration
of the object in experience—the various features or parts of the object must
provide experience of it as a coherent single thing—but this experience of
the object as a coherent unity need not involve semantic classification of the
object as an object of this or that sort.
3. What Justifies Binding?

So far I have talked about the binding strategies that the visual system uses to
put together the various properties of a single object. The question I want to
address now is: what causes and justifies our use of those strategies? Is there a
role for sortal concepts in explaining how we come to use those strategies, or
in explaining why those strategies are correct?
Consider, for example, the use of location as a binding parameter. As I said
earlier, it seems likely that the visual system uses location as a fundamental
parameter in binding together different features as features of the same object:
features at the same location are, as a first approximation, assigned to the same
object. Someone who thinks there is a foundational place for sortal concepts
in explaining how there can be conscious attention to objects may press the
question: Why is binding together features at the same location the right
procedure for the visual system to use? What causes and justifies the use of this
binding procedure?
The null hypothesis is that there is nothing which justifies the use of one
binding procedure rather than another. On this hypothesis, which binding
procedures the visual system uses is more fundamental than either the
question which objects there are in the environment, or the question which
sortal concepts the subject employs. It is more fundamental than the question
which objects there are in the environment, because, the argument runs,
John Campbell / 211
the very notion of the ‘environment’ is always relative to a particular type of
creature; it depends on the ecological niche the creature inhabits. And, on this
view, the way in which the perceptual system solves the binding problem is
one of the things that determines what kind of world the creature inhabits.
Moreover, on this view, which procedure the visual system uses to solve the
binding problem is also more fundamental than the question which sortal
concepts we use. For which sortal concepts we use depends partly on what our
perceptual skills are; the most fundamental sortal concepts are observational
concepts, which we can apply to objects on the basis of perception alone. But
then the way in which our perceptual system solves the binding problem is one
of the things which determines what kinds of sortal concepts we can have and
use. So our way of solving the binding problem cannot be justified by reference
to our use of one rather than another collection of sortal concepts. Rather,
our grasp of sortal concepts simply has to work with whatever solution the
visual system finds to the binding problem. In any area, there is always, I think,
a presumption in favour of the null hypothesis; but it is a startling thought,
that the choice of a particular procedure to solve the binding problem may be
quite unconstrained by anything other than the internal requirements of the
visual system.
We can see what the null hypothesis is saying by supposing we consider for a
moment what would have to be happening for the visual system to be solving
the binding problem in a radically non-standard way. Quine and Goodman
used to talk about spatiotemporally scattered objects, which they claimed
were just as real as anything else. So a typical spatiotemporally scattered object
might comprise the top of one chair plus the base of a lightbulb. Non-standard
binding would involve putting together the perceived properties of the top of
the chair together with the perceived properties of the base of the lightbulb as
properties of a single, albeit spatiotemporally scattered, object. These features
would be combined to give an ‘object token’, which could then be compared
to stored representations to determine its characteristics. We could have a still
more radically non-standard form of binding. The properties of the top of the
chair would all be bound, and the properties of the base of the lightbulb would
all be bound, on the picture I just gave; the odd part is just putting the two
together. But we could in principle have binding in which no two features
from the same location were put together. The redness at this location, the
squareness at that location, and the uprightness from a still further location
could all be put together to give a single ‘object token’, to be compared to stored
representations. The upshot would be a kind of collection of spatiotemporally
scattered tropes.
According to the null hypothesis, there is no justification to be given for
the visual system operating in the ordinary way rather than in the ways I
just described; there is no objective advantage in the standard approach. The
visual system proceeds in whatever way it does; that is a primitive datum.
The way in which the visual system proceeds will determine which object
tokens are constructed, and that in turn will determine what stored object
representations the subject has. This in turn will determine what the sorts
of objects are in the subject’s environment; it is up to the subject to use one
system of object representations rather than another to delineate which things
in the surroundings he is thinking and talking about. It is natural to protest, as
against the null hypothesis, that the way in which we actually bind objects has
objective advantages over these bizarre alternatives. But the natural suspicion
is that this protest is just a parochial, conservative reaction which elevates
habit into a kind of transcendent superiority. What reason could be given for
thinking that one way of binding features into objects is better than another?
I think we can immediately say that it is not believable that there are no norm-
ative constraints on binding. There do appear occasionally individuals whose
visual systems have problems with binding—see, for example, Friedman-Hill
et al. (1995), or Robertson et al. (1997). These patients are quite seriously
impaired. Though they may, for example, be above chance at saying which
features are present in a scene displayed to them, they will be at chance when
saying which combinations of features are present. Or, as in the case of the
patient described by Humphreys and Riddoch (1987), they may be able to copy
a drawing of a complex scene accurately, but be doing it without any identifica-
tion of the objects involved. It is difficult to accept that since there are no norms
of binding, these patients cannot be described as impaired. Rather, it seems that
there must be normativity here, since these patients are so palpably impaired.
However, you might have a view on which getting it right in your use of
particular binding procedures is simply a matter of doing it the same way as
everyone else. That is, you might have a view of binding which is like the
kind of account sometimes given of what it is to be going right or wrong
in English grammar. On this picture there is such a thing as getting it right
or wrong in grammar; but ultimately, rightness and wrongness are just a
question of whether you are in step with other people speaking the same
language. Similarly, you might have, as it were, a ‘community view’ of binding
John Campbell / 213
procedures, on which rightness or wrongness is simply a matter of agreement
or disagreement with others in your community. This view implies that the
only problem with a non-standard binding strategy, such as putting together
features at different locations, and the only problem with the impaired patients
I just mentioned, is social. Their only problem is that they bind differently
to other people. But that seems entirely inadequate as an analysis of the
problem. The problem with these subjects is that they cannot see the objects
around them.
The sortalist proposal, at this point, contains two elements. One is a causal
hypothesis: that our visual systems use the binding strategies they do because
we have to sortal concepts that we do. The other element in the sortalist
proposal is normative: that what defines the objective of the binding strategies
being used is the need to keep faith with the system of sortal concepts used
by the subject. It has to be said, though, that neither of these ideas seems
compelling. Given the similarities between human vision and vision in other
species, it seems somewhat unlikely that the use of particular binding strategies
in humans evolved in response to our possession of particular sortal concepts.
And it is difficult anyway to see how our possession of particular sortal
concepts could have come first; our grasp of the concepts of particular sorts
of observable objects depends on our abilities to perceive them, which in turn
depends on having the relevant binding strategies in use already. Grasp of
a system of sortal concepts thus seems to depend causally on possession of
a relevant set of binding strategies, which makes it hard to see how there could
also be a causal dependence in the other direction.
The second element in the sortalist proposal is also difficult to accept.
According to the second element, it is our system of sortal concepts that
defines the objectives of the binding strategies that we use. Since the system
of sortal concepts that we have presumably developed later than the binding
strategies that we use, the sortal concepts are in effect providing a kind of after
the fact justification for the use of those binding strategies, on this view. But if
we have in fact developed a system of binding strategies in virtue of which we
can see various sorts of objects, then we do not need the development of sortal
concepts to provide a justification for the use of those strategies. The objective
is simply to see the relevant objects, and binding strategies achieve that. To
develop a system of sortal concepts, consequently upon this achievement, and
then maintain that the objectives of the binding strategies were after all being
set by this system of sortal concepts, adds nothing. We already have it in place
that the objective of the binding strategies is to let us see the various sorts of
objects around us, and there is no place for a further level of justification.
4. Sortals as Orienting Attention

There are two quite simple points which together can make it seem that
there must be something right about the Delineation Thesis. One is that we
usually do use sortal concepts in demonstrative constructions; very often, in
discussion, you would not know which thing was in question unless your
interlocutor used a sortal. The second point is that if someone uses an incorrect
sortal classification in a demonstrative construction, then what they have said
cannot be regarded as correct as it stands; once the facts are known, the
statement has to be withdrawn and replaced by a use of the correct sortal
classification. These points together can make it seem that (a) sortal concepts
are indeed essential to demonstrative reference, and (b) cases in which there
seems to be singular reference but incorrect sortal classification are somehow
not cases in which the singular reference has after all been successful. But
actually, these points need to be independently explained and they do not in
the end offer any support to the Delineation Thesis. I will take them in turn.
On the first point, there is a role for sortal concepts in demonstrative
identification which is much less fundamental than the kind of role allotted
to them by the Delineation Thesis, but which is commonplace and pervasive.
When I commented on Quine’s discussion of pointing, I remarked on the
distinction between conscious attention to an object, and the control, or
causation, of conscious attention. We can distinguish between the factors in
virtue of which you can be said to be consciously attending to one particular
object rather than anything else, and the factors which caused you to be
attending to that object rather than anything else. For example, in the case of
vision, it is arguable that you attend to a specific object in virtue, in part, of
the fact that you are attending to the location the object is at. In contrast, I
might orient your attention to the object by pointing, or by physically turning
your head towards it while waggling the object in front of you. You can see
the distinction between the two cases. Suppose that attending to the object
is partly constituted by attending to the place, while the turning of your
head is merely a cause of your attending to the object. Then you could have
attended to the object, in the very same way, as a result of some other cause.
John Campbell / 215
For example, I could simply have pointed to the thing, or you might have
become interested in it spontaneously. The upshot could still have been that
you consciously attended to it in the very same way.
As I said, we do not in ordinary communication confine ourselves to
saying ‘this’ and ‘that’; we very often do use phrases of the form, ‘this F’,
where F is indeed a sortal term. Does this show that the Delineation Thesis
is after all correct? I think it does not. We can distinguish between two ways
in which a descriptive component can figure in demonstrative reference.
Suppose you and I are strolling through the park and I point my umbrella
and say something about ‘that clock to the left of the fountain’. How does the
descriptive component ‘to the left of the fountain’ relate to the demonstrative
‘that clock’ in this case? There are two possibilities. One is that it states a
descriptive condition which has to be met by any object, if it is to be the
reference of my term. So if there is no clock to the left of the fountain, I have
not referred to anything. The other possibility is that the phrase ‘to the left of
the fountain’ is just a way of orienting your attention, exactly on a par with
the flick of my umbrella, which aims to direct your attention onto an object
I have already identified, prior to my use of the phrase or the umbrella. The
identification of the object came first, and now I am trying, by hook or by
crook, to direct your attention to that thing, and it does not really matter how
I achieve that effect. So if the clock is not really to the left of the fountain, but
I have still managed to direct your attention to it by means of my use of the
phrase, then my term still refers and you understand what I have said.
In these terms, we can say that one commonplace role for sortals in
demonstrative identification is to orient attention to one object rather than
another. The sortal really does play a role in the singling out of the object: it
is what causes the orientation of attention to one thing rather than another.
If I say to you, ‘that is very old’, you will very often have no idea which thing
I am talking about until I supply the sortal, even if I point in roughly the
right direction. You use the sortal to orient your attention onto the right
object. But since the sortal here is functioning merely to orient your attention
onto the right object, it can play its role successfully even if the object is not
actually of that sort. That is why the notion of a ‘plant’ can be playing a role
in referring to an object which is not actually a plant; it is ‘good enough’ in
the sense that anyone looking in roughly the right direction will have their
attention directed onto the right object by the sortal, even though the thing
is not in fact a plant.
To say that the role of the sortal is merely to orient attention towards the
right object, though, is also to say that the use of the sortal is dispensable. You
could in principle have your attention oriented towards that object by some
other cause. This is what happens in the case of the long-preserved teacup.
Our descendants manage to orient their attention onto the thing without the
use of any sortals at all.
This place for the role of sortals does not give them the kind of constitutive
role that the Delineation Thesis envisages in making it the case that you
are consciously attending to one thing rather than another. It is merely an
external causal factor in the act of conscious attention. But it does explain why
typically you simply would not understand a demonstrative ‘this’ which was
not, implicitly or explicitly, accompanied by a sortal.
This point, about the role of sortals in demonstrative reference, should
not be interpreted as a point about the semantic treatment of complex
demonstratives in English. We can contrast simple demonstratives, such as
‘this’ and ‘that’, with complex demonstratives, such as ‘this river’ or ‘that
mountain’. Suppose that the contribution of the nominal ‘F’ in a phrase of
the form ‘that F’ were merely to orient attention to the demonstrated object.
Then the truth of a statement of the form ‘that F is G’, would not require that
the object referred to be F. But it does have to be acknowledged that when
you use a demonstrative ‘that F’ to identify an object which is not in fact F,
your statement cannot be regarded as correct. On the face of it, at any rate, if
I point to a sundial and say, ‘that clock is one hour slow’, then what I say is
not literally true; it implies that there is a clock which is one hour slow, and
there may be no such clock. One analysis of the situation is that in addition
to orienting attention, the nominal ‘F’ restricts what the demonstrative can
refer to (cf. Kaplan 1989a, 1989b). On the view I am recommending, it would
be surprising if ordinary English did work in this way. For demonstrative
reference, on the view I am recommending, does not intrinsically require that
you get it right about the sort of the object. But it would in principle be possible
for English to impose an extrinsic requirement on reference: that when the
simple demonstrative is coupled with a nominal, the nominal must apply to
the object for the simple demonstrative to refer to it.
It may be, though, that the nominal in a complex demonstrative contributes
to the truth-conditions of a sentence containing it otherwise than by imposing
a condition on the reference of the demonstrative. Strawson (1950) suggested
that a sentence of the form ‘that F is G’ says the same thing as a sentence
John Campbell / 217
of the form, ‘that is the F which is G’. On this interpretation, the simple
demonstrative refers without any conditions on reference being imposed by
the nominal, though the nominal may indeed serve the pragmatic function of
orienting attention. But the nominal does contribute to the truth-conditions
of the sentence. The sentence cannot be true unless the nominal applies to
the object to which the simple demonstrative refers.
In a recent discussion, Lepore and Ludwig (2000) provide an analysis of
the same general type, in which we have a self-standing use of a simple
demonstrative and a quantificational analysis of the role of the nominal:
The key to understanding demonstratives in complex demonstratives is to see the
concatenation of a demonstrative with a nominal, as in ‘That F’, as itself a form of
restricted quantification, namely as equivalent to ‘[The x: x is that and x is F]’. (Lepore
and Ludwig 2000: 229; cf. King 2001)
On this analysis, the simple demonstrative refers without the nominal playing
any role in fixing its reference. Such an analysis nonetheless acknowledges
that a statement of the form ‘that F is G’ cannot be true unless the object
referred to is F; this is the second point with which we had to deal. And this
provides no comfort to a proponent of the Delineation Thesis, which we must
abandon.
References
Farah, Martha J. (2000), The Cognitive Neuroscience of Vision (Oxford: Blackwell).
Fodor, J. A. (1998), ‘There and Back Again: A Review of Annette Karmiloff-Smith’s
Beyond Modularity’, in In Critical Condition, 127–42 (Cambridge, Mass.: MIT Press).
Friedman-Hill, S. R., Robertson, L. C., and Treisman, A. (1995), ‘Parietal Contri-
butions to Visual Feature Binding: Evidence from a Patient with Bilateral Lesions’,
Science, 269: 853–5.
Humphreys, Glyn W, and Riddoch, M. Jane (1987), To See but Not to See: A Case Study
of Visual Agnosia (London: Erlbaum).
Kaplan, David (1989a), ‘Demonstratives’, in Joseph Almog, John Perry, and Howard
Wettstein (eds.), Themes from Kaplan (Oxford: OUP), 481–563.
(1989b), ‘Afterthoughts’, in Joseph Almog, John Perry, and Howard Wettstein
(eds.), Themes from Kaplan (Oxford: OUP), 565–614.
King, Jeffrey C. (2001), Complex Demonstratives: A Quantificational Account (Cambridge,
Mass.: MIT Press).
Lepore, Ernest, and Ludwig, Kirk (2000), ‘The Semantics and Pragmatics of Complex
Demonstratives’, Mind, 109: 199–240.
Lissauer, H. (1890), ‘Ein fall von seelenblindheit nebst einem Beitrage zur Theorie
derselben’, Archiv fur Psychiatrie und Nervenkrankenheiten, 21: 222–70.
Palmer, Stephen E., and Rock, Irvin (1994), ‘Rethinking Perceptual Organization:
The Role of Uniform Connectedness’, Psychonomic Bulletin and Review, 1: 29–55.
Prinzmetal, William (1995), ‘Visual Feature Integration in a World of Objects’, Current
Directions in Psychological Science, 4: 90–4.
Quine, W. V. O. (1953), ‘Identity, Ostension and Hypostasis’, in From a Logical Point of
View (Cambridge, Mass.: Harvard University Press).
(1960), Word and Object (Cambridge, Mass.: MIT Press).
Robertson, L., Treisman, A., Friedman-Hill, S., and Grabowecky, M. (1997), ‘The
Interaction of Spatial and Object Pathways: Evidence from Balint’s Syndrome’,
Journal of Cognitive Neuroscience, 9: 254–76.
Strawson, P. F. (1950), ‘On Referring’, Mind, 59: 320–44.
Treisman, Anne (1996), ‘The Binding Problem’, Current Opinion in Neurobiology, 6: 171–8.
Zeki, Semir (1993), A Vision of the Brain (Oxford: Blackwells).
Part III
Personal Identity
10
Vagueness and Personal Identity
Keith Hossack
1. Unclarity of Personal Identity

If someone journeys by matter transfer booth, or undergoes a brain fission, it
can seem unclear whether he himself will survive the adventure. The well-
known thought-experiments about personal identity raise questions which
we cannot see clearly how to answer.
What is the explanation of irresoluble unclarity about personal identity?
We might say that sometimes there is no precise fact of the matter about
identity of persons; that would be to explain the unclarity as a case of objective
indefiniteness. We might instead say that in every case there is always a precise
fact, but that in some cases we do not know how to discover it; that would be
to explain the unclarity as simply our own ignorance.
If materialism is true then probably the explanation of the unclarity is
objective indefiniteness. One version of materialism says that persons are
biological beings. We already know enough about biology to know that
probably nature draws no sharp boundaries in the biological realm; that
means there can be borderline cases where there is no fact of the matter
whether a particular animal has survived, or whether a particular biological
organ such as a brain has survived. A different version of materialism identifies
a person’s life with a psychological process sustained by an underlying physical
I am grateful for very helpful comments to David Bell, David Papineau, Mark Sainsbury, and an
anonymous referee; and to audiences at Sheffield, Lampeter, and Sussex.
222 / Vagueness and Personal Identity
mechanism. Again, we already know enough about mechanisms to know that,
whether natural or artificial, probably there are no sharp boundaries in the
realm of mechanisms. So standard versions of materialism seem committed
to the view that personal identity can be objectively indefinite.
Cartesian dualism, in contrast, standardly denies that there can be borderline
cases of personal identity: in the realm of the soul, dualists think, the relevant
boundaries are always sharp. Dualism explains unclarity about personal
identity as due to our own ignorance, not objective indefiniteness.1 Historically,
dualism has been motivated less by abstract metaphysics than by issues in ethics
and the philosophy of religion. For example, Kant held that it is a requirement
of pure practical reason that moral agents be able to hope that if they so
act as to be worthy of happiness, then they will in fact attain happiness. But
happiness and worthiness of happiness seem uncorrelated in this present life,
so Kant argued that belief in a future life is practically necessary.2 But this
seems to require that the person who is to enjoy happiness in the future life
be definitely the one who had made himself worthy of happiness; it would
be unsatisfactory if it were less than definite whether the right person was
enjoying the happiness. Dualist resistance to indefiniteness of identity is the
reason that Reid and others were led to the initially odd-seeming doctrine
that a person is a monad, i.e. a simple thing that has no parts.3 The motivation is
that if we replace enough parts of a complex thing like the ship of Theseus, it
1
The epistemic theory of vagueness of Williamson (1994) invokes ignorance to explain
unclarity even in borderline cases. A dualist and a materialist who subscribes to the epistemic
theory will therefore be in agreement that unclear cases of personal identity are cases of ignorance.
Where they will differ is over whether the unclear cases are ever borderline cases of personal identity,
for the dualist will deny it, but the materialist will affirm it. According to Williamson, vagueness
and borderline cases are due to a special kind of ignorance which arises because of limitations of
our powers of conceptual discrimination (1994: 237). Williamson says that if a concept is vague, it
is because we are unable to discriminate it from other concepts we might have used instead, which
draw the line in a slightly different place from where we actually draw it. This inability of ours,
itself epistemic, reflects something ontological, namely the absence of a natural division where
the line would naturally fall: in the absence of a natural division, there is nothing to ‘stabilize’ the
extension of the concept; that is why the concept is vague (ibid. 231.) Thus even on the epistemic
theory, the materialist will say that personal identity can be objectively indefinite, because of an
absence of natural boundaries; whereas the dualist will deny this.
2
See Kant (1933: ‘The Canon of Pure Reason’, A820–31); (1993: ‘On Assent Arising from a
Need of Reason’, pp. 149–53).
3
‘A part of a person is a manifest absurdity. A person is something indivisible, and is what
Leibniz calls a monad’ (Reid 1983: Essay III, ch. 4).
Keith Hossack / 223
can seem indefinite whether it survives; whereas a simple thing is not subject
to this kind of indefiniteness.
The question whether personal identity is determinate therefore seems a
central issue in the debate between materialism and dualism. But can we even
make sense of objectively indefinite identity in the case of persons? Bernard
Williams has suggested that there is something baffling in the very idea of it.
Williams (1970) invites one to imagine one has fallen into the clutches of a
mad surgeon. The surgeon tells one that he will bring it about that tomorrow
there will be two human persons A and B, each with a plausible claim to be
oneself. A will be bodily continuous with oneself; B will be psychologically
continuous with oneself; and either A or B will be tortured tomorrow. One
naturally wonders whether one will still be surviving tomorrow, and if so
whether one will be in pain. In normal cases bodily and psychological criteria of
personal identity go together, but the point of Williams’s example is to oblige us
to consider a case where they come apart. It may seem tempting to say there is
no fact of the matter whether one will be A or B; for example, on the grounds
that our concepts are simply not designed to cover such situations. But
Williams thinks this would encounter an ‘extraordinary difficulty’. He writes:
there seems to be an obstinate bafflement in mirroring in my expectations a situation
in which it is conceptually undecidable whether I occur.4
The reason is as follows:

Central to the expectation of [situation] S is the thought of how it will be for me,
the imaginative projection of myself as participant in S. Suppose it is for conceptual
reasons undecidable whether it involves me or not. The subject has an incurable
difficulty about how he may think about S. If he engages in projective imaginative
thinking (about how it will be for him) he implicitly answers the unanswerable
question; if he thinks that he cannot engage in such thinking, he answers it in the
opposite direction.5
If the supposition of indefinite personal identity really faces an ‘incurable

difficulty’, we might think we could develop Williams’s point into an argument
against materialism generally. Since there can be borderline cases of bodily
continuity, and borderline cases of psychological continuity, it follows that
standard versions of materialism imply that there can be borderline cases
4 5
Williams (1973: 61). Ibid.
of personal identity, and hence that there can be indefiniteness of personal
identity. But if indefinite personal identity is so baffling as to be incoherent, a
version of materialism that implies it must itself be incoherent.
I shall call this the Bafflement Argument against materialism. Richard
Swinburne states it as follows:
An awkwardness for any such empiricist theory is that there will inevitably be
borderline cases. But there seems no intermediate possibility between a certain future
conscious being being that person, and not being that person.6
Geoffrey Madell also endorses it. He writes:

What I fear about the future pain is simply its being felt by me. We, rightly, find
unintelligible the idea that there could be a pain in the future which is in part mine and
in part not. I claim, then, that the empiricist view of personal identity is incoherent
in itself.7
The Bafflement Argument is open to the reply that it is not at all baffling to
learn that a person vaguely identical with oneself will be tortured tomorrow;
I know exactly what to expect, for I know how to describe it from the third-
person perspective, and what I must expect is that. If it is insisted that bafflement
remains because one cannot effect the imaginative projection of oneself into
the situation, the reply is that this is just false. One can project oneself quite
vividly into the pain that the person who may or may not be oneself will feel
tomorrow. So one can imagine the torture tomorrow ‘from the inside’, and
one can conceive of its being indefinite whether that will happen to oneself.
So why should it be insisted that we still cannot really make sense of indefinite
personal identity?
Despite this reply, it seems to me that the Bafflement Argument retains a
residual ability to puzzle us. What seems so hard to imagine is not just the
situation tomorrow, but one’s whole course of experience between today and
tomorrow. Consider a variant of the mad surgeon’s experiment, where he
permits only A to be alive tomorrow. One is told today that it is objectively
indefinite whether one is identical with A, and that A will be tortured
tomorrow. How is one to imagine one’s future course of experience? If one is
definitely not identical with A, one will not survive until tomorrow, and so
one will have a course of experience that will end before tomorrow. If one is
6 7
Swinburne (1974: 240). Madell (1989: 32–3).
Keith Hossack / 225
definitely identical with A, one will have a course of experience that extends
through today into torture tomorrow. Either of these courses of experience
is imaginable, but it is not so easy to imagine ‘from the inside’ a course of
experience that is somehow borderline between the two. To put it another
way, what one would like to know is whether the light of one’s own particular
consciousness—this consciousness—will be on or off tomorrow, and what
seems so baffling is the answer that it will not definitely be on, but it will not
definitely be off either.
In this paper I attempt a diagnosis of this bafflement. I argue that the
concept of a person is a theoretical concept of a theory that combines two
distinctive components, namely a subjective conception of conscious states,
and an objective conception of psychological agents. Exclusive concentration
on the objective component is misguided, but it is equally wrong to overem-
phasize the subjective component. I suggest that the latter mistake is what
underlies the Bafflement Argument: by focusing excessively on the subjective
component it induces a solipsistic perspective from which indefinite personal
identity is indeed baffling. But although indefinite identity may baffle the
solipsist, it need not baffle those whose concept of person gives proper weight
to the place of persons as part of the objective order.
2. ‘I’—the ‘Essential’ Indexical

What is a person? I shall assume that x is a person only if x has a mind, and
that x has a mind iff x has the capacity to know something. One way to know
something, and therefore to have a mind, is to instantiate a conscious state; for
I take it that a state S is conscious iff necessarily if x is in S then x is consciously
aware of x’s being in S.
To say that a state is conscious is to say that its subject is aware of it, and
hence is aware of what is in fact a state of itself. In that sense any conscious
mind is conscious of itself. It does not follow, however, that such a mind is
self-conscious, for it may lack a concept of its Self. Because of the possibility of
such conceptual deficiency, awareness of what is in fact a state of itself need
not give a mind self-awareness. There is no contradiction in supposing that
a complex event could have a simple mode of presentation: if event E is the
instantiation of state S by entity x, and M is a mode of presentation of E, it need
not follow that M has a part which presents S and a part which presents x. In
particular, then, a mental event might have a simple mode of presentation,
in which case it would fail to have parts which present either the instantiated
mental state or the instantiating mental subject.
The distinction between mere consciousness and full self-consciousness was
relied on by Locke to distinguish between minds generally and the special
kinds of minds that are persons. According to Locke, a person is:
a thinking intelligent Being, which has reason and reflection, and can think of it self
as it self, the same thinking being in different times and places.8
Here ‘it self ’ seems to be occurring as the English indirect reflexive, i.e. the
pronoun used in indirect speech to report the content expressed by a speaker
who refers to himself as ‘I’.9 Locke’s definition then amounts to the claim that a
being with a mind is a person if it is able to think of itself by means specifically of
the concept or mode of presentation that is expressed by the word ‘I’. A concept
is individuated by its cognitive role, which is a matter for epistemology. So on
Locke’s assumptions, the philosophy of the self is one of those rare cases where
metaphysics is best done by epistemological methods; if a self is that which can
grasp a self-concept, we study the Self best by studying the self-concept, and
therefore by studying its cognitive role.
The cognitive role of the self-concept expressed by ‘I’ has at least the three
following distinctive features:
(1) Warrant: My normal warrant for ‘I am in pain’ ceases to be a warrant
if a co-referring expression is substituted for ‘I’. I know by immediate
introspection that I am in pain, but even if I am A, I cannot know by
introspection that A is in pain—unless I know that I am A. The same
applies to other contents warranted by introspection.
(2) Desire: ‘Thank goodness I am not to be tortured!’ This expression of
relief will not be apt if a co-referring expression is substituted for ‘I’. It is
not the same to say ‘Thank goodness A is not to be tortured!’, even if I am
A. I am not relieved that B is to be tortured and A is not—what’s so good
about that? Here is a characteristic kind of non-contrastive relief which
can only be properly expressed in English by the first-person and its
cognates, thus: I am glad I will not be tortured myself (indirect reflexive).
(3) Action: Substitution of another mode of presentation for the first-
person can destroy the rationalization of action. Suppose I desire that the
8 9
Locke (1975: Bk. ii, ch. xxvii, § 9). Geach (1957).
Keith Hossack / 227
race should start, and believe that if I wave my arm the race will start. The
desire and belief jointly rationalize my trying to wave my arm, but they
do not rationalize my trying to wave A’s arm, unless I know that I am A.
Perry and others give a psychologistic account of this ‘essential indexical’
role of ‘I’-thoughts, by appeal to the supposed causal powers of first-personal
‘belief-states’.10 But we can give an account that is more Fregean in spirit
if we regard conscious action as rationalized by a practical syllogism whose
conclusion is the action. A practical syllogism is formally valid iff its action
conclusion and its belief premiss formally entail its desire premiss, e.g.:
(desire) the race starts
(belief ) if I wave my arm then the race starts
(action) I wave my arm
A syllogism rationalizes an agent’s conscious action only if it describes it
under that mode of presentation in which the action is given to the agent’s
consciousness. Now consciousness of waving my arm warrants my thought
that I am waving my arm, but it does not warrant the thought that A is
waving his arm, even if I am A. So the following practical syllogism is not
formally valid:
(belief ) if A waves his arm then the race starts
(action) I wave my arm
Here the action and the belief do not formally entail the thing desired.11
The action would get me what I want, but I do not know that, and cannot
rationally act on this belief and desire. Rationality can be restored here only
by adding as an extra premiss the belief ‘I am A’.
3. How Do I Know I Exist?

To investigate the self-concept, we must investigate its cognitive role. This
will include an investigation of what warrants a subject’s judgements about
himself. One judgement I am inclined to make about myself is that I exist:
what warrants that?
10 11
Perry (1979). cf. Campbell (1994: ch. 3).
I cannot come to know of my existence by ordinary perception. I see my
body, and thus have warrant for believing ‘This body exists’. Given some forms
of materialism, I know of what is in fact myself that it exists. But I need not yet
know that I exist, because the identity ‘I am that body’ would be informative for
me. Nor can I do better by turning my mental gaze within. Even if there really
were an inner gaze and, luckier than Hume, I managed to catch a glimpse of
my cartesian Ego, still I would not know that I exist, for the judgement ‘I am
that Ego’ would be informative for me.
I cannot know of my own existence apriori. It is of course apriori that any
token of ‘I exist’, including this one, is true. So I can know apriori of a token of
‘I exist’ which is in fact produced by me that it is true. But still I do not know
apriori that I exist, for I do not know apriori that I produced that token.
I cannot know of my existence by introspection. Following Reid I will use
‘consciousness’ to denote that immediate introspective knowledge one has of
one’s own conscious states.12 Then consciousness will be part of my warrant
for such knowledgeable judgements as ‘I am in pain’, or ‘I am thinking that
snow is white’. But mere consciousness cannot warrant my judgement that
I exist. For although consciousness informs me of the existence of a state of
what is in fact myself, it cannot inform me of the existence of my self; as is
shown by the existence of minds that are conscious but not self-conscious.
Descartes missed this point in his famous paralogism Cogito, ergo sum. But his
critics such as Hume and Kant did not miss it; they correctly insisted that mere
consciousness of a conscious state does not on its own warrant a belief in the
existence of the subject of the conscious states. As Lichtenberg put the point:
To say cogito is already to say too much, as soon as we translate it I think. (1990:
Notebook K, 18, p. 168)
According to Lichtenberg, all that is warranted on the basis of consciousness
alone is a report of the conscious states in a subjectless style:
We should say it thinks, just as we say it lightens. (ibid.)
Wittgenstein endorsed this manoeuvre. Discussing the word ‘I’ in Philosophical
Remarks he says:
It would be instructive to replace this way of speaking by another in which immediate
experience would be represented without using the personal pronoun; We could
12
‘Consciousness is that immediate knowledge we have of our present thoughts and purposes,
and in general of all the present operations of our minds’ (Reid 1983: I. 7).
Keith Hossack / 229
adopt the following way of representing matters: if I, L.W. have toothache, then that
is expressed by means of the proposition ‘There is toothache.’13
We can imagine constructing a Lichtenbergian language from our ordinary

language. Whenever ordinary language contains a predicate which I am
warranted in applying to myself by consciousness together with knowledge
that the conscious state is a state of myself, we introduce into the Lichtenbergian
language a corresponding subjectless sentence, for which the consciousness
alone is warrant enough. Then we can replace ‘I am thinking’ by ‘It thinks’,
‘I am in pain’ by ‘It pains’, and so on. Each Lichtenbergian sentence expresses
a content warrantable by consciousness alone; equivalently, it expresses an
introspective mode of presentation of the relevant conscious state.
A Lichtenbergian language is of course the Wittgensteinian ‘private lan-
guage’, which ‘only I myself can understand’.14 In what follows, I side-step
the question whether such a language is genuinely possible. All I require is the
more modest assumption that there exist introspective modes of presentation
of conscious states that are not modes of presentation of the Self. These are
the modes of presentation which the Lichtenbergian ‘private language’ would
express, supposing a private language were possible. A private language may
be a fiction, but it is a useful one here, since it provides a convenient means of
discussing these modes of presentation.
A Lichtenbergian sentence is, by a convenient pun, subjectless in a double
sense. First, it is subjectless because it is not a subject-predicate sentence. Instead
it has the holophrastic logical form of a Quinean ‘observation sentence’; the
sentence calculus is adequate for its formal representation. Secondly, a
Lichtenbergian sentence is subjectless because it expresses a conscious mental
state without explicit mention of the owner or subject of that state. Thus
someone could speak Lichtenbergian even if he had no concept of a mental
subject. He could state the occurrence of states of affairs which were in fact
states of a subject, without having a concept of the subject.
We can imagine someone following Descartes’s method of doubt, believing
only what he finds indubitable. He no longer believes in outer things. He
can even doubt his own existence, and pace Descartes the doubt is not self-
refuting—to say dubito ‘is already to say too much’. This skepticism is extreme
indeed. It goes beyond even what Russell called ‘solipsism of the present
13 14
Wittgenstein (1975: 88). Wittgenstein (1958b: §256).
moment’, i.e. the belief that only I and my present experiences are real. Our
skeptic goes further, for he lacks belief not only in the existence of other
minds, but also in his own.
Imagine our skeptic has been sunk in his skepticism so long that it is no
longer right to say he doubts his own existence; for he has now entirely lost the
very concept of himself. I shall call such a skeptic a ‘Lichtenbergian solipsist’.
(I say ‘solipsist’ to accord with what I take to be Wittgenstein’s usage, but
the terminology is of course a little misleading: the Lichtenbergian solipsist
does not think that he is the only person who exists, for he lacks so much as
the concept of his own or other minds.) Our solipsist follows Lichtenberg’s
procedure, always saying ‘It thinks’ instead of ‘I think’, and so on. Is he
hopelessly stuck in his solipsism? Or is there a way he can rationally come to
believe again that he exists? If we can discern such a way, that would throw
light on the tacit warrants we non-solipsists have for the belief ‘I exist’, and
hence clarify the cognitive role of the self-concept.
4. Beyond Solipsism of the Present Moment

A Lichtenbergian solipsist arrives at his epistemological predicament by
doubting all his aposteriori sources of knowledge other than consciousness. He
can attain knowledge of his own existence again only if he starts to draw on
the other sources of knowledge that he has.
If he is prepared to trust his memory, then he can have knowledge of the past.
He can arrive at a conception of the future by remembering that in the past the
present was not yet, i.e. was future. He can thus advance to a Lichtenbergian
sentential language enriched with tense operators like in the past and always.
A second source of knowledge he might come to trust is induction, i.e. the
ability to project a remembered regularity into the future. That enables him
to know Lichtenbergian psychological laws. For example:
Always, if it pains, it is desired that it does not pain.
Compare the subjectless meteorological law:
Always, if it rains, it pours.
Knowledge of a Lichtenbergian psychological law is knowledge of a genuine
law of subjective psychology, and as I shall argue later, such knowledge plays
a key role in our understanding of other minds. But no amount of such
Keith Hossack / 231
knowledge on its own will lead the solipsist to knowledge of the Self as one of
the things that objectively exist, for the very conclusive reason that our solipsist
may as yet completely lack the concept of a thing. The solipsist can arrive at this
concept by trusting another source of knowledge—abduction, i.e. inference to
the best explanation by means of a theory.
To think of himself as ‘the same thinking thing at different times and places’
(Locke), our solipsist will need to have concepts of objective locations in time
and space. Kant argued that one cannot have a conception of time and space
without a conception of the physical objects that occupy time and space; this
entails that Lockean self-consciousness requires concepts of physical objects.
We arrive at the claim that (for human beings) there can be no conception of
oneself without a conception of the physical world in which one is located.
Is the converse true, that one have no conception of the physical world
without a concept of oneself as located within it? Strawson and others have
argued that it is true, on the grounds that one could not regard one’s
experience as experience of independent objects without some rudimentary
theory of perception which relates the content of one’s experience to one’s
own location and the physical objects perceptible from that location.15 If so,
the transition from Lichtenbergian solipsism to knowledge of the physical
world would necessarily bring with it knowledge of one’s own existence.
The claim would appear to be that if our solipsist is to have experiential
knowledge of material objects in space and time, he will need to develop at
least a rudimentary physics in which physical objects, places, and times are the
theoretical entities. The solipsist’s evidence for the theory will be the course
of his own conscious experience, so to link the theoretical entities with his
consciousness, he will need to supplement the physics with ‘bridge laws’ or
‘correspondence rules’ that amount to a rudimentary psycho-physical theory
of perception. This theory must link physical states of affairs with states
accessible to consciousness, and the key point is that in the case of human
beings it must do so by linking conscious experience to physical impingements
on the theorist’s own body. Thus without something to play the role of his
own body in the theory, there can be no prospect of a thinker having warrant
for belief in physical objects.
If I have a concept of what is my body in this sense, do I have a concept of
myself? Must I take the adjective ‘my’ as the possessive case of the substantive ‘I?’
15
Strawson (1966). For a careful discussion of this and the ensuing literature, see Cassam (1997).
I think we must deny this if we wish to maintain a sharp division between mere
mindedness and being a person. Example: it might not be definitely wrong
to say that an animal can have the concept of its body, since it can acquire
through perception a warranted belief that, e.g., there is food to its right;
even so, it presumably remains definitely wrong to attribute to the animal a
Lockean concept of itself.
Our solipsist’s concept of his body will be embedded in a theory which posits
physical objects as the ‘theoretical entities’, with the conscious experiential
states of the theorist himself as the ‘observation sentences’. I see no reason
why this theory could not be developed by a solipsist who conceived of his
own conscious states only in the subjectless style of Lichtenberg. The solipsist
would acquire a concept of what is in fact his own body, and would think of
it under the special mode of presentation it has in virtue of its central role in
his theory of perception. We may suppose that the solipsist employs a special
name to express this special mode of presentation of his own body; perhaps
he adopts for it the name ‘Centre’.16 Our solipsist has attained to a minimal
physics, with an embedded theory of perception giving a special role to Centre.
But in view of the possibility of a Lichtenbergian ‘observation language’, his
doing all this does not entail that he has formed a concept of himself; he might
remain a solipsist.
In The Blue Book, Wittgenstein describes a solipsist who knows about the
physical world, and is aware of his own conscious states, but lacks so much as
the concept of his own or other minds.
[T]he man whom we call a solipsist, and who says that only his own experiences
are real . . . would say that it was inconceivable that experiences other than his own
were real.17
Despite his extensive or even complete knowledge of the physical world and
the psycho-physical laws relating his conscious states to the physical states of
his body, this solipsist still lacks the concept expressed by the word ‘I’.
5. The Solipsist as Rational Agent

Can this solipsist be a fully rational conscious agent, acting rationally upon his
beliefs and desires, and learning rationally from his experience? We saw that
16 17
Wittgenstein (1975: 89). Wittgenstein (1958a: 59).
Keith Hossack / 233
‘I’-thoughts have an ‘essential indexical’ cognitive role, which is indispensable
for our rationality. The solipsist has no self-concept, and therefore no ‘I’-
thoughts: but his Lichtenbergian sentences can play a corresponding ‘essential
indexical’ role in his rational economy.
(1) Warrant. Consciousness is the salient warrant for self-attributions like ‘I
am in pain’. The solipsist has just the same warrant for the corresponding
Lichtenbergian thought ‘It pains’.
(2) Desire. There is a particular non-contrastive relief I feel when I learn
I will not be tortured. The solipsist will feel the same non-contrastive
relief when he learns that it is not Centre that will be badly injured, for
from psycho-physical laws he knows he can infer ‘It will not pain’, and
be relieved at that. His relief is non-contrastive, in just the required way:
necessarily so, for he lacks the concepts of self and others, and so cannot
draw the distinction that a contrastive relief would require.
(3) Action. Action normally requires intentional movement of one’s body,
so if a practical syllogism is to be formally valid its belief premiss must
refer to the agent’s body under the mode of presentation whereby he
is conscious of it in voluntary movement. But that is the mode of
presentation under which the solipsist thinks of his body when he refers
to it as ‘Centre’. So the solipsist’s action can be properly rationalized by
a Lichtenbergian practical syllogism. For example, if he waves his arm in
order to start the race, we have:
(belief ) if Centre’s arm is waved, then the race starts
(action) Centre’s arm is waved
So even in the absence of a self-concept, our solipsist can think thoughts with
an ‘essential indexical’ rational role, and so he can be a rational agent.
6. A Functionalist Conception of Human Beings

At this stage about the only remaining difference between the Wittgensteinian
solipsist and ourselves is the solipsist’s continuing lack of a self-concept.
Because the solipsist has no concept of a self, he lacks the concept of a person,
and cannot be aware of other human beings as persons. However, he knows
of their bodies and can predict how their bodies will move by using physics.
Moreover there seems nothing to prevent the solipsist from acquiring a theory
of objective psychology, which will allow him more easily to predict how
human bodies will move.
An example of an objective psychology that is cognitively accessible to the
solipsist is functionalism; for example, the functionalist ‘folk psychology’ that
some say all normal human beings tacitly know. A functionalist theory says
that human beings are functional agents with bodies; perhaps human beings
are identical with their bodies, but this is a question functionalism need not
settle. The psychological states of a psychological agent explain why his body
moves as it does, in accordance with the functional role of each state. Such
a functionalism would be fully cognitively accessible to a Wittgensteinian
solipsist, but knowledge of the existence of psychological agents in the sense
of functionalism is not yet knowledge of other minds. Functionalism treats
pain as a psychological state, but it need not treat it as a conscious state. A fully
developed functionalism may include functional analogues of introspection
and qualitative character; for example, it may say that normally if x is in pain
then x reliably believes that x is in pain and desires that x is not in pain, but
this does not entail that the pain, the belief, or the desire are conscious states.
There need be nothing in a definition of functional pain that requires that
pain is a conscious state.18
7. Other Minds Theory

Although our solipsist is by now an expert physicist and an expert objective
psychologist, he still does not know that there are any other conscious beings
in the world. He does not know this because he still lacks the very concept of
a conscious being or a conscious state. He is aware of what are in fact his own
conscious states, but thinks of them only in the subjectless Lichtenbergian way.
To form a concept of the mental subject, the solipsist needs to conceive of
his subjectless ‘It pains’ as being made true by a fact of subject-predicate form,
18
Here I am assuming that the functionalist theory in question is within the cognitive grasp
of our solipsist. Might a more elaborate version of functionalism be able to link pain with
consciousness? A conscious state is one that its subject is aware of, i.e. a state such that he knows he
instantiates it. To build consciousness into ‘functionalism’ would therefore require a functionalist
analysis of knowledge itself; it seems excessive to require our solipsist to come up with such an
analysis, given that no philosopher has so far been able to do so.
Keith Hossack / 235
i.e. by the instantiation of some state by some entity. He must posit the existence
of a class of conscious states corresponding to each of his Lichtenbergian
sentences, and then he can posit a type of entity that can instantiate these
conscious states. Only then will he be able so much as to grasp the hypothesis
that conscious entities exist, and that he himself is one of them.
How can the solipsist find evidence for this hypothesis? He must link the
hypothesis in a theoretically fruitful way to the rest of what he knows.
He already knows that each normal human body is associated with an
agent in the sense of functionalism who ‘owns’ that body; in particular, he
knows there is an agent that ‘owns’ Centre. Suppose the solipsist calls this
agent the ‘Ego’. The solipsist can now propose to identify the Ego with the
entity that instantiates conscious pain when the Lichtenbergian sentence ‘It
pains’ is true. Similarly, the solipsist knows there is a state that realizes the
functional role of pain in human beings. He can propose to identify this
state with the state instantiated by the Ego when ‘It pains’ is true. He will
propose corresponding property identifications for other sentences of the
Lichtenbergian language.
A rich theory now emerges, in which the solipsist conceives of the Ego as
both an agent in the sense of functionalism and as the conscious subject of
the states of subjective consciousness expressed by Lichtenbergian sentences.
Because the Lichtenbergian sentences are indexical, the solipsist’s concept of
the Ego is an indexical one. Thus the solipsist has attained to a self-concept
at last, and he may as well start using the indexical word ‘I’ to express
it. By ‘I’ he expresses a theoretical concept of the following theory: I am
the ‘owner’ of Centre; I exist; whenever it pains, I am in pain, whenever it
thinks, I am thinking, etc. Such an I-concept will have the proper ‘essential
indexical’ rational role, since it will straightforwardly inherit that role from
the corresponding Lichtenbergian sentences. It seems plausible that this is in
fact the very concept expressed by the English indexical ‘I’.
Our solipsist has now arrived at a psychological theory that is more
powerful than functionalism alone. For by attributing conscious states to other
human beings, he enables the Lichtenbergian psychological laws he knows to
generalize to other minds. So now he has a new method of understanding
other people: he can put himself in the other person’s place by means of
projective imaginative thinking. The Lichtenbergian psychological laws he
knows from his own case can by this means be extended to apply to other
people too, and thus he comes to understand other people better than he
would if he were to rely on functionalism alone. The success of the method
is evidence of the correctness of its assumptions; the solipsist is therefore
warranted in believing in his own existence precisely because and to the extent
that he is warranted in believing in the existence of other persons. This is
the deep point of Wittgenstein’s discussion of the problem of other minds:
skepticism about other minds is an unstable position, because anyone seriously
skeptical about the existence of other people will undermine the evidence he
has for his own existence. Ontological commitment to oneself is inseparable
from ontological commitment to others.
8. Is Personal Identity Indefinite?

Wittgenstein suggests in The Blue Book that ‘as subject’ uses of the first-person
pronoun can be seen as immune to error through mis-identification of
the subject.
There is no question of recognizing a person when I say I have toothache. To ask ‘are
you sure it’s you who have pains?’ would be nonsensical. And now this way of stating
our idea suggests itself: that it is as impossible that in making the statement ‘I have
toothache’ I should have mistaken another person for myself, as it is to moan with
pain by mistake, having mistaken someone else for me.19
As Wittgenstein perhaps intended, if there is indeed such a phenomenon as

immunity to error through mis-identification, it can be neatly explained in the
light of the Lichtenbergian theory of the word ‘I’. For a person can be mistaken
in the ‘as subject’ belief ‘I am in pain’ only if he can be mistaken in what
warrants it. But the warrant is his awareness of pain under the introspective
mode of presentation expressed by the Lichtenbergian sentence ‘It pains’.
Since the Lichtenbergian sentence is subjectless, there seems no possibility of
mistakenly believing it as a result of mis-identifying its subject; therefore ‘as
subject’ uses of ‘I’ are immune from error arising from error about the identity
of the subject.
Keeping this point in mind, let us return to our consideration of the
Bafflement Argument against materialism. The argument says it is hard to
make sense of information that it is indefinite whether a person who will
19
Wittgenstein (1958a: 67).
Keith Hossack / 237
be in pain tomorrow will be me. The suggestion is that, from the first-
person perspective, there is something incoherent in the materialist claim that
personal identity can be indefinite. There is an analogy here with immunity
to error through mis-identification, which we can bring out by putting the
suggestion of the anti-materialists in the following way: ‘as subject’ uses of ‘I’
are immune from indefiniteness arising from indefiniteness about the identity
of the subject.
A neat Lichtenbergian explanation might again seem to be available.
Consider again the situation where the mad surgeon is going to bring it
about that the only candidate for being oneself tomorrow will be person A,
who will be a borderline case of bodily and/or psychological continuity with
oneself. One would like to know if one will still exist tomorrow. Since A will
be tortured tomorrow, and one will exist tomorrow only if one is identical
with A, then if one will exist tomorrow, one will be in pain tomorrow, and
so one’s corresponding Lichtenbergian sentence ‘It will pain tomorrow’ is
true now. So a way to express what one would like to know about one’s
survival is this: if A is tortured tomorrow, will it pain tomorrow? Here I
am expressing things from the Lichtenbergian perspective of a solipsist, but
intuitively this does seem to be, at least in the present case, exactly ‘what
matters in survival’.20 Now from a Lichtenbergian perspective, indefiniteness
of identity and survival is indeed baffling; for the sentence ‘If A is tortured
tomorrow, it will pain tomorrow’ seems immune from indefiniteness about
the subject, since it does not even employ a concept of the subject. Note that
the Lichtenbergian sentence is not otherwise indefinite, for even borderline
torture is definite pain.
When an opponent of materialism feels bafflement about how his personal
identity can be indefinite tomorrow, I suggest it is because he has slipped
into thinking of things in this Lichtenbergian way. The subjectless sentence
‘It will pain tomorrow’ does indeed appear immune from indefiniteness about
the subject. However, this appearance is misleading. For a vague object can
make a sentence vague even if the sentence itself expresses no concept of
the vague object. This can happen if the sentence concerned is indexical, for
an indexical sentence may rely on the context to contribute an object that
enters its truth-conditions, and if the context disobligingly contributes a vague
object, then the indexical sentence itself will be vague.
20
Parfit (1971).
Now non-solipsists know that Lichtenbergian sentences are indexical, so
they understand that the subject does indeed enter the truth-condition of ‘It
will pain tomorrow’:
A token of ‘It will pain tomorrow’ by S at t is true iff S is in pain the day
after t
So non-solipsists understand that a token of ‘It will pain tomorrow’ will be
indefinite if it is indefinite whether the person speaking at t is identical with
the person who is tortured the day after t. Of course, the solipsist is baffled to
understand this, for he lacks the concept of a person. But the fact that the
solipsist is baffled does not mean that the rest of us ought to be.
9. Conclusion
The Bafflement Argument says that from the first-person perspective there
is something incoherent in the materialist claim that personal identity can
be indefinite. My suggestion has been that the anti-materialist’s so-called
‘first-person perspective’ is not the genuinely first-person perspective of one
who conceives of himself as part of the objective order. Rather it is the merely
Lichtenbergian perspective of the solipsist, who does not conceive of himself at
all. When an anti-materialist feels bafflement about how his personal identity
can be indefinite tomorrow, I suggest it is because he is thinking of things from
this subjectless point of view. He can escape the bafflement easily enough by
going beyond solipsism and seeing Lichtenbergian sentences as indexical with
respect to the subject of experience.
My account of the self-concept turned on the identification of states of the
conscious subject with the properties that realize the functional properties of
objective psychology. If the account is correct, it will follow that in our present
state of knowledge we cannot answer the question whether personal identity
can be indefinite. For our present concept of the self combines a functional with
a subjective conception of conscious states, and neither of these can contribute
any information on the question of the real essence of consciousness. For
the time being at least, although we know what consciousness does, and
although we know what it is like to be conscious, we know nothing of what
consciousness is.
It may turn out that a theory of the real essence of consciousness can be
completed within biology, or within the theory of physical mechanisms; if
Keith Hossack / 239
so, then personal identity will have turned out to be indefinite. But for all
we presently know, the realm of consciousness may be one where nature in
fact does draw sharp boundaries. If so, personal identity will not after all be
objectively indefinite, and the most plausible versions of materialism will have
been refuted. But this is not something we can decide in advance apriori, either
by the Bafflement Argument or indeed by any other species of knowledge
argument for dualism.
Appendix: A Lichtenbergian Reconstruction of the

Bafflement Argument
Assume the scenario of Williams’s paper ‘The Self and the Future’. Subject S
is told today that tomorrow there will be a person A, who will definitely be
the only person to suffer pain tomorrow, and who is such that it is objectively
indefinite whether S = A. If materialism is true, what S is told is possibly true.
But it is not possibly true, and so materialism is false.
‘Proof’: Use ‘’ as the ‘definitely’ operator. What S is told is (1) and (2) below:
(1) ¬(S = A)& ¬¬(S = A)
(2) (∀x)(x will be in pain tomorrow ↔ x = A)
Assume you are definitely S.
(3) (I am S)
(4) (S = A ↔ I am A) (from (3))
Now ‘imaginatively project’ yourself into A’s situation tomorrow:
(5) (I am A ↔ will be in pain tomorrow) (from (2))
In view of the connection between ‘I’ and your Lichtenbergian private language:
(6) (I will be in pain tomorrow ↔ It will pain tomorrow)
The pain in question is to be torture, i.e. definite pain. Since the Lichtenbergian
sentence expresses no concept of the subject, it cannot be indefinite whether
it will pain tomorrow. So:
(7) (It will pain tomorrow) v ¬ (It will pain tomorrow)
We may assume that for the operator ‘’ the following rule of inference
is valid:
Definite Modus Ponens: p, (p → q) |- q
Then we have from (4), (5), and (6)
(8) (It will pain tomorrow) |- (S = A)
(9) ¬ (It will pain tomorrow) |- ¬ (S = A)
And hence by (7) we may conclude:
(10) (S = A) v ¬ (S = A)
(10) contradicts (1), so either what S was told is not possibly true, and so
materialism is false, or else the method of projective imagining (premisses (5 &
6)) is not possible in this case. But it would be ‘baffling’ to have a situation where
projective imagining is not possible, since it is constitutive of the ‘I’-concept.
So materialism is false.
References
Campbell, J. (1994), Past, Space and Self (Cambridge, Mass.: MIT Press).
Cassam, Q. (1997), Self and World (Oxford: OUP).
Geach, P. (1957), ‘On Beliefs about Oneself ’, Analysis, 18; rep. in his Logic Matters (Oxford:
Blackwell, 1972).
Kant, I. (1933), Critique of Pure Reason, trans. Norman Kemp Smith (London: MacMillan).
(1993), Critique of Practical Reason, trans. Lewis White Beck (Englewood Cliffs, NJ:
Prentice-Hall).
Lichtenberg, G. C. (1990), Aphorisms, trans. R. J. Hollingdale (Harmondsworth: Pen-
guin Classics).
Locke, J. (1975), An Essay concerning Human Understanding (Oxford: OUP).
Madell, G. (1989), ‘Personal Identity and the Mind/Body Problem’, in J. R. Smithies
and J. Beloff (eds.), The Case for Dualism (Charlottesville, Va.: University of Vir-
ginia Press).
Parfit, D. (1971), ‘Personal Identity’, Philosophical Review, 80: 3–27.
Perry, J. (1979), ‘The Problem of the Essential Indexical’, Noûs, 13: 3–21.
Reid, T. (1983), Essays on the Intellectual Powers of Man, excerpted in Inquiry and Essays, ed.
R. Beanblossom and K. Lehrer (Indianopolis: Hackett Publishing Company).
Strawson, P. (1966), The Bounds of Sense (London: Methuen).
Swinburne, R. G. (1974), ‘Personal Identity’, Proceedings of the Aristotelian Society, 74:
231–47.
Keith Hossack / 241
Williams, B. (1973), ‘The Self and the Future’ (1st pub. 1970), repr. in his Problems of the
Self (Cambridge: CUP).
Williamson, T. (1994), Vagueness (London: Routledge).
Wittgenstein, L. (1958a), The Blue and Brown Books (Oxford: Basil Blackwell).
(1958b), Philosophical Investigations, trans. E. Anscombe (Oxford: Basil Blackwell).
(1975), Philosophical Remarks, ed. R. Rhees, trans. R. Hargreaves and R. White
(Oxford: Basil Blackwell).
11
Is There a Bodily Criterion
of Personal Identity?
Eric T. Olson
1.
One of the main problems of personal identity is supposed to be how we relate
to our bodies. A few philosophers endorse what is called a ‘bodily criterion of
personal identity’: they say that we are our bodies, or that our identity over
time consists in the identity of our bodies. Many more deny this—typically on
the grounds that we can imagine ourselves coming apart from our bodies. But
both sides agree that the bodily criterion is an important view, which anyone
thinking about personal identity must consider.
I have never been able to work out what the bodily criterion is supposed to be.
Despite my best efforts, I have not found any clear position that plays the role
in debates on personal identity that everyone takes the bodily criterion to play.
What role is that? What is the bodily criterion supposed to be? Well, first it is
supposed to be a thesis about our bodies and how we relate to them. Second, it
is supposed to be, or imply, an account of what it takes for us to persist through
time. Specifically, it should imply that we go where our bodies go: it should
rule out our having different bodies at different times, or surviving without a
I thank Keith Allen, Andrew Deakin, Katherine Hawley, Jane Heal, Dan Hewitt, Diana Kendall,
Hugh Mellor, Trenton Merricks, Gonzalo Rodriguez-Pereyra, an anonymous referee, and the
audience at the 2000 St Andrews conference for comments on earlier drafts.
Eric T. Olson / 243
body. Third, the bodily criterion is supposed to be a substantive metaphysical
claim that is neither trivially true nor trivially false. That we have bodies is
uncontentious, or at least no more contentious than the existence of other
physical objects; but there is supposed to be room for disagreement about
whether we are our bodies. This is not just the old debate between dualism
and materialism: it is supposed to be possible for a materialist—someone who
thinks that we are made entirely of matter—to reject the bodily criterion by
denying that we are identical with our bodies.
Any thesis that deserves to be called the bodily criterion of personal identity
ought to have at least these three features. It should also be compatible with the
things that virtually all philosophers say about our bodies in other contexts.
For instance, both those who accept the bodily criterion and those who deny
it agree that a person’s body does not cease to exist simply because that person
loses the ability to think. The bodily criterion should respect this; otherwise
both sides of the debate would say, ‘that’s not what I meant by the bodily
criterion’. Let us try to formulate a thesis that meets this modest standard.
2.
The bodily criterion suggests two different thoughts: that a person’s identity
consists in the identity of something called her body, and that we are identical
with our bodies. These are not the same: Ayer, for instance, held the first but
not the second (1936: 194). But they are closely related, and it is no accident that
they are commonly affirmed or denied together. They are certainly supposed
to be compatible. (This is another thesis about bodies that both friends and
foes of the bodily criterion agree on.)
Let us begin with the first thought, that our identity is determined by the
identity of our bodies. Here is a typical statement:
Person A at time t1 is identical to person B at time t2 if and only if A and
B have the same body. (Garrett 1998: 45)
I take this to mean that one is, necessarily, that past or future person whose
body then is the very thing that is one’s body now:
Necessarily, for any person x existing at a time t and any person y existing
at another time t∗ , x = y if and only if the thing that is x’s body at t is the
thing that is y’s body at t∗ .
Let us call this the Standard Bodily Criterion, or Standard Criterion for short.
244 / A Bodily Criterion of Personal Identity?
Is this the bodily criterion? Well, suppose you were to lapse into an irreversible
vegetative state: your upper brain is destroyed, while the lower parts that direct
your ‘vegetative’ functions remain intact. The result is an animal that is fully
alive in the biological sense—as alive as an oyster is alive—but with no mental
features whatever. Most philosophers, I think, would say that your body still
exists in this case. Of course, the Standard Criterion does not itself say this.
It doesn’t say when something is the same body as an earlier one, and when
not. This raises a problem: the bodily criterion will tell us nothing about our
identity through time unless we have at least some idea of what it takes for
a person’s body to persist; yet no one has ever produced a serious account of
the identity conditions of human bodies. But never mind. In order to give the
bodily criterion a run for its money, let us pretend that we know what it takes
for a person’s body to persist. Specifically, let us accept that your body persists
as a vegetable when you lapse into an irreversible vegetative state.
Now what happens to you in this case? The bodily criterion ought to imply
that you go where your body goes. If your body persists as a vegetable, then
according to the bodily criterion you too ought to persist as a vegetable. But the
Standard Criterion implies no such thing. That is because it applies only in cases
where we have a person existing at one time and a person existing at another
time; and there is no person left behind after your upper brain is destroyed.
The usual view of personhood, anyway, is that a thing needs certain mental
properties, or at least the capacity to acquire them, in order to count as a
person. You might think that advocates of the bodily criterion should reject
any account of what it is to be a person that involves mental properties. But
there is no reason why they should. Such an account ought to be perfectly
compatible with the bodily criterion. There is no evident contradiction in
saying that beings with such-and-such mental properties are identical with
their bodies, or that they persist if and only if their bodies do. The bodily
criterion is not meant to be a definition of ‘person’, but rather an account of our
identity through time. Moreover, if the bodily criterion is to be in competition
with other accounts of personal identity, those accounts must all give identity
conditions for people in the same sense of the word; otherwise they will simply
be about different things. And what non-psychological account of personhood
could advocates of the bodily criterion give?
Now since a human vegetable has no mental properties and no capacity to
acquire them, it is presumably not a person. So the Standard Criterion implies
nothing about whether you could come to be a human vegetable—even
Eric T. Olson / 245
supposing that your body could come to be a vegetable. For the same reason,
it implies nothing about what happens to you when you die. Your body may
persist as a corpse when you die; but if a corpse is not a person, the Standard
Criterion does not imply that you persist as a corpse. It simply does not apply
here, for this is not a case in which we have ‘a person x existing at a time t and
a person y existing at another time t∗ ’. The Standard Criterion even allows
for you to have different bodies at different times, if at one of those times you
are not a person. Suppose you get a bad case of senile dementia—so bad that
you no longer count as a person. For all the Standard Criterion says, you may
end up as a demented non-person with a different body from the one you
have now.
The Standard Criterion tells us far less than the bodily criterion was supposed
to tell us. In fact the Standard Criterion is not really an account of what it
takes for us to persist through time at all, but only an account of what it takes
for us to persist as people—that is, what it takes for us to persist as long as we
remain people. It tells us nothing about what would happen to us if we ceased
to be people and became vegetables or corpses.
Now many philosophers assume that every person is a person essentially,
or at any rate that nothing could be a person at one time and a non-person
at another. That entails that no person could persist without remaining a
person. This assumption leads philosophers to ask what it takes for us to persist
through time by asking what it takes for a person existing at one time to be
identical with a person existing at another time; and it is this question that the
Standard Criterion purports to answer.
If being a person implies having mental properties, however, the assumption
that people are people essentially is questionable. It entails the contentious
metaphysical claim that whatever is a person at one time must necessarily
cease to exist (not just cease to be a person) if it loses all of its mental properties.
This is something that no advocate of the bodily criterion would accept.
The bodily criterion is supposed to imply that we may end as unthinking
vegetables or corpses, and hence (on the usual view of personhood) that we
are people only contingently. If we are to take the bodily criterion seriously,
then, we cannot assume at the outset that people are people essentially. We
cannot ask about our identity over time in a way that presupposes something
that is incompatible with the bodily criterion. Since the Standard Criterion
does make this assumption, it is not an acceptable statement of the bodily
criterion.
3.
We can avoid this problem by asking the right question about our identity
over time (Olson 1997: 21–6): Under what conditions is a person existing at
one time identical with anything, person or not, existing at another time? This
suggests what we might call the Revised Bodily Criterion:
Necessarily, any person x who exists at a time t is identical with something
y that exists at another time t∗ (whether or not y is a person then) if and
only if the thing that is x’s body at t is the thing that is y’s body at t∗ .
That is, you are that past or future being whose body then is the thing that is
your body now. If all people are essentially people, the standard and revised
criteria are equivalent; but we cannot assume that here.
But the Revised Criterion is not the bodily criterion either. Where the
Standard Criterion implied too little, the Revised Criterion implies too much.
The Standard Criterion implied nothing about whether you could one day
be a vegetable or a corpse. The Revised Criterion implies that you definitely
could not be; nor, for that matter, were you ever a foetus. And this is so even
if our bodies were once foetuses and may one day be vegetables or corpses.
Suppose you die peacefully and your body continues to exist as a corpse. Do
you still exist? The Revised Criterion says that you do if and only if something that
exists then has as its body the thing that is now your body: if the corpse has the
same body then as you have now, the corpse is you. Alternatively, if the corpse
is the very thing that is now your body, and if it is then the body of anything,
you are the thing whose body it then is. Otherwise you no longer exist.
(I assume that the corpse and the corpse’s body are the only candidates, after
your death, for being the thing that is now your body.) What, then, is the
corpse’s body? Or what thing is it the body of?
That depends on what it is for something to be someone’s (or something’s)
body. For any things x and y, under what conditions is x the body of y? This
important question is almost universally ignored. Philosophers speak freely
of ‘people’s bodies’ without saying a word about how they understand that
phrase. If there are such things as our bodies, however, and if it makes sense to
ask how we relate to them, this question must have an answer. It may be a vague
or a messy or a complicated answer: for instance, there may be pairs of objects
such that, for one reason or another, it is neither definitely true nor definitely
false that one is the body of the other. But a vague or messy or complicated
Eric T. Olson / 247
answer is still an answer. There must be such an account if the bodily criterion
of personal identity is to have any content at all. (Note that an account of
what it is for someone to be embodied—to have a body—does not yet tell us
which thing one’s body is. Nor does an account of what it is for a body to be
someone’s body, until we know what counts as a ‘body’ in the relevant sense.)
The most common account of what it is for something to be someone’s
body is roughly this: your body is that material object by means of which you
perceive and act in the physical world. More precisely, it is the largest such
object: your left hand is not your body, even though you use it to perceive
and act. You get knowledge of the physical world only by the way it impinges
upon your body, via your sense organs. And you can affect the physical world
only by affecting your body. To move your pencil, you have to move your
body; but you can move your body directly, just by intending to, and without
thereby moving anything else (save parts of it). I once called this the Cartesian
Account of what makes something someone’s body—not because it implies
Cartesian dualism, but because Descartes held something like it.
The Cartesian Account is probably not right as it stands: for one thing, it
appears to conflict with the conviction that limbs and organs that you can
neither move nor feel may still be parts of your body (Olson 1997: 145–8).
But let us ignore these worries and suppose that something like the Cartesian
Account is correct. The Revised Criterion then amounts to this:
Necessarily, any person x who exists at a time t is identical with something
y that exists at a time t∗ if and only if the largest material object that x can
move and feel directly at t is the largest material object that y can move
and feel directly at t∗ .
A moment’s reflection shows that this is not an acceptable statement of the
bodily criterion. A corpse cannot learn about the physical world by any means
whatever, let alone by virtue of changes in any particular material object. Nor
can it move any material object just by intending to move it, for it cannot
intend anything. Nothing is the corpse’s body on the Cartesian Account. For
similar reasons, the corpse is not the body of anything.
The Revised Criterion thus implies that you would cease to exist if you lost
all your mental capacities, for without them you would have no body, and the
Revised Criterion rules out your surviving without a body. You could never
come to be a corpse, or a vegetable, even if the thing that is now your body
came to be a corpse or a vegetable. Nor were you ever a foetus. During the
first half or two-thirds of its gestation, anyway, a human foetus has no mental
properties, and therefore no body. Nor was any foetus your body, for a foetus
(on the Cartesian Account) is not the body of anything. So you did not exist
when the foetus that produced you was developing in the womb, even if the
thing that is now your body existed then. You don’t always go where your
body goes. Your body—the thing that is now your body—exists at times
when you don’t exist. It follows that you are not identical with your body.
The Revised Criterion is actually incompatible with the idea that people are
their bodies. This is clearly not what the bodily criterion was supposed to be.
4.
We might try to avoid these troubles by moving to the second way of
understanding the bodily criterion, namely the idea that people are their
bodies. Call this Corporealism. As we have seen, the Standard Criterion does not
imply Corporealism, since for all it says your body might outlive you; and the
Revised Criterion is incompatible with it.
There are at least two different versions of Corporealism. Weak Corporealism
says that you are identical with that thing that is now your body. That is:
Necessarily, for any person x existing at a time t, x is the thing that is x’s
body at t.
Though this does not explicitly say what it takes for us to persist through time,
it implies that you are that past or future being that is identical with the thing
that is now your body:
Necessarily, for any person x existing at a time t and any y existing at
another time, x = y if and only if y is the thing that is x’s body at t.
So your identity is the identity of your body insofar as you and your body are
one and the same.
I doubt whether Weak Corporealism is what anyone means by the bodily
criterion, for Weak Corporealism allows for a person to survive without a
body. It even allows for people to exchange bodies. At any rate it does so on
the Cartesian Account of what it is for something to be someone’s body.
Suppose, as Weak Corporealism has it, that you are the thing that is now
your body. That is, you yourself are the largest material object through which
you now perceive the world, and which you can now move just by intending
to move. Now imagine that surgeons cut the sensory and motor nerves that
Eric T. Olson / 249
connect your brain with the rest of you, but leave intact the connections that
direct your life-sustaining functions. The resulting being would be alive but
entirely blind, deaf, numb, and paralysed. (I take this to be physically possible,
though it would no doubt be a delicate business in practice.) Presumably
you could survive this: the resulting being would be you. At any rate the
bodily criterion shouldn’t rule it out. On the Cartesian Account, however,
you would thereby cease to have a body, for there would no longer be any
material object that you could move or feel. Thus, you could survive without
having any body at all, even if you are identical with your body. This is not
because you are only contingently identical with your body, but because your
body only contingently has the property of being your body.
It is easy enough to alter the story so that you get a new body. Let the surgeons
hook up the motor and sensory nerves leaving your brain to the appropriate
nerves of my spinal cord (and my optic nerves, etc.) by wires or radio links, so
that you can move my limbs and perceive via my sense organs in just such
a way that the thing that was once my body comes to be your body. Adding
appropriate links between my brain and your spinal cord would make the thing
that was previously your body, my body. We could exchange bodies. You could
even outlive your body: if we destroyed the thing that comes to be your body in
this way, you would still exist. This is precisely the sort of thing that the bodily
criterion was meant to rule out. Yet it is compatible with Weak Corporealism.
We could avoid this by moving to Strong Corporealism, the view that each
person must, at every time when she exists, be identical with the thing that is
her body then:
Necessarily, for any person x existing at any time t and any y existing at
any time t∗ , x = y if and only if y is the thing that is x’s body at t and x is
the thing that is y’s body at t∗ .
That is, you are identical with that thing that is now your body and which is,
at every other time when it exists, its own body. The persistence of your body
as such is necessary and sufficient for you to persist.
This rules out your existing without a body or exchanging bodies. However,
it also rules out your existing at a time when you have no mental capacities.
For you would then be unable to act or perceive, and so nothing, not even
yourself, would count as your body. Strong Corporealism therefore implies
that you were never a foetus and could never come to be a vegetable or a
corpse. In fact your body was never a foetus and could never come to be a
vegetable or a corpse, since you are your body and you couldn’t do so. For
that matter, Strong Corporealism implies that both you and your body would
perish if surgeons disconnected the sensory and motor nerves leaving your
brain. No one sympathetic towards the bodily criterion would accept that.
Strong Corporealism is not what anyone meant by the bodily criterion either.
5.
We have not yet found a statement of the bodily criterion that does what that
view is supposed to do. The Revised Criterion implied that you could never
be a foetus or a vegetable or a corpse, even if your body is, and ruled out
our being identical with our bodies. Weak Corporealism allowed for people to
survive without a body and to change bodies. Strong Corporealism implied
that neither people nor their bodies were ever foetuses, or could come to be
vegetables or corpses, or could survive total paralysis. None of these views is
the bodily criterion we learned about as students.
Now many of these difficulties can be laid at the door of the Cartesian
Account of body ownership. So we might try a different account of what
makes something someone’s body. There are not many to choose from. The
literature offers two main rivals to the Cartesian Account.
Shoemaker counts a thing as your body if its relation to you is near enough to
what the Cartesian Account specifies.1 Someone who is completely paralysed
may still have a body, he says, if there is something that she would be able to
move just by intending to, were it not damaged or disabled. Roughly, x is y’s
body if and only if y would be able to move and feel x in the ‘Cartesian’ way if x
were intact.
This would rule out your surviving without a body or exchanging bodies
with someone else as a result of your sensory and motor nerves’ being cut
or re-routed, making Weak Corporealism look more like what the bodily
criterion was supposed to be. But Shoemaker’s account still appears to imply
that a foetus, vegetable, or corpse neither has a body nor is the body of
anything. A foetus is intact, yet until at least mid-gestation it cannot move or
feel anything, or be moved or felt in the relevant way. There is nothing the
foetus relates to in anything like the Cartesian way, and not just because a
1
(1976: 115). What he actually offers is an account of what it is to be embodied, not of what it
is for something to be someone’s body; so I extrapolate.
Eric T. Olson / 251
few nerves are cut. Nor is there anything that a vegetable or a corpse relates
to in anything like the way in which you relate to the thing you can now
move and feel. Nothing there is in any way capable of doing any moving or
feeling. If a foetus, vegetable, or corpse neither has a body nor is the body of
anything, then Weak Corporealism still implies that you could exist without
a body (assuming that the thing that is now your body was once a foetus or
could one day be a vegetable or corpse).
Even if Shoemaker’s account could be amended to avoid these problems, it
faces a more serious obstacle: it implies all by itself that we are identical with
our bodies. The implication is not merely that we are identical with our bodies
if the bodily criterion is true. On Shoemaker’s account, the very idea of what
it is for something to be someone’s body implies that we are our bodies. Or at
least it does so given an assumption that most philosophers, whether friends
or foes of the bodily criterion, would accept, namely that we are made of the
same matter as our bodies. That is, we are things of flesh and blood, and not
immaterial souls, bundles of perceptions, or what have you; and we are no
larger or smaller than our bodies. Call this Standard Materialism. It implies that
you yourself are the largest thing that you can move and feel in the direct
Cartesian way, or that you would be able to move and feel in that way if it were
intact. (At any rate you are one such thing. I will return to this complication in
the next section.) Hence, on Shoemaker’s account of what it is for something
to be someone’s body, you qualify as your own body on Standard Materialism.
But the bodily criterion is not supposed to follow from Standard Materialism.
Many Standard Materialists, including Shoemaker himself, deny that we are
our bodies and reject the bodily criterion root and branch. The view that we
are made of the same matter as our bodies but are nonetheless numerically
different from them is widely held. Shoemaker’s account of body ownership
would make this view self-contradictory. I take this to be unacceptable.
Tye (1980) proposes that a person’s body is the bearer of those physical and
spatio-temporal properties truly predicable of her in ordinary language. The
idea is this: we say truly, when we aren’t doing philosophy, that Margaret
Thatcher is human (in the biological sense), is female, is over five feet tall,
and so on. Even those who think that in the strict and philosophical sense
Thatcher is an immaterial substance or a bundle of perceptions will agree
that this naive description is in some sense accurate. What makes it accurate,
Tye says, is that these things are strictly true of the material object that is
Thatcher’s body. So that is what her body is.
This view is in some ways more promising than the Cartesian Account
and its variants. But it too makes Standard Materialism imply that we are
identical with our bodies; for in that case those physical and spatio-temporal
properties truly predicable of you in ordinary language really are, in the strict
philosophical sense, properties of you.
So neither of these alternative accounts of body ownership is any help in
formulating the bodily criterion.
6.
The content of the bodily criterion depends on what it is for something to be
someone’s body. We want an account of this relation that would enable the
bodily criterion to play the role it is supposed to play in debates about personal
identity. The account should not imply that most of those who claim to reject
the bodily criterion are in fact committed to it; in particular, it shouldn’t make
Standard Materialism a version of the bodily criterion. None of the accounts
of body ownership currently on offer meets this standard. Perhaps we can
do better by inventing one of our own. We seem to need something more
stringent than those we have considered: not just anything that you would be
able to move and feel in the Cartesian way if it were intact, or that bears those
physical and spatio-temporal properties truly predicable of you in ordinary
language, counts as your body. Here are three suggestions.
1. Your body is that body that bears those physical and spatio-temporal
properties truly predicable of you in ordinary language, or that you would be
able to move and feel in the Cartesian way if it were intact. The bodily criterion
would then be the claim that you are that body. Standard Materialists may agree
that you bear the relevant physical and spatio-temporal properties, but disagree
about whether you are a body; so they can disagree about whether you are
identical with your body, and hence about whether the bodily criterion is true.
But what is a body? Not just any material object, or even any connected
material object that contrasts with its surroundings, for according to Standard
Materialism you are a body in that sense. A human body, then. (Set aside worries
about non-human people.) But what is a human body? If it is just the body of
a human person, then the proposed amendment—replacing ‘thing’ in Tye’s
account with ‘body’—adds nothing.
Perhaps a human body is a human organism. The bodily criterion would
then be the claim that each of us is identical with a human animal: Animalism,
as it is sometimes called.
Eric T. Olson / 253
Animalism does have many of the features that the bodily criterion is
supposed to have. It is neither trivially true nor trivially false. It looks
consistent with Standard Materialism, but does not follow from it. And it
implies something about what it takes for us to persist through time, namely
that our persistence conditions are those of animals.
But Animalism is not the bodily criterion. It does not belong to the idea of
one’s body that it must be an animal, let alone a human animal. At any rate,
people often say that someone could have a partly or wholly inorganic body:
our bodies might one day include artificial limbs, or even be entirely robotic. But
no animal—no biological organism—could be partly or wholly inorganic.
If you cut off an animal’s limb and replace it with something inorganic, the
animal only gets smaller and has an inorganic prosthesis attached to it. So the
view that people might one day have partly or wholly inorganic bodies allows
for a person to be identical with his body without being an animal, or to be an
animal without being identical with his body. Many philosophers even say that
an animal’s body is always a different thing from the animal itself. They say that
an animal stops existing when it dies, but unless its death is particularly violent
its body persists as a corpse; or they say that an animal, by virtue of large-scale
changes of parts, can have different bodies at different times. That would mean
that no one could be both an animal and identical with his body. Animalism
would be incompatible with the bodily criterion. I am in no position to say
whether these claims about bodies are true. But they are not obviously false;
and they don’t seem to conflict with the very idea of one’s body.
So animalism is not a thesis about how we relate to our bodies. The idea
of one’s body doesn’t come into it. One could of course stipulate: ‘By ‘‘x’s
body’’ I shall mean the biological organism that bears those physical and
spatio-temporal properties truly predicable of x in ordinary language.’ But it
would be a stipulation, and no part of what anyone ordinarily means by ‘body’.
The bodily criterion was supposed to be about bodies in some ordinary sense
of the term. Thus, animalism is not the bodily criterion.
2. Some say that a person’s body by definition lacks any mental prop-
erties (Wiggins 1976: 152). I suppose it is absurd to suppose that Bertrand
Russell’s body denied the existence of God. The reason, on this view, is that
Russell’s body obviously did not have that belief. So we might say that your
body is that bearer of those physical and spatio-temporal properties truly
predicable of you in ordinary language (or that thing that you can move and
feel in the relevant way) that has no mental properties. Standard Materialists
could then deny that we are our bodies.
But this would make Corporealism trivially false. If part of what it is to
be one’s body were to be something that lacks any mental properties, then
it would be evident that we are not our bodies, for it is evident (eliminative
materialism aside) that we have mental properties. Yet no one thinks that the
bodily criterion is trivially false.
3. Perhaps your body is that bearer of those physical and spatio-temporal
properties truly predicable of you in ordinary language that has brute physical
(non-psychological) identity conditions.2 Many philosophers say that our
identity through time consists in some sort of mental continuity. They will
deny that we are identical with our bodies in this sense, even if they are Standard
Materialists. The Standard Materialists who take us to have brute physical
identity conditions are those who accept the bodily criterion. The bodily
criterion would be the claim that you are identical with your body in this sense.
This is the most promising attempt to state the bodily criterion that I know
of. For all that, it faces two serious technical problems. The first arises from
a snag in Tye’s account of body ownership and others that I have ignored up
to now.
What thing is it that bears those physical and spatio-temporal properties truly
predicable of you in ordinary language? Presumably a certain human organism
does. But many philosophers believe that other things do too. Any Standard
Materialist who denies that you are your body believes that the same matter can
make up more than one material object at once. That suggests that a certain
hunk of flesh or mass of matter now stands to ‘your’ organism as your body
stands to you, or as (according to a popular view) a hunk of clay stands to a clay
statue made from it. Owing to metabolic turnover, an organism coincides with
different masses of matter at different times. If, as many believe, you coincide
both with a mass of matter and with an organism, then there are at least two
things that bear those physical and spatio-temporal properties truly predicable
of you in ordinary language and have brute physical identity conditions.
This could mean either of two things. If your body must be the only thing
that has the relevant features, then nothing is your body. Yet both those who
accept the bodily criterion and those who deny it agree that we have bodies.
Or we might say that anything with the relevant features is your body—one
of your bodies, that is. So the animal and the mass of matter coinciding with
you would both be bodies of yours: you would have at least two bodies. Worse,
2
I owe this inspired suggestion to Katherine Hawley.
Eric T. Olson / 255
those bodies would have different identity conditions. Your ‘animal body’
would continue to exist as long as it remains biologically alive (so I should
argue, anyway). Your ‘matter body’ would remain your body only as long
as that mass of matter remains in human form: a fraction of a second. The
moment your metabolism assimilates or expels a particle, the mass would
begin to disperse. In a few months’ time it would be widely scattered. (If a
mass of matter must be connected, it exists only momentarily.) Which body
would it be that, according to the bodily criterion, is you, or whose identity
determines your identity?
Of course, there may be no such things as masses of matter. There may
be only one thing with brute physical identity conditions that bears those
physical and spatio-temporal properties truly predicable of you in ordinary
language. But the intelligibility of the bodily criterion ought not to turn on
this deep metaphysical issue. This problem plagues other accounts of what
makes something someone’s body as well: many philosophers believe that
there is more than one largest thing that you would be able to move and feel
in the Cartesian way if it were intact.
Here is the second problem. Some philosophers say that we are human
organisms, but deny that those organisms have any identity conditions, brute
physical or otherwise (Merricks 1998). Our identity over time does not consist
in anything other than itself. Nearly everyone would call this a version of the
bodily criterion. But it would not be a version of the bodily criterion on the
current proposal. Perhaps the bodily criterion is the view that we are those
things that have the relevant physical and spatio-temporal properties and lack
psychological identity conditions. But consider the view of Lowe (1996: 35 f.,
41 ff.) and Baker (2000) that we have no identity conditions, yet coincide with
human organisms numerically different from ourselves. Lowe and Baker deny
that this is a version of the bodily criterion, and no one has ever disputed this
description. Yet it would be a version of the bodily criterion on our revised
proposal, implying that we ourselves are our bodies.
I can see no good solution to either of these problems.
7.
The accounts of body ownership that Descartes, Shoemaker, and Tye propose
imply that the bodily criterion is not the controversial view that many
materialist philosophers oppose, but rather something that follows trivially
from Standard Materialism. And our attempts to find an account of body
ownership that avoided this problem merely turned up more trouble. So we
have not yet found the bodily criterion.
You might conclude from this that there is no one bodily criterion, but
rather a number of different bodily criteria that philosophers have failed
to distinguish. The problem, you might think, is merely that those who
ask whether we are our bodies, or whether our identity through time is
determined by the identity of our bodies, have not said which notion of ‘body’
they mean. But none of these versions is acceptable as a statement of the bodily
criterion. And if they really were versions of the bodily criterion, it ought to
be possible to say explicitly what they are versions of, and why. There ought
to be a more general thesis about how we relate to our bodies that the various
versions entail. This would be the bodily criterion we seek. Yet we have been
unable to find such a thesis. What could it mean, then, to say that Animalism
(for instance) is a version of the bodily criterion?
If there is any unifying idea behind the various views that are called versions
of the bodily criterion, it is that we are material things whose identity consists
in some sort of brute physical continuity:
Necessarily, any person x who exists at a time t is a material thing, and is
identical with any y that exists at another time t* if and only if x relates at
t in some brute physical way to y as it is at t*.
Given an appropriate specification of the ‘brute physical way’, this claim seems
perfectly intelligible. It is an important position in debates on personal identity.
And it seems to follow from most views that are called versions of the bodily
criterion.3 Is this what we were after?
I think not. The Anodyne Bodily Criterion, as we might call it, is the bodily
criterion in name only. It doesn’t mention bodies. That notion doesn’t come
into it. That was its virtue: if it were about our bodies and how we relate
to them, it would presumably be equivalent to one of the unsatisfactory
formulations we considered earlier. We needn’t talk about our bodies in order
to say that we are material and that our identity consists in brute physical
continuity. Introducing that notion only makes trouble. You can call brute
physical continuity ‘bodily’ continuity if you like, but you will only come to
grief if you try to understand it in terms of things called people’s bodies. If the
3
Merricks’s view, mentioned at the end of the previous section, is an exception.
Eric T. Olson / 257
bodily criterion is nothing more than the Anodyne Criterion, it is not a thesis
about bodies at all. Stated carefully, it is not the view that we are identical with
our bodies, or that our identity consists in the identity of our bodies. But then
we might as well say that there is no bodily criterion of personal identity.
8.
Let me now venture a diagnosis of the problem. I suspect that the reason
we have been unable to find the bodily criterion of personal identity is that
there is none. Why did we all think there was? Presumably it was because we
took phrases like ‘Wilma’s body’ to be names of objects. To say that Wilma
has a muscular body, we supposed, is to state a relation between Wilma and
a certain muscular object. Having a muscular body is like having a powerful
car, except that one has one’s body in a more intimate sense than one has
one’s car. Without some sort of assumption like this there can be no question
of how we relate to our bodies, and no bodily criterion worthy of the name.
Our inability to state the bodily criterion, if nothing else, suggests that this
linguistic assumption is false.
If this is hard to believe, consider a closely related term: mind. Does the
phrase ‘Wilma’s mind’ purport to name an object? When we say that Wilma
has a clever mind, are we stating a relation between Wilma and a certain clever
object? Is having a clever mind like having a clean shirt, except that one has
one’s mind in a more intimate sense than one has one’s shirt? I think most
philosophers would say no. Otherwise many metaphysical questions would
arise: What is this clever thing made of? How big is it? How clever is it? (As
clever as Wilma?) How exactly does it relate to Wilma? (Is it identical with her?
Or a part of her?) What is it about this mind that makes it Wilma’s, rather than
someone else’s? Could it have been the mind of someone else? And so on. These
are questions no one asks. But if minds were things that people in some sense
own, they would have to have answers.
This is not to deny that real questions are sometimes put in the language of
people’s minds. We may ask, for instance, whether a person’s identity consists
in the identity of her mind. But no one means this literally. It is only an
informal way of asking something we should put more carefully in other
terms: whether a person’s identity consists in some sort of psychological unity
and continuity, for instance. When we speak carefully, apparent reference
to the mind as an owned mental object disappears. To speak of someone’s
mind is presumably just to speak in a loose and colourful way of her mental
properties, where the notion of a mental property can be explained without
talking about things called minds. To say that Wilma has a clever mind is to
just say that Wilma is clever (with the implication, perhaps, that cleverness is a
mental property: though Wilma may be tall, we don’t say that her mind is tall).
I suggest that talk of people’s bodies is like talk of people’s minds. Our bodies
are no more physical things that we somehow have than our minds are mental
things that we have. To speak of a person’s body is just to speak in a loose
and colourful way about her ‘bodily’ properties, where the notion of a bodily
property can be explained without talking about things called people’s bodies.
To say that Wilma has a muscular body is just to say that Wilma is muscular.
(With the implication, perhaps, that being muscular is a bodily property. What
makes it absurd to say that Russell’s body denied the existence of God may be
the absurd implication that denying the existence of God is a bodily property.)
The human body is the subject matter of anatomy and physiology only in the
notional sense that the mind or psyche is the subject matter of psychology. The
true subjects of those sciences are the anatomical, physiological, and mental
properties of human beings.4 When we speak carefully, reference to the body
as a physical object whose relation to its owner is open to question disappears.
That is why the only acceptable statement of the bodily criterion is not about
bodies. We should no more expect there to be a bodily criterion of personal
identity relating people to their bodies than we should expect there to be a
‘mental’ criterion of personal identity relating people to things called minds.
If this is right, there is no meaningful question of how we relate to our
bodies, any more than there is a real question of how we relate to our minds.
There may be real questions that the phrase ‘how we relate to our bodies’
suggests: whether we are made entirely of matter, for instance, or whether
our identity through time consists in some sort of brute physical continuity.
But they are not really questions about things called bodies. There is no bodily
criterion of personal identity strictly so called. More generally, there is no
concept of a body as a thing that each of us has.
This would have wide implications. The notion of one’s body as an owned
object plays an essential role in a great deal of philosophical thought. The
landscape would look rather different without it.
4
Versions of this view are defended in van Inwagen (1980), Rosenberg (1998: 71–3), and Olson
(1997: 152 f.).
Eric T. Olson / 259
I am not confident that this diagnosis is correct. Perhaps I simply haven’t
been clever enough to understand the bodily criterion. But there is clearly
some sort of problem here. Those who believe that there is a bodily criterion of
personal identity ought to tell us what they think it is.
References
Ayer, A. J. (1936), Language, Truth, and Logic (London: Gollancz).
Baker, L. R. (2000), Persons and Bodies (Cambridge: CUP).
Garrett, B. (1998), Personal Identity and Self-Consciousness (London: Routledge).
Lowe, E. J. (1996), Subjects of Experience (Cambridge: CUP).
Merricks, T. (1998), ‘There are No Criteria of Identity over Time’, Noûs, 32: 106–24.
Olson, E. (1997), The Human Animal (New York: UP).
Rosenberg, J. (1998), Thinking Clearly about Death (2nd edn., Indianapolis: Hackett).
Shoemaker, S. (1976), ‘Embodiment and Behavior’, in A. Rorty (ed.), The Identities of
Persons (Berkeley and Los Angeles: University of California Press).
Tye, M. (1980), ‘In Defense of the Words ‘‘Human Body’’ ’, Philosophical Studies, 38:
177–82.
van Inwagen, P. (1980), ‘Philosophers and the Words ‘‘human body’’ ’, in van Inwagen
(ed.), Time and Cause (Dordrecht: Reidel).
Wiggins, D. (1976), ‘Locke, Butler, and the Stream of Consciousness’, in A. Rorty
(ed.), The Identities of Persons (Berkeley and Los Angeles: University of California Press).
INDEX
abstract objects apriori infallibilism 45–9, 50–2

and the Caesar problem 175–6 apriori necessities 51
numbers as 116, 117–19 apriori red flags 50
and the Standard Conception of apriori reflection 50, 52, 53–5
metaphysical necessity 21 Aristotelianism, and mathematical
action structuralism 135–6
and the ‘I’ self-concept 226–7 Aristotle, on mathematical
and the solipsist as rational agent 233 methodology 119–20
advanced modal claims, and modal Armstrong, D.M. 2, 26, 27, 118
fictionalism 66–71 axiom of extensionality 5, 151, 152, 165, 167
agnosics, and sortal concepts 209–10 and the Caesar problem 200
analytic necessity 15–16 Ayer, A.J. 243
analytical methodology, and the
Non-Standard Conception of backwards causation, and chance 104
metaphysical necessity 27–8 Bafflement Argument against materialism,
Animalism, and bodily criterion 252–3, 256 and personal identity 8, 224–5, 236–8,
Anodyne Bodily Criterion 256–7 239–40
Anselmian God, and the Non-Standard Baker, L.R. 255
Conception of metaphysical basic truth, and modal infallibilism 40–56
necessity 30–1 Benacerraf, Paul 136, 159, 161, 162
ante rem structuralism 4–5, 109, 110, 111, 112, and the Caesar problem 186–7, 188
113–21 ‘What Numbers Could Not Be’ 109, 113,
and the Caesar problem 125 114, 122
and eliminative structuralism 136, 143 binding, and sortal concepts 7, 205–14
and the governance of identity 164, 165, Blackburn, Simon, Oxford Dictionary of Philosophy
169–72 definition of essence 115, 116
and haecceities 140 bodily criterion of personal identity 8–9,
Hellman’s objections to 141–3 242–59
and the identity of indiscernibles 5, 132, 133, ‘animal bodies’ and ‘matter bodies’ 254–5
139 and Animalism 252–3, 256
and the individuation problem 136 Anodyne Bodily Criterion 256–7
and Keränen’s principle (STR) 169–70 and Corporealism 248–50, 249–50, 254
Keränen’s reply to Shapiro 5–6, 146–63 and minds 9, 257–8
aposteriori infallibilism 47, 49 Revised Bodily Criterion 246–8, 250
aposteriori realism, and the Caesar Shoemaker’s account of 9, 250–1
problem 177 Standard Bodily Criterion 243–5
appercetive agnosics, and sortal and Standard Materialism 251, 252, 253,
concepts 209–10 256
262 / Index
bodily criterion of personal identity (cont.) corpses, and bodily criterion 245, 246, 247, 251,
Tye’s account of 9, 251–2, 254 253
Brock, Stuart, objection to modal correct conceivability, and metaphysical
fictionalism 58–63 necessity 22–3
Burgess, J. 131 credences, and realism about chance 84–8
cross-structural identity in
Caesar problem 6–7, 174–202 mathematics 121–31
dimensions of the problem 176–7
epistemological 6, 177–80 Dedekind, Richard 110–11, 113, 143
and mathematical structuralism 125, 127, delineation thesis, and sortals 7, 203–5, 209,
128, 129 214, 215, 216, 217
meaning-theoretic 6–7, 177, 185–9, 198 demonstratives, and sortal concepts 7, 215–17
metaphysical 6, 177, Descartes, René 228, 229, 255
180–5 desire
neo-Fregean solution 194–9 and the ‘I’ self-concept 226
and object and identity 199–200 and the solipsist as rational agent 233
supervaluationist solution 190–3 determinism, and realism about chance 75,
what the problem is 175–89 78–9, 83, 91, 92
Campbell, John, ‘Sortals and the Binding Differential Class, of metaphysical
Problem’ 7, 203–18 necessity 20–1, 30, 35–6
cardinal numbers, identity of 6 discrimination, and the epistemological
Carnap, R. 162, 196 Caesar problem 177, 178–9
Cartesian Account, of bodily criterion 247, distance, and basic metaphysics 43
248, 249, 250–1, 252, 255–6 Divers, John and Hagen, Jason, ‘The Modal
Cartesian dualism, and personal identity 8, Fictionalist Predicament’ 3–4, 57–73
222–3 dualism
chance see realism about chance and bodily criterion 9, 243
circumscription problem, and the Caesar and mathematical structuralism 115
problem 193, 196 and personal identity 8, 222–3
classical mereology, and the Standard Dummett, M. 17
Conception of metaphysical necessity 19 and the Caesar problem 191, 192
common sense, and the epistemological and ‘mystical’ structuralism 114, 121
Caesar problem 177, 178, 179
conceptual necessity 15 Elga, A. 99, 101
concrete objects, and the Caesar problem 181, eliminative structuralism 136, 149, 161
182 enduring runabouts 2, 43–4
conditional probability, and chance 103–4 and placeshifting 44, 45
consciousness, and self-consciousness 226 and shapeshifting 44–5
contingency, limits of 1–2, 13–38 epistemological Caesar problem 6, 177–80
contingent apriori, and the Non-Standard epistemology
Conception of metaphysical necessity and metaphysics (ep-&-met tendency) 3,
28 53–4
contingent properties and personal identity 226
and mathematical structuralism 115 essential idexical role, of ‘I’ thoughts 225–7,
of numbers 121 233
Corporealism, and bodily criterion 248–50, essential properties, and mathematical
249–50, 254 structuralism 115–17, 120–1
Index / 263
Euclidean geometry and ontological realism 161, 162
and the governance of identity 169, 170–1, and realist structuralism 5–6, 149, 159
172 and the trivializing objection 150, 155–7
and mathematical structuralism 131–2, Hagen, Jason see Divers, John and Hagen,
138 Jason
Evans, G. 8 Hale, B.
generality constraint and the Caesar and the Caesar problem 198
problem 196–8 objection to modal fictionalism 59, 60, 63,
existence, and personal identity 227–30 64, 65–6, 71
happiness, and unclarity of personal
faithfulness constraint, and mathematical identity 22
structuralism 111–12, 123, 125, 129 Hellman, Geoffrey 111, 116, 131, 136
fictionalism see modal fictionalism Mathematics Without Numbers 109
Field, Hartry 21 modal eliminative structuralism 143
Fine, Kit 24 objection to ante rem structuralism
finite cardinal structures 131, 132, 141–2 141–3
and Keränen’s principle (STR) 169 Hossack, Keith, ‘Vagueness and Personal
finite ordinals, and mathematical Identity’ 7, 221–41
structuralism 122, 129, 131 Hume, D. 228
fit, and realism about chance 94–100, Humean reflection, on metaphysical
100–3 necessity 15
foetuses, and bodily criterion 246, 247–8, Hume’s Principle 6
249–50, 250–1 Humphreys, G.W. 209–10, 212
Frege, G. 125
and the Caesar problem 6–7, 175–7, 179, ‘I’ self-concept, and personal identity 8,
185, 191, 194–9, 200 225–7, 233
frequentist analyses of chance 89–91 identity
functionalist conception of human and the Caesar problem 6–7, 174–202
beings 233–4 cross-structural identity in
mathematics 121–31
Garrett, B. 243 governance of 164–74
generality constraint, and the Caesar and mathematical structuralism 131–44
problem 196–8 problem for realist structuralism 5–6,
genuine modal realism (GMR), and modal 146–63
fictionalism 3, 57–8, 63–72 trivial accounts of 155–7
genuine modality 40–1, 54 see also personal identity
genuine necessity 41 indeterminacy, and mathematical
genuine possibility, and intelligibility 55–6 structuralism 122–3, 127, 128–9,
Gestalt principles, and sortal concepts 7, 130
207–8 indeterministic statistical phenomena, and
golden mountain example, and the realism about chance 78–84
Non-Standard Conception of indiscernability, and mathematical
metaphysical necessity 31–2 structuralism 131–44
individuation
Haecceitism and the Caesar problem 6–7, 174–202
and the identity of indiscernibles 139–40 and the governance of identity 165–6,
and the Leibniz Principle 154 168–9
264 / Index
individuation task Lewis, David 32, 40, 42–3, 44–5, 46–7, 49
and the identity of indiscernibles 134–7, and advanced modal claims 66–7
139, 141 ‘best-system’ analysis of chance 4, 76–7, 85,
and Keränen’s reply to Shapiro 150–1 91–4, 100–1, 104
infallibilism, model infallibilism and basic chance and the problem of fit 94–7, 99,
truth 40–56 100–1
informed/correct conceivability, and Lichtenberg, G.C., and personal identity 228,
metaphysical necessity 22–3 229, 230, 231, 232, 234, 235–6, 237–8, 239
intelligibility, and metaphysical location, as a binding parameter 210
possibility 2–3, 40–56 Locke, John 226, 231
intrinsic property possession, and the Caesar logical modality 40
problem 195–6 logical necessity 15–16
intrinsics, and basic metaphysics 42–3 Lowe, E.J. 255
Inwagen, Peter van 21 Ludwig, K. 217
isomorphism theorem, and mathematical
structuralism 139 MacBride, Fraser 1–9, 40
‘The Julius Caesar Objection: More
Julius Caesar objection see Caesar problem Problematic than Ever’ 6–7, 174–202
McDermott, M. 89–90
Kant, I. 25, 222, 228, 231 McGee, Vann 124, 129
Kastin, Jonathan 112, 114, 116, 124, 146 Maddy, Penelope 118
Keränen, Jukka 112 Madell, Geoffrey 224
and eliminative structuralism 136 materialism
‘Identity Problem for Realist Structuralism and bodily criterion 9, 243, 251, 252, 253 256
II’ 5–6, 131, 146–63, 164, 165, 171, and personal identity 8, 221–2, 223–5,
172 236–8, 239–40
Keränen’s principle (STR) 139, 140, 141, 148, mathematical ontology, realist theory
152, 165, 166, 167, 168–70 of 160–3
and mathematical structuralism 131, mathematical structuralism 4–5, 109–46
132–4, 139 cross-structural identity 121–31
and Shapiro’s ‘The Governance of identity and indiscernability 131–44
Identity’ 6, 131, and mathematical properties 115
164–74 model eliminative structuralism 109
Kitcher, P. 186 see also ante rem structuralism
Kreisel, G. 130 mathematical truths, and metaphysical
Kripke coherence 56 necessity 15
Kripke truth, and modal infallibilism 2–3, 46, mathematicians
47, 48, 49, 50, 56 and natural numbers 124–5, 125–7, 129,
Kripkean necessities, and metaphysical 130–1
necessity 15, 18, 21 and philosophers 110
meaning-theoretic Caesar problem 6–7, 177,
laws of nature, and metaphysical 185–9, 198
necessity 34–5, 36, 37, 38 Mellor, D.H., and realism about chance 80, 81,
Leibniz, G.W. 184 84–5, 90
Leibniz Principle, and realist membership properties, and the governance
structuralism 148, 149, 152, 153–5 of identity 168, 168–9, 170
Lepore, E. 217 mereological universalism 43
Index / 265
metaphysical Caesar problem 6, 177, 180–5 nominals, and sortal concepts 216–17
metaphysical conception, of the Leibniz Non-Standard Conception of metaphysical
Principle 153–4 necessity 1, 2, 20, 21–32, 37–8
metaphysical modality 16, 29 coherence of 29–32
metaphysical necessity 1–2, 13–38 and the informal explanation 28–9
defining 13–14 and objections to the Standard
Differential Class of 20–1, 30, 35–6 Conception 33, 36
informal elucidation of 14–17 and physical necessity 37
Non-Standard Conception of 1, 2, 20, 21–8, Noonan, Harold 63
33, 36, 37–8
and physical necessity 36–7 object and identity, and the Caesar
Standard Conception of 1, 2, 17–32, 29, problem 199–200
32–6, 37–8 objects see abstract objects; sortal concepts
metaphysical possibility 17 Olson, Eric T., ‘Is There a Bodily Criterion of
and basic metaphysics 41–4, 53 Personal Identity?’ 8–9, 242–59
and intelligibility 2–3, 40–56 ontological realism, identity problem for 6,
metaphysics, and the governance of 160–3, 172
identity 165 ontology
Mill-Ramsey-Lewis (MRL) account of the laws and the governance of identity 164, 165
of nature, and metaphysical and metaphysical necessity 28, 35, 37
necessity 34–5, 36 orienting attention, sortals as 214–17
minds, and bodily criterion of personal other minds theory, and personal
identity 9, 257–8 identity 234–6
minimalism constraint, and mathematical The Others, and the Non-Standard
structuralism 111, 112, 113, 119, 129 Conception of metaphysical
modal eliminative structuralism 143 necessity 21–2, 24, 25, 26, 28, 29
modal fictionalism 3–4, 57–73
and advanced modal claims 66–71 Pairing axiom, and metaphysical
Brock–Rosen objection to 58–63 necessity 18–19, 19–20, 24–5, 26
and genuine modal realism (GMR) 3, 57–8, Parsons, Charles 123–4, 143
63–72 particulars, and the Non-Standard Conception
modal logic, and modal fictionalism 60–2, 72 of metaphysical necessity 26–7
modal realism 32 Percival, Philip, ‘On Realism about Chance’ 4,
modal status, and the Standard Conception of 74–106
metaphysical necessity 33 perdurance, and basic metaphysics 42
modality 1–4 Perry, J. 227
chance 4, 74–106 persistence, and basic metaphysics 41–2
and metaphysical necessity 1–2, 13–38 personal identity
metaphysics of 3–4 and the Bafflement Argument against
modal fictionalism 3–4, 57–73 materialism 8, 224–5, 236–8, 239–40
modal infallibilism and basic truth 40–56 bodily criterion of 8–9, 242–59
model eliminative structuralism 109 and the functionalist conception of human
beings 233–4
natural numbers, and mathematical and the ‘I’ self-concept 8, 225–7, 233
structuralism 124–5, 125–31 as indefinite 236–8
neo-Fregean solution, to the Caesar and knowledge of existence 227–30
problem 194–9 and other minds theory 234–6
266 / Index
personal identity (cont.) realism
and the solipsistic conception of the self 8, and genuine modality 41
225, 230–6 identity problem for ontological
unclarity of 8, 221–5 realism 160–3
vagueness and 8, 221–41 realism about chance 4, 74–106
physical necessity, and metaphysical analyses of chance 88–100
necessity 36–7 and backwards causation 104
physicalism, and intelligibility 55 defining 75
physics and determinism 75
and metaphysical necessity 29 and fit 94–100, 100–3
and realism about chance 79–80 frequentist analyses of 89–91
placeshifting, and enduring runabouts 44, and indeterministic statistical
45 phenomena 78–84
Platonism Lewis’s ‘best-system’ analysis of 4, 76–7, 85,
and the Caesar problem 198–9 91–4, 100–1, 104
and mathematical structuralism 117, 118, and temporally relative warrant 84–8
119–20, 141, 142 realist structuralism, identity problem
and metaphysical necessity 22 for 5–6, 146–63
pointing, and sortal concepts 7, 204, 214–15 referential realism, and mathematical
practical syllogisms, and the ‘I’ self-concept structuralism 124, 129
227 Reid, T. 222, 228
primitive resemblance theory, and the relational properties, and the governance of
Non-Standard Conception of identity 166–8
metaphysical necessity 27 relative conceivability, and metaphysical
Principal Principle, and realism about necessity 23
chance 85 Resnik, Michael
‘propensity’ theories of the ontology of and the identity of indiscernibles 139
chance 95 Mathematics as a Science of Patterns 109, 111, 114,
Putnam, H. 27 119, 121, 122
and natural numbers 125–6
qualitative similarity, and the Non-Standard restricted modality, and metaphysical
Conception of metaphysical necessity 36
necessity 26–7 Revised Bodily Criterion 246–8, 250
Quine, W.V.O. Riddoch, M.J. 209–10, 212
and mathematical structuralism 111, Rosen, Gideon
127–8, 137, 140 objection to modal fictionalism 58–63, 64
Quinean observation sentences 229 ‘The Limits of Contingency’ 1–2, 13–39,
and sortal concepts 7, 203–4, 211, 214 40, 52, 53
Quine–Kraut requirement, and mathematical Russell, B. 113, 179, 229–30
structuralism 137–8
Quinean conception, of the Leibniz semantic access agnosia 209
Principle 152, 153, 154–5 semantic indeterminacy, and the Caesar
problem 190–1
Railton, P., deductive normological semantic threshold, and the Caesar
mode 82–3 problem 187
rational agents, solipsists as 232–3 senile dementia, and bodily criterion
real modalities, and metaphysical necessity 16 245
Index / 267
set theory statistical phenomena, and realism about
and the axiom of extensionality 5, 151, 152, chance 78–84
165, 167 Strawson, P.F. 8, 216–17, 231
and the Caesar problem 200 Strong Corporealism, and bodily
and Keränen’s principle (STR) 169, 170 criterion 249–50
and mathematical structuralism 135, 138 Sturgeon, Scott
and metaphysical necessity 18, 22, 25, 38 ‘Moral Infallibilism and Basic Truth’ 2–3,
shape, and basic metaphysics 42–3 40–56
shapeshifting, and enduring runabouts 44–5 and the Non-Standard Conception of
Shapiro, Stewart metaphysical necessity 32
‘The Governance of Identity’ 6, 131, 164–74 sufficient reason principle, and the Caesar
Keränen’s reply to 5–6, 131, 146–63 problem 185
and the Leibniz Principle 148, 149, 152, supervaluationist solution, and the Caesar
153–5 problem 190–3
Philosophy of Mathematics: Structure and supervenience, and mathematical
Ontology 109, 146, 149, 155, 157 structuralism 117
and realist structuralism 157–9 Swinburne, Richard 224
‘Structure and Identity’ 4–5, 109–46, 165
and trivial accounts of identity 155–7 temporally relative warrant, and realism
trivializing objection 150–2 about chance 84–8
see also ante rem structuralism trope theory, and the Non-Standard
Shoemaker, S., account of body ownership 9, Conception of metaphysical necessity 27
250–1, 255–6 Tye, M., account of body ownership 9, 251–2,
similarity, and the Non-Standard Conception 254, 255–6
of metaphysical necessity 27
solipsistic conception of the self 8, 225, 230–3 UMC, and the Standard Conception of
and functionalism 233–4 metaphysical necessity 19–20
and other minds theory 234–6 unclarity of personal identity 221–5
sortal concepts 7, 203–18
and the binding problem 205–10 vagueness, and personal identity 8, 221–41
and the delineation thesis 7, 203–5, 209, vegetative states, and standard bodily
214, 215, 216, 217 criterion 244–5, 246, 247, 249–50, 251
justification for binding 7, 210–14 von Neumann numbers
as orienting attention 214–17 and the Caesar problem 186
sortal exclusion, and the Caesar problem 180, and mathematical structuralism 122, 124,
182–3 129, 143
spatiotemporally scattered objects, binding
of 211–12 warrant
Standard Bodily Criterion 243–5 and the ‘I’ self-concept 226
Standard Conception of metaphysical and the solipsist as rational agent 233
necessity 1, 2, 17–21, 29, 37–8 Weak Corporealism, and bodily
objections to 32–6 criterion 248–9, 250, 251
and physical necessity 37 Weber, Heinrich 110–11
Standard Materialism, and bodily Williams, Bernard
criterion 251, 252, 253, 256 on personal identity 8, 223–4
standards of identity and existence 162–3 ‘The Self and the Future’ 239
268 / Index
Williamson, T. 222 Zermelo system
Wittgenstein, L. 228–9, 230 and the Caesar problem 186
The Blue Book 232, 236 and mathematical structuralism 122
Wright, C. 198 Zermelo-Frankel set theory 150–1, 152

MacBride - Identity and Modality - Oxford University Press, USA. 0199285748

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

MacBride - Identity and Modality - Oxford University Press, USA. 0199285748

Încărcat de

Drepturi de autor:

Formate disponibile

Identity and Modality

MIND ASSOCIATION OCCASIONAL SERIES

This series consists of occasional volumes of original

CLARENDON PRESS · OXFORD

ISBN 0–19–928574–8 978–0–19–928574–7

1. The Limits of Contingency 13

2. Modal Infallibilism and Basic Truth 40

3. The Modal Fictionalist Predicament 57

4. On Realism about Chance 74

Part II: Identity and Individuation 107

5. Structure and Identity 109

6. The Identity Problem for Realist Structuralism II:

7. The Governance of Identity 164

8. The Julius Caesar Objection: More Problematic than Ever 174

9. Sortals and the Binding Problem 203

Part III: Personal Identity 219

10. Vagueness and Personal Identity 221

11. Is There a Bodily Criterion of Personal Identity? 242

This volume grew out of a conference held at the University of St Andrews

John Campbell isWillisS. andMarion Slusser Professor of Philosophy, University

The papers in this volume constellate about fundamental philosophical issues

Part II: Identity and Individuation

Part III: Personal Identity

1. What is Metaphysical Necessity?

3. A Question about the Informal Elucidation

4. The Standard Conception and the Differential Class

Existence claims elsewhere in mathematics, e.g. the existential principles

5. The Non-Standard Conception

6. The Two Conceptions and the Informal Explanation

7. Is the Non-Standard Conception Coherent?

Let God be Anselm’s God—a necessarily existing perfect spirit—and

8. Objections to the Standard Conception

φ (P) =df (φ → P).

9. Physical Necessity Reconsidered

2. Six Claims of Basic Metaphysics

And he polishes it thus:

Distance between things is intrinsic. Our third claim of Basic Metaphysics

2. Two Objections and One Beneﬁt

1 (H1) Not (Possibly P) Hyp

3. The First Development

4. The Second Development

1 (1) Not-P Hyp

‘Chance’ is a single-case, temporally relative, objective probability for which

1. Chance and Inference to the Best Explanation

1.1 Statistical Phenomena

1.2 Temporally Relative Warrant

(M) For all r, p and t: credence r in p at t is M-warranted iff the chance

2.1 Frequentist Analyses

2.2 Lewis’s ‘Best System’ Analysis of Chance

This proposal is an analysis of chance law, not an analysis of chance itself: it

Although this analysis has attracted much critical attention, most of it

2.3 The Problem of Fit

2.4 ‘Fit’ as a Primitive?

The purpose of this paper is to further articulate my preferred version

Quine’s argument here seems to be that since scientists (and mathematicians)

1. What Is (Ante Rem) Structuralism?

Kastin’s ‘Shapiro’s mathematical structuralism’ identiﬁes passages in my

My stated view at the time was that there are no cross-structural-identity

This passage comes on the heels of McGee’s use of Quine–Putnam considera-

3. Identity and Indiscernibility

I do not know what to make of this objection. In the terminology of my

Special thanks are due to Ken Manders and John McDowell.

1. Identity and Individuation

2. The Trivializing Objection