# Terence Parsolls

INDETERMINATE IDENTITY

Terence Parsons presents a lively and controversial study of philo-

sophical questions about identity. Is a person identical with that

person's body? If a ship has all its parts replaced, is the resulting

ship identical with the original ship? If the discarded parts are

reassembled, is the newly assembled ship identical with the origi-

nal ship? Because these puzzles remain unsolved, some people

believe that they are questions that have no answers, perhaps

because the questions are improperly formulated; they believe

that there is a problem with the language used to formulate them.

Parsons explores a different possibility: that such puzzles lack

answers because of the way the world is (or because of the way

the world is not); there is genuine indeterminacy of identity in the

world. He articulates such a view in detail and defends it from a

·host of criticisms that have been levelled against the very pos-

sibility of indeterminacy in identity.

Indeterlllinate

Identity

Metaphysics and Semantics

TERENCE PARSONS

CLARENDON PRESS· OXFORD

OXFORD

UNIVERSITY PRESS

Great Clarendon Street, Oxford OX2 6DP

Oxi'ord University Press is a department of the University or Oxford.

It furthers the University's objective of excellence ill research, scholarship,

and education by publishing worldwide in

Oxford New York

Athens Auckland Bangkok Bogota Buenos Aires Calcutta

Cape Town Chennai Oar es Salaam Delhi Florence Hong Kong Istanbul

Karachi Kuala Lumpur Madrid Melbourne Mexico City Mumbai

Nairobi Paris Sao Paulo Shanghai Singapore Taipei Tokyo Toronto Warsaw

and associated companies in Berlin Ibadan

Oxford is a registered trade mark of Oxford University Press

in the UK and certain other countries

Published in the United States

by Oxford University Press Inc., New York

© Terence Parsons 2000

The moral rights of the author have been asserted

Database right Oxford University Press (maker)

First published 2000

All rights reserved. No part of this publication may be reproduced.

stored in a retrieval system, or transmitted, in any fOflll or by any n1eans,

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 organizC:\tions. Enquiries concerning reproduction

outside the scope of the ahove shllUld 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 the same condition on any acquirer

British Library Cataloguing in Publication Data

Data available

Library of Congress Cataloging in Publication Data

Parsons, Terence.

Indeterminate identity: metaphysics and semantics / Terence Parsons.

p. cm.

Includes bibliographical references.

1. Identity (Philosophical concept) 1. Title.

BD236 .P36 2000 Ill'.82-dc21 00-034027

ISBN 0-19-825044-4 (alk. paper)

I 3 5 7 9 10 8 6 4 2

1ypcset by Best-set Typesetter Ltd., Hong Kong

Printed in Great Britain

on acid-free paper by

Biddies Ltd

Guildford & King's Lynn

To Peter Woodruff

PREFACE

I first addressed the topic of indeterminate identity in a short paper

in the mid-1980s, resulting in Parsons (1987). A few years later I

returned to the topic, and Peter Woodruff and I began discussing

it in detail. We believed that it is coherent to hold that identity

statements might be indeterminate, and that all of the a priori

proofs to the contrary are clearly question-begging. But there

might very well be other considerations that mitigate against this;

for example, the determinacy of identities might be presupposed

by other deeply held beliefs of ours. Besides, it is not an easy

idea to grasp. We decided that we stood in need of some overall

conception of what it would be like for some identities to be

indeterminate. These conversations led to a study of Venn-

Euler-like diagrams for "picturing" indeterminacy, diagrams that

appear in some of our joint papers and in Chapters 7-9 of this book.

These diagrams have often guided us in development of basic

positions on matters of logic and semantics. We also wanted to set

the basic theory on a sure footing, and this led in turn to an inves-

tigation of indeterminate set theory, which is summarized in

Chapter 11.

In Spring 1998 I had an opportunity to lecture on Indeterminate

Identity at the University of Salzburg. This required me to give a

systematic overview of the topic of indeterminate identity, starting

from scratch for people not already immersed in the literature. In

doing this, I needed to indicate how my/our views are to apply to

the identity puzzles that figure so prominently in the literature. I

also needed to work out how the various thoughts that we have

had on the topic mesh with one another, and with the ongoing lit-

erature on the subject. Those lectures led to this book. Although

the book is singly authored, much of its theoretical content was

originally developed in cooperation with Peter Woodruff. I am

solely responsible for matters of exposition, and for various

developments of the theory. Some of my own contributions (such

as the notion of super-resolutions) can be identified, but in many

cases I am not myself able to say whether I have new exposition of

viii Preface

old ideas, or new ideas. I also want to emphasize that many of the

general arguments given here in defence of the coherence of

indeterminate identity are already present in some form in the

literature. When I take a point to be common knowledge, I have

not tried to trace its exact origins.

I wish to express my thanks to the Salzburg Philosophy depart-

ment for its invitation to lecture there. Thanks are also due to those

Fall 1998 participants in the UCI Logic Workshop who monitored

and critiqued a presentation of a draft of this book, particularly:

Jason Alexander, Jeff Barrett, Gary Bell, Penelope Maddy, Patricia

Marino, and Kyle Stanford. My greatest debt, of course, is to Peter

Woodruff, who is almost indeterminately a co-author.

u.c. Irvine

1999

Terence Parsons

CONTENTS

/Inalytical Table of Contents

1. Introduction

2. Indeterminacy

3. Identity

4. The Evans Argument, Properties, and DDiff

5. Non-Conditional Disputations

6. Conditional Disputations

7. Understanding Indeterminacy

8. Counting Objects

9. Denoting Objects

10. Alternatives to Indeterminate Identity

11. Sets and Properties with Indeterminate Identity

12. Higher-Order Indeterminacy

Appendix: Evans on Indeterminacy

Reference

Index

x

1

11

31

45

56

87

107

134

150

160

181

195

204

215

219

ANALYTICAL TABLE

OF CONTENTS

1. INTRODUCTION

Identity puzzles persist without consensus on any solution. This

book explores the idea that they have no solutions because the

identities are indeterminate-there is no fact of the matter about

whether or not they hold. Four puzzles without determinate

answers are discussed throughout the book: Ci) is a person identi-

cal with that person's body? (ii) if a ship has its parts replaced and

the old parts are assembled into a ship, which of the two resulting

ships is identical to the original? (iii) if a person undergoes a crucial

change, is the person after the change identical with the person

before the change? (iv) given a cat with imprecise boundaries,

which cat-like thing with precise boundaries is identical to the

actual cat? The purpose of the book is to explore the view that

there are no answers to these questions because of indeterminacy

in the world (as opposed, for example, to imperfections in our

language).

2. INDETERMINACY

Indeterminacy in the world is a genuine option within idealism,

realism, or pragmatism. Indeterminacy pertains to states of affairs,

such as whether a certain object has or lacks a certain property. If

neither of these is the case, then the state of affairs is indeterminate,

and a sentence reporting it lacks truth-value. A semantics is

described for a logical notation in which sentences may lack truth-

value. It contains a connective'!' for determinate truth: the sentence

'!S' is true if'S' is true, and is false if'S' is either false or lacking in

truth-value. The sign 'v' represents indeterminacy: '''IS' is true if'S'

lacks truth-value and is otherwise false. The logic of such a language

Analytical Table of Contents XI

is described, in which validity means guaranteed preservation of

truth. Classical tautologies such as '5 V -S are not generally valid.

since they lack truth-value if '5' does, but '5 V -,!S' is always valid.

Classical indirect proof does not hold, since if'S' entails a contradic-

tion this proves only that '5' is not true, not that its negation '-,5' is

[rue. So you cannot infer '-,S' by indirect proof, but you can infer

'-,)5'. Likewise, contrapositive reasoning is not generally valid: if

you can infer' B' from 'A' you may not be able to infer '--,A' from

'-,B', though you will be able to infer '-,!A' from '-,B'.

3. IDENTITY

Real identity in the world is defined in terms of coincidence of

properties: objects are identical if they both determinately have and

determinately lack the same properties, and they are distinct if one

determinately has a property that the other determinalely lacks. If

neither of these holds, then it is indeterminate whether the objects

are identical. It is an empirical matter whether there is any indeter-

minacy at all, and an empirical matter whether such indeterminacy,

if it exists, extends to identity. This explanation of identity validates

Leibniz's Law: From 'a = b' and '$a' one can infer 'Q>b'. But it does

not validate the contrapositive form of this law: one might have 'Q>a'

and '-.,Q>b' both true without '-.,a = b' being true; this can happen if

'Q>x' does not express a property. The identity discussed here is not

"relative identity" because relative identity, unlike genuine identity,

does not validate Leibniz's Law. The four identity puzzles are

reviewed; in each case in which the identity is indeterminate, there

is a property determinately possessed by one object that is not

determinately possessed by the other, but there is no property for

which they determinately disagree.

4. THE EVANS ARGUMENT, PROPERTIES,

AND DDIFF

Gareth Evans's 1978 argument disproving the existence of inde-

terminate identity is discussed. On a simple analysis it appeals to

Xll Analytical Table of Contents

the (invalid) contrapositive version of Leibniz's Law; on a deeper

analysis it presumes incorrectly that the formula 'v(x = a)'

expresses a property (the property of being indeterminately identi-

cal with a). The proof is an RAA of the hypothesis that that formula

expresses a property. This fact is not due to non-extensionality

or to anything like the semantic paradoxes; instead it is akin to

the "paradoxes" of na"ive set theory, due to the fact that identity

is defined in terms of global quantification over all (worldly) pro-

perties. A test for whether a formula '<jlx' expresses a worldly

property is whether it satisfies the principle that joint (determinate)

satisfaction and (determinate) dissatisfaction of the formula

makes a "Definite Difference" in the identity of the objects in

question; that is, that there are no objects x and y such that '<jlx' &

'!.....,<jli are both true, along with 'n = y'. If a formula satisfies this

principle, then an abstract, 'Ax<jlx', constructed from it stands for a

worldly property; otherwise we can either say that it does not stand

for a property, or that it stands for a non-worldly "conceptual"

property.

5. NON-CONDITIONAL DISPUTATIONS

Various objections to indeterminate identity are not conclusive.

Quine's doctrine "No entity without identity" rules out only enti-

ties for which identity does not make sense; it is not aimed at inde-

terminacy. Holding on to bivalence for methodological reasons is

not compelling, since little of methodological value is lost in allow-

ing that claims may lack truth-value, and many physicists have

(rightly or wrongly) adopted this stance towards some phenomena

in quantum mechanics without hampering their methodology. Even

if bivalence is maintained within each science, identity puzzles may

still arise where sciences overlap. An argument of Salmon's is

reviewed; it is either a version of Evans's argument, already

discussed, or a challenge to indeterminate set theory, discussed in

Chapter 1l.

Sometimes puzzles arise from reading sentences non-literally. A

sentence may be read supervaluationally: it is treated as true (false)

if all ways of making its vocabulary determinate yield readings

Analytical Table of Contents xii

which are true (false). It may also be read super-resolutionally: ii

is treated as true (false) if all worlds that result from some way 01

making our own world completely determinate would make thE

sentence true (false). These options account for data used by som(

writers in an attempt to refute indeterminate identity. A purportec

refutation of non-bivalence by Williamson is discussed; it is arguec

I hat his claims that all contradictory sentences must be falSE

are plausible only if they are read in some special way, e.g

supervaluationally.

Several authors point out that if an object has indeterminatE

boundaries, this does not logically entail that it is

identical to something. Some mereological principles are giver

that would fill in this logical gap. Objects mayor may not bE

subject to these principles. An example due to Cook is aIlegec

to be hypothetically of this kind (given his assumptions);

disproof of indeterminate identity is examined and found to b,

inconclusive.

An argument by Noonan against an example (due to Broome

of indeterminate identity of clubs is shown to be implausible if thE

claims in it are read literally; those claims are plausible if reac

super-resolution ally, but then they do not conflict with indeter

minate identity.

6. CONDITIONAL DISPUTATIONS

Truth-conditions for conditionals are discussed, and thE

Lukasiewicz conditional is adopted; such a conditional i:

false when the antecedent is true and the consequent false, anc

true if the truth-value status of the consequent is at least as higl

as that of the antecedent (counting lack of truth-value as a statu:

intermediate between truth and falsity). A conditional 'If A ther

B' is to be read either with this conditional CA B') or giver

the "if-true" reading ('lA B'). The Lukasiewicz conditiona

satisfies modus ponens, modus tol/ens, hypothetical syllogism, anc

contraposition, but only a restricted form of conditional proof

taking 'A' as a premiss and deriving 'B' allows you to condition

alize and infer '!A B', but not 'A B'. (No non-bivalen

truth-status-functional conditional satisfies all of modus pOllens

xiv Analytical Table of Contents

modus tollens, and conditional proof, so one can do no better.) If

one wishes to state Leibniz's Law as a conditional, the bare

Lukasiewicz conditional form is incorrect, but the "if-true" version

is correct: 'la = b ~ (<j>a ~ <j>b)'. Broome gives a different version,

which is logically equivalent to this one. 10hnson criticizes these

versions, but gives a conditionalized-conditional version that is also

equivalent.

A second argument of Williamson is considered which defends

bivalent versions of the Tarski biconditionals; it is held that his

rationales for the biconditionals Clre subject to interpretCltion, Clnd

that interpretations that do not make them beg the question

rationalize only non-bivalent versions.

7. UNDERSTANDING INDETERMINACY

The terminology used in explaining indeterminate identity is meant

to be ordinary; 'identical' is not intended in a special sense. 'Inde-

terminate' may be definable within certain world views (such as

idealism), but is otherwise taken as Cl primitive. The biggest impedi-

ment to understanding views that invoke indeterminate identity is

our tendency to reason bivalently; this has nothing special to do

with identity.

One can picture situations involving indeterminacy using Venn-

like diagrams in which objects are represented by images with finite

size. An ob.iect is pictured as being indeterminately P if its image

lies partly inside of and partly outside of the region representing

property P; objects are pictured as being indeterminately identical

if their images properly overlap. Simple constraints on the pictur-

ing conventions entail that Leibniz's definition of identity in terms

of coincidence of properties is built into the picturing. Picturings of

the paradigm identity puzzles are given.

A more general notion of picturing can be defined that is not

necessarily two-dimensional; a simple condition yields a kind of

principle of plenitude for properties and the Leibnizian account of

identity. A process is given to refine pictures into more determinate

ones; under certain general conditions these refinements picture

the resolutions discussed earlier in giving super-resolutional read-

ings of sentences.

Analytical Table of Contents xv

8. COUNTING OBJECTS

I [ we try to count objects in the face of indeterminacy, we some-

times get no determinate answer; this is due to indeterminacy of

predication (producing indeterminacy regarding which objects are

supposed to be counted), or indeterminacy of identity (producing

indeterminacy regarding whether an object has already been

counted), or both. Familiar formulas are given for making cardi-

nality claims; e.g. "there are at least two <\>'s" is written as ':3x:3y

(--..x = y & <\>x & <\>y)'. It is shown how to get the "right" answers; e.g.

that in the ship case it is true that there are at least two ships, false

that there are more than three, and indeterminate whether there

are exactly two (or exactly three). Sometimes a question can be for-

mulated in two ways: either austerely, or with a determinacy con-

nective ('1') added; these formulations correspond to two natural

"right" answers. Super-resolutional readings also explain certain of

our intuitions.

9. DENOTING OBJECTS

Some authors have suggested that if it is indeterminate whether

a = b, then it is not possible to determinately denote a; as a conse-

quence, all identity sentences that lack truth-value suffer from

some semantic defect. When it is indeterminate whether a = b, it is

shown how to have singular terms that determinately denote a

without denoting b at all, and terms that determinately denote a

while also indeterminately denoting b, and terms that indetermi-

nately denote each. So all options are possible.

10. ALTERNATIVES TO INDETERMINATE

IDENTITY

Many writers believe that it is both possible and desirable to

account for the lack of truth-value of identity sentences by appeal

to indeterminacy in the semantics of our language or of the COIl-

xvi Analytical Table of Contents

cepts employed by our language. It is shown here that several

natural accounts of this sort do not yield a lack of truth-value in

the puzzle cases without also wrongly yielding lack of truth-value

in other cases as well. For example, in the case of a personal

disruption, it ought to be true that exactly one person entered the

room, and true that exactly one left, without its being either true

or false that the person who entered is the person who left.

An account is examined that employs supervaluations to analyse

indeterminacy in singular terms; as formulated, it does not seem to

give the right pattern of truth-values in all cases. A second account

is based loosely on a discussion by Lewis of how many cats exist in

a given region; the account (not endorsed by Lewis) is objection-

able; more importantly, it does not apply to the puzzle cases in

general. Last is an account due to Stalnaker according to which we

refine our concepts upon demand when faced with puzzle cases.

This is the most promising approach; versions of it are refuted, and

we are left uncertain whether it can be developed into a viable

al tema ti ve.

11. SETS AND PROPERTIES WITH

INDETERMINATE IDENTITY

Sets are things whose identities are defined in terms of their

members: they are identical if they both determinately have and

determinately lack the same members, they are distinct if one of

them determinately has a member that the other determinately

lacks, and otherwise it is indeterminate whether they are identical.

A distinction is made between worldly sets and conceptual sets; the

former but not the latter obey the DDiff principle for set mem-

bership: that if x is determinately a member of Sand y is determi-

nately not a member of S, then x is determinately distinct from y.

A theory of objects together with worldly sets of those objects is

formulated, and it is shown that there is indeterminacy of identity

between such sets if there is any worldly indeterminacy at all, even

if there is no indeterminacy of identity between objects. Salmon's

argument against indeterminacy is shown to fail within this theory.

If we construe worldly properties as extensional, then they can be

Analytical Table of Contents xvii

identified with the sets under discussion, and the resulting theory

is a formalization of the framework of objects and properties and

identity sketched in Chapters 2 and 3. The theory can be extended

to a transfinite hierarchy of sets, yielding an indeterminate version

of Zermelo-Fraenkel set theory; the classical ZF theory is satisfied

in a definable subdomain of this theory. Relations are definable as

sets of ordered pairs, using a refinement of the usual definition of

ordered pairs; this shows the consistency of an assumption made in

Chapter 3 that if R is a relation between individuals, then bearing

R to object 0 is a property.

A theory of conceptual sets results from removing the DDiff

condition placed on worldly sets.

12. HIGHER-ORDER INDETERMINACY

We have indeterminacy if there is a question that has no answer;

we have higher-order indeterminacy when it is indeterminate

whether a certain question has an answer. There might not be any

higher-order indeterminacy, but in case there is, two accounts of it

are offered. The theory from earlier chapters can be accommodated

within either account.

Appendix: Evans on indeterminacy

Some of Evans's remarks about modalitics near the end of his 1978

paper have puzzled many commentators. Following a suggestion of

Wiggins, it is shown that they make good sense if we assume that

Evans saw indeterminacy as a kind, or mode, of truth, so that 'vS'

may be true when'S' is true. On this reading, his argument is

not relevant to the thesis of indeterminate identity discussed in the

subsequent literature.

1

Introduction

1.1 IDENTITY PUZZLES

Throughout history philosophers have puzzled over questions of

identity:

Is a person identical with that person's body? Of course, if the

person lives on after the body no longer exists, they are dis-

tinct, and if the person ceases to exist at death but the body

continues in existence, they are distinct. But what if they

always coexist?

Suppose a ship sets sail, and while at sea it is completely

rebuilt, plank by plank; is the resulting ship with new parts the

ship that originally set sail? What if the discarded pieces of the

original ship are assembled into a ship; is that the ship that

originally set sail?

If a person has a brain transplant, or a memory transplant, or

... is the resulting person the same person who antedated the

operation, or has the old person ceased to exist, to be replaced

by another?

These, and a host of puzzles like them, persist without adequate

solutions. Each puzzle is actually a vague description of a spectrum

of cases, some of which give rise to one answer, and some of which

give rise to the opposite answer, with sufficient cases in between to

bewilder just about anyone, regardless of their instincts; thus the

puzzles persist.)

I This way of construing the puzzles carries out a theme of van Inwagen (1988).

which describes a cabinet in which a person undergoes disruptive changes, so that

you are uncertain whether the person who emerges is the person who entered. Van

Inwagen suggests that anyone can fine-tune what happens inside the cabinet so as

to yield a case about which the fine-tuner is uncertain.

2 Introduction

Consider the ship case. Sometimes we can disassemble an object

into parts and then reassemble those parts into the original object.

I may have to do this in order to get my favourite desk into a new

room; this is a way to move my desk into the new room, not a way

of destroying it and replacing it with a new desk. (If you are not

convinced by this, think of taking apart a blender to wash it and

then putting it back together again.)2 So sometimes reassembling

parts yields the original object. But sometimes I repair an object

by replacing a part; if I replace the radiator in my car with a new

one I think r still own a car that I bought several years ago, not one

that just sprang into existence. So sometimes replacing a part yields

the original object. The "ship of Theseus,,3 example above illus-

trates a case in which our judgements about identity are supposed

to be in clear conflict: the original ship has apparently been disas-

sembled and reassembled, and the original ship has apparently

been repaired by having new parts installed. But these cannot both

be true, because two non-identical ships have resulted.

Perhaps you see one of the options as being clearly superior to

the other? Then a minor adjustment in the case will bring them into

conflict:

If you are sure the ship with new parts is the original ship,

suppose that the new parts of the repaired ship are quite

unlike the parts being replaced, making the ship with new

parts unlike the original ship-as unlike as you need to

weaken your judgement of identity. (You may also wish to

suppose that the activity is part of a contest to see how crea-

tively the contestants can disassemble a ship and reassemble

it, thus favouring the reassembled ship as being the original

ship.)

If you are sure the "reassembled" ship is the original ship, just

reassemble most but not all of the discarded parts, mixing in a

2 If you are inclined to be sceptical about even such a case as this, consider what

you would say if asked where you purchased your blender. You would never say "I

didn't purchase it, I created it after the parts were last washed." So you must admit

that you talk as if an object can be reassembled. And you must admit that this is

serious talk, worth explaining, so the idea that what you say is plausible because it

is literally true is an option well worth exploring. That is enough to open the door

to theorizing on the identity question.

3 A ship example originates in Plutarch's Life of Theseus. This was updated by

adding a second ship by Hobbes in De Corpere, 2, It. [ have added the twist that

the repair/replacement takes place at sea; this is not essential to the example.

Introduction 3

few new ones. Adjust the number of parts upward or down-

ward until you are no longer sure about what to say. If neces-

sary, leave some of the original parts in the ship with (mostly)

new parts.

Some philosophers have been driven by examples of this sort to

provide a method for answering any such question. The most

popular methods are the simplest. For example, some propose that

having the same parts is necessary for identity, and any change of

parts whatever leads to non-identity; there is never any such thing

as repair by replacement of a part. Others propose that continuity

of size, shape, and function are required to maintain identity, so that

there is never any such thing as reassembly. These are extreme posi-

tions; they are often admired for the ingenuity that goes into

defending them, but they have few real adherents. Instead, most

philosophers try to devise a subtler criterion for identity preserva-

tion that avoids such extreme judgements, a criterion that will allow

us to say "Aha, that's it", whenever the criterion is applied. Such

methods (when not overly vague) give natural answers in the prob-

lematic cases, but other clever philosophers inevitably devise new

cases in which the "subtle" methods ride roughshod over our surest

judgements. Proponents of such views then must either refine them,

or explain why normal intelligent people are wrong to reject them.

A great deal of ingenuity has gone into the defence of and attack

on such views; none of that discussion will be reprised here. It is

clear that none of the methods has won popular acceptance, and

this motivates us to look elsewhere.

Because so many identity puzzles have remained unsolved for

centuries, some observers have been led to speculate that these are

questions that have no answers. But that raises other questions.

Why don't they have answers? What would it take for them to have

answers?

The most popular option in this century is that the questions

have no answers because they are improperly formulated, typically

because they incorporate a definite description that does not

uniquely denote. Sometimes this is plausible. It is plausible in the

"building" case. Suppose that Old Ivy Hall has an addition built

onto it, tripling it in size without altering the structure of the origi-

nal in any significant way, and the new large building is named Post-

modern Hall. My office is in the old part and yours is in the new

4 Introduction

part. Someone asks "Is the building in which Parsons has his office

the building in which you have your office?" It seems clear that this

has no unique correct answer, and that this is because the definite

description 'the building in which Parsons has his office' does not

uniquely denote. Buildings can be parts of other buildings, and Old

Ivy Hall is now a part of Postmodern Hall. My office is in Old Ivy

Hall, and also in Postmodern Hall, and these are not the same

building; one is part of the other. So there is no such thing as the

building in which I have my office. And so the identity question is

ill-formed.

This solution is plausible because buildings are parts of other

buildings. It is less plausible to think that ships are parts of

other ships (at least for normally designed ships), and quite

implausible that persons are parts of other persons. Yet this is

what it would take to solve the harder cases in parallel fashion.

For example, you could say that the definite description 'the

original ship' fails to uniquely denote, because there were actually

two original ships: one that was later repaired with new parts, and

one that was later reassembled from the original parts. This is not

plausible.

4

I do not believe that a systematic diagnosis of identity puzzles

in terms of imperfections in our language, or in the concepts

embodied in our language, will be satisfactory. This is certainly

a natural option, and one that I need to discuss (Chapter 10).

However, my main task is to explore an alternative that I find more

plausible.

1.2 WORLDLY INDETERMINACY

OF IDENTITY

I am inclined to think that an identity question can be completely

coherent and well formed and yet lack an answer because of the

way the world is (or because of the way the world is not). Not that

there is an unknown answer, but rather that there is no answer

at all. In the ship case above, the facts are these: there is a unique

original ship, there is a unique ship with new parts, and there is a

unique newly assembled ship. The ship with new parts is distinct

, 1 examine some closely related but more plausible views in Ch. 10.

Introduction 5

from (non-identical with) the newly assembled ship. But there is

no fact of the matter regarding whether the newly assembled

ship is the original ship, or whether the ship with new parts is the

original ship.

When there is no fact of the matter, a sentence reporting the

purported fact lacks truth-value. So the following sentences have

the indicated statuses:

the newly assembled ship = the ship with

new parts False

the newly assembled ship = the original ship No truth-value

the ship with new parts = the original ship No truth-value

These indeterminacies must cohere with certain others as well. It

is indeterminate whether the newly assembled ship formerly left

port when the original ship did, and likewise for the ship with new

parts. It is indeterminate whether the original ship now has new

parts. And so on.

Each of the sentences listed above non-defectively reports a state

of affairs, a state of affairs of identity. The first state of affairs is

made not to hold by the way the world is, but the second and third

are not either made to hold or made not to hold by the way the

world is. Thus there is genuine indeterminacy of identity in the

world. This is not an illusion generated by indeterminacy as to how

our language fits with the world; the indeterminacy is real.

There is now a growing literature on the question of worldly

indeterminacy of identity, stimulated almost wholly by a one-page

article by Gareth Evans (1978), giving a proof that there cannot be

genuine indeterminacy of identity in the world. This proof has been

attacked, defended, revised, expanded, and so on. The literature

on this subject has now matured to the point where an extended

look at the issues seems both timely and feasible. Discussion of

the issues so far has proceeded on a number of fronts, and on a

piecemeal basis. What does not yet exist is a single coherent

presentation of a position that articulates and defends worldly

indeterminacy of identity. That is the task of this book. There is little

contained here that has not already been argued in the literature

in one form or another. The contribution of this book is to choose

from among the positive views a subset that can be maintained

together, to express these in a common vocabulary, to assemble

6 Introduction

them in a presentation that can be mastered by someone new to

the topic, and to add a few details that may advance the issues.

In presenting the thesis of worldly indeterminacy of identity I

speak as an advocate on its behalf. This is not because I am con-

vinced it is true; indeed, if I am right, that is a contingent matter on

which nobody can be certain. But it makes for a more coherent

exposition if I take a definite stance on the issue. I am convinced

that the existence of worldly indeterminacy of identity is both

coherent and possible.

My goal is primarily to articulate the view in detail and in gen-

erality. The articulation of the view I take to be a matter of hypoth-

esis formulation, and this is immune from certain kinds of criticism.

For example, I cannot be said to beg the question merely by

formulating what one answer to the question is. The main task

of the book is just that: to formulate a position. I will also argue

that the position is immune to certain kinds of attacks that have

been levelled at it in the literature. Here I can be said to beg the

question-But only if I presume a point at issue. The bulk of this

book will be occupied with these two tasks: formulation of the

view and defences of its coherence and prima-facie plausibility in

the face of purported refutations. Neither of these tasks, even if

successful, will show the view to be true. For that, what is required

is to see whether it explains better than competing views do

why certain identity questions seem to be coherent and yet lack

answers. Some competing views will be discussed in Chapter 10, but

many will not. And so the ultimate fate of the view will be left

uncertain.

Methodologically, I take a Peircean perspective. 1 begin with

ordinary beliefs, which 1 will reject only if some reason is found to

challenge them. These are my tentative data: ordinary beliefs-such

as the belief that I have exactly one wife, that there is exactly one

dog in my back yard, and that exactly one ship set sail before

the problematic replacementlrepairlreassembly process. I reject

philosophical analyses that contradict these judgements, telling

me, for example, that I actually have several dogs, or that there is

not really any such thing as a dog-there are only basic particles

that swarm into dog-like shapes. It is also part of my methodology

that I do not take as data highly theoretical philosophical gen-

eralizations, such as "nothing is indeterminate", or "no two things

can be in the same place at once", or the opposites of such views.

Introduction 7

I see these as theoretical observations which are to be validated by

how well they conform to the data, as opposed to the other way

around. I may occasionally rely on some theoretical views myself,

hut only by accident, or when I see no other way to resolve an

issue.

I do not necessarily expect the reader to share this methodology,

Of to apply it exactly as I do; I articulate it here to clarify what I

will and will not be doing. Many proposed solutions to the puzzles

involve taking a stand that requires us to reject some of the appar-

ent data, for example, rejecting the view that exactly one person

entered the room in which the disruption took place. Let me call

any such position a "traditional solution". A traditional solution

explains how and why we should change our beliefs, and shows how

t he puzzles are resolved if we do so. I will not discuss such solu-

tions at all. There is an enormous literature devoted to this already,

and there is no need for me to duplicate it. Instead I focus on

proposed solutions that "preserve the data", solutions that explain

how it is that

The ordinary beliefs that we have about the identity puzzle

cases are literally true.

I also limit myself by the working assumption that

It is literally true that there is no answer to the identity

questions in the puzzle cases.

I will develop one such explanation, and I will discuss others.

1.4 A STOCK OF PUZZLES

I I will help to have at hand a small stock of cases to discuss, and an

indication of what I take the data to be in each.

The person/body: Assuming that a person exists when and only

when their body exists,S is each person identical with his/her

body?

; This assumption is crucial. For example, Stalnaker (1988: 354) says "we can

distinguish intimately related things such as an artifact and what it is made of, a

I'l'rson and his or her body, by distinguishing their temporal properties." The

I'L'rson/body puzzle discussed here is a hypothetical one: what if there were no

8 Introduction

Data: If I am alone in a room, then there is exactly one person

in the room, and exactly one human body in the room.

Working assumption: There is no answer to the question

whether the person in the room is the body in the room.

The ship: An assemblylrepair process takes place as described

above. Is the original ship identical with the ship with new

parts, or with the newly assembled ship (or neither)?

Data: Exactly one ship left port, and exactly two ships docked.

Working assumption: There is no answer to the question

whether the original ship is the ship with new parts, or whether

the original ship is the newly assembled ship.

The personal disruption: A person enters a room where

something disruptive happens to them that challenges our

judgements about personal identity. Is the person who entered

the room identical with the person who later leaves the room?

Data: Exactly one person entered the room, and exactly one

person left the room.

Working assumption: There is no answer to the question

whether the person who entered the room is the person who

left the room.

The cal:

6

It is unclear exactly what the parts of a cat are. For

example, it is unclear whether a molecule loosely attached to

the end of a hair that is engaged in falling out is part of the

cat. Consider all ways of answering such questions precisely.

Call the object (if any) that answers to any such precise

description of its parts a "p-cat". There are many p-cats, and

they are distinct from one another, because they all have

different parts. How is the cat related to the p-cats? Is it

identical to any of them? To none of them? (Are the p-cats

cats? If so, how many cats are on the table when we naIvely

think there is only one?)

Data: There is exactly one cat, and there are many p-cats.

Working assumption: For any given p-cat, there is no answer

to the question whether it is the cat.

clear examples of temporal properties in which they differ? Of course, if they do

not temporaJly coincide, then presumably one will be a person·at-t and the other

not a person-at-t, so they will be definitely distinct. The puzzle is: what if they always

coexist?

6 Patterned after an example in Lewis (1993).

·Introductioll 9

You are asked in each case to adjust the details of the example so as

la make the answer to the identity question most uncertain, based

on your convictions.l will assume that I am addressing a reader who

makes such adjustments to the cases under consideration.

For convenience later, 1 will refer to the above data together with

the working assumptions as "extended data". This is not to insist

that they are correct, but rather to emphasize that I will be trying

to account for them, not to argue for them.

The first and last of these questions involve cases of identity-at-

a-time, and the middle two are cases of identity-a cross-time. People

sometimes distinguish these, calling the first "coincidence" and

the second "persistence". I ignore the distinction because it is not

relevant to any of the issues that I address.

1.5 PLAN OF THE BOOK

First T will address what is involved in there being indeterminacy

in the world. Next, I will apply this to indeterminacy of identity. In

both of these enterprises we need to clarify not only what a par-

tially indeterminate world is like, but also how a language works

that correctly describes such a world. Then I look at a number of

arguments in the literature that attempt to show that positing inde-

terminacy of identity leads quickly to inconsistency. This is followed

by a discussion of some variations of the inconsistency argument

based on the logic and semantics of conditionals. Following this I

discuss the complaint that we cannot conceive of identity's being

indeterminate, and 1 give a useful "classical picturing" of situations

involving indeterminacy, including indeterminacy of identity. This

is followed by a discussion of how we can count objects whose iden-

tities are partially indeterminate, and of how we can uniquely refer

to such objects. Then certain alternative views are treated, views

that locate indeterminacy wholly in language or in our concepts;

the question is whether such a view can provide a better account

of the data than the view that posits real indeterminacy. After this

there is a discussion of how it is possible to develop a coherent

theory of sets that allows sets to have indeterminate members;

properties with indeterminate identity are also discussed. Finally I

10 Introduction

briefly discuss what impact higher-order indeterminacy might have

on this enterprise.

If I am successful, the reader will find a coherent statement of

what it would be like if there were indeterminacy in the world,

extending to indeterminacy of identity. (It would be what the world

would be like if the world were exactly as we naIvely think it is.)

2

Indeterminacy

2.1 WHY TAKE INDETERMINACY

SERIOUSLY?

The view under consideration sees the world as consisting of states

of affairs. Some of these definitely hold, some definitely do not

hold, and others are simply undetermined. Why should one think

that the world might be like this? Because it is a possibility. For all

we know, the world is like this.

Suppose that some form of idealism is correct, so that the world

is a thing created by our minds. Then since our minds are finite,

they are not likely to get around to finishing the job. So some

aspects of the world will be left uncreated, or undetermined, and

there will be indeterminacy. This option is sometimes foreclosed by

supposing that there is an absolute mind that fills in for us what-

ever we ourselves leave undetermined. But why suppose that the

absolute mind determines everything? If the absolute mind is

anything like the traditional Judeo-Christian-Islamic deity, then

this deity freely chooses both whether and what to create, so why

assume that (s)he determines everything? It takes a rather special

sort of deity to guarantee complete determinacy.

Ignore idealism. Michael Dummettl has made popular the idea

t hat a test for a realist view is whether that view embodies "biva-

lence", the condition that every claim about the world is true or

false. Suppose that a sentence expressing an undetermined state of

affairs lacks truth-value. (This is an assumption that I will indeed

make.) Then it appears that realism is incompatible with indeter-

minacy by definition! Perhaps this is so for Dummett's brand of

realism, but not for mine. I assume that realism includes any view

that how the world is is independent of my mind or how I conceive

I e.g., in Dummetl (1978: 145-65).

12 Indeterminacy

of it. If this is so, there is no a priori reason to assume bivalence,

and no reason to suppose that the world is completely determinate.

Why should it be?

Another outlook is pragmatism, the idea that "the world" is

whatever is revealed by scientific investigation and theorizing.

Quine (along with other pragmatists) holds a view something like

this. Quine sees the question of bivalence as a pragmatic method-

ological choice we can make. A commitment to bivalence in our

theorizing leads to methodological elegance and simplicity, while

making it difficult to reconcile recalcitrant data; an abandonment

of bivalence permits data to mesh more neatly with theory, while

complicating the theory. (1 am stating Quine's view here, not

endorsing it.) Thus bivalence is a choice. Quine himself chooses

bivalence over its "fuzzy and plurivalent alternatives" (Quine 1981:

94-5) but others may choose differently. Again, indeterminacy is an

option.

Other views could be surveyed,2 but the point is clear enough:

there are plenty of reasonable conceptions of what the world is

like that do not preclude at least some indeterminacy. And so we

should investigate what the consequences would be if there were

indeterminacy.

2.2 INDETERMINACY IN THE WORLD

I suppose that the world contains objects, properties of objects, and

relations among objects. For any given object 0 and property p

there are three options: either 0 definitely has p, or 0 definitely

lacks p, or it is indeterminate whether 0 has or lacks p. For any

given objects 01 and 02 and relation r there are also three options:

the object o[ definitely bears the relation r to 02, or 01 definitely

does not bear the relation r to 02, or it is indeterminate whether 01

bears r to 02'

For the moment

3

suppose we speak of 0 's having or lacking p as

I ef. Burgess (1990), especially §§JO-lJ, for a critique of certain contemporary

views that reject all indeterminacy for reasons that he finds less than compelling.

J I say "for the moment" because in the detailed formulation of the theory I

will not need to talk of states of affairs at all; I will only need to talk about objects,

properties, and relations.

Indeterminacy 13

a state of affairs. Then these states of affairs either determinately

hold (if 0 definitely has p), or determinately do not hold (if 0 def-

initely does not have p), or they are not determined either way. In

the last case the state of affairs is indeterminate. I emphasize that

it is the state of affairs that is indeterminate, not the objects or

properties or relations composing it. Indeterminacy emerges at the

level of states of affairs, and if we wish to call objects or properties

or relations themselves indeterminate, we will mean by that merely

that they enter into indeterminate states of affairs. There is nothing

in the view discussed here which lays the blame for indeterminacy

of states of affairs on either the objects or on the properties or

relations. I thus set aside certain discussions in the literature

addressing indeterminacy of identity which try to distinguish

whether it is objects that are indeterminate, or the relation of

identity itself.4 I do not know how to apply the notion of inde-

terminacy below the level at which we make whole judgements,

except derivatively; e.g. if one defines an indeterminate object as

one for which there is some object with which it is indeterminately

identical:

x is an indeterminate object =df 3y[it is indeterminate

whether x = y].

I am also neutral about whether or not properties and relations are

"intensional" entities, that is, about whether there can be two dis-

finct properties p and q that determinately hold of exactly the same

objects and determinately fail to hold of exactly the same objects.

It just happens that this issue does not bear on anything I have to

say. So I will not discuss it, except briefly in Chapter 11. However,

I am not neutral about realism for properties. The properties under

discussion here must be genuine constituents of the world, not

concepts in the mind or mere words. It is real properties in the

4 Sainsbury (1989) defines a "vague property" as one that neither definitely

applies nor definitely fails to apply to a sharp ohject, where a "sharp" object is one

that does not have two precisific<ltions such that some property definitely applies to

one but definitely does not belong to the other. He does not define 'precisification'.

Sainsbury (1994) argues more sceptically that there is no substantive thesis about

whether there are vague properties. He suggests that one could define a vague object

as one such that it is indeterminate whether some sharp property applies to it. but

he doubts that one can clarify what a "sharp" property is. 1 am inclined to agree.

Many writers suggest that there can be objects that possess genuinely vague bound-

aries, but the connection between this fact and indeterminate identity is debatable;

see Ch. 5.

14 Indeterminacy

world that determine whether objects (real objects in the world)

are or are not identical (really identical). So my theorizing is

avowedly realistic. I have in mind no particular theory of proper-

ties, except that they must be genuine universals-they must be the

sort of thing that are capable of being possessed by many objects.

My discussion of identity is neutral among many different versions

of such a view.

5

Nominalists and conceptualists often believe that they can

account for any phenomenon that realists can account for, by

showing how to take talk about real things in the world and

explain it in nominalistically or conceptualistically acceptable

ways. I have no objection to such projects, but they need to be

distinguished from what J am doing here. J am presenting a

realist theory that might provide the input to nominalist or con-

ceptualist reductions; I am not myself presenting a nominalist

or conceptualist theory. (1 do not know how to formulate the

theory 1 discuss here in terms acceptable to nominalism or

conceptualism. )

2.3 TRUTH-MAKING

States of affairs can make sentences true in different ways. In the

most dircct way, sentences may express states of affairs, e.g. the sen-

tence may consist of an atomic predicate that stands for a property

and a name that stands for an object. Such a sentence is true if the

object named determinately possesses the property, false if the

object determinately lacks the property, and it is otherwise lacking

in truth-value. The sentence is made true, or false, or neither,

directly by "the facts", i.e. by the status of how the object and prop-

erty are related. Other sentences are made true, false, or neither,

indirectly. For example, an existential "Something is F" whose

5 For a survey of various theories of properties see MeUor and Oliver (1997). I

take for granted that properties are universaJs (typically shared by many objects)

in various places in this text. A referee for OUP suggests that the indeterminate

identity view may be developed without this assumption. This may be true, but J

have not explored that option.

Indeterminacy 15

predicate 'F' stands for a property will be made true by a state of

affairs that determinately holds; it will be made false by the deter-

minate non-holding of all states of affairs involving any object and

the property F; and it will be made truth-valueless by there not

being any determinately holding state of affairs involving an object

and the property F, together with there being some states of affairs

of that form that do not either determinately hold or determinately

not hold.

In addition to predicates that stand for properties, there are

predicates that do not stand for properties. We will be able to prove

below that if there is indeterminacy of certain sorts, then some

predicates do not stand for properties, yet sentences containing

them will be made true (or false or neither) by how the world is.

For example, if 'F' stands for a property, then we might have a

predicate meaning 'is neither determinately F nor determinately

not F'. There is no reason to expect there to be a property answer-

ing to this latter predicate (though there might be)"; if there is not,

such a predicate does not stand for a property, though it is clear

how the truth of sentences containing it are validated (or not) indi-

rectly by the facts. When we discuss existing predicates arising from

philosophical examples, it may be uncertain whether they stand for

properties; we will sometimes need to discuss issues that are for-

mulated in terms of such predicates while being uncertain whether

they stand for properties.

In order to clarify in general how aspects of the world make sen-

tences about the world true (or false, or neither), we will need to

say in detail how our language works and how it is related to the

world. So we must discuss the semantics of the language that will

be used in stating our metaphysical theses; this is the business of

the next section. It turns out that we do not need to speak of states

of affairs in order to do this, so they will henceforth vanish from

our exposition. We do need to speak of objects, properties, and

relations.

n Typically, whether a predicate can stand for a property depends on how

the world goes. For example, if it is indeterminate whether a = b, then we can

show that the predicate 'being a thing such that it is indeterminate whether it is

/J' will be a predicate that cannot stand for a property; see Chs. 4 and 7. But if b

is not indeterminately identical with any object, there is no inconsistency in

holding that 'being a thing slIch that it is indeterminate whether it is b' stands for

a property.

16 Indeterminacy

2.4 LANGUAGE: SYNTAX AND SEMANTICS

The symbolism used herein consists mostly of the first-order

predicate calculus. It contains names of objects, and predicates that

are true (or false, or neither) of objects. There are connectives, and

there are quantifiers to express generality. I postpone introducing

a sign for identity until the next chapter, and I postpone introduc-

ing a conditional sign until Chapter 6; otherwise, what follows here

is the language that will be used throughout the book.

Names: I assume we have a stock of names that name objects.

Each name names a unique object. There are no denotationless

names, and no names whose denotation is ambiguous. I make these

assumptions for simplicity of discussion; the issues are sufficiently

complex that we should not have to worry about whether, for

example, a name lacks reference. (In Chapter 9, I discuss the claim

that names cannot unambiguously denote objects which enter into

indeterminate identities. My position is that they can, and I assume

here that they do.)

Predicates: A one-place predicate may stand for a property, or it

may not, and a two-place predicate may stand for a relation, or it

may not. We need to allow for both options. In either case I assume

that a predicate is true of some (or no) objects and is false of

some (or no) objects. I decree that there is no object such that the

predicate is true of it and also false of it. There may be objects

the predicate is neither true nor false of. (Two-place predicates

are true of/false of/neither true nor false of ordered pairs of

objects.)

If a predicate stands for a property, then it is true of the objects

that (determinately) possess the property, and it is false of those

objects that (determinately) lack the property. So if an object

neither determinately has nor determinately lacks the property,

the predicate will be neither true of nor false of it. (Similarly for

two-place predicates that stand for relations.)

Truth: TIle truth conditions for formulas and sentences are given

here in a simple form. Most people knowledgeable about logic will

know how to rriake this more rigorous (e.g. by defining satisfaction

of formulas by infinite sequences of objects assigned to the

variables), and most others will not be concerned.

Atomic sentences: An atomic sentence consists of a one-place

Indeterminacy 17

predicate and a name, or a two-place predicate and two names.

Such a sentence is true if the predicate is true of the object named

by the name (or the pair of objects named by the names), and false

if it is false of the object named by the name; otherwise the sen-

tence has no truth-value.

COllnectives and quantifiers: There are many different ways to

define connectives and quantifiers for a language in which some

sentences lack truth-value. We need to choose some particular con-

ventions in order to have a definite framework, while remaining

sensitive to the fact that other writers may interpret connectives

differently. I make my own choice here, and discuss alternatives

when they are relevant. For familiarity, I assume that all connec-

tives and quantifiers behave in the usual ways when the formulas

with which they combine possess truth-value. The quirks arise only

for cases without truth-value. For these J choose the Lukasiewicz

conventions, for their naturalness:

Negation: A negation of a true sentence is false, and a nega-

tion of a false sentence is true. If a sentence lacks truth-value,

so does its negation.

Conjunction: A conjunction is true if both conjuncts are true,

false if either is false, and otherwise lacking in truth-value.

Disjunction: A disjunction is true if either disjunct is true, false

if both are false, and otherwise lacking in truth-value.

These connectives are summed up in Tables 2.1 and 2.2, in

which the dashes stand for absence of truth-value.

Quantifiers: The quantifiers ::3 and V are understood as gener-

alizations of conjunction and disjunction:

::Jxcj>x is true jf cj>x is true for at least one assignment to x, false

if cj>x is false on every assignment to x, and otherwise ::3x<jlx

is lacking in truth-value.

\lxcj>x is true if cj>x is true for every assignment to x, false if

cj>x is false on at least one assignment to x, and otherwise

\lx<jlx is lacking in truth-value.

The determinate truth connective: It will be useful to have a

"non-classical" connective for truth. I use '!' for this. The con-

vention is:

'!S' is true jf 'S' is true, and otherwise false.

18 Indeterminacy

TABLE 2.1. TABLE 2.2.

A B A&B AvB A -.A

T T T T T F

T T

F T F T F T

T T

F F

T F F T

F F

F F F F

I call this the "truth" connective, though it is probably better called

the "determinately" connective. That is, you should read '!S' as

'Determinately, S'. The notion of truth gets in only in the metalan-

guage, when we state the truth-conditions for the elements of our

language: 'Determinately, S' is true if'S' is true, and false if'S' is

false or lacks truth-value.

Using this notation we can express other ideas. For example, the

falsehood of'S' is expressed by its negation being (determinately)

true: '!--S. We can also express indeterminacy. There is an

established sign for this in the literature, originating in Evans

(1978): we express a claim that S is indeterminate by writing 'vS',

defined as:

This means that it is not determined whether S. Thus vS is incom-

patible with S's being true and with S's being false. A sentence of

the form 'vS' is true if and only if'S' itself lacks truth-value. These

conventions are summed up in Table 2.3.

This is most of the language we will use throughout. The

language is not "bivalent" in the usual sense because it allows the

possibility that sentences may lack truth-value. But it is also not

"multivalent" as this is usually understood, since this usually means

that there are more than two truth-values. For the language used

Indeterminacy 19

TABLE 2.3.

(determinate) (determinate)

truth falsity indeterminacy

S !S

!--,s

vS

T T F F

F F T

F F T F

here, there are only two truth-values, true and false, though there

are three truth-value statuses: true, false, and neither.

7

2.5 DETERMINACY IN NATURAL

LANGUAGE ASSERTIONS

A sentence expressing an indeterminate state of affairs lacks truth-

value. When we assert sentences, we generally put them forth as

true, so an assertion of'S' will commit the assertor to the claim that

the state of affairs that S determinately holds.

s

This means that for

assertions it is generally redundant to use the adverb 'determi-

nately' to modify the whole sentence used to make the assertion.

With regard to subclauses of assertions, however, it mayor may not

be redundant to add 'determinately'. Since it is not always imme-

diately obvious whether it is redundant or not, I will occasionally

burden my sentences with redundant uses of 'definitely' and

'determinately' so that the reader will not have to pause to figure

out whether determinacy is entailed (or required) or not.

How do we deal with indeterminacy in real life discussion?

7 For most purposes it does not matter whether one assumes that a sentence

which is neither true nor false has no truth-value, as I do, or assumes that it has a

special "neuter" value. I take the former option merely because it does not require

explaining what an additional "truth"-value is which is neither truth nor falsehood.

S It is possible to commit oneself merely to the non-falsehood of the proposition

asserted, but this usually takes a special convention. See Woodruff (1970) for a study

of "hedged assertion" in formal languages, and Parsons (1984) for natural languages.

These options are especially relevant to work on the semantic paradoxes. Since the

paradoxes are not at issue here, I omit discussing them.

20 Indeterminacy

Presumably, if we think a claim is indeterminate we avoid making

either that claim or its negation (which is also indeterminate). But

what if we are called upon to take sides? We have one of two

resources. One option is to retreat to a meta-Ievel and say some-

thing like "You can't say one way or the other", or "That can't be

determined", or "There's no answer to that question", or some

variant of these. If we are philosophers, we will feel called upon to

say more, but in ordinary conversations such a comment will usually

suffice. A second option is to "deny" the claim, usually with a special

emphasis on negation; this is followed up, if need be, by denying the

opposite as well. If asked whether the ship that docked in a given

port is the same as the one that left the original port, you might say

"It's not the same, but it's not different either; there's just no telling",

or even "It's not the same, but it's not not the same either." Denying

that it is the same ship is not the same as asserting the negation of

the claim that it is the same ship, because that would make you

take sides, which is exactly what you think should not be done. So a

denial is either an assertion of a special kind of negation, or it is itself

a special kind of speech act-an act which denies the content "that

it is the same ship", as opposed to an act which asserts the different

content "that it is not the same ship". In this book I take the con-

servative option, which is to see this kind of denial as assertion of

the special kind of negation called "exclusion negation".9 Unlike

normal negation, the assertion of the exclusion negation of a claim

merely denies that claim, it does not endorse its opposite (its

ordinary negation). Whereas the ordinary negation of a claim S is

another assertion, 'not S', which lacks truth-value if S does,

exclusion negation is used in such a way that 'ex-not S' is an

assertion that is true when S lacks truth-value (Table 2.4).

This, then, is the kind of negation that we are trying to express

when we reject a claim with special intonation in order to distance

ourselves from taking sides regarding it. Exclusion negation is

9 It is hard to say which of these is the best 3ccount, or even if there is a differ-

ence between them. But some such account is needed, because there must be (and

is) an alternative to asserting the negation of a claim that you think is indetermi-

nate. I have argued in Parsons (1984) that in some recalcitrant cases (cases involv-

ing semantic paradox) we are forced sometimes to see denial as a speech act

separate from assertion, rather than see it as the ordinary assertion of a special kind

of negation. Since this book is not about the paradoxes, I here take the simpler line

of preserving a single kind of assertion and employing a special kind of negation

when needed. It is not a trivial task to repackage a theory developed using exclu-

sion negation into one that appeals to a special kind of denial.

S

T

F

Indeterminacy

TABLE 2.4.

negation

not S

F

T

exclusion negation

ex-not S

F

T

T

21

easily defined in terms we have available from above; the exclusion

negation of'S' is just '-,!S'. If we want to say in the object language

that a claim lacks truth-value, we just exnegate both it and its nega-

tion, saying '-,!S & -,!-,S'. Indeed, this is our proposal above for

introducing a connective for indeterminacy. This kind of notation

corresponds to things we actually say when forced to confront

claims in real life that we think are indeterminate.

'o

to Here and throughout this book I am making a simplification that is strictly

wrong, but convenient. This is to ignore certain instances of what Frege calls indi-

rect contexts. I argue in Parsons (1996, 1997) that whenever we have a that-clause

or a whether-clause we have a case of indirect discourse, a context in which we have

shifted to talk of the meaning of the contents of the clause. This is obvious in 'She

means that snow is white', or 'She believes that snow is white'; I think it is equally

true for 'It is true that snow is white' and 'It is not the case that snow is white'. (In

Fregean terms, the sentences' It is true that snow is white' or That snow is white is

true' attribute truth to a thought, which is referred to by the that-clause.) As a con-

sequence, there is no such connective as 'it is not the case that'; the phrase 'it is not

the case that' is a complex phrase that contains a connective, 'not', but that is not

itself a connective; it is more than a connective since it also creates an indirect

context. Nonetheless we are inclined to call it a connective because it resembles the

behaviour of one in spite of its complexity: the 'that' makes the phrase that follows

refer to its meaning, and the 'it is the case' correctly characterizes that meaning if

and only if the sentence following the 'that' is true when normally interpreted. As

a result, these two phenomena cancel each other out, and the complex phrase 'it is

not the case that' has the same overall effect as 'not' all by itself, with the exception

that the long phrase goes uniformly in front and the 'not' must usually go some-

place inside. Something like that is at work in the suggested readings above for the

determinate-truth connective. It is not strictly correct to read '!S' as 'it is true that

S', or 'determinately, S', because '!' is not itself to be pronounced at all; its effect in

speech shows up only as a special intonation or stress on the sentence that follows

it. Exclusion negation ' ~ ! ' is also just pronounced 'not', but with a special stress that

settles on the 'not' itself or some parl of the sentence that follows. Since stress is

rarely unambiguous in what it contributes to meaning, I will persist below in the

inaccurate but clearer policy of reading '!S' as 'it is (determinately) true that S' and

' ~ ! S ' as 'it is not (determinately) true that S', with the understanding that the nota-

tion is intended to be my way of symbolizing some uses of 'not S', as the occasion

demands.

22 Indeterminacy

2.6 LOGIC

When we wish to be precise in using a language that permits

truth-valueless sentences, the logical truths that result and the

inferences that are valid differ subtly from the ones we are used to

when we idealize and assume bivalence. There is nothing difficult

or mysterious about this, but because of lack of familiarity it may

not be obvious without some thought, and even some analysis, what

is permitted and what is not. This is especially so for people who

have been trained in logic, for they have developed quick instincts

for what is the case, and these instincts must sometimes be over-

come. In any event, it is worth taking some time to go over the

consequences for logic of admitting the possibility of lack of

truth-value.

Things are simple in one way: the language described above (and

used throughout this book) contains no non-extensional contexts.

I omit these because they have no bearing on any of the issues to

be discussed. So one may "quantify into" any context. That is, if

' ... a .. .' is a sentence containing a name 'a', the following is always

valid:

... a ...

3x(. .. x ... )

And when we introduce identity (in the next chapter), this infer-

ence will always be valid:

a=b

... a .. .

... b .. .

Validity: I assume that when we discuss validity we are discussing

truth-preservation. That is:

An argument is formally valid if and only if any instance of it

that has all true premisses also has a true conclusion.

One could mean different things than this by validity, and some

people do. For example, one might require of a valid argument

that it preserve truth and also preserve non-falsehood, so that no

instance of a valid argument can lead from premisses that lack

Indeterminacy 23

truth-value to a conclusion that is false.

lI

I do not require this

additional complication because I judge that this is never needed

to assess arguments that are actually given in the literature. (Of

course, if you are arguing from true premisses, the differences will

not matter.) This leaves open the possibility that in some of the

philosophical discussion addressed below there might be mis-

understanding due to a conflict of assumptions about what is meant

by validity. I think this does not actually happen, but the reader

should be on guard.

The semantics described in §2.4 determines the valid inferences.

These valid inferences may also be studied by using formal proofs

that are based on the semantics. I give a set of rules of proof here.

Except for abstraction principles, these include all of the rules of

proof that will be appealed to in this book.

In order to state all the rules here, J need to add a conditional

connective. I use the symbol '=>' for this connective. Its definition

is that '<\> => '1" is true if '<\>' is false or ''I'' is true or both '<\>' and ''I''

lack truth-value; it is false if '<\>' is true and ''I'' is false; it is other-

wise lacking in truth-value. This connective will be discussed and

its choice justified in Chapter 5; I include it here in order to state

a set of rules that will not need to be supplemented later. I also

include the rules for identity, though they will not be discussed until

the next chapter.

The following are (partly redundantY2 rules for a natural deduc-

tion system in which premisses are introduced at will, and sub-

proofs can appear at any later point. A line dominates a later line

if the dominating line is not within a previously completed sub-

proof. Almost any currently popular version of natural deduction

(modified as indicated below) will do. In the system given here,

when a rule indicates that something may be inferred from certain

given things, those things may be anywhere in the proof so long

as they dominate the line being derived.

Except for the quantifier/determinacy commutation rules and

the non-falsity rules for quantifier inferences, these rules look quite

!! Van Inwagen (1988) does this, as does Cowles (1994) following him. This must

be kept in mind when assessing any of the claims of those papers. My views are

much closer to the views expressed by van Inwagen and Cowles than one might

think, since apparently conHicting claims turn out not to conHict once the alternate

meanings of 'valid' are unpacked.

12 For ease of use, more rules are stated here than are logically required; e.g. RAA

can be inferred from CP and Rule T.

24 Indeterminacy

TABLE 2.5. Rules for a natural reduction system

Rule T: Infer anything on a line if it is tautologically implied by a set of

sentences on lines that dominate that line.

Rule RAA: If there is a subproof whose sole hypothesis is '(f and whose

last line is of the form '1jf&.....,Ijf', then the subproof may be followed by

'--,!(p'.

Rule CP: If there is a subproof whose sole hypothesis is '<jJ' and whose

last line is 'IV', then the subproof may be followed by '!<jJ ~ IV'.

Self-Identity: '\7'x(x == x)' may be written on any line.

Leibniz's Law: From's == t' and '<jJs' infer '<jJt',

where '4>t' results from '<ps' by replacing some or all occurrences of's' in

'<jJs' by't'.

Quantifier Rules: We adopt the usual rules of Universal Specification,

Universal Generalization, Existential Specification, Existential

Generalization.

Quantifier Rules for Non-falsity: We also adopt "non-falsity" versions of

the rules of Universal Specification, Universal Generalization,

Existential Specification, Existential Generalization by prefixing their

formulas with '.....,!.....,'. For example, the non-falsity version of Universal

Specification is that from '.....,!.....,\7'xFx' one may infer '.....,!.....,Fa'.

Quantifier Interchange: '3x<!J' may be interchanged with '.....,\7'x.....,<jJ' in any

context; likewise for '\7'x<!J' and '.....,3x.....,<!J'.

QuantifierlDeterminacy Commutativity: '!\7' x<!J' is interchangeable with

'\7'x!<P' everywhere. Likewise '!3x<jJ' and '3x!<jJ'.

normal. Their principal novelty lies in the modified reductio and

conditional proof rules, and in the fact that tautological implication

has different forms than in bivalent logic. Tautological implication

is defined as follows:

A set of sentences tautologically implies a given sentence if and

only if every assignment of truth-value statuses to the non-

molecular parts of the sentences which makes all the sentences

in the given set true also makes the given sentence true.

This definition relies on the truth-table accounts of the connectives,

including the non-standard '!' and 'v' connectives and the '=>'

conditional. Since these truth-tables are new, so are the resultant

tautological inferences. I summarize here some points of com-

parison with ordinary bivalent logic.

Indeterminacy 25

First, consider sentences of the ordinary predicate calculus as

described above tha t do not contain the connectives'!' or 'v' or '=>'.

Then:

1. Classical tautologies (those made without the use of the

special connectives) are not provable. Classical tautologies are

not tautologies here because of the possibility of lack of truth-

value. For example, A v -.A is not a tautology because if A

lacks truth-value, so does -.A, and so does the disjunction. The

"Law of Excluded Middle" does not obtain.

2. Most inferences from premisses, however, work in the usual

fashion. That is, if you have some premisses, and a conclusion

that classically follows from those premisses, then that infer-

ence is usually valid here. This is because 'follows' means that

if the premisses are true the conclusion must be true too.

Assuming that the premisses are true very often amounts to

assuming that there are no gaps in them in positions that could

make a classically valid conclusion lack truth-value. The com-

monest exceptions to this are cases of "irrelevant" inferences;

e.g. in bivalent logic you can infer 'A v -.A' from 'B v -.B', and

you cannot do that here. But you can infer 'A' from 'A v B' and

'-.B', and you can infer '-.A v -.B' from '--,(A & B)', and so

on.

3. Reductio ad absurdum: You can show that something is

not true by deriving a contradiction from it. But that does

not show that its negation is true. And so the proof tech-

nique of indirect proof (Rule RAA: show that '-.S' is true by

deriving a contradiction from'S') cannot take its usual form.

Illustration:

If you assume A&-.A as a hypothesis, you can easily infer

a contradiction from it (itself). Classical indirect proof

would then let you infer -.(A&-.A). But if A lacks truth-

value, so does -.(A&-.A), so you are not allowed to infer

that.

When the truth connective (the "determinately" connective) is

included, the following hold:

1. This connective does not interfere with any of the infer-

ences we just discussed. If you can prove something in a way

26 Indeterminacy

I just described, the presence of the other connectives does not

interfere with this.

2. The truth connective commutes with quantifiers. So you

needn't worry about properly ordering it with respect to the

quantifiers. For example, !3xFx will always have exactly the

same truth-value status as 3x!Fx.

3. The truth connective distributes across & and v:

!(A&B) is equivalent in truth-value status to !A&lB, and

!(AvB) is equivalent in truth-value status to !Av!B.

This connective does not, however, commute with negation;

--,lA is not generally equivalent to !--,A.

4. There are tautologies and logical truths involving '!'. For

example,A v --,!A is a tautology, and Vx(Fx v --,!Fx) is logically

true.

5. There is a form of indirect proof. If you derive a contradic-

tion from S, then although that doesn't show --,S, it does show

--,!S. So although this is not valid:

NO: l ~ ____ _

~

--,S

the following is valid:

Y E S : ~

I A &--,A

--,!S

Contrapositive reasoning: TI1e above principles are natural and

easy to recall. A subtler point is this: if you have a valid inference

pattern, contrapositive forms of that pattern need not be valid. For

example, if this is a valid form:

A

:.B

then you cannot assume that this form is valid:

--,B

:.--,A

J ndeterminacy 27

As a counter-instance, it is plain that this inference pattern is valid:

A

:.lA

because if the premiss is true, so is the sentence that says it is true.

But the contrapositive principle is not remotely compelling:

-,!A

:.---,A

Tt A lacks truth-value, then the premiss is true and the conclusion

is not true, because it lacks truth-value. This failure of contra-

positive reasoning does not rely on any esoteric doctrine; it is an

inevitable and natural result of admitting the possibility of a

sentence's lacking truth-value.

There is, of course, a reliable form of "near-contrapositive"

reasoning, using the truth connective. If this is valid:

A

:.B

then so is this:

In my opinion the major cause of ongoing controversy regarding

indeterminacy in general and indeterminacy of identity in paJ-ticu-

lar is our tendency to take for granted contraposilive reasoning

when using propositions that may lack truth-value. This type of

reasoning is so natural to us when dealing with truth-valued

claims that we instinctively pursue it when dealing with meaning-

ful claims that may lack truth-value, where it is straightforwardly

fallacious. I will touch on this point a number of times throughout

the book.

The point just made about contra positive reasoning depends on

the use of a negation connective that produces a sentence without

truth-value when appended to a sentence that lacks truth-value.

This is the sort of negation that is usually intended when people

discuss issues of indeterminacyY If one uses "exclusion" negation,

" This is based on my reading of the literature; others might disagree_ Fortunately.

people usually specify how they intend their negations to be taken_

28 Indeterminacy

things are different. Recall that the exclusion negation of a sen-

tence, produces a sentence that is true when'S' lacks truth-value:

ex-not S =df -.!S

The use of exclusion negation removes all truth-value gaps, and so

contra positive reasoning is valid using it. But exclusion negation

has idiosyncrasies of its own; for example, the law of double ex-

negation does not hold. (From 'ex-not ex-not S' one may not infer

'S', though the reverse is valid.) I find it more revealing to think of

exclusion negation as a way of denying the truth of a claim, and

writing this explicitly using '-,!' instead of a single connective 'ex-

not'. When discussing the views of others, I need to be careful to

match their own use of negation, and the reader needs to be alert

that no questions are begged because of unclarity over what kind

of negation is used.

2.7 WHAT IS NOT AT ISSUE

Certain topics are not discussed here, though they may be called to

mind by some of the discussion so far.

Semantic and/or set-theoretic paradoxes: Much work in the last

few decades on systems of logic that allow for truth-value gaps have

been motivated by the study of semantic paradoxes or, to a lesser

extent, by the paradoxes of set theory. Those topics are not under

consideration here.

Vagueness: People sometimes use 'vagueness' to express what I

call indeterminacy. I avoid the word 'vague' because it suggests a

very different study of one or more very different notions.

1. Vagueness is usually thought of as a property of language or

of concepts. Some people think that this kind of vagueness is to

blame for identity puzzles, but I do not, and we need to keep the

views straight. The topic here is indeterminacy in the world,

not vagueness of words or concepts. It may be difficult to dis-

tinguish these in practice, but it is essential to use terminology

that at least allows the distinction a chance to be made. 14

14 The use of 'vague' makes a transition between language and the world overly

natural. Consider Pelletier (1989). He quotes me in Parsons (1987) talking about

indeterminacy of truth-value, using 'indeterminacy' to describe this. He rewords this

as "vagueness, or indeterminateness" (Pelletier 1989: 492), and then he concludes

Indeterminacy 29

2. When people use 'vague' to characterize things in the

world, they often speak of vague objects or of whether the iden-

tity relation itself is vague. As explained early in this chapter,

I see indeterminacy as applying at the level of states of affairs,

and I do not attempt the additional step of blaming it on either

the objects or the properties or relations making up those

states of affairs. One can define a "vague object" as an object

x such that for some object y it is indeterminate whether

x = y. This, however, does not blame vagueness on x itself, as

opposed to y, or to the identity relation.

3. Studies of vagueness often include cases in which the

vagueness is due to a borderline case, where it is a matter of

degree whether a predicate applies to an object, whereas I

develop a theory in which indeterminacy is not a matter of

degree; in language it leads to a lack of truth-value, not to

degrees of truth-values. Perhaps this is an inadequate way

to view things, but it would be misleading to use a term that

suggests the theory is something it is not.

4. It is common to assume that along with vagueness comes

higher-order vagueness; not only is it vague whether this is

orange, it may be vague whether it is vague whether this is

orange. Indeterminacy is not like this, at least in the version

of the theory 1 discuss in most of the book. Higher-order

indeterminacy is discussed briefly in Chapter 12.

Fuzzy logic: Fuzzy logic is a study that has been much discussed

over the last couple of decades because it has been taken as a

slogan for a certain kind of computer programming; it refers to soft-

ware systems that are designed to allow for "risky" reasoning, infer-

ences to conclusions that are not deductively validated by the given

data. It sometimes concerns reasoning to conclusions that have def-

inite truth or falsity, but their truth or falsity is yet unknown, and

the scale of "truth-values" in fuzzy logic is often taken as repre-

senting degrees of knowledge. My enterprise is not epistemo-

logical at all. When 1 say that a sentence lacks truth-value this

is because the world gives it none, not because we are presently

ignorant of what it is.

"vagueness ought to be viewed as a semantic notion, and investigated by means of

Jifferent evaluation techniques." (ibid.). I agree totally with his conclusion, but the

subject has changed. One kind of vagueness is indeed a (semantic) notion essentially

involving language. Its study leaves untouched the indeterminacy I am investigating.

30 Indeterminacy

As originally developed, "fuzzy logic" has a technical definition;

it is a semantical system in which there are additional truth values

between 0 and 1, usually an infinite number of them, representing

degrees of truth. Again, I have only two truth-values: true and false,

and three possible truth-value statuses: true, false, and neither; I do

not think of "neither" as a degree of truth, though I am not sure

what difference this would make.

Non-extensional contexts and modality: These issues are not

directly relevant to anything discussed in this book. I discuss only

language with extensional contexts, and no modal notions are

invoked in the analysis.

3

Identity

3.1 WHAT IS IDENTITY?

When are objects a and b identical? T accept the Leibnizian doc-

trine that defines identity between objects in terms of coincidence

of their properties. If there is indeterminacy in the world, then it

may be indeterminate whether a and b coincide in the properties

that they share. Consequently, it will be indeterminate whether they

are identical. Specifically:

x is determinately identical with y if and only if: every prop-

erty that x determinately possesses y also determinately pos-

sesses (and vice versa) and every property that x detenninately

does not possess y also determinately does not possess (and

vice versa).

x is determinately not identical with y if and only if: there is

some property that x determinately possesses that y determi-

nately does not possess, or some property that x determinately

does not possess that y determinately possesses.

It is indeterminate whether x is identical with y if and only

if x and y are neither determinately identical nor deter-

minately not identical. This will happen where there is no

property that x determinately possesses that y determinately

does not possess, and vice versa, but there is at least one

property that x determinately possesses such that it is inde-

terminate whether y possesses it, or vice versa, or at least

one property that x determinately does not possess such

that it is indeterminate whether y possesses it, or vice

versa.

More succinctly:

32 Identity

x is determinately identical with y if and only if x and y deter-

minately possess and determinately lack exactly the same

properties.

x is determinately not identical with y if and only if there is

some property regarding whose possession x and y determi-

nately disagree.

Otherwise it is indeterminate whether x is identical with y.

1l1is account is meant to be a "real definition", not a "nominal def-

inition". That is, the account is not meant to explain what I take

'identical' to mean. I mean by 'identical' exactly what others mean

by it; this is the only way I know to guarantee that we are discussing

the same issue. I assert the equivalences above as truths about

identity. That is, the theory under examination is a theory that

embodies these equivalences as substantive claims about identity. I

I use 'distinct' for 'not identical'. (That is a nominal definition.)

Indeterminacy of identity between a and b thus coincides exactly

with indeterminacy of distinctness.

The account of identity given above in terms of properties says

nothing about relations. Suppose that x and y coincide completely

in terms of their properties, but differ determinately in terms of

their relations to objects?2 Then they should be distinct, not iden-

tical. So a better account than the one above would read:

x is determinately identical with y if and only if x and y deter-

minately possess and determinately lack exactly the same

properties, and determinately stand in, and determinately do

not stand in, the same relations to the same objects.

x is determinately not identical with y if and only if there is

some property regarding whose possession x and y determi-

I Here and throughout the text [ ignore the complication of lime, assuming e.g.

that lhe properties in question are things like "being yellow at tOO as opposed to

"being yellow", where the latter is a property that one and the same thing possesses

at one time and not at another. If I were to appeal to the latter notion of property,

identity would have to be defined in terms of "having all the same properties at all

the same times", instead of just "having all the same properties". I use the former

notion because the account is less complicated; I think that nothing of substance

turns on the choice.

2 An old illustration is of a universe consisting of nothing but two spheres of the

same size. composition, etc., existing some distance apart; the spheres are supposed

to be two objects that differ only with respect to their relations. (One is distant from

a certain object-the other sphere-and the other is nol distant from that object.)

Identity 33

nately disagree or some relation to some object regarding

which they disagree.

Otherwise it is indeterminate whether x is identical with y.

Officially I adopt this more refined account as the official view. But

it is clumsy to state, and there is a simple way to merge the options.

Suppose that for any relation R and any object 0 there is a prop-

erty P which is determinately possessed (or determinately dispos-

sessed) by exactly those objects that determinately bear R (or that

determinately do not bear R) to o. Then the accounts coincide.

I will indeed make this assumption. It is a controversial assump-

tion. but so far as J can see it is irrelevant to any of the issues about

indeterminate identity to be discussed. If it becomes relevant,

we should retreat to the account of identity that explicitly takes

relations into account.

Is identity itself a relation? It can he, but it need not be; the

assumptions made so far leave this open. This is primarily because

so little has been assumed. For example, I have not even assumed

that if P and Q are properties, then their conjunction is a property.

(I have not assumed that there is some property U that determi-

nately holds of exactly the objects that both P and Q determinately

hold of, and determinately fails to hold of exactly the objects that

either P or Q determinately fail to hold of.) Later I discuss appli-

cations in which I assume a principle of plenitude for properties

that will yield such conjunctive properties, as well as others, but

nothing controversial will turn on that. These questions about

how properties are related to relations, and to one another, are

important, but they are technical and they are distracting, and they

are not crucial to the ontological issues discussed here.

It should be clear that I reject as basic any account of identity in

terms of coincidence of the predicates in some language that

combine truly with the names of the objects in question. Such an

account can be wrong in two ways. First, if the language is too

impoverished, we may not have the words to express a difference

in properties between x and y when one exists. Second, there is no

automatic guarantee that predicates stand for properties, and if

they do not then it might be possible for such a predicate to

combine truly with a name of x and falsely with a name of y even

when it is not determinate whether x is y. 111is can (and will)

happen sometimes when the content of the predicates is partly

34 Identity

linguistic or semantic. Cases of this sort will be discussed exten-

sively below.

I take it as obvious that identity between objects in the world is

basically a matter of what the world is like, and only secondarily a

matter of how language functions. 1 know of no conclusive way to

argue this point, so I merely state it? This leaves me at odds with

certain philosophers who take the contrary view. For example,

Harold Noonan urges that identity be defined in terms of the

behaviour of predicates and names, and he suggests that if identity

is defined in terms of properties, and if satisfying a predicate does

not necessarily count as a property, then one is speaking merely of

'a kind of relative identity: a relation which ensures indiscernibility

of its terms in some, but not all, respects' (Noon an 1984: 118).

I think that Noonan is mistaken in this claim about relative

identity; this will be discussed below. For now, I state the conflict

so as to contrast my own position.

3.2 INDETERMINATE IDENTITY

The ontological picture introduced in the last chapter is neutral

about whether any indeterminacy obtains at all. It is consistent to

append to the account the assumption that every property is actu-

ally determinately possessed or determinately not possessed by

every object. It is also consistent to assume otherwise, and I will

continue with that assumption in effect, since otherwise the theory

has no interesting application.

If there is indeterminacy in the world, then it is still left

open whether it extends to identity. Perhaps there is a great deal

of indeterminacy, but for any objects a and b either their own

indeterminacies completely coincide or else they determinately

disagree about possession of some property. Then there would be

indeterminacy without indeterminacy of identity of objects.

4

We

cannot settle a priori whether this happens. I will presume that

3 See Keefe (1995: 185 ff) for considerations in favour of this position.

4 I leave open for the present whether there might be indeterminacy of identity

of sets of objects even when there is no indeterminacy of identity for the objects

themselves. A positive answer to this question is consistent and in teresting; it is

discussed in Ch. 11.

Identity 35

identity itself is sometimes indeterminate, for this is the interesting

case, but it is a contingent matter.

It is sometimes convenient to speak of an identity between two

objects as being indeterminate; this is always elliptical for the

proposition that it is indeterminate whether the objects in question

are identical.

3.3 LAWS OF IDENTITY

The semantics for the identity sign are:

5

The predicate '=' is true of exactly those pairs of objects <x,y>

such that x is determinately identical to y.

The predicate '=' is false of exactly those pairs of objects <x,y>

such that x is determinately not identical to y.

Given this explanation together with the semantical and meta-

physical theses laid out above the following two principles must

automatically hold for the identity sign:

Reflexivity: Any sentence of the form 'a = a' is logically true.

6

Likewise, the sentence ''v'x(x = x)' is logically true.

Law: The following holds, where ' ... t .. .' represents

any sentence containing the term 't' in one or more places:

s =t

... s ...

. . . t ...

That is, given any identity between terms, and any sentence

whatever containing one of those terms, you may infer the

same sentence with the one term replaced by the other in as

many of its occurrences as you like.

-' The first of these two clauses is emphasized in Stalnaker (1988: 350). When inde-

terminacy is possible, we must consider the "anti-extension" of a predicate as well

as its extension: identity is no exception.

6 Again, this is because only names with unique denotations are permitted in the

language. If there were denotationless names in the language, sentences of the form

'a a' would not be logically true. In Ch. 9, I argue that it is possible for a name to

have a unique denotation even when it names an object which is indeterminately

identical with some object that the name does not denote.

36 Identity

We can show that Leibniz's Law holds by an inductive proof. We

have three base cases:

l. Suppose first that the sentence ' ... s .. .' is atomic; for sim-

plicity, suppose it consists of a one-place predicate 'P followed by

's'. Assuming that's =: t' is true, the object named by's' is identical

to that named by 't'. Assuming that 'Fs' is true, it must be that 'F'

is true of the object named by's'. So 'F' is true of the object named

by 't'. So 'Ft' is true.

By similar reasoning, the inference holds when ' ... s . . .' is '-,Fs'

(just replace 'true' by 'false' and add a clause for negation).

By similar reasoning, the inference holds when ' ... s ... ' is ''IFs'.

Here, since ''IFs' is true, 'Fs' lacks truth-value, so 'P is neither true

nor false of the object named by's'. So it is neither true nor false

of the object named by 't'. So 'Ft' lacks truth-value, and ''1Ft' is true.

2. The law can be shown for more complex formulas by induc-

tion on the complexity of the formula.

This proof itself uses Leibniz's Law in the metalanguage, and this

might lead one to suspect that some question is being begged here.

Perhaps in some sense it is, but this is no different in principle than

any other law of logic. For example, when one shows that a certain

classical first-order system of logic obeys the principle of universal

instantiation, one uses universal instantiation in the metalanguage.

If one insists on more than this, then no system can ever be shown

to validate any law of logic. In the face of extreme logical scepti-

cism, the most that one can show is that certain principles cohere,

that is, that using the principle in the metalanguage does indeed

allow you validate it for the object language.

Since I intend the metaphysical theory under discussion to pos-

sibly be true of the actual world, then I must be using the same

logic in my metalanguage as in the object language.

Recall that validity means preservation of truth. If the identity

premiss is true, then the objects named by '5' and by 'I' are indis-

tinguishable with respect to any of their properties and relations.

Further, given that the language contains no non-extensional con-

texts, Leibniz's Law holds. Indeed, if Leibniz's Law were not to hold

for such an extensional language, this would cast serious doubt on

whether our sign of identity were actually expressing identity, as

opposed to some weaker relation.

(Leibniz's Law is expressed here as a principle of inference, as

opposed to a schema for conditional statements.1l1is is because we

Identity 37

have not as yet discussed conditionals. Conditionals bring with

them their own idiosyncrasies and controversies, and it is better

not to get those issues entangled with issues about Leibniz's Law.

Conditional formulations of Leibniz's Law are discussed in

Chapter 6.)

Leibniz's Law together with reflexivity lets us derive some of the

other well-known principles of identity, such as:

Symmetry:

s =t

:. r = s

Transitivity:

s = t

t=u

:. s =u

It is crucial, however, that contrapositive versions of Leibniz's Law

do not hold. We do not have:

... s ...

----,(. .. t ... )

:.--,s = t NO!

Recall the point made in the last chapter: if a pattern of inference

is valid, that does not guarantee that contrapositive versions of it

are valid. Leibniz's Law is a crucial example of this. As we will see

in the next chapter, if contrapositive versions of Leibniz's Law were

valid, it would be possible to disprove the truth of any assertion of

indeterminate identity. l1ms the theory under discussion must deny

the validity of the contrapositive version of Leibniz's Law.

But how can there be a counter-example to this principle? If the

premisses are both true, then it appears that sand t must disagree

with respect to the property expressed by the context ' ... x .. .'. But

then the objects denoted by's' and '( disagree with respect to that

property, and thus they are by definition determinately not identi-

cal, which is all that the conclusion states! The answer to this is that

the context ' ... x . . .' may not express a property, and it is only in

this sort of case that the principle fails. We will see examples of this

in the next chapter.

The following principle remains valid in any case:

38 Identity

... s ...

-,(oO . t oO .)

:.-,!S=l

So the loophole cases are ones in which sand t are neither deter-

minately identical nor determinately not identical; they will be

cases of indeterminate identity. With this in mind, it is easy to give

a specific counter-example to a contrapositive version of Leibniz's

Law. L ~ t the first premiss say that a is determinately identical with

a, and the second premiss say that a is not determinately identical

with b:

!a =a

-.!a = b

Obviously we do not want to conclude from the fact that a is not

determinately identical with b that it is (determinately) not identi-

cal with b:

la=a

-,la = b

:.-,a = b NO!

At least I do not want to conclude this, though it is clear that others

do. The theory of indeterminate identity must hold the contraposi-

tive of Leibniz's Law to be invalid, at least for some instances, such

as the one just displayed. In later chapters I will dispute various

arguments to the effect that the contrapositive of Leibniz's Law

must be valid.

By denying the validity of this principle I may be charged with

begging the question in favour of indeterminate identity. I coun-

tercharge that to assume the validity of the principle is to beg the

question on the other side. So we must let each side take sides, and

consider what independent considerations there might be to

resolve the issue.

(One might worry whether using the converse of Leibniz's Law

in the metalanguage would force the converse of Leibniz's Law to

hold for the object language. Yes, that is true. So it is only by adopt-

ing the desired logic in the metalanguage that I can say that the

semantics I have described works as I say. But this should be no

embarrassment. Again, I assume that my metalanguage works just

Identity 39

as my object language does. The remaining worry is whether I can

do all this consistently. That is not difficult. Consistency can be

established by an assignment of artificial entities to the language,

described in a classical set-theoretic setting. That can easily be done

by appeal to the 'pictures' described in §7.6.)

3.4 RELATIVE IDENTITY?

We may now return to Noonan's charge that defining identity in

terms of properties and relations instead of in terms of linguistic

interchangeability will yield only a kind of relative identity. 'Rela-

tive identity' is a phrase made popular by Peter Geach (in Geach

1967), who focused on the fact that we often wish to say things

like

'the first letter in "aardvark" is the same type as the second

one, but not the same token.'

Geach took the locution 'x is the same F as y' and called it

'relative identity', 'relative' because of the relativity of choice of

'F, and 'identity' because of the use of the word 'same'. It is clear

that relative identity is not the same as the non-relative identity

that I have been discussing. The most crucial difference between

them is that Geach's relative identity does not validate Leibniz's

Law; if it did, 'x is the same F as y' would entail 'x is the same G as

y', and this is supposed not to follow. So a good test for relative

identity is whether it does or does not sanction Leibniz's Law. The

identity I am discussing does sanction this law, and thus it cannot

be a species of relative identity. If Noonan means by 'relative

identity' what Geach means, his criticism does not apply to the

theory given here.

Noonan may not have meant 'relative identity' in Geach's sense.

He may merely have meant that identity defined in terms of prop-

erties and relations is just a kind of limited indiscernibility that does

not guarantee general indiscernibility. This is the crux of the matter;

if it is true that a = b, then there should not be any statement made

using one of these names (outside of non-extensional and quota-

tional contexts) that changes truth-value status when that name is

replaced by the other. Again, it is Leibniz's Law that is given as a

40 Identity

test for whether we are talking about real identity. On this test our

theory yields real identity.?

Here is a fuller quote from Noonan. He suggests that if identity

is defined in terms of properties, and if satisfying a predicate does

not necessarily count as a property, then one is speaking merely of

a kind of relative identity: a relation which ensures indiscernibility of its

terms in some, but not all, respects-in particular, not in respects only

expressible by predicates containing 'v' or synonyms of such predicates.

Noonan (1984: 118)

The implication here seems to be that an identity will not sanction

indiscernibility of its terms within contexts governed by the inde-

terminacy connective 'v'. But this is not so. At least, it is not so in

the theory under discussion here. The following is a valid pattern

no matter how the blanks are filled:

a=b

'1( ... a .. . )

:.'1( ... b .. .)'

Our identity does guarantee complete indiscernibility, even in

contexts containing 'v'. What Noonan has in mind may be a

converse principle: that if there is a discernibility in language, then

this guarantees distinctness. That is, that this is valid:

8

!( ... a ... )

!---,c .. b ... )

:.---,a =b

In Noonan (1990) he cites this principle, which he calls "the Prin-

ciple of the Diversity of the Definitely Dissimilar", and in Noonan

7 I do not mean to suggest that Leibniz's Law is itself beyond question. For

example, Zemach (1991) rejects it. He relativizes identity to ideologies, and holds

that objects may be identical with respect to one ideology but not with respect to

others. He adopts a kind of relative identity as a "substitute" for Leibniz's Law. His

theory of "vague objects" is thus quite different from the one J am developing.

8 I have simplified Noonan's formulation by representing his 'G> & "G>' as '!G>',

where '''G>' means 'it is determinately true or determinately false that'. These are

equivalent in truth-value status. Also, Noonan formulates the principle as a con-

ditional, whereas I use a principle of inference. This is sometimes a crucial

difference, for the conditional formulation is often a stronger principle. But when

the premisses (antecedents) are bivalent in virtue of form, as is true of any formula

that begins with '!', the various conditional formulations all coincide with the

principle of inference.

Identity 41

(1991) he calls this a "weaker" principle than Leibniz's Law. But it

is not weaker in the logical sense (the logical sense being that

Leibniz's Law entails it but not vice versa), for it is not entailed by

Leibniz's Law. The principle in question is just another equivalent

to the contrapositive of Leibniz's Law. (This is because putting a

determinate-truth connective'!' on the front of a whole premiss is

always redundant.)9

Perhaps the contrapositive of Leibniz's Law is valid; perhaps 110t.

But it is not obvious. It is the point at issue.

3.5 APPLICATIONS

According to the account of identity just described, if it is indeter-

minate whether a = b, then there is no property with respect to

which a and b disagree, but there is at least one property (typically,

many) that one of them determinately possesses (or determinately

does not possess) such that it is indeterminate whether the other

has it. If we think that we have a case of indeterminate identity,

then we should be able to point to some such property. I review

here the four identity puzzles cited above with this in mind.

Me and my body: The theory allows one to say that it is

indeterminate whether I am identical with my body. This

says little in detail about which properties I and my body

share. For example, suppose that my body takes up space,

and suppose that taking up space is a property. TIlen I myself

9 To make this dear in the case in question, suppose first that the contrapositive

version of Leibniz's Law holds; then so does Noonan's principle:

1. !( ... a ... )

2. !-,(. .. b ... )

3. ( ... a ... )

4. -{ .. b ... )

5. : . ...., a=b

Noonan's premiss

Noonan's premiss

From 1

From 2

By the contrapositive of Leibniz's Law

Second, suppose that Noonan's principle holds. Then so does the contrapositive

version of Leibniz's Law:

1. ( ... a ... )

2 . ....,( ... b .. .)

3. !( ... (/ .. .)

4. !-.(. .. b ... )

5. : . ...., {/ = b

Premiss of contrapositive of Leibniz's Law

Premiss of contrapositive of Leibniz's Law

From 1

From 2

By Noonan's principle

42 Identity

must either have the property of taking up space, or else it

must be indeterminate whether I have this property. Both

options are open. Conversely, if it is true that I think, and

if thinking is a property, then either my body thinks, or it is

indeterminate whether it thinks. Somewhere there must be

some indeterminacy of property in order for me and my body

to not be detenninately identical. But the case abstractly

described does not give details. If I and my body and noth-

ing else are in a certain room,1° then there is at least one thing

in the room and at most two, though it is indeterminate

whether there is exactly one thing, and indeterminate whether

there are two. (Such cardinality judgements are discussed in

Chapter 8.)

Perhaps I and my body differ only in modal properties.

Perhaps we share all non-modal properties, such as taking up

space, and thinking. But perhaps I could live on without my

body (although we both cease to exist at death, or God, con-

tingently, sees fit to resurrect my body for my comfort). What

does the theory say then? Well, if such modal properties are

properties, then if I have this property and my body does not,

we are distinct, and the puzzle has a traditional solution. If

modal predicates do not stand for properties, then the theory

does not yet take sides. The theory does not tell us what prop-

erties there are; it only says what the consequences are for

identity of possessing properties. I am inclined to view modal

properties as properties, but that is me, not the theory. The

issues discussed here are not likely to turn on considerations

of modalityY

The ship: Suppose that being located at a certain point in space

at a given time is having a property. Let 0 be the place from

which the original ship set sail at time to, and let d

1

and d

2

be

the places that the ship with new parts and the newly ass em-

In Of course, my body can't be in a room unless its parts are there too, so the

example is fanciful; I think, however, the point is clear.

II 111e options for how to do modal semantics are so numerous that just about

any view about indeterminate identity may be reconciled with just about any view

about whether modal predicates do or do not stand for properties. et. Keefe (1995)

and Stalnaker (1988: 355 ff) for some discussion. In my opinion, the additional com-

plexities (and complexities abound!) are all contributed by artefacts of modal

semantics, and shed no light on the basic question of whether identity might be inde-

terminate (in this world). This is why I do not discuss them.

Identity 43

bled ship respectively dock at time t

n

• Then the ship with new

parts determinately has the property of being at d

1

at t

n

, and

the newly assembled ship determinately lacks that property

(since it is determinately elsewhere then); as a result, the ship

with new parts and the newly assembled ship are determi-

nately distinct. It is indeterminate whether the original ship is

at d

,

at to> since otherwise it would be determinately distinct

from one of the resulting ships, and the puzzle would have a

traditional solution.

This example shows that indeterminate identity is not transi-

tive: the newly assembled ship is indeterminately identical with

the original ship, which is indeterminately identical with the

ship with new parts, but the newly assembled ship is not inde-

terminately identical with the ship with new parts. Before the

repair job began there was exactly one ship; afterwards there

are exactly two. All told, there are at least two and at most

three, with either exact answer being indeterminate.

The disrupted persan(s): It is indeterminate whether the

person who originally entered the room is the person who later

left the room. Assuming that leaving the room is a property or

entails the possession of certain properties, it is indeterminate

whether the person who entered the room left the room.

Exactly one person entered the room, and exactly one person

left the room. It is indeterminate whether there is exactly one

person overall or exactly two persons overall, but false that

there are fewer than one and false that there are more than

two.

The cat and the p-cats: There is exactly one cat on the table,

though probably millions of p-cats. For each p-cat, it is inde-

terminate whether it is identical with the cat. Since there is

only one cat, for each p-cat it is indeterminate whether it is a

cat (if it were determinately not a cat it would be determi-

nately distinct from the cat; if it were a cat, then by parity so

would be the others, and there would be millions of cats on

the table).

In describing these cases I have not only said how J think the theory

is supposed to work, I have also reviewed what I take to be some

of the data against which any theory should he tested. That is, if

44 Identity

you grant that each identity puzzle has no answer, then the rest of

the things that I say about the puzzling situations are what I think

any thoughtful person uncorrupted by a philosophical theory

would say. I say this here partly to urge agreement, but also to let

the reader know something about how subsequent discussion will

go: the theory under consideration has the virtue of agreeing with

these points, and competing theories are compelled to deny some

of them.

4

The Evans Argument, Properties, and DDiff

One should be sllspiciolls of any argument that purports to

get substantive metaphysical conclusions out of the logic of

identity.

Stalnaker (1988: 357)

4.1 THE EVANS ARGUMENT

In 1978, Gareth Evans published a one-page article, "Can there be

Vague Objects?" In spite of its title (vague objects), the aim of the

paper was to prove that no statements of indeterminate identity

can be true. Specifically, no statement of the form ''Va = b' can be

true. This paper has become a focal point for a growing body of lit-

erature, including the present book. It is thus important to consider

what Evans's argument is, and what it shows.

I set aside two potentially distracting issues:

First, it is not clear that Evans intended to address the question

that I and other commentators call indeterminacy of identity. There

is some reason to suspect that Evans saw indeterminacy as a sub-

class of truth, as opposed to a lack of truth-value. That is, he may

have intended ''VS' to mean" 'S' is true, but indeterminately so", as

opposed to meaning that'S' lacks truth-value altogether. If so, his

argumentation is irrelevant to most of the subsequent literature

that it inspired. But it is the discussion that the paper inspired that

I am interested in, so I will ignore the fact that Evans might have

meant something different. I speak of "the Evans argument" as an

argument aimed at refuting the possibility of indeterminacy of

identity as I (and most others) understand indeterminacy of iden-

tity. (I say more about the alternative interpretation of Evans in the

Appendix.)

46 Evans Argument, Properties, and DDiff

Second, as a number of commentators have pointed out, it is

obvious that a statement of the form 'a = b' can lack truth-value if

the singular terms making it up do not have clear and unambigu-

ous reference. Evans could not possibly give a conclusive argument

to refute this possibility. I agree with the majorityl in interpreting

his argument as trying to show that a statement of the form 'v(a =

b)' cannot be true if the singular terms making it up have clear and

unambiguous reference. So interpreted, the argument is a direct

attack on the central thesis of this book.

The argument: The argument is a reductio ad absurdum of

the view that a statement of the form ''la = b' is true. It goes as

foIlows:

2

(1) v(a = b)

(2) h[v(a=x)]b

(3) -,v(a = a)

(4) -,tu:[v(a = x)]a

(5) -,(a = b)

The hypothesis to be refuted

Abstraction from (1) ["Abstract

introduction"]

Truism

3

Reverse abstraction from (3)

["Abstract elimination"]

(2), (4) by Leibniz's Law

Before critiquing the argument, it needs to be completed. A

reductio argument should end in a contradiction, and this one does

not. Evans discusses this point and makes some complicated

remarks about it, which I ignore here.

4

For in the theory under

I Including Burgess (1989), Garrett (1988), and Lewis (1988).

2 Evans (1978: 208)-which is the only page of the article. I use Evans's notation

except for replacing his negation sign '-' with mine '-,', and using the slightly more

common notation 'AX[ . .. ]' for property abstraction than his 'xl .. .]'. He does not

explain either piece of notation, so one can assume that he means nothing unortho-

dox. I have also replaced 'x = a' by 'a = x' in lines (2) and (4); this bring the argu-

ment into conformity with the rationale that Evans gives for the steps. Without this

change the argument is missing two steps that appeal to the symmetry of identity

within indeterminacy contexts. This is a valid principle, but it complicates the proof,

which does not need to rely on it.

3 Evans does not call (3) a truism; he merely cites it without justification. In order

for (3) to be justified we need to assume that the term 'a' is not referentially flawed.

We may need a similar assumption about 'b' to justify step (2).

4 It is Evans's complex remarks about how to complete the proof that provide

the best evidence that he meant something different by indeterminacy than most

commentators (see the Appendix to this book). His discussion has inspired a

number of speculations about what he might have had in mind; cf. Gibbons (1982),

Pelletier (1984, 1989); nobody has come up with an interpretation of these remarks

which makes them plausible without changing the subject matter away from the

kind of indeterminacy we are discussing.

Evans Argument, Properties, and DDiff 47

discussion in this book, the argument is easy to complete. Since

validity is just truth preservation, if (5) is true so is (6):

(6) !--.(a = b).

As we have construed indeterminacy, line (1) means:

(I') --.!(a = b) & --.!--.(a = b),

and (6) clearly contradicts the second conjunct of (I'). So if

the proof is correct as given, it is easily fteshed out into a full

reductio.

s

TIle complexity of the argument is due partly to the presence of

the property abstracts. Their importance can be best understood if

we initially see what the argument would be like without them. So

remove the lines in which they occur, and consider the following

three-line variant:

(1) v(a=b)

(3) --.v(a=a)

(5) -,(a = b)

The hypothesis to be refuted

Truism

(1), (3) by Leibniz's Law

This shortened version of the argument gives the strategy in a

succinct form:

Step (1) says something about b: that b is indeterminately

identical to a. Step (3) denies this about a. So a and b cannot

be identical.

The response to this shortened form is clear. The argument does

not use Leibniz's Law at all, but rather its contra positive. This is the

principle discussed and rejected in the last chapter. The argument

thus begs the question.

However, we also noted in the last chapter that the contra-

positive of Leibniz's Law is valid if the formula in question

stands for a property. And this is the point of the additional

abstraction steps in the full argument; they introduce the required

properties:

, This fleshing out of the argument is not possible if Evans's negation is meant

10 be exclusion negation, in which case line (5) would mean what we would express

as '-.!(a = br. So there is an interpretation of the argument on which it fails to arrive

at a full reductio, thus making it irrelevant to the theory under discussion. Garrett

(1988) adopts this interpretation.

48 Evans Argument, Properties, and DDiff

(1) \l(a=b)

(2) A.x[\l(a = x)]b

(3) -,\l(a=a)

(4) -'Ax[\l(a=x)]a

(5) -,(a = b)

The hypothesis to be refuted

Abstraction from (1) ["Abstract

introduction"]

Truism

Reverse abstraction from (3)

["Abstract elimination"]

(2), (4) by Leibniz's Law

Evans says as much in justifying line (2); he says:6

(1) reports a fact about b which we may express by ascribing

to it the property 'A.x[\l(a = x)]'.

So the strategy is to see (1) as attributing a property to b, and make

this explicit in (2), and then to see (3) as denying that very same

property of a, and make that explicit in (4). Step (5) is the contra-

positive of Leibniz's Law, but applied in a special case in which it

is valid. So criticism of the full proof must focus on the transitions

from (1) to (2) and (3) to (4). And here it is apparent what must

be said by any defender of indeterminate identity: there is no prop-

erty of "being indeterminately identical to a", and so the property

abstract 'A.x[ \l( a = x)]' does not stand for a property. The inference

from (1) to (2) is fallacious, because there is no reason to think that

the sentence in (1) can be recast in terms of attribution of a prop-

erty to an object. And the proof itself gives a reason to think that

(I) cannot be so recast, for there are no other flaws in the proof,

and the proof attempts the impossible: to prove a priori the incon-

sistency of indeterminate identity.

Actually, there are two distinct ways to criticize the proof,

depending on how one interprets the use of property abstracts.

Way 1: In the first way of interpreting abstracts, one supposes that

the use of a property abstract is legitimate when, and only

when, that abstract actually stands for a property which holds

of the objects that satisfy the formula inside the abstract (and

determinately fails to hold of the objects that dissatisfy the formula,

and indeterminately holds of the objects such that it is indeter-

minate whether they satisfy the formula). This is the interpretation

employed in the previous paragraph. On this interpretation, the

fact that step (1) is true does not automatically tell you that

there is a property which reflects the semantic behaviour of the

6 Again I have replaced 'x = a' by '0 = x'; see note 2 of this chapter.

Evans Argument, Properties, and DDiff 49

formula in the abstract. On this interpretation, the inference

from (1) to (2) is fallacious. Way 2: In the second way of inter-

preting the abstracts, one does not assume that an abstract needs

to stand for a property; abstracts are just ways of reexpressing

other formulas that do not use this mode of expression, and the

semantics of formulas containing abstracts is completely parasitic

on that of the formulas one gets by eliminating the abstracts. If the

abstracts of Evans's proof are interpreted in this way, the inference

from (1) to (2) is above reproach (as is the inference from (3) to

(4»). But on this interpretation, step (5) is again fallacious. Recall

that the contra positive of Leibniz's Law does not hold for formu-

las in general, though it holds for the predication of genuine prop-

erties. On this second way of interpreting abstracts, the abstracts in

the proof are not guaranteed to stand for properties, and so step

(5) is fallacious. (The argument reduces to the shortened version

discussed above,f

7 There is yet a third way of interpreting the abstracts; one might require that

abstracts can only be used if they stand for properties, but not require that the prop-

erty that an abstract stands for holds of exactly those objects which determinately

satisfy the contained formula, and determinately fails to hold of exactly those

objects which dissatisfy the formula, and indeterminately holds of exactly those

objects which indeterminately satisfy the formula. Specifically, it is possible to give

a non-standard account of abstraction in which an abstract is always guaranteed to

stand for a property, but in such a way that the abstraction principles are not nec-

essarily satisfied. (This is similar to Frege's "chosen-object" treatment of definite

descriptions; a definite description always denotes an object, but not necessarily one

that the description part of the definite description describes.) We describe such a

non-standard interpretation of abstracts in Parsons and Woodruff (1995). This non-

standard interpretation of the abstracts does not improve Evans's proof, for on this

interpretation, although step (2) of the Evans argument indeed turns out to be valid,

step (4) becomes invalid.l ignore an extended discussion of this non-standard inter-

pretation because it does not improve the proof, and because the interpretation

itself is somewhat artificial.

(The non-standard interpretation is this: we suppose that an abstract stands for

whatever property determinately holds of exactly the objects that determinately

satisfy the contained formula. and indeterminately holds of exactly the objects that

indeterminately satisfy the formula or that do not determinately satisfy the formula

but that are indeterminately identical to objects which determinately satisfy the

formula; it determinately fails to hold of the rest. On the assumptions used in §4.3

below there will always exist such a property, and if the formula inside the abstract

does express a property (as described in §4.3),it will express a property which agrees

perfectly with the property assigned to the abstract. Otherwise, the formula will be

determinately dissatisfied by certain objects which the property assigned to the

abstract indeterminately holds of. This happens for the formula '''(a = x)'; it is deter-

minately dissatisfied by a (making step (3) true) but it is indeterminate whether

a possesses the property that the abstract stands for (making step (4) lack

truth-value). )

50 Evans Argument, Properties, and DDif(

4.2 WHY THERE IS NO PROPERTY

OF INDETERMINATE-IDENTITY-

WITH-AN-OBJECT

The argument above proves that belief in indeterminacy of identity

forces one to deny that there is a property of being indeterminately-

identical-with-o, where 0 is an object which is indeterminately

identical to something. But why should this be the case?

I t is apparent that a property abstract purportedly expressing

such a property must "quantify into" an indeterminacy connective;

the abstraction operator on the outside must bind a variable within

the scope of the indeterminacy connective:

This leads one to suspect that we might be dealing here with some

kind of non-extensionality, or some kind of modal fallacy.8 I think

this is not the case at all. As noted in the last chapter, the connec-

tives '!' and 'v' do not create non-extensional contexts; one may

freely existentially generalize on terms within their scopes, and one

may freely intersubstitute coreferential names. The phenomenon

at work here is unrelated to oddities of non-extensionality or of

semantic paradoxes; it is more closely associated with the para-

doxes of naive set theory. This can be seen without talking about

identity at all if we replace identity by its definition in terms of

properties. Instead of writing 'a = b', write:

VP(Pa {:=} Pb),

where the quantifier ranges over properties, and where the bicon-

ditional is defined to be true if both sides have the same truth-value

status (true, false, or neither) and false if they disagree in truth-

value (one side is true and the other is false). Then the indetermi-

nate identity of a with b is expressed as:

[*]vVP(Pa {:=} Pb).

The question whether there is a property of "being indeterminately

identical with a" is then the question whether the following abstract

stands for a property:

~ e.g. Burgess (1990: 269) suggests that if the Evans argument is not valid, then

one must hold that the indeterminacy connective introduces a referentially opaque

context.

Evans Argument, Properties, and DDiff 51

A.x[v\fP(Pa <=> Px)].

Suppose it does. Then we can refute [*]. For then a lacks that

property:

.....,A.x[v\fP(Pa <=> Px)]a

because this is false:

v\fP(Pa <=> Pa).

And b has that property:

AX[v\fP(Pa <=> Px)]b

because [*] is true.

So b has the property and a lacks it:

A.x[ v\fP(Pa <=> Px)]b & .....,AX[ v\fP(Pa <=> Px)]a.

Existentially generalizing on the abstract denoting the property

then yields the following as true:

~ P ( P b & .....,Pa).

But the truth of this sentence contradicts the truth of [*]. So if the

abstract stands for a property that is in the range of its own prop-

erty variable, we can show that [*] is not true.

The force behind the reasoning thus comes from the fact that

identity is defined in terms of what properties there are, and a

problematic property is defined using an abstract that quantifies

over those properties. The condition in the abstract is cleverly

designed to conflict with its yielding one of the properties quanti-

fied over (if any objects are indeterminately identical with a). The

reasoning thus resembles that of the RusselJ paradox in set theory.

(Identity between sets is defined in terms of what sets they have

as members, and a problematic set is defined using a set abstract

that quantifies over those sets. The condition in the set abstract is

cleverly designed to conflict with its yielding one of the sets quan-

tified over.)

I see no way around the facts discussed above. We thus have a

choice to make: either deny that indeterminate identity is possible

or deny that abstracts covertly employing quantification over prop-

erties (such as the one above) always pick out a property from

among the ones quantified over in the definition of identity. Either

52 Evans Argument, Properties, and DDiff

option is consistent. So the opponents of indeterminate identity can

happily give the Evans argument as an illustration of their point of

view, and defenders of indeterminate identity can happily reject its

abstraction steps as illustrative of their point of view. Neither of

these stances can refute the other side. My stance, of course, is the

latter.

4.3 DDIFF: THE CONDITION OF

DEFINITE DIFFERENCE

It is convenient to speak of abstracts as standing for or not

standing for properties, and of the formulas that generate such

abstracts as expressing (or not expressing) such properties. In

what circumstances does a formula of one free variable express

a property? The framework sketched earlier is silent about this

question for atomic predicates, but it is sometimes informative

about complex predicates. Coupled with our definition of identity,

the framework gives us a necessary condition for a formula to

express a property. Let <!>x be a formula with one free variable x.

Then a necessary condition for <!>x to express a property is that sat-

isfaction and/or dissatisfaction of it by objects make a definite dif-

ference in their identity. Following Peter WoodrufC I call this

condition "DDiff" for "definite difference". The condition says that

there are no objects x and y such that x definitely satisfies <!> and y

definitely dissatisfies <!> and yet it is indeterminate whether x is y.

In notation:

DDiff: ....,::3x::3y[!<!>x & !....,<jJy & vx = y].

Equivalently:lO

9 Woodruff and Parsons (1997). Woodruff is responsible for the name 'definite

difference'. Noonan (1990) and (1991) calls the principle that DDiff holds of every

formula the "Principle of the Diversity of the Definitely Dissimilar".

10 The clauses in the conditional form of DDiff contain determinacy and inde-

terminacy connectives that force both the antecedent and the consequent to be

truth-valued. So all conditionals that agree with the classical material conditional

for parts that have truth-value are equivalent in this context, and the choice of one

of these conditionals over another is not crucial here. For consistency, I have used

my favourite conditional connective, which is used throughout the text. (See §6.1

for discussion of the alternatives.)

Evans Argument, Properties, and DDiff 53

DDiff: '<ix'<iY[!<!Ix & !-,<!Iy => --on = y].

If <!Ix does not satisfy DDiff, then it cannot express a property. For

if it did express a property, that property would be definitely pos-

sessed by an object x and definitely lacked by an object y, and thus,

by our definition of identity, x would definitely not be identical with

y, contrary to the clause in DDiff of the form ''i7X = y'. So this is not

an independent thesis; it is just a logical consequence of the worldly

Leibnizian definition of identity that we are using."

In certain circumstances, we can prove that certain formulas do

not express properties, because we can prove that they do not

satisfy DDiff. The formula ''i7(a = x)' discussed above is an example

of this (if there is at least one object that is indeterminately iden-

tical to a). The converse is sometimes possible as well. Suppose that

a formula is constructed entirely from primitive predicates that

11 This can be seen as follows. Using property quantifiers, our definition of iden-

tity is:

a = b =dr VP(Pa <=> Pb).

With this definition. DDiff takes this form:

DDiff: <=> Py) & 'Q>x &

Now suppose that Q> does express a property, Q. TIlen the instance of DDiff for Q> is

equivalent to this form:

DDiff: !Qx&

where it is understood that 'Q' stands for a property, and thus is the sort of thing

that the quantifier 'V P' may be instantiated for. We need to show that this form of

DDiff must be true. For that, it is sufficient to show that a formula of this form

cannot be true:

vVP(Px <=> Py)& !Qx & hQy.

So suppose it is true. Then an inference based on the last two conjuncts yields:

vVP(Px<=> Py)& 3P(!Px&

But the last conjunct entails

...,VP(!Px <=>! Py).

Since the smallest ingredients of this formula are preceded by'!'. the whole formula

is bivalent in virtue of its form, and thus it entails

!-NP(! Px <=> !Py).

But this is inconsistent with 'vVP(Px<=>Py), when 'v' is expanded by its definition;

that is, it is inconsistent with:

...,!VP(Px <=> Py) & ....,!....,VP(Px <=> Py).

54 Evans Argument, Properties, and DDiff

stand for properties and relations, and that the logical symbols in

the formula include only the identity sign, the connectives &, v, -',

and the quantifiers ::3 and V. Then if <jlx is such a formula with one

free variable, it satisfies DDiff. Only if we include a connective such

as '!' or 'v' or ' ~ ' do we get formulas that do not satisfy DDiff12

(or if we start out with primitive predicates that do not stand for

properties ).

I will assume hereafter that any formula of one free variable that

satisfies DDiff expresses a property, and that abstracts constructed

from such formulas stand for those properties. This commits one to

a certain principle of plenitude for properties. If a formula satisfies

Ddiff, then we know it can express a property. But does it? That

depends on what we assume about properties. We can assume that

there are hardly any properties at all, e.g. that logical combinations

of predicates expressing properties typically do not themselves

express properties. On this assumption if 'Fx' expresses Fness and

'Cx' expressed Cness then there may be no property of being F

and C for' Fx & Cx' to express. Or we can assume instead that

there are lots of properties, and that logically complex formulas

(made without '!' or 'v' or ' ~ ' ) whose parts express properties typ-

ically also express properties. I find it convenient to use "property"

talk, and so I will indeed make the maximal assumption that any

formula expresses a property if it is constructed entirely from

atomic predicates that stand for properties, and from identity, and

the quantifiers, and the connectives '&', 'v', and '-,'. For the philo-

sophical issues discussed below, it won't matter whether we adopt

this principle or something weaker. This is because the important

point always turns out to be the converse: any formula that does

not satisfy DDiff cannot express a property.

4.4 ABSTRACTS AND PROPERTIES

In the discussion above it became clear that some abstracts, such

as the one employed in the Evans argument, cannot stand for one

of the properties in terms of which we define identity. We did not

show that the abstracts cannot stand for some other kind of prop-

12 See Keefe (1995) for a discussion of why we should not expect prefixing a

predicate with ',,' to preserve propertyhood.

Evans Argument, Properties. and DDiff 55

erty. In the account of indeterminacy that began in Chapter 2 we

talked about properties and relations "in the world"; this is an onto-

logical notion of property. People sometimes talk about properties

in another way, using 'concept' and 'property' interchangeably,

sometimes even construing properties as the meanings of predi-

cates. Suppose there are two sorts of things that are commonly

called "properties": real things in the world, on the one hand, and

parts of our conceptual apparatus for representing the world,

on the other. I am not certain this distinction can be clearly made,

but I think these two different ways of viewing things are quite

common. If the distinction can be made, then it is clear that the

theory I am discussing sees real identity in the world as arising with

the worldly properties, not the conceptual ones. When people feel

that abstracts must stand for properties, they may be thinking of

the other sort of properties, those that are part of our conceptual

apparatus. We can happily admit that Evans's abstract 'A.x[v(a =

x)]' expresses a conceptual property. But there is no reason that I

know of for assuming that conceptual properties validate the con-

trapositive of Leibniz's Law, which involves the worldly sort of

properties.

It is essential for my purposes that we distinguish abstracts that

can stand for worldly properties from those that cannot. It is much

less important what we decide to do with the ones that cannot stand

for mundane properties. One mayor may not want to say that such

an abstract stands for a conceptual property. (Conceptual proper-

ties are investigated a bit more in Chapter 11.) When T say without

qualification that a predicate does not stand for a property, it will

be the worldly sort of property that T have in mind.

5

Non-Conditional Disputations

In this chapter I discuss a number of attempts to prove that there

can be no such thing as indeterminate identity. The arguments

discussed here are limited to ones that do not turn on the logic

or semantics of conditionals. These are straightforward arguments,

mostly to the effect that positing indeterminate identity is incon-

sistent with some known truth. Other sorts of arguments are

confronted later, including arguments based on the logic of con-

ditionals, arguments asserting that indeterminate identity makes

it impossible to refer to objects, arguments to the effect that inde-

terminate identity cannot be understood, and arguments to the

effect that there are better accounts of the data.

It does not make sense to survey the whole literature on this

topic. I include a representative sampling of the sorts of arguments

that people commonly find persuasive. I begin with the three con-

siderations that, aside from Evans's proof, have been mentioned to

me most often by others.

5.1 QUINE: NO ENTITY WITHOUT IDENTITY

W. V. Quine has made famous the slogan "No entity without iden-

tity", and some people construe this as ruling out the possibility

that there might be objects between which a claim of identity has

no answer. Does Quine's slogan conflict with the view under

consideration here?

I don't find Quine's slogan, as intended by him, incompatible

with the possibility of indeterminate identity. The slogan is fuelled

by a discussion of merely possible entities in "On What There Is",

in which he expostulates:

Non-Conditional Disputations 57

Or, finally, is the concept of identity simply inapplicable to unactualized

possibles? But what sense can be found in talking of entities which cannot

meaningfully be said to be identical with themselves and distinct from one

another? Quine (1961: 4)

What Quine rejects here is a kind of entity to which the concept of

identity is inapplicable. An endorsement of indeterminate identity

does not deny that the concept of identity is applicable to objects,

it merely claims that in certain circumstances when you apply the

concept you get no answer. Not because the objects are the wrong

sort of things to which to apply identity, but just because that is the

way things go (or fail to go). Furthermore, all objects are identical

with themselves, and al\ are distinct from one another if 'another'

is read strongly to mean "definitely not the same". Further, it is

meaningful to ask of any objects a and b whether a is b, and to

expect a correct response from the alternatives yes, no, or neither

yes nor 110. Quine's slogan does not rule out indeterminate identity

because it does not address the issue.

Suppose we reinterpret the slogan so as to require a determinate

yes or no answer to an identity question whenever posed? Since

Quine did not propose the slogan under that interpretation, it is no

surprise that he gives no arguments for it under that interpretation.

The slogan is now simply a statement of what is at issue, and its

fame alone has no efficacy.

5.2 METHODOLOGICAL BTVALENCE

As mentioned in Chapter 1, Quine sees the question of bivalence

as a methodological choice. There is no fact of the matter about

whether our utterances are bivalent or not; there is only a decision

to be made about how to theorize. In any given science, it is pos-

sible to maintain a methodology committed to bivalence; it is

possible because it is always possible to interpret recalcitrant

data in such a way as to preserve our methodological choice. It is

also possible to abandon bivalence if that makes things go more

smoothly.

It should be clear that if it is a methodological choice that we

face, then this fact by itself does not refute indeterminacy. For a

refutation, we need to see why the choice should be made against

58 Non-Conditional Disputations

indeterminacy. I think that one factor that leads people to suppose

that one should choose bivalence is fears about what an alterna-

tive will bring. The first fear is that centuries of accomplishments

that are formulated under the assumption of bivalence will need to

be jettisoned. But this is not so. If you have a particular bit of

subject matter that is nicely accounted for by a theory based on

bivalence, then it needn't be abandoned at all. Giving up bivalence

as a general principle does not mean abandoning it in every case.

All that you need to do for your bivalent subtheory is to explicitly

assume bivalence of application for its primitive predicates. Then

no classical consequences are lost. You lose classical consequences

only when you do not assume bivalence for the claims with which

you are working. An example of both points is given in Chapter 11

where classical Zermelo-Fraenkel set theory is found as a bivalent

subtheory of a general non-bivalent theory of sets.

A second fear is that abandoning bivalence will leave you in a

position of not knowing what follows from what. When Quine

(1981: 94-5) rejects the "fuzzy and plurivalent alternatives" to biva-

lence, he may not have had in mind the clear and well-behaved

logic sketched in Chapter 2. It is true that if you simply abandon

bivalence without settling on any particular alternative at all, then

you open a Pandora's box of alternatives, including options that

hardly anybody understands, and that is not a wise methodology

for anybody. But that is not what is suggested in this book. The non-

bivalent framework used here is really quite conservative and well

behaved. It is especially important in this connection to recall that

reductio ad absurdum is retained in this framework as a refutation

technique. You can show that a claim is not true by deriving a con-

tradiction from it. You merely cannot use this to establish that its

negation is true, unless you have reasons to believe that the claim

is a bivalent one.

A mark of science is supposed to be that hypotheses are con-

firmed by their instances. I see no reason why this would be dif-

ferent in a non-bivalent framework.! Another mark is that a

hypothesis is a serious scientific option only if it is falsifiable (can

I It may be important in this connection to explore how probability theory might

be affected by non-bivalence. This can be done by replacing negation by exclusion

negation in tbe basic axioms (i.e. replace ' ~ S ' by ' ~ ! S ' ) . and expanding the avail-

able tautological equivalences by the principles of logic from Ch. 2. I have not

explored the philosophical consequences of doing this.

Non-Conditional Disputations 59

be shown to be false) within one's current methodology. In a non-

bivalent framework this must be liberalized to: a hypothesis is a

serious scientific option only if it is refutable (can be shown not to

be true) within one's current methodology.

It is possible to imagine that the world contains indeterminacies,

but why would anyone choose to do science on that basis? The

answer may lie in theoretical simplicity. It is conceivable that you

will find yourself confronted with data that can be accommodated

neatly with a non-bivalent theory, and not so neatly with a bivalent

one. This seems to have been the actual view of a substantial

number of people, both scientists and philosophers, working in

quantum mechanics. One reads over and over in this literature

that there is no truth to the matter of whether a certain particle

has simultaneously both position and momentum. I don't claim

that this view is correct; the foundations of quantum mechanics

are unsettled at this point in time. The point is rather that many

thoughtful scientists and philosophers have in actual fact been

willing to abandon bivalence for the sake of having the simplest

theory. Whether they were right to do so is beside the point. It is

their thoughtful willingness to do so that illustrates that abandon-

ing bivalence is a genuine and reasonable option in doing science.

Even if bivalence were chosen within the respective sciences by

individuals and communities of scientists, this cannot be decisive

about claims that occur between and across the individual sciences.

In fact, some of the most troublesome cases in the history of iden-

tity puzzles come from this locale.

2

Suppose a person undergoes a

kind of disruption, e.g. a partial brain transplant, that is easily

explainable in physical terms. There is no puzzle here about the

identity of physical systems with which physics deals, nor of the

psychological human development and interaction with which

psychology deals; the question of whether the same person emerges

is not a question in either physics or psychology. Bivalence within

these sciences leaves this puzzle unaddressed.

Perhaps the personal disruption is a question in sociology? Soci-

ology, based on normal occurrences, does not deal with such dis-

ruptions. And if such disruptions became commonplace, the science

would probably pose a new category in terms of which to study it-

speaking perhaps of disruptees. The indeterminacy would be thus

2 See Parsons (2000). Compare also Zemach (1991), who thinks of identity

puzzles as arising between ideologies, with bivalcnce holding within each ideology,

60 Non-Conditional Disputations

avoided, for such is the commitment of each science on its own (or

so we are supposing). But to avoid it is not to solve the identity

puzzle. Perhaps science works better by formulating bivalent theo-

ries using newly invented concepts that are carefully crafted to fit

into bivalent scientific laws. This does not refute indeterminacy, it

ignores it. We already knew that we could refocus our search for

scientific laws by employing new categories. The philosophical task

is to address the puzzles raised by the original ones. The tendency

of scientists to search for new concepts leaves unsettled the issues

posed in terms of other concepts. It is certainly not obvious how

old problems disappear-if they do so-when new categories are

introduced.

Other puzzles similarly arise in the interfaces; the person-and-

their-body at the interface of biology and some science of persons,

the cat and the p-cats at the interface of biology and physics, and

so on. Of course, one might hope that the sciences will eventually

grow together, so as to accomplish the much discussed reduction

of everything to physics, or something of this sort. But that is not

accomplished at present, and we don't know what form it will take

if it does occur, or whether and how identity puzzles will be

affected, if at all. In the meantime, a methodological commitment

to the search for bivalence in each science is not sufficient to close

the door on an exploration of other options for claims between the

sciences.

5.3 SALMON'S ARGUMENT BASED

ON ORDERED PAIRS

Nathan Salmon attempts to refute the possibility of indeterminate

identity, arguing as follows:

3

[S]uppose that there is a pair of entities x and y ... such that it is vague

(neither true nor false, indeterminate, there is no objective fact of the

matter) whether they are one and the very same thing. Then this pair <x,y>

is quite definitely not the same pair as <X,X>, since it is determinately true

that x is onc and the very same thing as itself. It follows that x and y must

, Some writers see Salmon's argument as identical in structure to Evans's (e.g.

Burgess (1990: n. 7). If it is, I already discussed it in the last chapter. Others see it

as a distinct argument, which is why I discuss it independently here.

Non-Conditional Disputations 61

be distinct. But then it is not vague whether they are identical or distinct.

Salmon (1981: 243)

The argument appears to have this structure:

4

(1) 'l(x=y)

(2) -''l(x = x)

(3) -,( <x,y> = <x,x»

(4) -,(x = y)

(5) -''l(x = y)

Hypothesis to be disproved

Truism

From (1) and (2)

From (3), by the theory of ordered

pmrs

From (4)

I agree with every step except for the inference from (1) and (2)

to (3). This is fallacious; if it is indeterminate whether x = y then it

is indeterminate whether the pair <x,y> is identical with the pair

<x,x>. No principle of bivalent logic or of bivalent set theory (or

ordered-pair theory) should be taken to validate the inference to

(3), since the inference crucially involves non-bivalency. Salmon

cites no reason for (3), so we can only speCUlate about how to

valida te it.

A natural way to think of filling in the proof would be to use

indirect proof, assuming the principles of ordered-pair theory: if the

pairs in (3) were identical, that would contradict (l) and (2). The

additional steps between (2) and (3) would then be:

(2a) Suppose <x,y> = <x,X>

(2b) Then x = y

(2c) -,'l(x=y)

(2d) 'l(x = y) & -''l(x = y)

(3) -,( <x,y> = <x,x»

Hypothesis for indirect

proof

From (2a) by the theory of

ordered pairs: pairs are

identical iff their first and

second members are.

From (2b) by preservation

of truth.

From (1) and (2c)

From (2a)-(2d) by indirect

proof.

Every step of this subproof is fine, but its conclusion is not. Indi-

rect proof is invalid in a logic that permits sentences that lack

< Following Salmon's words precisely would require us to place a determinately

sign '!' in front of lines (3) and (4). This complicates the notation without having

any particular effect on the validity of the steps.

62 Non-Conditional Disputations

truth-value, and so disproving (2a) is not sufficient to validate its

negation, it is only sufficient to conclude that (2a) is not true.

Instead of (3) you can conclude:

(3') -,!(<X,y> == <x,x» From (2a)-(2d) by indirect proof.

But this is not sufficient to complete the original proof.

There are a number of other ways to try to fill out the above rea-

soning, but J know of no alternatives that turn it into a conclusive

proof. This type of argument does however raise the question of

whether, and to what extent, ordinary principles of set theory may

be applied in the presence of indeterminate identity. If we had to

abandon set theory in the face of indeterminate identity, this would

be very serious, and one might then argue against indeterminate

identity on grounds of utility. In Chapter 11, I argue that set theory

meshes nicely with indeterminate identity.

5.4 SUPERVALUATIONS AND

SUPER-RESOLUTIONS

A popular way to deal with sentences that lack truth-value is to

invoke supervaluational readings. I do not appeal to supervalua-

tions in formulating the theory of this book. But I can't avoid

talking about them. This is for two reasons. First, I will discuss alter-

native accounts, and some of them are based on superva]uational

readings of sentences. So we need to know what supervaluational

readings are. Second, the theory I develop here without superval-

uations will sometimes contradict data-like sentences that we are

inclined to accept. Such contradiction fuels refutations of the

theory that are discussed in the remainder of this chapter. J will

suggest that we are inclined to accept some sentences in a way that

apparently contradicts the theory of indeterminate identity

because we naturally read them non-literally. We either tend to

read them supervaluationally or superresolutionally, and when they

are read in this way they do not contradict the theory.

5.4.1 Ways of reading sentences

In doing semantics one often encounters sentences that elude a

simple classification in terms of truth-value. For example, it is

Non-Conditional Disputations 63

apparent that when asked to judge the sentence 'The king of France

is not bald', some people see the sentence as obviously false, since

there is no king of France to be bald, others see it as true, since it

contradicts a false sentence, others (e.g. Frege) see it as lacking

truth-value, since there is no king of France to be bald or not, and

still others (e.g. Russell) see it as ambiguous between true and false.

Since these are reactions of native speakers who are in possession

of all the facts, and since the sentence is so simple, many pcople,

myself included, see the sentence as one whose surface syntax does

not single out a unique meaning. One gets different meanings

depending on how one interprets, or "reads", the sentence. Seman-

tic theory is then challenged to explain how these many different

readings may arise from such a simple sentence. In inspecting the

phenomenon of indeterminacy, we will have to confront examples

in which speakers naturally interpret sentences differently, and

these different "readings" will need to be accounted for.

I take for granted that the reader is familiar with scope ambigu-

ities, such as in a sentence like 'An observer saw every ship', which

can either mean that some one observer was such that (s)he saw

every ship, or that every ship was seen by some observer, though

not necessarily by the same observer. In this text we will have

occasion to focus on additional sources of multiple readings of

sentences.

5.4.2 Supervaluatiol1s

Supervaluations provide a way of reading sentences, a way that

sometimes gives truth-values to sentences that would otherwise

lack them. The supervaluation technique is typically used for one

of two purposes:

5

(1) One application is to ambiguous sentences. Call an

ambiguous sentence "supertrue" if it is true on all ways of

resolving the ambiguity and "superfalse" if it is false on all

ways of resolving the ambiguity. If the sentence comes out true

on some resolutions and false on others, then it is "supertruth-

valueless". This successfully mirrors ways that we often assess

5 Supervaluations were originally proposed in van Fraassen (1966). Fine applies

the technique to (semantic) vagueness. Fine (1975: 284) emphasizes the application

to ambiguity: "An ambiguous sentence is true if each of its disambiguations is true."

64 Non-Conditional Disputations

real-life utterances that are subject to different interpreta-

tions; we call such an utterance true without bothering to

resolve the ambiguity if it is true no matter how the ambigu-

ity is resolved.

(2) A second application of supervaluations is to sentences

which lack truth-value for reasons other than ambiguity. To

evaluate a sentence supervaluationally, you focus on the basic

units of the sentence that are responsible for its lack of truth-

value, and you consider how to flesh out these units consistent

with whatever determinacy is already there. For example, the

lack of truth-value may be because a predicate does not apply

either truly or falsely to certain objects, so you consider

how to extend its extension (the objects it is true of) and its

anti-extension (the objects it is false of) so as to cover

these additional objects. Or else the lack of truth-value is due

to the fact that some singular term lacks a referent, and

you consider various ways of arbitrarily assigning it a referent.

Call any way of extending the original semantics so as to get

a truth-value for every sentence an "extended valuation".

Then a sentence is supertrue if it is true on all extended valua-

tions, superfalse if it is false on all extended valuations, and

otherwise supertruthvalueless. The advantage of reading sen-

tences supervaluationally in this way is that sentences that are

not logically true because of the possibility of truth-value gaps

may be logically supertrue because of the way the gaps get

filled in. For example, classical tautologies are often not

logically true when gaps are allowed, but they are logically

supertrue, and this restores a measure of neatness to one's

logical system."

In theory formulation, I avoid supervaluational readings for the

same reason others invoke them: they make the assessment of a

sentence less directly a matter of how its parts relate to the world

than do literal readings. For example, 'Fa v -,Fa' is supertrue

n There are limits to the neatness of supervaluationallogic. For example, as with

my own approach, contra position of validity does not generally hold in super-

valuational logic. Note that the general theory of supervaluations is more com-

plicated than my survey of it indicates; complications abound when one has notation

in the object language for both 'true' and for 'supertrue'.l avoid this; my'!' converts

a true sentence into a true one and all others into a false one. It does not necessar-

ily convert a Sllpertrue sentence into a true one.

Non-Conditional Disputations 65

because of its logical form, quite apart from whether 'Fa' is true,

false, or neither. The connection between the truth-value status of

a sentence and how its parts relate to the world is typically simpler

to calculate than the connection between its supervaluational

status and how its parts relate to the world. Since I am primarily

concerned with depicting the world, the more direct semantics is

preferable. I also think that certain key claims involving indeter-

minacy are more easily formulated presuming a literal reading than

a supervaluational one, when supervaluational readings let logical

form wipe out differences of content. Simply put, I am interested

in assessing the causes of truth-value gaps, not in elegant ways of

avoiding such gaps by interpreting them away.

Although I avoid using supervaluations in formulating my own

theoretical views, I will certainly appeal to them on occasion to

explain why we tend to judge certain sentences true when my

theory says they lack truth-value. For example, if'S' lacks truth-

value, then on my view so does'S or not-S', yet we are inclined to

think that'S or not-S' is in some sense correct. This is exactly the

sort of data that supervaluations were introduced to account for;

we do tend to read complex sentences supervaluationally some-

times, and I have no reservations about agreeing with this. What

must be kept in mind is that a supervaluational reading is always

parasitic on another, which I call the "literal" reading. The theory

produces the literal reading, and that is the one I typically focus on.

You can always read the results supervaluationally. But you should

not reject a claim that a sentence, read literally, lacks truth-value,

just because the same sentence, read supervaluationally, has a

truth-value.

5.4.3 Super-resolutions

In discussing the worldly theory of indeterminacy, we will need to

consider an additional method of reading sentences. Supervalua-

tions are a nominalistic creation; one typically deals with pieces of

language and how they relate to the world, without any discussion

of properties and without assuming that the world itself is indeter-

minate. So when we try to combine the traditional method of super-

valuations with the theory under discussion here there are a couple

of oddities that need to be addressed. First, suppose that a predi-

cate 'P' stands for a property. Making 'P' more precise by an

66 Non-Conditional Disputations

extended valuation may result in its no longer standing for a prop-

erty. Since whether 'P' stands for a property or not may affect how

it interacts with identity, we need to worry about how this change

may affect the readings of sentences that contain it and that also

contain identities of objects that have the original property. (An

example will be given below.) Second, it is generally assumed in the

case of supervaluations that all of the "input" indeterminacy is

semantic, to be resolved by clarifying the referents of names and

the extensions of predicates. But in our framework, we may have

an identity statement without truth-value even though the refer-

ents of the names are clear (as discussed in Chapter 9). How should

a supervaluation treat such a statement? If identity can have its

extension "clarified" at will, then it loses its special relation to the

having and lacking of properties, but if its extension cannot be

altered, supervaluations will not produce supertruth when they are

expected to.

Here is an example that combines these two concerns. Suppose

that 'P' stands for a property, and consider the sentence:

Pa & -,pb & a = b.

This sentence might lack truth-value when read literally. (Suppose

that 'P' stands for the property p, which holds of Cl and neither holds

nor fails to hold of b, and that it is indeterminate whether a is b.

Then 'Pa' is true, and 'Pb' lacks truth-value, and so does 'a = b'.) It

is also the sort of sentence that could not be true because of its

form, and thus it is the sort of sentence that supervaluations are

supposed to address; we are supposed to be able to address the

apparent logical falsehood of the sentence by showing that

although it lacks truth-value, it is superfalse, that is, it is false on

every extended valuation that gives it a truth-value. But the method

of supervaluations alone does not give this result without supple-

mentation. If no constraints are placed on the extended valuations,

then there are extended valuations that make the sentence true,

and ones that make it false, resulting in a sentence which is

supertruthvalueless, just as the original is.

(To make the sentence false, just extend the extension of 'P' to

include b, and the second conjunct will then be false. To make it

non-false, just extend the anti-extension of 'P' to include b, which

makes the second conjunct true. That leaves the third conjunct, the

identity, to consider. If' P' still stands for a property, then the iden-

Non-Conditional Disputations 67

tity must be false, since the first two conjuncts force a to have that

property and b to lack it. But there is nothing in the method of

supervaluations to require that' P' still stands for a property; in

extending it we may have converted it to one of those predicates

that do not stand for properties, in which case the identity clause

is not yet determined. So how do we treat the identity? As men-

tioned above, we might either leave it alone, or we might adjust its

extension at will. If we leave it alone, the third conjuncl stays inde-

terminate, and we have produced an extended valuation that makes

the whole sentence lack truth-value. If we adjust its extension at

will, we can let it be true of a and b, and thus the whole sentence

comes out true. In either case, the extended valuation makes the

sentence non-false, and thus it is not superfalse as desired.)

What is desired is a technique similar to the bare supervaluation

approach that respects the details of the theory under discussion.

Here is one such. Suppose that the world is not completely deter-

minate. Now consider various ways in which it might become deter-

minate by determining undetermined states of affairs. That is, by

taking a property that neither determinately applies nor fails to

apply to any given object and making it determinately apply or

determinately not apply to that object. Call the result of such a way

of making world IV completely determinate a resolution of w. If ,.

is such a resolution, we can ask about the truth-values of sentences

that result from altering the semantics of predicates that stand for

properties so as to correspond to the new ways that those proper-

ties behave under r, and from readjusting the extension of the iden-

tity predicate as required by the Leibnizian definition of identity in

terms of properties and relations. Then in any resolution, identity

will be completely determinate for any pair of objects. If a sentence

contains only primitive predicates that stand for properties, and

identity, we can ask whether it is true in a resolution,. of IV. This is

to ask about truth resulting from resolution of the world without

alteration of the language that depicts the world. And we can ask

further which sentences are true in every such resolution. Call this

"super-resolved truth". Super-resolved truth is like the supertruth

due to supervaluations, except that it is truth that results from con-

sidering making the world more determinate, instead of truth that

results from making our language more determinate. In a large

number of cases, super-resolved truth will coincide with unresricted

superva]uational truth, even though they are arrived at quite

68 Non-Conditional Disputations

differently. But not always. The example discussed above is super-

resolvedly false. This is· because super-resolutions respect the

ontology. We alter the extension and anti-extension of' P' indirectly

by extending what the property that it stands for is true or false of.

If we make the world more determinate by letting the property p

above be false of b, then there is a property that holds of a and not

of b, and the identity is forced to be false.

7

The reason for keeping super-resolutions in mind is the same as

for unrestricted supervaluations: sometimes we make judgements

that certain sentences are true, or false, when the theory says that

they lack truth-value. In some cases (according to many philo-

sophicallogicians, and I agree), this is because we are reading the

sentences supervaluationally. And in some other cases in which

unrestricted supervaluations yield no (super)truth-value, I think we

are reading the sentences in a super-resolved way. Certainly I need

some explanation of why people uncorrupted by philosophical

theory are tugged in the direction of saying that for any two things,

either they're the same thing or they're not:

\ix\iy(x = Y v -.x = y).

I myself feel this pull, and when I ask myself why, I find that I am

inclined to say something like:

Well, however things are, they have to be like that.

If the 'are' is read as 'determinately are', then I am asking about

resolved truth, not truth simpliciter. The sentence '\:j x\i y(x = Y v -.x

= y)', read literally, lacks truth-value according to the theory, but it

is true on all resolutions; it is super-resolvedly true. (Its super-

valuational truth status depends on how the theory of supervalua-

tions is extended to handle worldly indeterminacy of identity.)

The remaining three topics in this chapter bring in molecular sen-

tences in one way or another, so let me summarize my position on

these. If both disjuncts of a disjunction lack truth-value, so does the

disjunction-if it is read straightforwardly as I have explained it

7 The super-resolutional account can be interpreted as an applicafion of the

supervaluational account; simply limit the extended valuations of names and pred-

icates to valuations that they could have in some resolwio17. 1 will continue to use

the term 'super-resolution' because it suggests the details of how the valuations are

10 be determined; I call the approach that uses all arbitrarily chosen extended

valuations the "unrestricted supervaluation" approach.

Non-Conditional Disputations 69

above. However it is sometimes natural to read a disjunction super-

valuationally. It is always natural to read it this way when the

statement has the form of an explicit instance of the Law of

Excluded Middle: "A or not A". Read straightforwardly, 'A or

not A' lacks truth-value when 'A' lacks truth-value. Read super-

valuationally, 'A or not A' is true, since any way of assigning a truth-

value to 'A' makes the whole disjunction true. The same holds for

super-resolutional readings. When I make statements myself I

always intend that they be taken straightforwardly, not supervalu-

ationally or super-resolutionally, but when considering statements

made in other contexts we must be alert to other ways of inter-

preting them.

5.5 AN ARGUMENT BY WILLIAMSON

AGAINST NON-BIVALENCE

Timothy Williamson's book Vagueness (Williamson (1994» is

devoted primarily to linguistic or conceptual issues about vague-

ness, but he touches briefly on the possibility of indeterminacy of

identity. Tracing through his reasoning,S the line of argument

depends crucially on his rejection of multivalent logic, and his

defence of bivalence. Although I employ only two truth-values, true

and false, I admit three truth-value statuses: true, false, and neither,

and variants of Williamson's arguments apply to this stance as

well.

WilIiamson's position is a broad one, not directed merely at inde-

terminacy of identity; he believes that there is no worldly indeter-

minacy of any kind. He phrases this as an attack on the idea that

statements might have some degree of truth other than 0 (complete

falsity) or 1 (complete truth). Anyone who believes in some other

degree of truth must give a coherent account of how our language

embodies this. He considers a view substantially like the one dis-

cussed here, in which something like a "truth-functional" language

is used. Specifically, he argues against the claim that: "the degree of

S WilIiamson discusses indeterminate identity in his §9.2 Delenninacy in the

World; he rejects it there (1994: 255) because it requires a formulation in terms of

many-valued logic, which has been rejected in §§4.11-4.14.1 focus here on the main

point of §4.l4.

70 Non-Conditional Disputations

truth of various compounds is a function of the degrees of truth of

their components" (1994: 135).

If we read 'degree of truth' as 'type of truth-value status' then

his discussion is pertinent to the view under consideration here. He

argues as follows:

Now imagine someone drifting off to sleep. The sentences 'He is awake'

and 'He is asleep' are vague. According to the degree theorist, as the

former falls in degree of truth, the latter rises. At some point they have the

same degree of truth, an intermediate one. By what has just been argued,

the conjunction 'He is awake and he is asleep' also has that intermediate

degree of truth. But how can that be? Waking and sleep by definition

exclude each other. 'He is awake and he is asleep' has no chance at all of

being true. Our man is not in an unclear area between the cases in which

the conjunction is true and those in which it is false, for there are no cases

of the former kind .... Since the conjunction in question is clearly incor-

rect, it should not have an intermediate degree of truth. It is clearly incor-

rect, although neither conjunct is; one must be careful to distinguish what

can be said of the conjunction from what can be said of each conjunct.

Thus degree-functionality fails for conjunction.

The same point can be made with 'He is not awake' in place of 'He is

asleep' .... How can an explicit contradiction be true to any degree other

than O? (ibid. 136)

I do not hold that any contradiction can be true to any degree other

than 0 (falsity). I do, however, hold that a contradiction can lack

truth-value if its parts do. Wil1iamson's discussion of 'He is awake

and he is not awake' needs to be addressed because it appears

to rule out lack of truth-value as well as intermediate degrees of

truth-value.

What happens if 'He is awake' lacks truth-value? I hold that the

conjunction 'He is awake and he is not awake' also lacks truth-

value. Williamson is right in questioning whether this is how a

person would normally intend such an utterance. But this is hard

to assess, since it isn't an utterance that one would normally make.

But suppose someone actually asserts 'He is awake and he is not

awake'; what would they be likely to mean? ll1e most natural inter-

pretation would be that they are saying this to emphasize that he

is sort of awake and sort of not awake. Under this interpretation,

both Williamson and J are wrong, for so used, the sentence would

be true. But on that interpretation it is not a contradiction at all;

this is because the 'sort of' modifies the meanings so that the second

Non-Conditional Disputations 71

conjunct does not deny what the first asserts. So under this inter-

pretation the utterance is not relevant to Williamson's query "How

can an explicit contradiction be true?"

This is not what Williamson intended to discuss. So let us fasten

on a reading in which the second conjunct is the negation of the

first, in a situation in which the first conjunct is as midway between

truth and falsity as you can get it. I am then comfortable in saying

that both conjuncts lack truth-value, and thc whole does as well.

Williamson thinks that the whole must be fully false. He says this

apparently because he thinks this must be true of any contradic-

tion: "How can an explicit contradiction be true to any degree other

than O?" The answer to his question is straightforward: "by having

parts that lack truth-value".

This answers Williamson's question, but not his concern, for

there really is something odd about seeing explicit contradictions

as indeterminate. This is probably because an explicit contradiction

cannot he true. This, however, does not make it false, unless it has

a truth-value. But suppose we want somehow to be able to con-

strue any sentence that cannot be true in such a way that it is

thereby false. This is what supervaluational readings are for; a sen-

tence read supervaluationally is automatically false (or a sentence

read normally is automatically superfalse) if its form prevents it

from being true, and any explicit contradiction is like this. It is

appropriate to read a sentence supervaluationally if you are

judging it based more on form than on content, which is what

Williamson urges that we do with explicit contradictions.

Succinctly put: Williamson objects to the lack of bivalence of any

view that admits "truth-value status functionality" because, so read,

some explicit contradictions are not false (they lack truth-value).

But any sentence can be read supervaluationally. So anyone who

has a potentially non-bivalent language has a language that can

read contradictions as false when you focus on form over content.

So the existence of instincts to call explicit contradictions false

cannot refute such a view.

Elsewhere (Chapter 5), Williamson argues that a consistent use

of supervaluational readings yields unnaturalness in some contexts.

I agree with many of the points he makes, though the discussion is

too wide-ranging to sum up here. But I draw a different lesson.

There is a wide variety of contexts in which definite truth and falsity

cannot be assumed, and no uniform approach to non-bivalent

72 Non-Conditional Disputations

language will be most natural in all of them. The best we can do is

to be clear about how our own language is to be taken, and to be

careful not to misconstrue others. Occasional contexts in which

truth-status-functional readings of connectives are unnatural

cannot tell against a theory formulated with such readings, so long

as one does not assert that such readings of connectives are the

ones intended by all speakers of natural language in all contexts.

And I make no such claim.

Williamson has other relevant arguments; I discuss one which

relies on the logic of conditionals in the next chapter, and I discuss

higher-order indeterminacy in Chapter 12.

5.6 FUZZY OBJECTS AND

INDETERMINATE IDENTITY

One widely discussed issue is not strictly a refutation, though it

often plays much the same rhetorical role. It turns on the relation

between the existence of "fuzzy objects" and the existence of

indeterminacy of identity. Several writers suggest that people

believe in indeterminacy of identity because they mistakenly think

that it follows from the existence of fuzzy objects. But it does not;

so a major rationale for believing in indeterminacy of identity

is thus undercut. If the only evidence one could ever have for

indeterminacy of identity is easily reconstruable as evidence for

fuzzy objects, the thesis of indeterminacy of identity could not ever

be validated.

An example: a mountain has indeterminate boundaries. Pick

a rise in the landscape with boundaries such that it is indeter-

minate whether they are the boundaries of the mountain. Does

it follow that it is indeterminate whether the rise is the moun-

tain? Not automatically; certainly not without some additional

argumentation.

A "fuzzy object" is an object with indistinct boundaries. The

indistinctness of boundary is usually construed as an indistinctness

in whether a certain portion of space is inside or outside the object

or not, and this is then recast as the question of whether the stuff

in the disputed region is or is not part of the object. If indistinct-

ness is interpreted as indeterminacy, a fuzzy object then becomes

Non-Conditional Disputations 73

an object such that it is indeterminate what its parts are. For our

purposes, then, a fuzzy object may be taken to be as follows:

x is a fuzzy object =df there is some y such that it is indetermi-

nate whether y is part of x.

As many writers either point out or argue at length,9 the existence

of a fuzzy object in this sense does not logically entail that there

arc objects such that it is indeterminate whether they are identical.

Most people leave it at that. But there is a bit more to be said.

For there are situations in which one can infer the indeterminate

identity from the existence of fuzzy objects.

5.6. J Inferring indeterminate identity from fuzzy objects

Although the existence of fuzzy objects does not entail that there

is any indeterminate identity, it may do so with the help of some

additional assumptions. Here are some assumptions that would

make the transition feasible. The assumptions are ingredients of a

theory of mereology (the study of parts and whales). We use 'part'

here in the sense in which each object is part of itself.

Mereological identity: For any objects x and y, x is identical

with y if and only if x and y have the same parts.

Transitivity: If x is part of y, and y part of z, then x is part

of z.

Mereological sums: For any objects x and y there is an object

z which is their mereological sum, in the sense that

(i) x is part of z, and

(ii) y is part of z, and

(iii) everything that x and y are both parts of, z is part of also.

Uniformity of indeterminate parts: If it is indeterminate

whether y is part of x, there is no part of the sum of x and y

which is determinately not part of x.

These assumptions are to be understood strongly. For example, the

claim of mereological identity is that 'x is identical to y' is to be true,

false, or neither, exactly when the claim 'x and y have the same

parts' is true,false, or neither. In mereological sums it is understood

that for any objects x and y it is determinately true that there is an

9 e.g. Burgess (1990).

74 Non-Conditional Disputations

object z which is the sum of x and y, where the explanation of

"sum" is definitive as just explained. I clarify these assumptions in

this way not because I believe that they are true when so clarified,

but rather because I am giving an example of assumptions that

would connect indeterminacy of parts with indeterminacy of iden-

tity. These assumptions are not true of objects in general, because

there are many sorts of objects that either have no parts in a

straightforward sense, or they have parts in a sense that does not

allow us to infer their identity from the fact that they share parts.

But in some limited domains, the assumptions may be quite plau-

sible. In such a domain one could infer indeterminate identity from

the existence of fuzzy objects. Consider an object a and some object

b where it is indeterminate whether b is a part of a. We could argue

as follows:

By mereological sums there is an object c which is the sum of

a and b. We will show that it is indeterminate whether c is iden-

tical to a. By mereological identity it will be sufficient to show

two things:

(1) There is something that is determinately a part of c and

not determinately a part of a. This rules out the possibility

that a and care determinately identical.

(2) There is nothing that is determinately part of a that is

determinately not part of c, and vice versa. This rules out the

possibility that a and care determinately distinct.

Arguments:

(J) By hypothesis it is indeterminate whether b is part of a,

and by construction of c as the sum of a and b, it is deter-

minate that b is part of c. So a is not determinately the same

as c.

(2) By construction of c as the sum of a and b and by tran-

sitivity it follows that every determinate part of a is a deter-

minate part of c. And by the uniformity of indeterminate

parts, there is no determinate subpart of the sum of a and b

(that is, no determinate subpart of c) that is determinately

not part of a.

So there are assumptions with which one can argue from the exis-

tence of fuzzy objects to the existence of indeterminacy of identity,

and they may be quite plausible when applied to certain sorts of

Non-Conditional Disputations 75

objects. But in other cases the assumptions may not be plausible at

all. For example, I am a person. It is up for grabs whether 1 am a

mereological object or not. I do indeed have a spatial location, but

it is unclear whether my identity is dependent on this, so the prin-

ciple of mereological identity may be false when applied to me. If

so, I may be a fuzzy object, but there may be nothing with which I

am indeterminately identical.

Are mountains mereological entities of which these assumptions

hold? I suspect not, though I am not certain. If they were, they

would yield hosts of examples of indeterminate identity.

5.7 COOK'S BUILDING

The following example involves considerations of parts and wholes.

Cook (1986) takes the position that the Evans argument is flawed,

but that there is nonetheless no such thing as indeterminate iden-

tity. To make this plausible he argues generally against indetermi-

nate identity, and he also explains how at least one apparent case

of indeterminate identity is instead merely a case of a fuzzy object.

Cook considers a building, somewhat different from the one I

discussed in Chapter 1. He assumes a structure with a kind of

dumbbell shape, consisting of two modules, A and B, connected by

a narrow walkway, C. Smith is lecturing somewhere within module

A, and lones within module B. The question will be whether the

building in which Smith is lecturing is the same as the building in

which lones is lecturing.

My own instinct is that there are three determinate buildings: A,

B, and ABC, with each of A and B being parts of ABC. If this is so,

there are two distinct buildings (B and ABC) in which lones is lec-

turing, and thus 'the building in which Jones is lecturing' is a defi-

nite description that has no unique reference, and the example

raises no genuine puzzle about identity. But Cook assumes other-

wise, and his view leads to an interesting case.

Let 'J' abbreviate 'the building in which lones is lecturing', and

'S' abbreviate 'the building in which Smith is lecturing'. Cook

assumes that 'J' has a unique reference, and so does'S', but it is

indeterminate exactly what the parts of J are, and likewise for S:

76 Non-Conditional Disputations

"The denotation is not indeterminate; the boundaries of what is

denoted are indeterminate" (1986: 182).

He rejects the idea that clarifying what counts as a building is

even relevant to the question of denotation or identity:

When we sharpen the notion of a building, we are not making the

denotation of 'the building in which Smith is lecturing' determinate;

we are making it determinate whether this building includes [B]. (If

the colour of my car is on the borderline between red and orange, then

it is indeterminate whether the colour of my car is red. 'The colour of

my car', however, has a determinate denotation. If we sharpen the notion

of being red, we don't make the denotation of 'the colour of my car'

determinate, we make it determinate whether the colour of my car is red.)

(ibid: 183)

Cook's position is that'S' and 'J' each uniquely denotes an object

with indeterminate parts, but there is no indeterminate identity of

objects.

5.7.1 An argument from fuzzy buildings to indeterminate identity

Let me digress for a moment to consider whether this might be a

case in which indeterminacy of parts actually does entail an inde-

terminate identity. Assume that the only parts of the building that

are relevant in this example are A, B, and C, and sums made of

them. We may also in this case make a crucial assumption that

cannot be made about all cases: that the buildings in question

would be identical if they determinately had exactly the same parts,

and that they would be determinately distinct if one of them deter-

minately had a part that the other determinately lacked. S has A

as a determinate part, since S is determinately the building in which

Smith is lecturing, and it is determinate that Smith is lecturing in

A; similarly J has B as a determinate part. Now we can argue as

follows:

Suppose that J is determinately identical to S. Then J and S

have the same parts. Since A is determinately a part of J, it is

also determinately a part of S, and since B is determinately a

part of S, it is determinately a part of 1. So each of J and Shave

both A and B as determinate parts. If we assume that in this

case there are no scattered buildings, C must be determinately

Non-Conditional 77

part of each as well.

1U

It follows that neither J nor S has any

indeterminate parts at all, contrary to the initial assumption.

So J cannot be determinately identical to S.

Suppose that J and S are determinately distinct. Then we

have determinately distinct objects J and S, with each of them

having a determinate part (B for J and A for S). They must

determinately disagree with respect to some part, for this is

necessary for them to be determinately distinct. By parity, C

must be part of both or part of neither or an indeterminate

part of each, so they will not disagree with respect to C. So it

must be that A is determinately not a part of J and B is deter-

minately not a part of S. Since they are fuzzy objects, they must

have indeterminate parts. As a result, it must be that C is an

indeterminate part of each. This is coherent, but entirely

unmotivated by our understanding of the setup. Why should it

be determinately false that B is part of S, yet indeterminate

whether C is? Indeed, suppose that we move the parts A

and B closer and closer together until C has shrunk to nothing

at all and A and B share only a doorway. Then there is no

remaining part to be an indeterminate part of both J and S,

and the example collapses. Yet it is just as plausible in that case

to say that J and S are fuzzy objects as it is in the original

case.

The mereological assumptions in this argument are certainly not

beyond challenge, but they seem to me plausible. I conclude that

this is a case in which the indeterminacy of parts does plausibly

lead to indeterminacy of identity. Or, it would if the parts were

indeterminate, which is what Cook assumes.

ll

Sometimes there can

be indeterminate part hood without indeterminacy of identity, and

sometimes there cannot be; one has to look at the details of the

case to tell.

10 We could also assume at this point that C is an indeterminate part of one

of the buildings, and argue by symmetry that it must be an indeterminate part of

the other as well. The conclusion would then follow that J and S agree in all their

parts.

" To be clear: I disagree with Cook that this is a case in which there is determi-

nate reference to buildings but in which it is indeterminate what their parts are. I

think that in this particular case one should deny that the descriptions have unique

denotations, and one should not conclude that there is indeterminacy of identity in

the world because of such a case.

78 Non-Conditional Di!>putations

5.7.2 Cook's refutation of indeterminacy of identity

I have argued that if Cook is right about J and S having indeter-

minate parts, they are indeterminately identical. But Cook has an

argument that this is not possible, and we need to look at that. Cook

rejects the Evans argument; he says that being indeterminate iden-

tical with a is a property, but that one cannot conclude from the

fact that b determinately possesses it and a does not that a and b

are distinct. (1 agree with this only on the assumption that he has

in mind what I above called conceptual properties.) But he thinks

a better argument against the indeterminacy of identity can be

given, one involving the law of excluded middle (LEM). Briefly, he

holds that indeterminacy of identity conflicts with LEM, but LEM

can be defended.

Here is the statement of the conflict between indeterminacy of

identity and LEM:

[IJf indeterminacy is in the world then LEM does not hold-objects can

be in three mutually exclusive states: they can be identical, they can be

non-identical, and they can be indeterminately identical. To defend LEM,

then, is to attack the view that objects can be indeterminately identical.

(Cook 1986: 183)

The defence of LEM is explained as follows:

The trick to seeing that borderline cases do not force us to give up LEM

is (a) seeing thal a disjunction can be true even though neither disjunct is

true, and (b) seeing that this does not prevent the disjunction's being

exhaustive. (ibid. 184)

Expanding on (a) he says:

[Olne will not be able to explain the truth conditions for the statement

that the car is red in a straightIorward truth-functional way. Here I think

one should appeal to something like true-uIlder-an-acceptable-sharpening.

Thus, one might say that 'red or orange' is true of an object if under all

acceptable sharpenings of 'red' and 'orange' either 'red' or 'orange' is true

of it. (ibid.)

He goes on to explain how 'J == S' can be seen as a similar case,

because the identity contains a concealed appeal to 'part of the

same building', which is vague in a way similar to the way in which

'red' is vague.

The immediate problem with this explanation is that it is no

longer clear why indeterminate identity is supposed to conflict with

Non-Conditional Disputations 79

LEM. The conflict was asserted above, but that explanation makes

it clear that the conflict arises when the disjunction 'a == b or not

a = b' is treated truth-functionally. However, the defence of LEM

rests on the claim that a disjunction may be true even if neither dis-

junct is true; this is to treat the disjunction non-truth-functionally,

and this undercuts the assertion that indeterminate identity

inevitably conflicts with LEM. Thus the argument against indeter-

minate identity vanishes.

The explanation, I think, is that Cook does not think that real

indeterminacy of identity admits of the same treatment as vague-

ness, and so when LEM is applied to a case of indeterminate iden-

tity, the truth-functional reading is the only one available. And he

is consistent in his treatment of the building case, for Cook sees

'J == S' as lacking in truth-value, but not because of indeterminacy

of identity; he thinks that this is only an apparent case of indeter-

minacy of identity. It is an identity statement with a concealed

vagueness in it, but a vagueness in 'part of', not a vagueness in '='.

Thus he may consistently hold that a disjunction may be true when

neither disjunct is true, if the lack of truth-value depends on vague-

ness, while also holding that if there were real indeterminacy of

identity, a disjunction made up of parts that lack truth-value for this

reason would still lack truth-value.

This puts us on common ground, at least for the language I have

adopted here: if a is indeterminately identical with b then 'a = b

or not a == b' really does lack truth-value, and "sharpenings" of

concepts or vague terms is irrelevant to this. I then accept this as

a counter-example to LEM, a law which Cook defends. But on

this understanding it appears to me that Cook has not defended

LEM for the case in point. He explains how apparent counter-

examples to LEM due to vague terms are not real counter-

examples, but this explanation does not apply to the case of

indeterminate identity. So violations of LEM due to indeterminacy

of identity have not been addressed, and indeterminacy of identity

remains unrefuted.

5.8 BROOME'S CLUB

In a paper objecting to Evans's argument, John Broome (1984)

proposes a plausible example of a case of indeterminate identity.

80 Non-Conditional

SCENARIO: A club comes into existence and continues for five years, at

which point it ceases functioning. A few years later a group of people,

including many members of the club previously mentioned, get together

and act as a club for an additional twenty-five years. The reader is invited

to fill in the details in such a way that it is indeterminate whether the earlier

club was revived when the new meetings began.

Broome suggests that this is a case in which there is an earlier club,

and a later club, with no fact of the matter as to whether they are

the same. (If they were the same, then it would be determinately

true that the earlier club was revived, and if they were not the same

it would be determinately false that the earlier club was revived.)

I am inclined to agree.

Noonan (1984) argues that this cannot be a case of indetermi-

nacy of identity. He asks that we consider the predicates:

lasted for at most five years

lasted for at least twenty-five years

lasted for at most five years or lasted for at least twenty-five

years

He claims that it is indeterminate whether the first predicate is true

of the earlier club, and indeterminate whether the second predi-

cate is true of the earlier club, but determinate that the third pred-

icate is true of the earlier club. He concludes that he cannot

understand how there can be such an object. The argument is

subtle, so I quote it in full:

[I]f'the earlier club is the later club' is indeterminate in truth value for [the]

reason [that the objects are indeterminately identical], then the predicate

'lasted for at most five years' will be neither determinately true nor deter-

minately false of the object denoted by the term 'the earlier club' (for if the

identity were false that predicate would be true of the earlier club, and if

the identity were true it would be false of the earlier club). Similarly, the

predicate 'lasted for at least twenty-five years' will be neither determinately

true nor determinately false of the earlier club (for if the identity were true

that predicate would be true of it, and if the identity were false it would be

false of it). On the other hand, the predicate 'lasted for at most five years or

lasted for at least twenty-five years' must be determinately true of the

earlier club (for the object determinately denoted by 'the earlier club' has

certainly lasted for at least five years on Broome's account, and there is no

other longer-Jived entity, apart from the latcr club-which has lasted for at

least twenty-five years-with which it might be identical, so its Iifespan

must either be a maximum of five, or a minimum of twenty-five, years). The

Non-Conditional Disputations 81

earlier club, then, on Broome's account, must be an object which determi-

nately satisfies the predicate 'lasted for at most five years or lasted for at

least twenty-five years' but neither determinately satisfies the predicate

'lasted for at most five years' nor determinately satisfies the predicate

'lasted for at least twenty-five years'. But I do not understand how there can

be such an object. (Noonan 1984: 119)

Noonan's bewilderment (as he goes on to explain) is not with the

general case of a disjunctive predicate being true of a thing when

neither disjunct is, for he believes that this occurs naturally with

borderline cases such as being either-orange-or-red; the puzzle-

ment is over examples like that above, where the description ('at

most five or at least twenty-five') prohibits such a borderline. I have

difficulty with such cases as well. But I am not convinced that we

have such a case at hand. T agree that each of these predications

are indeterminate:

The earlier club lasted for at most five years

The earlier club lasted for at least twenty-five years.

But I think the disjunctive predication is indeterminate as well:

It is indeterminate whether the earlier club lasted for at most

five years or at least twenty-five years.

Why does Noonan think the disjunctive predication is true? His

argument has this structure:

The predicate 'lasted for at most five years or lasted for at least

twenty-five years' must be determinately true of the earlier

club.

For: (i) the object determinately denoted by 'the earlier

club' has certainly lasted for at least five years,

and: (ii) there is no other longer-lived entity, apart from the

later club-which has lasted for at least twenty-five

years-with which the earlier club might be identical,

so: (iii) its lifespan must either be a maximum of five, or a

minimum of twenty-five, years.

T agree with (i), but (ii) is difficult to interpret. Part of the trouble

is the 'might' in:

(ii') there is no entity x such that x is distinct from the later

club and x lasts more than five years and such that the earlier

club might be x.

82 Non-Conditional Disputations

If the 'might' is read modally, then the conclusion (iii) does not

follow. If it is merely rhetorical, as in:

(ii') there is no entity x such that x is distinct from the later

club and x lasts more than five years and such that the earlier

club is x.

the clause is much too strong; it is equivalent to the claim that the

earlier club is identical with the later club. This yields the con-

clusion, but it begs the question.

Probably what is meant isY

(ii') there is no entity x such that x is distinct from the later

club and x lasts more than five years and such that it is not

determinately true that the earlier club is not x.

that is:

(ii") -,3x[-,(x == I) & (x lasts more than five years) &

-,!-,(e == x)].

But this is not true in the example envisaged. Not because it is false,

but because it is indeterminate. For there is an instance of (ii")-

the earlier club itself-which makes (ii') indeterminate in truth-

value. The instance is:

Indeterminate: -,(e == I) & (e lasts more than five years) &

-,!-,(e == e).

111is is indeterminate because each of its first two conjuncts are

indeterminate (and its last conjunct is true). Since this is indeter-

minate, its existential generalization:

3x[-,(x == I) & (x lasts more than five years) & -,!-,(e == x)]

is either indeterminate or true. But this is the negation of (ii") ,

which is thus indeterminate or false. So the argument is not

successful.

Of course, one might interpret 'apart from the later club' to mean

'determinate!y distinct from the later club. Then (ii') has the form:

-,3x[(!-,x == I) & (x lasts more than five years) & -,(!-,e == x)].

This does entail:

12 I am indebted here to an anonymous referee for OUP.

Non-Conditional Disputations 83

-,(!-,e = I) v -,(e lasts more than five years) v (!-,e = e).

The last disjunct drops off, yielding:

-,(!-,e = I) v -,(e lasts more than five years).

But the additional '!' in the first conjunct prevents us from

inferring:

(e = l) v -,( e lasts more than five years)

which is what we need to infer:

(e lasts a minimum of twenty-five years) v -,(e lasts more than

five years).

Instead, one gets:

-,!-,(e lasts a minimum of twenty-five years) v -,(e lasts more

than five years).

which is not the desired result.

I see no other way to interpret the point that makes it strong

enough to yield the conclusion without presupposing what is at

issue. (I return to a speculative interpretation later in this section.)

It is worth noting that if this kind of reasoning were to be per-

suasive, it would be a problem for indeterminacy in general, not

just for the indeterminacy of identity. For a conundrum like that

which Noonan presents occurs without identity being involved at

all. Notice that identity was not mentioned in the original scenario;

the question directly posed was whether or not the original club

was revived. Suppose we ignore 'the later club' and simply ask

whether the original club lasted at most five years, or at least

twenty-five years, or either at most five or at least twenty-five

years? It is easy to simulate a puzzle like Noonan's for the original

club:

[I]f it is indeterminate whether the original club was revived, then the pred-

icate 'lasted for at most five years' will be neither determinately true nor

determinately false of the object denoted by the term 'the original club'

(for if that predicate were determinately true of the original club, it would

be determinately false that the original club was revived, and jf it were

determinately false of the original club, it would be determinately true that

there was a revival). Similarly, the predicate 'lasted for at least twenty-five

years' will be neither determinately true nor determinately false of the

84 Non-Conditional Disputations

original club (for if that predicate were determinately true of the original

club, it would be determinately true that the original club was revived, and

if it were determinately false of the original club, it would be determinately

false that there was a revival). On the other hand, the predicate 'lasted for

at most five years or lasted for at least twenty-five years' must be deter-

minately true of the original club (for the object determinately denoted by

'the original club' has certainly lasted for at least five years on Broome's

account, and there is no club with a life-span between five and twenty-five

years, so its lifespan must either be a maximum of five, or a minimum of

twenty-five, years). The original club, then, on Broome's account, must be

an object which determinately satisfies the predicate 'lasted for at most five

years or lasted for at least twenty-five years' but neither determinately

satisfies the predicate 'lasted for at most five years' nor determinately

satisfies the predicate 'lasted for at least twenty-five years'. But how can

there be such an object?

If this reasoning is good, it shows there is no such thing as inde-

terminacy. But the central parenthesis here clearly involves an ille-

gitimate appeal to excluded middle. This is the part of the argument

that Noonan fills in with the remarks about identity examined

above.

5.B.l Super-resolutional readings

I could rest here, but I am bothered, for in spite of the inconclu-

siveness of Noonan's reasoning, I find I am tempted (somewhat) to

agree that the disjunction he cites is true:

The earlier club lasted for at most five years or at least

twenty-five years.

Why is this? When I try to examine my own instincts I find it is

partly my inclination to respond to disjunctive questions with the

Sherlock Holmes principle:

"When all the alternatives have been ruled out, whatever

remains must be true, no matter how unlikely."

And in the case of the disjunction, the alternative is ruled out; it is

ruled out that the earlier club lasted more than five but less than

twenty-five years. But the Holmes principle yields the disjunction

above only if "no answer" is not one of the alternatives. If it is one

of the options, the principle is misapplied.

Non-Conditional Disputations 85

Another closely related theme that moves me towards accepting

the disjunction is to read it:

No matter how things turn out, either the earlier club lasted

for at most five years or at least twenty-five years.

Something like this is a possible construal of the passage from

Noonan quoted above:

[the earlier club 1 has certainly lasted for at least five years ... , and there

is no other longer-lived entity, apart from the later club-which has lasted

for at least twenty-five years-with which it might be identical, so its lifes-

pan must either be a maximum of five, or a minimum of twenty-five, years.

It is indeed true that in any determinate extension of this world, the

lifespan of the earlier club must be either a maximum of five, or a

minimum of twenty-five, years. But to read the sentence in this way

is to read it super-resolutionally; the sentence is true in every res-

olution of the world. But if this is how we read it when we endorse

the disjunction, then we can understand how a disjunction can be

true when its parts are indeterminate. The parts are indeterminate

because there is no one way they turn out in every resolution, but

in every resolution, one of them turns out true, and that is why the

disjunction is super-resolutionally true.

So there are ways in which the disjunction may be true when its

parts lack truth-value, but their existence does not tell against the

indeterminate identity at all.

5.9 WHY CONSIDERING CONJUNCTIONS

OR DISJUNCTIONS IS INEVITABLY

INCONCLUSIVE

There is a theme to the last few arguments we have considered. In

each case, what is at issue is whether a certain claim can lack truth-

value because of what the world is like. The strategy is to focus not

on this claim, but on a complex sentence having that claim as a part,

for example, on a disjunction containing the claim as a disjunct, or

a conjunction containing it as a conjunct. The advantage of this is

supposed to be that we have clear instincts that the disjunction is

true (or that the conjunction is false), thus refuting the claim that

86 Non-Conditional Disputations

its disjuncts lack truth-value. This strategy has two weaknesses. The

first is that the data to the effect that the disjunction is true are not

compelling. The second is that when the disjunction is read so as to

make its truth compelling, this appears to force a reading in which

the disjunction is true in spite of the possible lack of truth-value of

its parts, and so one may not reason from the truth of the whole to

the truth-values of its parts. I don't know any way around this. If

there is a reading on which the disjunction is true, that seems to go

nowhere in attacking indeterminacy.

6

Conditional Disputations

Some people have thought that indeterminacy of identity can be

refuted by considerations gleaned from the logic of conditionals-

the logic of statements of the form 'If A then B'. I think that this

cannot be right; indeterminate identity can be stated and assessed

without the use of conditionals, and considering conditionals

merely brings in irrelevant complications. But others have thought

differently, and it is incumbent on me to provide a coherent

response to their objections. Besides, although it is possible to avoid

the use of conditionals, it is awkward.

I begin by saying what I mean by 'if ... then .. .', and then I

discuss constraints on other options for treating conditionals. We

then look at various ways in which Leibniz's Law can be formu-

lated as a conditional. This is followed by some ways in which the

logic of conditionals and biconditionals has been used in attempts

to refute indeterminate identity. (The sections after §6.1 are not

presupposed by later chapters.)

6.1 THE CONDITIONAL AND THE

BICONDITIONAL

For reasons given earlier I ignore supervaluational readings and I

concentrate entirely on conditionals for which the truth-value

status of the whole is determined by the truth-value statuses of the

parts. I take for granted that any feasible rendition of the condi-

tional will agree with the classical material conditional when the

parts of the conditional have truth-values.l also assume that a con-

ditional with a false antecedent and a truth-valueless consequent is

just as true as one with a false antecedent and a false consequent;

so it is true. Likewise for a conditional with a true consequent and

88 Conditional Disputations

an antecedent without truth-value. So any truth-table will be got

by filling in the cases left open in Table 6.l.

If we use the classical definition of the material conditional in

terms of disjunction and negation:

<jl ::J \jf =df -,<jl V \jf,

then our previous analysis of disjunction yields a conditional whose

truth-table is got by making all the open cases above be truth-value

gaps (Table 6.2).

This conditional is well defined, but it does not seem to reflect

what people have in mind when they choose to say 'if ... then .. .'

instead of'not ... or .. .'. At the very least, there is a much stronger

inclination to treat 'If S then S' as automatically true than to treat

'S v -,S' as automatically true. Doing this requires that the line in

the truth-table for 'if ... then .. .' with two gaps be T. I will assume

this in what follows. So the question is how to fill in the other two

empty cells in the table. I choose the most popular version, the so-

called "Lukasiewicz" interpretation, which makes the remaining

empty cells be gaps (Table 6.3).

I call this "sustaining if-then": the truth-value status of a condi-

tional is determined by how far the consequent drops below

the antecedent in truth-value status, counting T as highest and F

as lowest. If there is no drop at all, the conditional is true, if there

TABLE 6.1.

TABLE 6.2.

<P

\jf If <p then \jf

<p \jf <p :::> \jf

T T T

T T T

T ? ~

T

T F F

T F F

T T

T T

? ~

F ? ~

F

F T T

F T T

F T

F T

F F T

F F T

Conditional Disputations 89

is a drop all the way from T to F it is false, and otherwise the

conditional lacks truth-value. The resulting conditional vali-

dates most of the laws one naturally expects from a conditional,

such as modus ponens, modus tal/ens, hypothetical syllogism, and

contraposition.

The conditional just defined also yields a natural biconditional.

That is, if 'cj> <=> 'V' is defined as '( cj> ~ 'V)&('V ~ cj>)', then one obtains

a biconditional that is true when cj> and \jf have the same truth-value

status, false when they differ in truth-value, and otherwise indeter-

minate (Table 6.4).

The "if-true" use of conditionals: Woodruff (1969) points out that

there is a use of 'if ... then .. .', which he calls the "if-true" use, in

which what one means by saying 'if cj> then \jf' is 'if cj> is true then \jf'.

This is a hypothesis about people's behaviour in interpreting con-

ditionals, and a substantial justification of it would take more dis-

cussion than I can devote to it here, but I think it is right, and it is

important to take into account when assessing arguments that

people actually give in a non-classical setting. I regard this as a

special use of 'if ... then .. .'. This use is easily expressed with the

notation at hand: if someone says 'if cj> then \jf' intending the "if-

true" meaning, their claim should be symbolized as '!cj> ~ \jf'.

The "if-true" use of conditionals is important to capture because

we need to confront the fact that the literal reading of the

TABLE 6.3. TABLE 6.4.

(jJ \jI ( j J ~ \ j I (jJ \jI ( j J ~ \ j f

T T T T T T

T T

T F F T F F

T T T

T T

F F

F T T F T F

F T F

F F T F F T

90 Conditional Disputations

Lukasiewicz conditional, '$ => 'If', does not validate conditional

proof. This is easy to see from the conventions already adopted.

Suppose that $ lacks truth-value, and consider the following three-

line argument using conditional proof:

1.1j>

2 . ~

3. $ =>!$

Hypothesis for CP

From 1

By CP from 1-2 NO!

If this proof were good, '$ => W would be a truth of our logic, but

it is not; it lacks truth-value when $ does. So the '=>' connective

does not validate classical conditional proof. This is important, since

people often take conditional proof techniques for granted in their

reasoning. Woodruffs suggestion is that when people instinctively

use conditional proof in a non-classical setting they are assuming

that $ is true, and then deriving 'If from $, and then concluding cor-

rect/y(!) that they have shown 'if $ is true, then \jf,.1 In a classical

setting, this is no different from concluding 'if $ then \jf', and no

harm is done. But in a non-classical setting one needs to be attuned

to the difference. The difference is that conditional proof is not

valid for the literal reading of '=>', but it is valid for the "if-true"

reading of '=>'. We do not have:

NO:

but we do have:

YES:

!$ => \jf

I If this is done above, line 3 will read '!<I> ~ W, which is a truth of logic.

Conditional Disputations 91

6.2 CONSTRAINTS ON ALTERNATIVES

Why settle for a conditional that does not obey conditional proof

without giving it the "if-true" reading? Why not use a conditional

that satisfies all the laws we expect from a well-behaved condi-

tional? The answer is that no such conditional exists. We can do no

better than the options already discussed. Suppose that you want

a conditional that, like the material conditional, is true whenever

its antecedent is false, and whenever its consequent is true, and that

makes ' ~ => ~ ' always true. There is no such conditional definable

in terms of truth-value status that sanctions these three rules:

2

(i) modus ponens

(ii) modus tollens

(iii) conditional proof

The proof is simple. If we had classical conditional proof, we could

prove 'if ~ then ! ~ ' , as we showed above, so any sentence of this form

would be true. When <I> lacks truth-value, 'if <I> then !<i>' would then be

a true sentence with its antecedent lacking truth-value and its con-

sequent false. Since the truth-value status of the parts determine

that of the whole, any sentence with a truth-valueless antecedent

and a false consequent would be true. Let 'If A then B' be such a true

sentence. Then modus toUens would lead from the truths 'If A then

B' and 'not B' to the truth-valueless sentence 'not N.

So you cannot have everything you want. Nobody may fault the

conditional chosen in the last section merely because it does not

do everything one might want; that is not a reasonable goal. Like-

wise, one cannot propose an alternative treatment of conditionals

without also failing to achieve all desirabilities. Given these facts, 1

am inclined to see the Lukasiewicz conditional as the best of all,

since it validates both modus ponens and modus tollens, and we can

explain good versions of conditional proof as resulting from if-true

readings of it, that is, as proofs that correctly conclude a sentence

of the form 'if !<I> then 'If'.

2 Likewise, no such conditional satisfies all of (i) modus ponens. (ii) contraposi·

tion, (iii) conditional proof. Here is the argument. Conditional proof would allow

us to prove 'if A then !A', so that would be true. If A lacks truth-value, this would

yield truth for any conditional 'if G then H' where G lacks truth·value and 11 is

false. By contraposition, 'if ---.H then ---.G' would then be true, when ---.H is true and

---.G lacks truth·value. But then modus ponens would lead us from truths ('If ---.H

then ---.G' and '---.H') to a conclusion ('---.G') that lacks truth-value.

92 Conditional Disputations

6.3 LEIBNIZ'S LAW AS A CONDITIONAL

It has been suggested that indeterminate identity may be refuted

as follows. First, formulate Leibniz's Law in conditional form:

[LL?] If s = t then (<\ls iff <\ll).

Then use a simple variant of the Evans argument to refute inde-

terminate identity:

1. 'la = b

2. -''l(b=b)

3. -'('la=b iff'lb=b)

4. If a = b then

('la=biff'lb=b)

5. -,a = b.

Hypothesis for refutation

Logical truth

From 1,2 by truth-functional logic

[LL?]

From 3,4 by modus tollens,

con tradicting (1)

Whether this is a good argument or not depends on how the con-

nectives are symbolized. Using the Lukasiewicz conditional and

biconditional the steps are indeed valid:

1. 'la = b

2. -''l(b = b)

3. -'('la=b{::::}'lb=b)

4. a = b ==> (<\la {::::} <\lb)

5. -,a = b.

Hypothesis for refutation

Logical truth

From 1,2 by truth-functional logic

[LL?] using ==>

From 3,4 by modus tollens

So there appears to be a problem for indeterminate identity.

The source of the problem is the formulation of Leibniz's Law.

I have accepted a version of this law as a principle of inference,

and this automatically validates the following "if-true" version of

Leibniz's Law:

Leibniz's Law: !a = b ==> (<\la {::::} <\lb). [Good Version]

But I have rejected the contrapositive version of the law. Since the

'==>' conditional satisfies contraposition, it is easy to show that the

following simpler version of Leibniz's Law must not be accepted:

NO: a = b ==> (<\la {::::} <\lb). [Bad Version]

This is the version of Leibniz's Law used in the above proof, and it

must be rejected. If the good version of Leibniz's Law is used

instead, the last line of the proof does not follow by modus tollens,

nor does it follow by any acceptable inference pattern:

Conditional Disputations 93

1. 'Va = b

2. -,'V(b = b)

3.

4. !a = b :::::} (ljIa IjIb)

5. -,a = b

Instead, one can infer:

5'. -,!a = b.

Hypothesis for refutation

Logical truth

From 1,2 by truth-functional logic

Leibniz's Law as a conditional:

good version

From 3,4 by modus to liens NO!

But (5') is compatible with the claim that 'Va = b, and it achieves no

refutation of it.

Any defender of indeterminate identity is thus forced to accept

certain conditional formulations of Leibniz's Law, and reject

others. A good version is:

Leibniz's Law:

I see no difficulty in rejecting the other version above, only the

inevitable awkwardness that comes with using conditionals in a

framework that takes indeterminacy seriously.

In conversation, I have occasionally been confronted with the

following sort of argument:

Granted, we disagree about indeterminate identity, and we

don't seem to be able to find grounds to resolve our disagree-

ment directly. So we should turn instead to an independent dis-

cussion of conditionals, about which we have intuitions having

nothing to do with identity. We clearly expect conditionals

to be minimally well behaved, and our only assumption about

identity is that we expect Leibniz's Law to have a conditional

formulation. But then the proof at the beginning of this sec-

tion is valid, and it disproves the existence of indeterminate

identity.

The answer must be that there is indeed a conditional formulation

of Leibniz's Law, using the "if-true" reading of the conditional.

With that version of the Law, the above proof is fallacious, as dis-

cussed. If you insist on formulating Leibniz's Law using that same

conditional without the "if-true" reading, then I explicitly disagree,

but this is not a new issue; it is the same old question about the

validity of the contra positive of Leibniz's Law couched in terms of

94 Conditional Disputations

how to formulate Leibniz's Law as a conditional. And if you wish

to use some other conditional with which to formulate the Law,

then we must see what its logic is, and whether it can accommodate

the invalidity of the contrapositive of Leibniz's Law and also satisfy

modus tollens (which is needed in the proof above). No new issues

arise; we merely repackage old ones.

For example, suppose that you are inclined to agree with me on

issues concerning indeterminate identity, but you feel that there

must be a conditional form of Leibniz's Law which requires no

special "if-true" reading. That is easily accomplished. Define a new

connective with this truth-table (Table 6.5).

Then formulate Leibniz's Law as follows:

Leibniz's Law

This is completely adequate from my point of view, since it is

equivalent to the "if-true" reading of the version of Leibniz's Law

given above, and it itself requires no special reading. This new con-

nective validates both modus ponens and conditional proof. (It

does not, however, validate modus tollens, or contraposition, so it

does not save the flawed proof above. Again, you can't have

everything you want in a conditional.)

I assume (when this does not beg a question) that the principal

options for interpreting conditionals are Lukasiewicz's "sustaining"

TABLE 6.5.

<\> \If

<\> I ~ \If

T T T

T

T F F

T T

T

F T

F T T

F T

F F T

Conditional Disputations 95

if-then conditional, or that same conditional given its "if-true"

reading. If other writers use conditionals differently, I will need to

be sensitive to that fact. But nobody can appeal to the "naturally

desirable conditional": the one that is truth-status-functional and

that validates all natural logical principles, for no such conditional

exists if there are truth-value gaps. As a corollary, I will insist that

the awkwardness of anyone of the existing non-classical condi-

tionals not be used as an objection to some unrelated claim. Most

important, when we use a conditional, we must keep straight what

meaning is intended by it when a claim is made using it, so that the

meaning does not shift during discussion.

6.4 TWO EXAMPLES OF

ALTERNATIVE CONDITIONALS

I have suggested that discussion of formulations of condi-

tionals does not change anything of substance about indeterminate

identity; it only repackages the issues. Here are two illustrations of

this.

Broome (1984) discusses Evans's argument, and asks what sort

of conditional would be needed to express a correct version of

Leibniz's Law for use in the proof. He assumes that the Law should

take this form:

If a = b then (Ea iff Eb)

and he argues:

The ground of Leibniz's Law is that if a is identical to b then a and bare

one object, and an object has a property if and only if it has it. This licenses

us to say only that

If 'a = b' is true then '(Ea == Eb)' is true.

In other words, it rules out that 'a = b' should be true and 'Ea == Eb' false,

and also that 'a = b' should be true and 'Ea == Eb' undetermined. (1984: 9)

He then claims that in order to express Leibniz's Law in

conditional form, the conditional ':::>' in the Law must have one

of the two truth-tables (Table 6.6 which makes the conditional

be automatically bivalent or Table 6.7). Broome observes that

neither of these tables validates modus tollens, and that

96 Conditional Disputations

TABLE 6.6. TABLE 6.7.

$

\jI if $ then \jI

I»

\jI if I» then \jI

T T T T T T

T F T

T F F T F F

T T T T

T T

F T F T

F T T F T T

F T F T

F F T F F T

Evans's argument formulated with such a version of Leibniz's Law

fails.

3

Broome's second table is the table for the symbol 'c::>' introduced

temporarily above, the connective that embodies the "if-true"

reading of the conditional. Thus his proposal is almost the same

as the form I have suggested for a correct version of Leibniz's Law.

Almost the same; it depends on how we read the biconditional in

the consequent of Broome's version. It is natural to assume that

this is the conjunction of two conditionals,4 so Broome's version of

the law would likely be:

, Garrett (l988) adds that if this is correct then an argument from Wiggins (1986)

against indeterminate identity also fails. But Wiggins means something else by inde-

terminate identity (personal communication); he is disputing a view according to

which a statement might be indeterminate while having a truth-value. So the view

Wiggins disputes does not require formulation in a non-bivalent language, and his

own argument is entirely classical.

4 One might speculate that Broome just means the biconditional to be the three-

valued material biconditional '(jJ '" 'V' defined as '( (jJ&'V) v or equivalently

as v v (jJ)'. But these would produce a version of Leibniz's Law that is

much too strong. In the trivial case it would yield:

a = a c:> (Ea'" Ea).

Since 'a = a' is logically true, this requires that 'Ea'" Ea' also be logically true.

But when 'Ea' lacks truth-value, this (material) biconditional also lacks truth-

value.

Conditional Disputations 97

BROOME a = b c:> [(Ea c:> Eb) & (Eb c:> Ea)].

This differs from my own suggested reading, which can be

expressed as:

TP a = b c:> [(Ea =:} Eb) & (Eb =:} Ea)].

However, in spite of the difference in conditionals in the conse-

quent, it can be shown that these two versions are logically equiva-

lent.' I find my own version handier,6 but there is no difference in

content between the views being expressed. (Broome's is the first

5 It is easy to show that TP entails BROOME, since 'I» => '1" entails '!I» => '1",

which is 'ip 0;> '1". For the other way around, we give a conditional proof. Begin with:

1. a = b Hypothesis for conditional proof.

The following arc all instances of BROOME:

2. a = b <:> [(!Ea 0;> !Eb) & (!Eb <> lEa)]

3. a = b <> [(l,Ea <> !,Eh) & (!,Eb <> !'Ea)]

4. a = b <:> [(!<'Ea <.:> !<'Eb) & (!<'Eb y !<'Ea»)

Modus ponens then yields:

5. !Eh) & (!Eh <:> !Ea)

6. (!,Ea <> !,Eb) & !'Ea)

7. (!<'Ea c.'> IvEb) & (!<,Eb <> !<'Ea)

Changing 'I» <> 'V' to '!q, => '1" (urns these into:

8. (!lEa => !Eb) & (!lEb => !Ea)

9. (!!'Ea => !,Eb) & (!!'Eh => !'Ea)

10. (!!<,Ea => !<,Eb) & (!!"Eb => !"Ea)

Eliminating redundant repetitions of 'I' yields:

11. (!Ea => !Eb) & (!Eb => !Ea)

12. (!,Ea => !,Eb) & (!,Eb => ! ..... Ea)

13. (!<,Ea => l<'Eb) & (!"Eb => l"Ea)

111ese can now be reexpressed as biconditionals:

14. lEa <rl lEb

15. !,Ea <rl !,Eb

16. l<'Ea <rl !"Eb

These three together insure that 'Ea' and 'Eb' have exactly the same truth-value

status. This is sufficient for the truth of:

17. Ea <rl Eb

Then conditional proof (in its good form) from 1-17 yields:

18. la = b => (Ea <rl Eb) QED.

A similar proof can also be given using Broome's first truth-table for 'if then'.

6 e.g. in the preceding footnote it was trivial to show that TP entails BROOME,

but the other way round was complicated.

98 Conditional Disputations

conditional formulation of Leibniz's Law in the literature that I

know of that is congenial to indeterminacy of identity.)

10hnson (1989) objects strongly to Broome's formulation.

10hnson says that Broome

apparently confuses the object-language symbol for conditionals ':::J', with

a metalinguistic symbol invoking the truth of the antecedent and conse-

quent, such that he reads 'P :::J Q' as if it symbolized 'if P is true then Q is

true'. One serious consequence of this error is that Broome's truth tables

involve rejection of modus toliens; rather than resort to such extremes it

would be better to admit that incoherence of vague identity for which

Evans argues. (1989: 105-6)

(Johnson also objects (1989: 106-7) to my own earlier objections to

Evans's use of abstraction, and of my separation of Leibniz's Law

from its contrapositive.)

I don't see Broome as confusing object-language and metalan-

guage at all; he merely states in the metalanguage his opinion about

the conditions that are needed for the truth of the conditional in

Leibniz's Law. Does 10hnson disagree with these conditions? He

has his own proposal for how to formulate Leibniz's Law as a con-

ditional. He begins his discussion by criticizing this formulation of

the principle:

LL VyVz«(y == z) ~ [A.x(<I>x)(y) == A.x(<I>x)(z)]),

in which he assumes that the conditional is the three-valued ma-

terial conditional, where '<I> ~ \)I' is equivalent to '---,<1> v \)I', with the

truth-table (Table 6.8) displayed over (and, presumably, the

material biconditional is equivalent to the conjunction of two ma-

terial conditionals). lohnson agrees that someone who wishes to

attack indeterminate identity will be begging the question by for-

mulating Leibniz's Law in this way, and so he considers ways to

reformulate it. After some discussion, the version of Leibniz's Law

that he finds uncontroversial is the following, which is just like LL

except for the addition of a conditional antecedent making all con-

juncts determinate:

7

LLu VyVz([---,v(y = z) & ---,vA.x(<I>x)(y) & ---,vA.x(<I>x)(z)] ~

[(y = z) ~ [A.x(<I>x)(y) == AX(<I>X)(z)]]).

7 10hnson uses ''''IJ>' to mean that IJ> is determinate. I have replaced this in his

formula by , ~ , , < P ' , the equivalent claim that <p is not indeterminate.

Conditional Disputations 99

TABLE 6.8.

<P

IjI $::lljl

T T T

T

T F F

T T

F

F T T

F T

F F T

It is a straightforward matter to determine (by inspection of truth-

tables) that LLu is logically equivalent to the version of Leibniz's

Law proposed by Broome.

8

So 10hnson shows how to produce a

version of Leibniz's Law that does not disagree with Broome (or

myself) on any matter of substance. Does he also avoid what he

calls Broome's "resorting to extremes"? Only by looking the other

way. The connective that Broome uses for 'if ... then .. .' is, after

all, formulable, as Broome has clearly shown. It is also easily defin-

able using 10hnson's own terminology; Broome's '<\> <::> '1" (which is

equivalent to '!<\> ~ '1") is logically equivalent to 10hnson's '-,v<\> ::J

(<\>::J'1')', and one can easily see how this equivalence pattern

relates BROOME to LLu above. And although lohnson avoids

Broome's "extreme" conditional, he himself is forced to abandon

a conditional formulation of Leibniz's Law in favour of a condi-

tionalized conditional version of that law. An acceptable version of

the law as a simple conditional will require some conditional very

much like Broome's. Note also that if one objects to Broome's con-

ditional because it fails to satisfy modus tallens, it is equally objec-

tionable that 10hnson's conditional does not satisfy conditional

~ That is, this is easy to establish if we ignore Johnson's use of abstracts. If we do

not ignore his use of abstracts, lohnson's version of Leibniz's Law is considerably

weaker than Broome's version or mine. They become equivalent if one adopts unre-

stricted abstraction principles; see Ch. 4.

100 Conditional Disputations

proof. We have here a deadlock regarding how best to read condi-

tionals, but in the end no difference in the philosophical position

being formulated.

9

The fact that these very different approaches to how to formu-

late Leibniz's Law as a conditional by myself, Broome, and 10hnson

turn out to be logically equivalent suggests to me that there is a

consensus on the important topic-indeterminacy, and indetermi-

nacy of identity-in spite of differences of instinct about how to

treat conditionals.

6.S WILLIAMSON'S DEFENCE OF

THE TARSKI BICONDlTIONALS

There are a number of a priori proofs in the literature to establish

that there cannot be a failure of bivalence for meaningful declara-

tive sentences. In this section I discuss a recent version using bicon-

ditionals that has gained some currency. The argument is from

Williamson (1994: §7.2). The argument is put as a reductio ad absur-

dum of any specific claim to the effect that a meaningful declara-

tive utterance is neither true nor false. This is a bold move, since

the argument attempts to disprove the existence of truth-value gaps

arising from any source, not just those arising from indeterminacy

in the world; it attempts, for example, to disprove the existence of

truth-value gaps due to semantic indeterminacy, provided only that

the utterance in question is a declarative one that succeeds in

saying something meaningful.

Williamson is careful to relate his argument to utterances rather

than sentences, because context can change the semantics of a sen-

tence, and he is careful to allow that the account will work for utter-

ances of sentences that are not in English. But neither of these is

crucial to my discussion, so I simplify by talking as if sentences are

at issue, and I stick to English examples. I also avoid marginally

meaningful utterances ('The number 2 is green'), ones that are

unclear in scope ('Swans are white'), and ethical statements ('The

9 Notice also that Johnson's formulation of Leibniz's Law as LLu validates the

inference rule that I have called Leibniz's Law and fails to validate the inference

law that I have called the contrapositive of Leibniz's Law, in spite of Johnson's

objections to distinguishing these from one another.

Conditional Disputations 101

death penalty is immoral'), since these raise special problems of the

sort I am not focusing on.

Using Williamson's notation, I suppose that 'P' abbreviates a sen-

tence of English and that 'u' is a name of that sentence, e. g. a quo-

tation mark name. For example 'P' might be:

I am my body,

in which case 'u' would be:

'I am my body'.

The argument itself is expressed independently of any particular

choice for 'P' and 'u'. The argument is a short one:

l. Not: u is true or u is false Hypothesis for refutation

2a. u is true iff P Semantic fact

2b. u is false iff not P Semantic fact

3. Not: P or not P Substituting in (1) using

(2a) and (2b)

4. Not P and not not P DeMorgan's Law from (3)

The argument consists of a hypothesis for refutation, two semantic

facts, and then two inferences culminating in a contradiction, thus

refuting the initial hypothesis. The inference from (3) to (4) is

acceptable from my point of view, so I will focus on the earlier lines.

This reasoning is too good to be true. It refutes not only inde-

terminacy of identity, but indeterminacy of any meaningful declar-

ative utterance, thus demolishing decades of work in the logic and

semantics of sentences without truth-value. This is too much to

expect. But the reasoning is apparently compelling, and it is cer-

tainly incumbent on me to explain where and why the reasoning

goes wrong. This is not a simple task, since indeterminacy allows

for a number of options regarding what one means and what one

says. So there is no avoiding a survey of the options.

The two principal questions that need to be addressed are what

one means by the predicates 'true' and 'false', and what one means

by the biconditional 'if and only if. (I take for granted that we agree

on the meanings of 'not', 'and', and 'or'. There are options for these

as well, but I don't think that Williamson's argument turns on using

these differently from the way I do.)

The principal question about the use of 'true' and 'false' is

whether they are themselves intended to be used bivalently. I have

102 Conditional Disputations

implicitly adopted a bivalent use for them, in the sense that if a sen-

tence is lacking in truth-value it is determinately not true, as I use

'true' (and it is also determinately not false). Call this the "deter-

minate" use of 'true' and 'false'. There is also another use, explored

by several writers on the semantic paradoxes, starting with Martin

and Woodruff (1975), and Kripke (1975). Call this the "redun-

dancy" use. On this use, if a sentence lacks truth-value, then so does

the claim that it is true or false. So if a sentence lacks truth-value,

the claim that it is true is itself indeterminate. This is not my usage

here, but it is an important optionj so let me deal with it briefly. The

important thing to say about it is that holders of such a view will

agree with Williamson in rejecting (1), but they will disagree about

his claim that (1) is the right way to formulate an assertion of lack

of bivalence. Exactly how an assertion of lack of bivalence is to be

stated on this view is a matter of some debate, but (1) is certainly

not the way to do it. I have discussed elsewhere (in Parsons 1984)

how things should go on this option, and it would be too much of

a digression to recapitulate it here. It is important to realize that

this option exists, partly to be aware that there are other defences

of lack of bivalence than the one I will be giving, and to distinguish

this alternative approach from my own in this book. But from here

on I ignore this option.

Let us settle, then, on the bivalent use of 'true' and 'false', and

admit on that view that (1) does correctly formulate a claim of inde-

terminacy. That then puts the burden on (2a) and (2b). Whether I

am willing to endorse (2a) and (2b) depends on how the bicondi-

tional is read. There are a number of options here; 1 focus only on

the most relevant ones. One of these, of course, is the biconditional

that I have adopted above. I interpret '<jl <=> \jf' as being true if <jl and

\jf have exactly the same truth-value status, false if they disagree in

truth-value (if one is true and the other false) and otherwise truth-

valueless. On this interpretation I would not assert either (2a) or

(2b) unless I were sure that P had a truth-value. This is because

they would be Jacking in truth-value if P were itself lacking in truth-

value. For the claim that u is true would be false in that case, and

also the claim that u is false. So these would be biconditionals

without truth-value:

2a. u is true <=> P

2b. II is false <=> not P.

No truth-value

No truth-value

Conditional Disputations 103

I would thus view Williamson's argument as begging the question

by assuming these.

But these are just the old and venerable Tarski biconditionals,

unquestioned within a two-valued framework. Certainly they seem

to be trying to capture something that is right, and this should not

be swept under the rug. What is it about them that is right, and what

would I use in their place? A natural option would be the "iff-true"

reading of the biconditional. This is a parallel to the "if-true"

reading of the conditional, discussed earlier in this chapter. We con-

strue the biconditionals in (2a) and (2b) as conjunctions of the "if-

true" conditionals, so that we get

JO

2a. [u is true r:::> P] & [P r:::> u is true]

2b. [u is false r:::> not P] & [not Pr:::> u is false].

These are correct on anybody's view; the only question is whether

they say enough. They are also the natural conditional renditions

of the following "Tarski biconditional inference schemes", which

are also correct on anybody's view:

VALID:

VALID:

s

IS' is true

not 5

'5' is false

'5' is true

5

IS' is false

not 5

Williamson (1994: 300) notes with approval that Evans and

McDowell (1976), Machina (1976), and Peacocke (1981) all

endorse "disquotational schemas for truth". So do I, in the

restricted sense that I regard the inferences above as valid, and the

following conditionals as true without exception:

IS' is true <:>5

5 r:::> '5' is true.

If we read (2a) and (2b) as indicated, they are true according to the

theory I endorse. We thus have (1) as a denial of bivalence, and (2a)

and (2b) in forms that I endorse. However, with the versions of (2a)

and (2b) just given, (3) no longer follows. For the fact that if 'u is

true' is true, so is' P', and vice versa, does not mean that 'u is true'

10 In the notation originally introduced in Ch. 2, these are:

2a. [!(u is true) => P] & [JP => u is true]

2b. [!(u is false) => not PJ & [!(not P) => u is false].

104 Conditional Disputations

and 'P' have the same truth-value, since those connections obtain

even if 'P' does not have a truth-value at all and 'u is true' is false.

So without the assumption that 'P' has a truth-value, the substitu-

tion that leads to (3) is not verified. And the assumption that' P'

has a truth-value is what is at issue.

In summary, there is no doubt that the wording on line (1) of

Williamson's proof:

1. Not: u is true or u is false Hypothesis for refutation

can be interpreted so as to deny bivalence. And there is no doubt

that the wording on lines (2a) and (2b):

2a. u is true iff P

2b. u is false iff not P

Semantic fact

Semantic fact

can be interpreted so as to be endorsable by a person who believes

in the possibility of non-bivalence. But on these readings, line (3)

does not follow.

3. Not: P or not P Substituting in (1) using (2a) and (2b).

It is also possible to give each of these lines other readings, and on

some of these the conclusion does follow, but not in a way that

refutes anything that a believer in non-bivalence believes.

Williamson, of course, is alert to this sort of reply, and so he does

not rest content with merely giving the proof. He adds:

It might be replied that if u says that P and is neither true nor false, then,

'u is true' is false while' P' is neither true nor false, so that the two sides

of (2a) do not match in semantic value, and neither (2a) nor (T) is true.

A parallel reply might be made to (2b) and (F). The trouble with this

objection is that it does nothing to meet the rationale for (T) and (F). It

gives no hint, when u says that TW is thin, of any way in which u could

fail to be true, other than by TW failing to be thin, or of any way in

which u could fail to be false, other than by TW failing to be not thin. (1994:

190)

The principles (T) and (F) that Williamson alludes to are:

(T) If u says that P, then u is true if and only if P

(F) If u says that P, then u is false if and only if not P

I do indeed hold that it is possible for the two sides of (2a) not to

match in semantic value-the left-hand side might be false when

Conditional Disputations 105

the right has no truth-value at all-so the quoted comment is apt.

This shifts the onus of discussion to the rationale for (T) and (F).

Williamson states their rationale as follows:

The rationale for (T) and (F) is simple. Given that an utterance says

that TW is thin, what it takes for it to be true is just for TW to be thin,

and what it takes for it to be false is for TW not to be thin. No more

and no less is required. To put the condition for truth or falsity any higher

or lower would be to misconceive the nature of truth or falsity. (1994:

190)

I find this rationale completely compelling. The question,

however, is whether it validates (T) or (F) on the interpretation

that Williamson needs. Notice that (T) and (F) are biconditionals,

and the issues discussed above about biconditionals arise here too.

If they are given the "iff-true" interpretation, then they are com-

pletely acceptable but they are consistent with the view that

WilIiamson is disputing. So suppose that we give them the stronger

reading, so that they are false for any meaningful utterance that

lacks truth-value. Does the rationale the WilIiamson gives validate

them on that reading? That depends on how the rationale is under-

stood. How are we to interpret, e.g.:

(*) What it takes for 'TW is thin' to be true is just for TW to

be thin?

A t the very least, we need this to entail that either of these:

'TW is thin' is true

TW is thin

is inferrable from the other; that is, that both of these inferences

are valid:

'TW is thin' is true TW is thin

TW is thin 'TW is thin' is true

But they are both valid, and that is not sufficient in a language with

truth-value gaps for the previous argument to be valid. So the ratio-

nale for (T) and (F) is not sufficient to validate (T) and (F) if they

are read so as to entail bivalence.

Then perhaps one wants a stronger rationale. Something like that

can be provided. I suggest this:

106 Conditional Disputations

There are three options regarding the state of affairs of TW's

being thin: it might determinately hold, it might determinately

fail to hold, or neither. It is constitutive of the meaning of the

sentence 'TW is thin' that it match these options precisely, so

that:

(i) If the state of affairs determinately holds, the sentence

must be true.

(ii) If the state of affairs determinately fails to hold, the

sentence must be false.

(iii) If neither option is realized, the sentence must be

neither true nor false.

This is as strong a rationale as anyone might demand, though it

differs from Williamson's. It allows, rather than proscribing, failure

of truth-value. No doubt a defender of bivalence will claim that

option (iii) cannot be realized, but that is not to say that the ratio-

nale is false, it is just to say that it does not rule out lack of biva-

lence. If more than this is required by Williamson's understanding

of the rationale, then the rationale, so understood, is itself far from

obvious, and cannot itself be the foundation from which one can

argue in a neutral way about bivalence.

My aim here is not to establish that bivalence fails; it is merely

to defuse arguments to the effect that it cannot fail. To decide

whether it fails, we need to consider whether a hypothesis of lack

of truth-value provides a better over-all explanation of the entire

set of statements that are prima facie lacking in truth-value, and

this cannot be adequately assessed by means of an a priori argu-

ment on either side.

7

Understanding Indeterminacy

7.1 "I JUST DON'T UNDERSTAND .. "

People have told me that they do not understand indeterminate

identity. Sometimes 1 have told myself the same thing, so it is hard

to discount all the sources of the worry. The point of this chapter

is to see what can be said on this issue.

To begin, we need to divide the question. When someone says

that they do not understand indeterminacy in general, or inde-

terminate identity in particular, they may be making one of

two kinds of comment. The first is a rejection of indeterminacy,

while the second is merely a confession of lack of understanding.

The first is usually expressed as a comment to the effect that one

does not understand how identity could possibly be indeterminate;

the other is merely a report of bewilderment in thinking about

the issue. The first has the effect of a positive assertion rejecting

the claim that identity might be indeterminate, made emphatic by

a comment about understanding. The second is a mere confession

of lack of understanding. These invite different though related

responses.

A claim of the first sort can be put forward from a variety of

rhetorical stances. The most challenging stance is when the intent

IS:

"I don't believe it, because I can't conceive of its being true.

And you can't change my mind about this unless you can show

me how it is that identity is indeterminate."

The best 1 can do in response to a challenge of this sort is to point

out what things would be like if identity were sometimes indeter-

minate, e.g. in the classical identity puzzles, and then point out that

things are like this, and that alternative explanations have not been

108 Understanding Indeterminacy

found to be very satisfying. This is true of indeterminacy in general,

and of indeterminacy of identity in particular.

The second kind of comment indicates a stance, something like

this:

"I don't believe it, but I'm willing to consider it. But when

J ponder it, I feel genuinely bewildered."

This is a very serious challenge, and one that festers inside me as

well. It is what has motivated this whole study. I, too, feel the bewil-

derment (it comes and it goes), and this whole work is devoted to

addressing it.

Three things are needed for understanding. The first is to be sure

that the terminology being employed is understood; if this is not

accomplished, no view will have been presented to be understood

or not understood. The second is to address the fact that people

need to somehow "conceive of" what it "would be like" for the view

to be true. I am not sure what this amounts to, but it has a vener-

able and influential tradition-including, but not limited to, both

empiricist and rationalist theories that require a kind of mental pic-

turing for understanding. The third is to have a kind of working

knowledge of the theory; you need to know the sorts of situations

in which the view applies to things, and how it is supposed to apply

in those cases, and what the consequences are of its not applying

in those situations. I address the first two of these in the present

chapter; the third is addressed throughout the book. In the next

section I discuss the terminology employed in theorizing. After

that, I discuss "picturing" what it would be like for the theory to be

true.

7.2 UNDERSTANDING THE TERMINOLOGY

I intend my terminology to be completely normal. When I speak

of identity, and when I use the sign '=', or use 'is' in the sense of

identity, I mean exactly what everyone else means. My definition

of identity in Chapter 3 is not meant to confer meaning on the

word 'identical'; it is a substantive proposal about when this

relationship holds. The same is true of my use of the words 'true'

and 'false', and of 'property', 'object', 'refer', 'and', 'or', 'every', and

Understanding Indeterminacy 109

so on. Of course, each of these terms is subject to various

interpretations; they are used differently by different people. I

have either relied on context to home in on my intended usage, or

(e.g. in the case of the truth-table definitions of the connectives)

I have explicitly addressed which of the available meanings I had

in mind. These accounts are never meant to be stipulative in the

sense that I am inventing some new technical meaning from

scratch; they are rather meant to be stipulative in the sense that I

need to declare which of the several normal uses I intend. If I have

not succeeded in doing this in some cases, then that is an error in

detail in my exposition-it is a failed attempt to isolate my meaning

from a host of familiar ones, not a successful attempt to invent a

new idea.

Perhaps the familiar meanings are too corrupt to use; if so, my

exposition has failed, along with much of contemporary philoso-

phizing. For example, if WiIIiamson is right, then there is no gap at

all between a declarative utterance expressing something mean-

ingful and that utterance being true or false, and so there cannot

be a meaningful utterance without truth-value. And, if he and

others are right, this is somehow constitutive of the meaning of

'true'. And so I cannot possibly say that a meaningful utterance

lacks truth-value, meaning by 'true' what others mean by it, while

saying something that might be correct. I insist that I do mean what

others mean by 'true', and so the challenge must be an assertion

that what I have to say is wrong, but not an assertion that what I

have to say cannot be understood. If the critics are right, the most

they can claim is that what I have to say cannot be believed,

because understanding what I mean entails being unable to under-

stand what it would be like for it to be true. But they cannot claim

that what I have to say cannot be understood.

With regard to indeterminacy of identity, there are two key no-

tions to understand, 'determinate' and 'identical' as they occur in:

For some object x and some object y, it is not determinate

whether x is identical to y.

I have already commented on identity. Determinacy is for me a

primitive notion. It can be given more content within certain world-

views, such as idealist ones, but I have no transcendent account to

give of it. But it is clear, I think, how it is to function in the theory.

Of course the displayed statement in which these terms occur

110 Understanding Indeterminacy

involves quantifying into a whether-clause, and this easily invites

claims of incoherence. So the stage then shifts to a detailed account

of language and the world and the semantical relations between

them. Here, plenty of terminology enters, all of it problematic. But

none of it is specially problematic. Perhaps I am saying things using

this terminology that invites special responses in incredulity, but

I am not using either terminology or semantical constructions

in special ways. And so the issue is either belief, or difficulty of

picturing or conceiving of claims that are meaningfully expressed.

The question of meaningfulness arises not only when a theory is

formulated, but also when it is applied to problematic situations.

And I want to emphasize that in all of the applications the termi-

nology I use is normal. In particular, I use words of our native

tongue, such as 'person', 'ship', and 'cat'. Here I may be criticized

not for using unexplained terminology, but for using familiar

terminology that is not sufficiently scientific or philosophically

respectable. But in this respect I think the applications described

here are superior to approaches that require unexplained techni-

cal terminology, such as 'gen-identity', or that suggest that we need

completely new terminology to replace ordinary talk of objects, but

without providing the new terminology. This issue will be engaged

later (in Chapter 10); I raise it now to alert the reader to it and to

emphasize that the question of meaningfulness must be addressed

with respect to a whole theory including its applications, not just to

its theoretical core.

7.3 NON-BIVALENT REASONING

Probably the biggest impediment to understanding indeterminacy

of any kind is that it brings with it lack of truth-value, and this

affects the kind of logical inferences we can make. There is no

mystery about this at all, and the failure of certain inferences we

normally take for granted are easy to understand when scrutinized.

The problem comes when we do not scrutinize them, but instead

take bivalent reasoning for granted. This principally involves what

I have called contrapositive reasoning, which is not valid in a

non-bivalent setting. I have belaboured this so much elsewhere that

it does not make sense to do so again here. But it is important to

Understanding Indeterminacy 111

highlight this phenomenon, since (it seems to me) this is really the

root cause of most of our feelings of bewilderment when ponder-

ing indeterminacy. I have found that as the instinct to reason con-

trapositively lessens, the feeling of lack of understanding similarly

lessens. This is not a point that can be argued; you need to experi-

ence it yourself to get the effect.

7.4 PICTURING INDETERMINATE IDENTITY

The claims that lead to bewilderment are mostly claims to the effect

that it is indeterminate whether certain objects have certain prop-

erties and/or that certain objects are identical. There is much more

to the account than this, but if we can "picture" this part of it, the

rest will be easy.

There is a centuries-long tradition of using Venn-like diagrams

to picture objects possessing properties and to picture objects being

identical or distinct. Many people have found these diagrams con-

ceptually useful; they can be adapted for present purposes as well.

In the traditional diagrams, the extensions of properties are pic-

tured by certain subregions of a rectangular figure, and objects are

represented by points, as in Figure 7.1. An object represented by a

point (such as b) located inside a property-region (such as P) is pic-

tured as having the property in question; a point (such as a) located

outside the region represents an object that lacks the property.

Following Parsons and Woodruff (1995), I extend these con-

ventions by representing objects by small shaded regions of the

figure. An object-region, a, inside a property-region, P, represents

the object's having the property P, one completely outside the

property-region, b, represents an object's lacking P, and one lying

partially within and partially without the property-region, c, repre-

sents its being indeterminate whether the object has P (see Figure

7.2). If a region has no object-images in it, the picture itself is

neutral regarding whether there is an object present in it. But if we

are told that there are no objects other than the ones pictured, then

we know that empty regions represent properties that hold of no

objects.

I assume that there is at least one property that holds of no

objects. Its representing region can be anywhere in the picture, so

112 Understanding Indeterminacy

a

•

b

FIG. 7.1

c

FIG. 7.2

Understanding Indeterminacy 113

long as it overlaps no object image. For example, if there are only

three objects, then the circle labelled 'E' in Figure 7.3 represents a

property determinately not possessed by any object.

These pictures show what things would be like if there is inde-

terminacy; they do not yet address identity. In the classical pictur-

ing, identical objects are represented by coinciding points (or by

the same point; there is no difference) and distinct objects are rep-

resented by distinct points. I extend this convention by assuming

that objects represented by the same regions are represented as

being identical, objects represented by disjoint regions are repre-

sented as being distinct, and objects represented by partially over-

lapping regions are represented as being indeterminately identical.

Figure 7.4 represents a as identical to b, and represents both a and

b as distinct from c, whereas Figure 7.S represents a as distinct from

both band c, and represents its being indeterminate whether b is

c. Notice that a partial overlap of object-representatives does not

indicate uncertainty as to whether they are identical (we have no

means in a picture alone to indicate uncertainty). Partial overlap

means that the world is such that it is not determined whether the

represented objects are identical.

I require by edict that no object-representatives lie completely

inside others, or combinations of others, as pictured in Figure 7.6.

b

FIG. 7.3

114 Understanding Indeterminacy

a b c

FrG. 7.4

a b c

FIG.7.S

Each object representative gets a bit of its own space in the

diagram; that is, part of its region is not shared by any other

representative. l

Various principles are now built into the picturing conventions.

Both symmetry and reflexivity of identity are built in, as is the

symmetry and non-reflexivity of indeterminate identity. The trans-

itivity of identity is also built in. It is also built in that transitivity

of indeterminate identity fails, for a diagram can look like Figure

7.7.

I The possibility of making drawings that violate this principle is an artefact of

the model, just as is the possibility of drawing different representatives in different

colours, or labelling them with different fonts. I disallow such drawings because I

think they have no application to worlds such as ours. But the idea is not incoher-

ent.A referee for OUP has suggested that I might determinately possess every prop-

erty that my body determinately possesses, yet still not be determinately identical

with it, and this would be appropriately diagrammable by representing me as a sub-

region of the region that represents my body. TIlis would require that I be deter-

minately a body, which I do not grant, but certainly someone could believe this. My

worry about this idea is that it seems to me certain that if I am a body, then I am

my body (I am determinately identical with my body), which this view must dis-

agree with. Similarly, it is determinately true that there is exactly one human body

presently in my chair, but this would not be determinately true on the view under

consideration. (See Ch. 8 regarding cardinality claims.) The idea is coherent, and is

worth exploring, but I do not do so in this work.

Understanding Indeterminacy 115

FIG. 7.6

a c

b

FIG. 7.7

Suppose that we require that every region of the diagram repre-

sents the extension of some property. Then the ontological version

of the Leibnizian definition of identity also results from the pictur-

ing conventions: a and b are represented as identical if and only if

their pictures are included in, and are excluded from, and properly

overlap, exactly the same property representatives; they are repre-

sented as distinct if and only if some property representative

includes one of their representatives and excludes the other; and

they are represented as being indeterminately identical if and only

if some property representative totally includes or excludes one of

their representatives while properly overlapping the other.

The pictures are ontological in the following sense: the regions

in them can be used to represent extensions of properties, but they

cannot be presumed to represent the extension of a predicate that

does not stand for a property, unless that extension happens to

coincide with that of some property. Take for example the Evans

pseudo-property of being indeterminately identical to a. If there is

no indeterminacy of identity at all, or even if there is nothing inde-

terminately identical with a, then this predicate is coextensive with

any null property, and we have decided that that has an extension.

116 Understanding Indeterminacy

a b

FIG. 7.8

But suppose that a is in fact indeterminately identical with some

object b, so that their picture looks like Figure 7.8. Then it is

obvious that no region can completely include b while com-

pletely excluding a, which is what would be required for the

(pseudo-)property of being indeterminately identical with a to

have a picturable extension.

The case just given is typical in the following respect: it is a con-

tingent matter whether there is some property that is coextensive

with the predicate 'is indeterminately identical with a'. The same

goes, for example, for' being P or indeterminately P' when 'P' stands

for a property; this is picturable only if its extension and anti-

extension happen to coincide with those of some predicate that

stands for a property. Figure 7.9 depicts a possible state of affairs,

and 'being P or indeterminately P' cannot stand for a property in

this situation. For there would need to be a region that includes all

of that of b without including any of that of a, which is impossible?

2 In Parsons and Woodruff (1995) we considered what abstracts might correspond

to properties, and we gave a non-standard interpretation of "ontological abstracts"

that guaranteed that they always yield properties. This was artificial, and done for

theoretical reasons only. In terms of the diagrams, they worked like this: you take

the objects that determinately satisfy the formula of the abstract, and you draw a

minimal property-region that includes them; then you declare this to be the prop-

erty represented by the abstract. If the abstract satisfies DDiff, then what you get is

exactly right. If not, it is intuitively wrong, but it gives you some property-extension

as an answer. This is like Frege's idea for insuring that definite descriptions always

uniquely refer by requiring that they refer to the right object if there is one, and

otherwise letting them refer to a unique "supplied" object. I now don't like our pro-

posal much, not because it is incorrect in any way, or technically flawed, but because

of its artificiality. It was done in response to criticisms such as Johnson (1989: 106)

about restricting abstraction; the idea was to guarantee that abstracts always stand

for properties in the world, while-consistent with this-maximizing the extent to

which they stand for the "right"' property. But its artificiality makes it confusing, and

I have abandoned it here.

Understanding Indeterminacy 117

a

FIG. 7.9

I take it as a thesis of the theory under investigation that if a

situation is picturable, then it is metaphysically possible for things

in the world to be as the picture represents?

In the next section we look at the intended pictures of the situa-

tions in which the paradigm identity puzzles occur.

7.5 REPRESENTING THE PARADIGM PUZZLES

ME AND My BODY: The first paradigm puzzle presented in Chapter

1 centred around whether I am my body. If this is a case of worldly

indeterminacy, then a diagram for the situation should look like

Figure 7.10. The overlapping shaded circles represent the fact that

J This is not to rule out the fact that it is de re impossible for certain objects

to have certain configurations. For example, many people think that if a 01- b

then necessarily a 01- b, and so there is no possible situation represented by over-

lapping object images if those images represent a and b. But there is a possible

situation represented by such images if we are neutral about which objects they

represent.

118

Physical

objects

Understanding Indeterminacy

FrG. 7.10

it is indeterminate whether I am my body. The diagram represents

the fact that I am determinately a person, whereas it is indetermi-

nate whether my body is, and the fact that my body is determinately

a physical object, whereas it is indeterminate whether I am. The

diagram takes for granted that being a person is a property, and so

is being a physical object.

Figure 7.11 indicates that both I and my body are in my

office. In everyday discourse we talk as if I am where my body

is, and vice versa, in normal circumstances. (If there literally

were such a thing as out-of-body travel, circumstances would not

be normal, and I would not be where my body is. If this were to

occur, the identity issue would be resolved in the negative.) The

diagram is drawn taking this at face value. Some would suggest that

there is ambiguity in talk of location of both persons and their

bodies; if this is so, the region in the diagram represents the

disjunctive property: "being located in my office in the person sense

of location, or being located in my office in the physical object sense

of location".

THE DISRUPTED PERSON(S): TI1e second paradigm puzzle has a

diagram that it isomorphic to the person-body one. Figure 7.12

indicates that person a entered the room, though it is indetermi-

nate whether person b did, and that person b left the room, though

it is indeterminate whether person a did. Tt also indicates that both

are determinately persons, in the straightforward everyday sense of

'person'. The isomorphism between this diagram and the one for

Physical

objects

Things that

entered

Understanding Indeterminacy 119

FIG. 7.11

FIG. 7.12

Persons

Things in

my office

Things that

left

me and my body indicates the lack of relevance of time to the ques-

tion of indeterminacy of identity.

THE SHIPS: Call the original ship '0', the ship with new parts 'p',

and the newly assembled ship 'a'. Then the diagram for the ships is

Figure 7.13. The diagram indicates that it is indeterminate whether

120 Understanding Indeterminacy

Ships

FIG. 7.13

o is p, and indeterminate whether 0 is a, though it is determinate

that a is not p. It is determinate that 0 left port at such-and-such a

time, but indeterminate whether a did and indeterminate whether

p did. It is determinate that a docked at place A, false that p did,

and indeterminate whether 0 did. All of 0, a, and pare determi-

nately ships.

THE CATS: There are probably millions of p-cats, which makes

the picture hard to draw. Let us idealize and assume that there

are merely four. Then the diagram is as in Figure 7.14. Each

p-cat is determinately distinct from each of the others (since

they have determinately different parts). Each is such that it is

indeterminate whether it is the cat (represented by 'c'), and

also indeterminate whether it is a cat. All are determinately on the

table.

SOME PILES OF TRASH: None of our paradigm puzzles yield double

indeterminacies of identity, but these are clearly possible. Consider

a pile of trash PI at time t

l

, and a pile P2 at t

2

, and a pile Pi at t3.

Between tl and t

2

, suppose that pile PI has been blown around and

had some parts changed, enough to make you genuinely uncertain

about whether P2 is the same pile as PI' Likewise for P3 and P2'

It is consistent with this that you might judge that P3 is genuinely

Things

on the

table

Understanding Indeterminacy

FIG. 7.14

121

different from PI, but you also might judge that this is inde-

terminate as well. In the latter case, the diagram will look like

Figure 7.15.

7.6 PICTURING MADE PRECISE

Pictures are useful conceptual guides, but only if we can trust them

to guide us correctly. The discussion of pictures above has been

somewhat informal; it needs clarification so as to be clear just what

is and what is not a picture, and more discussion is needed about

exactly which aspects of pictures are picturing which aspects of the

world. The account also needs to be made more rigorous in order

to be able to show that it delivers the intended judgements about

the world that are being pictured. This is also useful in providing a

foundation for the account of worldly resolutions introduced

earlier. The remaining sections of this chapter are devoted to

working out the technical points that will achieve these goals. (Most

of the material in the remaining chapters of this book does not rely

on the points discussed here.)

122

Piles

of

trash

Understanding Indeterminacy

FIG. 7.15

Although we have spoken of our pictures as being two-

dimensional, the two-dimensional geometry is not really essential

to the representation, nor is the size and shape of the regions. A

simple set-theoretical construction can be given which entails the

principles discussed informally in the last section. This is simplest

if we begin by defining a robust complete picture of a world. I call

such a picture robust because it embodies a principle of plenitude

for properties; it is complete because it includes images of every

object in the world. The diagrams produced above are like snap-

shots of some portion of a robust complete picture.

Call the object-images icons, and call the regions that represent

properties pictensions. Then a robust complete picture consists of a

set ° of points, a set I of icons, which are non-empty subsets of 0,

and a set P of pictensions, which are also subsets of 0. What

more is required for such a combination to represent the world?

All that is required is the right relationship between identity and

the possession of properties; that is, what is required is that the

Understanding Indeterminacy 123

icons and pictensions be so related that they portray this relation-

ship properly. It turns out that a single condition will achieve every-

thing we desire. It is that for any two disjoint sets of icons there is

a pictension such that the first set of icons represents their objects

as determinately having the property represented by that picten-

si on (and no icon not in the first set does this), and the second set

of icons represents their objects as being indeterminate whether

they have the property represented by p (and no icons not in the

second set do this, except for icons that overlap icons in the first

set). Given our conventions for representation, this is expressible

set-theoretically as:

[C] For any disjoint sets sand r of icons, there is a pictension

p such that:

any icon i is a subset of p if and only if it is a member of s,

and any icon i that is not in s properly overlaps p if and

only if i is a member of r or i properly overlaps some member

ofs.

Condition [C] involving sand r is a kind of principle of pleni-

tude for properties. The condition entails certain of the constraints

that were posed above in a seemingly ad hoc fashion. For example,

it entails that no icon is completely contained inside others. For

suppose that icon i is completely inside of some other icons. Take

the set of those other icons to be s in the condition above. Then

the first clause of the condition is violated, because no picten-

sion can have exactly those icons as subsets: since i is "hidden"

by them, it too will be a subset of any set that has all of them as

subsets.

As another example, the condition entails that there is a com-

pletely empty property (it includes no icons and overlaps no icons);

just take both sand r above to be the empty set.

Most important, the condition yields an analogue of the Leibniz

account of identity in terms of properties. Suppose we clarify what

it means for an icon and a pictension to represent something about

the object and the property they represent as follows:

DEFINITION: In any picture, if i is an icon and p is a pictension,

then the picture

(i) represents i's object as determinately having p's property

iff i r;;;.p

124 Understanding Indeterminacy

(ii) represents j's object as determinately not having ps

property iff i and p are disjoint,

(iii) represents its being indeterminate whether i's object has

p's property iff i properly overlaps p.

We also clarify what it is for two icons to represent their objects as

identical or distinct:

DEFINITION: In any picture, if i and j are icons, then the picture

(i) represents i's object as being identical to j's object iff i = j

(ii) represents i's object as being distinct from j's object iff i

and j are disjoint

(iii) represents its being indeterminate whether i's object is

identical to j's object iff i properly overlaps j.

It then follows from condition [C] that in any picture, if i and j are

icons, then the picture

4

(i) represents i's object as being identical to ps object iff it

represents i and} as determinately having and determinately

not having exactly the same properties

(ii) represents i's object as being distinct from j's object iff

it represents i as determinately having a property that j

determinately does not have, or vice versa, and

(iii) represents its being indeterminate whether i's object is

identical to j's object iff neither (i) nor (ii) of the first defi-

nition hold.

An incomplete picture is simply one that does not represent

every object that there is. This is not an inherent trait of a picture;

, This is proved as follows.

For (i). the left-ta-right argument is trivial. For right-to-left: suppose that the

picture represents i and j as determinately having and determinately not having the

same properties; then they are subsets of exactly the same pie tensions. If i and j were

not identical, condition [e) applied to li) and the empty set would yield a picten-

sion that includes i and does not include j, contrary to the supposition.

For (ii): Left-ta-right: suppose that i represents its object as being distinct from

j's object; then they are disjoint. Then, by [e) applied to liJ and the empty set, there

is a pictension that includes i and excludes j, and so the picture represents i's object

as having a property thatj's object does not have. Right-ta-left: Suppose the picture

represents j's object as having a property that j's object does not have.1l1en some

pictension includes i and excludes j; and so i and j are disjoint, and they thus rep-

resent their objects as being distinct.

For (iii): this follows from (i) and (ii).

Understanding Indeterminacy 125

it is a relation between the picture and its intended application. On

the other hand. a non-robust picture is one that portrays fewer

properties than is required by condition [C]. This should be looked

at more closely. since condition [C] provides a stronger principle of

plenitude for properties than is required by the theory under dis-

cussion in this book. In fact. it requires a fairly neat and natural

principle. Recall that in §4.5 we discussed the conditions under

which an abstract 'AX[ <l>x r might stand for a property. The condition

is that DDiff be satisfied for <1>:

-,3x3y[!<jJx & !-,<j)y & 'VX = y].

Suppose we have a complete and robust picture of the world, and

<I> satisfies DDiff. Let s be the set of icons that represents objects

that determinately satisfy <1>, and let r be the set of icons that rep-

resents objects that indeterminately satisfy <1>. Sets sand r are dis-

joint. If condition [C] obtains, there is a pictension p that represents

a property P that is determinately possessed by all and only objects

that are pictured by the icons in s, and is indeterminately possessed

by all and only the objects pictured by the icons in r. Then <l>

expresses that property P, in the sense that it is determinately sat-

isfied by exactly the objects that determinately possess P (namely,

those depicted by an icon in s), and it is determinately dissatisfied

by exactly the objects that dispossess P (namely, those that are not

represented by any icon in s or r). So condition [e] turns the DDiff

condition from a test for whether a formula <I> may express a prop-

erty, into a test for whether <t> does express a property. This gives

concrete form to the linkage between what is true in the language

and what properties there are in the world.

For those who want to get along with fewer properties, we can

ignore robustness and define a (general) picture as we did above,

replacing [e] by this weaker condition:

[Weak C] (i) Icons i and j properly overlap if and only if

some pictension properly overlaps i and either totally

includes or totally excludes j, or vice versa.

(ii) Icons i and j are disjoint if and only if some pictension

totally includes i and totally excludes j. or vice versa.

(iii) No icon is totally included in the union of any set of

icons not containing it.

126 Understanding Indeterminacy

This condition entails the Leibniz definition of identity in terms of

properties while positing the bare minimum of properties for this

to hold.

Most of the figures in this book do not themselves satisfy either

of the conditions [C] or [Weak C]; this is because the diagrams are

intended to picture a portion of a world. They are to be taken as

partial pictures which can be fteshed out to a picture that satisfies

at least [Weak q, and that contains icons for the rest of the objects

that there are in the world, and contains pictensions for the rest of

the properties.

7.7 PICTURING RESOLUTIONS

Suppose we say that a picture is identity-determinate if none of its

icons properly overlaps. An identity-determinate picture represents

a situation with no indeterminacy of identity among the objects

represented, and an identity-determinate picture of the whole

world represents a world with no indeterminacy of identity at all.

Suppose we say that a picture is precise if each icon consists of

exactly one point. A precise picture represents a situation with no

indeterminacy of any kind, and a precise picture of the whole world

represents a completely determinate world.

There are precise pictures, and there are pictures that are not

precise but are identity-determinate, and there are pictures that are

neither. This corresponds to the idea that the world might be com-

pletely determinate, and it might be indeterminate but contain no

indeterminacy of identity, and it might be indeterminate and also

contain indeterminacy of identity.

In §5.4 we introduced the notion of a resolution of a world. If a

picture satisfies a certain condition (if it "covers all the options"),

it is straightforward to produce from it a picture of a resolution of

the world that the original picture represents. If a picture does not

cover all the options, it is equivalent to one that does. The remain-

der of this section characterizes how to turn any picture into one

that covers all the options, and how to turn a picture that covers

all the options into a "fully refined" picture, which will represent

a resolved world.

Understanding Indeterminacy 127

Say that a picture P covers all the options if and only if this is

satisfied:

For any set s of icons in P for which any two members of s

properly overlap each other, there is a non-empty region r (a

set of points) that is a subset of each icon in s and is disjoint

from every icon not in s.

In pictorial terms, this means that when objects properly overlap,

they produce as many overlapping regions as possible. When only

two icons overlap, this condition is automatically satisfied, but if

there are three or more, it mayor may not hold. For example, if a,

b, and c all properly overlap each other, then Figure 7.16 covers all

the options, and Figures 7.17 and 7.18 pictures do not cover all the

c

FIG. 7.16

128 Understanding Indeterminacy

c

FIG. 7.17

options. (The square region in Figure 7.18 is meant to illustrate the

mutual overlap of the three object icons.) If you just want to rep-

resent what things are like in the world, then it makes no differ-

ence whether you produce a picture that covers all the options or

not. For this is true:

If P is a picture, there is a picture P' that covers all the options

and that represents the same objects as having the same prop-

erties and standing in the same identity/distinctness relations

to one another.

You can always produce a picture that covers all the options from

one that does not by supplying some additional points to use to

make the additional regions needed for option-covering.

Consider pictures that cover all the options. In any such picture,

each icon is partitioned into cells by its overlap with other icons: if

Understanding Indeterminacy 129

c

FIe. 7.18

it overlaps no other icon. it consists of one cell, if it overlaps one

other icon it consists of two cells, if it overlaps two other icons it

consists of three cells or four, depending on whether the other icons

overlap each other. And so on. Given a picture P that covers all the

options, an identity-refinement results from P by replacing each icon

in P with one of its cells. Notice that icons always become more

determinate, not less, in an identity-refinement, and that identities

all become completely determinate. Given an identity-refinement

of P, a full refinement is produced by replacing each icon in the

identity-refinement by the unit set of one of the points in it. In a

full refinement, there is no indeterminacy of any kind.

Recall the resolutions of the world that were discussed in

Chapter 2. The full refinements of a picture correspond exactly to

the resolutions of the world represented by the picture. That is,

130 Understanding Indeterminacy

given a world and a picture that represents it, an object has a prop-

erty in a resolution of that world if and only if its icon is a subset

of the pictension representing that property in the corresponding

full refinement of the picture of the world.

So you can easily see what is true in every resolution of a

pictorial situation by expanding the picture (if necessary) to one

that covers all the options, and then examining every refinement of

that picture. (Refinements are discussed further in the next

chapter.)

7.8 PICTURING COMPLEX PROPERTIES

Little has been said above about complex properties, because

metaphysical issues about indeterminate identity do not appear

to depend on what view is taken about them. The following dis-

cussion concerns how pictures that explicitly represent properties

(by circled regions) implicitly represent complex properties by

geometric combinations of those regions. This discussion is not

relied on elsewhere in the book.

There are two ways in which our pictures are metaphysically

neutral, and this should be kept in mind lest some significance be

attributed to the pictures that our picturing conventions do not

embody. First, suppose we have a picture with two distinct but coex-

tensive pictensions p and q. That is, p and q are distinct sets of

points, but they include and exclude exactly the same object icons.

Condition [C) requires that there be a "complement" to p and q, a

pictension r that includes the icons that p and q exclude and

excludes the icons that p and q include. Suppose that there is only

one such pictension, r. Then p and q can represent distinct proper-

ties not both of which have a unique negation (if property nega-

tion obeys the principle that the negation of the negation of a

property is that property itself). We can picture a situation in which

p has a unique negation, r, but q has no negation: there is no prop-

erty whose negation is q, even though there is a property (namely,

r) which is coextensive with the negation q would have if q had a

negation. As a consequence of this, it is possible to have a picten-

sion, q, whose complement (the set of points not in q) is not itself

a pictension. In terms of our two-dimensional diagrams, this means

Understanding Indeterminacy 131

that if we have a circle in the diagram representing q, we may

not assume that the region outside the circle itself represents the

extension of a property-though some region that includes and

excludes exactly the same object images as that region does, does

represent a property. For the latter reason, it is harmless for most

purposes to assume that the region outside a circle represents a

property.

But the issue is more serious whcn wc consider con.iunctions and

disjunctions of properties. Consider the Figure 7.19. If all the

objects that exist are already pictured in this diagram, then it is

harmless to assume that the region enclosed by the two circles

together (the union of the circles) represents the disjunction of p

and q. After all, we know by condition [C) that there is some region

that encloses (excludes) exactly the same object images as are

enclosed (excluded) by the p region or the q region, and the

union of p and q does this. But the union of two property regions

does not automatically do this. For suppose we have instead

Figure 7.20. Here, the object pictured is totally included in the

union of the two property regions, but it is not included in either

FIG. 7.19

132 Understanding Indeterminacy

FrG. 7.20

region p or region q or both. So this is a diagram in which the union

of two property regions cannot represent the disjunction of those

properties.

Diagrams of this sort are usually easily avoided. Figure 7.21 rep-

resents the same states of affairs, and here we may take the union

of region p and region q to represent the disjunction of p and q.

Under certain conditions, we can always avoid diagrams of the

former sort. Call two pictures equivalent if the icons of one can be

mapped one-one to the icons of the other, and the pictensions of

the one may be mapped one-one to those of the other, such that

an icon in one picture is included in (or excluded from) a picten-

sion of that picture if and only if the correlated icon is included in

(excluded from) the correlated pictension in the other picture.

Suppose that we have a picture that is extensional, in the sense that

there are never two distinct pictensions that include and exclude

exactly the same icons. Then there is always an equivalent picture

in which the complement of any pictension is itself a pictension,

and the union and intersection of two pictensions are themselves

pictensions, and in which an icon is included in (excluded from) the

intersection of two pictensions if and only if it is included in

Understanding Indeterminacy 133

FIG. 7.21

(excluded from) each pictension, and an icon is included in

(excluded from) the union of two pictensions if and only if it is

included in (excluded from) at least one of the pictensions.

Similar results hold under certain natural conditions on non-

extensional pictures, but a further investigation of this phenom-

enon is of no apparent direct interest to our general metaphysical

enterprise, so it will not be discussed further.

8

Counting Objects

Can indeterminately identical objects be counted? Yes, they can.

You count them just like you count objects that are determinately

distinct from every object they are not determinately identical to.

The presence of indeterminacy merely means that questions about

how many objects are such-and-such will sometimes have no deter-

minate answers; this is true even if there is indeterminacy, but no

indeterminacy of identity at all. In some cases there are determi-

nate answers to such questions, and in some cases there are not.

Which cases are which? That is hard to sum up in the abstract. But

in most cases we find that we have pretty solid judgements about

how many objects are such-and-such, or we judge that there ought

not to be an answer. A good theory should agree with these judge-

ments, or it should give us reason to revise them. The theory under

consideration reveals different ways to interpret the claims at issue;

in every case, one of the ways makes the theory agree with the

judgements we are inclined to make.

8.1 TWO SOURCES OF INDETERMINACY

IN COUNTING

There are two ways in which indeterminacy complicates the process

of answering the question 'How many «>'s are there?' One compli-

cation arises when some of the <1>'s are indeterminately identical to

others, so you don't know whether to count them as one or as more

than one. Another complication arises when there are things which

are indeterminately «>; thus it is indeterminate whether you should

count them at all. These two complications are different from one

another, and they need to be addressed differently.

Begin with the complication due to indeterminacy of identity.

Things that

entered

Counting Objects

FIG. 8.1

135

Things that

left

Consider the case of the disrupted person, and imagine that there

is one person looking on, as in Figure 8.1. Suppose you are asked

how many people there are all told. You point to the observer, and

say "one". You point to the person entering, and say "two". You

wait a while, then you point to the person leaving, and you ...

what? The problem is that you must not count this person if you

have already counted them, but you must count him/her if you

haven't counted him/her yet. Since it is indeterminate whether the

person leaving is the person who entered, it is indeterminate

whether this is a person you have already counted. Thus it is inde-

terminate whether you are to count them now. It follows that the

question of how many persons there are all told has no correct

answer. We are not in complete ignorance, of course. We know that

any answer less than "two" is incorrect, since you haven't counted

everyone, and we know that any answer more than "three" is incor-

rect, since you must have counted someone twice. There are an infi-

nite number of incorrect answers, two indeterminate ones, and no

correct ones. This is disconcerting, perhaps, but fairly straightfor-

ward. In fact, in this case it seems clear that these are the right

things to say: any answer less than two or more than three is wrong,

and either "two" or "three" is such that it is indeterminate whether

it is correct.

136 Counting Objects

FIG. 8.2

Now consider a different kind of case: indeterminacy due to

predication. Suppose there are three people, one of whom is deter-

minately bald and two of whom are such that it is indeterminate

whether they are bald (Figure 8.2). You decide to count the people

who are bald. You point to the person who is determinately bald,

and you say "one", and then what? It appears that in this case any

answer less than one or more than three is definitely wrong, but the

answers "one", "two", or "three" should all have indeterminate

truth-value.

Now let us consider a case of predicational indeterminacy com-

bined with indeterminacy of identity. Recall the ships and what

they are like; they are pictured in Figure 8.3. Suppose we ask here

how many ships left port? The structure of the situation is like that

of the bald persons above, and so there apparently should be no

determinate answer. But then what about the judgement, which I

earlier took to be part of the extended data, that exactly one ship

left port? Something like that is correct, but how can it be? I suggest

that we can either count ships that left port, or ships that determi-

nately left port. In normal circumstances we take determinacy for

granted, and there is no relevant difference in our assertions

between ' ~ ' and 'detcrminately ~ ' . Also, since we aim to assert what

Ships

Left

port

Counting Objects 137

atB

FIG. 8.3

is (determinately) true, there is no difference between asserting'S'

and asserting 'determinately S'. But when a clause is embedded

inside an assertion, and when there is a possibility of indetermi-

nacy, then we need to distinguish whether we mean 'determinately

<1>' or just '<1>'. This is the case when we assert 'there are n <1>'s' in a

situation of indeterminacy. In the case at hand, there is a difference

between counting the ships that left port, or the ships that determi-

nately left port. The considerations raised above seem to show that

if we ask the first question, assuming that we do not intend the

qualifier 'determinately',1 then there is no answer. But if we mean

the second, we get an answer. This is because we point at the ship

leaving port and say "one"-because it is determinately leaving

port-and we do not count the other ships because we know that

they are not determinately ships that left port. A natural language

assertion may thus be interpreted either austerely, without an

assumption of determinacy, or with determinacy qualifying any of

I Using 'entered' without a determinacy modifier is not the same as saying 'deter-

minately or indeterminately entered'. The bare predicate 'entered' is neither true

nor false of certain things, and that is why there is a problem in counting; the

predicate 'either determinately or indeterminately entered' is either true or false

of each thing, and it raises no such problem.

138 Counting Objects

its otherwise unqualified parts. I call the former the "unadorned"

reading (the reading unadorned with determinacy) and the latter

the adorned reading.

It follows that when I catalogued the extended data about iden-

tity puzzles in Chapter 1, I used ambiguous terminology, hoping

that the reader would interpret it in the way I intended. I intended

to classify as data the fact that exactly one thing is determinately a

person who entered. In my experience, most people will have inter-

preted me in this way. But some others will not have, and they

should have been troubled by what I was saying there. What I now

hope is that everyone will now consider the ambiguity, and agree

with the data when interpreted in the way in which I intended. It

will turn out below that when there is a case of indeterminacy of

identity, a cardinality statement will typically agree with what I

have called the data when given the adorned reading, but not the

unadorned reading. The theory is thus most plausible if applied to

ordinary judgements with the adorned reading being the default

reading in cases where the theory posits indeterminacy of identity.

I will assume this hereafter. (The baldness case above is not one of

these, and so it is not subject to this default.)

This discussion has taken place at a somewhat informal level. We

need to be precise about whether our language actually works as I

have just said, and there are various interpretations to be surveyed.

The following sections spell this out.

8.2 ANALYSING CARDINALlTY CLAIMS

There are standard techniques for representing finite numerical

claims in terms of quantifiers, connectives, and identity. The sim-

plest and least controversial are judgements of the form 'there are

at least n These are analysable as follows:

11Iere is at least one

There are at least two 3x3y(x 7= y & &

There are at least three 3x3y3z(x 7= y & x 7= Z & y 7= Z

& & &

What about 'there are at most n The usual formulation of this

is roughly that any 17 + 1 are such that some of them are iden-

tical. That is, we have these analyses:

Counting Objects 139

There is at most one cj>:

There are at most two cj>'s:

'v'x'v'y(cj>x & cj>y ~ x = y)

'v'x'v'y'v'z(cj>x & cj>y & cj>x ~

x = y v y = z v x = z)

'There are exactly n cj>'s' is equivalent to the conjunction: 'there are

at least n cj>'s and there are at most n cj>'s'. Combining this with the

above analyses yields:

There is exactly one cj>:

There are exactly two cj>'s:

3xcj>x & 'v'x'v'y(cj>x & cj>y ~

x=y)

3x3y(x:;t y & cj>x & cj>y) &

'v'x'v'y'v' z( cj>x & cj>y & cj>x ~ x = y

v y = z v x = z)

These analyses are natural hypotheses about the meaning of car-

dinality claims. If we use them, then the theory under considera-

tion gives the right answers when they exist, and correctly entails

that there are no answers in the other cases. I detail this below. We

will also need to look at alternative formulations that give slightly

different results.

8.3 COUNTING ALL THE SHIPS

Consider the puzzling ships. Our na"ive judgement is that exactly

one ship left port and exactly two ships docked. But how many

ships were there all told? Not fewer than two, since two ships

docked, and not more than three, since there are none under con-

sideration except for the one that left port and the two that docked.

So any answer that is less than two or more than three should be

definitely wrong. Were there exactly two? That seems impossible

to answer; since the world does not see fit to determine whether

either of the later ships are the original ship, there should be no

answer. (If your instinct tells you that the right answer should be

"exactly two", see §8.S for a possible account of that instinct.) Were

there exactly three? Same response. Were there two or three? Here

is where the idiosyncrasies of disjunction creep in. In one sense, if

it is wrong to say there were two, and wrong to say there were three,

then it is wrong to say that there were two or three. In another

sense, we seem to want to say this is right. At least, we want to say

140 Counting Objects

something like "two-ta-three is the right range of answers". This,

too, will be addressed in §8.S.

Setting aside disjunctions, the theory confirms our naive judge-

ments about cardinalities. That is, it confirms them on the assump-

tion that it is indeterminate whether the original ship is the newly

assembled ship, and indeterminate whether it is the ship with new

parts, and determinate that the newly assembled ship is distinct

from the ship with new parts.

I begin with the question of how many ships there are all told.

The answer according to the theory is whatever we get by feeding

the extended data into the numerical formulas above. Here are the

relevant data. Let 0 be the original ship, a the newly assembled

ship, and p the ship with new parts. First, 0, a, and p are ships under

consideration:

D1 So & Sa & Sp.

And these are determinately all the ships that are under considera-

tion:

D2 Vx(Sx => !x == 0 v !x == a v !x == p).

Ship a is not ship p, and it is indeterminate whether 0 is either of

them:

D3 --,(a == p)

'Vo == a

'Vo == p.

D1-D3 then generate the answers to the "how many" question.

(The arguments that follow are elementary but a bit lengthy. They

are not much longer, however, than the same sorts of arguments

that are needed when there is no indeterminacy at all.)

Exactly one? It is false that there is exactly one ship, since it is

false that there is at most one. If it were not false that there is at

most one, this would not be false:

VxVy(Sx & Sy => x == y).

So every instance of it would not be false, including this one:

Sa & Sp => a == p.

But by D1, these are true:

Sa & Sp.

This entails that 'a == p' is not false, contrary to 03.

Counting Objects 141

Exactly two? It is indeterminate whether there are exactly two,

since it is true that there are at least two (infer '-,(a = p) & Sa &

Sp' from D1 and D3 and existentially generalize it), and indeter-

minate whether there are at most two. We can show that it is inde-

terminate whether there are at most two by refuting the other two

options: that it is true that there are at most two, and that it is false

that there are at most two.

Suppose it is true that there are at most two ships. That is, this is

true:

'ifx'ify'ifz(Sx & Sy & Sx => x = y v y = z v x = z).

This, together with D1 lets us infer that this is true:

o = a v a = p v 0 = p.

But D3 rules out the possibility that any of the disjuncts are true.

Suppose on the other hand that it is false that there are at most

two ships; this is false:

'ifx'ify'ifz(Sx & Sy & Sz => x:=: y v x = Z v y = z).

Then some instance is false. So there are entities e, f, and g, such

that this is true:

Se & Sf & Sg & e"# f & e"# g & f"# g.

From this and D2 we infer each of:

e=ove=ave=p

f=ovf=avf=p

g = 0 v g = a v g = p.

These yield six cases to consider (actually, twenty-seven cases, but

twenty-one are immediately ruled out by the non-identities above),

with each of them contradicting D3. For example, one of the cases is:

e = 0 & f = a & g = p.

This, with 'e"# f' from the above six-part conjunction, entails that

this is true:

contradicting:

'10 =a

from D3. The other five cases are parallel.

142 Counting Objects

Exactly three? It is indeterminate whether there are exactly

three, since it is indeterminate whether there are at least three (this

proof is similar to the proof that it is indeterminate whether there

are at most two) and true that there are at most four (proof left to

reader).

Exactly four? It is easily shown (using D3) that it is false that

there are at least four ships, and this is sufficient to refute the claims

that there are exactly four, exactly five, etc.

8.4 COUNTING THE SHIPS THAT LEFT PORT

When we counted ships above, we assigned a cardinality to the

extension of a predicate, 'ship', that is determinately true or deter-

minately false of each thing. As discussed earlier, new complica-

tions enter when this condition is not met. Suppose we ask how

many ships left port. Certainly the original ship left port; that is

determinately true. Since it is indeterminate whether the ship with

new parts is the original ship, it is indeterminate whether the ship

with new parts left port; likewise for the newly assembled ship.

(These inferences presume that leaving port is, or is equivalent to,

a property; they also presume that if either of the ships that docked

had left port, that ship would be the original ship. Each of these is

an additional assumption that may be questioned.) It appears then

that there might be no determinate answer regarding how many

ships left port. But I suggested above that exactly one ship left port.

Which is right?

On the analysis that we have so far, exactly one ship determi-

nately left port. But there is no answer to the "unadorned" ques-

tion of how many ships left port. To verify these claims we need

additional data about which ships left port. As indicated earlier, I

assume that 0 left port, and it is indeterminate whether a or p did:

D4 La & vLa & v Lp.

To show that exactly one ship determinately left port we need to

show that this is true:

2

2 By D4 it is determinate for each entity under discussion whether or not it is a

ship. Thus 'Sx' and '!Sx' are equivalent, and we need only consider one version; I

use the simpler.

Counting Objects 143

:Jx!(Sx & Lx) & 'lfx'lfy(!(Sx & !Lx) & !(Sy & !Ly) =:> x = y).

The first conjunct is easy since 01 and 04 yield that this is true:

So & La.

Since it is true, we can add the required'!', and existentially general-

ize to get the first conjunct. To see that the second conjunct is true,

we need to show that any instance that has a true antecedent also

has a true consequent, and that any instance that has an indetermi-

nate antecedent has a true or indeterminate consequent. Suppose,

first, that we have an instance where the antecedent is true:

!(Se & !Le) & !(Sf & !Lt).

By 02 we have both of:

e=ove=ave=p

f = 0 v f = a v f = p.

We can rule out any of these disjuncts being true except the first

on each line, because any of the others, by Leibniz's Law, would

lead from the displayed conjunction to theses that contradict 04.

But the first disjuncts taken together entail 'e = f', which makes the

consequent of the conditional true. It remains to show that any

instance that has an indeterminate antecedent has a true or inde-

terminate consequent. This is vacuously true, since the antecedent

cannot be indeterminate:

!(Se & !Le) & !(Sf & !Lt).

This is determinate in virtue of its form (because of the determi-

nate-truth connective).

If we interpret the question as how many ships left port, omitting

the determinacy modifier, then the proof would go the same right up

to the last step. But at that point the antecedent would be indeter-

minate if e = a and f = p; in that case the consequent is false, and so

the whole conditional is indeterminate. So on this reading it is inde-

terminate whether exactly one ship left port. This is the kind of case

that rationalizes the decision in §8.1 to take the adorned reading

(the one that includes the determinacy modifier) as the default

reading for claims involving indeterminate identity?

, See the appendix to this chapter for a comparison of these results with the ones

one would get on a classical analysis of the situation using supervaluations.

144 Counting Objects

8.5 VARIANT ANALYSES

In bivalent logic there are many different equivalent formulations

of cardinality claims. It is important to note that some of these are

not equivalent in a non-bivalent setting. In fact, there is a fairly

natural way to formulate 'there is exactly one <jl' which is plainly

wrong. It is the formulation that says 'something is <jl, and any <jl is

it':

:Jx[<jlx & Vy(<jly:::::} x = y)].

To see that this is wrong, notice that in the ship case this tells us

that it is indeterminate whether there is exactly one ship. The

formula that says there is exactly one ship is nonfalse because there

is something-the original ship-which is determinately a ship, and

such that every ship is either determinately or indeterminately

identical with it. But it is plainly false that there is exactly one ship,

since two distinct ships docked at the end of the trip. So this seem-

ingly natural formulation that is equivalent in bivalent logic to the

one we have used above is not equivalent in non-bivalent logic.

4

This will be important in assessing how definite descriptions work

in the next chapter.

8.6 SUPER-RESOLUTIONAL READINGS

There is no correct answer to the question about how many ships

there are all told; all answers except 'two' and 'three' are false, and

those are indeterminate. Thus this is also indeterminate:

There are two or three ships.

4 The equivalence in bivalent logic depends on the transitivity of identity. In the

present application, this turns into a case of the transitivity of indeterminate iden-

tity. But indeterminate identity is not transitive. The ship case refutes the transitiv-

ity of indeterminate identity, since one of the later ships is indeterminately identical

to the original ship, which is indeterminately identical to the other later ship, but

the later ships are distinct from one another.

Counting Objects 145

But some of us have some inclination to endorse this claim. This

appears to be a matter of interpreting the disjunction in accord

with the super-resolutional reading discussed in Chapters 3 and 5.

Consider all ways of making properties more determinate; then

a list of options is correct on the resolutional reading if and only

if each option is made true on some resolution, and every resolu-

tion does make one of the options true. In the case of the ships, any

resolution identifies ship 0 with ship a, or with ship p, or with

neither; the first two options produce two ships, and the last

produces three. So this kind of reading validates the assertion

that there are two or three ships, while leaving each disjunct

indeterminate.

There are people whose instincts tell them that there should

be exactly two ships all told. When asked to explain why they

think this, they tend to answer that however things go, the original

ship must be identical to one of the resulting ships; it is indeter-

minate which it is identical to, but there can be no third option

where the original ship just ceases to exist. I have not taken

this to be part of the extended data (because J think it is false),

and the theory alone does not yield this as an option. But it is

possible to accommodate it. If indeed it is genuinely impossible

for the original ship to cease to exist while being replaced by two

distinct completely new ships, then the resolution that turns out

that way is not a possible way the world could go. If that is so, it

should not be one of the resolutions that we count when evalu-

ating the truth of a claim on all resolutions. If it is ruled out as

a resolution, then the super-resolutional reading does indeed say

that there are exactly two ships overall. This is an automatic

product of the theory sketched here coupled with the claim that

it is impossible for the original ship to cease to exist and be

replaced by two others. I don't agree with that claim about

what the possibilities are, but if I did, I would say that the theory

correctly accommodates my view on the super-resolutional

reading.

s

5 A similar point applies to the question of how many ships left port. The

super-resolutional reading by itself leaves this indeterminate between one, two,

and three. But the answer is determinately one if you add that it is impossible

for either of the later ships to leave port while being distinct from the original

ship.

146 Counting Objects

8.7 PERSONS, CATS, AND PILES

OF TRASH, ETC.

The material discussed above carries over straightforwardly to the

other paradigm puzzles. The results are:

Me and my body: Suppose it is determinate that both I and my

body are in the office, and nothing else is there. Then there are

not less than one nor more than two things in the office; it is

indeterminate whether there is one and indeterminate whether

there is two. In the super-resolutional reading, there are either

one or two. For the unadorned reading of how many persons

there are in the office, we get the same answer, but for the

determinate-person reading we get the answer "exactly one".

The disrupted person(s): Exactly one person entered the room

in either the determinacy or the unadorned sense. It is false

that there were less than one or more than two persons, but

both 'one' and 'two' are indeterminate as answers. On the

resolved reading there are either one or two.

The cat: There is exactly one cat on the table in the determi-

nacy sense, which is what I consider to be the default case;

there is no answer on the unadorned reading.

The piles of trash: There is not less than one pile of trash nor

more than three; there is no determinate literal answer, but in

the resolved sense there are between one and three piles of

trash.

A more complicated case: Suppose there are two similar ships, a

and b, afloat near one another. Each ship has its planks replaced

with planks from the other, until the planks are completely inter-

changed, and we end up with a ship c whose location is continuous

with that of a's, which consists entirely of the planks from b, and a

ship d whose location is continuous with that of b's, which consists

entirely of the planks from a. One might want to argue that there

is no fact of the matter whether a is identical with c or with d, and

likewise for b. The picture would look like Figure 8.4. There are at

least two ships here, and not more than four, but it is indetermi-

nate whether there are two and indeterminate whether there are

three and indeterminate whether there are four. On the resolved

Counting Objects 147

Ships

b

FIG. 8.4

reading there are either two or three or four. (One might want to

argue on the basis of the symmetry of the example, that the reso-

lution into three is impossible. If we impose a constraint on the res-

olutions that they be symmetrical, then wc would conclude that

there are at least two ships, and not more then four, and not exactly

three; it would be indeterminate whether there are two and inde-

terminate whether there are four.)

Appendix: A Classical Analysis

Using Supervaluations

A referee for OUP has suggested a comparison of these results with the

ones one would get on a classical analysis of the situation using superval-

uations. The answer is that a variety of answers are obtainable, depending

148 Counting Objects

on how the supervaJuation technique is applied. Two sorts of choices must

be made in any application. First, there is the ontological one: since there

is no worldly indeterminacy of identity, one must decide what there is in

the world on which to base the extended valuations. A typical choice is a

mereological one, where the atomic ingredients of the situation under dis-

cussion are:

A: the object that coincides with the original ship up to the recon-

struction

B: the object that coincides with the ship with new parts starting from

reconstruction

C: the object that coincides with the ship with old parts starting from

reconstruction

There are also the mereological sums of these: AB, AC, BC, and ABC.

The second choice involves which are the acceptable extended valua-

tions of the predicates in question. For simplicity, I assume that on any

extended valuation the predicate 'left port' is true of A and of any sum

that includes A. (Without this constraint, we get many more options than

those detailed below.) The interesting question is what constraints to

impose on the extended valuations of 'ship'. For simplicity, I assume that

both BC and ABC are ruled out as potential ships, since they each require

there to be a ship that is in two places at once. The valuations then must

be chosen from among the following as possible extensions for 'ship':

<1>, IAj, IBj, Iq, IABj, IAq, lA, Bj, lA, C), lA, ABj, lA, Aq, IB, q,

IB, ABj, IB, AC), IC, ABj, IC, BC), lAB, AC), lA, B, C), lA, B, ABj, lA,

B, Aq, lA, C, ABj, lA, C, BC), lA, AB, Bq, IB, C, ABj, IB, C, Bq,

IB,AB,Aq, IC,AB, Bq, lA, B, C,ABj, IA,B,C, Bq, lA, B,AB, Bq,

lA, C, AB, BC), IB, C, AB, BC). lA, B, C, AB, BC)

If all of these are allowed (an implausible application) then each exact car-

dinality judgement for ships leaving port from zero to five is superinde-

terminate, and those from six up are superfalse. A more plausible option

is to require that some ship left port, and that at least two docked; this

reduces the possible extended valuations of 'ship' to:

(C, AB), (B, Aq, [AB, Aq, lA, B, C), [A, B, Aq, [A, C, AB], [A, C,

Bq, (A. AB, BC), IB, C, AB], IB, AB, AC), IC, AB, BC), lA, B, C, AB},

lA, B, C, BC), lA, B, AB, BC), lA, C, AB, Bq, (B, C, AB, BC), (A, B,

C,AB,BC)

With these options it is superfalse that no ship left port, and each exact

cardinality judgement for ships leaving port from one to five is superinde-

terminate. One might wish to also insist that on any extended valuation no

two ships ever overlap. This would reduce the extended valuations to:

Counting Objects 149

[A, B, Cl, [AB, C], [AC, BJ

Now it is supertrue that exactly one ship left port.

Instead of the above choices, one might allow ships to overlap but insist

that no ship that left port ceases to exist midway in the voyage; if this is

added to the edict that no ship is ever in two places, the possible extended

valuations are:

[AB, C], [AC, Bl, [AB, AC]

On this option it is superfalse that no ship left port, sllpcrindeterminate

whether one, or whether two, left port, and superfalse that three or more

left port.

(I suppose that one should not really rule out the option that there was

never more than one ship, namely, ABC. That leaves the last two options

just discussed unchanged for the question of how many ships left port,

though it would change thc results for the judgement about how many

ships docked.)

9

Denoting Objects

9.1 THE ISSUE

Suppose it is indeterminate whether a is identical to b. Can we then

refer to a without, in some sense, also referring to b? If not, since

reference is supposed to be unique, can we refer to a at all? This

sort of issue arises regularly in the literature. There are two dimen-

sions to the worry. The first is that if indeterminacy of identity pre-

vents unique reference, we may not be able to make determinate

claims about any object at all if it is indeterminately identical with

some object. The second is that if indeterminacy of identity auto-

matically brings with it indeterminacy of reference, it would seem

difficult to claim that there is indeterminacy in the world, as

opposed to it being an illusion generated by vagaries in the seman-

tical relations between language and the world. In this chapter I

address the question of how we may refer determinately to objects

that are indeterminately identical to "other" objects. In the next

chapter I examine the proposal that all indeterminacy of identity

is an illusion based on semantic indeterminacy.

It is easy to generate worries about denotation when objects are

indeterminately identical. Here is an example from Burgess (1990)

about a mountain Aphla which is (allegedly) indeterminately iden-

tical with a mountain Ateb. He suggests that it is wrong to hold that

the singular term 'Aphla' can denote a unique object in such a case.

He argues as follows:

If there is something with which Aphla is indefinitely identical, then there

is no (unique) thing which Aphla is. We cannot gloss '''Aphla' denotes

Aphla" as" 'Aphla' denotes Aphla, whatever it may be", we should have

to say" 'Aphla' denotes Aphla, whatever unique thing it may become". This

gives the game away, implying as it does, that there is no unique thing which

Aphla is already. If there is no unique thing which Aphla is already, then

Denoting Objects 151

the sense of 'Aphla' is not yet sufficiently definite to determine a unique

object for Aphla to be. That the sense of the name could be modified to

overcome this difficulty is no excuse for holding that we can regard this

task as already having been accomplished. The prolepsis involved in the

manceuvre is unintelligible. (1990: 271)

If Aphla is indeterminately identical with Ateb, then apparently

there is no unique thing that we denote by 'Aphla'. A similar point

is also articulated by van Inwagen (1988):

[I]f identity is indeed vague, then the semantical relation between name

and thing named must also be vague. If, for example, 'Alpha' definitely

names x, and it is neither definitely true nor definitely false that x = y, then

it seems inevitable to suppose that it is neither definitely true nor definitely

false that 'Alpha' names y. Our semantics must somehow reflect this

consequence of vague identity for the naming relation. (1988: 259)

Cowles (1994) goes further; in commenting on this passage from

Van Inwagen he says:

I think that van Inwagen is wrong to suggest that 'Alpha' could definitely

name x even when it is neither definitely true nor definitely false that x =

y. (1994: 147)

Determinate reference presupposes determinate individuation.ll1e case

of Alpha/Omega [the disrupted person(s)], however, is a case in which

determinate individuation is ex hypothesi absent. (ibid. 153)

Stalnaker (1988) makes a related point. Commenting on Salmon's

argument against indeterminate identity (Chapters 4 and 11), he

says:

[T]he argument shows that if it is indeterminate whether a = b, then it is

indeterminate what 'a' refers to, or what 'b' refers to. (1988: 354)

These comments raise severe questions about what objects we are

talking about when we use singular terms. The questions need to

be addressed.

9.2 NAMES

Suppose we see a ship leave port on Tuesday, and we dub it 'Saman-

tha's Pride'. On Wednesday we see a ship dock, and we dub it

'Kim4Ever'. (We also see another ship dock, which we dub some-

thing else.) Then we discover that in between there occurred the

152 Denoting Objects

sort of repair/assembly that leads us to conclude that there is no

fact of the matter about whether Samantha's Pride is Kim4Ever.

Suppose that we try to explain part of the situation by saying

that it is indeterminate whether Kim4Ever left port on Tuesday.

Are we then speaking indeterminately about Samantha's Pride?

If so, what are we saying about it? Are we saying indeterminately

that it is indeterminate whether Samantha's Pride left port on

Tuesday? But it is determinate that Samantha's Pride left port. It

seems wrong to say of Samantha's Pride, even indeterminately,

that it is indeterminate whether it left port on Tuesday. Is it then

indeterminate whether we have said something wrong? I hope not.

I think that I can say that it is determinate that Samantha's Pride

left port on Tuesday, and indeterminate whether Kim4Ever left

port on Tuesday, without even being indeterminately wrong.

The key question is this: If it is indeterminate whether Kim4Ever

is Samantha's Pride, have I indeterminately dubbed Samantha's

Pride 'Kim4Ever' when I dub Kim4Ever 'Kim4Ever'? I don't think

so. I have dubbed Kim4Ever 'Kim4Ever', and as a result the name

'Kim4Ever' as used by me determinately refers to Kim4Ever and

determinately does not refer to Samantha's Pride. And so I am not

speaking of Samantha's Pride at all, even indeterminately.

At least this is how 1 think I use the name. This is in spite of an

obvious argument to the contrary. Suppose that I (determinately)

use the name 'Kim4Ever' to refer to Kim4Ever, and (determi-

nately) do not use the name to refer to Samantha's Pride. Then it

seems that by Leibniz's Law, Kim4Ever and Samantha's Pride are

distinct, in spite of the fact that it is indeterminate (by hypothesis)

whether they are identical. Fortunately, this does not follow. This

is reasoning not by Leibniz's Law, but by a contrapositive version

of it. The contrapositive version holds only for contexts in which

the predicates stand for properties. But conceptual or semantic

predicates do not necessarily stand for worldly properties; such

predicates instead characterize parts of our conceptual apparatus

for representing the world. 'N refers to x' is a paradigm case of a

predicate that does not stand for a property of x.

What about the question raised by Burgess: Is there a unique

thing which Kim4Ever is, and which we denote by 'Kim4Ever'?

What does it mean to ask if there is a "unique thing"? A thing

is unique in some respect or other. Apparently what is being

addressed here are the following two questions about uniqueness:

Denoting Objects

Is there a unique thing which is Kim4Ever? That is, is there

exactly one thing which has the property, being Kim4Ever?

Is there a unique thing which is denoted by 'Kim4Ever'? That

is, is there exactly one thing which is denoted by 'Kim4Ever'?

The answers to these are implicit in the results of the last chapter.

There are two ways to read questions like these: unadorned, and

with the determinacy interpretation. On the unadorned reading,

the question sometimes has no answer; it has no answer because it

is indeterminate whether Kim4Ever is identical with Samantha's

Pride. On the determinacy reading it is answered yes. There is

exactly one thing which is determinately Kim4Ever.

As I use the name, the second question can also be answered

"yes". There is exactly one thing denoted by the name 'Kim4Ever';

it is the exactly-one-thing that answers to the determinacy reading

of the first question. You can reason otherwise by using a fallacious

contrapositive version of Leibniz's Law, but not otherwise.

This is also the reply to Van Inwagen's comments. If 'Kim4Ever'

definitely names Kim4Ever, and if it is indeterminate whether

Kim4Ever is Samantha's Pride, then it does not follow that it is

indeterminate whether 'Kim4Ever' names Samantha's Pride. Not

only does it not follow, it is not true.

The position I take here is a consistent one, but hardly one that

will convince someone who is wedded to the opposite view. Can I

do anything to win over such a person? Perhaps not if we discuss

only proper names. The semantics of proper names is controver-

sial, partly because the facts are so undetermined. The names them-

selves have no semantic structure at all, and so we need to look at

our usage and our intent. And in the area of indeterminacy, these

give us little uncontroversial guidance. Perhaps we can do better

by looking at something with more structure? With this hope, 1 turn

to a consideration of definite descriptions.

9.3 DEFINITE DESCRIPTIONS

Theories of definite descriptions fall into roughly two categories.

One, endorsed by Russell, is that definite descriptions are not

semantical units in themselves, and so they have no semantic

154 Denoting Objects

analysis. Instead, one analyses sentences that paraphrase the

descriptions; instead of trying to analyse 'the present queen of

England' in 'The present queen of England is old' one merely

analyses 'There is exactly one queen of England, and she is old.' It

is clear from the discussion of the preceding chapter that sentences

of this sort will often be determinately true if given the determi-

nacy interpretation even if the queen of England has previously

undergone a disruption that makes her indeterminately identical

with the former princess. That is, in the circumstances described,

this may be true:

3x[!(x is OE) & \ix\iy[!(x is OE) & !(y is OE) ~ x == y] &

x is old].

However, it may be worth discussing also those views that take

definite descriptions to be genuine semantical units, since this is the

area in which the discussion of semantic indeterminacy is often

located. There are dozens of theories of this sort, but they almost

all agree on the condition under which a definite description

denotes something, and what it denotes. The condition is:

1

[DJ: 'lX[<»X]' denotes 0 if and only if 0 is <» and at most one thing

is <».

In this condition, the 'if and only if' is to be read strongly (it is to

be read as our '<:::>'), in the sense that this is true, or false, or neither

true nor false:

'lX[<»X]' denotes 0

if and only if this is true, or false, or neither true nor false

respectively:

<»0 & \ix\iy[<»x & <»y ~ x == y].

For definiteness in discussing how this works, J focus on the case of

the disrupted person(s), and the definite description 'the person

who entered'. The metaphysical situation is the one pictured in

Figure 9.1. The question is whether the description 'the person who

entered' denotes a determinately, or indeterminately, and in either

case whether it also denotes h indeterminately. As in the previous

1 The second condition uses 'at most one thing is <\l' instead of 'exactly one thing

is <\l' because the former is simpler, and they are equivalent when conjoined with

'0 is <\l'. It also avoids the natural but wrong anaJysis discussed in §8.S.

Things that

entered

Denoting Objects 155

b

Persons

FIG. 9.1

chapter, there are two interpretations to consider: the unadorned

reading, and 'the thing that is determinately a person who entered'.

Between them they seem to capture the things people want to say

about cases of this sort.

9.3.1 Determinate denotation of a without indeterminate

denotation of b

First, consider the determinacy reading of ' the person who entered':

tx!(x is a person & x entered).

I say that this determinately denotes person a in Figure 9.1. It

denotes person a because this is true:

!(a is a person & a entered) & 'v'x'v'y[!(x is a person & x

entered) & !(y is a person & y entered) ::=} x = y].

The first conjunct is true since both 'a is a person' and 'a entered'

are determinately true. The second conjunct is true because it is a

universally quantified conditional with all true instances. Instanti-

ating both variables with a validates both antecedent and con-

sequent. Instantiating either variable with b or with anything

156 Denoting Objects

determinately different from a makes it true since this produces

a false antecedent (because only a is determinately a person that

entered).

Does the definite description denote b as well? No, it does not.

H determinately does not denote b because this is false:

!(b is a person & b entered) & 'lix'liy[!(x is a person & x

entered) & !(y is a person & y entered) ~ x = y].

It is false because the first conjunct is false; this is because of the

initial '1' which requires that both of the component conjuncts be

true, and 'b entered' lacks truth-value. So the definite description

determinately does not denote b. It is easy to see that it denotes

nothing else (except a) too, for the same reason. So here is a defi-

nite description that determinately denotes a and that does not

indeterminately denote anything at all. So there is such a thing as

determinate denotation unaffected by indeterminate denotation

even when the term determinately denotes a thing that is indeter-

minately identical to something.

So indeterminate denotation is not forced on us by every term

denoting a thing that is indeterminately identical with something.

But there is indeterminate denotation as well.

9.3.2 Determinate denotation of a with indeterminate

denotation of b

On the unadorned reading, 'the person who entered' determinately

denotes a but also indeterminately denotes b, just as the critics say.

Consider first a. The truth-value status of:

'u[x is a person & x entered], denotes a

is the same as the truth-value status of:

[a is a person & a entered] & 'lix'liy[x is a person & x entered

& y is a person & y entered ~ x == y].

The first conjunct of this is true, so consider the second. This is a

universal claim, and it requires all of its instances to be true for it

to be true. Clearly it is true if we instantiate the same thing for both

Denoting Objects 157

variables, since this makes the consequent true. Likewise it is true

for any instantiation other than a or b, since all such instances make

the antecedent false. The remaining option is to instantiate a for

one variable and b for the other; this makes the conditional true

since it makes both antecedent and consequent indeterminate, and

the ' ~ ' connective is true in such a case. (So this is an instance in

which our formulation of condition [0] may be controversial; see

below.) Thus the definite description determinately denotes a.

The description indeterminately denotes b. This is because this

is indeterminate in truth-value:

[b is a person & b entered] & 'v'x'v'y[x is a person & x entered

& y is a person & y entered ~ x == y].

The first conjunct is indeterminate because it is indeterminate

whether b entered. The second is true as argued above. So the

whole is without truth-value; it is indeterminate whether the defi-

nite description denotes b.

It is easy to show that the definite description detenninately does

not denote anything other than a or b. So the definite description

determinately denotes a, indeterminately denotes b, and denotes

nothing else.

This conclusion is dependent on a particular reading of condi-

tion [0], one that employs o u r ' ~ ' connective to symbolize 'nothing

other than z is (j>'. Clearly, this might be symbolized differently; for

example, it could be symbolized with the material conditional

(using '(j> ::J \V' for' ---,(j> v \V'):

[a is a person & a entered] & 'Vx'Vy[x is a person & x entered

& y is a person & y entered::J x = y].

If this is used in condition [0 1 then it alters the judgement of how

the definite description with the unadorned reading denotes a. It

now denotes a indeterminately instead of determinately, because

when the second conjunct is instantiated so as to yield a conditional

whose antecedent and consequent are both indeterminate, the

whole material conditional lacks truth-value. With this change

it turns out that the definite description indeterminately denotes

both a and b. (This change has no effect on the determinacy

reading, where the definite description is 'tx![x is a person & x

entered]'. )

158 Denoting Objects

There thus appear to be three reasonable ways to interpret the

English phrase 'the person who entered the room'. On one inter-

pretation it determinately denotes a and does not denote anything

indeterminately. On a second interpretation it determinately

denotes a and indeterminately denotes b. On a third interpretation

it indeterminately denotes both a and b.

9.4 SUMMARY OF THE VIEW

BEING DEFENDED

We have seen that there are clear cases in which a definite descrip-

tion determinately denotes an object x and determinately does not

denote an object y even though it is indeterminate whether x is y.

There are also cases in which indeterminacy of identity leads to

indeterminacy of what is denoted, perhaps-or perhaps not--even

indeterminacy coupled with determinacy. This has been shown by

considering definite descriptions; proper names are harder to draw

clear conclusions about, but I see no reason why they cannot do

what the definite descriptions do. (None of the examples 1 consid-

ered had to do with scope distinctions or with modalities, which are

the two main areas in which names and descriptions are alleged to

behave essentially differently.)

I take this to vindicate my own policy in formulating the theory

of indeterminate identity. I have presumed that each of the

singular terms in the theory I formulate determinately denotes

something, and does not indeterminately denote anything. I have

assumed this so that I could present the metaphysical issues

without becoming entangled in issues of semantic indeterminacy.

Since it is possible for singular terms to behave in this way, I am

justified in presuming that mine do. I will continue with this

policy.

It would be fascinating to pursue a study of the logic and seman-

tics of indeterminate denotation, expanding the language intro-

duced in Chapter 2 to include singular terms that do not denote

determinately. However, this would be an exercise in semantics or

logic that is distinct from the metaphysical issues addressed in this

book. Our issues are difficult enough already without these addi-

tional complications. For this reason, J will not introduce indeter-

Denoting Objects 159

minate denotation into the official language of the theory, and I will

not speak of it any more.

There remains the question of how to interpret data that is for-

mulated in natural language. When we report on puzzling cases, we

use wordings that come naturally to us. And when we consider that

some statements lack truth-value, we may do so pre-theoretically,

without being certain about why. So in considering the "data" I may

not be certain that the singular terms used in its formulation

uniquely refer. I do think that it is justifiable to assume this prima

facie, at least if there is no reason to believe otherwise, and this is

the policy I have followed so far. In the next chapter, I consider

some challenges to this policy.

10

Alternatives to Indeterminate Identity

Perhaps the most popular view about indeterminacy of identity is

that it is an illusion generated by inexactness in our language, or

in the system of concepts embodied in our language. The world

appears indeterminate because our linguistic/conceptual apparatus

does not yield determinate judgements, but the lack of determinacy

lies in the apparatus itself, not in the world it purports to articulate.

The point of this chapter is to investigate such ideas.

Proposals of this sort are often stated, and even more often

endorsed, but they are rarely implemented in any detail. In my

experience, they sound more plausible in the abstract than in the

concrete. It is a central theme of our heritage of work in twentieth-

century analytic philosophy that questions apparently about reality

are to be reconstrued as questions about the language with which

we describe such reality, and so a proposal to do that has great

instinctive appeal. But in any given case, such a proposal is still

merely a hypothesis, not a given fact, and it deserves to be exam-

ined like any other hypothesis. That is the point of this chapter.

It takes little reflection to see that some work needs to be done

in order to have a view worth examining. There needs to be a pos-

itive doctrine spelled out about how our language works, and then

about how it applies to the problematic situations. Further, it needs

to either account for the (extended) data with which we begin, or

show us how those data are wrong. For the latter purpose, it can't

just be an announcement that the opinions with which we begin

are naive and inherently untrustworthy. The opinions with which

we begin are our own; they are the principles upon which we base

all our decisions, and if we are to reject some of them it must be

on the basis of others of them, not on the basis of a general charge

of naivety. But then the work needs to be done to show which are

which.

The theme of this chapter will be that theories that account for

Alternatives to Indeterminate Identity 161

the traditional identity puzzles in terms of the nature of language,

or of concepts, are not compelling in their present forms. Of course,

there are identity statements that lack truth-value because of the

nature of language, but the identities in the traditional puzzle cases

are not of this sort. I will not attempt to prove that one could not

give a plausible account of this sort, because I cannot. My goal is

instead to establish the present absence of settled accounts of this

kind. Such accounts should be developed further in the face of

criticisms raised here, but it is appropriate that such development

be undertaken by those who believe in them. I will not try to carry

this out myself.

The available pieces of language that need clarifying are singu-

lar terms, the identity predicate itself, and general terms with which

the identity puzzles are stated. I consider them in that order. In

each case we need to see how it is that our use of language can

account for the data that lead to the identity puzzles, and how

understanding that language dissolves the puzzles while preserving

the data that lead to them.

10.1 CLARIFYING DENOTATION

WITH SUPERVALUATIONS

A view that seems quite natural to many writers supposes that inde-

terminacy of identity is an illusion fostered by the fact that one or

both of the singular terms composing the identity statement has

more than one potential referent. There is at least one thing that is a

potential referent of both the terms in common, and at least one that

is a potential referent of one term but not of the other. Since at least

one singular term lacks a unique referent, the identity statement

itself is essentially ambiguous, and it has no unique reading which

could yield a single truth-value. The statements that articulate the

data are affected in this way too, because of the lack of unique

reference of their terms, but if they are given supervaluational

readings, they come out true. They come out true because they are

true no matter how the references of the terms are clarified. This is

not so for the identity statement; some valuations (some ways of

singling out unique referents for the terms from among their multi-

ple potential referents) make it true and some make it false, and so

162 Alternatives to Indeterminate Identity

it lacks supervaluational truth-value. Thus we have a truth-value

gap for an identity statement, but without any indeterminacy of

identity in the world; the truth-value gap is simply a consequence

of semantic variation in the potential reference of the terms.

Here is an illustration of how this might work in a transparent

case. Suppose that Old Ivy Lodge, which is on Main Street, and

Structuralist Manor, which is around the corner on Cedar Avenue,

have been incorporated into a much larger structure, PostModern

Hulk, which occupies the corner of Main and Cedar. But tradition

is strong, and the old designations are still in use. Suppose that

Smith has an office in Old Ivy Lodge (and therefore in PostMod-

ern Hulk) and Jones has an office in Structuralist Manor (and thus

in PostModern Hulk). We are naturally inclined to assent to the

claim that the building in which Smith has her office is on Main

Street, and to the claim that the building in which Jones has his

office is on Cedar Avenue, but it seems that there is no answer to

the question whether the building in which Smith has her office is

the building in which Jones has his office.

The supervaluation approach to singular terms accounts for

these judgements as follows. Let'S' abbreviate 'the building in

which Smith has her office' and 'J' abbreviate 'the building in which

Jones has his office'. Then both'S' and 'J' have mUltiple potential

referents; the former potentially refers both to Old Ivy Lodge and

to PostModern Hulk, and the latter potentially refers both to Struc-

turalist Manor and to PostModern Hulk. We treat the statement'S

is on Main Street' as true because it is supervaluationally true. It is

supervaluationally true because there are two valuations, one for

each potential referent of'S', and both make the statement true.

For parallel reasons, 'J is on Cedar Avenue' is supervaluationally

true. But'S is l' lacks supervaluational truth-value, because there

are four valuations for it, corresponding to the various combina-

tions of the potential referents of the terms; three of these are false

and one is true, so there is no consensus among them, and there is

no supervaluational truth-value.

This seems to me to be a nice analysis of the situation, I but this

pattern of analysis does not necessarily carry over neatly to the

I Nice, but not compelling. I prefer to see the statements as all contextually

ambiguous. The fact that certain of them come out true no matter how the ambi-

guity is analysed explains why we may not botber to raise the issue of ambiguity in

a practical situation, not why the sentence has a definite truth-value.

Alternatives to Indeterminate Identity 163

identity puzzles. Let us consider one of them: the case of me and

my body. A diagnosis of this case would go as follows: both 'this

person' and 'this body' have multiple potential referents. They

share a potential referent; call this 'a'. There is also a potential ref-

erent of 'this person' which is not a potential referent of 'this body';

call it 'p'. And there is a potential referent of 'this body' which is

not a potential referent of 'this person'; call it 'b'. In evaluating the

identity statement 'this person = this body' some classical valua-

tions assign a to both terms, yielding a classical valuation that is

true, and some assign different things to the terms (e.g. one valua-

tion assigns p to 'this person' and a or b to 'this body'), yielding a

classical valuation that is false. As a result, the identity statement

is assessed as supertruthvalueless. Further, any allowable classical

valuation to both 'person' and 'this person' makes the former true

of the latter, so 'This person is a person' is supertrue; likewise for

'This body is a body'. But some allowable classical valuations to

both 'this person' and 'this body' assign a to 'this person' and a set

containing both a and b to 'body' (making 'This person is a body'

true), whereas some other allowable valuations do not assign either

a or p to the extension of 'body'; so 'This person is a body' is

supertruthvalueless, and likewise for 'This body is a person'. As a

result, one gets the required extended data.

This procedure is formally adequate.

2

My worry about it is not

, In a draft MS I argued-mistakenly-that there is a formal difficulty with this

solution; I am indebted to a referee for our for pointing this oul. The mistaken

objection goes as follows:

Since 'This person is a person' must come out supertrue, every potential refer-

ent of 'this person' must be in the extension of 'person', including entity {/. For

a parallel reason, a must also be in the extension of 'body'. It then follows that

some entity, namely, a, is in the extension of both 'person' and 'body', and so

at least some bodies are determinately persons.

The mistake is to speak of "[he extension of 'person'" as if this is a fixed thing.

Instead, the predicate 'person' has multiple potential extensions, and which one is

selected for a valuation can be made to depend on which potential referent is

selected for 'this person'. (Cr. Fine (1975) on penumbral connections.) So whenever

there is an assignment to the pair 'person' and 'this person', the thing assigned to

the latter can be in the extension assigned to the former. But this permits an assign-

ment of p to 'this person' together with an assignment of an extension to 'person'

which includes p but excludes a. Because of such assignments it will not be supertrue

that a is a person.

(I believe that there are difficulties in explaining how these assignments to

multiple pieces of language are jointly allowable; these reservations are implicit

in the discussion in §§1O.4-7.)

164 Alternatives to Indeterminate Identity

logical, but rather metaphysical. The procedure requires me to be

uncertain as to which of two things I refer to with 'this person' (that

is, with 'me') and equally uncertain as to which of two things I refer

to with 'this body'. The solution in fact requires (at least) three en-

tities to be the potential referents of the two singular terms. But

I can locate no such entities. I don't believe that there are any such

entities. When I consider what the world is actually like, I can find

at most two entities that are relevant to the puzzle: me and my

body. Perhaps these are not two entities, but only one; that is at

issue. But there certainly are not three candidates that I am choos-

ing among. At least, that is what I think. That is, I believe that this

situation is not at all like the building case above. There, I am

certain that there are multiple buildings in question, some of them

parts of othcrs. I believe that there are at least three of them, and

that there is at least some kind of uncertainty about which of them

are denoted by phrases like 'the building in which Smith has her

office'. But when I consider the case of myself and my body, I am

not able to convince myself that there are here three distinct en-

tities I am choosing among.

Others, of course, will disagree. Anyone who accepts an old and

once well-established view of the nature of persons would disagree.

On that view, there are three things to consider: two substances and

a composite of the two. One substance, the analogue of p above, is

my soul; the other substance, the analogue of b above is that mate-

rial object typically called (perhaps unclearly) my body. In addition

to these two substances, there is their combination, which corre-

sponds to a above. The notion of a person is unclear between

whether it applies to souls or to combinations of souls with bodies,

and the notion of a human body is unclear between whether it

refers merely to the material substance or to the combination, to

the body-informed-by-a-soul. On such a view, the supervaluational

account works perfectly. And it is easy to find more plausible,

modern views that also have the same structure. For example: p is

a person considered in abstraction from the matter that contin-

gently makes it up, b is that mereological (three-dimensional) sum

of matter that "constitutes" the person, and a is the constituted

person. My problem is that J do not believe in such theories. I see

at most two things that are at all relevant to the puzzle-me and

my body-and I am unsure whether they are the same or not. J just

don't see anything else. Rather, I see plenty of other things, such as

Alternatives to Indeterminate Identity 165

a certain set of molecules, but I am convinced that these other

things are neither me nor my body, and the problem of the identity

of me and my body remains.

I don't expect all others to share this perspective. It remains pos-

sible for some others to patiently explain to me that there really

are three or more entities relevant to the issue, and that I am just

not sufficiently aware of the way in which the semantics of my own

language works. I suspect, however, that there are many such expla-

nations, and although they will all show that I am wrong, they are

not likely to agree with one another. This, then, is the weakness I

see in holding that the identity puzzles are to be dissolved by an

application of the technique of supervaluations; there is not at

present any particular application that will command general

assent.

10.2 CLARIFYING THE IDENTITY PREDICATE

I am not aware of any viable proposals for solutions to the iden-

tity puzzles that result from clarifying the identity predicate itself.

However, there is an idea in Lewis (1993) that one might try to

tailor for this purpose. Lewis does not propose it for the purpose

of solving the identity puzzles, so this is entirely my own invention,

and it is probably a straw man. But it caught my interest, and so I

describe it here.

Lewis explores the case of the cat and the p-cats. He has a super-

valuational solution, which will be discussed later; he also has a

non-supervaluational solution, which I discuss here. Recall that

none of the p-cats is in exactly the same place as any other, for each

determinately differs from another with respect to the having of

some part. They all mostly overlap, but not quite. Lewis will not

acknowledge the existence of a cat which is indeterminately iden-

tical with each p-cat, for he does not believe in indeterminate iden-

tity, nor will he acknowledge a cat without determinate location.

His solution is to suppose that the p-cats are all that there are, and

that we need to account for talk of the cat in terms of talk of the

p-cats. The difficulty comes when we ask which of these entities is

a cat. Lewis suggests that each p-cat is literally a cat:

166 Alternatives to Indeterminate Identity

The constituters are cat-like in size, shape, weight, inner structure,

and motion. They vibrate and set the air in motion-in short, they purr

(especially when you pat them). Any way a cat can be at a moment,

cat-constituters also can be; anything a cat can do at a moment, cat-

constituters also can do. They are all too cat-like not to be cats.' (1993: 26)

Since each p-cat is determinately distinct from each of the others,

there turn out to be millions pf distinct things on the table, each of

which is a cat. This validates the statement 'There are millions of

cats on the table', and falsifies 'There is exactly one cat on the table.'

Lewis concludes that this is correct, and that he literally has many

pets.

It is easy to defuse certain objections to this view. For example,

the many cats are not difficult to feed, because you feed them all

at once. The difficulty is not a practical one, but a theoretical one.

The theory yields hosts of consequences that contradict conven-

tional wisdom; in each case the conclusion is that conventional

wisdom is wrong, but harmlessly so. The many cats are, in a sense,

as cheap to feed as a single cat, even though the reasons that led

us to posit many cats also leads us to posit many bowls of food. So

they cost a great deal to feed, but that is no worry since you pay

with a single ten-dollar bill. Actually, the bill itself is multiple, so

you indeed pay a great deal of money, but this is money you already

possess. This does not make you wealthy, since you also have such

high expenses. And so on.

Does this mean rejecting the data? In a sense yes, and in a

sense no. Lewis explains that although there are many cats, it is

almost the case that there is only one, and you almost feed them

only one bowl of food, from a bag that almost costs a moderate

amount of money, etc. Thus our initial data are incorrect, but are

almost correct. The reason it is almost correct, Lewis suggests, is

that when we count cats, we do not do so using identity, we do

so using spatial coincidence, and the cats all almost coincide. That

is, when we say that there is exactly one cat on the table we do

not mean:

A cat is on the table, and any cats x and y on the table are such

thatx=y.

3 Lewis calls the p-cats "constituters" in this quote since he is here arguing against

the theory which says that there is one cat which the p-cats all constitute.

Alternatives to Indeterminate Identity 167

We mean:

A cat is on the table, and any cats x and y on the table are such

that x coincides with y.

In everyday talk about cats the 'coincides' becomes 'almost coin-

cides', and that is good enough for practical purposes. The identity

puzzle about the cat is thus explained as a confusion about senses

of 'exactly one'. If we use true identity in counting cats, we con-

clude, paradoxically, that there are many cats on the table. But

when we make ordinary everyday judgements about how many

things there are, we do not mean this. Instead, we mean something

that is analysable in terms not of identity but of coincidence, and

in this sense there is exactly one cat on the table. Well, not exactly

exactly one, but almost exactly one.

Before examining this theory we should distinguish two versions

of it. We have explained p-cats in terms of their present near-

coincidence with the cat, but we have not discussed their histories.

There are broad p-cats and narrow p-cats.A narrow p-cat is a thing

with determinate parts that almost coincides with the cat for its

entire existence; a broad p-cat may, but need not, do this; a broad

p-cat almost coincides with the cat right now, but may diverge

widely from it in the past or the future. So there is a version of the

account in which we construe p-cats narrowly, and one in which we

construe them broadly, and these accounts differ substantially.

Lewis intends the broad construal. The quote above continues as

follows:

They are all too cat-like not to be cats. Indeed, they may have had unfe-

line pasts and futures, but that doesn't show that they are never cats; it only

shows that they do not remain cats for very long. (1993: 26)

This version of the theory is a reformist view that is not well suited

for explaining the apparent data as being almost correct. For there

are hosts of things that we say about cats that are not even almost

true if cats are p-cats in the broad sense. Cats tend to have a life-

time of over ten years, they habitually return if you feed them, they

grow larger for much of their life, etc. None of these generaliza-

tions is even almost true of most broad p-cats. Further, not only do

I think that there is exactly one cat on the table; I think that there

is exactly one cat on the table that I got at the pound, but although

168 Alternatives to Indeterminate Identity

the former is almost true, the latter is not.

4

So the broad construal

of p-cat, coupled with the claim that p-cats are (now) cats, is

straightforwardly inconsistent with many of our common beliefs.

(One could imagine it being made consistent by a massive project

of reinterpreting our ordinary sentences. One way to do this

might be as follows. We call something a maximal cat if it is a cat

throughout its existence and if it does not ever almost coincide

with something that is also a cat throughout its existence and exists

longer. Then 'Cats usually live over ten years' might be interpreted

as 'Maximal cats almost always live over ten years, and each cat,

whenever it is a cat, almost coincides with a maximal cat.' But

maximal cats are then our narrow p-cats, and this is really a theory

of narrow p-cats, with the broad ones thrown in as epicycles.)

In order to return to the task of this chapter-which is to explain

the data, not to reform them-l focus on the version of the theory

that appeals to narrow p-cats.

I will not take issue with whether an almost-right solution is good

enough to be a solution; if we have a solution that explains how the

data in question is almost right, that is pretty good-it is almost

good enough, and perhaps fully good enough. T have two other

worries about this solution. The first is that it is not almost correct

as it stands; the second is that even if it is almost correct, it does

not generalize to the other identity puzzles.

The reason the solution is problematic as it stands is that it sup-

plies a way to construe the data that almost validates that data,

without giving us any reason not to also construe the data in equally

good ways, where it does not do this. Suppose that we sometimes

interpret claims about how many things there are by substituting

coincidence for identity. This does not mean that we always do this.

In some cases, we cannot, for we make plenty of assertions about

how many things there are where spatial coincidence is not applic-

able: "There are two prime numbers between 3 and 9", "Three

theorems were formulated by McGuillicutty", "Four electronic

payments failed to reach the bank". When coincidence and iden-

tity coincide, we can also use coincidence. But when it is absolutely

clear and uncontroversial that near-coincidence diverges from

actual coincidence, in a situation in which actual coincidence coin-

4 This sort of fact will raise difficulties for any account of definite descriptions

based on this view.

Alternatives to Indeterminate Identity 169

cides with identity, identity rules, or it is at the very least a natural

option. This is easy enough to illustrate by focusing on the p-cats

themselves, setting aside the question of whether they are cats. How

many p-cats are there on the table? Presumably, millions, and there

is no sense at all in which there is exactly one p-cat; that claim is

not in any sense almost true, even though the p-cats almost coin-

cide. If p-cats are under discussion, we count them using real iden-

tity, and this is the only available interpretation. But the p-cats

are cats, and the cats are p-cats; there is no difference at all between

cats and p-cats on this view. So there are exactly as many cats as

p-cats. So there can be no difference in the answer to the question

"how many are there?".

This is perhaps a bit too facile as a refutation. The response will

be, of course, that there can be a difference in the answer, if the

interpretation of the question is sensitive to the vocabulary used in

it. Perhaps 'cat' suggests an ordinary context in which we count with

near-coincidence, and 'p-cat' suggests a technical context in which

we count with identity. Then there are two ways to construe the

question, and Lewis has made the natural choice.

This is true about popular usage in some related cases. For

example, in popular usage, there may be "nothing" in a room, when

we know there is air, dust, wallpaper, and so on. But we can clarify

our usage, making clear for example whether we are going to count

air molecules as things, or only larger solid entities, and when we

do this there are two interpretations of the question, with two dif-

ferent answers. We can do the same for the question about cats,

making clear whether by 'exactly one' we mean one that doesn't

differ much from the others or one, and there are no others. When

I do this, I don't get any interpretation of the question that yields

a plausible answer that there are actually millions of cats on the

table. But this is not a point amenable to argument, since this is the

point at which we will disagree about the data. Lewis has made it

clear that once the question is clarified, he believes that there are

literally many p-cats on the table, and since each p-cat is too cat-

like not to be a cat, it is only in a loose and popular sense that there

is only one cat there. I think that there is only one cat no matter

how strict we are in our wording.

A distinct question is whether this solution, if it is a solution, gen-

eralizes to other identity puzzles. Recall that Lewis does not say

that it does, and I think it does not. Consider the case of the

170 Alternatives to Indeterminate Identity

disrupted person. The supervaluational approach to the person

case is similar to the cat case, in that we are driven to suppose that

there are several persons in the same place at the same time. But

in the many-person case, the persons are not in almost the same

place; they are in exactly the same place. We find this problematic

since we are forced to say there are many with no differences at all

among them. So what if we replace the multiple persons in exactly

the same place by multiple p-persons in almost the same place?

That solves the problem of distinguishing them, since they all

occupy slightly different places. But distinguishing the p-persons

will not solve the puzzle. We would have a solution to the puzzle if

some p-persons persisted through the change and some did not.

But there is no reason to expect the disruption to cleave along the

lines that distinguish p-persons. For example, if the disruption con-

sists of a partial brain transplant, then all of the p-brain portions

get transplanted or none does. It is not that some do and some

do not. (Or, if that is what happens, it is unrelated to the issue of

identity.) So an appeal to p-persons does not solve this problem,

nor any of the other paradigm puzzles that we have discussed.

5

We

need some different approach.

10.3 INDEFINITE CONCEPTS

It is natural to hold that the concepts expressed by our words are

imprecise. Since the concepts associated with our words determine

the propositions that get expressed, and since the truth of these

propositions determines the truth of the sentences that express

them, if the concepts are imprecise, this may produce propositions

with no truth-value. Perhaps this is what is to blame for identity

puzzles with no answers. A solution to the puzzles might be to con-

sider how we imagine replacing our imprecise concepts by more

precise ones.

5 One might try to create an analogue to the cat case by considering the p-persons

at time t to be the temporal parts of the person that exist at t. A supervaluational

reading then tells you that there is one person here now, and there was one person

there then, but there is no answer to the question whether the person who is here

now is the person who was there then. However, this result follows even if there is

no disruption at all; thus there would be no answer to the question of whether

anyone now is the person born previously at a given time and place.

Alternatives to Indeterminate Identity 171

Of course, more needs to be said than this. Holding that our con-

cepts are imprecise does not, by itself, explain why they nonethe-

less allow us to express a great deal of exact information. The ship

puzzles arise not because we have no idea how many ships there

were originally, and finally, but because we think there was origi-

nally one and finally two. Somehow our concepts are good enough

for determinate judgements about these matters. So what is needed

is an explanation of how imprecise concepts sometimes do, and

sometimes do not, yield determinate judgements, and how some

appeal to precise concepts might shed light on this.

Here is one explanation. Our concept, say, of person is imprecise.

Ideally, it would be replaced by a more refined concept. But that

takes time, effort, and work, and we have other priorities. So we

use the word 'person' without refining the concept associated with

it, but with the understanding that there are indeed ways to refine

it, and we know a fair amount about what would result. In ordinary

everyday situations we don't bother to consider these refinements,

because there is no particular way to refine the concept that is

importantly different from any other way. This is because, for the

purpose of the judgement at hand, any refinement would produce

the same result. For example, when we say that one ship left port,

we know that there are many ways to refine the notion of a ship,

and that any way of doing this will validate the claim that one ship *

left port, where 'ship*' expresses the resulting refinement of the

unrefined concept of ship. We don't bother refining on the spot

because it wouldn't make any difference to the correctness of the

claim at issue. The identity puzzles are unusual cases in which it

does make a difference how we refine things. If we refine ship in

one way, it is correct to conclude that the original ship, is identical

with the shipl with new parts, where 'ship/ expresses one particu-

lar refined version of the concept ship, and if we refine in some

other way, it is correct to conclude that the original ship2 is identi-

cal with the shipz with new parts. Refining yet a third way lets us

conclude that the original ship, ceased to exist, and was replaced

with two new ship/so Since how we refine makes a difference to the

correctness of the identity judgement, there is no unique answer to

the original (unrefined) question.

This proposal is an application of supervaluations. A valuation of

a general term results from its coming to express a refined version

of the concept that it actually expresses. (If the term occurs twice,

172 Alternatives to Indeterminate Identity

we must be sure that the same valuation is used for each occur-

rence.) Then a judgement is intuitively true (and, so, part of our

extended data) if and only if it is supervaluationally true on this

account. In the ship case, the claim that exactly one ship left port

is supervaluationally true, since it is true no matter how the concept

ship is refined; likewise the claim that exactly two ships docked. But

any of the problematic identity judgements lack supervaluational

truth-value, since they come out true on some valuations (some

refinements) and false on others.

In § 10.1, I argued that no supervaluational assessment will solve

the problem of me and my body. However, the person/body puzzle

is a special one, and this may be too narrow a point to reject super-

valuational solutions in general. So it is worth looking more closely

at an intuitively motivated theory of this sort. An example of this

approach is proposed in Stalnaker (1988). Stalnaker considers a

puzzle about restaurants that is isomorphic (with respect to iden-

tities) with the ship puzzle. A restaurant is started called 'Book-

binders'. At some later date, there are two rival restaurants with

almost the same name, each claiming to be the original Book-

binders, and with a history posited that makes each identity claim

as plausible as the other. Stalnaker (1988: 353) supposes that "the

concept of restaurant is indeterminate in that it is indeterminate

what counts as the same restaurant at different times." On some

ways of arbitrarily refining our concept, the original "restaurant"

will become one later "restaurant", on other ways it will become

the other, and on yet others it will cease to exist, to be succeeded

by two new "restaurants". This creates indeterminacy in the term

'the original restaurant', but no indeterminacy in the world.

For this solution to be plausible it is important to emphasize that

we do not ordinarily think of ourselves as seeking a common

content of mUltiple precisifications of our imprecise concepts.

Rather, we use our imprecise concepts as they stand. The use of

supervaluations of precise versions is an analysis of how we would

rationalize our use of the imprecise concepts, if such rationaliza-

tion were called for. We would rationalize our use of the imprecise

concepts by granting that we are using imprecise concepts, and

granting that we are using them in situations in which it is not rel-

evant to our present concerns how to make them more precise-

it is not relevant because it doesn't matter how this is done, for the

Alternatives to Indeterminate Identity 173

same answer would result no matter how we do it consistent with

our intentions. What makes the identity puzzles special is that they

are cases in which identity statements can be made for which it does

matter how we make our concepts more precise. And the instinc-

tive recognition that making them precise in different ways would

yield divergent answers is what underlies our bewilderment when

confronting them.

That is the view. But why should we think that it works? How

can we be sure, for example, that it is true on any refinement that

exactly one ship left port? There is nothing about the view as it is

sketched to guarantee this. The theory must be fieshed out so as to

guarantee it. The theory needs to specify how to refine concepts,

perhaps something like this: Whenever we refine our sortal terms,

we do so in such a way that if a sortal term is true of extended

spatio-temporal objects, then nothing counts as a refinement of the

concept that it expresses unless the refined concept is never true of

two spatially coincident things in normal circumstances. (The 'in

normal circumstances' is there to avoid, for example, science-fiction

cases in which techniques have been developed to let objects pass

through one another, so that two of them might momentarily

coincide.) If it does this, then any refinement of the concept ship

will let you stand at a dock and know that there is exactly one

refined-ship in front of you no matter how you refine ship.

However, this proposal, all by itself, is not adequate. This is

because so few guidelines are given for refining our concepts. In

any particular case, this seems outright impossible to do.

6

To refine

the notion of ship we need to explain how to tell if there is a ship

present (or several ships present) in a given situation, and we need

to explain how to tell for any such refinement whether any given

refined-ship a is identical with any refined-ship b. If this is not

accomplished, there remains a set of identity puzzles for refined-

ships that are not solved by the technique. But concept refinement

needs to be implemented with the resources at our disposal. Since

the conceptual resources at our disposal consist entirely of unre-

fined concepts, it is difficult to see how to go about this task. It is

fi Criticism of this sort appears in Woodruff and Parsons (1997) and Parsons

(forthcoming) in somewhat primitive form. (Because of the "fest-conference" in

which it originated, the tone of writing in the latter was meant to be "festive"; no

disrespect was meant to the views criticized there.)

174 Alternatives to Indeterminate Identity

conceivable, but unlikely, that combinations of imprecise concepts

can in all cases yield precise ones. It is even more difficult to have

confidence that the end result will yield the hoped-for patterns

needed to solve the identity puzzles.

There is, however, an answer to this challenge. Recall that the

view is not that we already grasp and use a multiplicity of refined

concepts when we speak; we only need to be confident that our

concepts could be refined on demand. But even this is perhaps not

necessary. The view could be that we do not ever have to be

able to produce the refined concepts, we just speak as if we are

expressing precise concepts; we do this in the knowledge that all

attainable concepts are imprecise, but we have in mind, at least

vaguely, ways they might be made more precise, and we suppose

that there exist concepts which are precise enough, even though

they are perhaps in principle ungraspable by us. We have in mind

ideals of completely precise concepts. When we use a word, it is to

be taken as if it were expressing a precise concept, one which is

a refinement of the imprecise one we actually express. Generally,

it won't matter which refinement of our imprecise concept we

use, because in everyday matters they all will validate what we are

saying. But for applications far removed from present-day

concerns, it may matter. The identity puzzles are of this sort, which

is why they lack answers.

But, again, why assume that this will work as advertised? If we

consider the most obvious ways of refining our concepts, it will not

work at all. Suppose that the refined concepts work like this.

Instead of the imprecise notion of ship wherein it is often unclear

whether a ship survives replacement of its parts, or whether it sur-

vives disassembly and reassembly, we have various refined notions

of ship. Some of these permit their denotata to have many of their

parts replaced and go on existing, and some of them rule that if

their denotata lose parts, they thereby cease to exist, even if the

parts are replaced. Likewise for disassembly and reassembly. Some

refined notions of ship rule that their denotata come into existence

upon assembly, regardless of whether the parts were previously

parts of a refined-ship before being separated. And so on. These

various combinations of criteria for identity through time are

embodied in various refined notions of ship, the only requirement

being that none of them permits any indeterminacy in the question

of whether one of its denotata exists at a time and place that a ship-

Alternatives to Indeterminate Identity 175

like thing exists, or in whether one of its denotata existing at one

time and place is identical with one of its denotata existing at some

other time and place. However, if these are what refinements are

like, then the proposal is in trouble. For although it produces the

correct answer in the ship-puzzle case, it produces wrong answers

in apparently non-problematic cases. It produces indeterminacy

where we do not want it. Consider Queen X's ship that is buried

with her under her pyramid; this was the royal barge on which she

held court during her reign. When she died it was transported to

her burial chamber, where it lies available for her afterlife journey.

We would like to say that she held court on exactly one ship during

her reign, and that one ship is buried with her, and that the ship on

which she held court is the ship that is buried with her. But, unfor-

tunately, they had to take it apart to get it into the burial chamber,

where they reassembled it. And so the theory under consideration

agrees that she held court on one ship, and it agrees that one ship

is buried with her, but it holds that it is indeterminate whether the

ship that is buried with her is the ship on which she held court. This

is because on some refinements of the notion of ship the ship on

which she held court was transported to the burial chamber (by

disassembly and reassembly), and on other refinements it ceased

to exist upon disassembly and a new ship was assembled from its

parts. And so by the supervaluational approach, there is no truth-

value for the claim that the ship on which she held court is the ship

buried with her. But this is just plain wrong. And the wrongness

extends to a host of everyday cases. If I had to disassemhle my desk

to get it into my study when I moved, there is now exactly one desk

in my study, but the proposal says that there is no truth-value for

the claim that the desk in my study was originally purchased at a

thrift shop-for this is true on some refinements of the concept of

desk hut not on others. So this natural implementation of the

refined-concept view yields the wrong answers in cases in which

identity puzzles should not arise.

Obviously, we are working with too crude a notion of concept

refinement. We need to refine our notion of refinement. It was

apparently a mistake to suggest that there is a refined notion of

ship on which a ship survives disassembly and reassembly, and one

on which it does not. This is an overly simplistic proposal for what

a refined concept should be. Instead, we need to say that there is a

refinement of ship on which a ship survives disassembly and

176 Alternatives to Indeterminate Identity

reassembly when another ship continuous with and resembling the

original gets new parts, and another on which a ship ceases to exist

upon disassembly when another one continuous with it and resem-

bling it gets new parts. Or something like this. This appears to be an

ad hoc move, and it is not at all clear how to generalize it. But we

must be careful not to be too hasty with charges of ad hocness if

the resulting account gets the answers right. The ad hocness of an

account that gets the right answers might be the outer appearance

of a natural account that we have not yet been able to articulate.

So let us proceed with this idea: that our everyday judgements are

all to be assessed in terms of supervaluational readings, where the

extended valuations are produced by refining our concepts in what-

ever way it takes to get the desired answers. This may make the

theory too abstract to evaluate, but we won't know if we don't try.

Here is a second objection. It applies to all of the puzzles, but I

will discuss it in application to the refinements of the sortal notion

of a person. It goes as follows:

No matter how the notion of a person is refined, a person had

better be something that thinks, feels, plans, and makes com-

mitments to other people-otherwise the proposal will end up

attributing lack of truth-value to truisms such as that (normal)

people think, feel, plan, and make commitments. Now consider the

judgement in the disruption case:

The person who entered is the person who left.

This needs to turn out to lack truth-value. On the supervaluational

approach, this can happen only if the refinements in the concept of

person result in multiple potential referents for 'the person who

entered' andlor 'the person who left'. So for this to be a solution,

there need to be divergent refinements of person such that 'the

person who entered' ends up with divergent potential referents, at

least one of which left, and at least one of which did not. Now let

us call any object that is truly characterized by some refinement of

the concept of person a "refined-person".7 Then there were two

refined-people in the same place at the same time before entering.

So there were two thinking, feeling, beings there at the same time.

7 This is a technical notion. A refined-person is not a person who is refined. An

entity x is a refined-person if and only if there is a refinement of the concept of

person such that, under that refinement, a is a person.

Alternatives to Indeterminate Identity 177

But this is just as objectionable as the traditional solution to the

puzzle which held that there were two ordinary persons in the same

place at the same time. How can I reassure my wife by telling her

she is the only person I love, if she knows that I love many refined-

persons, that is, many female, thinking, feeling beings, and that J feel

committed to each of them? (Of course, I speak loosely of "my

wife", since that is unrefined talk, but I do have a refined-wife; in

fact, I may have several of them. What will they think?) The point

is that the solution at hand demands an expanded ontology, and

one expanded with things that give rise to the original puzzle all

over again. The identity puzzles all reappear with 'person' replaced

by 'thing that thinks, loves, plans, owes money on a mortgage, etc.'.

We are thus in danger of having a version of the traditional "many

persons in the same place" solution, with a slight change of vocab-

ulary (with 'person' replaced by an elaborate explanation of what

all refined-persons have in common).

There is a natural way to escape that objection.

R

The proposal

under discussion says that we need to consider refining our con-

cepts, but we have really only considered refining our sortal con-

cepts. Since the difficulties above involve reconstructing identity

puzzles with non-sortal concepts (thinking being, instead of

person), one obvious option is to consider refining these as well. At

first glance, this merely enhances the difficulty. Consider the person

standing in that doorway, and consider the female being standing

in that doorway. We would like to say that the female being stand-

ing in the doorway is the person standing in the doorway. But now

we appear to have (at least) two different unrefined concepts to

work with: female and person. If each of them resolves into many

refined concepts, we get twice-many combinations of them. Some

such combinations may render true the statement:

the female being in the doorway = the person in the doorway

but many will not. That is, given that there are many different

refined-persons standing in the doorway, corresponding to each of

them will be many different refined-female beings. But most of

them will not correspond. In general, if the person14 is identical with

A 1 am indebted to Calvin Normore for helpful conversation regarding the

material in the remainder of this chapter.

178 Alternatives to Indeterminate Identity

the female being14 and distinct from the person22, then one resolu-

tion will be:

the female being14 in the doorway = the personl4 in the

doorway

but another will be:

the female being14 m the doorway = the person22 in the

doorway.

Since the first of these is true, the second is false, and the super-

valuational reading with the refinements providing the valuations

will lack truth-value. Thus the apparent data is not preserved.

There is, however, a way to try to get this to work out. It is

sometimes speculated that any use of a non-sortal word as a noun

in language presupposes some underlying sorta1. So if you say 'red

thing', you must have in mind some sortal concept to flesh out

'thing'. This is controversial, but it has some credence. So suppose

that our uses of a word like 'female' are coordinated with a

sortal concept, such as person. Then we just need to rule that in

evaluating our utterances in terms of supervaluations, we must limit

the valuations to the "coordinated" ones. For example, we need to

suppose that 'the female being in the doorway' means 'the female

person in the doorway'. Then we go back to just refining the sortals,

as we did before, but we refine the implicit ones along with the

explicit ones. If we do this, the underlying structure of our problem

example is:

the female person in the doorway = the person in the doorway.

We also need to assume that a refined-person is female at a time if

and only if every refined-person that coincides with it then is also

female then. Now we can refine person any way we like; so long as

we refine uniformly, the above statement will come out supervalu-

ationally true. For although there are many refined-people in the

doorway, on any given refinement there is only one. Likewise, if we

also coordinate the refinements of 'think', 'feel', and so on, there is

only one thinking thing in the doorway, and that thinking thing is

the female there.

The catch lies in the constraint we need to impose on 'female' so

as to make the account work. When do we impose a constraint of

this sort, and how do we tell exactly which constraint to apply? J

Alternatives to Indeterminate Identity 179

am not sure how to do this. The category analysis is still only a

sketch of a solution. This becomes even clearer if we consider how

entities of different categories are to interact. Consider Stalnaker's

restaurant case, and imagine that I am a server in the original

restaurant, before the problematic split. There are at least two

refined-restaurants that employ me. There must be two refined-

restaurants here in order for the identity claims to be indetermi-

nate between the original restaurant and the later ones, and I must

work for each of them, for otherwise '1 work for the restaurant on

the corner of Main and Elm' will not be true. Now when 1 file my

income taxes, I need to submit a W2 form for each of my employ-

ers.1t thus appears that 1 must fill out (at least) two W2 forms. This

is because each refined-restaurant does literally employ me, and

they are literally distinct from one another. Coordinating cate-

gories does not reduce the number in this case since the category

of person under which I fall is not coordinated with the category

of restaurant under which they fall. So we conclude incorrectly that

1 need to fill out two W2 forms.

Or do I? Perhaps 'W2 form' needs refinement, and perhaps it

is coordinated with the category of restaurant. Of course, there

would not be a direct coordination of W2 form and restaurant,

but there might be a coordination of W2 form with employer, and

perhaps the category of restaurant is to be analysed as employer

that serves food. Then I fill out one W2 form, because the require-

ment to fill out one per employer is fulfilled no matter how we

refine both W2 form and employer, so long as we coordinate the

refinements. However, it is not reasonable to analyse restaurant

as a subcategory of employer, for there are other equally good

categories under which it fits: purchaser, sales-tax collector, retail

establishment, etc.

10.4 LEFT IN LIMBO ...

Ideally, I would now be able to say that I have refuted the linguis-

tic/semantic alternatives to the hypothesis that there is indetermi-

nacy in the world. But clearly I have not; I have only raised

problems for how to work out various versions of alternative views,

without any solid conclusions in principle about how they might be

180 Alternatives to Indeterminate Identity

altered and developed so as to take account of the problems. And

so we are left in limbo. But I hope to have shown that there does

not yet exist an alternative view about which we can be confident.

And so the worldly indeterminacy of identity view is deserving of

close consideration.

11

Sets and Properties with

Indeterminate Identity

11.1 IDENTITY CONDITIONS FOR SETS

Sets, whatever else they are like, are the sorts of things that can be

said to be identical to or distinct from one another. So they are

subject to all of the logical principles involving identity discussed

elsewhere in this work. But they are also subject to an additional

principle. What is most distinctive about sets-which distinguishes

them from other sorts of things-is that their identity is determined

by their membership. This is meant in the strongest possible sense,

in that:

(i) if everything that is determinately a member of set A

is determinately a member of set B, and vice versa, and

everything that is determinately not a member of set A is

determinately not a member of set B, and vice versa, then A

is determinately identical with B;

(ii) if something is determinately a member of set A but

determinately not a member of set B, or vice versa, then A and

Bare determinately distinct;

(iii) otherwise it is indeterminate whether set A is identical to

set B.

This is a spelling out of a point that can be put more simply by

saying that if A and B are sets, then the truth-value status of the

claim that they are identical:

A=B

should be exactly the same as the truth-value status of the claim

that they have exactly the same members:

182 Sets with Indeterminate Identity

Vx(x EA<=:} X E B).

In terms of a single principle, this equivalence is true:

[Set Essence] A = B <=:} Vx(x EA<=:} X E B).

I assume that any non-bivalent theory of sets will adopt this

as a basic axiom. A particular theory of sets will result by

adding to [Set Essence] some principles that say which sets exist,

usually by stating which conditions have sets corresponding to

them.

11.2 DDIFF FOR SET MEMBERSHIP

Before considering which sets exist, we have another question to

consider. This is whether sets are to be thought of as robust con-

stituents of the world, or whether they are merely part of the con-

ceptual apparatus in terms of which we view that world. We can

formulate theories about either sort of entity, but we need to be

clear about the choice when developing the theory. The difference

between these two conceptions of set is whether being a member

of a set is itself a worldly property that a thing has, or whether talk

of a thing being a member of a set is just about a kind of concep-

tual classification. If the former, then we can expect that member-

ship in a given set is one of the worldly properties in terms of which

we characterize worldly identity. If the latter, then we make no such

assumption. In the first case, I speak of worldly sets, and in the

second, conceptual sets. I postpone a consideration of conceptual

sets to §11.9, and consider worldly sets here.

Worldly sets obey the principle that if some thing x is determi-

nately a member of such a set, and if y is determinately not a

member of that set, then x is distinct from y. The principle is:

!x E Z & !--,y E Z ~ --,x = y.

This is equivalent to the claim that membership in a given set

satisfies principle DDiff from Chapter 4. So I give it that title:

[DDiff for E Set] !x E Z & !--,y E Z ~ --,x = y.

Sets with Indeterminate Identity 183

11.3 SETS OF OBJECTS

Suppose we wish to formulate a theory of objects and sets of

objects, with the understanding that sets do not themselves contain

sets as members. Such a theory is formulable by the adoption of

the two principles discussed above together with a principle of set

comprehension which says that any formula with a free variable

over objects determines a set if it satisfies DDiff for set member-

ship. Suppose that we temporarily use small letters for variables

over objects (and not over sets) and capital letters for variables

over sets (and not over objects). We assume the logical laws of iden-

tity for the identity sign between any size letters, along with the fol-

lowing axioms (with the understanding that they are all implicitly

universally quantified):

[Set Essence] X= Y ~ 'i/z(z E X ~ Z E Y).

[DDiff for E Set] !x E Z & !-,y E Z ~ !-,x = y.

[Comprehension]

[Sets aren't Objects]

3X'i1z(z E X ~ <pz) where <p is any

formula satisfying

DDiff and not

containing 'X'

free.

'i/x'i/Y -,(x = Y).

These assumptions all strike me as defensible for a theory of sets

of objects.

11.4 INDETERMINATE IDENTITY FOR SETS

If one wishes to combine such a set theory with a theory encom-

passing indeterminacy, then one must allow sets to be indetermi-

nately identical to one another even if there is no indeterminacy of

identity of individuals at all. This is because one can devise distinct

defining conditions for sets such that it is indeterminate whether

the defining conditions specify the same members or not; [Set

Essence] then entails that it is indeterminate whether the sets

184 Sets with indeterminate identity

defined from those conditions are the same. Here is a sketch of a

proof that addresses this phenomenon.

Suppose there is no indeterminacy of identity among objects

at all, but there is some indeterminacy; i.e. there is a formula

<jlx true only of individuals and such that it is indeterminate

whether <jla is true:

v<jla

Since there is no indeterminacy of identity of individuals, both

'<jlx' and '<jlx v x = a' satisfy DDiff. Applying [Comprehension],

there are sets A and B such that:

V'x(x E A q <jlx)

V'x(x E B q <jlx v x = a)

It is easy to show that a E B. Then we may show that it is

indeterminate whether A = B by refuting both the claim that

A = B and the claim that -,A = B. For, assuming the first claim

(A = B), we can show that a E A, and thus <jla, contradicting

v<jla. And assuming the second claim (--,A = B) lets us infer

that there is some determinate member of B that is deter-

minately not a member of A; this can only be a, and thus -,<jla,

again contradicting v<jla. As a result, the identity 'A = B' is

indeterminate.

Thus, the mere existence of indeterminacy among objects gener-

ates indeterminate identities among sets, even if no objects are

indeterminately identical at all. This is because any indeterminacy

can be used to provide conditions to define set membership, as in

the proof just given.

Of course, indeterminate identity among objects also leads to

indeterminate identity of sets. Suppose that it is indeterminate

whether a = b. Our axioms entail the existence of the "unit set" of

a, call it la], the set yielded by [Comprehension] applied to the

formula 'z = a'. This set has a as a determinate member and it has

everything determinately distinct from a as a determinate non-

member; anything that is indeterminately identical with a is such

that it is indeterminate whether that thing is a member of [al. In

particular it is indeterminate whether b E [al. There is also a "pair

set" of a and b, call it '[a,bl', yielded by [Comprehension] applied

to the formula 'z = a v Z = b'. This set has both a and b as determi-

Sets with Indeterminate Identity 185

nate members. It is easy to see that there is nothing that is deter-

minately a member of one of these sets:

la)

{a,b)

that is determinately not a member of the other. So they are not

determinately distinct. But they are also not determinately the

same, since b is determinately a member of the second, but not of

the first. So it is indeterminate whether they are the same:

'l(la] = (a,b]).

So if there is any indeterminacy at all there will be sets that are

indeterminately identical, whether individual objects are indeter-

minately identical or not.

Let us now review Salmon's argument to the effect that set

theory is incompatible with indeterminacy of identity of individu-

als. Changing ordered to unordered pairs (to pair-sets), the argu-

ment is:

(1) 'l(a=b)

(2) -''l(a = a)

(3) -,(la,bl = {a,al)

(4) -,(a = b)

(5) -''l(a = b)

Hypothesis to be disproved

Truism

From (1) and (2) [No!]

From (3), by [Set Essence]

From (4)

There is no way to derive (3) from (1) and (2). In order to derive

step (3) from [Set Essence], we would need to show that there is

some member of {a, b) that is not a member of {a,a}, or vice versa,

but there is no way to do that. Instead, we can show, using com-

prehension, that it is indeterminate whether {a,bl = {a,a}, as we did

above.

11.5 A ZERMELO-FRAENKEL-LIKE

HIERARCHY OF SETS

The logical power of set theory comes when we admit not only sets

of objects, but also sets of sets, and sets of those sets, and so on

186 Sets with Indeterminate Identity

without limit. The following theory is adapted from Woodruff and

Parsons (forthcoming). In it we suppose that sets are built up in

stages from objects, which in the spirit of set theory are here called

"individuals". At stage 0 are sets whose members are only individ-

uals. At stage 1 are sets whose members are themselves sets of indi-

viduals, perhaps along with some individuals. At each stage, sets

contain as members or indeterminate members only those things

that exist at earlier stages. Assumptions are made that yield a trans-

finite series of such stages, just as in Zermelo-Fraenkel set theory.

There are two constraints on this process that are automatic in the

bivalent formulation of classical Zermelo-Fraenkel but that need

articulating in our framework. One is the DDiff condition for sets:

if y is detenninately a member of X and z is determinately not a

member of X, then it is determinate that y 1= z. The other is that a

set may not contain indeterminate members with ranks higher than

the ranks of all of its determinate members. The theory that results

from these assumptions is "neat", in that sets fall into ranks deter-

mined by the stage at which they are constructed, and it is deter-

minate whether a given set is a member of a given rank. Further,

if it is indeterminate whether X = Y, then X and Y will be deter-

minately of the same rank, and if it is indeterminate whether

X E Y, then it is determinate that X falls into a rank immediately

below that of Y. Other sorts of assumptions might be made, but

they are not considered here. I

Our vocabulary includes at least the primitive two-place predi-

cates 'E' and '='. We assume a primitive name 'I' for the set of indi-

viduals and '0' for the set with no determinate or indeterminate

members. Hereafter we understand small letters to range over

everything (over individuals and sets) and capital letters to be vari-

ables restricted to sets (to non-individuals); so ''I1X[ ... X ... ]' is

short for ''I1x[--,x El==> ... x . . .]'. In order to formulate the rank

restriction it is convenient to have a primitive two-place predicate

'ranks', where 'ex ranks x' means that the ordinal ex is the rank of

entity x. The two axioms above, [Set Essence] and [DDIFF for

I The "neat'" falling into definite ranks is a consequence of the principle that a

set may not contain indeterminate members with ranks higher than the ranks of all

of its determinate members. Relaxing this constraint has far-reaching consequences,

such as unit sets that are two or more ranks above their members (see Woodruff

and Parsons forthcoming: §2). I do not know how to develop a systematic theory of

worldly sets without such a constraint.

Sets with Indeterminate Identity 187

E Set), along with the following are our axioms for IZFU ("Inde-

terminate Zermelo-Fraenkel set theory with Ur-elements").

I present the remaining axioms beginning with things of the

lowest rank.

[Bivalence of Individualness] x E I v -tX E I

[Individuals Lack Members? x El:::::> -,::3yy E x

[Empty Sets] -,::3y!<!Jy:::::> ::3S'v'y[y E S <=> yE 1& <!Jy]

[Empty Sets] generates sets with no determinate members and with

arbitrary sets of individuals as indeterminate members. Selecting

'<!Jy' to be '-,y = y' entails the usual empty set axiom of ZF:

::3S'v' y-,y E S.

This yields an "emptiest set", 0, a set with no determinate or inde-

terminate members:

[Emptiest Set] -,0 E I & 'v'y-,y E 0.

If there are any individuals at all, there will be sets indeterminately

identical to 0; they, unlike 0, have indeterminate (individual)

members.

The following structural axioms are straightforward from classi-

cal set theory.

The pair set of x and y, denoted by 'Ix,yj', is the set whose deter-

minate members are exactly x and y, and whose indeterminate

members are those things that are not determinately identical to x

or to y but are indeterminately identical to x or to y:

[Pairs] ::3S'v'u[u E S <=> u =X v u = y].

The union of a set of sets is the set whose members are members

of one of those sets:

[Union] ::3S'v'Z[Z E S <=> ::3u(z E U & u E X)].

The power set of a set is the set of subsets of that set. Define

"subset" as follows:

x <;;;; Y =d[ -tX E I & -,y E I & 'v'u(u EX:::::> U E y).

Then the power set axiom is:

[Power Set] ::3S'v'z[z E S <=> Z <;;;; X].

2 This axiom together with the previous one entail that I is a set ( a non-individ-

ual): ....,1 E 1.

188 Sets with Indeterminate Identity

The usual ZFaxiom of infinity assumes that there is a set con-

taining 0 and containing the "successor" of any of its members,

where the successor of Y is the set whose members are the

members of Y along with Y itself:

[Infinity] 3S[0E S & 'liY(!YE S & 'liU(UE

Z{::::}U=YvUE Y)])J.

As discussed above, our Replacement scheme requires rank

restrictions. To formulate the rank restrictions, we first define the

ordinals. As a preliminary, we call a set 'tight' if its indeterminate

members are limited to those things that are indeterminately iden-

tical to some determinate member of the set:

Tight(x) =df -,x E 1& 'liy[vy E X 3z(!z E x & vy = z)].

Transitive sets are ones whose determinate members are determi-

nate subsets:

Transitive(x)=df-,xE 1& 'liY(!YE

Ordinals are then defined as tight transitive sets whose determi-

nate members are all tight and transitive:

Ord(x) =df Tight(x) & Trans(x) &

'liy[!y E x Tight(y) & Trans(y)].

So-defined, being an ordinal is a bivalent condition:

THEOREM: Ord(x) v -,Ord(x).

To express ordinal comparison, we define:

x <y =df Ord(x) & Ord(y) & x E y.

X::::y=clfX<YVX=y.

In order for the ordinals to be well ordered we take this as an

axiom:

[Least Ordinal] 3x[Ord(x) & !x E S] 3x[Ord(x) & !x E

S & 'liy[Ord(y) & !y E S x ::::y]].

Hereafter we use small Greek letters to range over the ordinals. To

complete the structuring of the ordinals we adopt:

[Ordinal Non-Self-E] -,U E U.

Sets with Indeterminate Identity 189

The following axioms constrain rankings. The first two resem-

ble parts of a familiar definition of ranking by recursion, but

they are stated here as axioms because 'ranks' is primitive

notation:

[Individuals are Not Ranked] x E I ~ -.y ranks x.

[Ranking for Sets] Z ranks X<=> z is the least ordinal such

that \fy[3x(!x E X & y ranks x) ~

y < z].

As discussed earlier, we posit that no indeterminate member of a

set exceeds all the determinate members in rank:

[Rank Limitation of Indeterminate Members]

-.!-,x E X & a ranks x ~ 3y3P[!y E X & pranks y & a :s; Pl

Our final axiom scheme is [Replacement]. Suppose we have (i)

a bivalent relational formula x\JfY which (ii) is functional on a set

S, and suppose that x\JfY "projects" from S a condition satisfying

(iii) the rank constraint and (iv) DDIFF for set membership. Then

its range for domain S is a set:

[Replacement]

If:

(i) u\JfY v -.u\JfY

(ii) -.!-.u E S & -.!-.v E S & u\JfY & v\Jfz & U = v ~ Y = Z

(iii) -.!-.u E S & u\JfY ~ 3v3z[!v E S & v\Jfz & 3a3p[a ranks

Y & pranks Z & a:s; P]]

(iv) !3x(x E S & x\Jfu) & !-.3y(y E S & Y\Jfv) ~ -.ll = v

Then:

3XVy[y EX<=> 3z[z E S & z\JfY]].

In Woodruff and Parsons (forthcoming) the above axioms are

shown to be consistent if classical bivalent Zermelo-Fraenkel set

theory with ur-elements is consistent.

Theorems. The following are some useful theorems.

THEOREM: 3a[ a ranks X]

THEOREM: x ranks Y v -,x ranks Y

THEOREM: a ranks x & p ranks x ~ a = p.

THEOREM: YE X ~ 3a3p[a ranks Y & P ranks X & a < P].

THEOREM: -.!-.X = Y ~ 3a[a ranks X & a ranks Y]

190 Sets with Indeterminate Identity

[Separation] follows from [Replacement]. Given a set Sand

formula <jl, a separated set exists if <jl satisfies DDIFF with respect

to the members of S and if the appropriate rank constraint is

satisfied:

THEOREM: [Separation]

If (i) !(x E Z & <jlx) & !-,(y E Z & <jly) => -,.X = Y

(ii) -,!-,(Y E Z & <jlY) => 3x[!(x E Z & <jlx) & 3a3p[a

ranks x & pranks Y & a ;:>: P]],

then 3S'v'x[x E S ~ X E Z & <jlx].

[Foundation] holds in the following form:

THEOREM: 3yy EX=> 3y[-,!-,y E X & -,3z(z E y & Z EX)].

Determinate tight separation: [Separation] is restricted by the

DDIFF and rank constraints, which can be irksome in practice. If

we want a separated set and we care only about getting a set with

a certain determinate membership, then the constraints may be

ignored. For any set S and formula <jl we have:

THEOREM: [Determinate Tight Separation]: 3X[Tight(X) &

'v'z[!z E X ~ !(z E S & <jlz)]].

(Proof: Let the '<jlx' in [Separation] be '3y[!(y E S & <jly) & x = Yr.)

Tightenings: As a corollary, every set S has a "tightening":

THEOREM: 3X[Tight(X) & 'v'z[!z E X ~ !Z E S]].

11.6 RECOVERING CLASSICAL ZF

The above is the full general theory, designed to accommodate

both objects and sets that may be indeterminately identical. There

are complications, such as the constraints on Separation and

Replacement, but for familiar purposes, familiar techniques remain

valid. For example, for considering the foundations of classical

mathematics, one usually works with a pure version of ZF. Classi-

cal ZF is equivalent to a subtheory of IZFU. TIlat is, within IZFU

one can define the hereditarily pure tight sets, which correspond to

the pure sets of ZF. Define a hereditarily pure tight set of rank a

as follows.

Sets with Indeterminate Identity 191

HPT,,(x) <=> Tight(x) & Vy[!y E x ~ 3 ~ [ ~ < a & HPTp(y))).

The hereditarily pure tight sets are then given by:

HPT(x) =df 30,: HPT,,(x).

The HPT sets resemble the usual ones from ZF; the first few are:

One of rank 0:

One of rank 1:

Two of rank 2:

Four of rank 3:

etc.

o

[0}

[0], [0, [0]]

[0, [0}, [0, [0] }}, {0, [0, {0]] }, [10}, [0,

[0] } }, { [0, {0}] }

One can easily establish by induction that both identity and mem-

bership are bivalent relations between HPT sets. Further, suppose

that our primitive predicates are limited to 'E', '=', and 'ranks'. Then

if all quantifiers are relativized to HPT sets, the axioms given above

all hold, and yield HPT sets. For example, if 'x ~ y' is redefined as

'Vz(HTP(z) ~ (z E X ~ Z E y»' then the power set axiom holds,

and yields an HPT set. Most importantly, the restrictions on [Sepa-

ration] and [Replacement] concerning DDIFF and ranks vanish.

So one can simply take over all classical results of ZF into this

theory.

1l. 7 RELATIONS

In set theory, relations are usually represented as sets of ordered

pairs. We can do the same here, so long as we have the right con-

strual of ordered pairs. The usual account is to represent the

ordered pair (x,y) as follows:

(x,y) = [ {x},[x,y}}.

However, in our bivalent framework, this does not satisfy lhe basic

identity conditions for ordered pairs, given by:

(a,b) = (c,d) <=> a = c & b = d.

This fails when we have 'la = b & vc = b & !--.a = c. For then it

follows that 'la E Ib,c} and vc E {b,aj, and so [Set Essence] entails

192 Sets with Indeterminate Identity

that 'lla,b} = /c,b}.Another application of [Set Essence] then yields

that '11 [b},[a,b} J = I [bJ,[c,bJ}. But by the proposed definition of

ordered pairs we could then infer that 'le (b,a) = (b,c». This is wrong,

since !--,a = c.

A solution is to use a slightly more complicated definition of

ordered pairs:

(x,y) =df [x, y, [x,ly} J J.

On this construction, the identity conditions for ordered pairs are

satisfied.

In §3.1 we assumed that if r is any relation and x any object,

there is a property got by "plugging up" r with x: the property of

bearing r to x. This is validated in our framework for relations

among individuals. To show this, we need some additional

definitions:

OP(x) =dr ::3y::3z[x = (y,Z)]

Reln(r) =dr 'v'x[x E r ~ !OP(x)]

Domain(r,x) =df !Reln(r) & 'v'Z[Z E x ~ ::3y«Z,y) Er)]

Range(r,x) =df !Reln(r) & 'v'z[z E x ~ ::3y«y,z) Er)]

We can then prove:

Thm: !Reln(r) ~ ::3XDomain(r,X)

Thm: !Reln(r) ~ ::3XRange(r,X)

In the set theory developed here it is not always possible to produce

a set from a relation by "plugging up" one of its places with an indi-

vidual or a set; that is, if r is a relation and x an individual or set

then there may be no set whose elements are defined as being those

things y such that (x,y) E r. This is because such a putative set may

violate the condition called [Rank Limitation of Indeterminate

Members]. However, if r is a relation limited entirely to individ-

uals, it is impossible to violate this condition, and so we may always

produce a set from a relation in the described manner. So we do

have:

Thm: !Reln(r) & 'v'x'v'y[(x,y) E r ~ x E J & Y E I] ~ 3S'v'z[z E

S ~ (x,z) Er].

This is the principle that was appealed to in §3.1.

Sets with Indeterminate Identity 193

11.8 CONCEPTUAL SETS

We postponed the treatment of conceptual sets from § 11.2. Their

theory is a simple variation of the theory of worldly sets. In particu-

lar, since membership in them is not expected to make a differ-

ence to the identity of their members, we can eliminate [DDiff

E Set] from the set of axioms governingthem.This then lets us elimi-

nate the restriction in [Comprehension] (in the theory of objects

and sets) or in [Replacement] (in the full theory of the hierarchy

of sets) requiring that (j) satisfy DDIFF. If we do this, then these sets

cannot play the role of properties in the Leibnizian account of iden-

tity. They may, however, be identified with the conceptual proper-

ties mentioned in §4.4.

11.9 SETS AS PROPERTIES?

Suppose that we think that worldly properties are extensional,

in the sense that if properties p and q determinately hold of

exactly the same objects, and determinately do not hold of exactly

the same objects, they are identical. (And if one of them determi-

nately holds of an object that the other determinately does not

hold of, they are distinct. And otherwise, it is indeterminate

whether they are identical.) People usually do not consider

properties to be extensional in this sense, but that may be because

they are thinking of conceptual properties. If we distinguish

between conceptual properties and worldly properties, then I am

not aware of any compelling argument to the effect that worldly

properties are not ex tensional. Suppose that they are, and let us

temporarily use the notation 'x E Y' to mean that the individual

object x has the property Y. Then the theory developed above is a

theory of objects and their properties. It is a formalization of the

framework of the world put forth in Chapters 2 and 3, with the

additional thesis that properties are identical if they are possessed

by the same objects. (This is in addition to the principle that prop-

erties are identical if they themselves have the same properties.) It

also assumes the principle of plenitude for properties discussed in

Chapter 7.

194 Sets with Indeterminate Identity

If properties are not extensional, then the set theory described

above can be weakened to a theory of properties by changing the

major connective of Set Essence from a biconditional into a con-

ditional. So far as I can see, none of the earlier discussion turns on

the question of whether or not properties are extensional.

12

Higher-Order Indeterminacy

] 2.1 WHAT IS HIGHER-ORDER

INDETERMINACY?

Is there such a thing as "higher-order" indeterminacy? In the

context of the present book, the question arises in the following

form: Some identity puzzles have answers, and some do not: each

answer to a puzzle of the latter kind ("true", "false") is such that

it is indeterminate whether it is correct. So indeterminacy arises

because of questions that have no answers. Now are there identity

puzzles of which we wish to say that it is indeterminate whether

they have answers (i.e. whether any of "true", "false", or "indeter-

minate" is correct)? If so, this seems to be a kind of higher-order

indeterminacy; it is indeterminacy about indeterminacy.

Of course there are puzzles in which we are confused about

whether or not they have answers. But this may be only epislemic

uncertainty, not metaphysical indeterminacy. Is there any persua-

sive case to be made for meaningful questions of identity such that

it is metaphysically unsettled whether they have answers? I see no

natural application of this idea to certain of the questions discussed

earlier, such as whether I am or am not identical with my body. But

certain of the others seem to turn on matters of degree, and one

might want to appeal to higher-order indeterminacy here. Suppose

that in the ship case you are sure that if the ship with new parts

very much resembles the old ship, and if 95 per cent of the old

planks were used to construct the newly assembled ship, then there

is no answer regarding which later ship is identical to the original

ship. But suppose, instead, that 10 per cent of the planks are

retained in the ship with (mostly) new parts, and suppose that the

newly assembled ship is made of 80 per cent of the original planks,

with the other 10 per cent thrown away. Then perhaps you will be

196 Higher-Order Indeterminacy

uncertain whether or not this is a case in which there is no answer.

If you think that no additional information will help, you might

think that this is because of how the world is (or is not). And so

the same sort of consideration that drove us originally to think that

sometimes there are no answers to identity questions might lead us

to think that sometimes no answer is determined as to whether we

have an example of that sort.

One thing is clear: if we are driven in this way to second-order

indeterminacy, then third-order indeterminacy will follow as well,

with only increasing bewilderment to cloud our minds to this

consequence. And fourth-order, and so on, without limit. So we will

then have arbitrarily high levels of indeterminacy.

I suspect that there is no such thing as higher-order indetermi-

nacy. Where there appears to be higher-order indeterminacy, there

is instead epistemic uncertainty about whether there is determi-

nacy-or else we may be dealing with a case of linguistic vague-

ness (which is a quite different phenomenon, and one that I see as

much more difficult to address than indeterminacy). However it is

hard to be certain about this, so it is worth investigating what

higher-order indeterminacy would be like. In this chapter I explore

how to extend the theory of the previous chapters to include

higher-order indeterminacy.

First, there is a conceptual dilemma that needs to be addressed.

Suppose that we have a case in which we are tempted to say that

it is indeterminate whether S is true, false, or indeterminate. If it is

indeterminate whether S is true, false, or indeterminate, then it

seems to follow that it is indeterminate whether S is true or false,

and thus it seems to follow that S is (simply) indeterminate. So

higher-order indeterminacy collapses to ordinary indeterminacy,

and there is no additional higher-order indeterminacy after all.

Although I suspect that this may be the right answer, our task

here is to make sense of second-order indeterminacy, not to dispute

its existence. So we need to examine ways around this "collapse".

In particular, we must decide what status to attribute to a pro-

position such that it is indeterminate whether it is true, or false, or

indeterminate. It seems that this status cannot be simple indeter-

minacy, because then there is no higher-order indeterminacy, as

argued above. So there must be some additional status in addition

to: true, false, and indeterminate. Let me distinguish this additional

status with the term 'unsettledness'. A proposition is unsettled if

Higher-Order Indeterminacy 197

there is no answer to the question whether it is true, or false, or

indeterminate, and so a proposition which is unsettled is thereby

not either true, or false, or (simply) indeterminate. This idea is

explored in the next section.

An alternative might be to see second-order indeterminacy as

compatible with first-order determinacy (or indeterminacy). That is,

one might hold that it is indeterminate whether there is an answer

to the question, though perhaps there is (or is not) an answer to

the question. Again, using 'unsettledness' for second-order inde-

terminacy, a proposition may be unsettled and also have anyone

of the statuses true, false, or indeterminate. This alternative seems

odd to me, but it is a coherent one. I discuss it in the second section

following.

12.2 UNSETTLED NESS AS A STATUS

DISJOINT FROM TRUTH, FALSEHOOD,

AND INDETERMINACY

To keep the first-order and second-order notions separate, let me

continue to speak of the world determining (or not determining)

whether a state of affairs holds or not, and let me speak of it being

settled or not settled in each given case whether the world does or

does not determine whether a state of affairs holds. If the relevant

state of affairs is settled, then a proposition describing it will be

true, or false, or indeterminate; otherwise it is unsettled. (We can

no longer consider simple indeterminacy a case of having no truth-

value, since that exhausts the options and leaves no room for levels

of indeterminacy. Our old indeterminacy must now be a status

such as "definite indeterminacy", with unsettled ne ss a kind of

"indefinite indeterminacy".) There are now, in effect, four mutually

exclusive truth-statuses (though perhaps 'truth-status' is no longer

appropriate; there are now simply four statuses):

true

false

undetermined

unsettled

198 Higher-Order Indeterminacy

A proposition is second-order undetermined iff it is unsettled.

Our notation now needs to have its meaning extended so as to

address the new status. I suppose that if a proposition is settled,

then our language works as before. If it is unsettled, we need to

specify what status its negation has, and so on. There is a certain

arbitrariness of formulation here; for the sake of definiteness I

make some particular choices about notation. In parallel with

previous notation, I will use '§5' to mean that it is settled that 5

is true, with the intent that '§5' is false if 5 is unsettled. Then we

use our determinacy notation: '!5' to mean that 5 is determinately

true, but with the intent that' !5' is unsettled if 5 itself is unsettled.

With these two notions we can express the other natural options.

Using 'u' for the unsettled status, some relevant truth-conditions

are given in Table 12.1.

Let us say that our other connectives (conjunction, disjunction,

and negation) retain their old values when their constituents have

settled statuses (when they have statuses other than u), and that

the negation of an unsettled sentence is itself unsettled, and that a

conjunction (disjunction) with an unsettled part is false (true) if the

other part is false (true), and indeterminate (has status '-') if the

other part is indeterminate, and otherwise unsettled. The quanti-

fiers will again be generalizations of conjunction and disjunction. I

skip discussing conditionals for the reasons given earlier: they

introduce complexity without addressing any new metaphysical

Issues.

Identity must now mean coincidence in all respects. We must

assume that there are more ways in which properties are had by

objects, and say:

TABLE 12.1.

determinate settled scttlcdncss

truth truth indeterminacy §S v §--,S v

S !S §S

"S §"S

T T T F T

F F T T

F F F F T

u u F u F

Higher-Order Indeterminacy 199

Object a is (definitely) identical to object b iff for any

property P:

it is true that a possesses P iff it is true that b possesses P

it is false that a possesses P iff it is false that b possesses P

it is indeterminate whether a possesses P iff it is indeterminate

whether b possesses P

it is unsetlled whether a possesses P iff it is unsettled whether

b possesses P

Object a is (definitely) not identical to object b iff for some

property P:

it is true that a possesses P and it is false that b possesses P, or

vice versa

It is indeterminate whether object a is identical to object b iff

a is neither (definitely) identical with b nor (definitely) not

identical with b, and for any property P, it is settled whether a

possesses P iff it is settled whether b possesses P

It is unsettled whether object a is identical to object b iff for

any property P:

it is neither true nor false nor indeterminate whether a is

identical with b.

We now have a logic that incorporates second-order indeter-

minacy (i.e. unsettledness). Since assertion is still assertion of truth,

most of our natural inferences from premisses are unaffected by

the introduction of unsettledness. For example, A v B together

with ...,A still entails B, and so on. Indirect proof, naturally,

needs revising. But if you add premisses to the effect that all

sentences used in the proof are settled, (are true or false or inde-

terminate), then our old argumentation is totally unaltered. What-

ever you do, Leibniz's Law is still valid, and contrapositive versions

of it are not generally valid. So although some material of the

preceding chapters now needs fine-tuning, none of the major theses

is lost as a result of admitting higher-order indeterminacy in the

form of unsettledness. (Of course, if there are interesting identity

puzzles that turn on unsettledness, they need to be addressed from

scratch.)

This approach can be generalized to a theory with any number

n of levels of indeterminacy. One needs to have n truth-value

statuses in addition to T and F; call these Ub ... , u

ll

, with Ul being

the same as our earlier -, and U2 the same as u above. Introduce n

200 Higher-Order Indeterminacy

determinate-truth connectives §!, ... , §'" with §! = ! and §2 = § from

above. Each §k maps

TtoT,

F to F,

Ui to itself if i > k,

U

i

to F if i $; k.

If n = ffi, then there can be indeterminacy of every finite level.

12.3 UNSETTLEDNESS AS COMPATIBLE

WITH OTHER STATUSES

It will help if we have an intuitive picture for a kind of unsettled-

ness that is compatible, for example, with definite truth. Suppose

that we believe in indeterminacy because we are idealists: some

kind of mind (perhaps our own, or perhaps God's or ... ) makes

some things true, other things false, and leaves the rest open (inde-

terminate). Can we raise the question of its being indeterminate

whether something, S, is made determinate, independent of wheth-

er or not S is indeterminate? Yes, we easily can, if we presume a

superior mind which, for each proposition p, either decides that p

is to be true, or decides that p is to be false (by deciding that its

negation is to be true), or decides to let the primary mind deter-

mine whether or not p is to be true, or false, or neither. In the latter

case it is second-order indeterminate whether p is determinate. In

particular, suppose that the superior mind produces a (possibly)

partially indeterminate world, as discussed in earlier chapters. Then

the primary mind gets to produce a partial resolution of that world,

as discussed in §5.4; that is, the primary mind gets to make deter-

minate some (or all) of the states of affairs left indeterminate by

the superior mind. (So a partial resolution differs from a "resolu-

tion" in that not all states of affairs need to be made determinate.)

We call a proposition unsettled if it does not have a truth-value in

the first world, and indeterminate if it lacks a truth-value in the

second world.

To keep the first -order and second-order notions separate, let me

continue to speak of the primary mind (or world) determining (or

not determining) whether a state of affairs holds or not, and let me

Higher-Order Indeterminacy 201

speak of the superior mind bestowing pre-emptive truth or pre-

emptive falsehood on propositions. There are now, in effect, five

truth-statuses:

Pre-emptive status:

pre-emptive truth

pre-emptive falsehood

the superior being decrees truth

the superior being decrees

falsehood

Non-pre-emptive status:

(non-pre-emptive) truth

(non-pre-emptive) falsehood

no value

the subordinate being

decrees truth

the subordinate being de-

crees falsehood

the subordinate being does

not decide

A proposltJon is second-order undetermined iff it is neither

pre-emptively true nor false. Such a proposition may be true, or

false, or indeterminate.

Notation for the five statuses:

pre-emptive truth: pT

pre-emptive falsehood: pF

ordinary truth: T

ordinary falsehood: F

lack of truth-value:

Our linguistic notation needs to have its meaning extended. I will

introduce '(;1' for 'it is pre-emptively true that', and I shall interpret

determinate truth as either truth or pre-emptive truth. The truth-

conditions will then be as in Table 12.2.

It is natural to say that a conjunction is pre-emptively true if

both its parts are pre-emptively true, and pre-emptively false if

either part is pre-emptively false. If neither of these conditions

apply, it is non-pre-emptively true if each part is true or pre-

emptively true, and false if at least one part is false. A disjunction

is pre-emptively true if either part is pre-emptively true, and pre-

emptively false if both parts are pre-emptively false. Otherwise it

is non-pre-emptively true if either part is non-pre-emtively true,

and pre-emptively false if both parts are true (pre-emptively

or non-pre-emptively). The quantifiers are to be the natural

202 Higher-Order Indeterminacy

TABLE 12.2.

pre-emptive determinate

truth truth pre-emptiveness determinacy negation

S pS !S pS v p ~ S !Sv ! ~ S ~ S

pT pT pT pT pT pF

pF pF pF pT pT pT

T pF T pF T F

F pF F pF T T

pF F pF F

generalizations of conjunction and disjunction. Again, 1 skip

discussing conditionals.

We must now characterize identity. The most natural treatment

is to do so in a fashion parallel to that of the last section, so that

identity means complete coincidence of statuses of possession of

properties. But now that we have two kinds of truth, we will have

two kinds of complete coincidence: pre-emptive complete coinci-

dence, and non-pre-emptive complete coincidence. Informally, an

identity will be pre-emptively true iff for each property P, either P

pre-emptively holds of both a and b, or it pre-emptively fails to hold

of either. Truth in general (pre-emptive or non-pre-emptive) will

mean that for each property P, the status of whether P holds of a

is the same as the status of whether P holds of b. The remaining

conditions should parallel those of the last section.

We again have a logic that incorporates second-order indeter-

minacy (lack of pre-emptiveness). If assertion is still assertion of

some kind of truth, (pre-emptive or non-pre-emptive) most of our

natural inferences from premisses are unaffected by the introduc-

tion of the new level. For example, A v B together with -,A still

entails B, and so on. Indirect proof naturally needs revising. But if

you add premisses to the effect that all sentences used in the proof

are true or false or indeterminate, then the old argumentation is

recovered unaltered.

Leibniz's Law is still valid, and contrapositive versions of it are

not generally valid.

As in the preceding section, although some material of earlier

chapters now needs fine-tuning, none of the major theses is lost as

a result of admitting higher-order indeterminacy in the form of lack

Higher-Order Indeterminacy 203

of pre-emptive determinacy. (Also, if there are interesting identity

puzzles that turn on lack of pre-emptive determinacy, they need to

be addressed from scratch.)

To generalize this account to a theory with any number n of

levels of indeterminacy one needs n levels of pre-emptive truth and

falsehood; call these Tj, ... , TIl and F

j

, ••. , Fm where, for example,

TJ = T above and Tz = pT above. And we add n pre-emptive-truth

connectives tJ], ... , tJ,,, with tJI = ! and tJz = tJ from above. Each tJk

maps

- to Fb

Ti to itself if i ::::: k, and likewise for F

j

Ti to Fk if i < k, and likewise for F

j

•

12.4 CONCLUSION

I hope at least to have made a case that the possibility of higher-

order indeterminacy does not threaten the theorizing of the rest of

this book. This is because if there is no higher-order indeterminacy,

then the theory is unaffected, and if there is, the theory may be

adapted in one of the ways given above.

APPENDIX

Evans on Indeterminacy:

An Unorthodox Interpretation

In Chapter 4, I indicated that Evans (1978) is subject to various interpre-

tations, including ones irrelevant to the topic of this book. The purpose of

this appendix is to describe such an interpretation and explain why it is

plausible that this is what Evans had in mind.'

1. THE ISSUE

Evans (1978) proposed to refute the claim:

v(a = b),

which he reads metalinguistically as "the identity statement 'a = b' is

of indeterminate truth value". Most commentators' have understood

this to be a claim that the identity statement 'a = b' has no truth-value

(or has some value distinct from truth or falsity); the sentential operator 'v'

combines with a sentence to produce a larger sentence that is true

if the ingredient sentence lacks truth-value, and false otherwise. So the truth

of'v(a = b)' amounts to the lack of truth-value of ' a = b', and Evans meant

to argue that this cannot happen if the terms 'a' and ob' actually refer to

objects. I will call this the "standard interpretation" of Evans's paper.

The "non-standard" interpretation was first suggested to me by David

Wiggins.

3

It is not much endorsed in print.' On this interpretation, inde-

, I am indebted here to Peter Woodruff (as nsnal), and to Peter Smith and an

anonymous referee for Analysis for criticisms of an earlier flawed account of these

matters.

1 Including, among others: Broome (1984), Cook (1986), Cowles (1994), Garrelt

(1988), Johnson (1989), Keefe (1995), Lewis (1988), Noonan (1990), Parsons (1987),

Parsons and Woodruff (1995), Thomasson (1982), van Inwagen (1988), Woodruff

and Parsons (1997). Several authors have written commentaries that do not address

this issue specifically. Pelletier (1989: 482) considers the possibility that indetermi-

nacy might be consistent with having a truth-value, but rejects this as a reasonable

option (ibid. 485-6).

J Personal communication, 1987.

, Gibbons (1982) interprets Evans in this way, and concludes that the resulting

view is "strange". Pellctier (1989) quotes David Lewis as reporting a communica-

Evans on Indeterminacy 205

terminacy is not a lack of truth-value; indeterminate truths are a subset of

the truths. To take an illustration wholly unrelated to identity, one might

hold that it is not determinate whether it is raining, without thereby

denying that it is either true that it is raining or false that it is raining. (If

indeterminacy is contingency, this is a claim that many have found plausi-

hIe; likewise if indeterminacy means lack of knowledge.) On this inter-

pretation 'vS' means that S is neither definitely true nor definitely false,5

though it is either true or false.

Recall the central argument that occupied us in Chapter 4. It is a reduc-

tio in five steps:

(1) v(a=b)

(2) h[v(a =x)]b

(3) -,v(a = a)

(4) -,h[v(a =x)]a

(5) -,(a = b)

The hypothesis to be refuted

Abstraction from (1)

Truism

Reverse abstraction from (3)

(2), (4) by Leibniz's Law

The point of this proof is that from the assumption that it is indeterminate

whether a = b, we conclude that -,(a = h). According to Evans we thereby

"[contradict] the assumption from which we began". But how? Since line

(5) does not explicitly contradict (I), Evans explains how to go on to com-

plete the proof. He says:

If 'Indefinitely' and its dual 'Definitely' C",,') generate a modal

logic as strong as SS, (1)-(4) and, presumably, Leibniz's Law,

may each be strengthened with a 'Definitely' prefix, enabling us

to derive

(5') ",,-,(a = b)

which is straightforwardly inconsistent with (1).

These remarks have puzzled almost everyone who has read them. They

seem to suggest that Evans finds plausible the modal principles of system

S5 of modal logic with '",,' playing the role of necessity and 'v' the role of

possibility. But then consider line (3):

(3) -,v(a = a) Truism

This "truism" asserts the analogue of "it is not possible that a = a". This is

not a truism, it's the opposite of one. Further, if 'v' were the dual of' ",,' then

we could replace 'v' by '-,,,,,-,' in (3), getting:

tion fTOm Evans saying that he intended his signs for determinacy llnd indetermi-

nacy as modal ones; Pelletier argues that this option is, in essence, an error on

Evans's part.

5 Evans uses 'definitely' and 'determinately' as synonyms.

206 Appendix

(3') -,-,t>-,(a = a)

or

(3") t>-,(a = a).

And if 't>' plays the role of necessity in modal logic, this would entail

(3"') -,(a = a)

and we could derive this absurdity from a truism alone, without even using

the hypothesis that is supposed to be refuted. Clearly something is wrong

here.

The explanation is simply that Evans made a slip. Several commenta-

tors have speculated that this is so, and Evans confirmed it himself later.

6

But exactly what slip? He seems to have replaced an appeal to 'it is deter-

minate thal S' with 'it is determinate whether S'.TIle latter is the dual of the

connective '\7', but the former is what is needed for his remarks to make

sense. What he intended appears to be this, using a sideways triangle for

'definitely';

If 'not definitely not' and its dual 'Definitely' generate a

modal logic as strong as SS, (1)-(4) and, presumably, Leibniz's Law,

may each be strengthened with a 'Definitely' prefix, enabling us to

derive

(5')

which is straightforwardly inconsistent with (1).

Evans then supposes a modal logic in which definite truth (symbolized with

plays the role of necessary truth. The rightwards triangle for 'defi-

nitely' is analogous to our determinately true connective 'I', except that

Evans's symbol is a modal operator that, unlike 'I', combines with a sen-

tence having truth-value, making a true sentence with some of these and

a false one with others.

Indeterminacy is definable in this logic as:

Indeterminacy is thus an analogue of contingency, not of possibility.

Why is this interpretation better than the standard one? Because it

makes the argumentation of the article make better sense than on the stan-

dard reading. Of course, Evans may have been confused, in which case an

interpretation making good sense of the article may falsify it. Let us take

that possibility for granted; it is still of interest to see if there is an inter-

pretation that makes good sense.

6 In a letter to David Lewis, reported in Pelletier (1989: 482).

Evans on Indeterminacy

2. DIFFICULTIES WITH THE

STANDARD INTERPRETATION

207

There are four difficulties with the standard interpretation. The first diffi-

culty is that Evans cites Leibniz's Law while using a contrapositive version

of it. On the standard interpretation, he is reasoning within a non-bivalent

framework in which it is a fallacy to appeal to a principle to justify a con-

trapositive application of it.

The second difficulty is the unjustified assumption (emphasized in

Chapter 4) that the abstracts introduced in the proof stand for properties

that obey the abstraction principles and also stand for the real sorts of

properties in terms of which identity is defined.

The third difficulty is that, on the standard interpretation, there is a very

simple extension of the proof from step (5) to a step that completes the

reductio: infer 'o--,(a = b)' from (5). But Evans does not do this, and he

writes as if to do so is problematic; he implies that a more complex justi-

fication is needed, one that appeals to a fairly elaborate set of modal prin-

ciples, including the reduction principles of S5. Why?

The fourth difficulty is that Evans appeals to a modal extension of

the proof by alluding to a modal logic "as strong as S5". This clearly

assumes that the reader will know what S5 is. But on the standard

interpretation, the proof is given in a non-bivalent language, and thus

one needs to know what a formulation of S5 in non-bivalent logic is. This

is hardly something that any writer would expect readers to be familiar

with.

(There is a fifth difficulty, perhaps already implicit in the fourth. Evans

stated later' that he intended the indeterminacy symbols in the original

argument as modal operators. A typical modal operator combines with

truth-valued sentences to produce truth-valued sentences, not with sen-

tences that lack truth-value altogether. And a typical modal operator will

combine with some true sentences to produce truths and with others to

produce falsehoods. But the determinacy connective, on the standard inter-

pretation, does not do this.)

3. SOLUTIONS TO THE DIFFICULTIES

The non-standard interpretation avoids these difficulties. First, on the non-

standard interpretation, the language that Evans is using is bivalent.

7 According 10 Pelletier (1989).

208 Appendix

Indeterminacy is not lack of truth-value, it is possession of truth-value but

in a certain way. In a bivalent logic, if a principle is valid, so is its contra-

positive. So even though Evans does cite Leibniz's Law while using a con-

trapositive version of it, in bivalent logic the contrapositive version is

entailed by the Law itself. If this is an error at all, it is a minor infelicity in

exposition-there is no logical error at all.

Second, what about the abstracts? On the non-standard interpretation,

the abstraction steps are redundant; they could be skipped completely so

long as Leibniz's Law itself is thought to apply to all singular terms. For

(5) can be inferred directly from (1) and (3) by the (valid) contrapositive

version of Leibniz's Law. So why are the abstracts there at all? Answer:

They are there to make explicit that Leibniz's Law does indeed apply to

the terms in question. This can be done in one of two ways, and I am not

sure which Evans had in mind. First, the point of the reformulation could

be to show (claim) that it is not relevant that the names are within the

scopes of modal operators. Since the names are supposed to be rigid, and

occur de re, being within the scope of a modal operator is irrelevant to the

appeal to Leibniz's Law later on; the abstraction steps show this by remov-

ing the names from the scopes of the operators. On this interpretation,

the question of whether the abstracts refer to properties is not germane;

the abstraction is rather intended to give an equivalent form in which the

names in question occur in overtly extensional contexts. The second option

to suppose that Evans had in mind the kind of property called conceptual

in Chapter 4. If this is so, then the abstraction steps are completely justi-

fied (as argued in Chapter 4). And since the logic is now bivalent, the use

of Leibniz's Law does not depend on what kind of property the abstracts

stand for, so there is no fallacy.

Third, why not infer ',....,(a = b)' from '....,(a = b)' and be done with it?

Well, recall that the reason for inferring '!....,(a '" b)' from '....,(a = b)' in

Chapter 4 depended on interpreting '!' as the determinate-truth connec-

tive, which combines truly with any true sentence. But Evans's determi-

nacy operator, ',', does not combine truly with any true sentence; it

combines truly with some and falsely with others. So Evans does not have

available this rule of inference:

S

This is because the form'S & ....,.S' is consistent in his logic. Then why does

he say that line (5) contradicts line (1)? Because it does! But not because

any old sentence of the form'S' contradicts '....,.S·; line (5) is a special case

of this pattern, and has a special role in the proof. For this reason, it is

going to take some explanation of why the proof shows that line (5) con-

tradicts line (1). And that is why Evans gives such an explanation, couched

Evans on Indeterminacy 209

in terms of a sketch of how to expand the original proof so as to show this.

In fact, a correct proof can be given along the lines he sketches, using a

modal logic with the principles of SS. (That is, it is completely correct as a

proof; we will return in §3 to question whether it is reasonable for him to

presume that the SS principles hold.) The proof is within a bivalent logic,

which thus avoids the fourth difficulty mentioned above.

The proof is supposed to show that step (5) contradicts step (1). The idea

behind the proof is easy to explain with an analogy from modal logic.

Suppose we start with an assumption that '5 is possible', and then we derive

' .... ,s' from this. Does that or does it not refute the possibility of'S'? It does,

but for a subtle reason, since '-,5' does not in general contradict 'OS'.

Indeed, '....,S & OS' is generally a consistent form. But if you derive '....,5' from

'OS', using only principles of modal logic, that does refute 'OS'. If the pos-

sibility of S entails the falsehood of'S', then'S' is not possible. At least this

is so in the modal system SS. And that is Evans's strategy in his proof. The

proof does not show that (5) contradicts (1) because of the way their oper-

ators relate to one another; it shows that (5) contradicts (1) because (5)

must contradict (1) if it logically follows from (1), and the proof already

given in (1) .... (5) has shown precisely that.

The reasoning just sketched is not obvious, and so Evans shows how to

prove it from more basic principles. He doesn't give the proof, but he

sketches the strategy; it involves prefixing "definitely" operators to the

lines of the original proof. His technique works fine if one takes for granted

certain derived principles of SS, substituting 'definitely' for 'necessarily'.

They are these:

Principles from ss:

[Interchange] Definite equivalents may be interchanged anywhere.

[Definition] '"S' is (definitely) equivalent by definition to &

-,t>"S'

[SS-Reduction] Every sentence that begins with an operator (or the

negation of such a sentence) is definitely equivalent to the result of

prefixing that sentence with E.g. '"S' is equivalent to and

'....,,,S' is equivalent to

[NecessitationJ If'S' is proved from logical principles alone, you may

infer

[Rule-Necessitation) If'A' follows logically from 'B' and 'C', then

follows logically from and

Here is a proof that results by following Evans's directions. It is slightly

simpler than what he suggests, because r have eliminated the abstraction

steps; this assumes that Leibniz's Law applies in an unrestricted manner

to precise rigid names. and that 'a' and 'b' are examples of such names.

210 Appendix

1. 'l(a=b)

2 .• 'l(a = b)

3. a =a

4 .• (a=a)

5. -''l(a = a)

6. .-''l( a = a)

7 .• -.(a = b)

8. -.' (a = b) & -.o-.(a = b)

9 .• -.(a = b) & -.'-' (a = b)

The hypothesis to be refuted

From 1 by S5- Reduction

Logically true

Necessitation from 3.

From 4 by definition of ''1' and

propositional logic

From 5 by necessitation

From 2 and 6 by the rule-neces-

sitation of the contrapositive of

Leibniz's Laws

From 1 by definition of ''1'.

From 7 and 8 by propositionallogic

The original steps (1) and (3) are retained here, and the new steps 2, 4, and

6 are, as Evans says, the results of prefixing the old (1), (3), and (5) by the

'definitely' operator. The additional steps just spell out some of the addi-

tional details. So this proof is an implementation of the sketch that Evans

gives, and it does reach a full reductio by employing principles taken

directly from SS.

4. WHAT DOES THIS ACCOMPLISH?

What does Evans's paper show if it is interpreted as above? Clearly, it pro-

vides a formal refutation of any view that accepts the premisses and rules

of proof that are used. But which views do this?

8 In case this step looks fishy, here is how to get it from Leibniz's Law alone,

without appeal to any derived principles, by using the S5 principle for conditionals:

.[A::::> B]::::> [oA::::> .B].

We have:

2 .• v(a=b)

6. = a)

We get to 7 as follows:

6a. a=b::::> [v(a= b)::::>v(a=a)]

6b.

6c .• [v (a = b)::::> = a)::::> = b)]}

6d. ov(a = b)::::> = a)::::> = b)]

6e .• = a)::::> (a = b)]

6f. = a)::::> • = b)

7 .•

An instance of Leibniz's Law in

bivalent logic

From 6a, contraposing by

propositionallogic

From 6b by necessitation

From 6c by the S5 principle for

conditionals

From 2 and 6d by modus ponens

From 6e by the S5 principle for

conditionals

From 6 and 6f by modus ponens

Evans on Indeterminacy 211

The view discussed in this book is untouched by the argument. This is

because the argument is not relevant to it. The view discussed in this

book holds that indeterminacy in a state of affairs results in lack of

truth-value of any sentence articulating that state of affairs; the view

under attack in the argument holds that indeterminacy in a state of affairs

results in a truth-value for a sentence that articulates it, though not deter-

minately. These are too different for a refutation of one to carry over to

the other.

Evans's argument is sometimes discussed as a refutation of the view

that there can be vagueness in the world, at least, vagueness of identity.

It is a viable option (though certainly not inevitable) to think that vague

sentences have truth-values, and that of such statements, some are vague

and some are not. (For example, Williamson (1994) holds that all mean-

ingful statements have truth-values, and some are vague and some not.)

Suppose we take the antonym of 'vague' to be 'precise', and we take 'pre-

cisely' to mean 'precise and true'. If we then interpret the 'definitely' of

the argument above to mean 'precisely', it appears to be a view according

to which no identity statement with singular terms that are used de re can

be vague.

A natural response to the argument as a refutation of vagueness is to

say that it runs afoul of higher-order vagueness. A statement may be vague

without being precisely vague (it may be vague even if it is vague whether

it is vague). If so, the reduction rules of S5 would not apply to a logic of

vagueness, since you should not be able to infer ','VS' from ''VS'. However,

if this were the only objection to Evans's argument, the argument would

still be of great interest to the vagueness theorist, for two reasons. First,

the case against the S5 reduction principles is not that firm; for example,

Fine (1975: §5) develops systems for vagueness both with and without

these principles. Second, although the above argument does not refute

vague identities, it constrains them in a severe fashion. Here is why. The

reduction postulate of SS is used only once, in step 2 above. Suppose we

reclassify line 2 as a premiss. Then every step in the proof is valid, and it

results in a contradiction. What this now shows is not that line 1 is incon-

sistent, but that lines 1 and 2 together arc inconsistent. 111at is, the nega-

tion of line 2 follows from line 1. The proof shows, in other words, that this

is valid:

'V(a=b)

But the following is trivially valid (recall that ',S' entails'S'):

'V(a=b)

"",'....,'V(a = b)

212 Appendix

Conjoining these gives:

v(a=b)

= b) & = b)

which, by definition of 'v' is:

v(a=b)

vv(a = b)

The argument shows, in other words, that if it is vague whether a is b, it is

vague whether it is vague whether a is b. So higher-order vagueness is not

only possible, it is forced. Further, this argument iterates. Take the argu-

ment just given and replace 'v(a = b)' everywhere by 'vvCa = b)', and that

argument, slightly expanded, is also valid. So here are all the things that

can be inferred from the assumption that it is vague whether a is b:"

9 Here is a sketch of how these results may be proved. It is easiest 10 do this by

employing some lemmas. The logic in question is SS except that none of the reduc-

tion axioms is used. So in addition to the logic of the ordinary propositional calcu-

lus we assume the following principles:

Necessitation of theorems of logic:

Necessary truths are true:

Definition of 'v' in terms of '. '.

infer '.S' if'S' is proved by logical

principles alone.

from '.S' infer'S'.

Necessitated contrapositive of Leibniz's Law (see earlier footnote):

From .<I>a and .-,<I>b infer .-,a = b.

Lemma A: The following are all theorems of the logic:

II

p----,vva =a

[>---,v,,"a:::: a

etc.

The following proves the first two of these by a pattern of proof that is easily

generalizable:

1. a = a

2 .• a =a

3. -,-,[>a=a

4. -,-,r>a:::: a v -,-,I>---,a = a

5. -,(--,t> 0 :::: a & -,t>--,{{:::: a)

6. -,va=a

7. p---,va=ll

8. -,-,!>---,va =0

9. -,-,t>'VO:::: a v -,-,t>---,'Va:::: a

10. -,(-----,I>va=a&---,l>-----,va=a)

11. -,vva = a

12 .• -,vva = a

13.

Truth of logic

1, Necessitation

2, propositionallogic (Double negation)

3, propositionallogic (Addition)

4, propositionallogic (DeMorgans)

5, Definition of , v'

6, Necessitation f-

7, propositionallogic (Double negation)

8, propositional logic (Addition)

9, propositionallogic (DeMorgans)

10, Definition of'v'

11, Necessitation f-

,,(a=h)

",,(a = h)

"",,( a = b)

""",,( a = b)

"""",,( a = b)

Evans on Indeterminacy 213

If vagueness is associated in any way with difficulty in knowing (as most

vagueness theorists hold), then vagueness of identity entails maximal igno-

rance of every sort about itself: if it is vague whether a is h, it is vague (you

can't know?) whether it is vague, and it is vague (you can't know?) whether

it is vague whether it is vague, ... I don't know quite what to conclude

Lemma B: The following is always a correct inference:

<t> /:. ~ • .,<t>

Proof:

l.<t> Given

2. Suppose • .,<t>

3 . .,<t>

Hypothesis for refutation

4. <t>&.,<t>

By the "truth of necessities": .S /:. S

1,3 propositionallogic

5 . ., • .,<t> RAA 2-4

With these lemmas in hand we can show from va = b these all follow:

vva=b

vvva =b

We give a proof for the first two conclusions, establishing a pattern for the rest:

O. va=b

1 .• ~ v a = a

2. Suppose .va = b

3 .• ~ a = b

4. ~ • .,a=b

5. ~ . v a = b

6. ~ • .,va = b

7. ~ . v a = b & ., • .,va = b

8. vVil=b

9. t>---,'\lVa ={l

10. Suppose .vva = b

11 .• .,a = b

12. ~ • .,a = b

13. ~ . v v a = b

14. ~ • .,vva = b

15 . .,.vva = b & ., • .,vva = b

16. vvva = b

17.

Premiss

Lemma A

Hypothesis to be refuted

Necessitated contrapositive of Leibniz's

Law

0, definition of 'v'

RAA2-4

0, Lemma B

5,6 propositionallogic

7, definition of 'v' f-

Lemma A

Hypothesis to be refuted

Necessitated contrapositive of Leibniz's

Law

0, definition of 'v'

RAA 10-12

Lemma B

13,14 Propositi anal logic

15, definition of , v' f-

214 Appendix

from this, but it certainly gives vagueness of identities a status that other

vaguenesses do not have. (Recall, however, that the above proofs are given

within classical logic, and they will be irrelevant to at least some non-

classical accounts.)

REFERENCES

Broome, John (1984) "Indefiniteness in Identity", Analysis 44: 6-12.

Burgess, J. A. (1989) "Vague Identity: Evans Misrepresented", Analysis 49:

112-19.

--(1990) "Vague Objects and Indefinite Identity", Philosophical Studies

59: 263-87.

Cook, Monte (1986) "Indeterminacy of Identity", Analysis 46: 179-86.

Copeland, B. Jack (1995) "On Vague Identity, Fuzzy Objects, and Fractal

Boundaries", Southern lournal of Philosophy 33: 83-96.

Cowles, David (1994) "On Van Inwagen's Defense of Vague Identity",

Philosophical Perspectives 8: 137-58.

Dummett, Michael (1978) Truth and Other Enigmas (Cambridge, Mass:

Harvard University Press).

Evans, Gareth (1978) "Can There be Vague Objects?",Analysis 38: 208.

--and McDowell, J. H.( eds.) (1976) Truth and Meaning (Oxford: Oxford

University Press).

Fine, Kit (1975) "Vagueness, Truth, and Logic", Synlhese 30: 265-300.

French, Steven, and Krause, Decio (1995) "Vague Identity and Quantum

Non-Individuality", Analysis 55: 20-6.

GarreU, B. J. (1988) "Vagueness and Identity", Analysis 48: 130-4.

--(1991) "Vague Identity and Vague Objects", Nous 25: 341-51.

Geach, Peter (1967) "Identity", Review of Metaphysics 21: 3-13.

Gibbons, P. F. (1982) "The Strange Modal Logic of Indeterminacy",

Logique et Analyse 25: 443--6.

Johnson, Bruce (1989) "Is Vague Identity Incoherent?", Analysis 49:

103-12.

Keefe, Rosanna (l995) "Contingent Identity and Vague Identity",

Analysis 55: 183-90.

--and Smith, Peter (1996) Vagueness:A Reader (Cambridge, Mass,: MIT

Press).

Kripke, Saul (1975) "Outline of a Theory of Truth ",.lournal of Philosophy

72: 690--716.

Lewis, David (1988) "Vague Identity: Evans Misunderstood", Analysis 48:

128-30.

--(1993) "Many, but Almost One", in John Bacon, Keith Campbell,

and L10yd Reinhardt (eds.), Ontology, Causality and Mind: £$says in

Honour of D. M. Armstrong (Cambridge: Cambridge University Press):

23-42.

216 References

Lowe, E. 1. (1994) "Vague Identity and Quantum Indeterminacy",

Analysis 54: 110-14.

Machina, K. F. (1976) "Truth, Belief and Vagueness", Journal of

Philosophical Logic 5: 47-78.

Martin, R. L., and Woodruff, P. (1975) "On Representing 'True-in-L' in L".

Philosophia 5: 213-17.

Melior, D. H., and Oliver, Alex (eds.) (1997) Properties (Oxford, New York:

Oxford University Press).

Noonan, Harold (1982) "Vague Objects", Analysis 42: 3-6.

--(1984) "Indefinite Identity: A Reply to Broome", Analysis 44: 117-2l.

--(1990) "Vague Identity Yet Again", Analysis 50: 157-62.

--(1991) "Indeterminate Identity, Contingent Identity, and Abelardian

Predicates", Philosophical Quarterly 41: 183-93.

--(1995) "E. 1. Lowe on Vague Identity and Quantum Indeterminacy",

Analysis 55: 14-19.

Over, D. E. (1984) "Vague Objects and Identity", Analysis 49: 97-9.

Parsons, Terence (1984) "Assertion, Denial and the Liar Paradox",.lournal

of Philosophical Logic 13: 137-52.

--(1987) "Entities without Identity", Philosophical Perspectives 1: 1-19.

--(1996) "Fregean Theories of Truth and Meaning", in Matthias Schirn

(ed.), Frege: Importance and Legacy (Berlin, New York: Waiter de

Gruyter).

--(1997) "Meaning Sensitivity and Grammatical Structure", in M. L.

Dalla Chiara et af. (eds.), Structures and Norms in Science (Kluwer,

Dordrecht): 369-83.

--(2000) "Indeterminacy of Identity of Objects": An Exercise in

Metaphysical Aesthetics", in Alex Orenstein and Petr Kotatko (eds.),

Knowledge, Language, and Logic: Questions for Quine (Kluwer,

Dordrecht): 213-24.

--and Woodruff, Peter (1995) "Worldly Indeterminacy of Identity",

Proceedings of the Aristotelian Society 1995 (Winter): 171-9l.

Peacocke, Christopher (1981) "Are Vague Predicates Incoherent?",

Synthese 46: 121-4l.

Pelletier, Francis Jeffrey (1984) "The Not-so-Strange Modal Logic of

Indeterminacy", Logique et Analyse 27: 415-22.

--(1989) "Another Argument against Vague Objects", Journal oj"

Philosophy 86: 481-92.

Quine, W. V. (1981) "What Price Bivalence?", Journal of Philosophy 78:

90-5.

Rasmusssen, Stig (1986) "Vague Identity", Mind 95: 81-91.

Sainsbury, R. M. (1988-9) "Tolerating Vagueness", Proceedings of the

Aristotelian Society 89: 33-48.

--(1989) "What is a Vague Object?" Analysis 49: 99-103.

References 217

--(1991) "Is There Higher-Order Vagueness?", Philosophical Quarterly

41: 167-82.

--(1994) "Why the World Cannot be Vague", Southern Journal of

Philosophy 33 (Suppl.): 63-81.

Salmon, Nathan (1981) Reference and Essence (Princeton: Princeton

University Press).

Stalnaker, Robert (1986) "Counterparts and Identity", Midwest Studies in

Philosophy 11: 121-40.

--(1988) "Vague Identity", in D. F. Austin, Philosophical Analysis

(Dordrecht: Kluwer): 349-60.

Thomasson, Richmond (1982) "Identity and Vagueness", Philosophical

Studies 42: 329-32.

Tye, Michael (1990) "Vague Objects", Mind 99: 535-7.

Van Fraassen, Bas (1966) "Singular Terms, Truth-Value Gaps, and Free

Logic", Journal of Philosophy 63: 481-95.

Van Inwagen, Peter (1988) "How to Reason about Vague Objects",

Philosophical Topics 16: 255-84.

Wiggins, David (1986) "On Singling out an Object Determinately", in

P. Pettit and 1. McDowell (eds.), Subject, Thought, and Context (New

York: Oxford University Press): 169-80.

Williamson, Timothy (1994) Vagueness (London and New York:

Routledge ).

Woodruff, Peter W. (1969) Foundations of Three-valued Logic (Ann Arbor:

University Microfilms).

--(1970) "Logic and Truth Value Gaps", in Karel Lambert (ed.),

Philosophical Problems in Logic: 121-42.

--(forthcoming) " ... And of Sets".

--and Parsons, Terence (1997) "Indeterminacy of Identity of Objects

and Sets", Philosophical Perspectives: 11.

----(forthcoming) "Set Theory with Indeterminacy or Identity",

Notre Dame Journal of Formal Logic.

Zemach, Eddy (1991) "Vague Objects", Nous 25: 323-40.

INDEX

!, defined 17

11, defined 18

[>, defined 206

abstract 47-55,99,116,125,134,160,

176,

20

7-

8

abstraction 23,46-52, g8-g, 116, 164,

20

5--9

adorned reading 138, 143, 146,

153-7

assertion 19,20,37, 79, 102, 107, 109,

136,137,145,168, Igg, 202

biconditional 50,87--9,92,96-7, g8,

100-5, 194

bivalence 12,22,57-60,69,71,

100-6

bivalent 18,24,25,4°,53,57-61,71,

95-6, 102, 110, 144, 186-91,

2

0

7-

10

body 1,7,8,41,42,45,60,101,114,

117-

1

9,11

6

, 163-5, 172, 195

Broomc 79-84, 95-100, 204

building 3-4,75--9,162,164

Burgess 12,46,5°,60,73, 150, 152

cat 8,43,60, 110,120,146,165-70

coincidence 31, 33, 166-9, Ig8, 202

complex 16, 21, 36, 46, 52, ,54, 6,5,

85,13

0

,

2

°7

concept 4,9,13,28,60,79,160--1,

17

0

-7

refinement ofa concept 3, 171-9

conceptual property 55, 78

conceptualism 14

condition [e] 123-6, 130-1

conditional 9, 16, 23-4,36-7,4°,52,

56,72,87-100, 109, 143, 155,

157,194, H)8, 202, 210

conjunction 17, g3, 70, 85, 96, 98,

1,19,141,143,198,201,202

connective 16-18,21-8,40-1, s0-4,

72,90,92,94,96,99, 109, 13

8

,

143, 157, 194, 198,200, 203,

206-8

con traposition 64, 89, gl, 92, 94

contrapositive 26--8,37-8,41,47--9,

SS' 92-4, g8, 100, 110, IS2, IS3,

199, 202, 207-12

Cook 75-9, 204

counting g, 34, 40, 88, 134-7, 166--9

Cowles 23,151,204

data 6,7, g, 12, 29,43,56, ,57, 59, 62,

65,86, 136, Ig8, 140, 142, 14S,

159-7

2

,17

8

DDiff 45, 52-4,116,125,182-6,193

definite description 3,4,49,75, 116,

144, 153-8, 168

denial 20, 109

disjunction 17, 25, 68-9, 78--9, 84-8,

131,132,139, 14S, 198,201,202

disruption 7,8,59, 154,17°,176

Dummett 1 1

Evans 5, 18,45-56,60,75-9,92,

95-8, 10;), 115, 204-11

Fine 63, 163,211

Frege 21,49, 63, 116

fuzzy logic 29

fuzzy object 72-7

Garrell 46,47,96,2°4

Geach 39

Gibbons 47,204

higher-order indeterminacy 10, 72,

195-

20

3

higher-{)rder vagueness 29, 21 1

icon 122-33

idealism 11

if-true reading of conditional 89--96,

10

3

indeterminate object 13

Johnson 98-100, 116,204

Keefe 34,42, 54,

20

4

language 4,5,9,1,5,16,17,18,21,

22, 28, 29, 3

0

, 33, 34, 35, 3

6

, 38,

39,4

0

,64,65,67,69,7

1

,7

2

,79,

96,98, 105, 110, 125, 137, 138,

150, 158, 159, 160, 161, 163,

165, 178, 19

8

, 207

220 INDEX

Law of Excluded Middle 25,69, 78-g,

84

Leibniz's l.aw 24,35-41, 46-9, 55, 87,

g2-100, 143, 1.')2, 153, Ig9, 202,

20

5-

12

Lewis 8, 46, 165-g, 204, 206

Machina 103

McDowell 103

Melior & Oliver 14

mereological identity 73

mind 11,13,14,23,28,38,40,41,

47,58,65,68,78,88, 107, 109,

13°,174,178,196,200,201,204,

208

modality 30, 42, .')0, 204-9

multivalence 18

name 14, 16-17,22,33-.'),39,50,52,

G6,68, 101,151-3,158,172,186,

208, 209

negation 17, 18, 20, 21, 25-8, 36,

46--7,58,62,7 1,82,88,130,

Ig8, 200, 20g, 211, 212

exclusion negation 20,28,47,58

nominalism 14

nonextensional 22,30, 36, 39, 50,

133

Noonan :)4, 39-4 I, 52, 80-4, 204

numerical 138, 140

ordered pair 16,61,191,192

Parsons 4,19,20,21,28,49,52 , 59,

102, Ill, 116, 163, 173, 186, 189,

20

4

part 2-5, 6, 8, 21, 24,42-3, 49, 52,

54, 55, 72-g, 84-7, 9 1, I I I, J 14,

119,120,136,1,\8, 14°--z, 145,

148,152, IG4-5, 167, 170--6, 182,

189,195,198,201,204

p-cats 8, 43, 60, 120, 165-9

Peacocke 103

Pelletier 28, 47, 204, 206, 207

1,7, 8,43,4459,60,

104,110,118,135,136,138,146,

151-8,163,164,170-2,176-

g

pictension 122-6, 130--3

picture 9,34,39, 108, 110-- 17,

121-33, 200

pile of trash 120, 146

pragmatism 12

preempti\'e truth 201, 203

principle of plenitude 33, .'>4, 122,

12

3,

12

5,193

properly 7, 9, 12-16, 28-5.'), 6.')-8, 78,

95, 108, 111-18, 122-32, 142,

145,152,153,182,192-4,198-g,

202, 207, 208

quantifier 1 G, 17, 26, 53, 54, 138,

191,198,201

Quine 12, 56-8

RAA 23-.'), 2 12

rank of a set 186-91

realism 11, 13, 14

reference 9,16,21,46,56,75,77,

108, 116, 150-2, l.')g-64' 204,

208

refinemenl 129-30,171--9

relation 12-13, IS-16, 29, 32-4, 36,

39-40, S4-S, 66--7, 72, 110, 12.'),

128,150-1,191- 2,199

relative identity 34, 39-40

resolution, worldly 61-g, 84-S, 93,

121, 126--30, 144-7, zoo

restaurant 172, 179

robust 122,12.'),182

room 2,7,8,42-3, 118, 146, 158,

16

9

rules 23,24,74,91,95,141, 16g, 210,

211

Ss

20

5-

12

Sainsbury 13

Salmon 60-1, ISI, 18S

semanlic paradox 19-20,28,50, 102

set 1,9,28,34,50-1,58,61-2,123,

130, 181-94

tight set 188, 190-1

Set Essence 182-3,185-6,191-4

sharp object 13

ship 1-6,8,20,42-3,63,110,119,

13

6

-5

1

,17

1

--6,195

sortal 173, 176-8

Stalnaker 7,35,42,45,151,172,

179

state of affairs 5, 11-15. 19,29,67,

106,116,132,197,200,211

status 5, 14, 19, 2G, 30, 39-40, 50, 65,

68-72,87-91,95,97,102,156,

181,196- 202,214

supervaluation 62-8,71,87, 147-8,

161-5,170- 8

INDEX 221

tautology 25, 26,64

Thomassoll 204

truth-making 14

truth-value 5, 17-30, 36, 39-40, 45-6,

49-5

2

,62-7

1

,79,85-9

1

,95-7,

100-10,156-62,172,175-8,181,

199-201,204,206-8,211

understanding 21,45,77,79-81,85,

106-11, If) 1 , 171, 183, 186

unsettled 59, 6o, 195-200

vagueness 1, 3, 13, 28-9, 60-3, 6g,

78--g, 98,1.')1,196,211- 14

validity 22-3,25-8,36-8,40-1,

46-50, 61, 64, 90, 92-3, 103,

10.'),110,190 ,199,202,208,

211-12

Van Fraassen 63

Van Inwagen 1,23,151,153,204

Venn diagram 11 1

wife 6, 177

Wiggins 96, 204

Williamson 69-72, 100-6, 109,

21 I

WoodrufT 19,49,52, 89-go, 102, Ill,

ll6, 173, 180,189,204

world 4,.'),9-15, 28--g, 31, 34, 36,42,

55, 59, 64-5, 67-8, 77-8, 8.'), 100,

109-10, 113, I 16-17, 121-2,

125-613°,139,145,148,15°,

152, 160, 162, 164, 172, 179,

182,193,196,197,200,211

Zemach 40, 59

Zermelo-Fraenkel 58, 18.')-7,189

ZF, classical 190

INDETERMINATE IDENTITY

Terence Parsons presents a lively and controversial study of philosophical questions about identity. Is a person identical with that person's body? If a ship has all its parts replaced, is the resulting ship identical with the original ship? If the discarded parts are reassembled, is the newly assembled ship identical with the original ship? Because these puzzles remain unsolved, some people believe that they are questions that have no answers, perhaps because the questions are improperly formulated; they believe that there is a problem with the language used to formulate them. Parsons explores a different possibility: that such puzzles lack answers because of the way the world is (or because of the way the world is not); there is genuine indeterminacy of identity in the world. He articulates such a view in detail and defends it from a ·host of criticisms that have been levelled against the very possibility of indeterminacy in identity.

Indeterlllinate Identity

Metaphysics and Semantics

TERENCE PARSONS

CLARENDON PRESS· OXFORD

or under terms agreed with the appropriate reprographics rights organizC:\tions. Title. p. Hong Kong Printed in Great Britain on acid-free paper by Biddies Ltd Guildford & King's Lynn .OXFORD UNIVERSITY PRESS Great Clarendon Street. Oxford University Press. scholarship.82-dc21 00-034027 ISBN 0-19-825044-4 (alk. and education by publishing worldwide in Oxford New York Athens Auckland Bangkok Bogota Buenos Aires Calcutta Cape Town Chennai Oar es Salaam Delhi Florence Hong Kong Istanbul Karachi Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi Paris Sao Paulo Shanghai Singapore Taipei Tokyo Toronto Warsaw and associated companies in Berlin Ibadan Oxford is a registered trade mark of Oxford University Press in the UK and certain other countries Published in the United States by Oxford University Press Inc. Terence. Identity (Philosophical concept) 1. BD236 . paper) I 3 5 7 9 10 8 6 4 2 1ypcset by Best-set Typesetter Ltd. in any fOflll or by any n1eans. No part of this publication may be reproduced. Indeterminate identity: metaphysics and semantics / Terence Parsons. Oxford OX2 6DP Oxi'ord University Press is a department of the University or Oxford. 1. or as expressly permitted by law.P36 2000 Ill'. at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Parsons. Includes bibliographical references. It furthers the University's objective of excellence ill research.. stored in a retrieval system. New York © Terence Parsons 2000 The moral rights of the author have been asserted Database right Oxford University Press (maker) First published 2000 All rights reserved.. or transmitted. without the prior permission in writing of Oxford University Press. cm. Enquiries concerning reproduction outside the scope of the ahove shllUld be sent to the Rights Department.

To Peter Woodruff .

.

These conversations led to a study of VennEuler-like diagrams for "picturing" indeterminacy. much of its theoretical content was originally developed in cooperation with Peter Woodruff. This required me to give a systematic overview of the topic of indeterminate identity. starting from scratch for people not already immersed in the literature. and with the ongoing literature on the subject. In doing this. the determinacy of identities might be presupposed by other deeply held beliefs of ours. resulting in Parsons (1987). We believed that it is coherent to hold that identity statements might be indeterminate. but in many cases I am not myself able to say whether I have new exposition of . A few years later I returned to the topic. I also needed to work out how the various thoughts that we have had on the topic mesh with one another. Besides. and this led in turn to an investigation of indeterminate set theory. and that all of the a priori proofs to the contrary are clearly question-begging. Some of my own contributions (such as the notion of super-resolutions) can be identified. In Spring 1998 I had an opportunity to lecture on Indeterminate Identity at the University of Salzburg. I am solely responsible for matters of exposition. and Peter Woodruff and I began discussing it in detail. which is summarized in Chapter 11. Those lectures led to this book. Although the book is singly authored. it is not an easy idea to grasp. for example. and for various developments of the theory. I needed to indicate how my/our views are to apply to the identity puzzles that figure so prominently in the literature. We also wanted to set the basic theory on a sure footing. We decided that we stood in need of some overall conception of what it would be like for some identities to be indeterminate. diagrams that appear in some of our joint papers and in Chapters 7-9 of this book. These diagrams have often guided us in development of basic positions on matters of logic and semantics.PREFACE I first addressed the topic of indeterminate identity in a short paper in the mid-1980s. But there might very well be other considerations that mitigate against this.

or new ideas. who is almost indeterminately a co-author.c. and Kyle Stanford. Terence Parsons u. of course. particularly: Jason Alexander. I wish to express my thanks to the Salzburg Philosophy department for its invitation to lecture there. When I take a point to be common knowledge.viii Preface old ideas. Gary Bell. Penelope Maddy. Patricia Marino. is to Peter Woodruff. I have not tried to trace its exact origins. Jeff Barrett. Thanks are also due to those Fall 1998 participants in the UCI Logic Workshop who monitored and critiqued a presentation of a draft of this book. My greatest debt. I also want to emphasize that many of the general arguments given here in defence of the coherence of indeterminate identity are already present in some form in the literature. Irvine 1999 .

Understanding Indeterminacy 8. Introduction 2. and DDiff 5. Non-Conditional Disputations 6. Counting Objects 9. Properties. Higher-Order Indeterminacy 195 Appendix: Evans on Indeterminacy Reference Index 204 215 219 . Identity 4. Denoting Objects x 1 11 31 45 56 87 107 134 150 160 181 10. The Evans Argument. Alternatives to Indeterminate Identity 11. Indeterminacy 3.CONTENTS /Inalytical Table of Contents 1. Sets and Properties with Indeterminate Identity 12. Conditional Disputations 7.

is the person after the change identical with the person before the change? (iv) given a cat with imprecise boundaries. INTRODUCTION Identity puzzles persist without consensus on any solution. A semantics is described for a logical notation in which sentences may lack truthvalue. This book explores the idea that they have no solutions because the identities are indeterminate-there is no fact of the matter about whether or not they hold. The sign 'v' represents indeterminacy: '''IS' is true if'S' lacks truth-value and is otherwise false. which cat-like thing with precise boundaries is identical to the actual cat? The purpose of the book is to explore the view that there are no answers to these questions because of indeterminacy in the world (as opposed. and a sentence reporting it lacks truth-value. which of the two resulting ships is identical to the original? (iii) if a person undergoes a crucial change. such as whether a certain object has or lacks a certain property. or pragmatism. to imperfections in our language). and is false if'S' is either false or lacking in truth-value. If neither of these is the case. for example. 2.ANALYTICAL TABLE OF CONTENTS 1. Four puzzles without determinate answers are discussed throughout the book: Ci) is a person identical with that person's body? (ii) if a ship has its parts replaced and the old parts are assembled into a ship. then the state of affairs is indeterminate. INDETERMINACY Indeterminacy in the world is a genuine option within idealism. Indeterminacy pertains to states of affairs. The logic of such a language . It contains a connective'!' for determinate truth: the sentence '!S' is true if'S' is true. realism.

though you will be able to infer '-.!S' is always valid. It is an empirical matter whether there is any indeterminacy at all. but you can infer '-. PROPERTIES. Likewise. If neither of these holds. not that its negation '-. but '5 V -. but there is no property for which they determinately disagree. But it does not validate the contrapositive form of this law: one might have 'Q>a' and '-. This explanation of identity validates Leibniz's Law: From 'a = b' and '$a' one can infer 'Q>b'. Classical indirect proof does not hold. and an empirical matter whether such indeterminacy. in each case in which the identity is indeterminate. 3.Analytical Table of Contents XI is described. unlike genuine identity.B'.a = b' being true. AND DDIFF Gareth Evans's 1978 argument disproving the existence of indeterminate identity is discussed. then it is indeterminate whether the objects are identical. THE EVANS ARGUMENT. in which validity means guaranteed preservation of truth. 4.!A' from '-.S' by indirect proof.. and they are distinct if one determinately has a property that the other determinalely lacks. this can happen if 'Q>x' does not express a property. IDENTITY Real identity in the world is defined in terms of coincidence of properties: objects are identical if they both determinately have and determinately lack the same properties. since they lack truth-value if '5' does. extends to identity. contrapositive reasoning is not generally valid: if you can infer' B' from 'A' you may not be able to infer '--. So you cannot infer '-.)5'.5' is [rue. there is a property determinately possessed by one object that is not determinately possessed by the other. The four identity puzzles are reviewed. The identity discussed here is not "relative identity" because relative identity. does not validate Leibniz's Law. if it exists.A' from '-. since if'S' entails a contradiction this proves only that '5' is not true.. On a simple analysis it appeals to . Classical tautologies such as '5 V -S are not generally valid.Q>b' both true without '-.B'.

A test for whether a formula '<jlx' expresses a worldly property is whether it satisfies the principle that joint (determinate) satisfaction and (determinate) dissatisfaction of the formula makes a "Definite Difference" in the identity of the objects in question. since little of methodological value is lost in allowing that claims may lack truth-value. instead it is akin to the "paradoxes" of na"ive set theory. 'Ax<jlx'. An argument of Salmon's is reviewed.Xll Analytical Table of Contents the (invalid) contrapositive version of Leibniz's Law. on a deeper analysis it presumes incorrectly that the formula 'v(x = a)' expresses a property (the property of being indeterminately identical with a). constructed from it stands for a worldly property.. along with 'n = y'. or that it stands for a non-worldly "conceptual" property. If a formula satisfies this principle. NON-CONDITIONAL DISPUTATIONS Various objections to indeterminate identity are not conclusive. 5.<jli are both true. due to the fact that identity is defined in terms of global quantification over all (worldly) properties. Quine's doctrine "No entity without identity" rules out only entities for which identity does not make sense. that there are no objects x and y such that '<jlx' & '!. discussed in Chapter 1l. it is not aimed at indeterminacy. The proof is an RAA of the hypothesis that that formula expresses a property.. This fact is not due to non-extensionality or to anything like the semantic paradoxes. Holding on to bivalence for methodological reasons is not compelling. it is either a version of Evans's argument. already discussed.. identity puzzles may still arise where sciences overlap. or a challenge to indeterminate set theory. Even if bivalence is maintained within each science. Sometimes puzzles arise from reading sentences non-literally.. and many physicists have (rightly or wrongly) adopted this stance towards some phenomena in quantum mechanics without hampering their methodology. that is. then an abstract.. otherwise we can either say that it does not stand for a property. A sentence may be read supervaluationally: it is treated as true (false) if all ways of making its vocabulary determinate yield readings .

hypothetical syllogism. These options account for data used by som( writers in an attempt to refute indeterminate identity. modus tol/ens. Objects mayor may not bE subject to these principles. anc contraposition. this does not logically entail that it is indeterminatel~ identical to something. but not 'A ~ B'. (No non-bivalen truth-status-functional conditional satisfies all of modus pOllens . e. A conditional 'If A ther B' is to be read either with this conditional CA ~ B') or giver the "if-true" reading ('lA ~ B'). inconclusive. A purportec refutation of non-bivalence by Williamson is discussed. those claims are plausible if reac super-resolution ally.g supervaluationally. and thE Lukasiewicz conditional '~' is adopted. It may also be read super-resolutionally: ii is treated as true (false) if all worlds that result from some way 01 making our own world completely determinate would make thE sentence true (false). but only a restricted form of conditional proof taking 'A' as a premiss and deriving 'B' allows you to condition alize and infer '!A ~ B'. Several authors point out that if an object has indeterminatE boundaries. Some mereological principles are giver that would fill in this logical gap. The Lukasiewicz conditiona satisfies modus ponens. 6. hi~ disproof of indeterminate identity is examined and found to b. it is arguec I hat his claims that all contradictory sentences must be falSE are plausible only if they are read in some special way.Analytical Table of Contents xii which are true (false). but then they do not conflict with indeter minate identity. An argument by Noonan against an example (due to Broome of indeterminate identity of clubs is shown to be implausible if thE claims in it are read literally. anc true if the truth-value status of the consequent is at least as higl as that of the antecedent (counting lack of truth-value as a statu: intermediate between truth and falsity). An example due to Cook is aIlegec to be hypothetically of this kind (given his assumptions). CONDITIONAL DISPUTATIONS Truth-conditions for conditionals are discussed. such a conditional i: false when the antecedent is true and the consequent false.

but the "if-true" version is correct: 'la = b ~ (<j>a ~ <j>b)'. 10hnson criticizes these versions. . the bare Lukasiewicz conditional form is incorrect. under certain general conditions these refinements picture the resolutions discussed earlier in giving super-resolutional readings of sentences. One can picture situations involving indeterminacy using Vennlike diagrams in which objects are represented by images with finite size. Picturings of the paradigm identity puzzles are given. but gives a conditionalized-conditional version that is also equivalent.) If one wishes to state Leibniz's Law as a conditional.iect is pictured as being indeterminately P if its image lies partly inside of and partly outside of the region representing property P. 'identical' is not intended in a special sense. Broome gives a different version. UNDERSTANDING INDETERMINACY The terminology used in explaining indeterminate identity is meant to be ordinary.xiv Analytical Table of Contents modus tollens. Simple constraints on the picturing conventions entail that Leibniz's definition of identity in terms of coincidence of properties is built into the picturing. this has nothing special to do with identity. The biggest impediment to understanding views that invoke indeterminate identity is our tendency to reason bivalently. so one can do no better. which is logically equivalent to this one. a simple condition yields a kind of principle of plenitude for properties and the Leibnizian account of identity. An ob. A more general notion of picturing can be defined that is not necessarily two-dimensional. A second argument of Williamson is considered which defends bivalent versions of the Tarski biconditionals. 7. objects are pictured as being indeterminately identical if their images properly overlap. and conditional proof. A process is given to refine pictures into more determinate ones. it is held that his rationales for the biconditionals Clre subject to interpretCltion. but is otherwise taken as Cl primitive. 'Indeterminate' may be definable within certain world views (such as idealism). Clnd that interpretations that do not make them beg the question rationalize only non-bivalent versions.

When it is indeterminate whether a = b. Familiar formulas are given for making cardinality claims. then it is not possible to determinately denote a.. and indeterminate whether there are exactly two (or exactly three). we sometimes get no determinate answer. It is shown how to get the "right" answers. and terms that determinately denote a while also indeterminately denoting b. COUNTING OBJECTS I[ we try to count objects in the face of indeterminacy. DENOTING OBJECTS Some authors have suggested that if it is indeterminate whether a = b. 9. Super-resolutional readings also explain certain of our intuitions. these formulations correspond to two natural "right" answers. all identity sentences that lack truth-value suffer from some semantic defect. "there are at least two <\>'s" is written as ':3x:3y (--. So all options are possible. Sometimes a question can be formulated in two ways: either austerely. e. and terms that indeterminately denote each.g.g. that in the ship case it is true that there are at least two ships. or indeterminacy of identity (producing indeterminacy regarding whether an object has already been counted).Analytical Table of Contents xv 8. this is due to indeterminacy of predication (producing indeterminacy regarding which objects are supposed to be counted). e. it is shown how to have singular terms that determinately denote a without denoting b at all. 10. ALTERNATIVES TO INDETERMINATE IDENTITY Many writers believe that it is both possible and desirable to account for the lack of truth-value of identity sentences by appeal to indeterminacy in the semantics of our language or of the COIl- . as a consequence. or with a determinacy connective ('1') added. or both. false that there are more than three.x = y & <\>x & <\>y)'.

A distinction is made between worldly sets and conceptual sets. it ought to be true that exactly one person entered the room. versions of it are refuted. 11. and it is shown that there is indeterminacy of identity between such sets if there is any worldly indeterminacy at all. even if there is no indeterminacy of identity between objects. it does not seem to give the right pattern of truth-values in all cases.xvi Analytical Table of Contents cepts employed by our language. If we construe worldly properties as extensional. SETS AND PROPERTIES WITH INDETERMINATE IDENTITY Sets are things whose identities are defined in terms of their members: they are identical if they both determinately have and determinately lack the same members. the account (not endorsed by Lewis) is objectionable. For example. as formulated. Last is an account due to Stalnaker according to which we refine our concepts upon demand when faced with puzzle cases. and otherwise it is indeterminate whether they are identical. in the case of a personal disruption. This is the most promising approach. they are distinct if one of them determinately has a member that the other determinately lacks. then they can be . the former but not the latter obey the DDiff principle for set membership: that if x is determinately a member of Sand y is determinately not a member of S. and true that exactly one left. then x is determinately distinct from y. A theory of objects together with worldly sets of those objects is formulated. it does not apply to the puzzle cases in general. Salmon's argument against indeterminacy is shown to fail within this theory. and we are left uncertain whether it can be developed into a viable al tema ti ve. A second account is based loosely on a discussion by Lewis of how many cats exist in a given region. without its being either true or false that the person who entered is the person who left. It is shown here that several natural accounts of this sort do not yield a lack of truth-value in the puzzle cases without also wrongly yielding lack of truth-value in other cases as well. An account is examined that employs supervaluations to analyse indeterminacy in singular terms. more importantly.

On this reading. HIGHER-ORDER INDETERMINACY We have indeterminacy if there is a question that has no answer. Following a suggestion of Wiggins. . but in case there is. this shows the consistency of an assumption made in Chapter 3 that if R is a relation between individuals. The theory can be extended to a transfinite hierarchy of sets. using a refinement of the usual definition of ordered pairs. Appendix: Evans on indeterminacy Some of Evans's remarks about modalitics near the end of his 1978 paper have puzzled many commentators. we have higher-order indeterminacy when it is indeterminate whether a certain question has an answer.Analytical Table of Contents xvii identified with the sets under discussion. so that 'vS' may be true when'S' is true. Relations are definable as sets of ordered pairs. his argument is not relevant to the thesis of indeterminate identity discussed in the subsequent literature. 12. two accounts of it are offered. The theory from earlier chapters can be accommodated within either account. then bearing R to object 0 is a property. it is shown that they make good sense if we assume that Evans saw indeterminacy as a kind. yielding an indeterminate version of Zermelo-Fraenkel set theory. of truth. A theory of conceptual sets results from removing the DDiff condition placed on worldly sets. and the resulting theory is a formalization of the framework of objects and properties and identity sketched in Chapters 2 and 3. the classical ZF theory is satisfied in a definable subdomain of this theory. or mode. There might not be any higher-order indeterminacy.

.

. or . if the person lives on after the body no longer exists. plank by plank. they are distinct. regardless of their instincts. and if the person ceases to exist at death but the body continues in existence. persist without adequate solutions. .) I This way of construing the puzzles carries out a theme of van Inwagen (1988).. with sufficient cases in between to bewilder just about anyone. Each puzzle is actually a vague description of a spectrum of cases. is that the ship that originally set sail? If a person has a brain transplant. or has the old person ceased to exist. and some of which give rise to the opposite answer. and a host of puzzles like them. and while at sea it is completely rebuilt. which describes a cabinet in which a person undergoes disruptive changes.1 IDENTITY PUZZLES Throughout history philosophers have puzzled over questions of identity: Is a person identical with that person's body? Of course. some of which give rise to one answer. so that you are uncertain whether the person who emerges is the person who entered.1 Introduction 1. is the resulting ship with new parts the ship that originally set sail? What if the discarded pieces of the original ship are assembled into a ship. or a memory transplant. they are distinct. Van Inwagen suggests that anyone can fine-tune what happens inside the cabinet so as to yield a case about which the fine-tuner is uncertain. But what if they always coexist? Suppose a ship sets sail. thus the puzzles persist. is the resulting person the same person who antedated the operation. to be replaced by another? These.

I may have to do this in order to get my favourite desk into a new room.)2 So sometimes reassembling parts yields the original object. mixing in a 2 If you are inclined to be sceptical about even such a case as this. [ have added the twist that the repair/replacement takes place at sea. if I replace the radiator in my car with a new one I think r still own a car that I bought several years ago. and the original ship has apparently been repaired by having new parts installed. The "ship of Theseus.. suppose that the new parts of the repaired ship are quite unlike the parts being replaced. I created it after the parts were last washed. Sometimes we can disassemble an object into parts and then reassemble those parts into the original object. this is a way to move my desk into the new room. . making the ship with new parts unlike the original ship-as unlike as you need to weaken your judgement of identity." So you must admit that you talk as if an object can be reassembled. (If you are not convinced by this.) If you are sure the "reassembled" ship is the original ship. So sometimes replacing a part yields the original object. not one that just sprang into existence. this is not essential to the example.3 example above illustrates a case in which our judgements about identity are supposed to be in clear conflict: the original ship has apparently been disassembled and reassembled. consider what you would say if asked where you purchased your blender. Perhaps you see one of the options as being clearly superior to the other? Then a minor adjustment in the case will bring them into conflict: If you are sure the ship with new parts is the original ship. That is enough to open the door to theorizing on the identity question. It. But these cannot both be true. (You may also wish to suppose that the activity is part of a contest to see how creatively the contestants can disassemble a ship and reassemble it. 3 A ship example originates in Plutarch's Life of Theseus. This was updated by adding a second ship by Hobbes in De Corpere.2 Introduction Consider the ship case. think of taking apart a blender to wash it and then putting it back together again. because two non-identical ships have resulted. And you must admit that this is serious talk. 2. You would never say "I didn't purchase it. not a way of destroying it and replacing it with a new desk. worth explaining. just reassemble most but not all of the discarded parts. But sometimes I repair an object by replacing a part. thus favouring the reassembled ship as being the original ship. so the idea that what you say is plausible because it is literally true is an option well worth exploring.

leave some of the original parts in the ship with (mostly) new parts. Suppose that Old Ivy Hall has an addition built onto it. most philosophers try to devise a subtler criterion for identity preservation that avoids such extreme judgements. Instead. These are extreme positions. none of that discussion will be reprised here. or explain why normal intelligent people are wrong to reject them. and the new large building is named Postmodern Hall. there is never any such thing as repair by replacement of a part. Others propose that continuity of size.Introduction 3 few new ones. If necessary. they are often admired for the ingenuity that goes into defending them. so that there is never any such thing as reassembly. shape. Because so many identity puzzles have remained unsolved for centuries. and any change of parts whatever leads to non-identity. some observers have been led to speculate that these are questions that have no answers. and this motivates us to look elsewhere. Adjust the number of parts upward or downward until you are no longer sure about what to say. Some philosophers have been driven by examples of this sort to provide a method for answering any such question. tripling it in size without altering the structure of the original in any significant way. Sometimes this is plausible. typically because they incorporate a definite description that does not uniquely denote. that's it". whenever the criterion is applied. Proponents of such views then must either refine them. and function are required to maintain identity. Such methods (when not overly vague) give natural answers in the problematic cases. The most popular methods are the simplest. My office is in the old part and yours is in the new . some propose that having the same parts is necessary for identity. It is plausible in the "building" case. Why don't they have answers? What would it take for them to have answers? The most popular option in this century is that the questions have no answers because they are improperly formulated. But that raises other questions. It is clear that none of the methods has won popular acceptance. but other clever philosophers inevitably devise new cases in which the "subtle" methods ride roughshod over our surest judgements. but they have few real adherents. For example. a criterion that will allow us to say "Aha. A great deal of ingenuity has gone into the defence of and attack on such views.

you could say that the definite description 'the original ship' fails to uniquely denote. and also in Postmodern Hall. . and quite implausible that persons are parts of other persons. The ship with new parts is distinct . Yet this is what it would take to solve the harder cases in parallel fashion. 1 examine some closely related but more plausible views in Ch. but rather that there is no answer at all. and these are not the same building. However. This is certainly a natural option. and that this is because the definite description 'the building in which Parsons has his office' does not uniquely denote. or in the concepts embodied in our language. my main task is to explore an alternative that I find more plausible. and one that I need to discuss (Chapter 10). and Old Ivy Hall is now a part of Postmodern Hall.4 Introduction part. Not that there is an unknown answer. the facts are these: there is a unique original ship. This solution is plausible because buildings are parts of other buildings. because there were actually two original ships: one that was later repaired with new parts. My office is in Old Ivy Hall. And so the identity question is ill-formed.2 WORLDLY INDETERMINACY OF IDENTITY I am inclined to think that an identity question can be completely coherent and well formed and yet lack an answer because of the way the world is (or because of the way the world is not). Someone asks "Is the building in which Parsons has his office the building in which you have your office?" It seems clear that this has no unique correct answer. In the ship case above. Buildings can be parts of other buildings. and one that was later reassembled from the original parts. 4 I do not believe that a systematic diagnosis of identity puzzles in terms of imperfections in our language. and there is a unique newly assembled ship. For example. This is not plausible. 1. one is part of the other. will be satisfactory. It is less plausible to think that ships are parts of other ships (at least for normally designed ships). 10. So there is no such thing as the building in which I have my office. there is a unique ship with new parts.

That is the task of this book. defended. and so on. or whether the ship with new parts is the original ship. to express these in a common vocabulary. And so on. and likewise for the ship with new parts. Thus there is genuine indeterminacy of identity in the world. The literature on this subject has now matured to the point where an extended look at the issues seems both timely and feasible. Each of the sentences listed above non-defectively reports a state of affairs.Introduction 5 from (non-identical with) the newly assembled ship. The first state of affairs is made not to hold by the way the world is. There is little contained here that has not already been argued in the literature in one form or another. the indeterminacy is real. stimulated almost wholly by a one-page article by Gareth Evans (1978). revised. Discussion of the issues so far has proceeded on a number of fronts. but the second and third are not either made to hold or made not to hold by the way the world is. When there is no fact of the matter. What does not yet exist is a single coherent presentation of a position that articulates and defends worldly indeterminacy of identity. expanded. It is indeterminate whether the original ship now has new parts. to assemble . It is indeterminate whether the newly assembled ship formerly left port when the original ship did. The contribution of this book is to choose from among the positive views a subset that can be maintained together. This proof has been attacked. a sentence reporting the purported fact lacks truth-value. So the following sentences have the indicated statuses: the newly assembled ship = the ship with new parts the newly assembled ship = the original ship the ship with new parts = the original ship False No truth-value No truth-value These indeterminacies must cohere with certain others as well. There is now a growing literature on the question of worldly indeterminacy of identity. This is not an illusion generated by indeterminacy as to how our language fits with the world. giving a proof that there cannot be genuine indeterminacy of identity in the world. But there is no fact of the matter regarding whether the newly assembled ship is the original ship. a state of affairs of identity. and on a piecemeal basis.

and to add a few details that may advance the issues. Methodologically. But it makes for a more coherent exposition if I take a definite stance on the issue. indeed. The bulk of this book will be occupied with these two tasks: formulation of the view and defences of its coherence and prima-facie plausibility in the face of purported refutations. The main task of the book is just that: to formulate a position. Here I can be said to beg the question-But only if I presume a point at issue. 1 begin with ordinary beliefs. And so the ultimate fate of the view will be left uncertain. I will also argue that the position is immune to certain kinds of attacks that have been levelled at it in the literature. It is also part of my methodology that I do not take as data highly theoretical philosophical generalizations. for example. such as "nothing is indeterminate". I take a Peircean perspective. Neither of these tasks. For example. will show the view to be true. which 1 will reject only if some reason is found to challenge them. These are my tentative data: ordinary beliefs-such as the belief that I have exactly one wife. if I am right. This is not because I am convinced it is true. I am convinced that the existence of worldly indeterminacy of identity is both coherent and possible. that I actually have several dogs. I reject philosophical analyses that contradict these judgements. telling me. I cannot be said to beg the question merely by formulating what one answer to the question is. . In presenting the thesis of worldly indeterminacy of identity I speak as an advocate on its behalf. For that. My goal is primarily to articulate the view in detail and in generality.6 Introduction them in a presentation that can be mastered by someone new to the topic. and that exactly one ship set sail before the problematic replacementlrepairlreassembly process. or that there is not really any such thing as a dog-there are only basic particles that swarm into dog-like shapes. that there is exactly one dog in my back yard. Some competing views will be discussed in Chapter 10. what is required is to see whether it explains better than competing views do why certain identity questions seem to be coherent and yet lack answers. and this is immune from certain kinds of criticism. even if successful. but many will not. The articulation of the view I take to be a matter of hypothesis formulation. that is a contingent matter on which nobody can be certain. or the opposites of such views. or "no two things can be in the same place at once".

For example. by distinguishing their temporal properties.Introduction 7 I see these as theoretical observations which are to be validated by how well they conform to the data. and shows how t he puzzles are resolved if we do so. and there is no need for me to duplicate it. Of to apply it exactly as I do. and an indication of what I take the data to be in each. Instead I focus on proposed solutions that "preserve the data". I may occasionally rely on some theoretical views myself. Many proposed solutions to the puzzles involve taking a stand that requires us to reject some of the apparent data. This assumption is crucial. solutions that explain how it is that The ordinary beliefs that we have about the identity puzzle cases are literally true. I will develop one such explanation. The person/body: Assuming that a person exists when and only when their body exists." The I'L'rson/body puzzle discussed here is a hypothetical one: what if there were no . or when I see no other way to resolve an issue. There is an enormous literature devoted to this already. hut only by accident. Stalnaker (1988: 354) says "we can distinguish intimately related things such as an artifact and what it is made of. I do not necessarily expect the reader to share this methodology.S is each person identical with his/her body? . a I'l'rson and his or her body. rejecting the view that exactly one person entered the room in which the disruption took place. I will not discuss such solutions at all.4 A STOCK OF PUZZLES I I will help to have at hand a small stock of cases to discuss. I articulate it here to clarify what I will and will not be doing. for example. A traditional solution explains how and why we should change our beliefs. Let me call any such position a "traditional solution". and I will discuss others. I also limit myself by the working assumption that It is literally true that there is no answer to the identity questions in the puzzle cases. as opposed to the other way around. 1.

and exactly one person left the room. there is no answer to the question whether it is the cat. or whether the original ship is the newly assembled ship. The ship: An assemblylrepair process takes place as described above. and exactly two ships docked. Working assumption: There is no answer to the question whether the original ship is the ship with new parts. The puzzle is: what if they always coexist? 6 Patterned after an example in Lewis (1993). The personal disruption: A person enters a room where something disruptive happens to them that challenges our judgements about personal identity. because they all have different parts. how many cats are on the table when we naIvely think there is only one?) Data: There is exactly one cat. Is the original ship identical with the ship with new parts. if they do not temporaJly coincide. Working assumption: For any given p-cat. then there is exactly one person in the room. There are many p-cats. so they will be definitely distinct. Is the person who entered the room identical with the person who later leaves the room? Data: Exactly one person entered the room. and they are distinct from one another. The cal: 6 It is unclear exactly what the parts of a cat are.8 Introduction Data: If I am alone in a room. Call the object (if any) that answers to any such precise description of its parts a "p-cat". clear examples of temporal properties in which they differ? Of course. Consider all ways of answering such questions precisely. . For example. it is unclear whether a molecule loosely attached to the end of a hair that is engaged in falling out is part of the cat. and exactly one human body in the room. then presumably one will be a person·at-t and the other not a person-at-t. and there are many p-cats. How is the cat related to the p-cats? Is it identical to any of them? To none of them? (Are the p-cats cats? If so. Working assumption: There is no answer to the question whether the person in the room is the body in the room. Working assumption: There is no answer to the question whether the person who entered the room is the person who left the room. or with the newly assembled ship (or neither)? Data: Exactly one ship left port.

la 1. 1 will refer to the above data together with the working assumptions as "extended data". and of how we can uniquely refer to such objects. not to argue for them. I will apply this to indeterminacy of identity. Next. views that locate indeterminacy wholly in language or in our concepts. This is not to insist that they are correct. but rather to emphasize that I will be trying to account for them. Then I look at a number of arguments in the literature that attempt to show that positing indeterminacy of identity leads quickly to inconsistency. Finally I . In both of these enterprises we need to clarify not only what a partially indeterminate world is like. calling the first "coincidence" and the second "persistence". The first and last of these questions involve cases of identity-ata-time. This is followed by a discussion of some variations of the inconsistency argument based on the logic and semantics of conditionals. Then certain alternative views are treated. After this there is a discussion of how it is possible to develop a coherent theory of sets that allows sets to have indeterminate members.·Introductioll 9 You are asked in each case to adjust the details of the example so as make the answer to the identity question most uncertain.l will assume that I am addressing a reader who makes such adjustments to the cases under consideration. and the middle two are cases of identity-a cross-time. I ignore the distinction because it is not relevant to any of the issues that I address. and 1 give a useful "classical picturing" of situations involving indeterminacy. the question is whether such a view can provide a better account of the data than the view that posits real indeterminacy. including indeterminacy of identity. For convenience later. People sometimes distinguish these. but also how a language works that correctly describes such a world. properties with indeterminate identity are also discussed. based on your convictions.5 PLAN OF THE BOOK First T will address what is involved in there being indeterminacy in the world. Following this I discuss the complaint that we cannot conceive of identity's being indeterminate. This is followed by a discussion of how we can count objects whose identities are partially indeterminate.

(It would be what the world would be like if the world were exactly as we naIvely think it is. the reader will find a coherent statement of what it would be like if there were indeterminacy in the world.) . If I am successful.10 Introduction briefly discuss what impact higher-order indeterminacy might have on this enterprise. extending to indeterminacy of identity.

1 WHY TAKE INDETERMINACY SERIOUSLY? The view under consideration sees the world as consisting of states of affairs. Michael Dummettl has made popular the idea t hat a test for a realist view is whether that view embodies "bivalence". . so that the world is a thing created by our minds.. so why assume that (s)he determines everything? It takes a rather special sort of deity to guarantee complete determinacy. and others are simply undetermined. and there will be indeterminacy. Why should one think that the world might be like this? Because it is a possibility. For all we know. or undetermined. Then since our minds are finite. they are not likely to get around to finishing the job. So some aspects of the world will be left uncreated. some definitely do not hold. Suppose that some form of idealism is correct. Some of these definitely hold. This option is sometimes foreclosed by supposing that there is an absolute mind that fills in for us whatever we ourselves leave undetermined. then this deity freely chooses both whether and what to create. Suppose that a sentence expressing an undetermined state of affairs lacks truth-value. the world is like this. (This is an assumption that I will indeed make.) Then it appears that realism is incompatible with indeterminacy by definition! Perhaps this is so for Dummett's brand of realism. the condition that every claim about the world is true or false.2 Indeterminacy 2. But why suppose that the absolute mind determines everything? If the absolute mind is anything like the traditional Judeo-Christian-Islamic deity. in Dummetl (1978: 145-65). but not for mine. Ignore idealism.g. I assume that realism includes any view that how the world is is independent of my mind or how I conceive I e.

while making it difficult to reconcile recalcitrant data. Quine sees the question of bivalence as a pragmatic methodological choice we can make. Burgess (1990). 2. and relations. for a critique of certain contemporary views that reject all indeterminacy for reasons that he finds less than compelling. If this is so. indeterminacy is an option. or it is indeterminate whether 0 has or lacks p. or 0 definitely lacks p. And so we should investigate what the consequences would be if there were indeterminacy. and relations among objects. or it is indeterminate whether 01 bears r to 02' For the moment 3 suppose we speak of 0 's having or lacking p as I ef. the idea that "the world" is whatever is revealed by scientific investigation and theorizing. Quine himself chooses bivalence over its "fuzzy and plurivalent alternatives" (Quine 1981: 94-5) but others may choose differently. and no reason to suppose that the world is completely determinate. Quine (along with other pragmatists) holds a view something like this. properties of objects. while complicating the theory. J I say "for the moment" because in the detailed formulation of the theory I will not need to talk of states of affairs at all. or 01 definitely does not bear the relation r to 02.) Thus bivalence is a choice. Again. For any given objects 01 and 02 and relation r there are also three options: the object o[ definitely bears the relation r to 02. Why should it be? Another outlook is pragmatism. there is no a priori reason to assume bivalence.2 INDETERMINACY IN THE WORLD I suppose that the world contains objects.2 but the point is clear enough: there are plenty of reasonable conceptions of what the world is like that do not preclude at least some indeterminacy. Other views could be surveyed. properties. an abandonment of bivalence permits data to mesh more neatly with theory. (1 am stating Quine's view here. . not endorsing it. A commitment to bivalence in our theorizing leads to methodological elegance and simplicity. I will only need to talk about objects.12 Indeterminacy of it. especially §§JO-lJ. For any given object 0 and property p there are three options: either 0 definitely has p.

see Ch. Indeterminacy emerges at the level of states of affairs. He suggests that one could define a vague object as one such that it is indeterminate whether some sharp property applies to it. Many writers suggest that there can be objects that possess genuinely vague boundaries. In the last case the state of affairs is indeterminate. It just happens that this issue does not bear on anything I have to say.g.4 I do not know how to apply the notion of indeterminacy below the level at which we make whole judgements. except briefly in Chapter 11. e. if one defines an indeterminate object as one for which there is some object with which it is indeterminately identical: x is an indeterminate object whether x = y]. but he doubts that one can clarify what a "sharp" property is. . 1 am inclined to agree. not the objects or properties or relations composing it. The properties under discussion here must be genuine constituents of the world. However. except derivatively. He does not define 'precisification'. There is nothing in the view discussed here which lays the blame for indeterminacy of states of affairs on either the objects or on the properties or relations. Sainsbury (1994) argues more sceptically that there is no substantive thesis about whether there are vague properties. So I will not discuss it. I thus set aside certain discussions in the literature addressing indeterminacy of identity which try to distinguish whether it is objects that are indeterminate. where a "sharp" object is one that does not have two precisific<ltions such that some property definitely applies to one but definitely does not belong to the other. and if we wish to call objects or properties or relations themselves indeterminate. 5. Then these states of affairs either determinately hold (if 0 definitely has p). It is real properties in the 4 Sainsbury (1989) defines a "vague property" as one that neither definitely applies nor definitely fails to apply to a sharp ohject. I emphasize that it is the state of affairs that is indeterminate. or they are not determined either way. not concepts in the mind or mere words. or the relation of identity itself. =df 3y[it is indeterminate I am also neutral about whether or not properties and relations are "intensional" entities. or determinately do not hold (if 0 definitely does not have p). about whether there can be two disfinct properties p and q that determinately hold of exactly the same objects and determinately fail to hold of exactly the same objects. we will mean by that merely that they enter into indeterminate states of affairs. that is. I am not neutral about realism for properties. but the connection between this fact and indeterminate identity is debatable.Indeterminacy 13 a state of affairs.

J am presenting a realist theory that might provide the input to nominalist or conceptualist reductions. (1 do not know how to formulate the theory 1 discuss here in terms acceptable to nominalism or conceptualism. sentences may express states of affairs. directly by "the facts". indirectly. In the most dircct way. or neither. Other sentences are made true. My discussion of identity is neutral among many different versions of such a view. I am not myself presenting a nominalist or conceptualist theory. false if the object determinately lacks the property. by showing how to take talk about real things in the world and explain it in nominalistically or conceptualistically acceptable ways. e.g. except that they must be genuine universals-they must be the sort of thing that are capable of being possessed by many objects. i. Such a sentence is true if the object named determinately possesses the property. For example. but they need to be distinguished from what J am doing here. or false. the sentence may consist of an atomic predicate that stands for a property and a name that stands for an object. an existential "Something is F" whose 5 For a survey of various theories of properties see MeUor and Oliver (1997). I take for granted that properties are universaJs (typically shared by many objects) in various places in this text. and it is otherwise lacking in truth-value. A referee for OUP suggests that the indeterminate identity view may be developed without this assumption. This may be true. by the status of how the object and property are related. I have in mind no particular theory of properties. but J have not explored that option. false.3 TRUTH-MAKING States of affairs can make sentences true in different ways. So my theorizing is avowedly realistic.14 Indeterminacy world that determine whether objects (real objects in the world) are or are not identical (really identical). or neither. ) 2. I have no objection to such projects.e. 5 Nominalists and conceptualists often believe that they can account for any phenomenon that realists can account for. . The sentence is made true.

then we can show that the predicate 'being a thing such that it is indeterminate whether it is /J' will be a predicate that cannot stand for a property. such a predicate does not stand for a property. yet sentences containing them will be made true (or false or neither) by how the world is. and relations. For example. we will need to say in detail how our language works and how it is related to the world. We will be able to prove below that if there is indeterminacy of certain sorts. 4 and 7. there is no inconsistency in holding that 'being a thing slIch that it is indeterminate whether it is b' stands for a property. and it will be made truth-valueless by there not being any determinately holding state of affairs involving an object and the property F. if 'F' stands for a property. n Typically. if it is indeterminate whether a = b. So we must discuss the semantics of the language that will be used in stating our metaphysical theses. we will sometimes need to discuss issues that are formulated in terms of such predicates while being uncertain whether they stand for properties. this is the business of the next section. then we might have a predicate meaning 'is neither determinately F nor determinately not F'. But if b is not indeterminately identical with any object. whether a predicate can stand for a property depends on how the world goes.Indeterminacy 15 predicate 'F' stands for a property will be made true by a state of affairs that determinately holds. there are predicates that do not stand for properties. . if there is not. it will be made false by the determinate non-holding of all states of affairs involving any object and the property F. see Chs. or neither). It turns out that we do not need to speak of states of affairs in order to do this. In addition to predicates that stand for properties. In order to clarify in general how aspects of the world make sentences about the world true (or false. We do need to speak of objects. it may be uncertain whether they stand for properties. When we discuss existing predicates arising from philosophical examples. then some predicates do not stand for properties. For example. There is no reason to expect there to be a property answering to this latter predicate (though there might be)". together with there being some states of affairs of that form that do not either determinately hold or determinately not hold. though it is clear how the truth of sentences containing it are validated (or not) indirectly by the facts. properties. so they will henceforth vanish from our exposition.

) Predicates: A one-place predicate may stand for a property. My position is that they can.) Truth: TIle truth conditions for formulas and sentences are given here in a simple form. and most others will not be concerned. Atomic sentences: An atomic sentence consists of a one-place . and a two-place predicate may stand for a relation. I decree that there is no object such that the predicate is true of it and also false of it. or neither) of objects. and predicates that are true (or false. (Two-place predicates are true of/false of/neither true nor false of ordered pairs of objects. Names: I assume we have a stock of names that name objects. There may be objects the predicate is neither true nor false of. otherwise. for example. I make these assumptions for simplicity of discussion. It contains names of objects. (In Chapter 9. the issues are sufficiently complex that we should not have to worry about whether. or it may not. or it may not. a name lacks reference. and I assume here that they do. I postpone introducing a sign for identity until the next chapter. and I postpone introducing a conditional sign until Chapter 6. So if an object neither determinately has nor determinately lacks the property.16 Indeterminacy 2. the predicate will be neither true of nor false of it. and it is false of those objects that (determinately) lack the property. We need to allow for both options. (Similarly for two-place predicates that stand for relations.g. by defining satisfaction of formulas by infinite sequences of objects assigned to the variables). Most people knowledgeable about logic will know how to rriake this more rigorous (e.) If a predicate stands for a property. and no names whose denotation is ambiguous. There are connectives. I discuss the claim that names cannot unambiguously denote objects which enter into indeterminate identities. and there are quantifiers to express generality. what follows here is the language that will be used throughout the book. There are no denotationless names. then it is true of the objects that (determinately) possess the property.4 LANGUAGE: SYNTAX AND SEMANTICS The symbolism used herein consists mostly of the first-order predicate calculus. In either case I assume that a predicate is true of some (or no) objects and is false of some (or no) objects. Each name names a unique object.

and otherwise lacking in truth-value. For these J choose the Lukasiewicz conventions. . The convention is: '!S' is true jf 'S' is true. and false if it is false of the object named by the name. otherwise the sentence has no truth-value. I assume that all connectives and quantifiers behave in the usual ways when the formulas with which they combine possess truth-value. and discuss alternatives when they are relevant.2. and otherwise false.Indeterminacy 17 predicate and a name. If a sentence lacks truth-value. and a negation of a false sentence is true. We need to choose some particular conventions in order to have a definite framework.1 and 2. COllnectives and quantifiers: There are many different ways to define connectives and quantifiers for a language in which some sentences lack truth-value. I use '!' for this. For familiarity. so does its negation. for their naturalness: Negation: A negation of a true sentence is false. These connectives are summed up in Tables 2. false if cj>x is false on at least one assignment to x. I make my own choice here. Conjunction: A conjunction is true if both conjuncts are true. The determinate truth connective: It will be useful to have a "non-classical" connective for truth. and otherwise ::3x<jlx is lacking in truth-value. and otherwise \lx<jlx is lacking in truth-value. Such a sentence is true if the predicate is true of the object named by the name (or the pair of objects named by the names). \lxcj>x is true if cj>x is true for every assignment to x. The quirks arise only for cases without truth-value. false if either is false. false if cj>x is false on every assignment to x. while remaining sensitive to the fact that other writers may interpret connectives differently. false if both are false. Quantifiers: The quantifiers ::3 and V are understood as generalizations of conjunction and disjunction: ::Jxcj>x is true jf cj>x is true for at least one assignment to x. or a two-place predicate and two names. and otherwise lacking in truth-value. in which the dashes stand for absence of truth-value. Disjunction: A disjunction is true if either disjunct is true.

Thus vS is incompatible with S's being true and with S's being false. defined as: This means that it is not determined whether S. These conventions are summed up in Table 2.3. For example. A sentence of the form 'vS' is true if and only if'S' itself lacks truth-value. S'. That is. Using this notation we can express other ideas.1. since this usually means that there are more than two truth-values.2. the falsehood of'S' is expressed by its negation being (determinately) true: '!--S. originating in Evans (1978): we express a claim that S is indeterminate by writing 'vS'. We can also express indeterminacy. S' is true if'S' is true. though it is probably better called the "determinately" connective. when we state the truth-conditions for the elements of our language: 'Determinately. Indeterminacy TABLE 2. and false if'S' is false or lacks truth-value.18 TABLE 2. you should read '!S' as 'Determinately. For the language used . But it is also not "multivalent" as this is usually understood. A T B T A&B T AvB T T A T -. This is most of the language we will use throughout. The language is not "bivalent" in the usual sense because it allows the possibility that sentences may lack truth-value. The notion of truth gets in only in the metalanguage.A F T T F T F T T F T F T F F F F T F F F F F I call this the "truth" connective. There is an established sign for this in the literature.

Indeterminacy TABLE 2. I take the former option merely because it does not require explaining what an additional "truth"-value is which is neither truth nor falsehood. it mayor may not be redundant to add 'determinately'. or assumes that it has a special "neuter" value. we generally put them forth as true. I omit discussing them.3. so an assertion of'S' will commit the assertor to the claim that the state of affairs that S determinately holds. s This means that for assertions it is generally redundant to use the adverb 'determinately' to modify the whole sentence used to make the assertion. I will occasionally burden my sentences with redundant uses of 'definitely' and 'determinately' so that the reader will not have to pause to figure out whether determinacy is entailed (or required) or not. With regard to subclauses of assertions. but this usually takes a special convention. Since it is not always immediately obvious whether it is redundant or not. true and false.5 DETERMINACY IN NATURAL LANGUAGE ASSERTIONS A sentence expressing an indeterminate state of affairs lacks truthvalue.s F F T indeterminacy vS F T F F here. . though there are three truth-value statuses: true. as I do. and neither. false. When we assert sentences. and Parsons (1984) for natural languages. there are only two truth-values. S It is possible to commit oneself merely to the non-falsehood of the proposition asserted. These options are especially relevant to work on the semantic paradoxes. How do we deal with indeterminacy in real life discussion? 7 For most purposes it does not matter whether one assumes that a sentence which is neither true nor false has no truth-value. 7 2. See Woodruff (1970) for a study of "hedged assertion" in formal languages. however. 19 S T (determinate) truth !S T F F (determinate) falsity !--. Since the paradoxes are not at issue here.

because that would make you take sides.9 Unlike normal negation. or "That can't be determined". 'not S'. usually with a special emphasis on negation. it does not endorse its opposite (its ordinary negation). is the kind of negation that we are trying to express when we reject a claim with special intonation in order to distance ourselves from taking sides regarding it. One option is to retreat to a meta-Ievel and say something like "You can't say one way or the other". if need be. but it's not not the same either. you might say "It's not the same. exclusion negation is used in such a way that 'ex-not S' is an assertion that is true when S lacks truth-value (Table 2. by denying the opposite as well. we will feel called upon to say more. or even "It's not the same. A second option is to "deny" the claim. In this book I take the conservative option.4). This. as opposed to an act which asserts the different content "that it is not the same ship". I have argued in Parsons (1984) that in some recalcitrant cases (cases involving semantic paradox) we are forced sometimes to see denial as a speech act separate from assertion. but it's not different either. which is exactly what you think should not be done. which is to see this kind of denial as assertion of the special kind of negation called "exclusion negation". which lacks truth-value if S does. I here take the simpler line of preserving a single kind of assertion and employing a special kind of negation when needed. Whereas the ordinary negation of a claim S is another assertion. But what if we are called upon to take sides? We have one of two resources. If asked whether the ship that docked in a given port is the same as the one that left the original port. It is not a trivial task to repackage a theory developed using exclusion negation into one that appeals to a special kind of denial. the assertion of the exclusion negation of a claim merely denies that claim. there's just no telling". then. If we are philosophers.20 Indeterminacy Presumably. this is followed up." Denying that it is the same ship is not the same as asserting the negation of the claim that it is the same ship. or even if there is a difference between them. But some such account is needed. Exclusion negation is 9 It is hard to say which of these is the best 3ccount. or some variant of these. or it is itself a special kind of speech act-an act which denies the content "that it is the same ship". . but in ordinary conversations such a comment will usually suffice. if we think a claim is indeterminate we avoid making either that claim or its negation (which is also indeterminate). or "There's no answer to that question". rather than see it as the ordinary assertion of a special kind of negation. because there must be (and is) an alternative to asserting the negation of a claim that you think is indeterminate. Since this book is not about the paradoxes. So a denial is either an assertion of a special kind of negation.

This is to ignore certain instances of what Frege calls indirect contexts. If we want to say in the object language that a claim lacks truth-value. but with a special stress that settles on the 'not' itself or some parl of the sentence that follows. the phrase 'it is not the case that' is a complex phrase that contains a connective. which is referred to by the that-clause.) As a consequence. the exclusion negation of'S' is just '-. S'. with the exception that the long phrase goes uniformly in front and the 'not' must usually go someplace inside. 'o to Here and throughout this book I am making a simplification that is strictly wrong. we just exnegate both it and its negation. Something like that is at work in the suggested readings above for the determinate-truth connective. This is obvious in 'She means that snow is white'. and the complex phrase 'it is not the case that' has the same overall effect as 'not' all by itself. these two phenomena cancel each other out. This kind of notation corresponds to things we actually say when forced to confront claims in real life that we think are indeterminate. As a result. Exclusion negation '~!' is also just pronounced 'not'.S'. (In Fregean terms. this is our proposal above for introducing a connective for indeterminacy. but convenient. because '!' is not itself to be pronounced at all. 1997) that whenever we have a that-clause or a whether-clause we have a case of indirect discourse. as the occasion demands. Since stress is rarely unambiguous in what it contributes to meaning. or 'determinately. the sentences' It is true that snow is white' or That snow is white is true' attribute truth to a thought. S negation not S exclusion negation ex-not S F T T T F F T easily defined in terms we have available from above. Nonetheless we are inclined to call it a connective because it resembles the behaviour of one in spite of its complexity: the 'that' makes the phrase that follows refer to its meaning. there is no such connective as 'it is not the case that'. saying '-. Indeed. I will persist below in the inaccurate but clearer policy of reading '!S' as 'it is (determinately) true that S' and '~!S' as 'it is not (determinately) true that S'. or 'She believes that snow is white'.!S & -.Indeterminacy TABLE 21 2. it is more than a connective since it also creates an indirect context.!S'. . I think it is equally true for 'It is true that snow is white' and 'It is not the case that snow is white'. It is not strictly correct to read '!S' as 'it is true that S'. and the 'it is the case' correctly characterizes that meaning if and only if the sentence following the 'that' is true when normally interpreted. but that is not itself a connective. I argue in Parsons (1996. 'not'. a context in which we have shifted to talk of the meaning of the contents of the clause.4.!-. its effect in speech shows up only as a special intonation or stress on the sentence that follows it. with the understanding that the notation is intended to be my way of symbolizing some uses of 'not S'.

. for they have developed quick instincts for what is the case..22 Indeterminacy 2. In any event.6 LOGIC When we wish to be precise in using a language that permits truth-valueless sentences. I omit these because they have no bearing on any of the issues to be discussed... it is worth taking some time to go over the consequences for logic of admitting the possibility of lack of truth-value.. a .' is a sentence containing a name 'a'. . For example.. one might require of a valid argument that it preserve truth and also preserve non-falsehood.. and these instincts must sometimes be overcome. so that no instance of a valid argument can lead from premisses that lack . and some people do.. but because of lack of familiarity it may not be obvious without some thought. There is nothing difficult or mysterious about this.. One could mean different things than this by validity. .. and even some analysis.. So one may "quantify into" any context.. Things are simple in one way: the language described above (and used throughout this book) contains no non-extensional contexts. the logical truths that result and the inferences that are valid differ subtly from the ones we are used to when we idealize and assume bivalence. what is permitted and what is not. if ' .. This is especially so for people who have been trained in logic.. a . a . b . x . this inference will always be valid: a=b . . . That is. the following is always valid: ... . 3x(. That is: An argument is formally valid if and only if any instance of it that has all true premisses also has a true conclusion.) And when we introduce identity (in the next chapter). Validity: I assume that when we discuss validity we are discussing truth-preservation.

Almost any currently popular version of natural deduction (modified as indicated below) will do. as does Cowles (1994) following him.) This leaves open the possibility that in some of the philosophical discussion addressed below there might be misunderstanding due to a conflict of assumptions about what is meant by validity. I use the symbol '=>' for this connective. I include it here in order to state a set of rules that will not need to be supplemented later. I also include the rules for identity. My views are much closer to the views expressed by van Inwagen and Cowles than one might think. . these include all of the rules of proof that will be appealed to in this book. In order to state all the rules here. if you are arguing from true premisses.Indeterminacy 23 truth-value to a conclusion that is false. The following are (partly redundantY2 rules for a natural deduction system in which premisses are introduced at will. though they will not be discussed until the next chapter. these rules look quite !! Van Inwagen (1988) does this. The semantics described in §2. These valid inferences may also be studied by using formal proofs that are based on the semantics. In the system given here. RAA can be inferred from CP and Rule T. when a rule indicates that something may be inferred from certain given things. the differences will not matter. (Of course. it is false if '<\>' is true and ''I'' is false. but the reader should be on guard. J need to add a conditional connective. Except for abstraction principles. Except for the quantifier/determinacy commutation rules and the non-falsity rules for quantifier inferences. This must be kept in mind when assessing any of the claims of those papers.g. This connective will be discussed and its choice justified in Chapter 5.4 determines the valid inferences. since apparently conHicting claims turn out not to conHict once the alternate meanings of 'valid' are unpacked. and subproofs can appear at any later point. e. those things may be anywhere in the proof so long as they dominate the line being derived. Its definition is that '<\> => '1" is true if '<\>' is false or ''I'' is true or both '<\>' and ''I'' lack truth-value. A line dominates a later line if the dominating line is not within a previously completed subproof. more rules are stated here than are logically required. it is otherwise lacking in truth-value. I think this does not actually happen. lI I do not require this additional complication because I judge that this is never needed to assess arguments that are actually given in the literature. 12 For ease of use. I give a set of rules of proof here.

..3x.\7'xFx' one may infer '. Since these truth-tables are new. I summarize here some points of comparison with ordinary bivalent logic. Quantifier Rules: We adopt the usual rules of Universal Specification.. and in the fact that tautological implication has different forms than in bivalent logic... including the non-standard '!' and 'v' connectives and the '=>' conditional.. then the subproof may be followed by '!<jJ ~ IV'.......Fa'.. Rule RAA: If there is a subproof whose sole hypothesis is '(f and whose last line is of the form '1jf&.... For example.. Likewise '!3x<jJ' and '3x!<jJ'... Quantifier Rules for Non-falsity: We also adopt "non-falsity" versions of the rules of Universal Specification.....<jJ' in any context.. Tautological implication is defined as follows: A set of sentences tautologically implies a given sentence if and only if every assignment of truth-value statuses to the nonmolecular parts of the sentences which makes all the sentences in the given set true also makes the given sentence true... Existential Generalization by prefixing their formulas with '. so are the resultant tautological inferences.. Leibniz's Law: From's == t' and '<jJs' infer '<jJt'. Rules for a natural reduction system Rule T: Infer anything on a line if it is tautologically implied by a set of sentences on lines that dominate that line... where '4>t' results from '<ps' by replacing some or all occurrences of's' in '<jJs' by't'.\7'x.Ijf'. QuantifierlDeterminacy Commutativity: '!\7'x<!J' is interchangeable with '\7'x!<P' everywhere.. Their principal novelty lies in the modified reductio and conditional proof rules. likewise for '\7'x<!J' and '..24 TABLE Indeterminacy 2. Rule CP: If there is a subproof whose sole hypothesis is '<jJ' and whose last line is 'IV'.!(p'..... Self-Identity: '\7'x(x == x)' may be written on any line. Quantifier Interchange: '3x<!J' may be interchanged with '..... normal. This definition relies on the truth-table accounts of the connectives..... Existential Generalization.<!J'. Existential Specification.!..'..!.. Universal Generalization..... .. then the subproof may be followed by '--.. the non-falsity version of Universal Specification is that from '. Universal Generalization...!. Existential Specification.5..

and so on. and you can infer '-. For example.S' is true by deriving a contradiction from'S') cannot take its usual form. 3. in bivalent logic you can infer 'A v -.B'. When the truth connective (the "determinately" connective) is included. so you are not allowed to infer that. work in the usual fashion.A as a hypothesis. the following hold: 1.g.A.(A & B)'. so does -.A v -.(A&-. This is because 'follows' means that if the premisses are true the conclusion must be true too. and a conclusion that classically follows from those premisses.B' from '--. you can easily infer a contradiction from it (itself).Indeterminacy 25 First.A). Then: 1. e. Classical tautologies are not tautologies here because of the possibility of lack of truthvalue. Reductio ad absurdum: You can show that something is not true by deriving a contradiction from it. The "Law of Excluded Middle" does not obtain.A is not a tautology because if A lacks truth-value. That is. But you can infer 'A' from 'A v B' and '-. and you cannot do that here. Most inferences from premisses.B'. if you have some premisses. consider sentences of the ordinary predicate calculus as described above tha t do not contain the connectives'!' or 'v' or '=>'. however.A). But if A lacks truthvalue. Classical indirect proof would then let you infer -. so does -.(A&-. And so the proof technique of indirect proof (Rule RAA: show that '-. But that does not show that its negation is true. If you can prove something in a way . This connective does not interfere with any of the inferences we just discussed. The commonest exceptions to this are cases of "irrelevant" inferences. and so does the disjunction. A v -. 2.A' from 'B v -. Illustration: If you assume A&-. then that inference is usually valid here. Classical tautologies (those made without the use of the special connectives) are not provable. Assuming that the premisses are true very often amounts to assuming that there are no gaps in them in positions that could make a classically valid conclusion lack truth-value.

S ~ ____ _ the following is valid: YES:~ I A &--. So you needn't worry about properly ordering it with respect to the quantifiers.!Fx) is logically true. and !(AvB) is equivalent in truth-value status to !Av!B.lA is not generally equivalent to !--.!S.A .B :.--. A subtler point is this: if you have a valid inference pattern. --.A v --. If you derive a contradiction from S. however. The truth connective commutes with quantifiers.S. commute with negation. it does show --. For example.!S Contrapositive reasoning: TI1e above principles are natural and easy to recall. contrapositive forms of that pattern need not be valid. 2. then although that doesn't show --. This connective does not. 4. 3.A. if this is a valid form: A :. The truth connective distributes across & and v: !(A&B) is equivalent in truth-value status to !A&lB. 5. So although this is not valid: NO: l~ --. the presence of the other connectives does not interfere with this. and Vx(Fx v --. For example. !3xFx will always have exactly the same truth-value status as 3x!Fx.A --.26 Indeterminacy I just described.B then you cannot assume that this form is valid: --.!A is a tautology. There are tautologies and logical truths involving '!'. There is a form of indirect proof. For example.

The point just made about contra positive reasoning depends on the use of a negation connective that produces a sentence without truth-value when appended to a sentence that lacks truth-value. of course. But the contrapositive principle is not remotely compelling: -. it is plain that this inference pattern is valid: A :.!A :.lA because if the premiss is true. a reliable form of "near-contrapositive" reasoning. There is.---.B then so is this: In my opinion the major cause of ongoing controversy regarding indeterminacy in general and indeterminacy of identity in paJ-ticular is our tendency to take for granted contraposilive reasoning when using propositions that may lack truth-value. I will touch on this point a number of times throughout the book. This failure of contrapositive reasoning does not rely on any esoteric doctrine. using the truth connective. where it is straightforwardly fallacious. so is the sentence that says it is true. then the premiss is true and the conclusion is not true. people usually specify how they intend their negations to be taken_ .Jndeterminacy 27 As a counter-instance.A Tt A lacks truth-value. it is an inevitable and natural result of admitting the possibility of a sentence's lacking truth-value. If this is valid: A :. This is the sort of negation that is usually intended when people discuss issues of indeterminacyY If one uses "exclusion" negation. " This is based on my reading of the literature. others might disagree_ Fortunately. This type of reasoning is so natural to us when dealing with truth-valued claims that we instinctively pursue it when dealing with meaningful claims that may lack truth-value. because it lacks truth-value.

The topic here is indeterminacy in the world. Vagueness is usually thought of as a property of language or of concepts. Consider Pelletier (1989). Those topics are not under consideration here. Recall that the exclusion negation of a sentence.28 Indeterminacy things are different. 1. and so contra positive reasoning is valid using it. (From 'ex-not ex-not S' one may not infer 'S'. I avoid the word 'vague' because it suggests a very different study of one or more very different notions. When discussing the views of others. the law of double exnegation does not hold. and writing this explicitly using '-.) I find it more revealing to think of exclusion negation as a way of denying the truth of a claim. and we need to keep the views straight. He rewords this as "vagueness. 14 14 The use of 'vague' makes a transition between language and the world overly natural. but I do not. using 'indeterminacy' to describe this. Semantic and/or set-theoretic paradoxes: Much work in the last few decades on systems of logic that allow for truth-value gaps have been motivated by the study of semantic paradoxes or.!' instead of a single connective 'exnot'. and then he concludes . though the reverse is valid. But exclusion negation has idiosyncrasies of its own. I need to be careful to match their own use of negation. but it is essential to use terminology that at least allows the distinction a chance to be made. by the paradoxes of set theory. Some people think that this kind of vagueness is to blame for identity puzzles. He quotes me in Parsons (1987) talking about indeterminacy of truth-value. or indeterminateness" (Pelletier 1989: 492). 2. for example. though they may be called to mind by some of the discussion so far. not vagueness of words or concepts.!S The use of exclusion negation removes all truth-value gaps. to a lesser extent. It may be difficult to distinguish these in practice.7 WHAT IS NOT AT ISSUE Certain topics are not discussed here. Vagueness: People sometimes use 'vagueness' to express what I call indeterminacy. produces a sentence that is true when'S' lacks truth-value: ex-not S =df -. and the reader needs to be alert that no questions are begged because of unclarity over what kind of negation is used.

not to degrees of truth-values. 3. it refers to software systems that are designed to allow for "risky" reasoning. but it would be misleading to use a term that suggests the theory is something it is not. . but the subject has changed. however. One kind of vagueness is indeed a (semantic) notion essentially involving language. It sometimes concerns reasoning to conclusions that have definite truth or falsity. as opposed to y. and investigated by means of Jifferent evaluation techniques. not because we are presently ignorant of what it is. Fuzzy logic: Fuzzy logic is a study that has been much discussed over the last couple of decades because it has been taken as a slogan for a certain kind of computer programming. My enterprise is not epistemological at all." (ibid. Higher-order indeterminacy is discussed briefly in Chapter 12. and I do not attempt the additional step of blaming it on either the objects or the properties or relations making up those states of affairs. As explained early in this chapter. When people use 'vague' to characterize things in the world. it may be vague whether it is vague whether this is orange. at least in the version of the theory 1 discuss in most of the book. Indeterminacy is not like this. I see indeterminacy as applying at the level of states of affairs.). whereas I develop a theory in which indeterminacy is not a matter of degree. they often speak of vague objects or of whether the identity relation itself is vague. and the scale of "truth-values" in fuzzy logic is often taken as representing degrees of knowledge. in language it leads to a lack of truth-value. "vagueness ought to be viewed as a semantic notion. not only is it vague whether this is orange. but their truth or falsity is yet unknown. I agree totally with his conclusion. Perhaps this is an inadequate way to view things.Indeterminacy 29 2. When 1 say that a sentence lacks truth-value this is because the world gives it none. Studies of vagueness often include cases in which the vagueness is due to a borderline case. This. Its study leaves untouched the indeterminacy I am investigating. One can define a "vague object" as an object x such that for some object y it is indeterminate whether x = y. 4. It is common to assume that along with vagueness comes higher-order vagueness. where it is a matter of degree whether a predicate applies to an object. or to the identity relation. inferences to conclusions that are not deductively validated by the given data. does not blame vagueness on x itself.

. though I am not sure what difference this would make. I have only two truth-values: true and false. and no modal notions are invoked in the analysis. I discuss only language with extensional contexts. Non-extensional contexts and modality: These issues are not directly relevant to anything discussed in this book. and neither. it is a semantical system in which there are additional truth values between 0 and 1. false. and three possible truth-value statuses: true. usually an infinite number of them. representing degrees of truth. Again. "fuzzy logic" has a technical definition.30 Indeterminacy As originally developed. I do not think of "neither" as a degree of truth.

x is determinately not identical with y if and only if: there is some property that x determinately possesses that y determinately does not possess. It is indeterminate whether x is identical with y if and only if x and y are neither determinately identical nor determinately not identical. or vice versa. and vice versa. then it may be indeterminate whether a and b coincide in the properties that they share. or at least one property that x determinately does not possess such that it is indeterminate whether y possesses it. Consequently. This will happen where there is no property that x determinately possesses that y determinately does not possess. Specifically: x is determinately identical with y if and only if: every property that x determinately possesses y also determinately possesses (and vice versa) and every property that x detenninately does not possess y also determinately does not possess (and vice versa). or vice versa. More succinctly: .3 Identity 3. it will be indeterminate whether they are identical. but there is at least one property that x determinately possesses such that it is indeterminate whether y possesses it. If there is indeterminacy in the world.1 WHAT IS IDENTITY? When are objects a and b identical? T accept the Leibnizian doctrine that defines identity between objects in terms of coincidence of their properties. or some property that x determinately does not possess that y determinately possesses.

and determinately stand in. that lhe properties in question are things like "being yellow at tOO as opposed to "being yellow". assuming e. x is determinately not identical with y if and only if there is some property regarding whose possession x and y determinately disagree. I think that nothing of substance turns on the choice.g. etc. not identical. 1l1is account is meant to be a "real definition".) Indeterminacy of identity between a and b thus coincides exactly with indeterminacy of distinctness. where the latter is a property that one and the same thing possesses at one time and not at another. I assert the equivalences above as truths about identity. composition. I use the former notion because the account is less complicated. That is. 2 An old illustration is of a universe consisting of nothing but two spheres of the same size. That is.) . Suppose that x and y coincide completely in terms of their properties. (That is a nominal definition. Otherwise it is indeterminate whether x is identical with y. identity would have to be defined in terms of "having all the same properties at all the same times". the account is not meant to explain what I take 'identical' to mean. and determinately do not stand in.32 Identity x is determinately identical with y if and only if x and y deter- minately possess and determinately lack exactly the same properties. I mean by 'identical' exactly what others mean by it. If I were to appeal to the latter notion of property. The account of identity given above in terms of properties says nothing about relations. not a "nominal definition". the theory under examination is a theory that embodies these equivalences as substantive claims about identity. the spheres are supposed to be two objects that differ only with respect to their relations.. instead of just "having all the same properties". but differ determinately in terms of their relations to objects?2 Then they should be distinct. So a better account than the one above would read: x is determinately identical with y if and only if x and y determinately possess and determinately lack exactly the same properties. this is the only way I know to guarantee that we are discussing the same issue. x is determinately not identical with y if and only if there is some property regarding whose possession x and y determiI Here and throughout the text [ ignore the complication of lime. the same relations to the same objects. (One is distant from a certain object-the other sphere-and the other is nol distant from that object. I I use 'distinct' for 'not identical'. existing some distance apart.

Officially I adopt this more refined account as the official view. It should be clear that I reject as basic any account of identity in terms of coincidence of the predicates in some language that combine truly with the names of the objects in question. but so far as J can see it is irrelevant to any of the issues about indeterminate identity to be discussed. Second. 111is can (and will) happen sometimes when the content of the predicates is partly . and there is a simple way to merge the options. But it is clumsy to state. but nothing controversial will turn on that. and determinately fails to hold of exactly the objects that either P or Q determinately fail to hold of. It is a controversial assumption. and to one another. I have not even assumed that if P and Q are properties. and they are not crucial to the ontological issues discussed here. then their conjunction is a property. we should retreat to the account of identity that explicitly takes relations into account. as well as others. Then the accounts coincide. I will indeed make this assumption. Such an account can be wrong in two ways. These questions about how properties are related to relations. the assumptions made so far leave this open. but they are technical and they are distracting. Suppose that for any relation R and any object 0 there is a property P which is determinately possessed (or determinately dispossessed) by exactly those objects that determinately bear R (or that determinately do not bear R) to o.) Later I discuss applications in which I assume a principle of plenitude for properties that will yield such conjunctive properties. we may not have the words to express a difference in properties between x and y when one exists. are important.Identity 33 nately disagree or some relation to some object regarding which they disagree. and if they do not then it might be possible for such a predicate to combine truly with a name of x and falsely with a name of y even when it is not determinate whether x is y. but it need not be. For example. (I have not assumed that there is some property U that determinately holds of exactly the objects that both P and Q determinately hold of. there is no automatic guarantee that predicates stand for properties. If it becomes relevant. Otherwise it is indeterminate whether x is identical with y. if the language is too impoverished. This is primarily because so little has been assumed. First. Is identity itself a relation? It can he.

but not all. and if satisfying a predicate does not necessarily count as a property. For now. and I will continue with that assumption in effect. 1 know of no conclusive way to argue this point. then one is speaking merely of 'a kind of relative identity: a relation which ensures indiscernibility of its terms in some. If there is indeterminacy in the world. It is also consistent to assume otherwise. but for any objects a and b either their own indeterminacies completely coincide or else they determinately disagree about possession of some property. 11. A positive answer to this question is consistent and in teresting. 3 4 . and only secondarily a matter of how language functions. For example. Perhaps there is a great deal of indeterminacy. respects' (Noon an 1984: 118). since otherwise the theory has no interesting application. Then there would be indeterminacy without indeterminacy of identity of objects. I take it as obvious that identity between objects in the world is basically a matter of what the world is like. Harold Noonan urges that identity be defined in terms of the behaviour of predicates and names. I leave open for the present whether there might be indeterminacy of identity of sets of objects even when there is no indeterminacy of identity for the objects themselves. 3. I think that Noonan is mistaken in this claim about relative identity. so I merely state it? This leaves me at odds with certain philosophers who take the contrary view. it is discussed in Ch. this will be discussed below. and he suggests that if identity is defined in terms of properties. 4 We cannot settle a priori whether this happens. Cases of this sort will be discussed extensively below.34 Identity linguistic or semantic. I will presume that See Keefe (1995: 185 ff) for considerations in favour of this position. then it is still left open whether it extends to identity. I state the conflict so as to contrast my own position. It is consistent to append to the account the assumption that every property is actually determinately possessed or determinately not possessed by every object.2 INDETERMINATE IDENTITY The ontological picture introduced in the last chapter is neutral about whether any indeterminacy obtains at all.

Given this explanation together with the semantical and metaphysical theses laid out above the following two principles must automatically hold for the identity sign: Reflexivity: Any sentence of the form 'a = a' is logically true..y> such that x is determinately identical to y. for this is the interesting case. When indeterminacy is possible. where ' . The predicate '=' is false of exactly those pairs of objects <x. t . . 6 Again. Leibl1iz:~ Law: The following holds.. you may infer the same sentence with the one term replaced by the other in as many of its occurrences as you like. given any identity between terms. .. I argue that it is possible for a name to have a unique denotation even when it names an object which is indeterminately identical with some object that the name does not denote. If there were denotationless names in the language.. That is. . but it is a contingent matter.. 3. this is always elliptical for the proposition that it is indeterminate whether the objects in question are identical... In Ch.y> such that x is determinately not identical to y. this is because only names with unique denotations are permitted in the language.. . sentences of the form 'a ~ a' would not be logically true. and any sentence whatever containing one of those terms. the sentence ''v'x(x = x)' is logically true. we must consider the "anti-extension" of a predicate as well as its extension: identity is no exception. t . 9.. It is sometimes convenient to speak of an identity between two objects as being indeterminate. . s .' represents any sentence containing the term 't' in one or more places: s =t .3 LAWS OF IDENTITY The semantics for the identity sign are: 5 The predicate '=' is true of exactly those pairs of objects <x. 6 Likewise.Identity 35 identity itself is sometimes indeterminate. -' The first of these two clauses is emphasized in Stalnaker (1988: 350).

. . By similar reasoning. So 'F' is true of the object named by 't'. that is. it must be that 'F' is true of the object named by's'.36 Identity We can show that Leibniz's Law holds by an inductive proof. 'Fs' lacks truth-value.Fs' (just replace 'true' by 'false' and add a clause for negation). By similar reasoning. since ''IFs' is true. and this might lead one to suspect that some question is being begged here. one uses universal instantiation in the metalanguage. that using the principle in the metalanguage does indeed allow you validate it for the object language. this would cast serious doubt on whether our sign of identity were actually expressing identity. Recall that validity means preservation of truth. if Leibniz's Law were not to hold for such an extensional language. s . the inference holds when ' . Leibniz's Law holds. Assuming that 'Fs' is true. as opposed to some weaker relation.. If the identity premiss is true. Here. then no system can ever be shown to validate any law of logic.. then the objects named by '5' and by 'I' are indistinguishable with respect to any of their properties and relations. Further..' is atomic. the object named by's' is identical to that named by 't'.. the most that one can show is that certain principles cohere. then I must be using the same logic in my metalanguage as in the object language. For example. The law can be shown for more complex formulas by induction on the complexity of the formula. (Leibniz's Law is expressed here as a principle of inference. So 'Ft' lacks truth-value. This proof itself uses Leibniz's Law in the metalanguage. So 'Ft' is true. Since I intend the metaphysical theory under discussion to possibly be true of the actual world. . so 'P is neither true nor false of the object named by's'. Indeed. Perhaps in some sense it is. s .1l1is is because we . as opposed to a schema for conditional statements. for simplicity. and ''1Ft' is true... If one insists on more than this. suppose it consists of a one-place predicate 'P followed by 's'. Assuming that's =: t' is true. Suppose first that the sentence ' . ' is ''IFs'. when one shows that a certain classical first-order system of logic obeys the principle of universal instantiation.. s . given that the language contains no non-extensional contexts. but this is no different in principle than any other law of logic. 2. In the face of extreme logical scepticism. .' is '-. We have three base cases: l. the inference holds when ' . So it is neither true nor false of the object named by 't'..

x . however. that does not guarantee that contrapositive versions of it are valid. We will see examples of this in the next chapter.. But then the objects denoted by's' and '( disagree with respect to that property....) Leibniz's Law together with reflexivity lets us derive some of the other well-known principles of identity.s = t NO! Recall the point made in the last chapter: if a pattern of inference is valid. it would be possible to disprove the truth of any assertion of indeterminate identity. t . . . Conditionals bring with them their own idiosyncrasies and controversies. s .(. l1ms the theory under discussion must deny the validity of the contrapositive version of Leibniz's Law. .. . The following principle remains valid in any case: . ) :. x . and it is only in this sort of case that the principle fails. r = s Transitivity: s=t t=u :. if contrapositive versions of Leibniz's Law were valid... Conditional formulations of Leibniz's Law are discussed in Chapter 6. We do not have: .. such as: Symmetry: s =t :. and thus they are by definition determinately not identical.--. Leibniz's Law is a crucial example of this. then it appears that sand t must disagree with respect to the property expressed by the context ' .' may not express a property.Identity 37 have not as yet discussed conditionals.. s =u It is crucial. But how can there be a counter-example to this principle? If the premisses are both true. As we will see in the next chapter. which is all that the conclusion states! The answer to this is that the context ' ... and it is better not to get those issues entangled with issues about Leibniz's Law.. that contrapositive versions of Leibniz's Law do not hold.'. ----.

By denying the validity of this principle I may be charged with begging the question in favour of indeterminate identity.-. (One might worry whether using the converse of Leibniz's Law in the metalanguage would force the converse of Leibniz's Law to hold for the object language. s . Yes. I assume that my metalanguage works just . L~t the first premiss say that a is determinately identical with a. I countercharge that to assume the validity of the principle is to beg the question on the other side. Identity -. they will be cases of indeterminate identity. t oO .. The theory of indeterminate identity must hold the contrapositive of Leibniz's Law to be invalid. But this should be no embarrassment. such as the one just displayed.!a = b Obviously we do not want to conclude from the fact that a is not determinately identical with b that it is (determinately) not identical with b: la=a -.!S=l So the loophole cases are ones in which sand t are neither determinately identical nor determinately not identical. at least for some instances.la = b :. With this in mind.a = b NO! At least I do not want to conclude this.... that is true. it is easy to give a specific counter-example to a contrapositive version of Leibniz's Law.(oO .-. In later chapters I will dispute various arguments to the effect that the contrapositive of Leibniz's Law must be valid. and consider what independent considerations there might be to resolve the issue. So it is only by adopting the desired logic in the metalanguage that I can say that the semantics I have described works as I say. Again. and the second premiss say that a is not determinately identical with b: !a =a -. though it is clear that others do.38 . So we must let each side take sides.) :.

That is not difficult. If Noonan means by 'relative identity' what Geach means. 'relative' because of the relativity of choice of 'F. if it is true that a = b. his criticism does not apply to the theory given here. it is Leibniz's Law that is given as a . Noonan may not have meant 'relative identity' in Geach's sense.4 RELATIVE IDENTITY? We may now return to Noonan's charge that defining identity in terms of properties and relations instead of in terms of linguistic interchangeability will yield only a kind of relative identity. who focused on the fact that we often wish to say things like 'the first letter in "aardvark" is the same type as the second one. So a good test for relative identity is whether it does or does not sanction Leibniz's Law. The identity I am discussing does sanction this law. The remaining worry is whether I can do all this consistently. Again. 'x is the same F as y' would entail 'x is the same G as y'. and 'identity' because of the use of the word 'same'. Consistency can be established by an assignment of artificial entities to the language. He may merely have meant that identity defined in terms of properties and relations is just a kind of limited indiscernibility that does not guarantee general indiscernibility. but not the same token. and thus it cannot be a species of relative identity. That can easily be done by appeal to the 'pictures' described in §7. and this is supposed not to follow.' Geach took the locution 'x is the same F as y' and called it 'relative identity'. if it did.6. then there should not be any statement made using one of these names (outside of non-extensional and quotational contexts) that changes truth-value status when that name is replaced by the other. It is clear that relative identity is not the same as the non-relative identity that I have been discussing. The most crucial difference between them is that Geach's relative identity does not validate Leibniz's Law. described in a classical set-theoretic setting.) 3.Identity 39 as my object language does. 'Relative identity' is a phrase made popular by Peter Geach (in Geach 1967). This is the crux of the matter.

. for the conditional formulation is often a stronger principle. it is not so in the theory under discussion here. He adopts a kind of relative identity as a "substitute" for Leibniz's Law. the various conditional formulations all coincide with the principle of inference. a .c . b . What Noonan has in mind may be a converse principle: that if there is a discernibility in language. On this test our theory yields real identity. where '''G>' means 'it is determinately true or determinately false that'.. Also. . His theory of "vague objects" is thus quite different from the one J am developing.a =b In Noonan (1990) he cites this principle... He suggests that if identity is defined in terms of properties.. not in respects only expressible by predicates containing 'v' or synonyms of such predicates. Zemach (1991) rejects it. He relativizes identity to ideologies.---. . even in contexts containing 'v'.. ) !---. The following is a valid pattern no matter how the blanks are filled: a=b '1( . b ... and if satisfying a predicate does not necessarily count as a property.. and in Noonan 7 I do not mean to suggest that Leibniz's Law is itself beyond question. But when the premisses (antecedents) are bivalent in virtue of form. 8 I have simplified Noonan's formulation by representing his 'G> & "G>' as '!G>'. whereas I use a principle of inference..40 Identity test for whether we are talking about real identity. which he calls "the Principle of the Diversity of the Definitely Dissimilar". then this guarantees distinctness.) :. .'1(. and holds that objects may be identical with respect to one ideology but not with respect to others.. That is. as is true of any formula that begins with '!'. This is sometimes a crucial difference.? Here is a fuller quote from Noonan. respects-in particular.)' Our identity does guarantee complete indiscernibility. At least. These are equivalent in truth-value status. For example. but not all.. But this is not so. ) :. a . Noonan formulates the principle as a conditional. then one is speaking merely of a kind of relative identity: a relation which ensures indiscernibility of its terms in some. Noonan (1984: 118) The implication here seems to be that an identity will not sanction indiscernibility of its terms within contexts governed by the indeterminacy connective 'v'.. that this is valid: 8 !(.

.. a ..Identity 41 (1991) he calls this a "weaker" principle than Leibniz's Law. 3. b ... 2. ... !( . (This is because putting a determinate-truth connective'!' on the front of a whole premiss is always redundant.. . but there is at least one property (typically. This says little in detail about which properties I and my body share.. if it is indeterminate whether a = b. then there is no property with respect to which a and b disagree. suppose that my body takes up space. 4. ) !-. It is the point at issue. b . a ... : ..(. If we think that we have a case of indeterminate identity. For example. b .. Me and my body: The theory allows one to say that it is indeterminate whether I am identical with my body. 3. 2.)9 Perhaps the contrapositive of Leibniz's Law is valid. then we should be able to point to some such property. But it is not obvious. many) that one of them determinately possesses (or determinately does not possess) such that it is indeterminate whether the other has it.. perhaps 110t. ( . . 5. ) : ... Then so does the contrapositive version of Leibniz's Law: 1. ... a .... ) ( . 4. I review here the four identity puzzles cited above with this in mind. suppose that Noonan's principle holds. suppose first that the contrapositive version of Leibniz's Law holds... then so does Noonan's principle: 1....( .) !-... The principle in question is just another equivalent to the contrapositive of Leibniz's Law.(.. and suppose that taking up space is a property. 3... {/ = b Premiss of contrapositive of Leibniz's Law Premiss of contrapositive of Leibniz's Law From 1 From 2 By Noonan's principle .. But it is not weaker in the logical sense (the logical sense being that Leibniz's Law entails it but not vice versa). for it is not entailed by Leibniz's Law. . ) -{ . ) 5. (/ .. TIlen I myself 9 To make this dear in the case in question. ..5 APPLICATIONS According to the account of identity just described.) !( .. ) .. b .... a=b Noonan's premiss Noonan's premiss From 1 From 2 By the contrapositive of Leibniz's Law Second.

Both options are open. then if I have this property and my body does not. the point is clear. or God.1° then there is at least one thing in the room and at most two. Somewhere there must be some indeterminacy of property in order for me and my body to not be detenninately identical.42 Identity must either have the property of taking up space. But the case abstractly described does not give details. and if thinking is a property. then the theory does not yet take sides. and shed no light on the basic question of whether identity might be indeterminate (in this world). Perhaps we share all non-modal properties. But perhaps I could live on without my body (although we both cease to exist at death. then either my body thinks. sees fit to resurrect my body for my comfort). and thinking. not the theory. or it is indeterminate whether it thinks. The issues discussed here are not likely to turn on considerations of modalityY The ship: Suppose that being located at a certain point in space at a given time is having a property. If I and my body and nothing else are in a certain room. or else it must be indeterminate whether I have this property. though it is indeterminate whether there is exactly one thing. my body can't be in a room unless its parts are there too.) Perhaps I and my body differ only in modal properties. This is why I do not discuss them. et. contingently. and indeterminate whether there are two. The theory does not tell us what properties there are. I am inclined to view modal properties as properties. however. What does the theory say then? Well. but that is me. If modal predicates do not stand for properties. Let 0 be the place from which the original ship set sail at time to. . (Such cardinality judgements are discussed in Chapter 8. and the puzzle has a traditional solution. and let d 1 and d 2 be the places that the ship with new parts and the newly ass em- In Of course. Conversely. In my opinion. if such modal properties are properties. so the example is fanciful. the additional complexities (and complexities abound!) are all contributed by artefacts of modal semantics. if it is true that I think. it only says what the consequences are for identity of possessing properties. we are distinct. such as taking up space. I think. II 111e options for how to do modal semantics are so numerous that just about any view about indeterminate identity may be reconciled with just about any view about whether modal predicates do or do not stand for properties. Keefe (1995) and Stalnaker (1988: 355 ff) for some discussion.

and the newly assembled ship determinately lacks that property (since it is determinately elsewhere then). For each p-cat. Before the repair job began there was exactly one ship. and the puzzle would have a traditional solution. Exactly one person entered the room. it is indeterminate whether it is identical with the cat. That is. with either exact answer being indeterminate. there are at least two and at most three.Identity 43 bled ship respectively dock at time t n • Then the ship with new parts determinately has the property of being at d 1 at t n . but the newly assembled ship is not indeterminately identical with the ship with new parts. The cat and the p-cats: There is exactly one cat on the table. as a result. though probably millions of p-cats. I have also reviewed what I take to be some of the data against which any theory should he tested. All told. for each p-cat it is indeterminate whether it is a cat (if it were determinately not a cat it would be determinately distinct from the cat. at to> since otherwise it would be determinately distinct from one of the resulting ships. This example shows that indeterminate identity is not transitive: the newly assembled ship is indeterminately identical with the original ship. The disrupted persan(s): It is indeterminate whether the person who originally entered the room is the person who later left the room. if it were a cat. but false that there are fewer than one and false that there are more than two. then by parity so would be the others. the ship with new parts and the newly assembled ship are determinately distinct. afterwards there are exactly two. which is indeterminately identical with the ship with new parts. Since there is only one cat. and exactly one person left the room. if . It is indeterminate whether there is exactly one person overall or exactly two persons overall. and there would be millions of cats on the table). In describing these cases I have not only said how J think the theory is supposed to work. It is indeterminate whether the original ship is at d . it is indeterminate whether the person who entered the room left the room. Assuming that leaving the room is a property or entails the possession of certain properties.

44 Identity you grant that each identity puzzle has no answer. I say this here partly to urge agreement. but also to let the reader know something about how subsequent discussion will go: the theory under consideration has the virtue of agreeing with these points. . and competing theories are compelled to deny some of them. then the rest of the things that I say about the puzzling situations are what I think any thoughtful person uncorrupted by a philosophical theory would say.

Stalnaker (1988: 357) 4. "Can there be Vague Objects?" In spite of its title (vague objects). it is not clear that Evans intended to address the question that I and other commentators call indeterminacy of identity. and DDiff One should be sllspiciolls of any argument that purports to get substantive metaphysical conclusions out of the logic of identity. Specifically. That is. his argumentation is irrelevant to most of the subsequent literature that it inspired. no statement of the form ''Va = b' can be true. as opposed to meaning that'S' lacks truth-value altogether. as opposed to a lack of truth-value. so I will ignore the fact that Evans might have meant something different. I set aside two potentially distracting issues: First. There is some reason to suspect that Evans saw indeterminacy as a subclass of truth. the aim of the paper was to prove that no statements of indeterminate identity can be true. It is thus important to consider what Evans's argument is. But it is the discussion that the paper inspired that I am interested in. but indeterminately so". Gareth Evans published a one-page article. I speak of "the Evans argument" as an argument aimed at refuting the possibility of indeterminacy of identity as I (and most others) understand indeterminacy of identity. (I say more about the alternative interpretation of Evans in the Appendix. If so. Properties.1 THE EVANS ARGUMENT In 1978. he may have intended ''VS' to mean" 'S' is true. This paper has become a focal point for a growing body of literature.4 The Evans Argument. and what it shows. including the present book.) .

'. Gibbons (1982). (4) by Leibniz's Law it needs to be completed. 4 It is Evans's complex remarks about how to complete the proof that provide the best evidence that he meant something different by indeterminacy than most commentators (see the Appendix to this book). Evans could not possibly give a conclusive argument to refute this possibility.. so one can assume that he means nothing unorthodox. This is a valid principle. I have also replaced 'x = a' by 'a = x' in lines (2) and (4). but it complicates the proof. as a number of commentators have pointed out. So interpreted. Evans discusses this point remarks about it. and using the slightly more common notation 'AX[ . cf. which I ignore I 2 Including Burgess (1989).(a = b) The hypothesis to be refuted Abstraction from (1) ["Abstract introduction"] Truism 3 Reverse abstraction from (3) ["Abstract elimination"] (2).. We may need a similar assumption about 'b' to justify step (2). and DDiff Second. It goes as foIlows: 2 (1) v(a = b) (2) h[v(a=x)]b (3) -.46 Evans Argument. Without this change the argument is missing two steps that appeal to the symmetry of identity within indeterminacy contexts. The argument: The argument is a reductio ad absurdum of the view that a statement of the form ''la = b' is true. . the argument is a direct attack on the central thesis of this book. which does not need to rely on it. ]' for property abstraction than his 'xl . 3 Evans does not call (3) a truism. he merely cites it without justification. reductio argument should end in a not. this bring the argument into conformity with the rationale that Evans gives for the steps. 1989).]'. nobody has come up with an interpretation of these remarks which makes them plausible without changing the subject matter away from the kind of indeterminacy we are discussing. . and this one does and makes some complicated here. . A contradiction. His discussion has inspired a number of speculations about what he might have had in mind. 4 For in the theory under Before critiquing the argument. it is obvious that a statement of the form 'a = b' can lack truth-value if the singular terms making it up do not have clear and unambiguous reference. I agree with the majorityl in interpreting his argument as trying to show that a statement of the form 'v(a = b)' cannot be true if the singular terms making it up have clear and unambiguous reference.tu:[v(a = x)]a (5) -. In order for (3) to be justified we need to assume that the term 'a' is not referentially flawed. Evans (1978: 208)-which is the only page of the article. Properties. Pelletier (1984. Garrett (1988). He does not explain either piece of notation.v(a = a) (4) -. I use Evans's notation except for replacing his negation sign '-' with mine '-. and Lewis (1988).

and DDiff 47 discussion in this book. (3) by Leibniz's Law This shortened version of the argument gives the strategy in a succinct form: Step (1) says something about b: that b is indeterminately identical to a. So a and b cannot be identical. thus making it irrelevant to the theory under discussion.Evans Argument. if (5) is true so is (6): (6) !--. we also noted in the last chapter that the contrapositive of Leibniz's Law is valid if the formula in question stands for a property.(a = b). and consider the following three-line variant: (1) v(a=b) (3) --. Since validity is just truth preservation. they introduce the required properties: .(a = b) The hypothesis to be refuted Truism (1). This fleshing out of the argument is not possible if Evans's negation is meant be exclusion negation.v(a=a) (5) -. the argument is easy to complete. and (6) clearly contradicts the second conjunct of (I').(a = b). The argument thus begs the question. So if the proof is correct as given. but rather its contra positive. . However. The argument does not use Leibniz's Law at all.!--. 10 br. And this is the point of the additional abstraction steps in the full argument.!(a = So there is an interpretation of the argument on which it fails to arrive at a full reductio. s TIle complexity of the argument is due partly to the presence of the property abstracts. line (1) means: (I') --. As we have construed indeterminacy. The response to this shortened form is clear. in which case line (5) would mean what we would express as '-. Step (3) denies this about a. Properties. This is the principle discussed and rejected in the last chapter. So remove the lines in which they occur. Their importance can be best understood if we initially see what the argument would be like without them. it is easily fteshed out into a full reductio.!(a = b) & --. Garrett (1988) adopts this interpretation.

On this interpretation. And here it is apparent what must be said by any defender of indeterminate identity: there is no property of "being indeterminately identical to a".x[\l(a = x)]'. because there is no reason to think that the sentence in (1) can be recast in terms of attribution of a property to an object. and indeterminately holds of the objects such that it is indeterminate whether they satisfy the formula).x[\l(a = x)]b (3) -. and make this explicit in (2). Step (5) is the contrapositive of Leibniz's Law. that abstract actually stands for a property which holds of the objects that satisfy the formula inside the abstract (and determinately fails to hold of the objects that dissatisfy the formula. The inference from (1) to (2) is fallacious. and make that explicit in (4). see note 2 of this chapter. and so the property abstract 'A. he says:6 (1) reports a fact about b which we may express by ascribing to it the property 'A.x[ \l( a = x)]' does not stand for a property.(a = b) The hypothesis to be refuted Abstraction from (1) ["Abstract introduction"] Truism Reverse abstraction from (3) ["Abstract elimination"] (2).\l(a=a) (4) -'Ax[\l(a=x)]a (5) -. Actually. And the proof itself gives a reason to think that (I) cannot be so recast. and only when.48 Evans Argument. . (4) by Leibniz's Law Evans says as much in justifying line (2). but applied in a special case in which it is valid. one supposes that the use of a property abstract is legitimate when. So the strategy is to see (1) as attributing a property to b. there are two distinct ways to criticize the proof. and DDiff (1) \l(a=b) (2) A. Way 1: In the first way of interpreting abstracts. Properties. the fact that step (1) is true does not automatically tell you that there is a property which reflects the semantic behaviour of the 6 Again I have replaced 'x = a' by '0 =x'. This is the interpretation employed in the previous paragraph. for there are no other flaws in the proof. So criticism of the full proof must focus on the transitions from (1) to (2) and (3) to (4). and then to see (3) as denying that very same property of a. and the proof attempts the impossible: to prove a priori the inconsistency of indeterminate identity. depending on how one interprets the use of property abstracts.

l ignore an extended discussion of this non-standard interpretation because it does not improve the proof. the inference from (1) to (2) is above reproach (as is the inference from (3) to (4»). ) . one does not assume that an abstract needs to stand for a property. Way 2: In the second way of interpreting the abstracts. step (4) becomes invalid. a definite description always denotes an object.3). and because the interpretation itself is somewhat artificial. But on this interpretation. and so step (5) is fallacious. and indeterminately holds of exactly the objects that indeterminately satisfy the formula or that do not determinately satisfy the formula but that are indeterminately identical to objects which determinately satisfy the formula. This happens for the formula '''(a = x)'. and if the formula inside the abstract does express a property (as described in §4. although step (2) of the Evans argument indeed turns out to be valid. On this second way of interpreting abstracts.) We describe such a non-standard interpretation of abstracts in Parsons and Woodruff (1995). On the assumptions used in §4. This nonstandard interpretation of the abstracts does not improve Evans's proof. (The non-standard interpretation is this: we suppose that an abstract stands for whatever property determinately holds of exactly the objects that determinately satisfy the contained formula. the inference from (1) to (2) is fallacious. abstracts are just ways of reexpressing other formulas that do not use this mode of expression. On this interpretation. and the semantics of formulas containing abstracts is completely parasitic on that of the formulas one gets by eliminating the abstracts. and indeterminately holds of exactly those objects which indeterminately satisfy the formula. step (5) is again fallacious. but not necessarily one that the description part of the definite description describes. one might require that abstracts can only be used if they stand for properties.f 7 There is yet a third way of interpreting the abstracts. but not require that the property that an abstract stands for holds of exactly those objects which determinately satisfy the contained formula. Otherwise. but in such a way that the abstraction principles are not necessarily satisfied. Recall that the contra positive of Leibniz's Law does not hold for formulas in general. though it holds for the predication of genuine properties. it determinately fails to hold of the rest. and DDiff 49 formula in the abstract. the formula will be determinately dissatisfied by certain objects which the property assigned to the abstract indeterminately holds of. it is possible to give a non-standard account of abstraction in which an abstract is always guaranteed to stand for a property. for on this interpretation.Evans Argument. Specifically. the abstracts in the proof are not guaranteed to stand for properties. and determinately fails to hold of exactly those objects which dissatisfy the formula. (The argument reduces to the shortened version discussed above. (This is similar to Frege's "chosen-object" treatment of definite descriptions. it is determinately dissatisfied by a (making step (3) true) but it is indeterminate whether a possesses the property that the abstract stands for (making step (4) lack truth-value). If the abstracts of Evans's proof are interpreted in this way.it will express a property which agrees perfectly with the property assigned to the abstract.3 below there will always exist such a property. Properties.

the abstraction operator on the outside must bind a variable within the scope of the indeterminacy connective: This leads one to suspect that we might be dealing here with some kind of non-extensionality. then one must hold that the indeterminacy connective introduces a referentially opaque context. the connectives '!' and 'v' do not create non-extensional contexts. Then the indeterminate identity of a with b is expressed as: [*]vVP(Pa {:=} Pb). and DDif( 4.2 WHY THERE IS NO PROPERTY OF INDETERMINATE-IDENTITYWITH-AN-OBJECT The argument above proves that belief in indeterminacy of identity forces one to deny that there is a property of being indeterminatelyidentical-with-o. write: VP(Pa {:=} Pb). As noted in the last chapter. Properties. The question whether there is a property of "being indeterminately identical with a" is then the question whether the following abstract stands for a property: ~ e. Instead of writing 'a = b'.g. where 0 is an object which is indeterminately identical to something. But why should this be the case? I t is apparent that a property abstract purportedly expressing such a property must "quantify into" an indeterminacy connective. it is more closely associated with the paradoxes of naive set theory. This can be seen without talking about identity at all if we replace identity by its definition in terms of properties. and one may freely intersubstitute coreferential names. or neither) and false if they disagree in truthvalue (one side is true and the other is false). where the quantifier ranges over properties. Burgess (1990: 269) suggests that if the Evans argument is not valid.50 Evans Argument. or some kind of modal fallacy. one may freely existentially generalize on terms within their scopes. The phenomenon at work here is unrelated to oddities of non-extensionality or of semantic paradoxes. . false.8 I think this is not the case at all. and where the biconditional is defined to be true if both sides have the same truth-value status (true.

For then a lacks that property: .. and a problematic set is defined using a set abstract that quantifies over those sets.. But the truth of this sentence contradicts the truth of [*].A.. 51 Suppose it does. So b has the property and a lacks it: A. Either . The reasoning thus resembles that of the RusselJ paradox in set theory.. Properties.x[v\fP(Pa <=> Px)].) I see no way around the facts discussed above. Then we can refute [*].. The force behind the reasoning thus comes from the fact that identity is defined in terms of what properties there are.x[v\fP(Pa <=> Px)]a because this is false: v\fP(Pa <=> Pa).. And b has that property: AX[v\fP(Pa <=> Px)]b because [*] is true.. So if the abstract stands for a property that is in the range of its own property variable. We thus have a choice to make: either deny that indeterminate identity is possible or deny that abstracts covertly employing quantification over properties (such as the one above) always pick out a property from among the ones quantified over in the definition of identity.AX[v\fP(Pa <=> Px)]a..Pa). and DDiff A. we can show that [*] is not true. Existentially generalizing on the abstract denoting the property then yields the following as true: ~P(Pb & . (Identity between sets is defined in terms of what sets they have as members..x[ v\fP(Pa <=> Px)]b & . and a problematic property is defined using an abstract that quantifies over those properties.. The condition in the abstract is cleverly designed to conflict with its yielding one of the properties quantified over (if any objects are indeterminately identical with a).. The condition in the set abstract is cleverly designed to conflict with its yielding one of the sets quantified over...Evans Argument...

and defenders of indeterminate identity can happily reject its abstraction steps as illustrative of their point of view.::3x::3y[!<!>x & !. So the opponents of indeterminate identity can happily give the Evans argument as an illustration of their point of view. and the choice of one of these conditionals over another is not crucial here. 10 The clauses in the conditional form of DDiff contain determinacy and indeterminacy connectives that force both the antecedent and the consequent to be truth-valued. and DDiff option is consistent. which is used throughout the text. Coupled with our definition of identity. Woodruff is responsible for the name 'definite difference'. I have used my favourite conditional connective.<jJy & vx = y].... Equivalently:lO 9 Woodruff and Parsons (1997). (See §6...52 Evans Argument. the framework gives us a necessary condition for a formula to express a property. For consistency. and of the formulas that generate such abstracts as expressing (or not expressing) such properties. Then a necessary condition for <!>x to express a property is that satisfaction and/or dissatisfaction of it by objects make a definite difference in their identity. 4. Properties.. The condition says that there are no objects x and y such that x definitely satisfies <!> and y definitely dissatisfies <!> and yet it is indeterminate whether x is y. Following Peter WoodrufC I call this condition "DDiff" for "definite difference". is the latter.) .. My stance. In notation: DDiff: .. In what circumstances does a formula of one free variable express a property? The framework sketched earlier is silent about this question for atomic predicates. Let <!>x be a formula with one free variable x. Neither of these stances can refute the other side. Noonan (1990) and (1991) calls the principle that DDiff holds of every formula the "Principle of the Diversity of the Definitely Dissimilar". of course.3 DDIFF: THE CONDITION OF DEFINITE DIFFERENCE It is convenient to speak of abstracts as standing for or not standing for properties. So all conditionals that agree with the classical material conditional for parts that have truth-value are equivalent in this context.1 for discussion of the alternatives. but it is sometimes informative about complex predicates.

If <!Ix does not satisfy DDiff...VP(Px <=> Py)..VP(!Px <=>! Py). that property would be definitely possessed by an object x and definitely lacked by an object y. x would definitely not be identical with y. and DDiff 53 DDiff: '<ix'<iY[!<!Ix & !-.. then it cannot express a property. that is. the whole formula is bivalent in virtue of its form." In certain circumstances. So this is not an independent thesis. we can prove that certain formulas do not express properties. by our definition of identity. because we can prove that they do not satisfy DDiff. The converse is sometimes possible as well. it is just a logical consequence of the worldly Leibnizian definition of identity that we are using. it is sufficient to show that a formula of this form cannot be true: vVP(Px <=> Py)& !Qx & hQy... it is inconsistent with: .. when 'v' is expanded by its definition. and thus it entails !-NP(! Px <=> !Py).. With this definition. For that. But this is inconsistent with 'vVP(Px<=>Py). TIlen the instance of DDiff for equivalent to this form: DDiff: ~3x3y[vVP(Px<=>Py)& !Qx& !~QyJ. For if it did express a property. is where it is understood that 'Q' stands for a property... our definition of identity is: a = b =dr VP(Pa <=> Pb). Suppose that a formula is constructed entirely from primitive predicates that 11 This can be seen as follows.Evans Argument. and thus. The formula ''i7(a = x)' discussed above is an example of this (if there is at least one object that is indeterminately identical to a). contrary to the clause in DDiff of the form ''i7X = y'.. But the last conjunct entails . . Then an inference based on the last two conjuncts yields: vVP(Px<=> Py)& 3P(!Px& !~Py).!. So suppose it is true. and thus is the sort of thing that the quantifier 'V P' may be instantiated for.. Using property quantifiers. DDiff takes this form: DDiff: ~3x3y[vVP(Px <=> Py) & 'Q>x & !~Q>y]... Properties. Q> Now suppose that Q> does express a property.<!Iy => --on = y]. Q. We need to show that this form of DDiff must be true.!VP(Px <=> Py) & . Since the smallest ingredients of this formula are preceded by'!'.

v. and that the logical symbols in the formula include only the identity sign. and the connectives '&'. 'v'. Only if we include a connective such as '!' or 'v' or '~' do we get formulas that do not satisfy DDiff12 (or if we start out with primitive predicates that do not stand for properties ). 4. . This commits one to a certain principle of plenitude for properties. and DDiff stand for properties and relations. it satisfies DDiff.'. and that logically complex formulas (made without '!' or 'v' or '~') whose parts express properties typically also express properties. and so I will indeed make the maximal assumption that any formula expresses a property if it is constructed entirely from atomic predicates that stand for properties.g.54 Evans Argument. For the philosophical issues discussed below. and the quantifiers ::3 and V. I will assume hereafter that any formula of one free variable that satisfies DDiff expresses a property. it won't matter whether we adopt this principle or something weaker.4 ABSTRACTS AND PROPERTIES In the discussion above it became clear that some abstracts. then we know it can express a property. We can assume that there are hardly any properties at all. such as the one employed in the Evans argument. that logical combinations of predicates expressing properties typically do not themselves express properties. Then if <jlx is such a formula with one free variable. We did not show that the abstracts cannot stand for some other kind of prop12 See Keefe (1995) for a discussion of why we should not expect prefixing a predicate with '. Properties. If a formula satisfies Ddiff. cannot stand for one of the properties in terms of which we define identity. Or we can assume instead that there are lots of properties. and from identity. On this assumption if 'Fx' expresses Fness and 'Cx' expressed Cness then there may be no property of being F and C for' Fx & Cx' to express. This is because the important point always turns out to be the converse: any formula that does not satisfy DDiff cannot express a property.' to preserve propertyhood. and '-. But does it? That depends on what we assume about properties. the connectives &. I find it convenient to use "property" talk.. and the quantifiers. and that abstracts constructed from such formulas stand for those properties. e. -'.

It is much less important what we decide to do with the ones that cannot stand for mundane properties. One mayor may not want to say that such an abstract stands for a conceptual property. this is an ontological notion of property. on the one hand. (Conceptual properties are investigated a bit more in Chapter 11. not the conceptual ones. It is essential for my purposes that we distinguish abstracts that can stand for worldly properties from those that cannot. using 'concept' and 'property' interchangeably. Properties. We can happily admit that Evans's abstract 'A. I am not certain this distinction can be clearly made. People sometimes talk about properties in another way. But there is no reason that I know of for assuming that conceptual properties validate the contrapositive of Leibniz's Law. If the distinction can be made.) When T say without qualification that a predicate does not stand for a property. Suppose there are two sorts of things that are commonly called "properties": real things in the world. and parts of our conceptual apparatus for representing the world. on the other. those that are part of our conceptual apparatus.Evans Argument. sometimes even construing properties as the meanings of predicates. When people feel that abstracts must stand for properties. In the account of indeterminacy that began in Chapter 2 we talked about properties and relations "in the world". it will be the worldly sort of property that T have in mind. and DDiff 55 erty.x[v(a = x)]' expresses a conceptual property. but I think these two different ways of viewing things are quite common. . which involves the worldly sort of properties. then it is clear that the theory I am discussing sees real identity in the world as arising with the worldly properties. they may be thinking of the other sort of properties.

incompatible with the possibility of indeterminate identity. including arguments based on the logic of conditionals. and some people construe this as ruling out the possibility that there might be objects between which a claim of identity has no answer.1 QUINE: NO ENTITY WITHOUT IDENTITY W. arguments to the effect that indeterminate identity cannot be understood. Does Quine's slogan conflict with the view under consideration here? I don't find Quine's slogan. 5. aside from Evans's proof. The slogan is fuelled by a discussion of merely possible entities in "On What There Is". in which he expostulates: . mostly to the effect that positing indeterminate identity is inconsistent with some known truth.5 Non-Conditional Disputations In this chapter I discuss a number of attempts to prove that there can be no such thing as indeterminate identity. and arguments to the effect that there are better accounts of the data. as intended by him. These are straightforward arguments. It does not make sense to survey the whole literature on this topic. The arguments discussed here are limited to ones that do not turn on the logic or semantics of conditionals. V. I begin with the three considerations that. have been mentioned to me most often by others. Other sorts of arguments are confronted later. arguments asserting that indeterminate identity makes it impossible to refer to objects. Quine has made famous the slogan "No entity without identity". I include a representative sampling of the sorts of arguments that people commonly find persuasive.

it is possible to maintain a methodology committed to bivalence. all objects are identical with themselves. It is also possible to abandon bivalence if that makes things go more smoothly. and its fame alone has no efficacy. Not because the objects are the wrong sort of things to which to apply identity.Non-Conditional Disputations 57 Or. then this fact by itself does not refute indeterminacy. Quine's slogan does not rule out indeterminate identity because it does not address the issue. and al\ are distinct from one another if 'another' is read strongly to mean "definitely not the same". and to expect a correct response from the alternatives yes. it is meaningful to ask of any objects a and b whether a is b. Further. Quine sees the question of bivalence as a methodological choice. 5. but just because that is the way things go (or fail to go). The slogan is now simply a statement of what is at issue. it is possible because it is always possible to interpret recalcitrant data in such a way as to preserve our methodological choice. it merely claims that in certain circumstances when you apply the concept you get no answer.2 METHODOLOGICAL BTVALENCE As mentioned in Chapter 1. finally. Suppose we reinterpret the slogan so as to require a determinate yes or no answer to an identity question whenever posed? Since Quine did not propose the slogan under that interpretation. In any given science. is the concept of identity simply inapplicable to unactualized possibles? But what sense can be found in talking of entities which cannot meaningfully be said to be identical with themselves and distinct from one another? Quine (1961: 4) What Quine rejects here is a kind of entity to which the concept of identity is inapplicable. we need to see why the choice should be made against . Furthermore. or neither yes nor 110. An endorsement of indeterminate identity does not deny that the concept of identity is applicable to objects. there is only a decision to be made about how to theorize. It should be clear that if it is a methodological choice that we face. no. it is no surprise that he gives no arguments for it under that interpretation. For a refutation. There is no fact of the matter about whether our utterances are bivalent or not.

An example of both points is given in Chapter 11 where classical Zermelo-Fraenkel set theory is found as a bivalent subtheory of a general non-bivalent theory of sets. he may not have had in mind the clear and well-behaved logic sketched in Chapter 2. All that you need to do for your bivalent subtheory is to explicitly assume bivalence of application for its primitive predicates. 2. When Quine (1981: 94-5) rejects the "fuzzy and plurivalent alternatives" to bivalence. It is especially important in this connection to recall that reductio ad absurdum is retained in this framework as a refutation technique. and that is not a wise methodology for anybody. You can show that a claim is not true by deriving a contradiction from it. replace '~S' by '~!S'). unless you have reasons to believe that the claim is a bivalent one. including options that hardly anybody understands. A second fear is that abandoning bivalence will leave you in a position of not knowing what follows from what. I think that one factor that leads people to suppose that one should choose bivalence is fears about what an alternative will bring. I see no reason why this would be different in a non-bivalent framework. If you have a particular bit of subject matter that is nicely accounted for by a theory based on bivalence. and expanding the available tautological equivalences by the principles of logic from Ch. I have not explored the philosophical consequences of doing this. then it needn't be abandoned at all. A mark of science is supposed to be that hypotheses are confirmed by their instances. then you open a Pandora's box of alternatives.e.! Another mark is that a hypothesis is a serious scientific option only if it is falsifiable (can I It may be important in this connection to explore how probability theory might be affected by non-bivalence. You lose classical consequences only when you do not assume bivalence for the claims with which you are working. It is true that if you simply abandon bivalence without settling on any particular alternative at all. . The first fear is that centuries of accomplishments that are formulated under the assumption of bivalence will need to be jettisoned. But that is not what is suggested in this book. Giving up bivalence as a general principle does not mean abandoning it in every case. You merely cannot use this to establish that its negation is true. This can be done by replacing negation by exclusion negation in tbe basic axioms (i. Then no classical consequences are lost. The nonbivalent framework used here is really quite conservative and well behaved.58 Non-Conditional Disputations indeterminacy. But this is not so.

the foundations of quantum mechanics are unsettled at this point in time. And if such disruptions became commonplace. nor of the psychological human development and interaction with which psychology deals.g.Non-Conditional Disputations 59 be shown to be false) within one's current methodology. 2 Suppose a person undergoes a kind of disruption. does not deal with such disruptions. some of the most troublesome cases in the history of identity puzzles come from this locale. One reads over and over in this literature that there is no truth to the matter of whether a certain particle has simultaneously both position and momentum. e. that is easily explainable in physical terms. who thinks of identity puzzles as arising between ideologies. a partial brain transplant. This seems to have been the actual view of a substantial number of people. the question of whether the same person emerges is not a question in either physics or psychology. In a nonbivalent framework this must be liberalized to: a hypothesis is a serious scientific option only if it is refutable (can be shown not to be true) within one's current methodology. In fact. Whether they were right to do so is beside the point. working in quantum mechanics. The point is rather that many thoughtful scientists and philosophers have in actual fact been willing to abandon bivalence for the sake of having the simplest theory. and not so neatly with a bivalent one. It is possible to imagine that the world contains indeterminacies. but why would anyone choose to do science on that basis? The answer may lie in theoretical simplicity. with bivalcnce holding within each ideology. It is their thoughtful willingness to do so that illustrates that abandoning bivalence is a genuine and reasonable option in doing science. Perhaps the personal disruption is a question in sociology? Sociology. Compare also Zemach (1991). It is conceivable that you will find yourself confronted with data that can be accommodated neatly with a non-bivalent theory. The indeterminacy would be thus 2 See Parsons (2000). Even if bivalence were chosen within the respective sciences by individuals and communities of scientists. Bivalence within these sciences leaves this puzzle unaddressed. based on normal occurrences. There is no puzzle here about the identity of physical systems with which physics deals. I don't claim that this view is correct. the science would probably pose a new category in terms of which to study itspeaking perhaps of disruptees. . both scientists and philosophers. this cannot be decisive about claims that occur between and across the individual sciences.

Then this pair <x. so as to accomplish the much discussed reduction of everything to physics.3 SALMON'S ARGUMENT BASED ON ORDERED PAIRS Nathan Salmon attempts to refute the possibility of indeterminate identity. Of course. This does not refute indeterminacy. The philosophical task is to address the puzzles raised by the original ones. since it is determinately true that x is onc and the very same thing as itself. for such is the commitment of each science on its own (or so we are supposing). it ignores it. But that is not accomplished at present.y> is quite definitely not the same pair as <X.. The tendency of scientists to search for new concepts leaves unsettled the issues posed in terms of other concepts. or something of this sort. 5.60 Non-Conditional Disputations avoided. Perhaps science works better by formulating bivalent theories using newly invented concepts that are carefully crafted to fit into bivalent scientific laws. if at all. and we don't know what form it will take if it does occur. Burgess (1990: n. It is certainly not obvious how old problems disappear-if they do so-when new categories are introduced.X>. and so on. In the meantime. which is why I discuss it independently here.g.. Other puzzles similarly arise in the interfaces. 7). there is no objective fact of the matter) whether they are one and the very same thing. I already discussed it in the last chapter. the person-andtheir-body at the interface of biology and some science of persons. such that it is vague (neither true nor false. If it is. a methodological commitment to the search for bivalence in each science is not sufficient to close the door on an exploration of other options for claims between the sciences. We already knew that we could refocus our search for scientific laws by employing new categories. Others see it as a distinct argument. or whether and how identity puzzles will be affected. . arguing as follows: 3 [S]uppose that there is a pair of entities x and y . But to avoid it is not to solve the identity puzzle. the cat and the p-cats at the interface of biology and physics. one might hope that the sciences will eventually grow together. It follows that x and y must . Some writers see Salmon's argument as identical in structure to Evans's (e. indeterminate.

No principle of bivalent logic or of bivalent set theory (or ordered-pair theory) should be taken to validate the inference to (3). . if it is indeterminate whether x = y then it is indeterminate whether the pair <x.y> = <x.( <x.y> is identical with the pair <x. Salmon cites no reason for (3). This is fallacious.Non-Conditional Disputations 61 be distinct. so we can only speCUlate about how to valida te it. by the theory of ordered pmrs From (4) I agree with every step except for the inference from (1) and (2) to (3).( <x. Salmon (1981: 243) The argument appears to have this structure: 4 (1) (2) (3) (4) 'l(x=y) -''l(x = x) -.'l(x=y) (2d) 'l(x = y) & -''l(x = y) (3) -.y> = <x. From (2b) by preservation of truth. From (1) and (2c) From (2a)-(2d) by indirect proof. but its conclusion is not.x>. This complicates the notation without having any particular effect on the validity of the steps. But then it is not vague whether they are identical or distinct. assuming the principles of ordered-pair theory: if the pairs in (3) were identical.(x = y) (5) -''l(x = y) Hypothesis to be disproved Truism From (1) and (2) From (3).y> = <x. that would contradict (l) and (2). (2c) -. Indirect proof is invalid in a logic that permits sentences that lack < Following Salmon's words precisely would require us to place a determinately sign '!' in front of lines (3) and (4). The additional steps between (2) and (3) would then be: (2a) Suppose <x.x» -.X> (2b) Then x = y Hypothesis for indirect proof From (2a) by the theory of ordered pairs: pairs are identical iff their first and second members are. A natural way to think of filling in the proof would be to use indirect proof.x» Every step of this subproof is fine. since the inference crucially involves non-bivalency.

x» From (2a)-(2d) by indirect proof. 5. ordinary principles of set theory may be applied in the presence of indeterminate identity. I argue that set theory meshes nicely with indeterminate identity.4. the theory I develop here without supervaluations will sometimes contradict data-like sentences that we are inclined to accept. and one might then argue against indeterminate identity on grounds of utility. Second. this would be very serious. but J know of no alternatives that turn it into a conclusive proof.1 Ways of reading sentences In doing semantics one often encounters sentences that elude a simple classification in terms of truth-value. For example. In Chapter 11. First. Instead of (3) you can conclude: (3') -. But I can't avoid talking about them. and some of them are based on superva]uational readings of sentences. So we need to know what supervaluational readings are.!(<X. If we had to abandon set theory in the face of indeterminate identity.y> == <x. I do not appeal to supervaluations in formulating the theory of this book. it is only sufficient to conclude that (2a) is not true. This is for two reasons. We either tend to read them supervaluationally or superresolutionally. I will discuss alternative accounts. 5. it is . But this is not sufficient to complete the original proof. Such contradiction fuels refutations of the theory that are discussed in the remainder of this chapter. and to what extent. and when they are read in this way they do not contradict the theory. and so disproving (2a) is not sufficient to validate its negation. J will suggest that we are inclined to accept some sentences in a way that apparently contradicts the theory of indeterminate identity because we naturally read them non-literally. There are a number of other ways to try to fill out the above reasoning.4 SUPERVALUATIONS AND SUPER-RESOLUTIONS A popular way to deal with sentences that lack truth-value is to invoke supervaluational readings.62 Non-Conditional Disputations truth-value. This type of argument does however raise the question of whether.

since it contradicts a false sentence. The supervaluation technique is typically used for one of two purposes: 5 (1) One application is to ambiguous sentences." . If the sentence comes out true on some resolutions and false on others. One gets different meanings depending on how one interprets. others (e.2 Supervaluatiol1s Supervaluations provide a way of reading sentences. we will have to confront examples in which speakers naturally interpret sentences differently. In this text we will have occasion to focus on additional sources of multiple readings of sentences. since there is no king of France to be bald or not. Fine applies the technique to (semantic) vagueness. the sentence. 5. though not necessarily by the same observer. then it is "supertruthvalueless". Russell) see it as ambiguous between true and false. and since the sentence is so simple.Non-Conditional Disputations 63 apparent that when asked to judge the sentence 'The king of France is not bald'. This successfully mirrors ways that we often assess 5 Supervaluations were originally proposed in van Fraassen (1966).g. or that every ship was seen by some observer. a way that sometimes gives truth-values to sentences that would otherwise lack them. Since these are reactions of native speakers who are in possession of all the facts. see the sentence as one whose surface syntax does not single out a unique meaning.4. Call an ambiguous sentence "supertrue" if it is true on all ways of resolving the ambiguity and "superfalse" if it is false on all ways of resolving the ambiguity. many pcople.g. which can either mean that some one observer was such that (s)he saw every ship. some people see the sentence as obviously false. Fine (1975: 284) emphasizes the application to ambiguity: "An ambiguous sentence is true if each of its disambiguations is true. I take for granted that the reader is familiar with scope ambiguities. Frege) see it as lacking truth-value. myself included. others see it as true. Semantic theory is then challenged to explain how these many different readings may arise from such a simple sentence. since there is no king of France to be bald. and these different "readings" will need to be accounted for. such as in a sentence like 'An observer saw every ship'. and still others (e. or "reads". In inspecting the phenomenon of indeterminacy.

For example. and you consider how to flesh out these units consistent with whatever determinacy is already there. as with my own approach. and otherwise supertruthvalueless. . contra position of validity does not generally hold in supervaluational logic. For example.l avoid this. superfalse if it is false on all extended valuations. complications abound when one has notation in the object language for both 'true' and for 'supertrue'. you focus on the basic units of the sentence that are responsible for its lack of truthvalue. my'!' converts a true sentence into a true one and all others into a false one. Or else the lack of truth-value is due to the fact that some singular term lacks a referent. 'Fa v -. we call such an utterance true without bothering to resolve the ambiguity if it is true no matter how the ambiguity is resolved. but they are logically supertrue. It does not necessarily convert a Sllpertrue sentence into a true one. the lack of truth-value may be because a predicate does not apply either truly or falsely to certain objects. Note that the general theory of supervaluations is more complicated than my survey of it indicates. and you consider various ways of arbitrarily assigning it a referent. The advantage of reading sentences supervaluationally in this way is that sentences that are not logically true because of the possibility of truth-value gaps may be logically supertrue because of the way the gaps get filled in. I avoid supervaluational readings for the same reason others invoke them: they make the assessment of a sentence less directly a matter of how its parts relate to the world than do literal readings. For example. (2) A second application of supervaluations is to sentences which lack truth-value for reasons other than ambiguity. and this restores a measure of neatness to one's logical system. To evaluate a sentence supervaluationally. For example." In theory formulation. Then a sentence is supertrue if it is true on all extended valuations. Call any way of extending the original semantics so as to get a truth-value for every sentence an "extended valuation".64 Non-Conditional Disputations real-life utterances that are subject to different interpretations.Fa' is supertrue n There are limits to the neatness of supervaluationallogic. so you consider how to extend its extension (the objects it is true of) and its anti-extension (the objects it is false of) so as to cover these additional objects. classical tautologies are often not logically true when gaps are allowed.

5. What must be kept in mind is that a supervaluational reading is always parasitic on another. false. read literally. we will need to consider an additional method of reading sentences.Non-Conditional Disputations 65 because of its logical form. But you should not reject a claim that a sentence.4. Simply put. read supervaluationally. Supervaluations are a nominalistic creation. lacks truth-value. the more direct semantics is preferable. I will certainly appeal to them on occasion to explain why we tend to judge certain sentences true when my theory says they lack truth-value. The connection between the truth-value status of a sentence and how its parts relate to the world is typically simpler to calculate than the connection between its supervaluational status and how its parts relate to the world. quite apart from whether 'Fa' is true. and I have no reservations about agreeing with this. Since I am primarily concerned with depicting the world. First. then on my view so does'S or not-S'. or neither. not in elegant ways of avoiding such gaps by interpreting them away. one typically deals with pieces of language and how they relate to the world. This is exactly the sort of data that supervaluations were introduced to account for. For example. we do tend to read complex sentences supervaluationally sometimes. when supervaluational readings let logical form wipe out differences of content. The theory produces the literal reading. I am interested in assessing the causes of truth-value gaps. has a truth-value. You can always read the results supervaluationally. Although I avoid using supervaluations in formulating my own theoretical views. Making 'P' more precise by an .3 Super-resolutions In discussing the worldly theory of indeterminacy. I also think that certain key claims involving indeterminacy are more easily formulated presuming a literal reading than a supervaluational one. suppose that a predicate 'P' stands for a property. yet we are inclined to think that'S or not-S' is in some sense correct. without any discussion of properties and without assuming that the world itself is indeterminate. just because the same sentence. So when we try to combine the traditional method of supervaluations with the theory under discussion here there are a couple of oddities that need to be addressed. and that is the one I typically focus on. if'S' lacks truthvalue. which I call the "literal" reading.

That leaves the third conjunct.) It is also the sort of sentence that could not be true because of its form. we may have an identity statement without truth-value even though the referents of the names are clear (as discussed in Chapter 9).) Second. just extend the extension of 'P' to include b. Here is an example that combines these two concerns. How should a supervaluation treat such a statement? If identity can have its extension "clarified" at will. (An example will be given below. If no constraints are placed on the extended valuations. then the iden- . resulting in a sentence which is supertruthvalueless. then there are extended valuations that make the sentence true. But in our framework.66 Non-Conditional Disputations extended valuation may result in its no longer standing for a property. and the second conjunct will then be false. to consider. just as the original is. (To make the sentence false. and thus it is the sort of sentence that supervaluations are supposed to address. and so does 'a = b'. just extend the anti-extension of 'P' to include b. and consider the sentence: Pa & -. If' P' still stands for a property. and 'Pb' lacks truth-value. it is false on every extended valuation that gives it a truth-value. to be resolved by clarifying the referents of names and the extensions of predicates. that is. we need to worry about how this change may affect the readings of sentences that contain it and that also contain identities of objects that have the original property. Since whether 'P' stands for a property or not may affect how it interacts with identity. Suppose that 'P' stands for a property. the identity. But the method of supervaluations alone does not give this result without supplementation. supervaluations will not produce supertruth when they are expected to. and ones that make it false. (Suppose that 'P' stands for the property p. it is generally assumed in the case of supervaluations that all of the "input" indeterminacy is semantic. then it loses its special relation to the having and lacking of properties. but if its extension cannot be altered. This sentence might lack truth-value when read literally. which makes the second conjunct true. we are supposed to be able to address the apparent logical falsehood of the sentence by showing that although it lacks truth-value. To make it non-false. which holds of Cl and neither holds nor fails to hold of b. it is superfalse. and that it is indeterminate whether a is b. Then 'Pa' is true.pb & a = b.

we might either leave it alone. But there is nothing in the method of supervaluations to require that' P' still stands for a property.) What is desired is a technique similar to the bare supervaluation approach that respects the details of the theory under discussion. by taking a property that neither determinately applies nor fails to apply to any given object and making it determinately apply or determinately not apply to that object. since the first two conjuncts force a to have that property and b to lack it. This is to ask about truth resulting from resolution of the world without alteration of the language that depicts the world. in extending it we may have converted it to one of those predicates that do not stand for properties.. and thus the whole sentence comes out true. If we leave it alone. In a large number of cases. instead of truth that results from making our language more determinate. And we can ask further which sentences are true in every such resolution.Non-Conditional Disputations 67 tity must be false. In either case. Here is one such. That is. of IV. Then in any resolution. If we adjust its extension at will. Call this "super-resolved truth". Suppose that the world is not completely determinate. and from readjusting the extension of the identity predicate as required by the Leibnizian definition of identity in terms of properties and relations. If . we can ask about the truth-values of sentences that result from altering the semantics of predicates that stand for properties so as to correspond to the new ways that those properties behave under r. except that it is truth that results from considering making the world more determinate. we can ask whether it is true in a resolution. and we have produced an extended valuation that makes the whole sentence lack truth-value. So how do we treat the identity? As mentioned above. or we might adjust its extension at will. If a sentence contains only primitive predicates that stand for properties. we can let it be true of a and b. identity will be completely determinate for any pair of objects. the extended valuation makes the sentence non-false. Now consider various ways in which it might become determinate by determining undetermined states of affairs. the third conjuncl stays indeterminate. is such a resolution. in which case the identity clause is not yet determined. super-resolved truth will coincide with unresricted superva]uational truth. even though they are arrived at quite . Super-resolved truth is like the supertruth due to supervaluations. and identity. and thus it is not superfalse as desired.. Call the result of such a way of making world IV completely determinate a resolution of w.

however things are. either they're the same thing or they're not: \ix\iy(x = Y v -. then there is a property that holds of a and not of b.x = y)'. and when I ask myself why. I call the approach that uses all arbitrarily chosen extended valuations the "unrestricted supervaluation" approach. simply limit the extended valuations of names and predicates to valuations that they could have in some resolwio17. . But not always. when the theory says that they lack truth-value. it is super-resolvedly true. The example discussed above is superresolvedly false. In some cases (according to many philosophicallogicians. If the 'are' is read as 'determinately are'. I think we are reading the sentences in a super-resolved way. or false.x = y).68 Non-Conditional Disputations differently. Certainly I need some explanation of why people uncorrupted by philosophical theory are tugged in the direction of saying that for any two things. 1 will continue to use the term 'super-resolution' because it suggests the details of how the valuations are 10 be determined. And in some other cases in which unrestricted supervaluations yield no (super)truth-value. I find that I am inclined to say something like: Well. I myself feel this pull. not truth simpliciter. The sentence '\:j x\i y(x = Y v -. read literally. this is because we are reading the sentences supervaluationally. lacks truth-value according to the theory. then I am asking about resolved truth.) The remaining three topics in this chapter bring in molecular sentences in one way or another. (Its supervaluational truth status depends on how the theory of supervaluations is extended to handle worldly indeterminacy of identity. but it is true on all resolutions. and I agree). We alter the extension and anti-extension of' P' indirectly by extending what the property that it stands for is true or false of. 7 The reason for keeping super-resolutions in mind is the same as for unrestricted supervaluations: sometimes we make judgements that certain sentences are true. so let me summarize my position on these. so does the disjunction-if it is read straightforwardly as I have explained it 7 The super-resolutional account can be interpreted as an applicafion of the supervaluational account. they have to be like that. If both disjuncts of a disjunction lack truth-value. If we make the world more determinate by letting the property p above be false of b. This is· because super-resolutions respect the ontology. and the identity is forced to be false.

but when considering statements made in other contexts we must be alert to other ways of interpreting them. It is always natural to read it this way when the statement has the form of an explicit instance of the Law of Excluded Middle: "A or not A". but he touches briefly on the possibility of indeterminacy of identity.l4. and his defence of bivalence. false.Non-Conditional Disputations 69 above. true and false. WilIiamson's position is a broad one.S the line of argument depends crucially on his rejection of multivalent logic. Read straightforwardly. Read supervaluationally. he rejects it there (1994: 255) because it requires a formulation in terms of many-valued logic. in which something like a "truth-functional" language is used. not directed merely at indeterminacy of identity.14.1 focus here on the main point of §4. and neither. which has been rejected in §§4.2 Delenninacy in the World.5 AN ARGUMENT BY WILLIAMSON AGAINST NON-BIVALENCE Timothy Williamson's book Vagueness (Williamson (1994» is devoted primarily to linguistic or conceptual issues about vagueness. 'A or not A' is true. he believes that there is no worldly indeterminacy of any kind. He considers a view substantially like the one discussed here. not supervaluationally or super-resolutionally. 5. Tracing through his reasoning. Although I employ only two truth-values. When I make statements myself I always intend that they be taken straightforwardly. 'A or not A' lacks truth-value when 'A' lacks truth-value. since any way of assigning a truthvalue to 'A' makes the whole disjunction true. However it is sometimes natural to read a disjunction supervaluationally. and variants of Williamson's arguments apply to this stance as well. he argues against the claim that: "the degree of S WilIiamson discusses indeterminate identity in his §9. Specifically.11-4. . He phrases this as an attack on the idea that statements might have some degree of truth other than 0 (complete falsity) or 1 (complete truth). Anyone who believes in some other degree of truth must give a coherent account of how our language embodies this. The same holds for super-resolutional readings. I admit three truth-value statuses: true.

. Wil1iamson's discussion of 'He is awake and he is not awake' needs to be addressed because it appears to rule out lack of truth-value as well as intermediate degrees of truth-value. for so used. Thus degree-functionality fails for conjunction. At some point they have the same degree of truth. 'He is awake and he is asleep' has no chance at all of being true. By what has just been argued. The same point can be made with 'He is not awake' in place of 'He is asleep' . however. He argues as follows: Now imagine someone drifting off to sleep. this is because the 'sort of' modifies the meanings so that the second . According to the degree theorist. Our man is not in an unclear area between the cases in which the conjunction is true and those in which it is false. How can an explicit contradiction be true to any degree other than O? (ibid... Under this interpretation. But suppose someone actually asserts 'He is awake and he is not awake'. as the former falls in degree of truth. What happens if 'He is awake' lacks truth-value? I hold that the conjunction 'He is awake and he is not awake' also lacks truthvalue.. Williamson is right in questioning whether this is how a person would normally intend such an utterance. hold that a contradiction can lack truth-value if its parts do. although neither conjunct is. 136) I do not hold that any contradiction can be true to any degree other than 0 (falsity). it should not have an intermediate degree of truth.. what would they be likely to mean? ll1e most natural interpretation would be that they are saying this to emphasize that he is sort of awake and sort of not awake. since it isn't an utterance that one would normally make. the latter rises. one must be careful to distinguish what can be said of the conjunction from what can be said of each conjunct. If we read 'degree of truth' as 'type of truth-value status' then his discussion is pertinent to the view under consideration here. But how can that be? Waking and sleep by definition exclude each other. But on that interpretation it is not a contradiction at all. Since the conjunction in question is clearly incorrect. the conjunction 'He is awake and he is asleep' also has that intermediate degree of truth. an intermediate one. The sentences 'He is awake' and 'He is asleep' are vague. But this is hard to assess. both Williamson and J are wrong.. the sentence would be true.70 Non-Conditional Disputations truth of various compounds is a function of the degrees of truth of their components" (1994: 135). It is clearly incorrect. for there are no cases of the former kind . I do.

so read. So under this interpretation the utterance is not relevant to Williamson's query "How can an explicit contradiction be true?" This is not what Williamson intended to discuss. for there really is something odd about seeing explicit contradictions as indeterminate. some explicit contradictions are not false (they lack truth-value). He says this apparently because he thinks this must be true of any contradiction: "How can an explicit contradiction be true to any degree other than O?" The answer to his question is straightforward: "by having parts that lack truth-value". I am then comfortable in saying that both conjuncts lack truth-value. unless it has a truth-value. But suppose we want somehow to be able to construe any sentence that cannot be true in such a way that it is thereby false. There is a wide variety of contexts in which definite truth and falsity cannot be assumed. and no uniform approach to non-bivalent . So let us fasten on a reading in which the second conjunct is the negation of the first. This answers Williamson's question. It is appropriate to read a sentence supervaluationally if you are judging it based more on form than on content. This is probably because an explicit contradiction cannot he true. however. So the existence of instincts to call explicit contradictions false cannot refute such a view. and thc whole does as well. I agree with many of the points he makes. But any sentence can be read supervaluationally. in a situation in which the first conjunct is as midway between truth and falsity as you can get it. Succinctly put: Williamson objects to the lack of bivalence of any view that admits "truth-value status functionality" because. but not his concern. which is what Williamson urges that we do with explicit contradictions. Williamson thinks that the whole must be fully false. Williamson argues that a consistent use of supervaluational readings yields unnaturalness in some contexts. a sentence read supervaluationally is automatically false (or a sentence read normally is automatically superfalse) if its form prevents it from being true. So anyone who has a potentially non-bivalent language has a language that can read contradictions as false when you focus on form over content. and any explicit contradiction is like this. But I draw a different lesson. This is what supervaluational readings are for. though the discussion is too wide-ranging to sum up here. Elsewhere (Chapter 5). This. does not make it false.Non-Conditional Disputations 71 conjunct does not deny what the first asserts.

It turns on the relation between the existence of "fuzzy objects" and the existence of indeterminacy of identity. a fuzzy object then becomes . 5. A "fuzzy object" is an object with indistinct boundaries. The best we can do is to be clear about how our own language is to be taken. and I discuss higher-order indeterminacy in Chapter 12. Does it follow that it is indeterminate whether the rise is the mountain? Not automatically. Occasional contexts in which truth-status-functional readings of connectives are unnatural cannot tell against a theory formulated with such readings. certainly not without some additional argumentation. Pick a rise in the landscape with boundaries such that it is indeterminate whether they are the boundaries of the mountain.72 Non-Conditional Disputations language will be most natural in all of them. The indistinctness of boundary is usually construed as an indistinctness in whether a certain portion of space is inside or outside the object or not. and this is then recast as the question of whether the stuff in the disputed region is or is not part of the object.6 FUZZY OBJECTS AND INDETERMINATE IDENTITY One widely discussed issue is not strictly a refutation. I discuss one which relies on the logic of conditionals in the next chapter. so a major rationale for believing in indeterminacy of identity is thus undercut. An example: a mountain has indeterminate boundaries. And I make no such claim. Several writers suggest that people believe in indeterminacy of identity because they mistakenly think that it follows from the existence of fuzzy objects. But it does not. If indistinctness is interpreted as indeterminacy. If the only evidence one could ever have for indeterminacy of identity is easily reconstruable as evidence for fuzzy objects. the thesis of indeterminacy of identity could not ever be validated. Williamson has other relevant arguments. and to be careful not to misconstrue others. so long as one does not assert that such readings of connectives are the ones intended by all speakers of natural language in all contexts. though it often plays much the same rhetorical role.

x is identical with y if and only if x and y have the same parts. Mereological identity: For any objects x and y. and y part of z. We use 'part' here in the sense in which each object is part of itself. For example. For our purposes. These assumptions are to be understood strongly. z is part of also. and (iii) everything that x and y are both parts of. in the sense that (i) x is part of z. there is no part of the sum of x and y which is determinately not part of x. But there is a bit more to be said.g. The assumptions are ingredients of a theory of mereology (the study of parts and whales). Most people leave it at that. Uniformity of indeterminate parts: If it is indeterminate whether y is part of x.Non-Conditional Disputations 73 an object such that it is indeterminate what its parts are. For there are situations in which one can infer the indeterminate identity from the existence of fuzzy objects. As many writers either point out or argue at length.false. 5.6. Here are some assumptions that would make the transition feasible. exactly when the claim 'x and y have the same parts' is true. J Inferring indeterminate identity from fuzzy objects Although the existence of fuzzy objects does not entail that there is any indeterminate identity. the claim of mereological identity is that 'x is identical to y' is to be true. a fuzzy object may be taken to be as follows: x is a fuzzy object =df there is some y such that it is indeterminate whether y is part of x. then. In mereological sums it is understood that for any objects x and y it is determinately true that there is an 9 e. false. . or neither. it may do so with the help of some additional assumptions. Mereological sums: For any objects x and y there is an object z which is their mereological sum. then x is part of z. Transitivity: If x is part of y.9 the existence of a fuzzy object in this sense does not logically entail that there arc objects such that it is indeterminate whether they are identical. Burgess (1990). or neither. and (ii) y is part of z.

I clarify these assumptions in this way not because I believe that they are true when so clarified. but rather because I am giving an example of assumptions that would connect indeterminacy of parts with indeterminacy of identity. This rules out the possibility that a and care determinately identical. So a is not determinately the same as c. and they may be quite plausible when applied to certain sorts of . and vice versa. there is no determinate subpart of the sum of a and b (that is. We could argue as follows: By mereological sums there is an object c which is the sum of a and b. (2) By construction of c as the sum of a and b and by transitivity it follows that every determinate part of a is a determinate part of c. These assumptions are not true of objects in general. (2) There is nothing that is determinately part of a that is determinately not part of c. We will show that it is indeterminate whether c is identical to a. But in some limited domains. no determinate subpart of c) that is determinately not part of a. So there are assumptions with which one can argue from the existence of fuzzy objects to the existence of indeterminacy of identity. This rules out the possibility that a and care determinately distinct. or they have parts in a sense that does not allow us to infer their identity from the fact that they share parts. because there are many sorts of objects that either have no parts in a straightforward sense. In such a domain one could infer indeterminate identity from the existence of fuzzy objects. Consider an object a and some object b where it is indeterminate whether b is a part of a. and by construction of c as the sum of a and b. it is determinate that b is part of c. where the explanation of "sum" is definitive as just explained. By mereological identity it will be sufficient to show two things: (1) There is something that is determinately a part of c and not determinately a part of a.74 Non-Conditional Disputations object z which is the sum of x and y. And by the uniformity of indeterminate parts. the assumptions may be quite plausible. Arguments: (J) By hypothesis it is indeterminate whether b is part of a.

The question will be whether the building in which Smith is lecturing is the same as the building in which lones is lecturing. It is up for grabs whether 1 am a mereological object or not. Smith is lecturing somewhere within module A. and his view leads to an interesting case. but that there is nonetheless no such thing as indeterminate identity. Let 'J' abbreviate 'the building in which lones is lecturing'. they would yield hosts of examples of indeterminate identity. and ABC. But in other cases the assumptions may not be plausible at all. but it is indeterminate exactly what the parts of J are. If so. He assumes a structure with a kind of dumbbell shape. 5.7 COOK'S BUILDING The following example involves considerations of parts and wholes. I may be a fuzzy object. But Cook assumes otherwise. connected by a narrow walkway. though I am not certain. somewhat different from the one I discussed in Chapter 1. consisting of two modules. and thus 'the building in which Jones is lecturing' is a definite description that has no unique reference. If they were. Cook assumes that 'J' has a unique reference. Cook considers a building. I am a person. If this is so. and the example raises no genuine puzzle about identity. and 'S' abbreviate 'the building in which Smith is lecturing'. To make this plausible he argues generally against indeterminate identity. Cook (1986) takes the position that the Evans argument is flawed. with each of A and B being parts of ABC. C. Are mountains mereological entities of which these assumptions hold? I suspect not. I do indeed have a spatial location. and lones within module B. and so does'S'. My own instinct is that there are three determinate buildings: A. but it is unclear whether my identity is dependent on this. and likewise for S: . so the principle of mereological identity may be false when applied to me.Non-Conditional Disputations 75 objects. but there may be nothing with which I am indeterminately identical. there are two distinct buildings (B and ABC) in which lones is lecturing. and he also explains how at least one apparent case of indeterminate identity is instead merely a case of a fuzzy object. A and B. For example. B.

We may also in this case make a crucial assumption that cannot be made about all cases: that the buildings in question would be identical if they determinately had exactly the same parts. C must be determinately . If we sharpen the notion of being red.76 Non-Conditional Disputations "The denotation is not indeterminate. B. Since A is determinately a part of J. If we assume that in this case there are no scattered buildings.1 An argument from fuzzy buildings to indeterminate identity Let me digress for a moment to consider whether this might be a case in which indeterminacy of parts actually does entail an indeterminate identity. the boundaries of what is denoted are indeterminate" (1986: 182).) (ibid: 183) Cook's position is that'S' and 'J' each uniquely denotes an object with indeterminate parts. and that they would be determinately distinct if one of them determinately had a part that the other determinately lacked. it is determinately a part of 1. 'The colour of my car'. (If the colour of my car is on the borderline between red and orange. Now we can argue as follows: Suppose that J is determinately identical to S. So each of J and Shave both A and B as determinate parts. but there is no indeterminate identity of objects. we don't make the denotation of 'the colour of my car' determinate. then it is indeterminate whether the colour of my car is red. Then J and S have the same parts. and it is determinate that Smith is lecturing in A. has a determinate denotation. and since B is determinately a part of S. we are making it determinate whether this building includes [B]. it is also determinately a part of S. similarly J has B as a determinate part. since S is determinately the building in which Smith is lecturing. and sums made of them. He rejects the idea that clarifying what counts as a building is even relevant to the question of denotation or identity: When we sharpen the notion of a building. however.7. we are not making the denotation of 'the building in which Smith is lecturing' determinate. 5. S has A as a determinate part. we make it determinate whether the colour of my car is red. and C. Assume that the only parts of the building that are relevant in this example are A.

" To be clear: I disagree with Cook that this is a case in which there is determinate reference to buildings but in which it is indeterminate what their parts are. for this is necessary for them to be determinately distinct. but they seem to me plausible.Non-Conditional Di~putations 77 part of each as well. yet indeterminate whether C is? Indeed. it would if the parts were indeterminate. one has to look at the details of the case to tell. I think that in this particular case one should deny that the descriptions have unique denotations. Then there is no remaining part to be an indeterminate part of both J and S. 10 We could also assume at this point that C is an indeterminate part of one of the buildings. suppose that we move the parts A and B closer and closer together until C has shrunk to nothing at all and A and B share only a doorway. and argue by symmetry that it must be an indeterminate part of the other as well. but entirely unmotivated by our understanding of the setup. Then we have determinately distinct objects J and S. they must have indeterminate parts. C must be part of both or part of neither or an indeterminate part of each. . it must be that C is an indeterminate part of each. As a result. ll Sometimes there can be indeterminate part hood without indeterminacy of identity. So J cannot be determinately identical to S. Since they are fuzzy objects. I conclude that this is a case in which the indeterminacy of parts does plausibly lead to indeterminacy of identity. and one should not conclude that there is indeterminacy of identity in the world because of such a case. By parity. and the example collapses. with each of them having a determinate part (B for J and A for S). So it must be that A is determinately not a part of J and B is determinately not a part of S. 1U It follows that neither J nor S has any indeterminate parts at all. The mereological assumptions in this argument are certainly not beyond challenge. and sometimes there cannot be. so they will not disagree with respect to C. They must determinately disagree with respect to some part. Why should it be determinately false that B is part of S. Yet it is just as plausible in that case to say that J and S are fuzzy objects as it is in the original case. This is coherent. which is what Cook assumes. Suppose that J and S are determinately distinct. The conclusion would then follow that J and S agree in all their parts. contrary to the initial assumption. Or.

But Cook has an argument that this is not possible. (ibid. then. he says that being indeterminate identical with a is a property. Here is the statement of the conflict between indeterminacy of identity and LEM: [IJf indeterminacy is in the world then LEM does not hold-objects can be in three mutually exclusive states: they can be identical. and we need to look at that.) He goes on to explain how 'J == S' can be seen as a similar case. Here I think one should appeal to something like true-uIlder-an-acceptable-sharpening.2 Cook's refutation of indeterminacy of identity I have argued that if Cook is right about J and S having indeterminate parts. but that one cannot conclude from the fact that b determinately possesses it and a does not that a and b are distinct.) But he thinks a better argument against the indeterminacy of identity can be given. is to attack the view that objects can be indeterminately identical. Cook rejects the Evans argument. and (b) seeing that this does not prevent the disjunction's being exhaustive. (Cook 1986: 183) The defence of LEM is explained as follows: The trick to seeing that borderline cases do not force us to give up LEM is (a) seeing thal a disjunction can be true even though neither disjunct is true. (1 agree with this only on the assumption that he has in mind what I above called conceptual properties. they are indeterminately identical. one involving the law of excluded middle (LEM). he holds that indeterminacy of identity conflicts with LEM. Briefly. which is vague in a way similar to the way in which 'red' is vague. Thus. 184) Expanding on (a) he says: [Olne will not be able to explain the truth conditions for the statement that the car is red in a straightIorward truth-functional way. they can be non-identical. one might say that 'red or orange' is true of an object if under all acceptable sharpenings of 'red' and 'orange' either 'red' or 'orange' is true of it. and they can be indeterminately identical. To defend LEM. (ibid.7. The immediate problem with this explanation is that it is no longer clear why indeterminate identity is supposed to conflict with . because the identity contains a concealed appeal to 'part of the same building'. but LEM can be defended.78 Non-Conditional Di!>putations 5.

Thus the argument against indeterminate identity vanishes. the truth-functional reading is the only one available. is that Cook does not think that real indeterminacy of identity admits of the same treatment as vagueness. the defence of LEM rests on the claim that a disjunction may be true even if neither disjunct is true. this is to treat the disjunction non-truth-functionally. This puts us on common ground. for Cook sees 'J == S' as lacking in truth-value. I then accept this as a counter-example to LEM. But on this understanding it appears to me that Cook has not defended LEM for the case in point. Thus he may consistently hold that a disjunction may be true when neither disjunct is true. but not because of indeterminacy of identity. I think. and this undercuts the assertion that indeterminate identity inevitably conflicts with LEM. It is an identity statement with a concealed vagueness in it. So violations of LEM due to indeterminacy of identity have not been addressed. a disjunction made up of parts that lack truth-value for this reason would still lack truth-value. not a vagueness in '='. while also holding that if there were real indeterminacy of identity. if the lack of truth-value depends on vagueness. but this explanation does not apply to the case of indeterminate identity.8 BROOME'S CLUB In a paper objecting to Evans's argument.Non-Conditional Disputations 79 LEM. The conflict was asserted above. And he is consistent in his treatment of the building case. and "sharpenings" of concepts or vague terms is irrelevant to this. However. and indeterminacy of identity remains unrefuted. and so when LEM is applied to a case of indeterminate identity. The explanation. . but a vagueness in 'part of'. He explains how apparent counterexamples to LEM due to vague terms are not real counterexamples. but that explanation makes it clear that the conflict arises when the disjunction 'a == b or not a = b' is treated truth-functionally. at least for the language I have adopted here: if a is indeterminately identical with b then 'a = b or not a == b' really does lack truth-value. 5. John Broome (1984) proposes a plausible example of a case of indeterminate identity. a law which Cook defends. he thinks that this is only an apparent case of indeterminacy of identity.

80 Non-Conditional Di~putations SCENARIO: A club comes into existence and continues for five years. or a minimum of twenty-five. get together and act as a club for an additional twenty-five years. A few years later a group of people. at which point it ceases functioning. years). He asks that we consider the predicates: lasted for at most five years lasted for at least twenty-five years lasted for at most five years or lasted for at least twenty-five years He claims that it is indeterminate whether the first predicate is true of the earlier club. and if the identity were true it would be false of the earlier club). so I quote it in full: [I]f'the earlier club is the later club' is indeterminate in truth value for [the] reason [that the objects are indeterminately identical]. and there is no other longer-Jived entity. and if they were not the same it would be determinately false that the earlier club was revived. The . then it would be determinately true that the earlier club was revived. Noonan (1984) argues that this cannot be a case of indeterminacy of identity. the predicate 'lasted for at most five years or lasted for at least twenty-five years' must be determinately true of the earlier club (for the object determinately denoted by 'the earlier club' has certainly lasted for at least five years on Broome's account. and indeterminate whether the second predicate is true of the earlier club. apart from the latcr club-which has lasted for at least twenty-five years-with which it might be identical. He concludes that he cannot understand how there can be such an object. so its Iifespan must either be a maximum of five. Similarly. and if the identity were false it would be false of it). The reader is invited to fill in the details in such a way that it is indeterminate whether the earlier club was revived when the new meetings began. and a later club. then the predicate 'lasted for at most five years' will be neither determinately true nor determinately false of the object denoted by the term 'the earlier club' (for if the identity were false that predicate would be true of the earlier club. Broome suggests that this is a case in which there is an earlier club. but determinate that the third predicate is true of the earlier club. The argument is subtle.) I am inclined to agree. (If they were the same. On the other hand. the predicate 'lasted for at least twenty-five years' will be neither determinately true nor determinately false of the earlier club (for if the identity were true that predicate would be true of it. including many members of the club previously mentioned. with no fact of the matter as to whether they are the same.

the puzzlement is over examples like that above. I have difficulty with such cases as well. so: (iii) its lifespan must either be a maximum of five. on Broome's account. must be an object which determinately satisfies the predicate 'lasted for at most five years or lasted for at least twenty-five years' but neither determinately satisfies the predicate 'lasted for at most five years' nor determinately satisfies the predicate 'lasted for at least twenty-five years'. Why does Noonan think the disjunctive predication is true? His argument has this structure: The predicate 'lasted for at most five years or lasted for at least twenty-five years' must be determinately true of the earlier club. For: (i) the object determinately denoted by 'the earlier club' has certainly lasted for at least five years. where the description ('at most five or at least twenty-five') prohibits such a borderline. . Part of the trouble is the 'might' in: (ii') there is no entity x such that x is distinct from the later club and x lasts more than five years and such that the earlier club might be x. But I think the disjunctive predication is indeterminate as well: It is indeterminate whether the earlier club lasted for at most five years or at least twenty-five years. for he believes that this occurs naturally with borderline cases such as being either-orange-or-red. T agree that each of these predications are indeterminate: The earlier club lasted for at most five years The earlier club lasted for at least twenty-five years. or a minimum of twenty-five. but (ii) is difficult to interpret. then. But I do not understand how there can be such an object. apart from the later club-which has lasted for at least twenty-five years-with which the earlier club might be identical. But I am not convinced that we have such a case at hand.Non-Conditional Disputations 81 earlier club. years. and: (ii) there is no other longer-lived entity. (Noonan 1984: 119) Noonan's bewilderment (as he goes on to explain) is not with the general case of a disjunctive predicate being true of a thing when neither disjunct is. T agree with (i).

(!-. This does entail: 12 I am indebted here to an anonymous referee for OUP. the clause is much too strong. 111is is indeterminate because each of its first two conjuncts are indeterminate (and its last conjunct is true). one might interpret 'apart from the later club' to mean 'determinate!y distinct from the later club.(e == x)].!-. which is thus indeterminate or false. Since this is indeterminate.!-. that is: (ii") -.!-. then the conclusion (iii) does not follow.e == x)].82 Non-Conditional Disputations If the 'might' is read modally. but it begs the question. Probably what is meant isY (ii') there is no entity x such that x is distinct from the later club and x lasts more than five years and such that it is not determinately true that the earlier club is not x.(x == I) & -. .3x[(!-. Not because it is false.(e == I) & (e lasts more than five years) & -.(e == e).(x == I) & (x lasts more than five years) & -. (x lasts more than five years) & But this is not true in the example envisaged. Of course.3x[-.x == I) & (x lasts more than five years) & -. If it is merely rhetorical. it is equivalent to the claim that the earlier club is identical with the later club.(e == x)] is either indeterminate or true. For there is an instance of (ii")the earlier club itself-which makes (ii') indeterminate in truthvalue. its existential generalization: 3x[-. But this is the negation of (ii") . as in: (ii') there is no entity x such that x is distinct from the later club and x lasts more than five years and such that the earlier club is x. The instance is: Indeterminate: -. but because it is indeterminate. So the argument is not successful. This yields the conclusion. Then (ii') has the form: -.

) It is worth noting that if this kind of reasoning were to be persuasive.(e lasts more than five years).(e lasts a minimum of twenty-five years) v -. yielding: 83 -.(!-. not just for the indeterminacy of identity. Suppose we ignore 'the later club' and simply ask whether the original club lasted at most five years.(!-. it would be determinately false that the original club was revived.Non-Conditional Disputations -.e = I) v -. the question directly posed was whether or not the original club was revived.(e lasts more than five years) v (!-. Notice that identity was not mentioned in the original scenario.e = e). I see no other way to interpret the point that makes it strong enough to yield the conclusion without presupposing what is at issue. then the predicate 'lasted for at most five years' will be neither determinately true nor determinately false of the object denoted by the term 'the original club' (for if that predicate were determinately true of the original club. (I return to a speculative interpretation later in this section. or either at most five or at least twenty-five years? It is easy to simulate a puzzle like Noonan's for the original club: [I]f it is indeterminate whether the original club was revived.(e lasts more than five years). Instead. or at least twenty-five years. one gets: -.e = I) v -.!-.(e lasts more than five years).( e lasts more than five years) which is what we need to infer: (e lasts a minimum of twenty-five years) v -. For a conundrum like that which Noonan presents occurs without identity being involved at all. the predicate 'lasted for at least twenty-five years' will be neither determinately true nor determinately false of the . it would be a problem for indeterminacy in general. Similarly. But the additional '!' in the first conjunct prevents us from inferring: (e = l) v -. The last disjunct drops off. it would be determinately true that there was a revival). which is not the desired result. and jf it were determinately false of the original club.

but I am bothered. on Broome's account. But the central parenthesis here clearly involves an illegitimate appeal to excluded middle.84 Non-Conditional Disputations original club (for if that predicate were determinately true of the original club. But how can there be such an object? If this reasoning is good. years). no matter how unlikely.B. it shows there is no such thing as indeterminacy.l Super-resolutional readings I could rest here. But the Holmes principle yields the disjunction above only if "no answer" is not one of the alternatives." And in the case of the disjunction. the predicate 'lasted for at most five years or lasted for at least twenty-five years' must be determinately true of the original club (for the object determinately denoted by 'the original club' has certainly lasted for at least five years on Broome's account. The original club. it is ruled out that the earlier club lasted more than five but less than twenty-five years. or a minimum of twenty-five. must be an object which determinately satisfies the predicate 'lasted for at most five years or lasted for at least twenty-five years' but neither determinately satisfies the predicate 'lasted for at most five years' nor determinately satisfies the predicate 'lasted for at least twenty-five years'. so its lifespan must either be a maximum of five. This is the part of the argument that Noonan fills in with the remarks about identity examined above. whatever remains must be true. it would be determinately true that the original club was revived. I find I am tempted (somewhat) to agree that the disjunction he cites is true: The earlier club lasted for at most five years or at least twenty-five years. On the other hand. Why is this? When I try to examine my own instincts I find it is partly my inclination to respond to disjunctive questions with the Sherlock Holmes principle: "When all the alternatives have been ruled out. If it is one of the options. then. 5. the principle is misapplied. it would be determinately false that there was a revival). . for in spite of the inconclusiveness of Noonan's reasoning. the alternative is ruled out. and there is no club with a life-span between five and twenty-five years. and if it were determinately false of the original club.

but in every resolution. either the earlier club lasted for at most five years or at least twenty-five years. for example. years. and there is no other longer-lived entity. or a conjunction containing it as a conjunct. then we can understand how a disjunction can be true when its parts are indeterminate. So there are ways in which the disjunction may be true when its parts lack truth-value. so its lifespan must either be a maximum of five. on a disjunction containing the claim as a disjunct. The advantage of this is supposed to be that we have clear instincts that the disjunction is true (or that the conjunction is false). In each case. years. the sentence is true in every resolution of the world.Non-Conditional Disputations 85 Another closely related theme that moves me towards accepting the disjunction is to read it: No matter how things turn out.. . thus refuting the claim that . It is indeed true that in any determinate extension of this world. 5. The parts are indeterminate because there is no one way they turn out in every resolution. the lifespan of the earlier club must be either a maximum of five. what is at issue is whether a certain claim can lack truthvalue because of what the world is like. The strategy is to focus not on this claim. but on a complex sentence having that claim as a part. Something like this is a possible construal of the passage from Noonan quoted above: [the earlier club1has certainly lasted for at least five years . but their existence does not tell against the indeterminate identity at all. or a minimum of twenty-five. one of them turns out true. apart from the later club-which has lasted for at least twenty-five years-with which it might be identical.9 WHY CONSIDERING CONJUNCTIONS OR DISJUNCTIONS IS INEVITABLY INCONCLUSIVE There is a theme to the last few arguments we have considered. But to read the sentence in this way is to read it super-resolutionally. or a minimum of twenty-five.. But if this is how we read it when we endorse the disjunction. and that is why the disjunction is super-resolutionally true.

that seems to go nowhere in attacking indeterminacy. The second is that when the disjunction is read so as to make its truth compelling. and so one may not reason from the truth of the whole to the truth-values of its parts. . this appears to force a reading in which the disjunction is true in spite of the possible lack of truth-value of its parts. I don't know any way around this. This strategy has two weaknesses. The first is that the data to the effect that the disjunction is true are not compelling. If there is a reading on which the disjunction is true.86 Non-Conditional Disputations its disjuncts lack truth-value.

so it is true. I take for granted that any feasible rendition of the conditional will agree with the classical material conditional when the parts of the conditional have truth-values. it is awkward. and then I discuss constraints on other options for treating conditionals. I begin by saying what I mean by 'if . and considering conditionals merely brings in irrelevant complications. then .1 THE CONDITIONAL AND THE BICONDITIONAL For reasons given earlier I ignore supervaluational readings and I concentrate entirely on conditionals for which the truth-value status of the whole is determined by the truth-value statuses of the parts.. and it is incumbent on me to provide a coherent response to their objections. This is followed by some ways in which the logic of conditionals and biconditionals has been used in attempts to refute indeterminate identity. We then look at various ways in which Leibniz's Law can be formulated as a conditional. Likewise for a conditional with a true consequent and .6 Conditional Disputations Some people have thought that indeterminacy of identity can be refuted by considerations gleaned from the logic of conditionalsthe logic of statements of the form 'If A then B'.1 are not presupposed by later chapters.) 6. But others have thought differently. I think that this cannot be right.. indeterminate identity can be stated and assessed without the use of conditionals.l also assume that a conditional with a false antecedent and a truth-valueless consequent is just as true as one with a false antecedent and a false consequent.. . although it is possible to avoid the use of conditionals. Besides. (The sections after §6.'.

. I call this "sustaining if-then": the truth-value status of a conditional is determined by how far the consequent drops below the antecedent in truth-value status. if there TABLE 6. the conditional is true. then . .1. there is a much stronger inclination to treat 'If S then S' as automatically true than to treat 'S v -.. which makes the remaining empty cells be gaps (Table 6. I will assume this in what follows.. counting T as highest and F as lowest.' with two gaps be T. This conditional is well defined. Doing this requires that the line in the truth-table for 'if . So any truth-table will be got by filling in the cases left open in Table 6..' instead of'not ..S' as automatically true. If there is no drop at all.. .<jl V \jf. I choose the most popular version.88 Conditional Disputations an antecedent without truth-value. So the question is how to fill in the other two empty cells in the table.. If <p then T ?~ \jf <p \jf <p :::> \jf T T T T F T F T ?~ ?~ T T T T T F T F T F F F F T F F F F T T T T F T T T F . then our previous analysis of disjunction yields a conditional whose truth-table is got by making all the open cases above be truth-value gaps (Table 6.2). or . At the very least...2.'.3). . <P \jf TABLE 6. then .l. but it does not seem to reflect what people have in mind when they choose to say 'if . the socalled "Lukasiewicz" interpretation. If we use the classical definition of the material conditional in terms of disjunction and negation: <jl ::J \jf =df -.

such as modus ponens. The "if-true" use of conditionals: Woodruff (1969) points out that there is a use of 'if . and otherwise the conditional lacks truth-value. (jJ \jI (jJ~\jI TABLE 6. their claim should be symbolized as '!cj> ~ \jf'. and contraposition. This use is easily expressed with the notation at hand: if someone says 'if cj> then \jf' intending the "iftrue" meaning. The "if-true" use of conditionals is important to capture because we need to confront the fact that the literal reading of the TABLE 6.. The resulting conditional validates most of the laws one naturally expects from a conditional.4. if 'cj> <=> 'V' is defined as '( cj> ~ 'V)&('V ~ cj>)'. modus tal/ens. . but I think it is right.4). This is a hypothesis about people's behaviour in interpreting conditionals.. and it is important to take into account when assessing arguments that people actually give in a non-classical setting... then . hypothetical syllogism. then one obtains a biconditional that is true when cj> and \jf have the same truth-value status. The conditional just defined also yields a natural biconditional. false when they differ in truth-value. and a substantial justification of it would take more discussion than I can devote to it here. then .. That is. in which what one means by saying 'if cj> then \jf' is 'if cj> is true then \jf'.'. . and otherwise indeterminate (Table 6.3.. which he calls the "if-true" use. (jJ \jI (jJ~\jf T T T T T F T F T T T T T T T F T F T F F F F T T T T F F F F T F T F F . I regard this as a special use of 'if .'.Conditional Disputations 89 is a drop all the way from T to F it is false.

it lacks truth-value when $ does. line 3 will read '!<I> ~ W.90 Conditional Disputations Lukasiewicz conditional.~ 3. '$ => W would be a truth of our logic.. does not validate conditional proof. We do not have: NO: but we do have: YES: !$ => \jf I If this is done above. . $ =>!$ 1. and no harm is done. and then concluding correct/y(!) that they have shown 'if $ is true. The difference is that conditional proof is not valid for the literal reading of '=>'.1j> Hypothesis for CP From 1 By CP from 1-2 NO! If this proof were good. and then deriving 'If from $. Suppose that $ lacks truth-value. This is easy to see from the conventions already adopted.1 In a classical setting. This is important. since people often take conditional proof techniques for granted in their reasoning. which is a truth of logic. then \jf. but it is not. But in a non-classical setting one needs to be attuned to the difference. Woodruffs suggestion is that when people instinctively use conditional proof in a non-classical setting they are assuming that $ is true. but it is valid for the "if-true" reading of '=>'. '$ => 'If'. So the '=>' connective does not validate classical conditional proof. this is no different from concluding 'if $ then \jf'. and consider the following threeline argument using conditional proof: 2.

and whenever its consequent is true.G' would then be true. that is. There is no such conditional definable in terms of truth-value status that sanctions these three rules: 2 (i) modus ponens (ii) modus tollens (iii) conditional proof The proof is simple. is true whenever its antecedent is false. We can do no better than the options already discussed. By contraposition. (ii) contraposi· tion. and that makes '~ => ~' always true. Conditional proof would allow us to prove 'if A then !A'. Here is the argument. Since the truth-value status of the parts determine that of the whole. this would yield truth for any conditional 'if G then H' where G lacks truth·value and 11 is false. one cannot propose an alternative treatment of conditionals without also failing to achieve all desirabilities. But then modus ponens would lead us from truths ('If ---. so that would be true. so any sentence of this form would be true. since it validates both modus ponens and modus tollens. 1 am inclined to see the Lukasiewicz conditional as the best of all.G') that lacks truth-value. . that is not a reasonable goal. Likewise. So you cannot have everything you want. If we had classical conditional proof. (iii) conditional proof. 'if ---.Conditional Disputations 6. Given these facts. no such conditional satisfies all of (i) modus ponens. Then modus toUens would lead from the truths 'If A then B' and 'not B' to the truth-valueless sentence 'not N.G lacks truth·value. Nobody may fault the conditional chosen in the last section merely because it does not do everything one might want.H') to a conclusion ('---. any sentence with a truth-valueless antecedent and a false consequent would be true. Suppose that you want a conditional that. like the material conditional. 'if <I> then !<i>' would then be a true sentence with its antecedent lacking truth-value and its consequent false.2 CONSTRAINTS ON ALTERNATIVES 91 Why settle for a conditional that does not obey conditional proof without giving it the "if-true" reading? Why not use a conditional that satisfies all the laws we expect from a well-behaved conditional? The answer is that no such conditional exists.H then ---.H is true and ---.G' and '---. as we showed above. as proofs that correctly conclude a sentence of the form 'if !<I> then 'If'. 2 Likewise. when ---. and we can explain good versions of conditional proof as resulting from if-true readings of it. we could prove 'if ~ then !~'. Let 'If A then B' be such a true sentence. If A lacks truth-value. When <I> lacks truth-value.H then ---.

the last line of the proof does not follow by modus tollens. -'('la=b iff'lb=b) 4. nor does it follow by any acceptable inference pattern: .a = b. con tradicting (1) Whether this is a good argument or not depends on how the connectives are symbolized. it is easy to show that the following simpler version of Leibniz's Law must not be accepted: NO: a = b ==> (<\la {::::} <\lb). If the good version of Leibniz's Law is used instead.a = b. Then use a simple variant of the Evans argument to refute indeterminate identity: 1. Using the Lukasiewicz conditional and biconditional the steps are indeed valid: 1.92 Conditional Disputations 6. First. 4. I have accepted a version of this law as a principle of inference. -''l(b = b) -'('la=b{::::}'lb=b) a = b ==> (<\la {::::} <\lb) -. -.3 LEIBNIZ'S LAW AS A CONDITIONAL It has been suggested that indeterminate identity may be refuted as follows.2 by truth-functional logic [LL?] using ==> From 3.2 by truth-functional logic [LL?] From 3. 'la = b 2. and this automatically validates the following "if-true" version of Leibniz's Law: Leibniz's Law: !a = b ==> (<\la {::::} <\lb). The source of the problem is the formulation of Leibniz's Law. Since the '==>' conditional satisfies contraposition.4 by modus tollens So there appears to be a problem for indeterminate identity.4 by modus tollens. 'la = b 2. and it must be rejected. -''l(b=b) 3. 5. 3. If a = b then ('la=biff'lb=b) 5. Hypothesis for refutation Logical truth From 1. [Good Version] But I have rejected the contrapositive version of the law. Hypothesis for refutation Logical truth From 1. [Bad Version] This is the version of Leibniz's Law used in the above proof. formulate Leibniz's Law in conditional form: [LL?] If s = t then (<\ls iff <\ll).

and reject others. With that version of the Law. I see no difficulty in rejecting the other version above. We clearly expect conditionals to be minimally well behaved. we disagree about indeterminate identity.2 by truth-functional logic Leibniz's Law as a conditional: good version From 3. The answer must be that there is indeed a conditional formulation of Leibniz's Law. I have occasionally been confronted with the following sort of argument: Granted. then I explicitly disagree.a = b Instead. the above proof is fallacious. But then the proof at the beginning of this section is valid.4 by modus to liens NO! 5'.'V(b = b) 3. -. only the inevitable awkwardness that comes with using conditionals in a framework that takes indeterminacy seriously. -. In conversation.('Va=b~'Vb=b) 4. -. but this is not a new issue. it is the same old question about the validity of the contra positive of Leibniz's Law couched in terms of . and we don't seem to be able to find grounds to resolve our disagreement directly. Any defender of indeterminate identity is thus forced to accept certain conditional formulations of Leibniz's Law. So we should turn instead to an independent discussion of conditionals. using the "if-true" reading of the conditional. and it disproves the existence of indeterminate identity.!a = b. one can infer: 93 Hypothesis for refutation Logical truth From 1. But (5') is compatible with the claim that 'Va = b. If you insist on formulating Leibniz's Law using that same conditional without the "if-true" reading. as discussed.Conditional Disputations 1. -. A good version is: Leibniz's Law: !a=b:::::}(ljIa~ljIb). 'Va = b 2. about which we have intuitions having nothing to do with identity. and our only assumption about identity is that we expect Leibniz's Law to have a conditional formulation. and it achieves no refutation of it. !a = b :::::} (ljIa ~ IjIb) 5.

validate modus tollens. And if you wish to use some other conditional with which to formulate the Law. No new issues arise. but you feel that there must be a conditional form of Leibniz's Law which requires no special "if-true" reading.94 Conditional Disputations how to formulate Leibniz's Law as a conditional. and it itself requires no special reading. This new connective validates both modus ponens and conditional proof.5). since it is equivalent to the "if-true" reading of the version of Leibniz's Law given above. That is easily accomplished. suppose that you are inclined to agree with me on issues concerning indeterminate identity. so it does not save the flawed proof above. we merely repackage old ones. and whether it can accommodate the invalidity of the contrapositive of Leibniz's Law and also satisfy modus tollens (which is needed in the proof above). Define a new connective with this truth-table (Table 6. then we must see what its logic is. or contraposition. (It does not. you can't have everything you want in a conditional. Again. <\> \If T <\> I~ \If T T T T F T F T T T T T T F F F F T F . Then formulate Leibniz's Law as follows: Leibniz's Law This is completely adequate from my point of view.) I assume (when this does not beg a question) that the principal options for interpreting conditionals are Lukasiewicz's "sustaining" TABLE 6. however. For example.5.

I will insist that the awkwardness of anyone of the existing non-classical conditionals not be used as an objection to some unrelated claim. for no such conditional exists if there are truth-value gaps. He assumes that the Law should take this form: If a = b then (Ea iff Eb) and he argues: The ground of Leibniz's Law is that if a is identical to b then a and bare one object.4 TWO EXAMPLES OF ALTERNATIVE CONDITIONALS I have suggested that discussion of formulations of conditionals does not change anything of substance about indeterminate identity. it only repackages the issues. when we use a conditional. and also that 'a = b' should be true and 'Ea == Eb' undetermined.Conditional Disputations 95 if-then conditional. As a corollary. But nobody can appeal to the "naturally desirable conditional": the one that is truth-status-functional and that validates all natural logical principles. (1984: 9) He then claims that in order to express Leibniz's Law in conditional form. or that same conditional given its "if-true" reading. I will need to be sensitive to that fact. Broome observes that neither of these tables validates modus tollens. If other writers use conditionals differently. the conditional ':::>' in the Law must have one of the two truth-tables (Table 6. it rules out that 'a = b' should be true and 'Ea == Eb' false.7). so that the meaning does not shift during discussion. Broome (1984) discusses Evans's argument. 6. and an object has a property if and only if it has it. Here are two illustrations of this. In other words. and that .6 which makes the conditional be automatically bivalent or Table 6. we must keep straight what meaning is intended by it when a claim is made using it. Most important. and asks what sort of conditional would be needed to express a correct version of Leibniz's Law for use in the proof. This licenses us to say only that If 'a = b' is true then '(Ea == Eb)' is true.

6. the connective that embodies the "if-true" reading of the '~' conditional. But when 'Ea' lacks truth-value.4 so Broome's version of the law would likely be: . it depends on how we read the biconditional in the consequent of Broome's version. But these would produce a version of Leibniz's Law that is much too strong.96 TABLE 6. . 4 One might speculate that Broome just means the biconditional to be the threevalued material biconditional '(jJ '" 'V' defined as '( (jJ&'V) v (~<l>&~'V)'. Since 'a = a' is logically true. Almost the same. he is disputing a view according to which a statement might be indeterminate while having a truth-value. and his own argument is entirely classical.7. It is natural to assume that this is the conjunction of two conditionals. So the view Wiggins disputes does not require formulation in a non-bivalent language. this (material) biconditional also lacks truthvalue. Garrett (l988) adds that if this is correct then an argument from Wiggins (1986) against indeterminate identity also fails. But Wiggins means something else by indeterminate identity (personal communication). $ T T T \jI if $ then T F F T T T T T T \jI I» T T T \jI if I» then T F T T \jI T F T F T F T F T T T F F F T F F F F T F T Evans's argument formulated with such a version of Leibniz's Law fails. Conditional Disputations TABLE 6. or equivalently as '(~(jJ v 'V)&(~'V v (jJ)'. Thus his proposal is almost the same as the form I have suggested for a correct version of Leibniz's Law. 3 Broome's second table is the table for the symbol 'c::>' introduced temporarily above. this requires that 'Ea'" Ea' also be logically true. In the trivial case it would yield: a = a c:> (Ea'" Ea).

it can be shown that these two versions are logically equivalent. Begin with: 1..Eb) & (!!'Eh => !'Ea) 10.Ea <> !. l<'Ea <rl lEb <rl !.> '1". The following arc all instances of BROOME: 2. 6 e. which is 'ip 0. This differs from my own suggested reading. (!.Ea => !<. la = b => (Ea <rl Eb) QED. => '1" (urns these into: 8.Eb) & (!'Eh~) !'Ea) 7.Eb <> !'Ea)] 4.Ea => !. a =b Hypothesis for conditional proof..Eh) & (!.Ea 16. A similar proof can also be given using Broome's first truth-table for 'if then'. since 'I» => '1" entails '!I» => '1".g. in spite of the difference in conditionals in the consequent.> !Eb) & (!Eb <> lEa)] 3..Eb <> !<'Ea) Changing 'I» <> 'V' to '!q.Eb) & (!. Ea <rl Eb Then conditional proof (in its good form) from 1-17 yields: 18.Ea <> !. which can be expressed as: TP a = b c:> [(Ea =:} Eb) & (Eb =:} Ea)].6 but there is no difference in content between the views being expressed.Conditional Disputations 97 BROOME a = b c:> [(Ea c:> Eb) & (Eb c:> Ea)]. (!Ea => !Eb) & (!Eb => !Ea) 12. in the preceding footnote it was trivial to show that TP entails BROOME.'> IvEb) & (!<. a = b <> [(l.Ea => l<'Eb) & (!"Eb => l"Ea) 111ese can now be reexpressed as biconditionals: 14. (!!<. we give a conditional proof. For the other way around.Eb => !.Eb <rl !"Eb These three together insure that 'Ea' and 'Eb' have exactly the same truth-value status. . lEa 15.Eb) & (!!"Eb => !"Ea) Eliminating redundant repetitions of 'I' yields: 11. (!<. Ea) 13.' I find my own version handier. a = b <:> [(!<'Ea <. (!lEa => !Eb) & (!lEb => !Ea) 9. (!<'Ea c. but the other way round was complicated.. (!!'Ea => !. (Broome's is the first 5 It is easy to show that TP entails BROOME. (IEa~) !Eh) & (!Eh <:> !Ea) 6. This is sufficient for the truth of: 17.:> !<'Eb) & (!<'Eb y !<'Ea») Modus ponens then yields: 5. a = b <:> [(!Ea 0. !. (!. However.

8) displayed over (and. and so he considers ways to reformulate it. such that he reads 'P :::J Q' as if it symbolized 'if P is true then Q is true'. in which he assumes that the conditional is the three-valued material conditional. Does 10hnson disagree with these conditions? He has his own proposal for how to formulate Leibniz's Law as a conditional. the material biconditional is equivalent to the conjunction of two material conditionals). he merely states in the metalanguage his opinion about the conditions that are needed for the truth of the conditional in Leibniz's Law. lohnson agrees that someone who wishes to attack indeterminate identity will be begging the question by formulating Leibniz's Law in this way.x(<I>x)(y) & ---.v(y = z) & ---. the equivalent claim that <p is not indeterminate. and of my separation of Leibniz's Law from its contrapositive.vA. 10hnson says that Broome apparently confuses the object-language symbol for conditionals ':::J'. He begins his discussion by criticizing this formulation of the principle: LL VyVz«(y == z) ~ [A.x(<I>x)(y) == AX(<I>X)(z)]]). After some discussion.) I don't see Broome as confusing object-language and metalanguage at all. the version of Leibniz's Law that he finds uncontroversial is the following. which is just like LL except for the addition of a conditional antecedent making all conjuncts determinate: 7 LLu VyVz([---.~.x(<I>x)(z)]). One serious consequence of this error is that Broome's truth tables involve rejection of modus toliens. rather than resort to such extremes it would be better to admit that incoherence of vague identity for which Evans argues. ~ 7 10hnson uses ''''IJ>' to mean that IJ> is determinate. I have replaced this in his formula by . (1989: 105-6) (Johnson also objects (1989: 106-7) to my own earlier objections to Evans's use of abstraction.98 Conditional Disputations conditional formulation of Leibniz's Law in the literature that I know of that is congenial to indeterminacy of identity. where '<I> ~ \)I' is equivalent to '---.vA. with a metalinguistic symbol invoking the truth of the antecedent and consequent.<P'.x(<I>x)(z)] [(y = z) ~ [A. presumably. with the truth-table (Table 6.) 10hnson (1989) objects strongly to Broome's formulation.<1> v \)I'.. .x(<I>x)(y) == A.

and one can easily see how this equivalence pattern relates BROOME to LLu above. he himself is forced to abandon a conditional formulation of Leibniz's Law in favour of a conditionalized conditional version of that law. 8 So 10hnson shows how to produce a version of Leibniz's Law that does not disagree with Broome (or myself) on any matter of substance. it is equally objectionable that 10hnson's conditional does not satisfy conditional ~ That is. then . Does he also avoid what he calls Broome's "resorting to extremes"? Only by looking the other way. .' is. <P 99 IjI $::lljl T T T T T F T F T F F F F T T T T F It is a straightforward matter to determine (by inspection of truthtables) that LLu is logically equivalent to the version of Leibniz's Law proposed by Broome.. see Ch. formulable. this is easy to establish if we ignore Johnson's use of abstracts.8.. If we do not ignore his use of abstracts. Note also that if one objects to Broome's conditional because it fails to satisfy modus tallens.Conditional Disputations TABLE 6. They become equivalent if one adopts unrestricted abstraction principles. 4. . It is also easily definable using 10hnson's own terminology. lohnson's version of Leibniz's Law is considerably weaker than Broome's version or mine. Broome's '<\> <::> '1" (which is equivalent to '!<\> ~ '1") is logically equivalent to 10hnson's '-. after all. An acceptable version of the law as a simple conditional will require some conditional very much like Broome's.v<\> ::J (<\>::J'1')'. The connective that Broome uses for 'if . as Broome has clearly shown. And although lohnson avoids Broome's "extreme" conditional..

We have here a deadlock regarding how best to read conditionals. 9 The fact that these very different approaches to how to formulate Leibniz's Law as a conditional by myself. so I simplify by talking as if sentences are at issue. In this section I discuss a recent version using biconditionals that has gained some currency. to disprove the existence of truth-value gaps due to semantic indeterminacy. But neither of these is crucial to my discussion.S WILLIAMSON'S DEFENCE OF THE TARSKI BICONDlTIONALS There are a number of a priori proofs in the literature to establish that there cannot be a failure of bivalence for meaningful declarative sentences. and I stick to English examples. provided only that the utterance in question is a declarative one that succeeds in saying something meaningful. The argument is put as a reductio ad absurdum of any specific claim to the effect that a meaningful declarative utterance is neither true nor false. since the argument attempts to disprove the existence of truth-value gaps arising from any source. and indeterminacy of identity-in spite of differences of instinct about how to treat conditionals. This is a bold move. it attempts. ones that are unclear in scope ('Swans are white'). . in spite of Johnson's objections to distinguishing these from one another. and he is careful to allow that the account will work for utterances of sentences that are not in English. Broome.2). because context can change the semantics of a sentence. The argument is from Williamson (1994: §7. 6.100 Conditional Disputations proof. Williamson is careful to relate his argument to utterances rather than sentences. for example. but in the end no difference in the philosophical position being formulated. and ethical statements ('The 9 Notice also that Johnson's formulation of Leibniz's Law as LLu validates the inference rule that I have called Leibniz's Law and fails to validate the inference law that I have called the contrapositive of Leibniz's Law. not just those arising from indeterminacy in the world. I also avoid marginally meaningful utterances ('The number 2 is green'). and 10hnson turn out to be logically equivalent suggests to me that there is a consensus on the important topic-indeterminacy.

since these raise special problems of the sort I am not focusing on. thus refuting the initial hypothesis. There are options for these as well. u is true iff P 2b. But the reasoning is apparently compelling. and what one means by the biconditional 'if and only if. and then two inferences culminating in a contradiction. g. a quotation mark name. and it is certainly incumbent on me to explain where and why the reasoning goes wrong. I have . but indeterminacy of any meaningful declarative utterance. It refutes not only indeterminacy of identity. The argument is a short one: Not: u is true or u is false 2a. So there is no avoiding a survey of the options. so I will focus on the earlier lines. in which case 'u' would be: 'I am my body'. but I don't think that Williamson's argument turns on using these differently from the way I do. The argument itself is expressed independently of any particular choice for 'P' and 'u'. This reasoning is too good to be true. For example 'P' might be: I am my body. This is not a simple task. The two principal questions that need to be addressed are what one means by the predicates 'true' and 'false'. The inference from (3) to (4) is acceptable from my point of view.) The principal question about the use of 'true' and 'false' is whether they are themselves intended to be used bivalently. 'and'. e. This is too much to expect. Not P and not not P Hypothesis for refutation Semantic fact Semantic fact Substituting in (1) using (2a) and (2b) DeMorgan's Law from (3) The argument consists of a hypothesis for refutation.Conditional Disputations 101 death penalty is immoral'). u is false iff not P 3. 4. (I take for granted that we agree on the meanings of 'not'. two semantic facts. and 'or'. Using Williamson's notation. I suppose that 'P' abbreviates a sentence of English and that 'u' is a name of that sentence. thus demolishing decades of work in the logic and semantics of sentences without truth-value. since indeterminacy allows for a number of options regarding what one means and what one says. Not: P or not P l.

is the biconditional that I have adopted above. Whether I am willing to endorse (2a) and (2b) depends on how the biconditional is read. No truth-value No truth-value . There is also another use. but they will disagree about his claim that (1) is the right way to formulate an assertion of lack of bivalence. But from here on I ignore this option. So if a sentence lacks truth-value. One of these. Call this the "redundancy" use. on the bivalent use of 'true' and 'false'. of course. but (1) is certainly not the way to do it. if a sentence lacks truth-value. as I use 'true' (and it is also determinately not false). 1 focus only on the most relevant ones. Call this the "determinate" use of 'true' and 'false'. false if they disagree in truth-value (if one is true and the other false) and otherwise truthvalueless. This is because they would be Jacking in truth-value if P were itself lacking in truthvalue. For the claim that u is true would be false in that case. then so does the claim that it is true or false. That then puts the burden on (2a) and (2b). This is not my usage here. explored by several writers on the semantic paradoxes. II is false <=> not P. and admit on that view that (1) does correctly formulate a claim of indeterminacy. It is important to realize that this option exists.102 Conditional Disputations implicitly adopted a bivalent use for them. I have discussed elsewhere (in Parsons 1984) how things should go on this option. On this interpretation I would not assert either (2a) or (2b) unless I were sure that P had a truth-value. Let us settle. then. On this use. starting with Martin and Woodruff (1975). So these would be biconditionals without truth-value: 2a. I interpret '<jl <=> \jf' as being true if <jl and \jf have exactly the same truth-value status. and to distinguish this alternative approach from my own in this book. and it would be too much of a digression to recapitulate it here. in the sense that if a sentence is lacking in truth-value it is determinately not true. the claim that it is true is itself indeterminate. and also the claim that u is false. u is true <=> P 2b. There are a number of options here. Exactly how an assertion of lack of bivalence is to be stated on this view is a matter of some debate. and Kripke (1975). but it is an important optionj so let me deal with it briefly. partly to be aware that there are other defences of lack of bivalence than the one I will be giving. The important thing to say about it is that holders of such a view will agree with Williamson in rejecting (1).

the only question is whether they say enough. For the fact that if 'u is true' is true. Machina (1976). So do I. They are also the natural conditional renditions of the following "Tarski biconditional inference schemes". . But these are just the old and venerable Tarski biconditionals. which are also correct on anybody's view: VALID: s IS' is true '5' is true VALID: not 5 '5' is false 5 IS' is false not 5 Williamson (1994: 300) notes with approval that Evans and McDowell (1976). and this should not be swept under the rug. in the restricted sense that I regard the inferences above as valid. However. they are true according to the theory I endorse. and Peacocke (1981) all endorse "disquotational schemas for truth".Conditional Disputations 103 I would thus view Williamson's argument as begging the question by assuming these. and (2a) and (2b) in forms that I endorse. [u is true r:::> P] & [P r:::> u is true] 2b. does not mean that 'u is true' 10 In the notation originally introduced in Ch. These are correct on anybody's view. If we read (2a) and (2b) as indicated. and what would I use in their place? A natural option would be the "iff-true" reading of the biconditional. Certainly they seem to be trying to capture something that is right. We thus have (1) as a denial of bivalence. [u is false r:::> not P] & [not Pr:::> u is false]. with the versions of (2a) and (2b) just given. unquestioned within a two-valued framework. so that we get JO 2a. This is a parallel to the "if-true" reading of the conditional. discussed earlier in this chapter. 2. We construe the biconditionals in (2a) and (2b) as conjunctions of the "iftrue" conditionals. these are: 2a. so is' P'. and the following conditionals as true without exception: IS' is true <:>5 5 r:::> '5' is true. [!(u is true) => P] & [JP => u is true] 2b. [!(u is false) => not PJ & [!(not P) => u is false]. (3) no longer follows. What is it about them that is right. and vice versa.

Williamson. And the assumption that' P' has a truth-value is what is at issue. He adds: It might be replied that if u says that P and is neither true nor false. It gives no hint. of course. And there is no doubt that the wording on lines (2a) and (2b): 2a. 3. But on these readings. u is false iff not P Semantic fact Semantic fact can be interpreted so as to be endorsable by a person who believes in the possibility of non-bivalence. and on some of these the conclusion does follow. the substitution that leads to (3) is not verified. line (3) does not follow. and so he does not rest content with merely giving the proof. there is no doubt that the wording on line (1) of Williamson's proof: 1. A parallel reply might be made to (2b) and (F).104 Conditional Disputations and 'P' have the same truth-value. Not: P or not P Substituting in (1) using (2a) and (2b). other than by TW failing to be thin. then u is false if and only if not P I do indeed hold that it is possible for the two sides of (2a) not to match in semantic value-the left-hand side might be false when . The trouble with this objection is that it does nothing to meet the rationale for (T) and (F). is alert to this sort of reply. 'u is true' is false while' P' is neither true nor false. but not in a way that refutes anything that a believer in non-bivalence believes. In summary. then. since those connections obtain even if 'P' does not have a truth-value at all and 'u is true' is false. It is also possible to give each of these lines other readings. or of any way in which u could fail to be false. Not: u is true or u is false Hypothesis for refutation can be interpreted so as to deny bivalence. of any way in which u could fail to be true. (1994: 190) The principles (T) and (F) that Williamson alludes to are: (T) If u says that P. So without the assumption that 'P' has a truth-value. so that the two sides of (2a) do not match in semantic value. then u is true if and only if P (F) If u says that P. when u says that TW is thin. other than by TW failing to be not thin. and neither (2a) nor (T) is true. u is true iff P 2b.

So the rationale for (T) and (F) is not sufficient to validate (T) and (F) if they are read so as to entail bivalence. that is. I suggest this: .Conditional Disputations 105 the right has no truth-value at all-so the quoted comment is apt.: (*) What it takes for 'TW is thin' to be true is just for TW to be thin? A t the very least. Given that an utterance says that TW is thin. If they are given the "iff-true" interpretation. No more and no less is required. Then perhaps one wants a stronger rationale. (1994: 190) I find this rationale completely compelling. is whether it validates (T) or (F) on the interpretation that Williamson needs. then they are completely acceptable but they are consistent with the view that WilIiamson is disputing. what it takes for it to be true is just for TW to be thin. So suppose that we give them the stronger reading. and the issues discussed above about biconditionals arise here too. and that is not sufficient in a language with truth-value gaps for the previous argument to be valid. Notice that (T) and (F) are biconditionals. however. so that they are false for any meaningful utterance that lacks truth-value. Something like that can be provided. that both of these inferences are valid: 'TW is thin' is true TW is thin TW is thin 'TW is thin' is true But they are both valid. Does the rationale the WilIiamson gives validate them on that reading? That depends on how the rationale is understood. How are we to interpret. Williamson states their rationale as follows: The rationale for (T) and (F) is simple. This shifts the onus of discussion to the rationale for (T) and (F). To put the condition for truth or falsity any higher or lower would be to misconceive the nature of truth or falsity. e. we need this to entail that either of these: 'TW is thin' is true TW is thin is inferrable from the other. The question. and what it takes for it to be false is for TW not to be thin.g.

(ii) If the state of affairs determinately fails to hold. This is as strong a rationale as anyone might demand. it is merely to defuse arguments to the effect that it cannot fail. It is constitutive of the meaning of the sentence 'TW is thin' that it match these options precisely. It allows. and cannot itself be the foundation from which one can argue in a neutral way about bivalence. so understood. is itself far from obvious. but that is not to say that the rationale is false. then the rationale. we need to consider whether a hypothesis of lack of truth-value provides a better over-all explanation of the entire set of statements that are prima facie lacking in truth-value. failure of truth-value. though it differs from Williamson's. To decide whether it fails. rather than proscribing. the sentence must be neither true nor false. No doubt a defender of bivalence will claim that option (iii) cannot be realized. or neither.106 Conditional Disputations There are three options regarding the state of affairs of TW's being thin: it might determinately hold. the sentence must be false. . If more than this is required by Williamson's understanding of the rationale. My aim here is not to establish that bivalence fails. it might determinately fail to hold. the sentence must be true. and this cannot be adequately assessed by means of an a priori argument on either side. (iii) If neither option is realized. so that: (i) If the state of affairs determinately holds. it is just to say that it does not rule out lack of bivalence.

When someone says that they do not understand indeterminacy in general. because I can't conceive of its being true. Sometimes 1 have told myself the same thing. or indeterminate identity in particular. the other is merely a report of bewilderment in thinking about the issue. e. and that alternative explanations have not been ." The best 1 can do in response to a challenge of this sort is to point out what things would be like if identity were sometimes indeterminate. they may be making one of two kinds of comment. in the classical identity puzzles. we need to divide the question. so it is hard to discount all the sources of the worry. A claim of the first sort can be put forward from a variety of rhetorical stances.1 "I JUST DON'T UNDERSTAND . The first is a rejection of indeterminacy.g. The first has the effect of a positive assertion rejecting the claim that identity might be indeterminate..7 Understanding Indeterminacy 7. These invite different though related responses. and then point out that things are like this. The second is a mere confession of lack of understanding. To begin. The point of this chapter is to see what can be said on this issue. The most challenging stance is when the intent IS: "I don't believe it. while the second is merely a confession of lack of understanding. " People have told me that they do not understand indeterminate identity. The first is usually expressed as a comment to the effect that one does not understand how identity could possibly be indeterminate. made emphatic by a comment about understanding. And you can't change my mind about this unless you can show me how it is that identity is indeterminate.

7. The third is to have a kind of working knowledge of the theory. The second kind of comment indicates a stance. I am not sure what this amounts to. I feel genuinely bewildered. I discuss "picturing" what it would be like for the theory to be true. 'every'. something like this: "I don't believe it. 'object'. the third is addressed throughout the book.108 Understanding Indeterminacy found to be very satisfying. and this whole work is devoted to addressing it. you need to know the sorts of situations in which the view applies to things." This is a very serious challenge. The second is to address the fact that people need to somehow "conceive of" what it "would be like" for the view to be true. This is true of indeterminacy in general. but I'm willing to consider it. it is a substantive proposal about when this relationship holds. The same is true of my use of the words 'true' and 'false'. or use 'is' in the sense of identity. When I speak of identity. and when I use the sign '='. but not limited to. and of 'property'. 'or'. My definition of identity in Chapter 3 is not meant to confer meaning on the word 'identical'. In the next section I discuss the terminology employed in theorizing. 'refer'. and of indeterminacy of identity in particular. and how it is supposed to apply in those cases. It is what has motivated this whole study. 'and'.2 UNDERSTANDING THE TERMINOLOGY I intend my terminology to be completely normal. Three things are needed for understanding. I mean exactly what everyone else means. feel the bewilderment (it comes and it goes). both empiricist and rationalist theories that require a kind of mental picturing for understanding. The first is to be sure that the terminology being employed is understood. and . I. I address the first two of these in the present chapter. if this is not accomplished. But when J ponder it. and one that festers inside me as well. and what the consequences are of its not applying in those situations. no view will have been presented to be understood or not understood. too. but it has a venerable and influential tradition-including. After that.

but I have no transcendent account to give of it. With regard to indeterminacy of identity. 'determinate' and 'identical' as they occur in: For some object x and some object y. and so there cannot be a meaningful utterance without truth-value. my exposition has failed. in the case of the truth-table definitions of the connectives) I have explicitly addressed which of the available meanings I had in mind. Determinacy is for me a primitive notion. And so I cannot possibly say that a meaningful utterance lacks truth-value. it is not determinate whether x is identical to y. there are two key notions to understand. I have either relied on context to home in on my intended usage. they are rather meant to be stipulative in the sense that I need to declare which of the several normal uses I intend. Of course. Perhaps the familiar meanings are too corrupt to use. along with much of contemporary philosophizing. how it is to function in the theory. then that is an error in detail in my exposition-it is a failed attempt to isolate my meaning from a host of familiar ones. they are used differently by different people.g. These accounts are never meant to be stipulative in the sense that I am inventing some new technical meaning from scratch. And. For example. Of course the displayed statement in which these terms occur . meaning by 'true' what others mean by it. It can be given more content within certain worldviews. and so the challenge must be an assertion that what I have to say is wrong. I think. then there is no gap at all between a declarative utterance expressing something meaningful and that utterance being true or false.Understanding Indeterminacy 109 so on. such as idealist ones. But they cannot claim that what I have to say cannot be understood. But it is clear. if he and others are right. or (e. but not an assertion that what I have to say cannot be understood. if WiIIiamson is right. not a successful attempt to invent a new idea. the most they can claim is that what I have to say cannot be believed. while saying something that might be correct. If the critics are right. If I have not succeeded in doing this in some cases. this is somehow constitutive of the meaning of 'true'. I have already commented on identity. because understanding what I mean entails being unable to understand what it would be like for it to be true. if so. each of these terms is subject to various interpretations. I insist that I do mean what others mean by 'true'.

all of it problematic. not just to its theoretical core. plenty of terminology enters. The question of meaningfulness arises not only when a theory is formulated. But none of it is specially problematic. And I want to emphasize that in all of the applications the terminology I use is normal. but also when it is applied to problematic situations. 'ship'. and this affects the kind of logical inferences we can make. but I am not using either terminology or semantical constructions in special ways. The problem comes when we do not scrutinize them. I use words of our native tongue. So the stage then shifts to a detailed account of language and the world and the semantical relations between them. and the failure of certain inferences we normally take for granted are easy to understand when scrutinized.3 NON-BIVALENT REASONING Probably the biggest impediment to understanding indeterminacy of any kind is that it brings with it lack of truth-value. In particular. Perhaps I am saying things using this terminology that invites special responses in incredulity. Here. Here I may be criticized not for using unexplained terminology. but for using familiar terminology that is not sufficiently scientific or philosophically respectable. But in this respect I think the applications described here are superior to approaches that require unexplained technical terminology. and 'cat'. 7. or that suggest that we need completely new terminology to replace ordinary talk of objects. which is not valid in a non-bivalent setting. and this easily invites claims of incoherence.110 Understanding Indeterminacy involves quantifying into a whether-clause. There is no mystery about this at all. but instead take bivalent reasoning for granted. but without providing the new terminology. such as 'person'. But it is important to . I have belaboured this so much elsewhere that it does not make sense to do so again here. This principally involves what I have called contrapositive reasoning. such as 'gen-identity'. And so the issue is either belief. I raise it now to alert the reader to it and to emphasize that the question of meaningfulness must be addressed with respect to a whole theory including its applications. or difficulty of picturing or conceiving of claims that are meaningfully expressed. This issue will be engaged later (in Chapter 10).

4 PICTURING INDETERMINATE IDENTITY The claims that lead to bewilderment are mostly claims to the effect that it is indeterminate whether certain objects have certain properties and/or that certain objects are identical. represents the object's having the property P. represents an object's lacking P. as in Figure 7. c. you need to experience it yourself to get the effect. I have found that as the instinct to reason contrapositively lessens. a. There is a centuries-long tradition of using Venn-like diagrams to picture objects possessing properties and to picture objects being identical or distinct. An object represented by a point (such as b) located inside a property-region (such as P) is pictured as having the property in question. the picture itself is neutral regarding whether there is an object present in it. There is much more to the account than this. the rest will be easy. I extend these conventions by representing objects by small shaded regions of the figure. I assume that there is at least one property that holds of no objects. Following Parsons and Woodruff (1995). An object-region. they can be adapted for present purposes as well. the feeling of lack of understanding similarly lessens. b. represents its being indeterminate whether the object has P (see Figure 7.2). since (it seems to me) this is really the root cause of most of our feelings of bewilderment when pondering indeterminacy.1. so . In the traditional diagrams. inside a property-region. Many people have found these diagrams conceptually useful. then we know that empty regions represent properties that hold of no objects. But if we are told that there are no objects other than the ones pictured. 7.Understanding Indeterminacy 111 highlight this phenomenon. and one lying partially within and partially without the property-region. the extensions of properties are pictured by certain subregions of a rectangular figure. one completely outside the property-region. P. and objects are represented by points. Its representing region can be anywhere in the picture. If a region has no object-images in it. but if we can "picture" this part of it. a point (such as a) located outside the region represents an object that lacks the property. This is not a point that can be argued.

7.1 b c FIG.112 Understanding Indeterminacy a • FIG.2 . 7.

Figure 7. For example. whereas Figure 7.Understanding Indeterminacy 113 long as it overlaps no object image. Partial overlap means that the world is such that it is not determined whether the represented objects are identical. and represents its being indeterminate whether b is c. I extend this convention by assuming that objects represented by the same regions are represented as being identical. if there are only three objects. I require by edict that no object-representatives lie completely inside others. b FIG.3 represents a property determinately not possessed by any object. and objects represented by partially overlapping regions are represented as being indeterminately identical. there is no difference) and distinct objects are represented by distinct points.4 represents a as identical to b. 7. then the circle labelled 'E' in Figure 7. identical objects are represented by coinciding points (or by the same point. In the classical picturing.6. or combinations of others. objects represented by disjoint regions are represented as being distinct. as pictured in Figure 7.S represents a as distinct from both band c. they do not yet address identity. Notice that a partial overlap of object-representatives does not indicate uncertainty as to whether they are identical (we have no means in a picture alone to indicate uncertainty). These pictures show what things would be like if there is indeterminacy.3 . and represents both a and b as distinct from c.

My worry about this idea is that it seems to me certain that if I am a body. Similarly. it is determinately true that there is exactly one human body presently in my chair.4 c a b FIG. that is. but this would not be determinately true on the view under consideration. but I do not do so in this work. But the idea is not incoherent.S c Each object representative gets a bit of its own space in the diagram. which I do not grant. part of its region is not shared by any other representative. or labelling them with different fonts. I The possibility of making drawings that violate this principle is an artefact of the model. (See Ch. Both symmetry and reflexivity of identity are built in. I disallow such drawings because I think they have no application to worlds such as ours. 7. then I am my body (I am determinately identical with my body). and is worth exploring.) The idea is coherent.7.114 Understanding Indeterminacy a b FrG. which this view must disagree with. TIlis would require that I be determinately a body. just as is the possibility of drawing different representatives in different colours.7. for a diagram can look like Figure 7. yet still not be determinately identical with it. 8 regarding cardinality claims. . but certainly someone could believe this.A referee for OUP has suggested that I might determinately possess every property that my body determinately possesses. l Various principles are now built into the picturing conventions. as is the symmetry and non-reflexivity of indeterminate identity. It is also built in that transitivity of indeterminate identity fails. The transitivity of identity is also built in. and this would be appropriately diagrammable by representing me as a subregion of the region that represents my body.

they are represented as distinct if and only if some property representative includes one of their representatives and excludes the other. and we have decided that that has an extension. If there is no indeterminacy of identity at all. or even if there is nothing indeterminately identical with a. and they are represented as being indeterminately identical if and only if some property representative totally includes or excludes one of their representatives while properly overlapping the other. exactly the same property representatives. 7. and are excluded from. 7. and properly overlap.6 a b FIG. but they cannot be presumed to represent the extension of a predicate that does not stand for a property. unless that extension happens to coincide with that of some property. Then the ontological version of the Leibnizian definition of identity also results from the picturing conventions: a and b are represented as identical if and only if their pictures are included in. .7 c Suppose that we require that every region of the diagram represents the extension of some property.Understanding Indeterminacy 115 FIG. then this predicate is coextensive with any null property. The pictures are ontological in the following sense: the regions in them can be used to represent extensions of properties. Take for example the Evans pseudo-property of being indeterminately identical to a.

116 Understanding Indeterminacy a FIG. I now don't like our proposal much. and you draw a minimal property-region that includes them.8. and otherwise letting them refer to a unique "supplied" object. then what you get is exactly right. and 'being P or indeterminately P' cannot stand for a property in this situation.8 b But suppose that a is in fact indeterminately identical with some object b. then you declare this to be the property represented by the abstract. for' being P or indeterminately P' when 'P' stands for a property.9 depicts a possible state of affairs. 7. which is what would be required for the (pseudo-)property of being indeterminately identical with a to have a picturable extension. while-consistent with this-maximizing the extent to which they stand for the "right"' property. the idea was to guarantee that abstracts always stand for properties in the world. The case just given is typical in the following respect: it is a contingent matter whether there is some property that is coextensive with the predicate 'is indeterminately identical with a'. for example. and I have abandoned it here. so that their picture looks like Figure 7. Then it is obvious that no region can completely include b while completely excluding a. If the abstract satisfies DDiff. . In terms of the diagrams. and we gave a non-standard interpretation of "ontological abstracts" that guaranteed that they always yield properties. not because it is incorrect in any way. This was artificial. It was done in response to criticisms such as Johnson (1989: 106) about restricting abstraction. but because of its artificiality. Figure 7. The same goes. If not. This is like Frege's idea for insuring that definite descriptions always uniquely refer by requiring that they refer to the right object if there is one. For there would need to be a region that includes all of that of b without including any of that of a. it is intuitively wrong. but it gives you some property-extension as an answer. this is picturable only if its extension and antiextension happen to coincide with those of some predicate that stands for a property. they worked like this: you take the objects that determinately satisfy the formula of the abstract. or technically flawed. But its artificiality makes it confusing. which is impossible? 2 In Parsons and Woodruff (1995) we considered what abstracts might correspond to properties. and done for theoretical reasons only.

. then a diagram for the situation should look like Figure 7. 7. then it is metaphysically possible for things in the world to be as the picture represents? In the next section we look at the intended pictures of the situations in which the paradigm identity puzzles occur.5 REPRESENTING THE PARADIGM PUZZLES My BODY: The first paradigm puzzle presented in Chapter 1 centred around whether I am my body.10. If this is a case of worldly indeterminacy. and so there is no possible situation represented by overlapping object images if those images represent a and b. The overlapping shaded circles represent the fact that ME AND J This is not to rule out the fact that it is de re impossible for certain objects to have certain configurations.b then necessarily a 01.Understanding Indeterminacy 117 a FIG.b. But there is a possible situation represented by such images if we are neutral about which objects they represent. For example. many people think that if a 01.9 I take it as a thesis of the theory under investigation that if a situation is picturable. 7.

and I would not be where my body is. though it is indeterminate whether person b did. Some would suggest that there is ambiguity in talk of location of both persons and their bodies. In everyday discourse we talk as if I am where my body is. or being located in my office in the physical object sense of location". (If there literally were such a thing as out-of-body travel. Figure 7. in normal circumstances. if this is so. The diagram takes for granted that being a person is a property. If this were to occur. and vice versa. in the straightforward everyday sense of 'person'.) The diagram is drawn taking this at face value. the region in the diagram represents the disjunctive property: "being located in my office in the person sense of location. Figure 7. THE DISRUPTED PERSON(S): TI1e second paradigm puzzle has a diagram that it isomorphic to the person-body one. Tt also indicates that both are determinately persons.12 indicates that person a entered the room. and so is being a physical object. circumstances would not be normal.10 it is indeterminate whether I am my body. The isomorphism between this diagram and the one for . The diagram represents the fact that I am determinately a person. and the fact that my body is determinately a physical object. whereas it is indeterminate whether my body is. the identity issue would be resolved in the negative.11 indicates that both I and my body are in my office. though it is indeterminate whether person a did.118 Understanding Indeterminacy Physical objects FrG. 7. whereas it is indeterminate whether I am. and that person b left the room.

the ship with new parts 'p'. The diagram indicates that it is indeterminate whether . Then the diagram for the ships is Figure 7.13. and the newly assembled ship 'a'. 7.Understanding Indeterminacy 119 Physical objects Persons Things in my office FIG. THE SHIPS: Call the original ship '0'.11 Things that entered Things that left FIG. 7.12 me and my body indicates the lack of relevance of time to the question of indeterminacy of identity.

but indeterminate whether a did and indeterminate whether p did. Each is such that it is indeterminate whether it is the cat (represented by 'c'). and a pile P2 at t 2 and a pile Pi at t3. Let us idealize and assume that there are merely four. false that p did. and also indeterminate whether it is a cat. All are determinately on the table. and indeterminate whether 0 is a. Each p-cat is determinately distinct from each of the others (since they have determinately different parts). THE CATS: There are probably millions of p-cats. though it is determinate that a is not p. but these are clearly possible.120 Understanding Indeterminacy Ships FIG. Then the diagram is as in Figure 7. suppose that pile PI has been blown around and had some parts changed. . Between tl and t2 . and indeterminate whether 0 did.13 o is p. which makes the picture hard to draw.14. It is determinate that a docked at place A. SOME PILES OF TRASH: None of our paradigm puzzles yield double indeterminacies of identity. It is determinate that 0 left port at such-and-such a time. 7. and pare determinately ships. enough to make you genuinely uncertain about whether P2 is the same pile as PI' Likewise for P3 and P2' It is consistent with this that you might judge that P3 is genuinely . All of 0. a. Consider a pile of trash PI at time t l.

7. (Most of the material in the remaining chapters of this book does not rely on the points discussed here. In the latter case. 7.) . The account also needs to be made more rigorous in order to be able to show that it delivers the intended judgements about the world that are being pictured.Understanding Indeterminacy 121 Things on the table FIG. but only if we can trust them to guide us correctly. The remaining sections of this chapter are devoted to working out the technical points that will achieve these goals. it needs clarification so as to be clear just what is and what is not a picture.15. and more discussion is needed about exactly which aspects of pictures are picturing which aspects of the world. This is also useful in providing a foundation for the account of worldly resolutions introduced earlier.6 PICTURING MADE PRECISE Pictures are useful conceptual guides. The discussion of pictures above has been somewhat informal. the diagram will look like Figure 7. but you also might judge that this is indeterminate as well.14 different from PI.

that is. Call the object-images icons. 7. A simple set-theoretical construction can be given which entails the principles discussed informally in the last section. I call such a picture robust because it embodies a principle of plenitude for properties.15 Although we have spoken of our pictures as being twodimensional. a set I of icons. what is required is that the ° . the two-dimensional geometry is not really essential to the representation. and a set P of pictensions. which are also subsets of 0.122 Understanding Indeterminacy Piles of trash FIG. This is simplest if we begin by defining a robust complete picture of a world. it is complete because it includes images of every object in the world. nor is the size and shape of the regions. and call the regions that represent properties pictensions. The diagrams produced above are like snapshots of some portion of a robust complete picture. What more is required for such a combination to represent the world? All that is required is the right relationship between identity and the possession of properties. which are non-empty subsets of 0. Then a robust complete picture consists of a set of points.

The condition entails certain of the constraints that were posed above in a seemingly ad hoc fashion.p . this is expressible set-theoretically as: [C] For any disjoint sets sand r of icons. It turns out that a single condition will achieve everything we desire.. Suppose we clarify what it means for an icon and a pictension to represent something about the object and the property they represent as follows: DEFINITION: In any picture. it entails that no icon is completely contained inside others.. Given our conventions for representation.Understanding Indeterminacy 123 icons and pictensions be so related that they portray this relationship properly. there is a pictension p such that: any icon i is a subset of p if and only if it is a member of s. For suppose that icon i is completely inside of some other icons. if i is an icon and p is a pictension. It is that for any two disjoint sets of icons there is a pictension such that the first set of icons represents their objects as determinately having the property represented by that pictensi on (and no icon not in the first set does this). the condition yields an analogue of the Leibniz account of identity in terms of properties. it too will be a subset of any set that has all of them as subsets. Condition [C] involving sand r is a kind of principle of plenitude for properties. Then the first clause of the condition is violated. As another example. Take the set of those other icons to be s in the condition above. For example. then the picture (i) represents i's object as determinately having p's property iff i r. just take both sand r above to be the empty set. because no pictension can have exactly those icons as subsets: since i is "hidden" by them. Most important. and the second set of icons represents their objects as being indeterminate whether they have the property represented by p (and no icons not in the second set do this. and any icon i that is not in s properly overlaps p if and only if i is a member of r or i properly overlaps some member ofs. the condition entails that there is a completely empty property (it includes no icons and overlaps no icons).. except for icons that overlap icons in the first set).

. the left-ta-right argument is trivial.124 Understanding Indeterminacy (ii) represents j's object as determinately not having ps property iff i and p are disjoint. by [e) applied to liJ and the empty set. Right-ta-left: Suppose the picture represents j's object as having a property that j's object does not have. or vice versa. there is a pictension that includes i and excludes j. contrary to the supposition. . and (iii) represents its being indeterminate whether i's object is identical to j's object iff neither (i) nor (ii) of the first definition hold. This is proved as follows. For (i). and so i and j are disjoint. We also clarify what it is for two icons to represent their objects as identical or distinct: DEFINITION: In any picture. If i and j were not identical. It then follows from condition [C] that in any picture. if i and j are icons. if i and j are icons. and they thus represent their objects as being distinct. Then. An incomplete picture is simply one that does not represent every object that there is. For right-to-left: suppose that the picture represents i and j as determinately having and determinately not having the same properties. For (ii): Left-ta-right: suppose that i represents its object as being distinct from j's object. For (iii): this follows from (i) and (ii). condition [e) applied to li) and the empty set would yield a pictension that includes i and does not include j. then they are subsets of exactly the same pie tensions. then the picture 4 (i) represents i's object as being identical to ps object iff it represents i and} as determinately having and determinately not having exactly the same properties (ii) represents i's object as being distinct from j's object iff it represents i as determinately having a property that j determinately does not have.1l1en some pictension includes i and excludes j. and so the picture represents i's object as having a property thatj's object does not have. This is not an inherent trait of a picture. (iii) represents its being indeterminate whether i's object has p's property iff i properly overlaps p. then they are disjoint. then the picture (i) represents i's object as being identical to j's object iff i = j (ii) represents i's object as being distinct from j's object iff i and j are disjoint (iii) represents its being indeterminate whether i's object is identical to j's object iff i properly overlaps j.

replacing [e] by this weaker condition: [Weak C] (i) Icons i and j properly overlap if and only if some pictension properly overlaps i and either totally includes or totally excludes j. Let s be the set of icons that represents objects that determinately satisfy <1>. Recall that in §4. Sets sand r are disjoint. and <I> satisfies DDiff. If condition [C] obtains.3x3y[!<jJx & !-. those that are not represented by any icon in s or r). (ii) Icons i and j are disjoint if and only if some pictension totally includes i and totally excludes j. (iii) No icon is totally included in the union of any set of icons not containing it. those depicted by an icon in s). and it is determinately dissatisfied by exactly the objects that dispossess P (namely. since condition [C] provides a stronger principle of plenitude for properties than is required by the theory under discussion in this book. So condition [e] turns the DDiff condition from a test for whether a formula <I> may express a property. or vice versa. there is a pictension p that represents a property P that is determinately possessed by all and only objects that are pictured by the icons in s.<j)y & 'VX = y]. This should be looked at more closely. and is indeterminately possessed by all and only the objects pictured by the icons in r.5 we discussed the conditions under which an abstract 'AX[ <l>x might stand for a property. In fact. into a test for whether <t> does express a property.Understanding Indeterminacy 125 it is a relation between the picture and its intended application. . a non-robust picture is one that portrays fewer properties than is required by condition [C]. Then <l> expresses that property P. For those who want to get along with fewer properties. it requires a fairly neat and natural principle. The condition is that DDiff be satisfied for <1>: r -. in the sense that it is determinately satisfied by exactly the objects that determinately possess P (namely. or vice versa. On the other hand. we can ignore robustness and define a (general) picture as we did above. This gives concrete form to the linkage between what is true in the language and what properties there are in the world. Suppose we have a complete and robust picture of the world. and let r be the set of icons that represents objects that indeterminately satisfy <1>.

and contains pictensions for the rest of the properties. 7. Suppose we say that a picture is precise if each icon consists of exactly one point. . it is equivalent to one that does. and there are pictures that are neither. This corresponds to the idea that the world might be completely determinate. and an identity-determinate picture of the whole world represents a world with no indeterminacy of identity at all. this is because the diagrams are intended to picture a portion of a world.7 PICTURING RESOLUTIONS Suppose we say that a picture is identity-determinate if none of its icons properly overlaps. and there are pictures that are not precise but are identity-determinate. An identity-determinate picture represents a situation with no indeterminacy of identity among the objects represented. A precise picture represents a situation with no indeterminacy of any kind. If a picture does not cover all the options. The remainder of this section characterizes how to turn any picture into one that covers all the options. and it might be indeterminate but contain no indeterminacy of identity. In §5. and how to turn a picture that covers all the options into a "fully refined" picture. and a precise picture of the whole world represents a completely determinate world. There are precise pictures.126 Understanding Indeterminacy This condition entails the Leibniz definition of identity in terms of properties while positing the bare minimum of properties for this to hold. They are to be taken as partial pictures which can be fteshed out to a picture that satisfies at least [Weak q. Most of the figures in this book do not themselves satisfy either of the conditions [C] or [Weak C]. and it might be indeterminate and also contain indeterminacy of identity. it is straightforward to produce from it a picture of a resolution of the world that the original picture represents. and that contains icons for the rest of the objects that there are in the world. which will represent a resolved world.4 we introduced the notion of a resolution of a world. If a picture satisfies a certain condition (if it "covers all the options").

For example. they produce as many overlapping regions as possible.18 pictures do not cover all the c FIG. but if there are three or more. this means that when objects properly overlap. b.17 and 7. there is a non-empty region r (a set of points) that is a subset of each icon in s and is disjoint from every icon not in s. 7.16 covers all the options. this condition is automatically satisfied.Understanding Indeterminacy 127 Say that a picture P covers all the options if and only if this is satisfied: For any set s of icons in P for which any two members of s properly overlap each other. if a. and Figures 7. and c all properly overlap each other.16 . it mayor may not hold. In pictorial terms. then Figure 7. When only two icons overlap.

) If you just want to represent what things are like in the world.18 is meant to illustrate the mutual overlap of the three object icons. each icon is partitioned into cells by its overlap with other icons: if . You can always produce a picture that covers all the options from one that does not by supplying some additional points to use to make the additional regions needed for option-covering. 7. In any such picture.17 options. then it makes no difference whether you produce a picture that covers all the options or not. (The square region in Figure 7. For this is true: If P is a picture.128 Understanding Indeterminacy c FIG. Consider pictures that cover all the options. there is a picture P' that covers all the options and that represents the same objects as having the same properties and standing in the same identity/distinctness relations to one another.

Understanding Indeterminacy

129

c

FIe. 7.18

it overlaps no other icon. it consists of one cell, if it overlaps one other icon it consists of two cells, if it overlaps two other icons it consists of three cells or four, depending on whether the other icons overlap each other. And so on. Given a picture P that covers all the options, an identity-refinement results from P by replacing each icon in P with one of its cells. Notice that icons always become more determinate, not less, in an identity-refinement, and that identities all become completely determinate. Given an identity-refinement of P, a full refinement is produced by replacing each icon in the identity-refinement by the unit set of one of the points in it. In a full refinement, there is no indeterminacy of any kind. Recall the resolutions of the world that were discussed in Chapter 2. The full refinements of a picture correspond exactly to the resolutions of the world represented by the picture. That is,

130

Understanding Indeterminacy

given a world and a picture that represents it, an object has a property in a resolution of that world if and only if its icon is a subset of the pictension representing that property in the corresponding full refinement of the picture of the world. So you can easily see what is true in every resolution of a pictorial situation by expanding the picture (if necessary) to one that covers all the options, and then examining every refinement of that picture. (Refinements are discussed further in the next chapter.)

7.8 PICTURING COMPLEX PROPERTIES Little has been said above about complex properties, because metaphysical issues about indeterminate identity do not appear to depend on what view is taken about them. The following discussion concerns how pictures that explicitly represent properties (by circled regions) implicitly represent complex properties by geometric combinations of those regions. This discussion is not relied on elsewhere in the book. There are two ways in which our pictures are metaphysically neutral, and this should be kept in mind lest some significance be attributed to the pictures that our picturing conventions do not embody. First, suppose we have a picture with two distinct but coextensive pictensions p and q. That is, p and q are distinct sets of points, but they include and exclude exactly the same object icons. Condition [C) requires that there be a "complement" to p and q, a pictension r that includes the icons that p and q exclude and excludes the icons that p and q include. Suppose that there is only one such pictension, r. Then p and q can represent distinct properties not both of which have a unique negation (if property negation obeys the principle that the negation of the negation of a property is that property itself). We can picture a situation in which p has a unique negation, r, but q has no negation: there is no property whose negation is q, even though there is a property (namely, r) which is coextensive with the negation q would have if q had a negation. As a consequence of this, it is possible to have a pictension, q, whose complement (the set of points not in q) is not itself a pictension. In terms of our two-dimensional diagrams, this means

Understanding Indeterminacy

131

that if we have a circle in the diagram representing q, we may not assume that the region outside the circle itself represents the extension of a property-though some region that includes and excludes exactly the same object images as that region does, does represent a property. For the latter reason, it is harmless for most purposes to assume that the region outside a circle represents a property. But the issue is more serious whcn wc consider con.iunctions and disjunctions of properties. Consider the Figure 7.19. If all the objects that exist are already pictured in this diagram, then it is harmless to assume that the region enclosed by the two circles together (the union of the circles) represents the disjunction of p and q. After all, we know by condition [C) that there is some region that encloses (excludes) exactly the same object images as are enclosed (excluded) by the p region or the q region, and the union of p and q does this. But the union of two property regions does not automatically do this. For suppose we have instead Figure 7.20. Here, the object pictured is totally included in the union of the two property regions, but it is not included in either

FIG.

7.19

132

Understanding Indeterminacy

FrG. 7.20

region p or region q or both. So this is a diagram in which the union of two property regions cannot represent the disjunction of those properties. Diagrams of this sort are usually easily avoided. Figure 7.21 represents the same states of affairs, and here we may take the union of region p and region q to represent the disjunction of p and q. Under certain conditions, we can always avoid diagrams of the former sort. Call two pictures equivalent if the icons of one can be mapped one-one to the icons of the other, and the pictensions of the one may be mapped one-one to those of the other, such that an icon in one picture is included in (or excluded from) a pictension of that picture if and only if the correlated icon is included in (excluded from) the correlated pictension in the other picture. Suppose that we have a picture that is extensional, in the sense that there are never two distinct pictensions that include and exclude exactly the same icons. Then there is always an equivalent picture in which the complement of any pictension is itself a pictension, and the union and intersection of two pictensions are themselves pictensions, and in which an icon is included in (excluded from) the intersection of two pictensions if and only if it is included in

7. so it will not be discussed further. . Similar results hold under certain natural conditions on nonextensional pictures. but a further investigation of this phenomenon is of no apparent direct interest to our general metaphysical enterprise.21 (excluded from) each pictension.Understanding Indeterminacy 133 FIG. and an icon is included in (excluded from) the union of two pictensions if and only if it is included in (excluded from) at least one of the pictensions.

8

Counting Objects

Can indeterminately identical objects be counted? Yes, they can. You count them just like you count objects that are determinately distinct from every object they are not determinately identical to. The presence of indeterminacy merely means that questions about how many objects are such-and-such will sometimes have no determinate answers; this is true even if there is indeterminacy, but no indeterminacy of identity at all. In some cases there are determinate answers to such questions, and in some cases there are not. Which cases are which? That is hard to sum up in the abstract. But in most cases we find that we have pretty solid judgements about how many objects are such-and-such, or we judge that there ought not to be an answer. A good theory should agree with these judgements, or it should give us reason to revise them. The theory under consideration reveals different ways to interpret the claims at issue; in every case, one of the ways makes the theory agree with the judgements we are inclined to make.

8.1 TWO SOURCES OF INDETERMINACY IN COUNTING There are two ways in which indeterminacy complicates the process of answering the question 'How many «>'s are there?' One complication arises when some of the <1>'s are indeterminately identical to others, so you don't know whether to count them as one or as more than one. Another complication arises when there are things which are indeterminately «>; thus it is indeterminate whether you should count them at all. These two complications are different from one another, and they need to be addressed differently. Begin with the complication due to indeterminacy of identity.

Counting Objects

135

Things that entered

Things that left

FIG. 8.1

Consider the case of the disrupted person, and imagine that there is one person looking on, as in Figure 8.1. Suppose you are asked how many people there are all told. You point to the observer, and say "one". You point to the person entering, and say "two". You wait a while, then you point to the person leaving, and you ... what? The problem is that you must not count this person if you have already counted them, but you must count him/her if you haven't counted him/her yet. Since it is indeterminate whether the person leaving is the person who entered, it is indeterminate whether this is a person you have already counted. Thus it is indeterminate whether you are to count them now. It follows that the question of how many persons there are all told has no correct answer. We are not in complete ignorance, of course. We know that any answer less than "two" is incorrect, since you haven't counted everyone, and we know that any answer more than "three" is incorrect, since you must have counted someone twice. There are an infinite number of incorrect answers, two indeterminate ones, and no correct ones. This is disconcerting, perhaps, but fairly straightforward. In fact, in this case it seems clear that these are the right things to say: any answer less than two or more than three is wrong, and either "two" or "three" is such that it is indeterminate whether it is correct.

136

Counting Objects

FIG.

8.2

Now consider a different kind of case: indeterminacy due to predication. Suppose there are three people, one of whom is determinately bald and two of whom are such that it is indeterminate whether they are bald (Figure 8.2). You decide to count the people who are bald. You point to the person who is determinately bald, and you say "one", and then what? It appears that in this case any answer less than one or more than three is definitely wrong, but the answers "one", "two", or "three" should all have indeterminate truth-value. Now let us consider a case of predicational indeterminacy combined with indeterminacy of identity. Recall the ships and what they are like; they are pictured in Figure 8.3. Suppose we ask here how many ships left port? The structure of the situation is like that of the bald persons above, and so there apparently should be no determinate answer. But then what about the judgement, which I earlier took to be part of the extended data, that exactly one ship left port? Something like that is correct, but how can it be? I suggest that we can either count ships that left port, or ships that determinately left port. In normal circumstances we take determinacy for granted, and there is no relevant difference in our assertions between '~' and 'detcrminately ~'. Also, since we aim to assert what

Counting Objects

137

Ships

Left port

atB

FIG.

8.3

is (determinately) true, there is no difference between asserting'S' and asserting 'determinately S'. But when a clause is embedded inside an assertion, and when there is a possibility of indeterminacy, then we need to distinguish whether we mean 'determinately <1>' or just '<1>'. This is the case when we assert 'there are n <1>'s' in a situation of indeterminacy. In the case at hand, there is a difference between counting the ships that left port, or the ships that determinately left port. The considerations raised above seem to show that if we ask the first question, assuming that we do not intend the qualifier 'determinately',1 then there is no answer. But if we mean the second, we get an answer. This is because we point at the ship leaving port and say "one"-because it is determinately leaving port-and we do not count the other ships because we know that they are not determinately ships that left port. A natural language assertion may thus be interpreted either austerely, without an assumption of determinacy, or with determinacy qualifying any of

I Using 'entered' without a determinacy modifier is not the same as saying 'determinately or indeterminately entered'. The bare predicate 'entered' is neither true nor false of certain things, and that is why there is a problem in counting; the predicate 'either determinately or indeterminately entered' is either true or false of each thing, and it raises no such problem.

most people will have interpreted me in this way. (The baldness case above is not one of these. The simplest and least controversial are judgements of the form 'there are at least n ~'s'. But some others will not have. I will assume this hereafter. The following sections spell this out. I used ambiguous terminology. but not the unadorned reading. connectives. a cardinality statement will typically agree with what I have called the data when given the adorned reading.138 Counting Objects its otherwise unqualified parts. and identity. The theory is thus most plausible if applied to ordinary judgements with the adorned reading being the default reading in cases where the theory posits indeterminacy of identity.2 ANALYSING CARDINALlTY CLAIMS There are standard techniques for representing finite numerical claims in terms of quantifiers. It follows that when I catalogued the extended data about identity puzzles in Chapter 1. we have these analyses: .) This discussion has taken place at a somewhat informal level. In my experience. It will turn out below that when there is a case of indeterminacy of identity. and agree with the data when interpreted in the way in which I intended. and there are various interpretations to be surveyed. and they should have been troubled by what I was saying there. These are analysable as follows: 11Iere is at least one ~: There are at least two ~'s: There are at least three ~'s: 3x~x 3x3y(x 7= y & ~x & ~y) 3x3y3z(x 7= y & x 7= Z & y 7= Z & ~x & ~y & ~z) What about 'there are at most n ~'s'? The usual formulation of this is roughly that any 17 + 1 ~'s are such that some of them are identical. We need to be precise about whether our language actually works as I have just said. 8. I call the former the "unadorned" reading (the reading unadorned with determinacy) and the latter the adorned reading. and so it is not subject to this default. hoping that the reader would interpret it in the way I intended. What I now hope is that everyone will now consider the ambiguity. That is. I intended to classify as data the fact that exactly one thing is determinately a person who entered.

we seem to want to say this is right.S for a possible account of that instinct. see §8. Were there two or three? Here is where the idiosyncrasies of disjunction creep in.) Were there exactly three? Same response. So any answer that is less than two or more than three should be definitely wrong. and wrong to say there were three. If we use them. 8. then it is wrong to say that there were two or three.Counting Objects There is at most one cj>: There are at most two cj>'s: 139 'v'x'v'y(cj>x & cj>y ~ x = y) 'v'x'v'y'v'z(cj>x & cj>y & cj>x ~ x = y v y = z v x = z) 'There are exactly n cj>'s' is equivalent to the conjunction: 'there are at least n cj>'s and there are at most n cj>'s'. But how many ships were there all told? Not fewer than two. since there are none under consideration except for the one that left port and the two that docked.3 COUNTING ALL THE SHIPS Consider the puzzling ships. In another sense. and not more than three. Our na"ive judgement is that exactly one ship left port and exactly two ships docked. if it is wrong to say there were two. Combining this with the above analyses yields: There is exactly one cj>: There are exactly two cj>'s: 3xcj>x & 'v'x'v'y(cj>x & cj>y ~ x=y) 3x3y(x:. since the world does not see fit to determine whether either of the later ships are the original ship. We will also need to look at alternative formulations that give slightly different results. I detail this below. At least. (If your instinct tells you that the right answer should be "exactly two". since two ships docked. In one sense. and correctly entails that there are no answers in the other cases. then the theory under consideration gives the right answers when they exist. we want to say . Were there exactly two? That seems impossible to answer.t y & cj>x & cj>y) & 'v'x'v'y'v' z( cj>x & cj>y & cj>x ~ x = y v y = z v x = z) These analyses are natural hypotheses about the meaning of cardinality claims. there should be no answer.

The answer according to the theory is whatever we get by feeding the extended data into the numerical formulas above. a. however. (The arguments that follow are elementary but a bit lengthy. . 0 Ship a is not ship p. than the same sorts of arguments that are needed when there is no indeterminacy at all. these are true: Sa & Sp. will be addressed in §8. This. So every instance of it would not be false. If it were not false that there is at most one. and indeterminate whether it is the ship with new parts.(a == p) 'Vo == a 'Vo == p. this would not be false: VxVy(Sx & Sy => x == y). and it is indeterminate whether them: is either of D3 --.) Exactly one? It is false that there is exactly one ship. and p the ship with new parts. This entails that 'a == p' is not false. First. Here are the relevant data. I begin with the question of how many ships there are all told. D1-D3 then generate the answers to the "how many" question. They are not much longer. too. and determinate that the newly assembled ship is distinct from the ship with new parts. But by D1. a the newly assembled ship. and p are ships under consideration: D1 So & Sa & Sp.S. That is. it confirms them on the assumption that it is indeterminate whether the original ship is the newly assembled ship. the theory confirms our naive judgements about cardinalities.140 Counting Objects something like "two-ta-three is the right range of answers". Setting aside disjunctions. And these are determinately all the ships that are under consideration: D2 Vx(Sx => !x == 0 v !x == a v !x == p). Let 0 be the original ship. since it is false that there is at most one. contrary to 03. 0. including this one: Sa & Sp => a == p.

That is. but twenty-one are immediately ruled out by the non-identities above). Then some instance is false. this is true: 'ifx'ify'ifz(Sx & Sy & Sx => x = y v y = z v x = z). such that this is true: Se & Sf & Sg & e"# f & e"# g & f"# g. and indeterminate whether there are at most two. Suppose on the other hand that it is false that there are at most two ships. For example.Counting Objects 141 Exactly two? It is indeterminate whether there are exactly two. entails that contradicting: '10 =a from D3. with each of them contradicting D3. Suppose it is true that there are at most two ships. f. This. But D3 rules out the possibility that any of the disjuncts are true. this is true: f = a & g = p. . and that it is false that there are at most two. The other five cases are parallel. with 'e"# f' from the 0 & above six-part conjunction. and g. since it is true that there are at least two (infer '-. These yield six cases to consider (actually.(a = p) & Sa & Sp' from D1 and D3 and existentially generalize it). together with D1 lets us infer that this is true: o =av a=p v 0 = p. one of the cases is: e= This. twenty-seven cases. So there are entities e. From this and D2 we infer each of: e=ove=ave=p f=ovf=avf=p g = 0 v g = a v g = p. this is false: 'ifx'ify'ifz(Sx & Sy & Sz => x:=: y v x = Z v y = z). We can show that it is indeterminate whether there are at most two by refuting the other two options: that it is true that there are at most two.

But there is no answer to the "unadorned" question of how many ships left port. But I suggested above that exactly one ship left port. and this is sufficient to refute the claims that there are exactly four. and it is indeterminate whether a or p did: D4 La & vLa & v Lp. Which is right? On the analysis that we have so far. Suppose we ask how many ships left port. etc. new complications enter when this condition is not met. As discussed earlier.142 Counting Objects Exactly three? It is indeterminate whether there are exactly three.) It appears then that there might be no determinate answer regarding how many ships left port. a property. I use the simpler. . exactly five. Since it is indeterminate whether the ship with new parts is the original ship. it is indeterminate whether the ship with new parts left port. Thus 'Sx' and '!Sx' are equivalent. To verify these claims we need additional data about which ships left port. and we need only consider one version. 8. Certainly the original ship left port. 'ship'. they also presume that if either of the ships that docked had left port. since it is indeterminate whether there are at least three (this proof is similar to the proof that it is indeterminate whether there are at most two) and true that there are at most four (proof left to reader). As indicated earlier. To show that exactly one ship determinately left port we need to show that this is true: 2 2 By D4 it is determinate for each entity under discussion whether or not it is a ship. we assigned a cardinality to the extension of a predicate. that is determinately true or determinately false of each thing. Exactly four? It is easily shown (using D3) that it is false that there are at least four ships.4 COUNTING THE SHIPS THAT LEFT PORT When we counted ships above. Each of these is an additional assumption that may be questioned. exactly one ship determinately left port. I assume that 0 left port. likewise for the newly assembled ship. that is determinately true. (These inferences presume that leaving port is. that ship would be the original ship. or is equivalent to.

So on this reading it is indeterminate whether exactly one ship left port. This is vacuously true. we can add the required'!'. would lead from the displayed conjunction to theses that contradict 04. This is determinate in virtue of its form (because of the determinate-truth connective). If we interpret the question as how many ships left port. But the first disjuncts taken together entail 'e = f'. It remains to show that any instance that has an indeterminate antecedent has a true or indeterminate consequent. and existentially generalize to get the first conjunct. But at that point the antecedent would be indeterminate if e = a and f =p. To see that the second conjunct is true. that we have an instance where the antecedent is true: !(Se & !Le) & !(Sf & !Lt). in that case the consequent is false. first. and that any instance that has an indeterminate antecedent has a true or indeterminate consequent. by Leibniz's Law. since the antecedent cannot be indeterminate: !(Se & !Le) & !(Sf & !Lt). which makes the consequent of the conditional true. Since it is true. then the proof would go the same right up to the last step.1 to take the adorned reading (the one that includes the determinacy modifier) as the default reading for claims involving indeterminate identity? . Suppose. See the appendix to this chapter for a comparison of these results with the ones one would get on a classical analysis of the situation using supervaluations. because any of the others. We can rule out any of these disjuncts being true except the first on each line. omitting the determinacy modifier.Counting Objects 143 :Jx!(Sx & Lx) & 'lfx'lfy(!(Sx & !Lx) & !(Sy & !Ly) =:> x = y). . By 02 we have both of: e=ove=ave=p f = 0 v f = a v f = p. The first conjunct is easy since 01 and 04 yield that this is true: So & La. This is the kind of case that rationalizes the decision in §8. we need to show that any instance that has a true antecedent also has a true consequent. and so the whole conditional is indeterminate.

and such that every ship is either determinately or indeterminately identical with it. The formula that says there is exactly one ship is nonfalse because there is something-the original ship-which is determinately a ship. . The ship case refutes the transitivity of indeterminate identity. But indeterminate identity is not transitive. there is a fairly natural way to formulate 'there is exactly one <jl' which is plainly wrong. In fact. since two distinct ships docked at the end of the trip. all answers except 'two' and 'three' are false. notice that in the ship case this tells us that it is indeterminate whether there is exactly one ship. It is important to note that some of these are not equivalent in a non-bivalent setting. 4 The equivalence in bivalent logic depends on the transitivity of identity. It is the formulation that says 'something is <jl. this turns into a case of the transitivity of indeterminate identity.144 Counting Objects 8. To see that this is wrong. Thus this is also indeterminate: There are two or three ships. which is indeterminately identical to the other later ship.6 SUPER-RESOLUTIONAL READINGS There is no correct answer to the question about how many ships there are all told. So this seemingly natural formulation that is equivalent in bivalent logic to the one we have used above is not equivalent in non-bivalent logic. In the present application. since one of the later ships is indeterminately identical to the original ship. but the later ships are distinct from one another. But it is plainly false that there is exactly one ship. and those are indeterminate. and any <jl is it': :Jx[<jlx & Vy(<jly:::::} x = y)]. 4 This will be important in assessing how definite descriptions work in the next chapter. 8.5 VARIANT ANALYSES In bivalent logic there are many different equivalent formulations of cardinality claims.

This appears to be a matter of interpreting the disjunction in accord with the super-resolutional reading discussed in Chapters 3 and 5. or with ship p. The super-resolutional reading by itself leaves this indeterminate between one. but if I did. or with neither. When asked to explain why they think this. they tend to answer that however things go. then the super-resolutional reading does indeed say that there are exactly two ships overall. it is indeterminate which it is identical to. the original ship must be identical to one of the resulting ships. I would say that the theory correctly accommodates my view on the super-resolutional reading. If that is so. . and the last produces three. and every resolution does make one of the options true. the first two options produce two ships.Counting Objects 145 But some of us have some inclination to endorse this claim. If it is ruled out as a resolution. and three. and the theory alone does not yield this as an option. This is an automatic product of the theory sketched here coupled with the claim that it is impossible for the original ship to cease to exist and be replaced by two others. If indeed it is genuinely impossible for the original ship to cease to exist while being replaced by two distinct completely new ships. but there can be no third option where the original ship just ceases to exist. But it is possible to accommodate it. In the case of the ships. There are people whose instincts tell them that there should be exactly two ships all told. But the answer is determinately one if you add that it is impossible for either of the later ships to leave port while being distinct from the original ship. then the resolution that turns out that way is not a possible way the world could go. two. I have not taken this to be part of the extended data (because J think it is false). it should not be one of the resolutions that we count when evaluating the truth of a claim on all resolutions. s 5 A similar point applies to the question of how many ships left port. So this kind of reading validates the assertion that there are two or three ships. any resolution identifies ship 0 with ship a. while leaving each disjunct indeterminate. then a list of options is correct on the resolutional reading if and only if each option is made true on some resolution. I don't agree with that claim about what the possibilities are. Consider all ways of making properties more determinate.

and a ship d whose location is continuous with that of b's. In the super-resolutional reading. and we end up with a ship c whose location is continuous with that of a's. One might want to argue that there is no fact of the matter whether a is identical with c or with d. There are at least two ships here.7 PERSONS. and not more than four. For the unadorned reading of how many persons there are in the office. but for the determinate-person reading we get the answer "exactly one". there is no determinate literal answer. there is no answer on the unadorned reading. The material discussed above carries over straightforwardly to the other paradigm puzzles. which consists entirely of the planks from b. AND PILES OF TRASH. It is false that there were less than one or more than two persons. The results are: Me and my body: Suppose it is determinate that both I and my body are in the office. ETC. The disrupted person(s): Exactly one person entered the room in either the determinacy or the unadorned sense. but it is indeterminate whether there are two and indeterminate whether there are three and indeterminate whether there are four. until the planks are completely interchanged. On the resolved . and likewise for b. CATS. The piles of trash: There is not less than one pile of trash nor more than three. a and b. we get the same answer.146 Counting Objects 8. which consists entirely of the planks from a. On the resolved reading there are either one or two. it is indeterminate whether there is one and indeterminate whether there is two. and nothing else is there. but in the resolved sense there are between one and three piles of trash. Then there are not less than one nor more than two things in the office. which is what I consider to be the default case. there are either one or two. afloat near one another. The cat: There is exactly one cat on the table in the determinacy sense. Each ship has its planks replaced with planks from the other. but both 'one' and 'two' are indeterminate as answers. The picture would look like Figure 8. A more complicated case: Suppose there are two similar ships.4.

that the resolution into three is impossible.4 reading there are either two or three or four. (One might want to argue on the basis of the symmetry of the example. If we impose a constraint on the resolutions that they be symmetrical. then wc would conclude that there are at least two ships. it would be indeterminate whether there are two and indeterminate whether there are four. depending . 8. and not more then four.) Appendix: A Classical Analysis Using Supervaluations A referee for OUP has suggested a comparison of these results with the ones one would get on a classical analysis of the situation using supervaluations.Counting Objects 147 Ships b FIG. The answer is that a variety of answers are obtainable. and not exactly three.

A more plausible option is to require that some ship left port. C. Aq. lA. ABj. AB. BC). Aq. Bq. One might wish to also insist that on any extended valuation no two ships ever overlap. B. B. AB. lA. BC. and each exact cardinality judgement for ships leaving port from one to five is superindeterminate. AC. lA.AB. IBj.AB. IB. (B. C. IB. C. AB. BC). C. lA. [A. BC). IB. AB. where the atomic ingredients of the situation under discussion are: A: the object that coincides with the original ship up to the reconstruction B: the object that coincides with the ship with new parts starting from reconstruction C: the object that coincides with the ship with old parts starting from reconstruction There are also the mereological sums of these: AB. Bq. since they each require there to be a ship that is in two places at once. C. Bj. For simplicity. IC. Aq. AB.Aq. AC). Bq. C. Bq. B. ABj. IB. and ABC. and those from six up are superfalse. This would reduce the extended valuations to: . IB. First. B.) The interesting question is what constraints to impose on the extended valuations of 'ship'. C). lA. C. there is the ontological one: since there is no worldly indeterminacy of identity.B. BC). AC). this reduces the possible extended valuations of 'ship' to: (C. IC. I assume that both BC and ABC are ruled out as potential ships. Bq. IABj. C. Bq. B. IB. and that at least two docked. B. ABj. Bq. IB. lA. (Without this constraint. lA. one must decide what there is in the world on which to base the extended valuations. ABj. C. IA. A typical choice is a mereological one.AB. The valuations then must be chosen from among the following as possible extensions for 'ship': <1>. AB). lA. BC).ABj. B. AB. (A. B. lA. IB. BC). B. Two sorts of choices must be made in any application. we get many more options than those detailed below. [AB. For simplicity. BC). lA. lA. AB. IC. q. C. IAq. IB. C).148 Counting Objects on how the supervaJuation technique is applied. lA. AB. C. Aq. lA. AB. C. lAB. lA. BC) If all of these are allowed (an implausible application) then each exact cardinality judgement for ships leaving port from zero to five is superindeterminate. C. C. (B. B. AB.C. AB]. B. lA. lA. IAj. AB}.BC) With these options it is superfalse that no ship left port. Iq. lA. [A. AB]. C). BC). C. lA. Aq. C. [A. BC).AB. B. (A. ABj. The second choice involves which are the acceptable extended valuations of the predicates in question. I assume that on any extended valuation the predicate 'left port' is true of A and of any sum that includes A. AC). lA. ABj. IC.

) .Counting Objects [A. Instead of the above choices. one might allow ships to overlap but insist that no ship that left port ceases to exist midway in the voyage. Bl. C]. That leaves the last two options just discussed unchanged for the question of how many ships left port. left port. BJ 149 Now it is supertrue that exactly one ship left port. (I suppose that one should not really rule out the option that there was never more than one ship. namely. and superfalse that three or more left port. [AB. C]. sllpcrindeterminate whether one. Cl. [AC. ABC. though it would change thc results for the judgement about how many ships docked. [AB. if this is added to the edict that no ship is ever in two places. B. the possible extended valuations are: [AB. AC] On this option it is superfalse that no ship left port. or whether two. [AC.

It is easy to generate worries about denotation when objects are indeterminately identical. in some sense. then there is no (unique) thing which Aphla is. whatever unique thing it may become". it would seem difficult to claim that there is indeterminacy in the world. whatever it may be". This gives the game away. also referring to b? If not.1 THE ISSUE Suppose it is indeterminate whether a is identical to b. The second is that if indeterminacy of identity automatically brings with it indeterminacy of reference. since reference is supposed to be unique. as opposed to it being an illusion generated by vagaries in the semantical relations between language and the world. that there is no unique thing which Aphla is already. He suggests that it is wrong to hold that the singular term 'Aphla' can denote a unique object in such a case. implying as it does. we may not be able to make determinate claims about any object at all if it is indeterminately identical with some object. we should have to say" 'Aphla' denotes Aphla. In this chapter I address the question of how we may refer determinately to objects that are indeterminately identical to "other" objects. The first is that if indeterminacy of identity prevents unique reference. Here is an example from Burgess (1990) about a mountain Aphla which is (allegedly) indeterminately identical with a mountain Ateb.9 Denoting Objects 9. We cannot gloss '''Aphla' denotes Aphla" as" 'Aphla' denotes Aphla. can we refer to a at all? This sort of issue arises regularly in the literature. He argues as follows: If there is something with which Aphla is indefinitely identical. then . Can we then refer to a without. In the next chapter I examine the proposal that all indeterminacy of identity is an illusion based on semantic indeterminacy. There are two dimensions to the worry. If there is no unique thing which Aphla is already.

9. Our semantics must somehow reflect this consequence of vague identity for the naming relation. is a case in which determinate individuation is ex hypothesi absent. then the semantical relation between name and thing named must also be vague. (1988: 354) These comments raise severe questions about what objects we are talking about when we use singular terms. and we dub it 'Kim4Ever'. (We also see another ship dock. 153) Stalnaker (1988) makes a related point. and it is neither definitely true nor definitely false that x = y. On Wednesday we see a ship dock.Denoting Objects 151 the sense of 'Aphla' is not yet sufficiently definite to determine a unique object for Aphla to be. If. (ibid.2 NAMES Suppose we see a ship leave port on Tuesday. which we dub something else. and we dub it 'Samantha's Pride'. The prolepsis involved in the manceuvre is unintelligible. in commenting on this passage from Van Inwagen he says: I think that van Inwagen is wrong to suggest that 'Alpha' could definitely name x even when it is neither definitely true nor definitely false that x = y. Commenting on Salmon's argument against indeterminate identity (Chapters 4 and 11). That the sense of the name could be modified to overcome this difficulty is no excuse for holding that we can regard this task as already having been accomplished. for example. then apparently there is no unique thing that we denote by 'Aphla'. he says: [T]he argument shows that if it is indeterminate whether a = b. A similar point is also articulated by van Inwagen (1988): [I]f identity is indeed vague. or what 'b' refers to. (1994: 147) Determinate reference presupposes determinate individuation. however. 'Alpha' definitely names x.) Then we discover that in between there occurred the . (1988: 259) Cowles (1994) goes further. (1990: 271) If Aphla is indeterminately identical with Ateb. The questions need to be addressed.ll1e case of Alpha/Omega [the disrupted person(s)]. then it seems inevitable to suppose that it is neither definitely true nor definitely false that 'Alpha' names y. then it is indeterminate what 'a' refers to.

what are we saying about it? Are we saying indeterminately that it is indeterminate whether Samantha's Pride left port on Tuesday? But it is determinate that Samantha's Pride left port. and indeterminate whether Kim4Ever left port on Tuesday. and which we denote by 'Kim4Ever'? What does it mean to ask if there is a "unique thing"? A thing is unique in some respect or other. this does not follow. And so I am not speaking of Samantha's Pride at all. Are we then speaking indeterminately about Samantha's Pride? If so. that it is indeterminate whether it left port on Tuesday. even indeterminately. Then it seems that by Leibniz's Law. The contrapositive version holds only for contexts in which the predicates stand for properties. What about the question raised by Burgess: Is there a unique thing which Kim4Ever is. Fortunately. Apparently what is being addressed here are the following two questions about uniqueness: . but by a contrapositive version of it. I have dubbed Kim4Ever 'Kim4Ever'. and (determinately) do not use the name to refer to Samantha's Pride. Is it then indeterminate whether we have said something wrong? I hope not. 'N refers to x' is a paradigm case of a predicate that does not stand for a property of x. Suppose that we try to explain part of the situation by saying that it is indeterminate whether Kim4Ever left port on Tuesday. The key question is this: If it is indeterminate whether Kim4Ever is Samantha's Pride. It seems wrong to say of Samantha's Pride.152 Denoting Objects sort of repair/assembly that leads us to conclude that there is no fact of the matter about whether Samantha's Pride is Kim4Ever. such predicates instead characterize parts of our conceptual apparatus for representing the world. Suppose that I (determinately) use the name 'Kim4Ever' to refer to Kim4Ever. have I indeterminately dubbed Samantha's Pride 'Kim4Ever' when I dub Kim4Ever 'Kim4Ever'? I don't think so. without even being indeterminately wrong. in spite of the fact that it is indeterminate (by hypothesis) whether they are identical. This is in spite of an obvious argument to the contrary. This is reasoning not by Leibniz's Law. and as a result the name 'Kim4Ever' as used by me determinately refers to Kim4Ever and determinately does not refer to Samantha's Pride. even indeterminately. But conceptual or semantic predicates do not necessarily stand for worldly properties. Kim4Ever and Samantha's Pride are distinct. At least this is how 1 think I use the name. I think that I can say that it is determinate that Samantha's Pride left port on Tuesday.

Denoting Objects

Is there a unique thing which is Kim4Ever? That is, is there exactly one thing which has the property, being Kim4Ever? Is there a unique thing which is denoted by 'Kim4Ever'? That is, is there exactly one thing which is denoted by 'Kim4Ever'? The answers to these are implicit in the results of the last chapter. There are two ways to read questions like these: unadorned, and with the determinacy interpretation. On the unadorned reading, the question sometimes has no answer; it has no answer because it is indeterminate whether Kim4Ever is identical with Samantha's Pride. On the determinacy reading it is answered yes. There is exactly one thing which is determinately Kim4Ever. As I use the name, the second question can also be answered "yes". There is exactly one thing denoted by the name 'Kim4Ever'; it is the exactly-one-thing that answers to the determinacy reading of the first question. You can reason otherwise by using a fallacious contrapositive version of Leibniz's Law, but not otherwise. This is also the reply to Van Inwagen's comments. If 'Kim4Ever' definitely names Kim4Ever, and if it is indeterminate whether Kim4Ever is Samantha's Pride, then it does not follow that it is indeterminate whether 'Kim4Ever' names Samantha's Pride. Not only does it not follow, it is not true. The position I take here is a consistent one, but hardly one that will convince someone who is wedded to the opposite view. Can I do anything to win over such a person? Perhaps not if we discuss only proper names. The semantics of proper names is controversial, partly because the facts are so undetermined. The names themselves have no semantic structure at all, and so we need to look at our usage and our intent. And in the area of indeterminacy, these give us little uncontroversial guidance. Perhaps we can do better by looking at something with more structure? With this hope, 1 turn to a consideration of definite descriptions.

9.3 DEFINITE DESCRIPTIONS Theories of definite descriptions fall into roughly two categories. One, endorsed by Russell, is that definite descriptions are not semantical units in themselves, and so they have no semantic

154

Denoting Objects

analysis. Instead, one analyses sentences that paraphrase the descriptions; instead of trying to analyse 'the present queen of England' in 'The present queen of England is old' one merely analyses 'There is exactly one queen of England, and she is old.' It is clear from the discussion of the preceding chapter that sentences of this sort will often be determinately true if given the determinacy interpretation even if the queen of England has previously undergone a disruption that makes her indeterminately identical with the former princess. That is, in the circumstances described, this may be true: 3x[!(x is OE) & \ix\iy[!(x is OE) & !(y is OE) ~ x == y] & x is old]. However, it may be worth discussing also those views that take definite descriptions to be genuine semantical units, since this is the area in which the discussion of semantic indeterminacy is often located. There are dozens of theories of this sort, but they almost all agree on the condition under which a definite description denotes something, and what it denotes. The condition is: 1 [DJ: 'lX[<»X]' denotes 0 if and only if 0 is <» and at most one thing is <».

In this condition, the 'if and only if' is to be read strongly (it is to be read as our '<:::>'), in the sense that this is true, or false, or neither true nor false:

'lX[<»X]' denotes

0

if and only if this is true, or false, or neither true nor false respectively:

<»0

& \ix\iy[<»x & <»y ~ x == y].

For definiteness in discussing how this works, J focus on the case of the disrupted person(s), and the definite description 'the person who entered'. The metaphysical situation is the one pictured in Figure 9.1. The question is whether the description 'the person who entered' denotes a determinately, or indeterminately, and in either case whether it also denotes h indeterminately. As in the previous

1 The second condition uses 'at most one thing is <\l' instead of 'exactly one thing is <\l' because the former is simpler, and they are equivalent when conjoined with '0 is <\l'. It also avoids the natural but wrong anaJysis discussed in §8.S.

Denoting Objects

155

Things that entered

b

Persons

FIG.

9.1

chapter, there are two interpretations to consider: the unadorned reading, and 'the thing that is determinately a person who entered'. Between them they seem to capture the things people want to say about cases of this sort.

**9.3.1 Determinate denotation of a without indeterminate denotation of b
**

First, consider the determinacy reading of ' the person who entered':

tx!(x is a person & x entered).

**I say that this determinately denotes person a in Figure 9.1. It denotes person a because this is true:
**

!(a is a person & a entered) & 'v'x'v'y[!(x is a person & x entered) & !(y is a person & y entered) ::=} x = y].

The first conjunct is true since both 'a is a person' and 'a entered' are determinately true. The second conjunct is true because it is a universally quantified conditional with all true instances. Instantiating both variables with a validates both antecedent and consequent. Instantiating either variable with b or with anything

156

Denoting Objects

determinately different from a makes it true since this produces a false antecedent (because only a is determinately a person that entered). Does the definite description denote b as well? No, it does not. H determinately does not denote b because this is false:

**!(b is a person & b entered) & 'lix'liy[!(x is a person & x entered) & !(y is a person & y entered) ~ x = y].
**

It is false because the first conjunct is false; this is because of the initial '1' which requires that both of the component conjuncts be true, and 'b entered' lacks truth-value. So the definite description determinately does not denote b. It is easy to see that it denotes nothing else (except a) too, for the same reason. So here is a definite description that determinately denotes a and that does not indeterminately denote anything at all. So there is such a thing as determinate denotation unaffected by indeterminate denotation even when the term determinately denotes a thing that is indeterminately identical to something. So indeterminate denotation is not forced on us by every term denoting a thing that is indeterminately identical with something. But there is indeterminate denotation as well.

9.3.2 Determinate denotation of a with indeterminate denotation of b

On the unadorned reading, 'the person who entered' determinately denotes a but also indeterminately denotes b, just as the critics say. Consider first a. The truth-value status of:

**'u[x is a person & x entered], denotes a
**

is the same as the truth-value status of:

**[a is a person & a entered] & 'lix'liy[x is a person & x entered & y is a person & y entered ~ x == y].
**

The first conjunct of this is true, so consider the second. This is a universal claim, and it requires all of its instances to be true for it to be true. Clearly it is true if we instantiate the same thing for both

Denoting Objects

157

variables, since this makes the consequent true. Likewise it is true for any instantiation other than a or b, since all such instances make the antecedent false. The remaining option is to instantiate a for one variable and b for the other; this makes the conditional true since it makes both antecedent and consequent indeterminate, and the '~' connective is true in such a case. (So this is an instance in which our formulation of condition [0] may be controversial; see below.) Thus the definite description determinately denotes a. The description indeterminately denotes b. This is because this is indeterminate in truth-value:

**[b is a person & b entered] & 'v'x'v'y[x is a person & x entered & y is a person & y entered ~ x == y].
**

The first conjunct is indeterminate because it is indeterminate whether b entered. The second is true as argued above. So the whole is without truth-value; it is indeterminate whether the definite description denotes b. It is easy to show that the definite description detenninately does not denote anything other than a or b. So the definite description determinately denotes a, indeterminately denotes b, and denotes nothing else. This conclusion is dependent on a particular reading of condition [0], one that employs our'~' connective to symbolize 'nothing other than z is (j>'. Clearly, this might be symbolized differently; for example, it could be symbolized with the material conditional (using '(j> ::J \V' for' ---,(j> v \V'):

**[a is a person & a entered] & 'Vx'Vy[x is a person & x entered & y is a person & y entered::J x = y].
**

If this is used in condition [0 1then it alters the judgement of how the definite description with the unadorned reading denotes a. It now denotes a indeterminately instead of determinately, because when the second conjunct is instantiated so as to yield a conditional whose antecedent and consequent are both indeterminate, the whole material conditional lacks truth-value. With this change it turns out that the definite description indeterminately denotes both a and b. (This change has no effect on the determinacy reading, where the definite description is 'tx![x is a person & x entered]'. )

158

Denoting Objects

There thus appear to be three reasonable ways to interpret the English phrase 'the person who entered the room'. On one interpretation it determinately denotes a and does not denote anything indeterminately. On a second interpretation it determinately denotes a and indeterminately denotes b. On a third interpretation it indeterminately denotes both a and b.

9.4 SUMMARY OF THE VIEW BEING DEFENDED We have seen that there are clear cases in which a definite description determinately denotes an object x and determinately does not denote an object y even though it is indeterminate whether x is y. There are also cases in which indeterminacy of identity leads to indeterminacy of what is denoted, perhaps-or perhaps not--even indeterminacy coupled with determinacy. This has been shown by considering definite descriptions; proper names are harder to draw clear conclusions about, but I see no reason why they cannot do what the definite descriptions do. (None of the examples 1 considered had to do with scope distinctions or with modalities, which are the two main areas in which names and descriptions are alleged to behave essentially differently.) I take this to vindicate my own policy in formulating the theory of indeterminate identity. I have presumed that each of the singular terms in the theory I formulate determinately denotes something, and does not indeterminately denote anything. I have assumed this so that I could present the metaphysical issues without becoming entangled in issues of semantic indeterminacy. Since it is possible for singular terms to behave in this way, I am justified in presuming that mine do. I will continue with this policy. It would be fascinating to pursue a study of the logic and semantics of indeterminate denotation, expanding the language introduced in Chapter 2 to include singular terms that do not denote determinately. However, this would be an exercise in semantics or logic that is distinct from the metaphysical issues addressed in this book. Our issues are difficult enough already without these additional complications. For this reason, J will not introduce indeter-

Denoting Objects

159

minate denotation into the official language of the theory, and I will not speak of it any more. There remains the question of how to interpret data that is formulated in natural language. When we report on puzzling cases, we use wordings that come naturally to us. And when we consider that some statements lack truth-value, we may do so pre-theoretically, without being certain about why. So in considering the "data" I may not be certain that the singular terms used in its formulation uniquely refer. I do think that it is justifiable to assume this prima facie, at least if there is no reason to believe otherwise, and this is the policy I have followed so far. In the next chapter, I consider some challenges to this policy.

or show us how those data are wrong. In my experience. it can't just be an announcement that the opinions with which we begin are naive and inherently untrustworthy. they are the principles upon which we base all our decisions. not a given fact. Proposals of this sort are often stated. Further. and even more often endorsed. not in the world it purports to articulate. It is a central theme of our heritage of work in twentiethcentury analytic philosophy that questions apparently about reality are to be reconstrued as questions about the language with which we describe such reality. For the latter purpose. The theme of this chapter will be that theories that account for . and then about how it applies to the problematic situations. and so a proposal to do that has great instinctive appeal. it needs to either account for the (extended) data with which we begin. they sound more plausible in the abstract than in the concrete. It takes little reflection to see that some work needs to be done in order to have a view worth examining. That is the point of this chapter. not on the basis of a general charge of naivety. But then the work needs to be done to show which are which. The opinions with which we begin are our own. and if we are to reject some of them it must be on the basis of others of them. but they are rarely implemented in any detail. But in any given case. such a proposal is still merely a hypothesis. or in the system of concepts embodied in our language. The world appears indeterminate because our linguistic/conceptual apparatus does not yield determinate judgements. There needs to be a positive doctrine spelled out about how our language works.10 Alternatives to Indeterminate Identity Perhaps the most popular view about indeterminacy of identity is that it is an illusion generated by inexactness in our language. The point of this chapter is to investigate such ideas. but the lack of determinacy lies in the apparatus itself. and it deserves to be examined like any other hypothesis.

or of concepts. I consider them in that order. There is at least one thing that is a potential referent of both the terms in common. The statements that articulate the data are affected in this way too. but it is appropriate that such development be undertaken by those who believe in them. the identity predicate itself. but if they are given supervaluational readings. and it has no unique reading which could yield a single truth-value. and general terms with which the identity puzzles are stated. but the identities in the traditional puzzle cases are not of this sort. there are identity statements that lack truth-value because of the nature of language. are not compelling in their present forms.1 CLARIFYING DENOTATION WITH SUPERVALUATIONS A view that seems quite natural to many writers supposes that indeterminacy of identity is an illusion fostered by the fact that one or both of the singular terms composing the identity statement has more than one potential referent. They come out true because they are true no matter how the references of the terms are clarified.Alternatives to Indeterminate Identity 161 the traditional identity puzzles in terms of the nature of language. and so . My goal is instead to establish the present absence of settled accounts of this kind. some valuations (some ways of singling out unique referents for the terms from among their multiple potential referents) make it true and some make it false. they come out true. because I cannot. I will not attempt to prove that one could not give a plausible account of this sort. and how understanding that language dissolves the puzzles while preserving the data that lead to them. Since at least one singular term lacks a unique referent. Of course. In each case we need to see how it is that our use of language can account for the data that lead to the identity puzzles. because of the lack of unique reference of their terms. the identity statement itself is essentially ambiguous. The available pieces of language that need clarifying are singular terms. and at least one that is a potential referent of one term but not of the other. I will not try to carry this out myself. This is not so for the identity statement. Such accounts should be developed further in the face of criticisms raised here. 10.

The supervaluation approach to singular terms accounts for these judgements as follows. 'J is on Cedar Avenue' is supervaluationally true.162 Alternatives to Indeterminate Identity it lacks supervaluational truth-value. but without any indeterminacy of identity in the world. Let'S' abbreviate 'the building in which Smith has her office' and 'J' abbreviate 'the building in which Jones has his office'. which is around the corner on Cedar Avenue. and both make the statement true. which is on Main Street. three of these are false and one is true. Then both'S' and 'J' have mUltiple potential referents. and Structuralist Manor. have been incorporated into a much larger structure. because there are four valuations for it. But'S is l' lacks supervaluational truth-value. I but this pattern of analysis does not necessarily carry over neatly to the I Nice. and the old designations are still in use. Here is an illustration of how this might work in a transparent case. which occupies the corner of Main and Cedar. I prefer to see the statements as all contextually ambiguous. It is supervaluationally true because there are two valuations. not why the sentence has a definite truth-value. We are naturally inclined to assent to the claim that the building in which Smith has her office is on Main Street. But tradition is strong. Suppose that Smith has an office in Old Ivy Lodge (and therefore in PostModern Hulk) and Jones has an office in Structuralist Manor (and thus in PostModern Hulk). the truth-value gap is simply a consequence of semantic variation in the potential reference of the terms. one for each potential referent of'S'. . so there is no consensus among them. the former potentially refers both to Old Ivy Lodge and to PostModern Hulk. The fact that certain of them come out true no matter how the ambiguity is analysed explains why we may not botber to raise the issue of ambiguity in a practical situation. PostModern Hulk. but not compelling. and there is no supervaluational truth-value. and to the claim that the building in which Jones has his office is on Cedar Avenue. Thus we have a truth-value gap for an identity statement. corresponding to the various combinations of the potential referents of the terms. Suppose that Old Ivy Lodge. but it seems that there is no answer to the question whether the building in which Smith has her office is the building in which Jones has his office. This seems to me to be a nice analysis of the situation. For parallel reasons. We treat the statement'S is on Main Street' as true because it is supervaluationally true. and the latter potentially refers both to Structuralist Manor and to PostModern Hulk.

But some allowable classical valuations to both 'this person' and 'this body' assign a to 'this person' and a set containing both a and b to 'body' (making 'This person is a body' true). As a result. As a result. including entity {/. Because of such assignments it will not be supertrue that a is a person. the identity statement is assessed as supertruthvalueless. But this permits an assignment of p to 'this person' together with an assignment of an extension to 'person' which includes p but excludes a. and so at least some bodies are determinately persons. call this 'a'.4-7. yielding a classical valuation that is true. call it 'p'. call it 'b'. likewise for 'This body is a body'. (I believe that there are difficulties in explaining how these assignments to multiple pieces of language are jointly allowable. yielding a classical valuation that is false. any allowable classical valuation to both 'person' and 'this person' makes the former true of the latter. The mistake is to speak of "[he extension of 'person'" as if this is a fixed thing. I am indebted to a referee for our for pointing this oul. For a parallel reason. Fine (1975) on penumbral connections. a. Instead. Further.) . And there is a potential referent of 'this body' which is not a potential referent of 'this person'.) So whenever there is an assignment to the pair 'person' and 'this person'. every potential referent of 'this person' must be in the extension of 'person'. so 'This person is a person' is supertrue. is in the extension of both 'person' and 'body'. and likewise for 'This body is a person'. This procedure is formally adequate. The mistaken objection goes as follows: Since 'This person is a person' must come out supertrue. so 'This person is a body' is supertruthvalueless. one gets the required extended data. They share a potential referent. these reservations are implicit in the discussion in §§1O. In a draft MS I argued-mistakenly-that there is a formal difficulty with this solution. a must also be in the extension of 'body'. whereas some other allowable valuations do not assign either a or p to the extension of 'body'. and which one is selected for a valuation can be made to depend on which potential referent is selected for 'this person'. In evaluating the identity statement 'this person = this body' some classical valuations assign a to both terms. There is also a potential referent of 'this person' which is not a potential referent of 'this body'. namely. Let us consider one of them: the case of me and my body. the thing assigned to the latter can be in the extension assigned to the former. the predicate 'person' has multiple potential extensions.g. It then follows that some entity. one valuation assigns p to 'this person' and a or b to 'this body').Alternatives to Indeterminate Identity 163 identity puzzles. and some assign different things to the terms (e. A diagnosis of this case would go as follows: both 'this person' and 'this body' have multiple potential referents. (Cr. 2 My worry about it is not .

My problem is that J do not believe in such theories. but only one. I believe that this situation is not at all like the building case above. And it is easy to find more plausible. One substance. that is what I think. that is at issue. But I can locate no such entities. There. The solution in fact requires (at least) three entities to be the potential referents of the two singular terms. In addition to these two substances. Perhaps these are not two entities. which corresponds to a above. is my soul. I am not able to convince myself that there are here three distinct entities I am choosing among. and the notion of a human body is unclear between whether it refers merely to the material substance or to the combination. Anyone who accepts an old and once well-established view of the nature of persons would disagree. there is their combination. the other substance. b is that mereological (three-dimensional) sum of matter that "constitutes" the person. The notion of a person is unclear between whether it applies to souls or to combinations of souls with bodies. I can find at most two entities that are relevant to the puzzle: me and my body. there are three things to consider: two substances and a composite of the two. modern views that also have the same structure. to the body-informed-by-a-soul. with 'me') and equally uncertain as to which of two things I refer to with 'this body'. and that there is at least some kind of uncertainty about which of them are denoted by phrases like 'the building in which Smith has her office'. the supervaluational account works perfectly. such as . the analogue of p above. I see plenty of other things. but rather metaphysical.164 Alternatives to Indeterminate Identity logical. Rather. On that view. will disagree. I believe that there are at least three of them. Others. some of them parts of othcrs. and a is the constituted person. I don't believe that there are any such entities. I am certain that there are multiple buildings in question. the analogue of b above is that material object typically called (perhaps unclearly) my body. When I consider what the world is actually like. But there certainly are not three candidates that I am choosing among. J just don't see anything else. But when I consider the case of myself and my body. The procedure requires me to be uncertain as to which of two things I refer to with 'this person' (that is. of course. At least. For example: p is a person considered in abstraction from the matter that contingently makes it up. On such a view. That is. I see at most two things that are at all relevant to the puzzle-me and my body-and I am unsure whether they are the same or not.

The difficulty comes when we ask which of these entities is a cat. for each determinately differs from another with respect to the having of some part. Recall that none of the p-cats is in exactly the same place as any other. and so I describe it here. and that we need to account for talk of the cat in terms of talk of the p-cats. which will be discussed later. I don't expect all others to share this perspective. 10. they are not likely to agree with one another. and although they will all show that I am wrong. But it caught my interest.Alternatives to Indeterminate Identity 165 a certain set of molecules. there is not at present any particular application that will command general assent. for he does not believe in indeterminate identity. It remains possible for some others to patiently explain to me that there really are three or more entities relevant to the issue. which I discuss here. Lewis explores the case of the cat and the p-cats. and that I am just not sufficiently aware of the way in which the semantics of my own language works. then. Lewis suggests that each p-cat is literally a cat: . is the weakness I see in holding that the identity puzzles are to be dissolved by an application of the technique of supervaluations. They all mostly overlap. but I am convinced that these other things are neither me nor my body. Lewis will not acknowledge the existence of a cat which is indeterminately identical with each p-cat. His solution is to suppose that the p-cats are all that there are. However. Lewis does not propose it for the purpose of solving the identity puzzles. This. I suspect. but not quite.2 CLARIFYING THE IDENTITY PREDICATE I am not aware of any viable proposals for solutions to the identity puzzles that result from clarifying the identity predicate itself. nor will he acknowledge a cat without determinate location. He has a supervaluational solution. there is an idea in Lewis (1993) that one might try to tailor for this purpose. and the problem of the identity of me and my body remains. and it is probably a straw man. however. that there are many such explanations. so this is entirely my own invention. he also has a non-supervaluational solution.

inner structure. but that is no worry since you pay with a single ten-dollar bill. This does not make you wealthy. Lewis explains that although there are many cats. They vibrate and set the air in motion-in short. etc. and in a sense no. The theory yields hosts of consequences that contradict conventional wisdom. This validates the statement 'There are millions of cats on the table'. The reason it is almost correct. even though the reasons that led us to posit many cats also leads us to posit many bowls of food. since you also have such high expenses. And so on. but this is money you already possess. weight. the bill itself is multiple. and any cats x and y on the table are such thatx=y. The difficulty is not a practical one. we do so using spatial coincidence. Any way a cat can be at a moment.' (1993: 26) Since each p-cat is determinately distinct from each of the others. because you feed them all at once. They are all too cat-like not to be cats. from a bag that almost costs a moderate amount of money. the many cats are not difficult to feed. but a theoretical one. there turn out to be millions pf distinct things on the table. and you almost feed them only one bowl of food.166 Alternatives to Indeterminate Identity The constituters are cat-like in size. and that he literally has many pets. so you indeed pay a great deal of money. as cheap to feed as a single cat. it is almost the case that there is only one. . For example. It is easy to defuse certain objections to this view. So they cost a great deal to feed. when we say that there is exactly one cat on the table we do not mean: A cat is on the table. they purr (especially when you pat them). cat-constituters also can be. but are almost correct. and the cats all almost coincide. anything a cat can do at a moment. in a sense. each of which is a cat. we do not do so using identity.' Lewis concludes that this is correct. That is. and falsifies 'There is exactly one cat on the table. shape. Lewis suggests. Thus our initial data are incorrect. Actually. is that when we count cats. 3 Lewis calls the p-cats "constituters" in this quote since he is here arguing against the theory which says that there is one cat which the p-cats all constitute. Does this mean rejecting the data? In a sense yes. but harmlessly so. in each case the conclusion is that conventional wisdom is wrong. The many cats are. and motion. catconstituters also can do.

and these accounts differ substantially. a broad p-cat almost coincides with the cat right now. So there is a version of the account in which we construe p-cats narrowly. we conclude. we mean something that is analysable in terms not of identity but of coincidence. If we use true identity in counting cats. There are broad p-cats and narrow p-cats. but that doesn't show that they are never cats. etc. they may have had unfeline pasts and futures. but almost exactly one. paradoxically.A narrow p-cat is a thing with determinate parts that almost coincides with the cat for its entire existence. that there are many cats on the table. but may diverge widely from it in the past or the future. None of these generalizations is even almost true of most broad p-cats. Instead.Alternatives to Indeterminate Identity 167 We mean: A cat is on the table. But when we make ordinary everyday judgements about how many things there are. it only shows that they do not remain cats for very long. and that is good enough for practical purposes. In everyday talk about cats the 'coincides' becomes 'almost coincides'. and one in which we construe them broadly. Lewis intends the broad construal. Before examining this theory we should distinguish two versions of it. do this. but need not. Cats tend to have a lifetime of over ten years. but although . The quote above continues as follows: They are all too cat-like not to be cats. The identity puzzle about the cat is thus explained as a confusion about senses of 'exactly one'. we do not mean this. but we have not discussed their histories. they grow larger for much of their life. not only do I think that there is exactly one cat on the table. Indeed. For there are hosts of things that we say about cats that are not even almost true if cats are p-cats in the broad sense. not exactly exactly one. and any cats x and y on the table are such that x coincides with y. a broad p-cat may. and in this sense there is exactly one cat on the table. Well. I think that there is exactly one cat on the table that I got at the pound. (1993: 26) This version of the theory is a reformist view that is not well suited for explaining the apparent data as being almost correct. they habitually return if you feed them. Further. We have explained p-cats in terms of their present nearcoincidence with the cat.

The reason the solution is problematic as it stands is that it supplies a way to construe the data that almost validates that data. . with the broad ones thrown in as epicycles. Suppose that we sometimes interpret claims about how many things there are by substituting coincidence for identity. whenever it is a cat. I will not take issue with whether an almost-right solution is good enough to be a solution. One way to do this might be as follows. that is pretty good-it is almost good enough. (One could imagine it being made consistent by a massive project of reinterpreting our ordinary sentences. Then 'Cats usually live over ten years' might be interpreted as 'Maximal cats almost always live over ten years. When coincidence and identity coincide. where it does not do this. coupled with the claim that p-cats are (now) cats. and this is really a theory of narrow p-cats. The first is that it is not almost correct as it stands. the second is that even if it is almost correct.' But maximal cats are then our narrow p-cats. it does not generalize to the other identity puzzles. is straightforwardly inconsistent with many of our common beliefs. This does not mean that we always do this.) In order to return to the task of this chapter-which is to explain the data. and perhaps fully good enough. and each cat. for we make plenty of assertions about how many things there are where spatial coincidence is not applicable: "There are two prime numbers between 3 and 9". the latter is not. T have two other worries about this solution. we can also use coincidence. not to reform them-l focus on the version of the theory that appeals to narrow p-cats. "Three theorems were formulated by McGuillicutty". if we have a solution that explains how the data in question is almost right. we cannot. without giving us any reason not to also construe the data in equally good ways. in a situation in which actual coincidence coin4 This sort of fact will raise difficulties for any account of definite descriptions based on this view. In some cases. We call something a maximal cat if it is a cat throughout its existence and if it does not ever almost coincide with something that is also a cat throughout its existence and exists longer. "Four electronic payments failed to reach the bank".168 Alternatives to Indeterminate Identity the former is almost true. almost coincides with a maximal cat. 4 So the broad construal of p-cat. But when it is absolutely clear and uncontroversial that near-coincidence diverges from actual coincidence.

This is easy enough to illustrate by focusing on the p-cats themselves. How many p-cats are there on the table? Presumably. Then there are two ways to construe the question. and 'p-cat' suggests a technical context in which we count with identity. and I think it does not. For example. or only larger solid entities.Alternatives to Indeterminate Identity 169 cides with identity. But the p-cats are cats. dust. and since each p-cat is too catlike not to be a cat. since this is the point at which we will disagree about the data. making clear for example whether we are going to count air molecules as things. Lewis has made it clear that once the question is clarified. in popular usage. I don't get any interpretation of the question that yields a plausible answer that there are actually millions of cats on the table. identity rules. making clear whether by 'exactly one' we mean one that doesn't differ much from the others or one. The response will be. that claim is not in any sense almost true. and there is no sense at all in which there is exactly one p-cat. and there are no others. generalizes to other identity puzzles. and when we do this there are two interpretations of the question. with two different answers. even though the p-cats almost coincide. When I do this. if it is a solution. millions. it is only in a loose and popular sense that there is only one cat there. we count them using real identity. This is true about popular usage in some related cases. or it is at the very least a natural option. he believes that there are literally many p-cats on the table. wallpaper. So there can be no difference in the answer to the question "how many are there?". and the cats are p-cats. and Lewis has made the natural choice. there is no difference at all between cats and p-cats on this view. if the interpretation of the question is sensitive to the vocabulary used in it. A distinct question is whether this solution. and so on. of course. But we can clarify our usage. Recall that Lewis does not say that it does. But this is not a point amenable to argument. and this is the only available interpretation. So there are exactly as many cats as p-cats. when we know there is air. setting aside the question of whether they are cats. Consider the case of the . that there can be a difference in the answer. We can do the same for the question about cats. I think that there is only one cat no matter how strict we are in our wording. there may be "nothing" in a room. This is perhaps a bit too facile as a refutation. Perhaps 'cat' suggests an ordinary context in which we count with near-coincidence. If p-cats are under discussion.

they are in exactly the same place. 10. We find this problematic since we are forced to say there are many with no differences at all among them. .170 Alternatives to Indeterminate Identity disrupted person. 5 One might try to create an analogue to the cat case by considering the p-persons at time t to be the temporal parts of the person that exist at t. 5 We need some different approach. the persons are not in almost the same place. So what if we replace the multiple persons in exactly the same place by multiple p-persons in almost the same place? That solves the problem of distinguishing them. and since the truth of these propositions determines the truth of the sentences that express them. It is not that some do and some do not. if the disruption consists of a partial brain transplant. since they all occupy slightly different places. Since the concepts associated with our words determine the propositions that get expressed. this result follows even if there is no disruption at all.) So an appeal to p-persons does not solve this problem.3 INDEFINITE CONCEPTS It is natural to hold that the concepts expressed by our words are imprecise. thus there would be no answer to the question of whether anyone now is the person born previously at a given time and place. if that is what happens. A supervaluational reading then tells you that there is one person here now. it is unrelated to the issue of identity. then all of the p-brain portions get transplanted or none does. if the concepts are imprecise. A solution to the puzzles might be to consider how we imagine replacing our imprecise concepts by more precise ones. However. but there is no answer to the question whether the person who is here now is the person who was there then. For example. (Or. We would have a solution to the puzzle if some p-persons persisted through the change and some did not. But in the many-person case. The supervaluational approach to the person case is similar to the cat case. and there was one person there then. But there is no reason to expect the disruption to cleave along the lines that distinguish p-persons. Perhaps this is what is to blame for identity puzzles with no answers. nor any of the other paradigm puzzles that we have discussed. this may produce propositions with no truth-value. But distinguishing the p-persons will not solve the puzzle. in that we are driven to suppose that there are several persons in the same place at the same time.

. and we have other priorities. say. Ideally. of person is imprecise. and how some appeal to precise concepts might shed light on this. Holding that our concepts are imprecise does not. The identity puzzles are unusual cases in which it does make a difference how we refine things. by itself. when we say that one ship left port. and finally. (If the term occurs twice. The ship puzzles arise not because we have no idea how many ships there were originally. So what is needed is an explanation of how imprecise concepts sometimes do. any refinement would produce the same result. But that takes time. Our concept. and sometimes do not. more needs to be said than this. where 'ship*' expresses the resulting refinement of the unrefined concept of ship. and work. This proposal is an application of supervaluations. Here is one explanation. Somehow our concepts are good enough for determinate judgements about these matters. If we refine ship in one way. it is correct to conclude that the original ship. Refining yet a third way lets us conclude that the original ship. is identical with the shipl with new parts. A valuation of a general term results from its coming to express a refined version of the concept that it actually expresses. where 'ship/ expresses one particular refined version of the concept ship. and if we refine in some other way.Alternatives to Indeterminate Identity 171 Of course. We don't bother refining on the spot because it wouldn't make any difference to the correctness of the claim at issue. In ordinary everyday situations we don't bother to consider these refinements. and that any way of doing this will validate the claim that one ship * left port. for the purpose of the judgement at hand. but with the understanding that there are indeed ways to refine it. yield determinate judgements. because there is no particular way to refine the concept that is importantly different from any other way. there is no unique answer to the original (unrefined) question. ceased to exist. and we know a fair amount about what would result. it would be replaced by a more refined concept. it is correct to conclude that the original ship2 is identical with the shipz with new parts. but because we think there was originally one and finally two. and was replaced with two new ship/so Since how we refine makes a difference to the correctness of the identity judgement. explain why they nonetheless allow us to express a great deal of exact information. This is because. effort. For example. So we use the word 'person' without refining the concept associated with it. we know that there are many ways to refine the notion of a ship.

since they come out true on some valuations (some refinements) and false on others. part of our extended data) if and only if it is supervaluationally true on this account. the person/body puzzle is a special one. But any of the problematic identity judgements lack supervaluational truth-value. so. The use of supervaluations of precise versions is an analysis of how we would rationalize our use of the imprecise concepts. Rather. we use our imprecise concepts as they stand. likewise the claim that exactly two ships docked. and this may be too narrow a point to reject supervaluational solutions in general. This creates indeterminacy in the term 'the original restaurant'. for the . I argued that no supervaluational assessment will solve the problem of me and my body. So it is worth looking more closely at an intuitively motivated theory of this sort.172 Alternatives to Indeterminate Identity we must be sure that the same valuation is used for each occurrence. Stalnaker considers a puzzle about restaurants that is isomorphic (with respect to identities) with the ship puzzle. A restaurant is started called 'Bookbinders'. the claim that exactly one ship left port is supervaluationally true.) Then a judgement is intuitively true (and. However. An example of this approach is proposed in Stalnaker (1988).1. on other ways it will become the other. In § 10. and with a history posited that makes each identity claim as plausible as the other. and granting that we are using them in situations in which it is not relevant to our present concerns how to make them more preciseit is not relevant because it doesn't matter how this is done." On some ways of arbitrarily refining our concept. there are two rival restaurants with almost the same name. At some later date. Stalnaker (1988: 353) supposes that "the concept of restaurant is indeterminate in that it is indeterminate what counts as the same restaurant at different times. each claiming to be the original Bookbinders. and on yet others it will cease to exist. but no indeterminacy in the world. We would rationalize our use of the imprecise concepts by granting that we are using imprecise concepts. since it is true no matter how the concept ship is refined. to be succeeded by two new "restaurants". the original "restaurant" will become one later "restaurant". if such rationalization were called for. For this solution to be plausible it is important to emphasize that we do not ordinarily think of ourselves as seeking a common content of mUltiple precisifications of our imprecise concepts. In the ship case.

The theory needs to specify how to refine concepts. If this is not accomplished.Alternatives to Indeterminate Identity 173 same answer would result no matter how we do it consistent with our intentions. so that two of them might momentarily coincide. for example. (The 'in normal circumstances' is there to avoid. this seems outright impossible to do. Since the conceptual resources at our disposal consist entirely of unrefined concepts. In any particular case. What makes the identity puzzles special is that they are cases in which identity statements can be made for which it does matter how we make our concepts more precise. there remains a set of identity puzzles for refinedships that are not solved by the technique. 6 To refine the notion of ship we need to explain how to tell if there is a ship present (or several ships present) in a given situation. science-fiction cases in which techniques have been developed to let objects pass through one another. and we need to explain how to tell for any such refinement whether any given refined-ship a is identical with any refined-ship b. this proposal. all by itself. (Because of the "fest-conference" in which it originated.) If it does this. This is because so few guidelines are given for refining our concepts. the tone of writing in the latter was meant to be "festive". that it is true on any refinement that exactly one ship left port? There is nothing about the view as it is sketched to guarantee this. And the instinctive recognition that making them precise in different ways would yield divergent answers is what underlies our bewilderment when confronting them. then any refinement of the concept ship will let you stand at a dock and know that there is exactly one refined-ship in front of you no matter how you refine ship. we do so in such a way that if a sortal term is true of extended spatio-temporal objects. The theory must be fieshed out so as to guarantee it. It is fi Criticism of this sort appears in Woodruff and Parsons (1997) and Parsons (forthcoming) in somewhat primitive form. However. But concept refinement needs to be implemented with the resources at our disposal. That is the view. it is difficult to see how to go about this task. no disrespect was meant to the views criticized there. But why should we think that it works? How can we be sure. then nothing counts as a refinement of the concept that it expresses unless the refined concept is never true of two spatially coincident things in normal circumstances. is not adequate. for example.) . perhaps something like this: Whenever we refine our sortal terms.

it is to be taken as if it were expressing a precise concept. Some refined notions of ship rule that their denotata come into existence upon assembly. Recall that the view is not that we already grasp and use a multiplicity of refined concepts when we speak. The identity puzzles are of this sort. because in everyday matters they all will validate what we are saying. But. it will not work at all. Suppose that the refined concepts work like this. It is even more difficult to have confidence that the end result will yield the hoped-for patterns needed to solve the identity puzzles. and we suppose that there exist concepts which are precise enough. regardless of whether the parts were previously parts of a refined-ship before being separated. even though they are perhaps in principle ungraspable by us. We have in mind ideals of completely precise concepts. The view could be that we do not ever have to be able to produce the refined concepts. an answer to this challenge. but we have in mind. why assume that this will work as advertised? If we consider the most obvious ways of refining our concepts. we have various refined notions of ship. And so on. the only requirement being that none of them permits any indeterminacy in the question of whether one of its denotata exists at a time and place that a ship- . we only need to be confident that our concepts could be refined on demand. When we use a word. they thereby cease to exist. however. Likewise for disassembly and reassembly. or whether it survives disassembly and reassembly. again. There is. one which is a refinement of the imprecise one we actually express. But even this is perhaps not necessary. These various combinations of criteria for identity through time are embodied in various refined notions of ship. that combinations of imprecise concepts can in all cases yield precise ones. Instead of the imprecise notion of ship wherein it is often unclear whether a ship survives replacement of its parts. even if the parts are replaced. Some of these permit their denotata to have many of their parts replaced and go on existing. at least vaguely. But for applications far removed from present-day concerns. we just speak as if we are expressing precise concepts. Generally. and some of them rule that if their denotata lose parts. it won't matter which refinement of our imprecise concept we use.174 Alternatives to Indeterminate Identity conceivable. which is why they lack answers. but unlikely. it may matter. ways they might be made more precise. we do this in the knowledge that all attainable concepts are imprecise.

and that the ship on which she held court is the ship that is buried with her. It produces indeterminacy where we do not want it. But this is just plain wrong.Alternatives to Indeterminate Identity 175 like thing exists. it produces wrong answers in apparently non-problematic cases. This is because on some refinements of the notion of ship the ship on which she held court was transported to the burial chamber (by disassembly and reassembly). And so the theory under consideration agrees that she held court on one ship. then the proposal is in trouble. where they reassembled it. there is now exactly one desk in my study. It was apparently a mistake to suggest that there is a refined notion of ship on which a ship survives disassembly and reassembly. So this natural implementation of the refined-concept view yields the wrong answers in cases in which identity puzzles should not arise. this was the royal barge on which she held court during her reign. or in whether one of its denotata existing at one time and place is identical with one of its denotata existing at some other time and place. When she died it was transported to her burial chamber. Obviously. Instead. Consider Queen X's ship that is buried with her under her pyramid. This is an overly simplistic proposal for what a refined concept should be. We need to refine our notion of refinement. And the wrongness extends to a host of everyday cases. and that one ship is buried with her. and one on which it does not. if these are what refinements are like. we need to say that there is a refinement of ship on which a ship survives disassembly and . they had to take it apart to get it into the burial chamber. but it holds that it is indeterminate whether the ship that is buried with her is the ship on which she held court. and on other refinements it ceased to exist upon disassembly and a new ship was assembled from its parts. we are working with too crude a notion of concept refinement. And so by the supervaluational approach. but the proposal says that there is no truth-value for the claim that the desk in my study was originally purchased at a thrift shop-for this is true on some refinements of the concept of desk hut not on others. unfortunately. If I had to disassemhle my desk to get it into my study when I moved. We would like to say that she held court on exactly one ship during her reign. But. where it lies available for her afterlife journey. and it agrees that one ship is buried with her. However. there is no truthvalue for the claim that the ship on which she held court is the ship buried with her. For although it produces the correct answer in the ship-puzzle case.

Or something like this. So there were two thinking. plan. Here is a second objection. The ad hocness of an account that gets the right answers might be the outer appearance of a natural account that we have not yet been able to articulate. So for this to be a solution. This appears to be an ad hoc move. beings there at the same time. and makes commitments to other people-otherwise the proposal will end up attributing lack of truth-value to truisms such as that (normal) people think. and another on which a ship ceases to exist upon disassembly when another one continuous with it and resembling it gets new parts. An entity x is a refined-person if and only if there is a refinement of the concept of person such that. feeling. This may make the theory too abstract to evaluate. this can happen only if the refinements in the concept of person result in multiple potential referents for 'the person who entered' andlor 'the person who left'. It goes as follows: No matter how the notion of a person is refined. a is a person. a person had better be something that thinks. But we must be careful not to be too hasty with charges of ad hocness if the resulting account gets the answers right. Now consider the judgement in the disruption case: The person who entered is the person who left. but I will discuss it in application to the refinements of the sortal notion of a person.7 Then there were two refined-people in the same place at the same time before entering. On the supervaluational approach. It applies to all of the puzzles. 7 This is a technical notion. This needs to turn out to lack truth-value. and it is not at all clear how to generalize it. under that refinement. feel. there need to be divergent refinements of person such that 'the person who entered' ends up with divergent potential referents. plans.176 Alternatives to Indeterminate Identity reassembly when another ship continuous with and resembling the original gets new parts. . and make commitments. where the extended valuations are produced by refining our concepts in whatever way it takes to get the desired answers. So let us proceed with this idea: that our everyday judgements are all to be assessed in terms of supervaluational readings. A refined-person is not a person who is refined. Now let us call any object that is truly characterized by some refinement of the concept of person a "refined-person". and at least one of which did not. feels. at least one of which left. but we won't know if we don't try.

with a slight change of vocabulary (with 'person' replaced by an elaborate explanation of what all refined-persons have in common). The identity puzzles all reappear with 'person' replaced by 'thing that thinks. but I do have a refined-wife. At first glance. I may have several of them. We would like to say that the female being standing in the doorway is the person standing in the doorway. instead of person). Some such combinations may render true the statement: the female being in the doorway = the person in the doorway but many will not. and consider the female being standing in that doorway. What will they think?) The point is that the solution at hand demands an expanded ontology. . thinking. that is. In general. I speak loosely of "my wife". and one expanded with things that give rise to the original puzzle all over again. loves.'. But now we appear to have (at least) two different unrefined concepts to work with: female and person. this merely enhances the difficulty. if the person14 is identical with A 1 am indebted to Calvin Normore for helpful conversation regarding the material in the remainder of this chapter. Consider the person standing in that doorway. feeling beings. we get twice-many combinations of them. Since the difficulties above involve reconstructing identity puzzles with non-sortal concepts (thinking being. if she knows that I love many refinedpersons. There is a natural way to escape that objection. in fact. one obvious option is to consider refining these as well. How can I reassure my wife by telling her she is the only person I love. etc. given that there are many different refined-persons standing in the doorway. owes money on a mortgage. That is. R The proposal under discussion says that we need to consider refining our concepts. We are thus in danger of having a version of the traditional "many persons in the same place" solution.Alternatives to Indeterminate Identity 177 But this is just as objectionable as the traditional solution to the puzzle which held that there were two ordinary persons in the same place at the same time. since that is unrefined talk. and that J feel committed to each of them? (Of course. many female. If each of them resolves into many refined concepts. plans. corresponding to each of them will be many different refined-female beings. but we have really only considered refining our sortal concepts. But most of them will not correspond.

For example. When do we impose a constraint of this sort.178 Alternatives to Indeterminate Identity the female being14 and distinct from the person22. there is only one thinking thing in the doorway. This is controversial. and the supervaluational reading with the refinements providing the valuations will lack truth-value. as we did before. but it has some credence. Now we can refine person any way we like. Then we go back to just refining the sortals. we must limit the valuations to the "coordinated" ones. and so on. We also need to assume that a refined-person is female at a time if and only if every refined-person that coincides with it then is also female then. and how do we tell exactly which constraint to apply? J . however. we need to suppose that 'the female being in the doorway' means 'the female person in the doorway'. the above statement will come out supervaluationally true. a way to try to get this to work out. the second is false. such as person. if we also coordinate the refinements of 'think'. then one resolution will be: the female being14 in the doorway = the personl4 in the doorway but another will be: the female being14 m the doorway = the person22 in the doorway. so long as we refine uniformly. So suppose that our uses of a word like 'female' are coordinated with a sortal concept. Thus the apparent data is not preserved. on any given refinement there is only one. If we do this. For although there are many refined-people in the doorway. 'feel'. So if you say 'red thing'. The catch lies in the constraint we need to impose on 'female' so as to make the account work. Since the first of these is true. Then we just need to rule that in evaluating our utterances in terms of supervaluations. but we refine the implicit ones along with the explicit ones. It is sometimes speculated that any use of a non-sortal word as a noun in language presupposes some underlying sorta1. the underlying structure of our problem example is: the female person in the doorway = the person in the doorway. and that thinking thing is the female there. There is. Likewise. you must have in mind some sortal concept to flesh out 'thing'.

etc. and perhaps the category of restaurant is to be analysed as employer that serves food. 10. without any solid conclusions in principle about how they might be . I need to submit a W2 form for each of my employers. Of course. and imagine that I am a server in the original restaurant. for there are other equally good categories under which it fits: purchaser. and they are literally distinct from one another. Consider Stalnaker's restaurant case. This becomes even clearer if we consider how entities of different categories are to interact.. Ideally. but there might be a coordination of W2 form with employer. The category analysis is still only a sketch of a solution. However. So we conclude incorrectly that 1 need to fill out two W2 forms. I have only raised problems for how to work out various versions of alternative views.4 LEFT IN LIMBO . because the requirement to fill out one per employer is fulfilled no matter how we refine both W2 form and employer. This is because each refined-restaurant does literally employ me. for otherwise '1 work for the restaurant on the corner of Main and Elm' will not be true. sales-tax collector. Or do I? Perhaps 'W2 form' needs refinement. retail establishment. it is not reasonable to analyse restaurant as a subcategory of employer. and I must work for each of them. But clearly I have not. there would not be a direct coordination of W2 form and restaurant. Coordinating categories does not reduce the number in this case since the category of person under which I fall is not coordinated with the category of restaurant under which they fall. before the problematic split.Alternatives to Indeterminate Identity 179 am not sure how to do this. Now when 1 file my income taxes. so long as we coordinate the refinements. and perhaps it is coordinated with the category of restaurant. I would now be able to say that I have refuted the linguistic/semantic alternatives to the hypothesis that there is indeterminacy in the world. There are at least two refined-restaurants that employ me. Then I fill out one W2 form.1t thus appears that 1 must fill out (at least) two W2 forms. There must be two refinedrestaurants here in order for the identity claims to be indeterminate between the original restaurant and the later ones..

And so the worldly indeterminacy of identity view is deserving of close consideration.180 Alternatives to Indeterminate Identity altered and developed so as to take account of the problems. And so we are left in limbo. But I hope to have shown that there does not yet exist an alternative view about which we can be confident. .

and everything that is determinately not a member of set A is determinately not a member of set B. But they are also subject to an additional principle. whatever else they are like. or vice versa. This is meant in the strongest possible sense. and vice versa.11 Sets and Properties with Indeterminate Identity 11. So they are subject to all of the logical principles involving identity discussed elsewhere in this work. then A is determinately identical with B. This is a spelling out of a point that can be put more simply by saying that if A and B are sets. and vice versa. in that: (i) if everything that is determinately a member of set A is determinately a member of set B. then the truth-value status of the claim that they are identical: A=B should be exactly the same as the truth-value status of the claim that they have exactly the same members: .1 IDENTITY CONDITIONS FOR SETS Sets. (iii) otherwise it is indeterminate whether set A is identical to set B. What is most distinctive about sets-which distinguishes them from other sorts of things-is that their identity is determined by their membership. are the sorts of things that can be said to be identical to or distinct from one another. (ii) if something is determinately a member of set A but determinately not a member of set B. then A and Bare determinately distinct.

usually by stating which conditions have sets corresponding to them.182 Vx(x EA<=:} Sets with Indeterminate Identity X E B). and if y is determinately not a member of that set. We can formulate theories about either sort of entity. then we can expect that membership in a given set is one of the worldly properties in terms of which we characterize worldly identity. . This is equivalent to the claim that membership in a given set satisfies principle DDiff from Chapter 4. Worldly sets obey the principle that if some thing x is determinately a member of such a set. and consider worldly sets here. then we make no such assumption. If the former. A particular theory of sets will result by adding to [Set Essence] some principles that say which sets exist. and in the second. So I give it that title: [DDiff for E Set] !x E Z & !--.2 DDIFF FOR SET MEMBERSHIP Before considering which sets exist. In terms of a single principle. then x is distinct from y. or whether they are merely part of the conceptual apparatus in terms of which we view that world. I postpone a consideration of conceptual sets to §11. or whether talk of a thing being a member of a set is just about a kind of conceptual classification.x = y. The difference between these two conceptions of set is whether being a member of a set is itself a worldly property that a thing has. If the latter. The principle is: !x E Z & !--. I speak of worldly sets.y E Z ~ --. conceptual sets.x = y. but we need to be clear about the choice when developing the theory. this equivalence is true: [Set Essence] A =B <=:} Vx(x EA<=:} X E B). In the first case. 11.9.y E Z ~ --. This is whether sets are to be thought of as robust constituents of the world. we have another question to consider. I assume that any non-bivalent theory of sets will adopt this as a basic axiom.

4 INDETERMINATE IDENTITY FOR SETS If one wishes to combine such a set theory with a theory encompassing indeterminacy.(x = Y). [Comprehension] 3X'i1z(z X ~ <pz) where <p is any formula satisfying DDiff and not containing 'X' free. We assume the logical laws of identity for the identity sign between any size letters. then one must allow sets to be indeterminately identical to one another even if there is no indeterminacy of identity of individuals at all. [Set Essence] then entails that it is indeterminate whether the sets . This is because one can devise distinct defining conditions for sets such that it is indeterminate whether the defining conditions specify the same members or not. Suppose that we temporarily use small letters for variables over objects (and not over sets) and capital letters for variables over sets (and not over objects). along with the following axioms (with the understanding that they are all implicitly universally quantified): [Set Essence] [DDiff for E X= Y ~ 'i/z(z E X ~Z E E Y). Set] !x E Z & !-.y E Z ~ !-.3 SETS OF OBJECTS Suppose we wish to formulate a theory of objects and sets of objects. 11. with the understanding that sets do not themselves contain sets as members. [Sets aren't Objects] 'i/x'i/Y-.Sets with Indeterminate Identity 183 11.x = y. These assumptions all strike me as defensible for a theory of sets of objects. Such a theory is formulable by the adoption of the two principles discussed above together with a principle of set comprehension which says that any formula with a free variable over objects determines a set if it satisfies DDiff for set membership.

Suppose there is no indeterminacy of identity among objects at all. There is also a "pair set" of a and b. Suppose that it is indeterminate whether a = b. call it la].bl'.A = B. as in the proof just given. contradicting v<jla. Of course. Then we may show that it is indeterminate whether A = B by refuting both the claim that A = B and the claim that -. the set yielded by [Comprehension] applied to the formula 'z = a'. This is because any indeterminacy can be used to provide conditions to define set membership. indeterminate identity among objects also leads to indeterminate identity of sets. This set has both a and b as determi- . there is a formula <jlx true only of individuals and such that it is indeterminate whether <jla is true: v<jla Since there is no indeterminacy of identity of individuals. Thus. the mere existence of indeterminacy among objects generates indeterminate identities among sets. call it '[a. In particular it is indeterminate whether b E [al. and thus <jla. again contradicting v<jla. Our axioms entail the existence of the "unit set" of a. For. and thus -. we can show that a E A.184 Sets with indeterminate identity defined from those conditions are the same.<jla.A = B) lets us infer that there is some determinate member of B that is determinately not a member of A. i. this can only be a. As a result. anything that is indeterminately identical with a is such that it is indeterminate whether that thing is a member of [al. even if no objects are indeterminately identical at all. the identity 'A = B' is indeterminate. there are sets A and B such that: V'x(x V'x(x E E A B q q <jlx) <jlx v x = a) It is easy to show that a E B. assuming the first claim (A = B). This set has a as a determinate member and it has everything determinately distinct from a as a determinate nonmember. both '<jlx' and '<jlx v x = a' satisfy DDiff. And assuming the second claim (--. but there is some indeterminacy. Applying [Comprehension]. yielded by [Comprehension] applied to the formula 'z = a v Z = b'. Here is a sketch of a proof that addresses this phenomenon.e.

but not of the first. or vice versa.(la.b) that is determinately not a member of the other. by [Set Essence] From (4) There is no way to derive (3) from (1) and (2). since b is determinately a member of the second. and so on . So they are not determinately distinct.al) -.5 A ZERMELO-FRAENKEL-LIKE HIERARCHY OF SETS The logical power of set theory comes when we admit not only sets of objects. and sets of those sets. but there is no way to do that. we can show.a}. But they are also not determinately the same. b) that is not a member of {a.bl = {a. as we did above. the argument is: (1) (2) (3) (4) (5) 'l(a=b) -''l(a = a) -. that it is indeterminate whether {a. but also sets of sets. It is easy to see that there is nothing that is determinately a member of one of these sets: la) {a. So it is indeterminate whether they are the same: 'l(la] = (a. 11. whether individual objects are indeterminately identical or not. Let us now review Salmon's argument to the effect that set theory is incompatible with indeterminacy of identity of individuals.(a = b) -''l(a = b) Hypothesis to be disproved Truism From (1) and (2) [No!] From (3).b]). we would need to show that there is some member of {a. Instead.Sets with Indeterminate Identity 185 nate members.a}. In order to derive step (3) from [Set Essence].bl = {a. So if there is any indeterminacy at all there will be sets that are indeterminately identical. using comprehension. Changing ordered to unordered pairs (to pair-sets).

x . In it we suppose that sets are built up in stages from objects. then X and Y will be determinately of the same rank. which in the spirit of set theory are here called "individuals". just as in Zermelo-Fraenkel set theory. if it is indeterminate whether X = Y. The following theory is adapted from Woodruff and Parsons (forthcoming). Relaxing this constraint has far-reaching consequences. . . sets contain as members or indeterminate members only those things that exist at earlier stages. The theory that results from these assumptions is "neat". and it is determinate whether a given set is a member of a given rank. perhaps along with some individuals. We assume a primitive name 'I' for the set of individuals and '0' for the set with no determinate or indeterminate members. then it is determinate that y 1= z. Assumptions are made that yield a transfinite series of such stages. The two axioms above.. and if it is indeterminate whether X E Y.. such as unit sets that are two or more ranks above their members (see Woodruff and Parsons forthcoming: §2). At each stage. so ''I1X[ . Other sorts of assumptions might be made. I Our vocabulary includes at least the primitive two-place predicates 'E' and '='..]'. The other is that a set may not contain indeterminate members with ranks higher than the ranks of all of its determinate members. where 'ex ranks x' means that the ordinal ex is the rank of entity x. in that sets fall into ranks determined by the stage at which they are constructed. There are two constraints on this process that are automatic in the bivalent formulation of classical Zermelo-Fraenkel but that need articulating in our framework. X .. One is the DDiff condition for sets: if y is detenninately a member of X and z is determinately not a member of X. ]' is short for ''I1x[--. but they are not considered here. In order to formulate the rank restriction it is convenient to have a primitive two-place predicate 'ranks'.. then it is determinate that X falls into a rank immediately below that of Y. At stage 0 are sets whose members are only individuals. .. Further. I do not know how to develop a systematic theory of worldly sets without such a constraint. [Set Essence] and [DDIFF for I The "neat'" falling into definite ranks is a consequence of the principle that a set may not contain indeterminate members with ranks higher than the ranks of all of its determinate members. At stage 1 are sets whose members are themselves sets of individuals.186 Sets with Indeterminate Identity without limit. Hereafter we understand small letters to range over everything (over individuals and sets) and capital letters to be variables restricted to sets (to non-individuals).x El==> .

...0 E I & 'v'y-. The following structural axioms are straightforward from classical set theory. is the set whose determinate members are exactly x and y.y E 0.y E S. [Bivalence of Individualness] [Individuals Lack Members? [Empty Sets] x x E Iv -tX E I E El:::::> -. and whose indeterminate members are those things that are not determinately identical to x or to y but are indeterminately identical to x or to y: [Pairs] ::3S'v'u[u E S <=> u =X v u = y]. have indeterminate (individual) members. X].yj'... they. 2 This axiom together with the previous one entail that I is a set ( a non-individual): . denoted by 'Ix. . This yields an "emptiest set".::3yy E x -..y = y' entails the usual empty set axiom of ZF: ::3S'v'y-. Selecting '<!Jy' to be '-.Sets with Indeterminate Identity E 187 Set). along with the following are our axioms for IZFU ("Indeterminate Zermelo-Fraenkel set theory with Ur-elements").. The power set of a set is the set of subsets of that set.1 E 1.y E I & 'v'u(u EX:::::> U E y). I present the remaining axioms beginning with things of the lowest rank.::3y!<!Jy:::::> ::3S'v'y[y S <=> yE 1& <!Jy] [Empty Sets] generates sets with no determinate members and with arbitrary sets of individuals as indeterminate members.. there will be sets indeterminately identical to 0. Y =d[ -tX E I & -. unlike 0.. The pair set of x and y.. Then the power set axiom is: [Power Set] ::3S'v'z[z E S <=> Z <. a set with no determinate or indeterminate members: [Emptiest Set] -. 0. Define "subset" as follows: x <. The union of a set of sets is the set whose members are members of one of those sets: [Union] ::3S'v'Z[Z E S <=> ::3u(z E U & u E X)]. If there are any individuals at all.

So-defined. To complete the structuring of the ordinals we adopt: [Ordinal Non-Self-E] -. In order for the ordinals to be well ordered we take this as an axiom: [Least Ordinal] 3x[Ord(x) & !x E S] ~ 3x[Ord(x) & !x S & 'liy[Ord(y) & !y E S ~ x ::::y]]. Transitive sets are ones whose determinate members are determinate subsets: Transitive(x)=df-. To express ordinal comparison. E Hereafter we use small Greek letters to range over the ordinals. where the successor of Y is the set whose members are the members of Y along with Y itself: [Infinity] 3S[0E S & 'liY(!YE S~3Z[ZE S & 'liU(UE Z{::::}U=YvUE Y)])J.x E 1& 'liy[vy E X ~ 3z(!z E x & vy = z)]. As discussed above. X::::y=clfX<YVX=y.U E U.Ord(x). we call a set 'tight' if its indeterminate members are limited to those things that are indeterminately identical to some determinate member of the set: Tight(x) =df -. To formulate the rank restrictions. Ordinals are then defined as tight transitive sets whose determinate members are all tight and transitive: Ord(x) =df Tight(x) & Trans(x) & 'liy[!y E x ~ Tight(y) & Trans(y)]. .188 Sets with Indeterminate Identity The usual ZFaxiom of infinity assumes that there is a set containing 0 and containing the "successor" of any of its members. As a preliminary.xE 1& 'liY(!YE x~ !y~x). we define: x <y =df Ord(x) & Ord(y) & x E y. our Replacement scheme requires rank restrictions. being an ordinal is a bivalent condition: THEOREM: Ord(x) v -. we first define the ordinals.

x ranks Y THEOREM: a ranks x & p ranks x ~ a = p. As discussed earlier.X = Y ~ 3a[a ranks X & a ranks Y] P]. Then its range for domain S is a set: [Replacement] If: (i) u\JfY v -. THEOREM: 3a[ a ranks X] THEOREM: x ranks Y v -. and suppose that x\JfY "projects" from S a condition satisfying (iii) the rank constraint and (iv) DDIFF for set membership. THEOREM: YE X ~ 3a3p[a ranks Y & P ranks X & a < THEOREM: -. Pl Our final axiom scheme is [Replacement].3y(y E S & Y\Jfv) ~ -.Sets with Indeterminate Identity 189 The following axioms constrain rankings. . Z ranks X<=> z is the least ordinal such that \fy[3x(!x E X & y ranks x) ~ y < z].!-. In Woodruff and Parsons (forthcoming) the above axioms are shown to be consistent if classical bivalent Zermelo-Fraenkel set theory with ur-elements is consistent. Theorems. we posit that no indeterminate member of a set exceeds all the determinate members in rank: [Rank Limitation of Indeterminate Members] -.v E S & u\JfY & v\Jfz & U = v ~ Y = Z (iii) -.u E S & u\JfY ~ 3v3z[!v E S & v\Jfz & 3a3p[a ranks Y & pranks Z & a:s.!-. but they are stated here as axioms because 'ranks' is primitive notation: [Individuals are Not Ranked] [Ranking for Sets] x E I ~ -. The following are some useful theorems. The first two resemble parts of a familiar definition of ranking by recursion.u E S & -.!-.!-.x E X & a ranks x ~ 3y3P[!y E X & pranks y & a :s. P]] (iv) !3x(x E S & x\Jfu) & !-.!-.ll = v Then: 3XVy[y EX<=> 3z[z E S & z\JfY]]. Suppose we have (i) a bivalent relational formula x\JfY which (ii) is functional on a set S.y ranks x.u\JfY (ii) -.

Given a set Sand formula <jl. familiar techniques remain valid. There are complications. which can be irksome in practice. every set S has a "tightening": THEOREM: 3X[Tight(X) & 'v'z[!z E X ~ !Z E S]]. which correspond to the pure sets of ZF.3z(z y & Z EX)]. If we want a separated set and we care only about getting a set with a certain determinate membership. [Foundation] holds in the following form: THEOREM: 3yy EX=> 3y[-.:>: P]]. one usually works with a pure version of ZF. for considering the foundations of classical mathematics. then the constraints may be ignored. For any set S and formula <jl we have: THEOREM: [Determinate Tight Separation]: 3X[Tight(X) & 'v'z[!z E X ~ !(z E S & <jlz)]].y E X & -. Classical ZF is equivalent to a subtheory of IZFU. (Proof: Let the '<jlx' in [Separation] be '3y[!(y E S & <jly) & x = Yr.(Y E Z & <jlY) => 3x[!(x E Z & <jlx) & 3a3p[a ranks x & pranks Y & a .. . Define a hereditarily pure tight set of rank a as follows.!-. 11. For example. designed to accommodate both objects and sets that may be indeterminately identical. such as the constraints on Separation and Replacement. E Determinate tight separation: [Separation] is restricted by the DDIFF and rank constraints. but for familiar purposes.X = Y (ii) -.) Tightenings: As a corollary. within IZFU one can define the hereditarily pure tight sets. a separated set exists if <jl satisfies DDIFF with respect to the members of S and if the appropriate rank constraint is satisfied: THEOREM: [Separation] If (i) !(x E Z & <jlx) & !-. then 3S'v'x[x E S ~ X E Z & <jlx].6 RECOVERING CLASSICAL ZF The above is the full general theory.190 Sets with Indeterminate Identity [Separation] follows from [Replacement].(y E Z & <jly) => -.!-. TIlat is.

Most importantly.. the axioms given above all hold. [0}. [0.aj. So one can simply take over all classical results of ZF into this theory.c} and vc E {b.a = c. [0] } }. if 'x ~ y' is redefined as 'Vz(HTP(z) ~ (z E X ~ Z E then the power set axiom holds. the restrictions on [Separation] and [Replacement] concerning DDIFF and ranks vanish.d) <=> a = c & b = d. and yields an HPT set. this does not satisfy lhe basic identity conditions for ordered pairs. The usual account is to represent the ordered pair (x. and 'ranks'. However. {0]] }. so long as we have the right construal of ordered pairs. [0. '='.. [10}.[x. [0. We can do the same here. suppose that our primitive predicates are limited to 'E'.y}}. the first few are: One of rank 0: One of rank 1: Two of rank 2: Four of rank 3: etc. {0. 7 RELATIONS In set theory. and yield HPT sets.y) = [ {x}. {0}] } y»' 1l.Sets with Indeterminate Identity HPT. [0] }}. One can easily establish by induction that both identity and membership are bivalent relations between HPT sets. For then it follows that 'la E Ib. The hereditarily pure tight sets are then given by: HPT(x) =df 30.b) = (c.(x) <=> Tight(x) & Vy[!y E 191 x ~ 3~[~ < a & HPTp(y))).y) as follows: (x. For example. Further. o [0} [0]. Then if all quantifiers are relativized to HPT sets. The HPT sets resemble the usual ones from ZF. This fails when we have 'la = b & vc = b & !--.: HPT. relations are usually represented as sets of ordered pairs. given by: (a. { [0. and so [Set Essence] entails .(x). in our bivalent framework. [0. [0]] [0.

a) = (b. [x. the identity conditions for ordered pairs are satisfied.192 Sets with Indeterminate Identity that 'lla.Z)] 'v'x[x E r ~ !OP(x)] =df Reln(r) =dr Domain(r.y) E r.x) Range(r. So we do have: Thm: !Reln(r) & 'v'x'v'y[(x. On this construction. if r is a relation and x an individual or set then there may be no set whose elements are defined as being those things y such that (x.y) S ~ (x.X) In the set theory developed here it is not always possible to produce a set from a relation by "plugging up" one of its places with an individual or a set. we need some additional definitions: OP(x) =dr ::3y::3z[x = (y.a = c. But by the proposed definition of ordered pairs we could then infer that 'le (b.[a. However. and so we may always produce a set from a relation in the described manner.z) Er].1.b} J = I [bJ. In §3. y. .Another application of [Set Essence] then yields that '11 [b}. This is wrong.b} = /c. This is because such a putative set may violate the condition called [Rank Limitation of Indeterminate Members].x) !Reln(r) & 'v'Z[Z E x ~ ::3y«Z. To show this.c».y) =df [x.[c. since !--. if r is a relation limited entirely to individuals. it is impossible to violate this condition.1 we assumed that if r is any relation and x any object.b}.z) Er)] We can then prove: Thm: !Reln(r) Thm: !Reln(r) ~ ~ ::3XDomain(r. there is a property got by "plugging up" r with x: the property of bearing r to x.bJ}.y) Er)] =df !Reln(r) & 'v'z[z E x ~ ::3y«y. E r~x E J & Y E I] ~ 3S'v'z[z E This is the principle that was appealed to in §3. that is.X) ::3XRange(r.ly} JJ. This is validated in our framework for relations among individuals. A solution is to use a slightly more complicated definition of ordered pairs: (x.

in the sense that if properties p and q determinately hold of exactly the same objects.2. If we do this. Then the theory developed above is a theory of objects and their properties.4. since membership in them is not expected to make a difference to the identity of their members. Their theory is a simple variation of the theory of worldly sets. (This is in addition to the principle that properties are identical if they themselves have the same properties. but that may be because they are thinking of conceptual properties.) It also assumes the principle of plenitude for properties discussed in Chapter 7.This then lets us eliminate the restriction in [Comprehension] (in the theory of objects and sets) or in [Replacement] (in the full theory of the hierarchy of sets) requiring that (j) satisfy DDIFF. . They may. then I am not aware of any compelling argument to the effect that worldly properties are not ex tensional. we can eliminate [DDiff E Set] from the set of axioms governingthem. and let us temporarily use the notation 'x E Y' to mean that the individual object x has the property Y. then these sets cannot play the role of properties in the Leibnizian account of identity. however.8 CONCEPTUAL SETS 193 We postponed the treatment of conceptual sets from § 11. be identified with the conceptual properties mentioned in §4. If we distinguish between conceptual properties and worldly properties. Suppose that they are. they are distinct.9 SETS AS PROPERTIES? Suppose that we think that worldly properties are extensional. and determinately do not hold of exactly the same objects.Sets with Indeterminate Identity 11. it is indeterminate whether they are identical. with the additional thesis that properties are identical if they are possessed by the same objects. (And if one of them determinately holds of an object that the other determinately does not hold of. And otherwise. In particular.) People usually do not consider properties to be extensional in this sense. 11. It is a formalization of the framework of the world put forth in Chapters 2 and 3. they are identical.

. none of the earlier discussion turns on the question of whether or not properties are extensional.194 Sets with Indeterminate Identity If properties are not extensional. then the set theory described above can be weakened to a theory of properties by changing the major connective of Set Essence from a biconditional into a conditional. So far as I can see.

But suppose. not metaphysical indeterminacy. and one might want to appeal to higher-order indeterminacy here. Of course there are puzzles in which we are confused about whether or not they have answers. instead. But certain of the others seem to turn on matters of degree. But this may be only epislemic uncertainty. such as whether I am or am not identical with my body.12 Higher-Order Indeterminacy ] 2.1 WHAT IS HIGHER-ORDER INDETERMINACY? Is there such a thing as "higher-order" indeterminacy? In the context of the present book. Then perhaps you will be . "false") is such that it is indeterminate whether it is correct. Is there any persuasive case to be made for meaningful questions of identity such that it is metaphysically unsettled whether they have answers? I see no natural application of this idea to certain of the questions discussed earlier. "false". or "indeterminate" is correct)? If so.e. and suppose that the newly assembled ship is made of 80 per cent of the original planks. and some do not: each answer to a puzzle of the latter kind ("true". it is indeterminacy about indeterminacy. then there is no answer regarding which later ship is identical to the original ship. and if 95 per cent of the old planks were used to construct the newly assembled ship. whether any of "true". this seems to be a kind of higher-order indeterminacy. Now are there identity puzzles of which we wish to say that it is indeterminate whether they have answers (i. Suppose that in the ship case you are sure that if the ship with new parts very much resembles the old ship. the question arises in the following form: Some identity puzzles have answers. with the other 10 per cent thrown away. that 10 per cent of the planks are retained in the ship with (mostly) new parts. So indeterminacy arises because of questions that have no answers.

false. So we need to examine ways around this "collapse". then it seems to follow that it is indeterminate whether S is true or false. and there is no additional higher-order indeterminacy after all. Suppose that we have a case in which we are tempted to say that it is indeterminate whether S is true. or indeterminate. we must decide what status to attribute to a proposition such that it is indeterminate whether it is true. Let me distinguish this additional status with the term 'unsettledness'. not to dispute its existence. And so the same sort of consideration that drove us originally to think that sometimes there are no answers to identity questions might lead us to think that sometimes no answer is determined as to whether we have an example of that sort. without limit. you might think that this is because of how the world is (or is not). So there must be some additional status in addition to: true. our task here is to make sense of second-order indeterminacy. there is a conceptual dilemma that needs to be addressed. false. false. So higher-order indeterminacy collapses to ordinary indeterminacy. so it is worth investigating what higher-order indeterminacy would be like.196 Higher-Order Indeterminacy uncertain whether or not this is a case in which there is no answer. then third-order indeterminacy will follow as well. and so on. and thus it seems to follow that S is (simply) indeterminate. If you think that no additional information will help. If it is indeterminate whether S is true. and indeterminate. In this chapter I explore how to extend the theory of the previous chapters to include higher-order indeterminacy. In particular. or indeterminate. One thing is clear: if we are driven in this way to second-order indeterminacy. A proposition is unsettled if . and one that I see as much more difficult to address than indeterminacy). as argued above. Although I suspect that this may be the right answer. It seems that this status cannot be simple indeterminacy. because then there is no higher-order indeterminacy. there is instead epistemic uncertainty about whether there is determinacy-or else we may be dealing with a case of linguistic vagueness (which is a quite different phenomenon. However it is hard to be certain about this. Where there appears to be higher-order indeterminacy. with only increasing bewilderment to cloud our minds to this consequence. And fourth-order. or false. or indeterminate. I suspect that there is no such thing as higher-order indeterminacy. First. So we will then have arbitrarily high levels of indeterminacy.

This alternative seems odd to me. FALSEHOOD.Higher-Order Indeterminacy 197 there is no answer to the question whether it is true. Again. or false. or indeterminate. An alternative might be to see second-order indeterminacy as compatible with first-order determinacy (or indeterminacy). If the relevant state of affairs is settled. four mutually exclusive truth-statuses (though perhaps 'truth-status' is no longer appropriate. though perhaps there is (or is not) an answer to the question. AND INDETERMINACY To keep the first-order and second-order notions separate. a proposition may be unsettled and also have anyone of the statuses true. I discuss it in the second section following. one might hold that it is indeterminate whether there is an answer to the question.2 UNSETTLED NESS AS A STATUS DISJOINT FROM TRUTH. This idea is explored in the next section. there are now simply four statuses): true false undetermined unsettled . or false. false. That is. Our old indeterminacy must now be a status such as "definite indeterminacy". (We can no longer consider simple indeterminacy a case of having no truthvalue. or (simply) indeterminate. since that exhausts the options and leaves no room for levels of indeterminacy. but it is a coherent one. with unsettled ne ss a kind of "indefinite indeterminacy". then a proposition describing it will be true. or indeterminate.) There are now. in effect. 12. let me continue to speak of the world determining (or not determining) whether a state of affairs holds or not. and so a proposition which is unsettled is thereby not either true. or false. otherwise it is unsettled. using 'unsettledness' for second-order indeterminacy. and let me speak of it being settled or not settled in each given case whether the world does or does not determine whether a state of affairs holds. or indeterminate.

and say: TABLE 12. but with the intent that' !5' is unsettled if 5 itself is unsettled.198 Higher-Order Indeterminacy A proposition is second-order undetermined iff it is unsettled. Identity must now mean coincidence in all respects. Our notation now needs to have its meaning extended so as to address the new status. with the intent that '§5' is false if 5 is unsettled. and otherwise unsettled. and that a conjunction (disjunction) with an unsettled part is false (true) if the other part is false (true). I will use '§5' to mean that it is settled that 5 is true. for the sake of definiteness I make some particular choices about notation. I skip discussing conditionals for the reasons given earlier: they introduce complexity without addressing any new metaphysical Issues. determinate truth settled truth scttlcdncss indeterminacy S T !S T §S T "S F T §S v §--. The quantifiers will again be generalizations of conjunction and disjunction. In parallel with previous notation. and that the negation of an unsettled sentence is itself unsettled. some relevant truth-conditions are given in Table 12.S §"S T T T v F u F F u F F F F u F . Then we use our determinacy notation: '!5' to mean that 5 is determinately true. If it is unsettled. we need to specify what status its negation has. and indeterminate (has status '-') if the other part is indeterminate. I suppose that if a proposition is settled. then our language works as before. Let us say that our other connectives (conjunction. disjunction. and so on. Using 'u' for the unsettled status. We must assume that there are more ways in which properties are had by objects. With these two notions we can express the other natural options.1. There is a certain arbitrariness of formulation here. and negation) retain their old values when their constituents have settled statuses (when they have statuses other than u).1.

and for any property P.. Introduce n ll . So although some material of the preceding chapters now needs fine-tuning. and so on. . none of the major theses is lost as a result of admitting higher-order indeterminacy in the form of unsettledness.. unsettledness). We now have a logic that incorporates second-order indeterminacy (i. they need to be addressed from scratch. . if there are interesting identity puzzles that turn on unsettledness. it is settled whether a possesses P iff it is settled whether b possesses P It is unsettled whether object a is identical to object b iff for any property P: it is neither true nor false nor indeterminate whether a is identical with b. and U2 the same as u above. One needs to have n truth-value statuses in addition to T and F. u with Ul being the same as our earlier -.e. or vice versa It is indeterminate whether object a is identical to object b iff a is neither (definitely) identical with b nor (definitely) not identical with b. needs revising. most of our natural inferences from premisses are unaffected by the introduction of unsettledness. and contrapositive versions of it are not generally valid. . then our old argumentation is totally unaltered. .Higher-Order Indeterminacy 199 Object a is (definitely) identical to object b iff for any property P: it is true that a possesses P iff it is true that b possesses P it is false that a possesses P iff it is false that b possesses P it is indeterminate whether a possesses P iff it is indeterminate whether b possesses P it is unsetlled whether a possesses P iff it is unsettled whether b possesses P Object a is (definitely) not identical to object b iff for some property P: it is true that a possesses P and it is false that b possesses P. Since assertion is still assertion of truth.) This approach can be generalized to a theory with any number n of levels of indeterminacy. (Of course..A still entails B. Whatever you do. call these Ub . Leibniz's Law is still valid. For example. (are true or false or indeterminate). naturally. But if you add premisses to the effect that all sentences used in the proof are settled. A v B together with . Indirect proof.

S. F to F. as discussed in earlier chapters. Ui to itself if i > k.200 Higher-Order Indeterminacy determinate-truth connectives §!. let me continue to speak of the primary mind (or world) determining (or not determining) whether a state of affairs holds or not. with definite truth. k. Suppose that we believe in indeterminacy because we are idealists: some kind of mind (perhaps our own. for example.. If n = ffi. other things false. In the latter case it is second-order indeterminate whether p is determinate. for each proposition p. or decides to let the primary mind determine whether or not p is to be true.) We call a proposition unsettled if it does not have a truth-value in the first world. either decides that p is to be true. or decides that p is to be false (by deciding that its negation is to be true). = ! and §2 = § from then there can be indeterminacy of every finite level. ) makes some things true. and let me . or neither. Then the primary mind gets to produce a partial resolution of that world. the primary mind gets to make determinate some (or all) of the states of affairs left indeterminate by the superior mind. To keep the first -order and second-order notions separate. Each §k maps TtoT.. Can we raise the question of its being indeterminate whether something. . (So a partial resolution differs from a "resolution" in that not all states of affairs need to be made determinate.3 UNSETTLEDNESS AS COMPATIBLE WITH OTHER STATUSES It will help if we have an intuitive picture for a kind of unsettledness that is compatible. that is. . In particular. §'" with §! above. and indeterminate if it lacks a truth-value in the second world. 12. or perhaps God's or . we easily can.4. independent of whether or not S is indeterminate? Yes. U i to F if i $.. if we presume a superior mind which. or false.. as discussed in §5. and leaves the rest open (indeterminate). suppose that the superior mind produces a (possibly) partially indeterminate world. is made determinate.

in effect. and pre-emptively false if both parts are true (pre-emptively or non-pre-emptively). or indeterminate. Notation for the five statuses: pre-emptive truth: pre-emptive falsehood: ordinary truth: ordinary falsehood: lack of truth-value: pT pF T F Our linguistic notation needs to have its meaning extended. There are now. I will introduce '(.1' for 'it is pre-emptively true that'. Otherwise it is non-pre-emptively true if either part is non-pre-emtively true. Such a proposition may be true. it is non-pre-emptively true if each part is true or preemptively true.Higher-Order Indeterminacy 201 speak of the superior mind bestowing pre-emptive truth or preemptive falsehood on propositions. and preemptively false if both parts are pre-emptively false. and pre-emptively false if either part is pre-emptively false. and I shall interpret determinate truth as either truth or pre-emptive truth. If neither of these conditions apply. The quantifiers are to be the natural . and false if at least one part is false. The truthconditions will then be as in Table 12. It is natural to say that a conjunction is pre-emptively true if both its parts are pre-emptively true.2. or false. five truth-statuses: Pre-emptive status: pre-emptive truth pre-emptive falsehood Non-pre-emptive status: (non-pre-emptive) truth the superior being decrees truth the superior being decrees falsehood the subordinate being decrees truth the subordinate being decrees falsehood the subordinate being does not decide (non-pre-emptive) falsehood no value A proposltJon is second-order undetermined iff it is neither pre-emptively true nor false. A disjunction is pre-emptively true if either part is pre-emptively true.

Leibniz's Law is still valid. (pre-emptive or non-pre-emptive) most of our natural inferences from premisses are unaffected by the introduction of the new level. Again. As in the preceding section.2. and non-pre-emptive complete coincidence. A v B together with -. 1 skip discussing conditionals. If assertion is still assertion of some kind of truth. either P pre-emptively holds of both a and b.202 Higher-Order Indeterminacy TABLE 12. But now that we have two kinds of truth. so that identity means complete coincidence of statuses of possession of properties. although some material of earlier chapters now needs fine-tuning. Truth in general (pre-emptive or non-pre-emptive) will mean that for each property P. none of the major theses is lost as a result of admitting higher-order indeterminacy in the form of lack . The most natural treatment is to do so in a fashion parallel to that of the last section. We again have a logic that incorporates second-order indeterminacy (lack of pre-emptiveness). Indirect proof naturally needs revising. The remaining conditions should parallel those of the last section. and contrapositive versions of it are not generally valid. we will have two kinds of complete coincidence: pre-emptive complete coincidence. We must now characterize identity. For example. and so on. an identity will be pre-emptively true iff for each property P. or it pre-emptively fails to hold of either. S pT pF T F pre-emptive truth pS pT pF pF pF pF determinate truth !S pT pF T F F pre-emptiveness pS v p~S pT pT pF pF pF determinacy !Sv !~S pT pT T T F negation ~S pF pT F T generalizations of conjunction and disjunction. Informally. But if you add premisses to the effect that all sentences used in the proof are true or false or indeterminate. the status of whether P holds of a is the same as the status of whether P holds of b.A still entails B. then the old argumentation is recovered unaltered.

. TIl and F j . with tJI = ! and tJz = tJ from above..to Fb Ti to itself if i ::::: k. This is because if there is no higher-order indeterminacy.. And we add n pre-emptive-truth connectives tJ]. and likewise for F j • 12.4 CONCLUSION I hope at least to have made a case that the possibility of higher- order indeterminacy does not threaten the theorizing of the rest of this book.. call these Tj. the theory may be adapted in one of the ways given above. TJ = T above and Tz = pT above. for example. tJ. . .Higher-Order Indeterminacy 203 of pre-emptive determinacy. and if there is. . . if there are interesting identity puzzles that turn on lack of pre-emptive determinacy. Each tJk maps .. then the theory is unaffected.. Fm where. . . and likewise for F j Ti to Fk if i < k. (Also. they need to be addressed from scratch. • • .) To generalize this account to a theory with any number n of levels of indeterminacy one needs n levels of pre-emptive truth and falsehood.

Most commentators' have understood this to be a claim that the identity statement 'a = b' has no truth-value (or has some value distinct from truth or falsity). 1987. Parsons and Woodruff (1995). The "non-standard" interpretation was first suggested to me by David Wiggins. I indicated that Evans (1978) is subject to various interpretations. Noonan (1990).' On this interpretation. and Evans meant to argue that this cannot happen if the terms 'a' and ob' actually refer to objects. Several authors have written commentaries that do not address this issue specifically. van Inwagen (1988). J Personal communication. Woodruff and Parsons (1997). but rejects this as a reasonable option (ibid. Parsons (1987). among others: Broome (1984). Pellctier (1989) quotes David Lewis as reporting a communica- . and concludes that the resulting view is "strange". Lewis (1988). inde. THE ISSUE Evans (1978) proposed to refute the claim: v(a = b). Cook (1986). I will call this the "standard interpretation" of Evans's paper. 3 It is not much endorsed in print. and false otherwise. Johnson (1989). I am indebted here to Peter Woodruff (as nsnal). and to Peter Smith and an anonymous referee for Analysis for criticisms of an earlier flawed account of these matters. Gibbons (1982) interprets Evans in this way. Thomasson (1982). So the truth of'v(a = b)' amounts to the lack of truth-value of ' a = b'. Garrelt (1988). Cowles (1994). including ones irrelevant to the topic of this book. 485-6).' 1.APPENDIX Evans on Indeterminacy: An Unorthodox Interpretation In Chapter 4. 1 Including. Pelletier (1989: 482) considers the possibility that indeterminacy might be consistent with having a truth-value. . The purpose of this appendix is to describe such an interpretation and explain why it is plausible that this is what Evans had in mind. which he reads metalinguistically as "the identity statement 'a = b' is of indeterminate truth value". Keefe (1995). the sentential operator 'v' combines with a sentence to produce a larger sentence that is true if the ingredient sentence lacks truth-value.

' in (3).-. Evans explains how to go on to complete the proof. indeterminate truths are a subset of the truths. To take an illustration wholly unrelated to identity.h[v(a =x)]a -.. They seem to suggest that Evans finds plausible the modal principles of system S5 of modal logic with '".(a = h). This is not a truism. These remarks have puzzled almost everyone who has read them. if 'v' were the dual of' ".. Recall the central argument that occupied us in Chapter 4. 5 Evans uses 'definitely' and 'determinately' as synonyms. But then consider line (3): (3) -. we conclude that -. (1)-(4) and. Further... It is a reductio in five steps: (1) (2) (3) (4) (5) v(a=b) h[v(a =x)]b -.(a = b) which is straightforwardly inconsistent with (1). enabling us to derive (5') "..v(a = a) Truism This "truism" asserts the analogue of "it is not possible that a = a".5 though it is either true or false.' then we could replace 'v' by '-.. But how? Since line (5) does not explicitly contradict (I).(a = b) The hypothesis to be refuted Abstraction from (1) Truism Reverse abstraction from (3) (2). Pelletier argues that this option is.') generate a modal logic as strong as SS. presumably.. it's the opposite of one.v(a = a) -. .' playing the role of necessity and 'v' the role of possibility.Evans on Indeterminacy 205 terminacy is not a lack of truth-value. in essence. (4) by Leibniz's Law The point of this proof is that from the assumption that it is indeterminate whether a = b. According to Evans we thereby "[contradict] the assumption from which we began". likewise if indeterminacy means lack of knowledge.) On this interpretation 'vS' means that S is neither definitely true nor definitely false.-.. getting: tion fTOm Evans saying that he intended his signs for determinacy llnd indeterminacy as modal ones. Leibniz's Law. an error on Evans's part. without thereby denying that it is either true that it is raining or false that it is raining. this is a claim that many have found plausihIe. (If indeterminacy is contingency. He says: If 'Indefinitely' and its dual 'Definitely' C". one might hold that it is not determinate whether it is raining. may each be strengthened with a 'Definitely' prefix.

= a) And if 't>' plays the role of necessity in modal logic.(a and we could derive this absurdity from a truism alone. Leibniz's Law. Evans may have been confused. and Evans confirmed it himself later. Indeterminacy is definable in this logic as: Indeterminacy is thus an analogue of contingency. in which case an interpretation making good sense of the article may falsify it. combines with a sentence having truth-value. The explanation is simply that Evans made a slip.TIle latter is the dual of the connective '\7'. Several commentators have speculated that this is so. using a sideways triangle for 'definitely'. Of course. unlike 'I'. not of possibility. (1)-(4) and. making a true sentence with some of these and a false one with others.(a Appendix = a) or (3") t>-. this would entail (3"') -. it is still of interest to see if there is an interpretation that makes good sense. Let us take that possibility for granted. 6 But exactly what slip? He seems to have replaced an appeal to 'it is determinate thal S' with 'it is determinate whether S'. What he intended appears to be this.(a=b) which is straightforwardly inconsistent with (1). except that Evans's symbol is a modal operator that.-. The rightwards triangle '~' for 'definitely' is analogous to our determinately true connective 'I'. but the former is what is needed for his remarks to make sense. If 'not definitely not' and its dual 'Definitely' ('~') generate a modal logic as strong as SS. without even using the hypothesis that is supposed to be refuted. enabling us to derive (5') ~-. reported in Pelletier (1989: 482).206 (3') -. 6 In a letter to David Lewis. may each be strengthened with a 'Definitely' prefix. Why is this interpretation better than the standard one? Because it makes the argumentation of the article make better sense than on the standard reading.(a = a). presumably. . Evans then supposes a modal logic in which definite truth (symbolized with '~') plays the role of necessary truth. Clearly something is wrong here.t>-.

On the standard interpretation. does not do this. he implies that a more complex justification is needed.) 3. one that appeals to a fairly elaborate set of modal principles. A typical modal operator combines with truth-valued sentences to produce truth-valued sentences. Evans stated later' that he intended the indeterminacy symbols in the original argument as modal operators. not with sentences that lack truth-value altogether. on the standard interpretation. This clearly assumes that the reader will know what S5 is. there is a very simple extension of the proof from step (5) to a step that completes the reductio: infer 'o--. and he writes as if to do so is problematic. perhaps already implicit in the fourth.(a = b)' from (5). the language that Evans is using is bivalent. including the reduction principles of S5. on the standard interpretation. on the nonstandard interpretation. But Evans does not do this. The first difficulty is that Evans cites Leibniz's Law while using a contrapositive version of it. Why? The fourth difficulty is that Evans appeals to a modal extension of the proof by alluding to a modal logic "as strong as S5". First. This is hardly something that any writer would expect readers to be familiar with. But the determinacy connective. the proof is given in a non-bivalent language. The second difficulty is the unjustified assumption (emphasized in Chapter 4) that the abstracts introduced in the proof stand for properties that obey the abstraction principles and also stand for the real sorts of properties in terms of which identity is defined. But on the standard interpretation.Evans on Indeterminacy 2. he is reasoning within a non-bivalent framework in which it is a fallacy to appeal to a principle to justify a contrapositive application of it. And a typical modal operator will combine with some true sentences to produce truths and with others to produce falsehoods. (There is a fifth difficulty. DIFFICULTIES WITH THE STANDARD INTERPRETATION 207 There are four difficulties with the standard interpretation. and thus one needs to know what a formulation of S5 in non-bivalent logic is. SOLUTIONS TO THE DIFFICULTIES The non-standard interpretation avoids these difficulties. 7 According 10 Pelletier (1989). . The third difficulty is that.

... Second.(a '" b)' from '. So why are the abstracts there at all? Answer: They are there to make explicit that Leibniz's Law does indeed apply to the terms in question.'.. it is a minor infelicity in exposition-there is no logical error at all. For this reason.. the abstraction steps show this by removing the names from the scopes of the operators. and has a special role in the proof.(a = b)' in Chapter 4 depended on interpreting '!' as the determinate-truth connective.. Then why does he say that line (5) contradicts line (1)? Because it does! But not because any old sentence of the form'S' contradicts '.. it is going to take some explanation of why the proof shows that line (5) contradicts line (1).. so there is no fallacy. then the abstraction steps are completely justified (as argued in Chapter 4). why not infer '. what about the abstracts? On the non-standard interpretation. and I am not sure which Evans had in mind. Third.. So even though Evans does cite Leibniz's Law while using a contrapositive version of it.. On this interpretation. For (5) can be inferred directly from (1) and (3) by the (valid) contrapositive version of Leibniz's Law. in bivalent logic the contrapositive version is entailed by the Law itself.. if a principle is valid. This can be done in one of two ways...S' is consistent in his logic. it is possession of truth-value but in a certain way. The second option to suppose that Evans had in mind the kind of property called conceptual in Chapter 4. Since the names are supposed to be rigid. recall that the reason for inferring '!. they could be skipped completely so long as Leibniz's Law itself is thought to apply to all singular terms. But Evans's determinacy operator. And that is why Evans gives such an explanation. so is its contrapositive.. the point of the reformulation could be to show (claim) that it is not relevant that the names are within the scopes of modal operators.(a = b)' from '.S·.. First.... '... being within the scope of a modal operator is irrelevant to the appeal to Leibniz's Law later on. And since the logic is now bivalent.. If this is an error at all..(a = b)' and be done with it? Well. In a bivalent logic. the abstraction steps are redundant... the question of whether the abstracts refer to properties is not germane. which combines truly with any true sentence.208 Appendix Indeterminacy is not lack of truth-value.. the use of Leibniz's Law does not depend on what kind of property the abstracts stand for. the abstraction is rather intended to give an equivalent form in which the names in question occur in overtly extensional contexts. does not combine truly with any true sentence. couched .. and occur de re. line (5) is a special case of this pattern. So Evans does not have available this rule of inference: S This is because the form'S & . If this is so.. it combines truly with some and falsely with others.

it is completely correct as a proof.~S [Interchange] Definite equivalents may be interchanged anywhere. The reasoning just sketched is not obvious. He doesn't give the proof.S & OS' is generally a consistent form.. it involves prefixing "definitely" operators to the lines of the original proof. but he sketches the strategy.... Indeed.. because r have eliminated the abstraction steps. The idea behind the proof is easy to explain with an analogy from modal logic.. Does that or does it not refute the possibility of'S'? It does.. a correct proof can be given along the lines he sketches... It is slightly simpler than what he suggests.5' from 'OS'. In fact. '. then'S' is not possible.. it shows that (5) contradicts (1) because (5) must contradict (1) if it logically follows from (1). .. '"S' is equivalent to '~"S'. Suppose we start with an assumption that '5 is possible'.~S'. [NecessitationJ If'S' is proved from logical principles alone. His technique works fine if one takes for granted certain derived principles of SS.t>"S' & [SS-Reduction] Every sentence that begins with an operator (or the negation of such a sentence) is definitely equivalent to the result of prefixing that sentence with '~'.. since '-.. The proof is supposed to show that step (5) contradicts step (1).... They are these: Principles from ss: '. and that 'a' and 'b' are examples of such names.. using a modal logic with the principles of SS. If the possibility of S entails the falsehood of'S'. that does refute 'OS'.. you may infer .. and the proof already given in (1) . and so Evans shows how to prove it from more basic principles. [Definition] '"S' is (definitely) equivalent by definition to -. (5) has shown precisely that.. [Rule-Necessitation) If'A' follows logically from 'B' and 'C'.S' is equivalent to '~.. And that is Evans's strategy in his proof.g....Evans on Indeterminacy 209 in terms of a sketch of how to expand the original proof so as to show this. this assumes that Leibniz's Law applies in an unrestricted manner to precise rigid names. but for a subtle reason... But if you derive '. which thus avoids the fourth difficulty mentioned above. we will return in §3 to question whether it is reasonable for him to presume that the SS principles hold. At least this is so in the modal system SS. Here is a proof that results by following Evans's directions. and then we derive ' . E.5' does not in general contradict 'OS'. using only principles of modal logic...) The proof is within a bivalent logic.S'. and '.s' from this. The proof does not show that (5) contradicts (1) because of the way their operators relate to one another.. then '~A' follows logically from '~B' and '~C.. (That is.. substituting 'definitely' for 'necessarily'.

So this proof is an implementation of the sketch that Evans gives.B].• (a=a) 5. and (5) by the 'definitely' operator. • ~(a=b) .-''l(a = a) 7. (3).o-.(a = b) 9. .• -.~v(a = a)::::> • ~(a = b) 7. v(a=b)::::>l~v(a=a)::::>~(a=b)] 6c .• 'l(a = b) 3.(a = b) Appendix The hypothesis to be refuted From 1 by S5. From 4 by definition of ''1' and propositional logic From 5 by necessitation From 2 and 6 by the rule-necessitation of the contrapositive of Leibniz's Laws From 1 by definition of ''1'. But which views do this? 8 In case this step looks fishy. . • [v (a = b)::::> [~v(a = a)::::> ~(a = b)]} 6d. 4. and it does reach a full reductio by employing principles taken directly from SS. 4.210 1.' (a = b) & -. it provides a formal refutation of any view that accepts the premisses and rules of proof that are used. and 6 are.• [~v(a = a)::::> ~ (a = b)] 6f. and the new steps 2. the results of prefixing the old (1).[A::::> B]::::> [oA::::> . -''l(a = a) 6. here is how to get it from Leibniz's Law alone.Reduction Logically true Necessitation from 3. contraposing by propositionallogic From 6b by necessitation From 6c by the S5 principle for conditionals From 2 and 6d by modus ponens From 6e by the S5 principle for conditionals From 6 and 6f by modus ponens 6b. . -.(a = b) & -.'-' (a = b) The original steps (1) and (3) are retained here. From 7 and 8 by propositionallogic 8. We have: 2. as Evans says. without appeal to any derived principles. 'l(a=b) 2 .~v(a = a) We get to 7 as follows: 6a.• -. ov(a = b)::::> . WHAT DOES THIS ACCOMPLISH? What does Evans's paper show if it is interpreted as above? Clearly.[~v(a = a)::::> ~(a = b)] 6e . a=b::::> [v(a= b)::::>v(a=a)] An instance of Leibniz's Law in bivalent logic From 6a. • v(a=b) 6. a =a 4 . The additional steps just spell out some of the additional details. by using the S5 principle for conditionals: .

First. Second. This is because the argument is not relevant to it. the reduction rules of S5 would not apply to a logic of vagueness. at least. The view discussed in this book holds that indeterminacy in a state of affairs results in lack of truth-value of any sentence articulating that state of affairs. Williamson (1994) holds that all meaningful statements have truth-values.. since you should not be able to infer '. It is a viable option (though certainly not inevitable) to think that vague sentences have truth-values. the case against the S5 reduction principles is not that firm. and it results in a contradiction. and some are vague and some not. Then every step in the proof is valid. the negation of line 2 follows from line 1. for two reasons. If so. If we then interpret the 'definitely' of the argument above to mean 'precisely'. in other words. for example. Evans's argument is sometimes discussed as a refutation of the view that there can be vagueness in the world. it constrains them in a severe fashion. that this is valid: 'V(a=b) But the following is trivially valid (recall that '. although the above argument does not refute vague identities. the view under attack in the argument holds that indeterminacy in a state of affairs results in a truth-value for a sentence that articulates it. However. if this were the only objection to Evans's argument... A statement may be vague without being precisely vague (it may be vague even if it is vague whether it is vague). vagueness of identity.) Suppose we take the antonym of 'vague' to be 'precise'.. and we take 'precisely' to mean 'precise and true'. Fine (1975: §5) develops systems for vagueness both with and without these principles.S' entails'S'): 'V(a=b) "". A natural response to the argument as a refutation of vagueness is to say that it runs afoul of higher-order vagueness.'V(a = b) . The proof shows. it appears to be a view according to which no identity statement with singular terms that are used de re can be vague.'VS' from ''VS'. Suppose we reclassify line 2 as a premiss. in step 2 above. What this now shows is not that line 1 is inconsistent. some are vague and some are not. These are too different for a refutation of one to carry over to the other. The reduction postulate of SS is used only once.Evans on Indeterminacy 211 The view discussed in this book is untouched by the argument. Here is why. and that of such statements. the argument would still be of great interest to