You are on page 1of 29

Indeterminancy of Identity of Objects and Sets Author(s): Peter W. Woodruff and Terence D. Parsons Source: Nos, Vol.

31, Supplement: Philosophical Perspectives, 11, Mind, Causation, and World (1997), pp. 321-348 Published by: Blackwell Publishing Stable URL: http://www.jstor.org/stable/2216136 . Accessed: 28/12/2010 02:41
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=black. . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing is collaborating with JSTOR to digitize, preserve and extend access to Nos.

http://www.jstor.org

Philosophical Perspectives, 11, Mind, Causation, and World,1997

INDETERMINACY OF IDENTITY OF OBJECTS AND SETS

Peter W. Woodruffand TerenceD. Parsons University of California,Irvine 1 Might Identity be Indeterminate?' The purpose of this essay is to explore the idea that identity might be indeterminate.That is, that certain questions of identity have no answer,not because of an inadequacyin the language in which they are framed, but because of genuine indeterminacyin the world. In section 1 we describe this idea in general terms, in section 2 we give a classical "picturing"of the situation designed to allay fears thatthe idea is incoherent,in section 3 we discuss Evans-like attempts to disprove the view, in section 4 we extend the accountto sets of objects, and in section 5 we discuss a classical alternativeto the view of indeterminacydescribed in the first three sections. 1.] Examples of indeterminateidentity The hypothesis of indeterminacyof identity arises naturallyas a plausible solution to puzzles in which identity questions seem to have no answers. Examples in the literaturefall into three classes. First, there is a question of identity over time when there is a simple disruptionof some kind. For example, a person receives a new brain having the old memories, or a new set of memories is inserted into an existing brain,or...Thereseems to be a single person underdiscussion beforethe disruption, a single personunderdiscussionafterthe disruption, and and the question arises as to whetherwe are dealing with one person or two. Put in terms of identity, is person a (the person before the disruption)identical with person b (the person after the disruption)?Cleverly designed cases will undermine any definite answer to the question, and one possible reaction to this is to hypothesize thatthis is because the world leaves the answerindeterminate(while leaving determinatethe claim that there is exactly one person before the disruption, and exactly one after the disruption). Additional complexities arise in cases of splitting across time. If a ship is repairedby having its parts replaced one at a time, and if the replaced parts are reassembled, is the repairedship b identical to the original ship a, or is the reas-

322 / Peter W. Woodruffand TerenceD. Parsons

sembled ship c identical to a? Here we have a non-identity to complicate the


picture:

btc & aMb & a=c. Again, there is the option that the question marks do not indicate uncertainty abouta definite answer,but ratherthe lack of an answer,and again not because of any deficiency in language, but ratherbecause of indeterminacyin the world. A thirdpopularsort of case arises at-a-time.Cats have indeterminate boundaries;theremay be no answerto the questionwhethera molecule loosely attached to the tip of a hair that is midway in the process of being shedded is a part of the cat or not. Considera cat, and consider the various cat-like objects that overlap it and that have determinateboundaries. (E.g. a catlike thing that definitely includes the molecule, a catlike thing thatdefinitely excludes it, a cat-like thing that includes that molecule but excludes a certain other one, and so on.) Call these "p-cats."Then the p-cats are definitely distinct from one another,but there may be no answer to the question of whether the cat itself is identical with p-cat number9, and so on for all the others. In this case the patternis:
PI P2 & P 1P3 &. ..& cat pI & cat

p2 &...

These three sorts of examples arise from differentpuzzles and motivations;what they have in common is the possibility that identity claims may have no answer. We will addresswhat they have in common, by exploring the view that the questionable identities are ordinaryidentities that are not made determinateby the world, while the obvious non-identities and identities are made determinateby the world. 1.2 Indeterminacyin the world What would it be like for there to be genuine indeterminaciesin the world, not due to indeterminacyof how our language relates to the world? Here is a description: The world consists of some objects, and some propertiesand relations, with the objects possessing (or not possessing) propertiesand standingin (or not standingin) relations. Call these possessings and standings-in"statesof affairs."Then some but not all such states of affairsare determinedby how the world is. Many people are willing to accept this possiblity with one reservation;they hold thatidentityis an exception to the claim thatstates of affairscan be indeterminate. If an identity sentence lacks truthvalue, then that must be because there is some indeterminacyin what its singularterms refer to. For identity in the world could not possibly be indeterminate;that would be incoherent. The purpose of this

Indeterminacyof Identity / 323

paper is to oppose this view. There is no incoherency in the view that identity itself is sometimes indeterminate. We can anchorthe debate somewhat if we accept the traditionalLeibnizian definition of identity between a and b as indistinguishabilityof a from b in terms of the actual properties and relations that they have or stand in. Whether this holds in a particularcase may be undeterminedby the facts. If so, a and b will actuallybe neitherdeterminatelyidenticalnordeterminatelydistinct.This will be a case of indeterminacyof identity. The frameworkof properties and relations sketchedabove is neutralaboutwhetherthis ever actuallyhappens.It will happen if a definitely has every property that b definitely has (and vice versa), and a definitely lacks every propertythat b definitely lacks (and vice versa), but there is some propertythata definitely has or definitely lacks when it is indeterminate whether b has it (or vice versa). 2 Describing the Theory Many people sincerely doubtthe cogency of indeterminate identity.They are confused by the notion, and descriptions such as those above, while helpful, do not completely clarify things, because the descriptionssound as incoherentas the view being defended. In general, there are two ways to defend a view whose coherence is at issue. One way is to describe the view in its own terms, and to produce sufficient discussion of problem cases from this point of view that it begins to become clear that the view suffers from no internal incoherence. This is frustratingto those who genuinely have difficulty in comprehendingit. The second way is to picture the view in neutralterms. Below, we give a way to do this for indeterminateidentity: to picture indeterminateidentity within a classical bivalent conceptual scheme that makes no use of indeterminateidentity,just as a possible worlds modelling of modal logic can be stated in language containing no modalities. This picturing proves that the proposed view is coherent, because it can be coherently modelled by the classical view. Incoherencies in the view under discussion, if they existed, would show up as inconsistencies in the picturing.
2.1 Picturing indeterminacy

We can picture mundaneindeterminacyusing diagrams similar to Euler or Venndiagrams,in which objects are representednot by points but by small filled regions. We use circles to representthe extensions of properties,as in Venn diagrams. Then an object is pictured as having a property if its image is wholly inside the property'sextension, andit is picturedas lacking a propertyif its image is wholly outside a property's extension. When its image overlaps the boundary whetherthe object of the property'sextension, thatmeans thatit is indeterminate picturedhas the property:

324 / Peter W. Woodruffand TerenceD. Parsons

Fig-ext of p

Lacksp p

/ Neither has p nor lacks p

p ~~~~~Has

It is easy to extend this kind of picturingto identity of objects; objects a and b are pictured as identical if they are representedby the same image, and they are pictured as distinct if they are representedby disjoint images. If their images whether properlyoverlap,this pictures(the stateof affairs)thatit is indeterminate a and b are the same:

A is identical to A

It is indeterminate
whether B = C

A is distinct from B
Withthis convention, Leibniz's definition of identity is nicely captured.It is easy to see that if the images picture two objects as distinct, then the images do not overlap, and a circle can be drawn to representthe extension of a propertythat one of them wholly has and thatthe otherlacks. So objects representedas distinct will definitely differ in their properties. If the images coincide then no circle representingthe extension of a propertycan distinguishthe object(s) in any way.

Indeterminacyof Identity / 325

So objects representedas identical must agree completely in which properties they have, which they lack, and which they are indeterminatewith respect to. Suppose finally that the objects are pictured by means of overlapping images. Then neither of the above cases hold; no property can wholly distinguish one from the other(since any circle thatencloses one image must enclose at least part of the other), but there are propertiesthatone has or lacks and it is indeterminate whetherthe otherhas or lacks it (just drawa circle thattotally encloses one image but not all of the other image). So if having a propertymeans being in the extension of that property,these diagramspicturethe kind of indeterminacyof states of affairs(including states of affairs involving identity) that were describedat the outset of this essay. Further, the diagrams validate Leibniz's account of identity. At least they do so if any region in the diagramis the extension of a property.And that assumptionmeshes with everythingwe have said so far. (We'll make a weakerassumptionbelow, but it will not change the way in which the diagramsvalidate Leibniz's definition.) Nothing aboutthese diagramsexcludes the possibility that one object image might fall totally inside another,or inside a group of others. By fiat, we prohibit this from happeningfor the applicationsdiscussed in this paper:

No objects are completely inside others


We prohibit these cases because in all of the applications of the hypothesis of indeterminate identityto solve puzzles from the literaturesuch (onto)logical containmentsof objects in othersnever arises as a possibility. Such overlapsgenerate radicallydifferentconceptualschemes; these areworthexploring,but we will not do so here.2 With diagramsof this sort we can picture the three types of problem situations from the literatureas follows:

326 / Peter W. Woodruffand TerenceD. Parsons

Person a

M person b

Ship b f ship c & ship a

M ship b & ship a M ship c

pcat1 #pcat2 pcat1 & #pca

&

pcat, fpcat2 & pcat, fpcat3 &..& cat=?pcatj & cat=?pcat2 &..

Indeterminacyof Identity / 327 2.2 A more rigorous formulation

If we are to nail down the logic of the situation, we need a more rigorous formulationof the picturingdescribed above. Here is one, within a classical bivalent theory. Suppose that the image space of the picture consists of a set D of points, called ontons. The term 'onton' is purely suggestive; the points form the ontological basis of the space. Ontons are purely artificial, much as possible worlds are artificial in modal model theory. We suppose thatin any total picturingof reality thereare a numberof images of objects; each image is a set of ontons. The only constraintwe impose on such a modelling is that each object image contain some ontons (at least two) that are not in any other object. The ontons in an image that are not in any other image constitute the core of that image. Object images representidentical objects if they consist of the same ontons, they representdistinct objects if they share no ontons, and they representindeterminately identical objects if they share some ontons but do not share some others. Each propertyon D has a figurative extension, which is a set of ontons, and each relationon D has a figurativeextension which is a pairof sets of ontons. An object image i pictures its object as having a propertyp iff every onton in i is in the figurative extension of p. An object image i pictures its object as lacking a propertyp iff no onton in i is in the figurativeextension of p. And an object image i pictures its object as neither having nor lacking a propertyp iff some of its ontons are in the figurative extension of p and some are not. (Similar conditions apply to pairs of objects and relations.) Assume for the momentthatevery set of ontons is the figurativeextension of some property.Then the following versions of Leibniz's account of identity turn out to be true on this account: a is (definitely) identical to b iff a and b both have and lack the same
properties

a (definitely) differs from b iff a has some property that b lacks, or vice versa It is indeterminate whethera is identical to b iff thereis no propertysuch that a has it and b lacks it, and no propertysuch that b has it and a lacks it, and thereis some propertythatone of them has or lacks and such thatthe otheris indeterminatewith respect to having it. Finally, we suppose that there is such a picture that accurately pictures reality. That is, we assume that there are indeed objects and properties, and that any object image pictures a unique object, and that every propertyhas a figurative extension. An object has (or lacks) a property iff it is pictured as having (or lacking) it.

328 / Peter W. Woodruffand TerenceD. Parsons

2.3 Semantics In discussing a metaphysical thesis about identity it is importantto distinguish issues aboutlanguage from issues aboutthe world. There is no way around discussing both. And so here is a sample semantics that takes account of the worldly indeterminacyof states of affairs, including states of affairs concerning identity. Most of the semantics is stated in terms that do not appeal to our modelling at all; so much of it makes sense on any account of indeterminateidentity. 1. We assume thateach atomic predicateis trueof certainobjects, andis false of certain otherobjects, and is thus neithertruenor false of the remainder,if any. An atomic predicatemay or may not express a property.If it expresses a property, then it must be true of exactly those objects that have the propertyand false of exactly those objects that lack the property.(Similarly for relations.) 2. Satisfaction: (i) An object o satisfies 'Px' if 'P' is trueof o; o dissatisfies 'Px' if 'P' is false of o, and otherwise o neither satisfies nor dissatisfies 'Px'. (ii) 'x=y' is satisfied by a pair of objects if they both have and lack the same properties, dissatisfied if one of them has a property that the other lacks; otherwise it is neither satisfied nor dissatisfied by that pair of objects. 3. We have to make some arbitrary choices about how our connectives and quantifierswork, in orderto have any logical terminology at all. We shall use the commonest ones, which are these: --A is true if A is false, false if A is true, and truthvaluelessif A lacks truth value. A&B is true if A and B are both true, false if either A or B is false, and otherwise lacks truthvalue. (3x)A is trueif A is satisfied by at least one object, false if A is dissatisfied by every object, and otherwise neithertrue nor false. VA is true if A lacks truthvalue, and is otherwise false. As usual, we define AvB as (-A&-B). It is also convenient to define a "truth" operator,'>', such that '>A' is true if 'A' itself is true, and is false if A is either false or lacking in truthvalue. Its definition is: >A
=df

A & --VA.

In the Appendix we state a set of rules for the logic of indeterminateidentity, based on the above explanations. 3 Applications Any theory that allows for indeterminacyin the world places constraintson how language might relate to the world. Suppose that we wish to add to our lan-

Indeterminacyof Identity / 329

guage some lambdaabstractsto form complex predicates,such as 'Xx[Ax& Bx]', i.e. the predicateof 'being both A and B.' If the language is sufficiently rich then we cannotassume thatany such abstractrefersto a propertywhose applicationto objects is perfectlycharacterized the usualway by lambdaabstraction, prinin the ciple that '?(a)' is interchangeable with 'Xx[1(x)](a)' in all extensional contexts. Such a powerful assumptionleads to paradoxes(like the Russell paradox)whenever the languageis sufficiently rich. Constraintsof this sort arewell-known, and people are used to restrictingeither the abstractionaxiom or quantificationover propertiesto avoid paradoxes. Here is a less familiar constraint:one cannot take for grantedthat lambda abstractsthat bind variablesin contexts governed by the indeterminacyoperator '7' standfor propertiesand also fully satisfy lambdaabstraction.This may seem to be a dry technicality,but it is a crucial one. This is because Evans' 1978 argument and all argumentssimilarto it (thatis, virtually all argumentson this matter in the literature)beg the question by ignoring this point-by assuming without argument that such abstracts automatically stand for properties and that they simultaneously satisfy the abstractionprinciple. We are not in a position here to find fallacies in argumentsfor these joint assumptions, for such argumentsare never given.3 But we can say with some confidence that the assumption that abstractsboth standfor propertiesand satisfy lambdaabstraction arbitrary cannot be provedfrom the generalview of the world statedin section 1. It is an additional assumption, and one that must be rejected by any proponent of indeterminate identity.Indeed, Evans' argumentproves this. 3.1 Evans'Argumentand Leibniz'sLaw To get clear on this, consider the Evans argument.It is a proof that if there were a case of indeterminacyof identity, the supposed indeterminateidentity could be shown to be a determinatenon-identity.Thatis, from the claim thatthere are objects a and b such that it is indeterminatewhethera is b one can prove that a is not b: 1. 2. 3. 4. 5. 6. Hypothesis. From 1 by abstraction. V-7(b-b) (Definite) truthof self-identity. From 3 by abstraction. -Axx[7(x=b)](b) (3P)[P(a) & -1P(b)] Conjoin 2 and 4 and existentially generalize. From 5 together with the definition of identity. --(a=b) V(a=b) Xx[V(x=b)](a)

Thus from the assumptionthat it is not determinatewhethera is b, we prove that


a is not b.

Several authorspoint out that the argumentas stated is too strong; surely identity statements can lack truth values if there is linguistic vagueness in the singularterms.4We agree with most commentatorsthatEvans intendedhis argument to prove that if the singularterms are not themselves vague, then there can

330 / Peter W. Woodruffand TerenceD. Parsons

be no truth-valuelessidentity statements.If the only escape from the argumentis to hold that the terms are linguistically vague, then the argumentsuccessfully disproves the possibility of indeterminateidentity in the world. Since one can coherentlymaintainthatit is indeterminate whethera is b, we regard this proof as establishing instead that the fault may be taken to lie elsewhere, in particular,either in the abstractionsteps or in the existential generalization step. We cannot assume that there is a genuine property ('not being determinatelyidenticalto b') thatbehaves in accordancewith the full abstraction principle.To avoid assuming this, we have two options. One is to assume thatthe principle of abstractionalways holds for predicates,but to leave it open whether the resulting predicates stand for properties. If we do this, the above argument begs the question at step 5. The other option is to take for grantedthat abstracts standfor properties,but rejectthe principleof abstractionas always providingthe conditions under which such properties hold of objects. Then the abstraction steps 2 and 4 may be unjustified. Recall what is at issue. It is not whetherpropertyabstractsare meaningful; that is a conceptual point, and says nothing about what there is. The issue is whether for every such abstractthere is a property in the world for which the abstractstands. In the case of identity, the issue of how identity behaves in the world is not a conceptual matter,it is an ontological one. It is characterizedin terms of properties and relations, not in terms of concepts or meanings. And so assuming that a propertyabstractis meaningful does not mean that it stands for a property.The abstractused in the argumentdoes not stand for a property,or it stands for a property that is not fully equivalent with the subformulain the abstract. Evans' own argumentappearsto avoid these considerations,since that argument does not quantify over properties at all. Evans apparentlybypasses any appeal to propertiesby moving directly from lines 2 and 4 to line 6, citing Leibniz's Law: From a=b and ...a...
infer ...b...

This is a good argumentform. If one of your premises is 'a=b', then you are assuming that 'a=b' is true. But if it is true, it is determinatelytrue, and a determinately true identity should sanction interchangeabilityof its terms (assuming thatthere are no non-extensionalcontexts at issue). We agree completely, and we note that the metaphysicalaccount of identity sketched above sanctions this version of Leibniz's Law. But Evans does not appealto Leibniz's Law; he appealsto its contrapositive.It is essential to distinguish Leibniz's Law, which is valid for anyone who is discussing real identity,from contrapositiveversions of it, which we dispute.This is because most attackson indeterminate identityin the literature employ principles that are contrapositiveversions of Leibniz's Law, such as the following:

Indeterminacyof Identity / 331


From ... a...

-1(...b...) infer aVb.

and

Ourframeworkdoes not endorsethis principle,even thoughit does endorseLeibniz' s Law. It is hard to shake the idea that if an inference is valid, its contrapositive should be too. But contrapositiondoes not hold for systems that allow indeterminacy, even if indeterminateidentity is ruled out. In fact, it does not even hold for systems that presume that the world is completely determinatebut that language is sometimes so flawed as to allow the formationof sentences that do not have truthvalue. Suppose there is some sentence S in some languge that lacks truthvalue, and suppose thatthis languagecontainsthe connective '7' thatforms true sentences from truth-valuelessones and false sentences from truth valued ones. Then the following is certainly valid:
S/... -,7S

But its contrapositiveis not even remotely plausible:


VS /.*.

-is.

(When we discuss validity we mean thatif the premise is true,the conclusion has to be truetoo. This notion of validity is the rightone to focus on because it is what is at issue in all of the philosophical argumentson the topic. ) Leibniz's Law is a principle of our system, but its contrapositive holds only for certain "wellbehaved"cases.5
3.2 Abstracts

None of the above discussion entails that we cannot have abstractsin our language.6There are two options for handling abstracts. One is to insure that abstractsstandfor properties,thereby abandoningfull abstractionprinciples (as proved above); the otheris to hold onto the full abstractionprinciplesbut without assuming that abstractsalways stand for properties. Call the first 'ontological abstraction,'since such abstractsare guaranteedby the semantics to stand for properties,and call the second 'conceptualabstraction,'since these abstractsare guaranteedto reproducethe conceptualcontent of the formulasfrom which they are generated.Both can be added to the language. where the asteriskon the Ontological abstractionis symbolized by X*x[FDx], A indicates the ontological loading. Ontological abstractionis interpretedas follows: X*x[(Fx]stands for a propertywhose figurative extension contains all ontons that are in objects that (definitely) satisfy 'Fx plus some but not all ontons in the core of each object thatneithersatisfies nor dissatisfies

332 / Peter W. Woodruffand TerenceD. Parsons


Conceptual abstraction (using a 'c' for 'conceptual') produces complex predicates that work as follows: Xcx[(ix] is a predicate that is true of an object o iff o satisfies IDx,and is false of o iff o dissatisfies 1'x. What logical principles are validated by these accounts? To answer this, we need to make some distinctions that are not needed in a classical setting. First, there is the question of the validity of the ordinary abstraction inferences: ABSTRACT INTRODUCTION:

(Da X[. x](a)


ABSTRACT ELIMINATION:

Xx[1'x](a) (Da
These principles are valid without restriction for either sort of abstract even if (D contains instances of the indeterminacy operator. That is, in any such case, if the Thus all of our abstracts are minimally premise is true, so is the conclusion. well-behaved. But in a non-classical setting, we can ask another question: must the conclusions of the above arguments be indeterminate (or true) if the premises are indeterminate? The answer is yes for conceptual abstracts for both inferences, since the conclusion must be indeterminate if the premise is. For ontological abstracts, this also holds for abstract introduction: if the original is indeterminate then the conclusion will be indeterminate too. But abstract elimination for ontological positive abstracts can take you from indeterminate to false. This means that the contraof ontological abstract elimination:

--a 8,X*x[(Dx](a)
can have a true premise and an indeterminate conclusion, and it is thus not valid. This is the form of the inference in step 4 in the argument above. In summary, the Evans argument can either be taken to appeal to conceptual abstracts or to ontological ones. If the abstracts are conceptual, they do not necessarily stand for properties, and the fallacy lies in assuming that they do (step 5). If the abstracts are ontological, the fallacy lies in assuming that the contrapositive of abstract elimination is valid (step 4). 4 Sets of Objects Sets are important to discuss because there are arguments based on set theory both for and against the claim that there is indeterminate identity. The point of this section is to extend our picturing to include sets, and to assess these arguments.

Indeterminacyof Identity / 333

Sets of objects are the extensions of properties of objects, and thus we already have them picturedin our model: a set image is any figurative extension. An object is picturedas (definitely) being a memberof a set if the object image lies totally within the figurative extension that picturesthe set, and it is pictured as (definitely) not being a member if its image is totally outside that figurative extension. If a set image lies partly within and partly outside of a figurative extension, thatportraysa situationin which it is indeterminate whetherthe object is or is not a memberof the set.8 The point of this section is to develop a precise accountof sets of objects and explore their indeterminateidentities.
4.1 Principles of set theory

It is unnaturalto conceive of every region (every set of ontons) in the diagrams described above as representinga distinct possible extension, since some regions differ in completely artificial ways. Suppose, for example, part of our picture looks like this:

If there areno otherobjects in the logical vicinitiy, the following differentpicture should not picture anythingdifferent:

334 / Peter W. Woodruffand TerenceD. Parsons

Let us say that two regions are ontologically equivalent if each wholly encloses the same object images as the other and each wholly excludes the same object images as the other (and thus each partially includes and partially excludes the same object images as the other). Ontologically equivalent regions should representthe same extension. The conceptuallysimplest way to accomplishthis is to select from each class of ontologically equivalent regions one member of the class to representall of them. Call this the canonical extension for that class. The fact that there is more than one region in a given equivalence class is a purely artificial product of our picturing conventions, and has no ontological significance, so we can just speak in terms of the canonical ones. Canonical extensions picture sets of objects. Suppose that we call any canonical extension a set image, and we characterize membership of an object in a

set as follows: o is definitely a memberof set S iff o's image is totally included in S o is definitely not a member of set S iff o's image is totally disjoint from S o is neitherdefinitely a memberof set S or definitely not a memberof S iff o' s image is partly included in S and partly outside of S We say that two set images representthe same set if they are in fact the same canonicalextension; they representdistinct sets if thereis an object image totally within one that is totally outside the other;otherwise it is indeterminatewhether the sets they representare the same. What principles of set theory do these assumptions yield? To express the strongestprinciples, we need a biconditional strongerthan the materialone. We would like a biconditionalA=B to be true when A and B have the same truth value or both lack truthvalue, false when they have opposite truthvalues, and lacking in truthvalue when either of A or B has a truthvalue and the other does not. Such a biconditionalcan be defined by: A?B =df (A&B) v (-,A&-,B) v (VA&VB). Using this connective, the following principleof set extensionality is always true in our modelings: A=B = Vx[xEA = xEB] To say that a biconditional of this sort is true is to make the strongest possible claim about the equivalence of its two sides; it means that they are true together, false together, and also lack truth value together. Suppose we write A-B for >[A=B]-

Indeterminacyof Identity / 335

Then saying that the former principle is validated by our semantics as true is equivalent to saying that this is true: STRONGEXTENSIONALITY: A=B - Vx[xEA = xEB] Thus our principle of set extensionality has the strongest formulationthat one could hope for. The formulationallows that membershipin a set might be either determinateor indeterminate,and the right hand side of the majorbiconditional says of sets A and B that any object is either determinatelyin both A and B or determinatelynot in eitherA or B or is such that it is indeterminatewhetherit is a memberof either.And the whole sentence says that whetheror not this coincidence in membershipholds for all objects is definitive of whetherA and B are the same; if the coincidence always holds, then A and B are the same, if it does not always hold, thenA and B are distinct, and if it is indeterminate whetherit always holds then it is indeterminatewhetherA and B are the same. The identity conditions for sets are thus determinedexactly in terms of the membershipof objects in them. Whataboutset comprehension?Is there a set correspondingexactly to every formula?That is, is this principle true: STRONGCOMPREHENSION: 3SVx[xES - (Dx]. This principle holds for a wide range of cases, but not all. Suppose that (F is the sort of formulatypically employed in set theory;in particular,suppose thatevery atomic predicate in (F has an extension, and that (F is built up out of names and atomic predicatesand identitiesby means of the Boolean connectives &, v, and the quantifiers3 and V. Then the above formula holds for (D.But if (Dcontains some predicates that do not have extensions, or contains non-classical connectives, such as 'V', then it may not hold. Here is another way to view things. Let us introduce set abstracts,which mimic ontological propertyabstracts,i.e. with the following semantics: {x:( } stands for the smallest canonical extension that contains every onton in every object that satisfies (Dand contains some but not all ontons in the core of every object that neither satisfies nor dissatisfies (D. (Recall that the core of an object is the set of ontons in the object that are not in anotherother object, and we assumed above that each object has a core with at least two ontons.) In our modeling there is always a unique canonical extension that satisfies the above condition, so such abstractsalways pick out unique sets. Further,no matterhow (Dis formulated,the abstractsalways satisfy these classical conditions:

336 / Peter W. Woodruffand TerenceD. Parsons

CLASSICALCOMPREHENSION: (Da
aE{x:iFxl

aE {x:iFx}
.-. (Da

However, the strongerthree-valuedcondition STRONGCOMPREHENSION: Vy[yE{xAfx} _ (Dy] does not always hold. Again, it holds for the typical formulas of classical set theory,but not necessarily for formulas containing non-classical terminology.9 We have confined our discussion here to sets of objects, because these can be pictured.We speculate that it is possible to extend the account to a full extended set theory of the usual sort, and to recover from this theory of indeterminatesets a subfragmentthat mirrorsthe classical Zermelo-Frankeluniverse.'0 4.2 An argumentfromset theory against indeterminateidentity Salmon 1981 gives the following argumentagainstthe possiblity of indeterminate identity for objects:11 1. 2. 3. 4. Suppose Vx=y But --Vx=x So {x,y)f{x,x) So xOy.

There is nothing wrong with 1, which is a supposition,or 2, which is validatedby our modelling. But how do we get to 3? Since there are no intermediatesteps given, one can either speculate aboutthe fill-in reasoning, orjust evaluate things directly. In our modeling, hypothesis 1 entails: 3'. V({x,yj}{x,x}). It does not validate 3. Since manypeople have endorsedSalmon's argument,it is temptingto try to see what is behind their intuitions.And there is a naturalidea in the wings: Leibniz's Law, which is often cited in this connection. But the argumentdoes not appeal to Leibniz's Law, it appeals to its contrapositive,which is not valid. 4.3 An argument from set theory infavor of indeterminateidentity The implementationof set theoryin this frameworkis importantfor a second reason. Suppose that there is no indeterminacyof identity among objects at all, but suppose that there is some property indeterminacy;e.g. it is indeterminate whethera is P, where 'P' stands for some property.Then consider the sets:

Indeterminacyof Identity / 337

A= {x:Px} B ={x:Px v x=a} It is an immediateconsequence of our modeling that it is indeterminatewhether A=B. So any kind of indeterminacyfor objects leads inevitablyto indeterminacy of identity for sets. (This is then a kind of indeterminacythat differs from the three sorts of examples given at the beginning of this paper.)Indeterminacyof identity is much harderto avoid that most of its opponents suppose. Is thisjust an artefactof our model? We thinknot, for argumentscan be given for this conclusion that do not exploit the model at all. Considerfirst the following two proofs thatjointly show that 'Pa' cannot be indeterminate: 1. 2. 3. 4. 5. SupposeA=B. Then Vx[Px = Px v x=a] Then Pa = Pa v a=a Pa -,VPa

Comprehensionand Extensionality Instantiation Tautologically implied by 3 4, Meaning of 'V'

1. Suppose A*B. 2. Then -,Vx[Px = Px v x=a] 3. Then [-,IPb= Pb v b=a] for some object b 4. EitherPb & -1(Pb v b=a) or -1Pb & (Pb v b=a) 5. The left disjunct of 4 is inconsistent. 6. -1Pb&b=a 7. -1Pa 8. -,VPa

Comprehensionand Extensionality QuantifierEquiv. TautImpl by 3 Taut 4,5,TautImpl 6, Leibniz's Law (not its contrapositive!) 7, Meaning of 'V'

These argumentsassume nothing about indeterminacyof identity or about our models. But suppose that 'Pa' lacks truth value. Then each proof becomes a reductio of its hypothesis, and so the proofs jointly refute both A=B and A*B. This is not an inconsistency, but only because there is a thirdoption: the refuted sentences lack truthvalue. Thus, the only way to maintainthe indeterminacyof 'Pa' in the presence of normalprinciples of set theory is to admit the indeterminacy of 'A=B'.2

So the assumptionthatPa is indeterminateleads to the conclusion thatthere The obvious way out of this argumentis are sets whose identity is indeterminate. to abandon the possiblity of doing set theory on the assumption that there is genuine indeterminacyin the world (that is, giving up ordinaryinstances of set comprehensionor extensionality), or to insist thatall indeterminacy(notjust that involving identity)be disavowed as a matterof theoreticalsimplicity.Quine 1981 seems to offer this dilemma, and to favor the latter choice. But the dilemma is a

338 / Peter W. Woodruffand TerenceD. Parsons

false one. The argumentgiven above shows that indeterminacyof set identity is a consequence of worldly indeterminacyregardingobjects possessing properties; it does not show that a theory that accomodates these indeterminacies is less simple thanone thattries to get aroundthem. Indeed,as we have triedto show, the even if set identity can be theory of sets of objects is simple and straightforward complexityhere,thereis only unfamiliarity. indeterminate. Thereis no insuperable Further,if simplicity is the issue, then one really needs to examine the simplicity of ways aroundadmittingindeterminacy,otherthanjust tryingto look the other way. These usually don't bother people because these ways haven't been articulatedand subjected to criticism. We suspect that if this were done, the indeterminateidentity view would look at least as good. 5 A Determinate World The object images discussed in the two preceding sections will look to some people like blurryparts of a pictureof a world that may not itself be blurry.This is a theme that is found throughoutthe literature:that apparentindeterminacyof (perhaps,but not identity is a productof a deficiency in the conceptualapparatus necessarily, of our language) that we use to representa world that is itself completely determinate.One version of this is the idea that identity puzzles are to be solved by "refiningour concepts."(Cf. Stalnaker1988 for a view something like this.) In this section we criticize this idea. of Some writerson identityhave suggested thatapparent indeterminacy identity is due to unclarityin our concepts. If our concepts are clarified, we discover a world in which identity is completely determinate.When there is a genuine puzzle aboutidentity,this is because the key concept(s) appealedto in the puzzle can be clarifiedin differentways. For example, if our concept of a ship is clarified in one way, then this identifies the original ship with the reassembledship, and if it is clarifiedin anotherway, thenthis identifies the original ship with the repaired ship. This is not quite accurate,since in terms of unclarifiedconcepts the phrase 'the original ship' does not pick out a ship at all. The more accuratething to say is that if our clarified concepts are ship, and ship2, then 'the original ship' does not denote anything at all, and instead, we have that the original ship, = the reassembled ship, the original ship2 the repairedship2 These are equally true in the actual world, and there is no indeterminacyof identity to confound us. This view is not compatible with the indeterminateidentity view. For one thing, the refined-conceptview requiresa largerontology thanthe indeterminate identify view. The total numberof ships in the indeterminateidentity model is more than one and less than three. (It is indeterminatewhetherthere are two or three.) The refined concepts view requiresmore objects. Let place-o be the early

Indeterminacyof Identity / 339

location of the original ship, place-a the later location of the repairedship, and place-s the later location of the reassembled ship. The notion of ship can be refined so that the following are all simultaneouslytrue: the original ship, = the reassembled ship1, and the original ship1 0 the repaired ship,. So a ship, is located at both place-o and place-s, and a (different) ship, is located at place-a. the original ship2 = the repairedship2,and the original ship2 0 the reassembled ship2.So a ship2is located at both place-o and place-a, and a (different) ship2 is located at place-s. the original ship3 0 the reassembled ship3, and the repaired ship3 * the reassembled ship3. So a ship3 is located at place-o, and a different ship3 is located at place-s, and a (still different) ship3is located at place-a. This requiresfive "ships"in the actualworld (using the unrefinednotion of ship), more than any of the refinements of the indeterminateidentity view. Of course, we shouldn'ttalk thatway, since ship is a notion in need of refinement.Wherewe thoughtthere was one ship (i.e. where "theoriginal ship"was) there are actually three ship-like things: a ship, (which laterends up where "thereassembledship" is), a ship2(which laterends up where "therepairedship"is), and a ship3(which ceases to exist during the repair/reassembly).All of these are distinct from one another.(There are also multiple ships later on, two of them in each location.) These multiplicities are either the strengthor the weakness of this alternative. This sort of burgeoningontology may be untroublesomefor ships, but it is more difficult in the case of people who undergodisruptivechanges. We need to ask what are the conditions for refinementof the notion of person. Some natural answers that come to mind are that people are things that think, they are things that pay taxes, and they are things that we marry.If these sorts of paradigmsdo not hold then it is not plausible to claim thatwe are refining the notion of person, we are simply changingthe subject.Now refinementsin the notion of person will not end up with there being multiple persons in one place at a time, since our concept of person stands in need of refinement, and should be dispensed with. But (like the ship case) the refinementswill requirethere to be in a single place both a person, and a person2.If these "refinedpeople" have the characteristics that are essential to refinementsof persons, then in a given place will be located two or more things that think and feel and desire, two or more things that pay taxes (do they both pay the same taxes, or have there been duplicate payments, or...?), and, in one case, two or more things that you married.This is disconcerting to say the least. There are two problemshere. The immediateproblemis that the view under considerationis not a developed view at all; we don't really have an option to consider,but many options, dependingon what refined things are supposedto be like. The options need to be spelled out. We need to know, for example, whether

340 / Peter W. Woodruffand TerenceD. Parsons we get two person-like things that share a body, or two body-like things too. The second problem is the suspicion that if the view is spelled out it will be incoherent. But it is premature to address that before seeing the options. A third issue lurks. We still need to make sense out of our ordinary talk of identity, in which we speak as if there is a single person in a given place at a given time. Even if this is an illusion, based on a sloppy use of unclear concepts, the sloppy illusory talk seems to be systematic, and one would like an account of it in terms of the unsloppy realistic view. But when this is done, perhaps the precise reconstruction of the sloppy talk about inexact identity will be what we all mean now when we apparently talk about identity. And perhaps this reconstructed identity will be indeterminate. And then perhaps the indeterminate identity view will turn out to be true. Derivatively true, perhaps, but true nonetheless.

Appendix: Modellings and Rules


In this appendix we make precise the three-valued logic we are dealing with, by giving two (equivalent)semanticmodellings and an axiomatic formulation.The syntax of the languages has already been indicated: a first-order language whose primitives are negation (-), conjunction(&) and the existential quantifier(3), as well as identity (=) and the additionaloperator< (non-falsity).13From the latterwe may define V (indeterminacy), A (determinacy),and > (definite truth)as follows:
Vb =df

>D

AO+ = df-__V =dfV<--

1. Two Semantic Modellings


We will begin by formalizing the semantics given in the paper.Then we will give an apparentlymore general three-valuedsemantics, but will show thatthe two are equivalent in the sense that they recognize the same inferences as valid. a. Ontic models i. Basic notions An ontic model for our language will be a tripleM = (0, D, v) such that 0 is a set (called the set of ontons of M), D a set of subsets of 0 (the objects of M) and v an interpretationwhich assigns to each individual constant an object in D and to each n-ary predicatea function from Dn to {t, u, f}. An assignment in D is a mappingfrom the variablesof our language to objects in D; if a is an assignment and dED, a(d/x) is like a save for assigning d to x. An assignmentdeterminesfor each singularterm a denotationin D and for each,formulaa truth-valuein {t, u, f}; we write vae for the value of the expression e and define this recursively as follows: 14 voa[t] = Pt if t is a constant, otherwise at.
va[P(tl, .. t)] t = P(Pa[tl], .*-, val[tn])

Indeterminacyof Identity / 341 v-a[s=t] = t if vza[s] = va[t], = f if voa[s] n voa[t] = 0, and

= u otherwise.
= t if v'a[b] = f, = f if va[b] = t, and

= u otherwise
var[O&q] = t if vza[b] = va[g] = t, = f if voa[b] = f or voa[q] = f, and

= u otherwise.

voa[3xb] = t if for some dED, voa(d/x)[b] = t,

= f if for all d, voa(d/x)[(b]= f, and = u otherwise.

We say thata formula4 is a consequence of formulasE (E k b)iff 4 is truewhenever (i.e. on every model and assignment on which) all of E is. ii. Nuclear models In sections 2.1 and 2.2 we made the assumptionthatevery object has a core subset of at least two ontons which belong to no other object. Let us call ontic models which satisfy this assumptionnuclear models. It might appearthat limiting ourselves to nuclear models is a strong assumption,but we shall see that they are in fact as general as those above, and lead to the same theory. iii. Figurative extensions Up to this point we have not mentioned figurative extensions (Section 2.2), nor the propertieswhich have them. Indeed, as noted in section 2.3, the modelling given above is independentof these notions. But we can introducethem in the following way. Let p be a three-valuedfunction on D (call such functions "concepts"for the time being). Then we say that a set S of ontons is a figurative extension of p iff for all dED, p(d)=t iff dCS andp(d) =f iff dnS =0. A concept is a propertyiff it has a figurative extension. It is worthnoting thatwe can give a more directcharacterization properties.To this of end, let the center of a concept p be the set of objects which satisfy it; thepenumbrabe the set of objects which neither satisfy nor dissatisfy it; and thefringe be the set of objects which overlap some member of the center but do not themselves belong to the center.15 Then Proposition 1: If p is a property, the fringe of p is a subset of the penumbra of p. In nuclear domains, the converse holds as well: a concept whose penumbra

contains its fringe is a property.

Proof: If p has a figurative extension, say S, then every member of the center is a subset of S, hence every member of the fringe overlaps S, hence does not the condition and D is nuclear. Take S to be the union of the center of p, together with exactly one element from the core of each penumbral object.

dissatisfy p and so belongs to the penumbra.Conversely, suppose p satisfies

342 / Peter W. Woodruffand TerenceD. Parsons Then if p(d) = t, d is in the center and hence a subset of S. If p(d) = u, then by construction d overlaps S; but also, since D is nuclear there will be some core element of d which is not in S. Finally, if p(d) = f, d is neither penumbral nor central; since (by assumption) the fringe of p is contained in the penumbra, d cannot be in the fringe, hence cannot overlap any central object. The only way, then, that it could overlap S is to contain a core element of some penumbral object, but then it would have to be that object, which is impossible. The preceding discussion has been restricted(for simplicity) to singularyproperties as opposed to relations,but the generalizationis straightforward. one n-tupleof objects Let Overlap (Include) another iff it overlaps (includes) at each coordinate. Then all of the above definitions may be repeatedwith capitalizedconcepts, andthe new version of Proposition 1 remains true.

b. General models
Ontic models give considerable structureto the objects of their domain. An apparently more general approachis to leave these objects unanalyzed,taking a general model to be simply a pair (D, v), with v as before. The one thing which must be modified is the recursive clause for identity, the only one which uses the ontic structure.We must now of provide more directly for the interpretation =. Thus = is treatedlike any binarypredicate, except that we require =1)
=2)

v'=(d,e)=tiffd=e
v=(d,e) = f iff v =(e,d) = f

Clause =1) is as before; clause =2), on the otherhand,is not a definition (as such it would be circular) but only a constrainton v. Given = 1), =2) is equivalent to postulating the symmetryof v= as a function: v=(d,e) = v=(e,d). Because of the symmetryof disjointness,it is evident thatevery ontic model (0, D, P) contains an equivalent general model (D, v). But what may be surprisingis that ontic models, even nuclear ontic models, are fully as general as (so-called) general models.'6 Proposition 2: For every general model (D, P) there is a nuclear model (0, D*, P*) and a bijection * from D onto D* such that for all terms t, (Pat)* = P*a*t, and for all formulas b, Pao = P*a*4o.l7 Proof: Given (D, P), let a and b be distinct objects not in D, and let 0 consist of the set of (unordered) pairs of objects in DU { a, b}.'8 For dED, let d* have as members {a, d}, {b, d}, and all {e, d} such that eED and P=(d,e) 0 f. Then d* is a subset of 0, and we can choose D* to be the set of all d* for dED. Note that if d* = e* then { a, d} Ee*; since there is exactly one set in e* which contains a, we must have {a ,d } = {a, e }; since a must be distinct from both d and e, we must then have d = e.19 Thus * maps D 1-1 onto D*. We will need the following
Lemma: v=(d, e) = f iff d*ne*
= 0.

Indeterminacyof Identity / 343

Proof of Lemma: Suppose v=(d, e) = f. Then dOe and {d, e1Xd*. Now suppose {g, h} Ed*nle*. Then (since every memberof d* contains d) either d = g or d = h, and similarly,eithere = g or e = h. Since d ande are distinct, it follows that we have d = g and e = h or v.v., and hence {d, e } = {g, h I E d* after all. By reductio, the condition of the lemma follows. Conversely, if v=(d, e) 0 f, then by construction {d, elEd*; furthermore,by =2) and constructionwe have {d, e}Ee*. Hence d*ne* 0 0, q.e.d. for the lemma.
We may now prove the conclusion of the propositionby inductionon terms and formulas. The only thing which really needs to be checked is that truth-valuesof identities are preserved.To this end, let d be vas and e be vat. We may assume by inductive hypothesis that v*a*s = d* and ,A*a*t= e*. It will suffice to show that v=(d, e) = v*a*s=t. Note first that v=(d, e) = t iff d = e iff (by the I-lness of *) d* = e* iff (by the above assumptions and the ontic semantics) v*a*s=t = t. By the lemma, v=(d, e) = f iff d*ne* = 0 iff (by assumption)v*a*snlr*a*t = 0 iff v*a*s=t = f, q.e.d. It is a corollaryof Prop.2 thatontic and generalmodels determinethe same notion of consequence. For suppose, say, thatX k 4k thatthereis some general model in which E but is true and b not. Then that model would determine by Prop. 2 an ontic model which satisfied E but not 4, contraryto our supposition. c. Artifacts of the model It is clear from our results that very little use is made of the fact that the objects of an ontic model are sets of ontons. In fact, the only thing that matters is the ways in which objects overlap (something they could not do if they were not sets); overlaps are used to determine whether objects are distinct or merely indeterminatelyidentical. Everything else about objects is an "artifactof the model" in the same way as is the fuzziness of a string model of a triangle:it is not part of what is being representedby the model. To make this point precise, let an onticframe be a pair (0, D), D partof the power set of 0. A model (0, D, v) is said to be over (0, D). The theory of a frame is the set of all formulas valid in the frame, i.e. true on every assignmentin every model over the frame. Two frames may be called overlap equivalentif there is a 1-1 map h from D onto D' such that for all d, e in D, doe iff h(d)oh(e). Then we have

Proposition 3: Overlapequivalent frames have the same theory.


The proof of this is very similar to that of proposition 2, consisting in setting up a correspondencebetween interpretations over the two frames, such that correspondinginterpretationsassign the same truthvalues; overlap equivalence is used to handle identity formulas. Reasoning similar to that in the proof of Proposition2 also lets us establish:

Proposition4: Every ontic frame is overlap equivalent to a nuclear frame.


This is why the restrictionto nuclearframes is inessential. (Enriching our language with conceptual or ontic abstractsadds no essential expressive power; Propositions 3 and 4 remain true for the expanded language.)20

344 / Peter W. Woodruffand TerenceD. Parsons

2. Natural Deduction for Indeterminate Identity


We now give a syntactical presentationof the set of valid inferences captured (in common) by the two modellingsjust given. These arenaturaldeductionrules of two sorts, direct and indirect.The direct rules are typified by &elim, written b&O kF and read "4 is derivable from +&Xf". Indirectrules are typified by reductio, e.g.

E F <-,+
read"if if and <-1 / areboth derivablefrom E and4),then <-4) is derivablefrom E alone." The exact implementationof naturaldeduction(Fitch, Quine, Kalish/Montague,etc.) will be left to the reader's imagination. The presence of < (non-falsity) in the above rules is characteristicfor three-valued naturaldeduction, being requiredto mesh with the fact that asserting a line in a proof always means asserting it as true. In the following rules, 4)(t/x) means the resultof substitutingt for x whereverx is free in 4);it is assumed that bound variables are rewritten as necessary to avoid clashes in carryingout such substitution.

a. Basic three-valuedlogic
Rules for Conjunction
&intro: &efim: ob, F ?b&/

O&O k ?>

? & k qj

<&intro:
<&elim:

<sb, <q F <(?b&0) <(b&0)


F <b <(b&q/) F <b

Rules for Negation21

RAA:
Rules for Non-Falsity22
<intro: b F <b

Indeterminacyof Identity / 345 Rules for Quantification 3Eintro: b(t/x) F 3xX

Eelim:

<3intro:

<b(t/x)

F <3xb

< 3 elim:

<3 x

(provided y is not free in the conclusions of the elimination rules). b. Theories of identity The following two rules characterizedeterminatelytrue identity. Reflexivity: Leibniz' Law:
Ft = t

b(s/x), s = t F b(t/x)

To these we add the symmetryof weak (non-false) identity: Weak Symmetry: <(s = t) F <(t = s).

Proof of the following proposition can be obtained along the usual Henkin lines, making 4 truein the canonical model if it belongs to the maximal set, false if <10does not so belong, and u otherwise. For details see Woodruff 1969 and (in preparation). Proposition 3. The above system of rules is sound and complete with respect to the notion of consequence defined in 1.1.2 above. These rules make obvious what was implicit in our semantics: the theory of true identity is just classical identity theory, while the theory of weak identity is simply the This means that most of the interest of three-valued theory of a symmetrical relation.23 identity theory lies in the propertiesof weak identity,and it is here thatone might look for interestingnew concepts and axioms.24 Let us just mention two possible additions. version of LL which we call the principle of Definite The first is the "contraposed" Difference:25 Definite Difference: b(s/x), --,b(t/x) F -i(s = t)

346 / Peter W. Woodruffand TerenceD. Parsons This is equivalent to the following replacementprinciple for weak identity: LL*: ?b(s/x), <(s=t)
F <?b(t/x)

and to Evans' principle: Evans: V(s = t) F (s = t),

the last formulationbeing the conclusion of Evans' famous argument.As we have already observed, DDiff does not follow from LL, but only from LL*, and the failure to note this is the Achilles' heel of Evans' argument.In fact, that DDiff is valid in a particularmodel for a context b(x) amountsto saying that the concept expressed by this context is a property (see 1.1.2 above). Those whose intuitions agree with Evans' will want to add one of these principles to those we have espoused. Any one of them togetherwith the other rules is equivalent to the determinacyof identity: F -,V(s = t). The second possible additionis a genuine developmentof the theoryof weak identity, one which is especially relevantto ontic models. We call it the principleof Definite Identity of Weak Indiscernibles(DWI): DWI: Vx(<(s = x)
X4

<(t = x)) k s =t

The adoptionof this axiom leads to a rich theory of the ontological structureof indefinite objects (see the note after Proposition4 above). Notes 1. This paper recapitulatesparts of Parsons & Woodruff 1995 in slightly more technical form, and extends the discussion to a considerationof indeterminateidentity for sets. 2. These are touched on in Woodruff(in progress). 3. E.g. Pelletier 1989 claims that rejection of these principles is based on incorrectly construingthe indeterminacyoperatoras a modal operator.(This may be true of some people's rejection of them, but it is not true of us.) He does not go on to show that the principles are true. 4. See Broome 1984, Cook 1986, Burgess 1989, Garrett1988, Noonan 1982, 1984, Rasmussen 1986, Thomason 1982. 5. Its contrapositiveholds for any formulawhose predicates all stand for properties,and that contains nothing else except identities, the connectives and &, and quantifiers. "Gap-filling"connnectives such as 'V' or '>'can spoil the validity of the principle. 6. We thus disagree with Parsons 1980:14 if he is interpretedas saying that we cannot have abstractsin a language that contains the indeterminacyoperator.(That interpretation may be strongerthan he intended.) 7. Such a figurative extension will always exist; the "some but not all" condition is satisfiable because of our earlierassumptionthatevery object has a core with at least two ontons. It is possible for more thanone figurativeextension to meet the conditiongiven in this clause for ontological abstracts,but when this happens the distinct figurative extensions completely include and completely exclude exactly the same objects. 8. Do not confuse the fact that a figurative extension is literally a set of ontons with the fact that that figurative extension is an image of a set of objects. This is no different

Indeterminacyof Identity / 347 from Venn diagrams, in which sets of spatial points on the page represent sets of objects that are not themselves spatial points. 9. The conditionholds for some of these as well. It is a necessary and sufficient condition for this condition to hold thatthere are no objects x and y such that it is indeterminate whether x=y and such that x satisfies CF and y dissatisfies (D. 10. This is describedin Parsons & Woodruff"Indeterminate Theory"(in progress). Set 11. Salmon uses orderedpairs instead of pair-sets, but this is not an essential change. 12. The argumentas stated makes use of set abstracts,but this is not essential. The reasoning can be duplicatedso as to derive ']X 3YVX=Y' from 'VPa' withoutusing set abstractsin any proofs. 13. We choose non-falsity,ratherthanindeterminacyas primitivebecause of its role in the naturaldeduction system to be given later. The propositionalconnectives we use are not functionally complete for threevalued logic; they suffice to express all those functions which in the terminology of Langholm 1988 are determinableand persistent. We could make them complete by adding a constant * with the value u, but there appearsto be little motivation in the present context for such a constant. Indeed, on the principles we espouse it appears likely that all cases of indeterminacywill be contingent. 14. The semantics for identity follows the discussion of the paper.The clauses for & and are those of the so-called Kleene strong connectives (Kleene 1952, Section 64). 15. Van Inwagen 1988:261-62 uses the term 'frontier' in a mannersimilar to our 'penumbra', and uses 'fringe-referent'differently from but suggestive of our 'fringe'. 16. The constructionat the heart of the following proposition is similar to one used, for slightly differentpurposes, in van Inwagen 1988. and assignments induced by * in the 17. Here P* and a* are the maps on interpretations usual way. For instance, a*x =df (ax)*, v*P(d*) =df vP(d). 18. Thereis an assumptioninvolved here thattherewill always be distinct objects a, b not belonging to D. If we take ourbackgroundset theoryto be ZFC, andrecall thatD must be a set, there will always be such objects (for instance, D and ID}). 19. This argumentshows incidentallythat Ia, d} (and, for similarreasons, Ib, d}, will not belong to e* for any e-d, so that D* satisfies the nuclearitycondition. 20. Things would change radically if we were able to express more of the set-theoretic structureof objects in our language. Suppose, for instance, that we require enough objects in D to define inclusion in terms of overlap:d c e if for all gED, god only if goe. Then the following axiom will be verified:
Vxy. Vz(<x=z
<-4 <y=z)
-*

x=y

and we can define "inclusion"of one object in anotherby


XEY =df VZ- <X=Z
-*

<y=Z.

Then the existence or non-existence of cores can be stated in the language, and nuclearityis no longer an artifact.We intendto investigate these richermodels in a future paper.
21. To save space, we make the following convention:

4 and A-

o are opposites, as are

--b and AO. -4 stands for any opposite of b. 22. In view of the ubiquitousrole of < in this system it would perhapsbe more perspicuous to treatthese rules as structuralrules in the sense of Curry 1963, p.186.

348 / Peter W. Woodruffand Terence D. Parsons 23. The first of these claims is true in a strong sense; if we restrictourselves to formulas constructedout of atomic formulasof the form >s=t, then all other occurrencesof <, V, etc. can be eliminatedand (for these formulas)classical predicatelogic holds. Thus the pure theory of true identity (in this sense) is just isomorphic to classical pure identity theory. On the otherhand,the theoryof weak identityis wholly unconstrained except for symmetry, with the consequence that, in contrast to classical pure identity theory, three-valuedpure identity theory is undecidable. 24. Essentially the only axioms which can be added to pure classical identity theory are statementsabout the size of the domain, for instance: 'VxVyx=y'. 25. Incidentally,though from the proof theoreticpoint of view it is preferableto concentrateon weak identity,from a philosophical standpointthereis considerableevidence of notion. thattruedifference(the contradictory weak identity)is the morefundamental

References
Broome, John. (1984) Indefinitenessin Identity,Analysis 44, 6-12. Analysis 49, 112-19. Burgess, J.A. (1989) Vague Identity:Evans Misrepresented, Cook, Monte. (1986) Indeterminacyof Identity,Analysis 46, 179-86. Curry,H. B. (1963) Foundations of MathematicalLogic. New York,McGraw-Hill. Evans, Gareth.(1978) Can There be Vague Objects?Analysis 38, 208. Garrett,B. J. (1988) Vagueness and Identity,Analysis 48, 130-34. Analysis 49, 103-12. Johnson, Bruce. (1989) Is Vague Identity Inhoherent? Kleene, Stephen Cole. (1952) Introductionto Metamathetmatics.New York,Van Nostrand. Langholm,Tore. (1988) Partiality, Truthand Persistence, CSLI, Stanford Noonan, Harold. (1982) Vague Objects, Analysis 42, 3-6. Noonan, Harold. (1984) Indefinite Identity:A Reply to Broome, Analysis 44, 117-21. Noonan, Harold.(1990) Vague IdentityYet Again, Analysis 50, 157-62. Parsons,Terence. (1987) Entities without Identity,Philosophical Perspectives 1, 1-19. Parsons,Terenceand Woodruff,Peter.(1995) WorldlyIndeterminacyof Identity,Proceedings of the Aristotelian Society, Winter 1995, 171-91. Pelletier, Francis Jeffrey. (1989) AnotherArgumentAgainst Vague Objects, Journal of Philosophy lxxxvi, 481-92. Rasmusssen, Stig. (1986) Vague Identity,Mind 95, 81-91. Salmon, Nathan. (1981) Reference and Essence. Princeton,PrincetonU. Press. Thomason, Richmond. (1982) Identity and Vagueness, Philosophical Studies 42, 329-32. VanInwagen, Peter.(1988) How to Reason aboutVagueObjects,Philosophical TopicsXVI, 255-84. Woodruff,Peter W. (1969) Foundations of Three-valuedLogic. University Microfilms, Ann Arbor Woodruff,Peter.(1970) Logic and TruthValue Gaps, in KarelLambert(ed) Philosophical Problems in Logic, 121-42. Woodruff,Peter.Formal Systems for IndeterminateIdentity (in preparation).

You might also like