The British Society for the Philosophy of Science

Natural Kinds Author(s): D. H. Mellor Reviewed work(s): Source: The British Journal for the Philosophy of Science, Vol. 28, No. 4 (Dec., 1977), pp. 299312 Published by: Oxford University Press on behalf of The British Society for the Philosophy of Science Stable URL: http://www.jstor.org/stable/686875 . Accessed: 01/03/2013 16:21
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
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.

Oxford University Press and The British Society for the Philosophy of Science are collaborating with JSTOR to digitize, preserve and extend access to The British Journal for the Philosophy of Science.

http://www.jstor.org

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Brit. J. Phil. Sci. 28 (1977), 299-312 Printedin GreatBritain

299

Natural Kinds*
by D. H. MELLOR i Some notable philosophers have recently used new arguments to revive essentialism, and have prescribed their essences for a variety of metaphysical fears and ailments. The essence of a self has been said to guarantee its ancestry; mental essence has been promoted as a sure defence against materialism; and diamonds have been warranted in all possible worlds against being paste. I mistrust these prescriptions, especially the claims made for their active ingredients: possible worlds and necessary identity. However, I don't mean here to resist all applications of these notions, nor to dispute all forms of essentialism. I mean only to supply an antidote to the natural kind essences widely advertised by Professors
Kripke ([*971], [1972]) and Putnam [19751-

Kripke and Putnam claim that natural kinds have essential properties; that is, properties which nothing can lack and still be of the kind. The kinds involved include the traditional natural kinds: elements and compounds like gold and water, and biological and botanical species like tigers and elm trees. Modern essences, however, come in a wider range. Essential properties are claimed also, for example, for temperatures and lengths.1 These do not traditionally form natural kinds, but it will be convenient here to stretch the term to match the doctrine. Properties alleged to be essential typically involve the microstructure of things. Having atomic number 79 is said to be the essential property of
gold (Kripke [1972], PP. 327), being H20 the essential property of water

(Putnam [1975], p. 233). Genetic makeup similarly provides essential properties for animals and plants,2 and mean molecular kinetic energy for
Received 3 November 1976 * Versions of this paper have been given to seminars at Columbia, Stanford, the Australian National, Cambridge and Oxford Universities; and in August 1975 to the Annual Conference of the Australasian Association of Philosophy. I am much indebted to all those who have commented on the paper on these and other occasions. x 'It is going to be necessary that heat is the motion of molecules' (Kripke [I1971], p. 16o). The proposition 'The standard meter rod (S) is I meter long' Kripke ([1972], p. 275) takes to be a contingent a priori truth-a priori because S sets the standard, contingent because S might have been shorter, longer or non-existent. But if S in fact fixed the extension of 'i metre' as Putnam prescribes (see below), then whatever shared properties in fact make distant objects the same length as S would be essential properties of being I metre long. 2 . would not be .. animals with the appearance of cats but reptilic internal structure ... cats, but "fools cats"' (Kripke [I972], p. 321).
U

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

300 D. H. Mellor temperature. These essential properties of natural kinds are supplied by the natural sciences: 'In general, science attempts, by investigating basic structural traits, to find the nature, and thus the essence (in the philosophical sense), of the kind' (Kripke [1972], p. 330). The necessity of essential properties is metaphysical, not epistemic. The claim is that things of a kind have its essential properties in all possible worlds, not that its essential properties are knowable a priori.1 In particular it is not supposed to be analytic to ascribe its essential properties to things of a kind. Kinetic theory gives the essence of temperature, not the meaning of 'temperature'. Essentialism needn't therefore dispute Quinean critiques of the analytic-synthetic distinction on the one hand, nor on the other need it plague theoretical conflict with the problems of incommensurability (Feyerabend [1962], Kuhn [1962]) or indeterminacy of translation (Quine [I960]): Note that on the present view, scientific discoveries of species essence do not constitute a 'change of meaning' . . . We need not ever assume that the biologist's denial that whales are fish shows his 'concept of fishhood'to be different from that of the layman; he simply corrects the layman, discovering that 'whalesaremammals,not fish' is a necessarytruth. Neither 'whalesaremammals' nor 'whales are fish' was supposed to be a priori or analyticin any case (Kripke P. [I1972], 330). So proponents of rival theories are not doomed by essentialism to Kuhn's [1970] and Feyerabend's [1970] dialogue of the deaf. We need not, therefore, continue that prolonged dialogue here. Essentialism about kinds has various sources. It derives partly from the plausibility of examples such as those given above. Some among the properties common to things of a kind undoubtedly matter more than others. In particular, some will be more central than others to a theory which explains the properties and relations of things of that kind. I consider in section 7 whether properties being important in that sense is in fact best explained by, and thereby lends support to, the claim that they are essential properties of the kind; and I conclude that it is not. Individual essences are a classic source of essentialism about kinds. Some kinds may provide criteria for the reidentification of things of that kind, such that no thing of the kind could survive change in the specified respect. It is arguable, for example, that a man could not survive the loss z
1 Kripke [I971], pp. 150-1; [I972], pp. 260-3. I have used Dummett's term for the latter

notion, since I incline to accept his account, on which "epistemicnecessityis a stronger but notion . . . a statementmay be [metaphysically] not epistemicallynecessary,but the converse could not occur. Kripke, however, claims the propertiesof being a priori and being necessaryto be quite independent"(Dummett [1973] p. 121).

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Natural Kinds 301 or drastic transformation of the body whose spatio-temporal continuity settles questions of human reidentification. I don't think that is true, but even if it were, it would follow neither that any men, nor that all men, must have bodies of the kind specified. In the first place, John's inability as a man to becomea beetle is compatible with the possibility of his always having been one; in the second, there could still have been men (other than the men there actually are) who lacked these bodily features and would be reidentified over time by different criteria. Anyway, arguments from individual essence, if they worked at all, would not work for kinds, such as water and gold, that provide no criteria for the reidentification of things. Perhaps there are characteristics a gold cup cannot come to have, but that will not show that other gold objects cannot have them. There are, I believe, no sound inferences from individual essences to kind essences: but that point is not new, and I need not argue it further here. My concern here is with two other, newly fashionable, arguments for essentialism about kinds. One, due to Putnam [1975], derives essentialism directly from a theory about how the extension of kind terms is fixed. The other, due to Kripke ([i97I], [I1972]), derives it indirectly via a theory of the singular reference of natural kind and other seemingly general terms, from whose necessary self-identity essentialism is taken to follow: When we have discovered that heat is molecular motion we've discovered an identification which gives us an essential property of this phenomenon. We have discovered a phenomenon which in all possible worlds will be molecular motion - which could not have failed to be molecular motion, because that's what the phenomenonis (Kripke [1972], p. 326). Putnam's theory of the extension of kind terms, and Kripke's theory of their reference, are alike in denying traditional accounts that make the reference (or extension) of terms a function inter alia of something like their Fregean sense.1 As applied to kinds in particular, the new theories deny that the extension of kind terms is any function of descriptions believed by their users to be true of things of the kind (Putnam [1975], p. 22z). Fregeans, who believe the contrary, need not of course deny that there are non-analytic essences of kinds (Dummett [1973], p. 117), but Fregean theories of how kind terms get their extension give no especial reason to think there are any. Fregean theories in general, and description theories of kind terms in particular, yield necessity only as a by-product
1

I say 'inter alia' because Fregean reference or extension is obviously a function also of context (e.g. in indirect speech, according to Frege, a name refers to what is normally its sense) and of what the world contains (e.g. whether 'gold' applies to my tiepin depends on whether my tiepin is gold). Cf. Dummett [1973], ch. 5, 9.

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

302

D. H. Mellor

of analyticity. (Had our Fregean sense of 'water' made us apply it only to what we believe 'H-20' applies to, then-at least for us in this worldwater would in all possible worlds have been H20. But that, as we have seen, is not why essentialists think being H2O is of the essence of water.) To provide essence without analyticity, an alternative is needed to the Fregean sense of what water is. Putnam to that end tells two tales designed to bury Frege, as a prelude to recommending his own essentialist account of natural kinds. In fact, Putnam buries Frege alive and well; and that fact must be shown first, before we can profitably turn to the deficiencies of Putnam's and Kripke's rival theories. (My arguments against Putnam's interpretation of his two tales overlap with those of Zymach [1976], but my further purposes make it desirable to restate them here in my own way.) 3 Putnam's tales are aimed at the idea of a kind term's extension being any Fregean function of its users' beliefs. The tales therefore present cases where such an extension differs for two groups of users with relevantly identical beliefs. First, we suppose a Twin Earth somewhere, which is just like Earth except for a different microstructure, XYZ, of what they too call 'water'. Macroscopically XYZ is indistinguishable from H20, and it plays just the same part in Twin Earth life that HO20does here. By 195o, however, it has become common knowledge on each planet that the other lives off different stuff. But back in 175o no one knew about the microstructure of water, and each planet had identical beliefs about the stuff they so called. Yet "the extension of the term 'water' was just as much H20 on Earth in 1750 as in 1950; and the extension of the term 'water' was just as much XYZ on Twin Earth in 1750 as in 1950" (Putnam [1975], p. 224). Now the local stuff was no doubt in the extension of 'water' as used on each planet in 1750. If the stuff on the other planet was different in kind, it was presumably not in fact in the term's extension, even though the users then would have mistakenly thought it was. Hence the antiFregean conclusion: 'water' can have different extensions in the same world for different users who give it the same Fregean sense. I agree that 'water' had (tenselessly) the same extension in 175o as it had in 1950; what I deny is that at either time that extension was different on Earth and on Twin Earth. The fact that Twin Earth's 195o beliefs about local water differed from ours doesn't begin to show that the extension of their term 'water' differed from that of ours. It doesn't even follow that the senses of the term differed; and if they did, the whole point of the sense/reference distinction is to allow sameness of reference (or extension) to accompany difference of sense. It is indeed quite plain to

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Natural Kinds 303 my Fregean eye that in 1950, as in 1750, 'water' had the same extension on Twin Earth as it had here. There was water on both planets alike, and there still is. We simply discovered that not all water has the same microstructure; why should it? Because its microstructure is an essential property of water? Well, that is what's in question. Fregeans need not resort to science fiction to recommend their reading of this tale. There is a perfect precedent in the discovery of isotopes. If Zymach's ([1976], p. I2z) heavy waters are too rare and exotic to convince, try the two common isotopes of chlorine. Note that in these real cases the various isotopes occur together in natural samples; they aren't segregated onto separate planets. It is therefore undeniable that the extension of 'chlorine' included both isotopes before their discovery, and so presumably includes both isotopes now.' What Putnam must say is that chlorine and water have been found not to be natural kinds after all, but rather mixtures of natural kinds. But in that case, as Zymach ([1976], p. I22) observes, it will very likely turn out that we have no natural kind terms. Anyway, the pertinent point is that the first Twin Earth tale doesn't compel that conclusion, which it would have to do to dispose of Frege. The Fregean reading that Putnam overlooks is prima facie at least as plausible as his own. Putnam, however, has another Twin Earth tale that won't admit this Fregean reading. This time we suppose that aluminium and molybdenum are practically indistinguishable, that molybdenum is as common on Twin Earth as aluminium is here, and consequently that on Twin Earth all our Earthly uses of the two metals are interchanged. Moreover, like Americans, Twin Earth men don't call aluminium 'aluminium'; they (unlike Americans) call it 'molybdenum'. The term 'aluminium' they reserve for molybdenum. Now most people, both here and there, can't tell the metals apart; call these people 'laymen'. Laymen here are therefore in the same psychological state about aluminium that laymen there are about what they also call 'aluminium'. Yet the extension of 'aluminium' as laymen use the term there is indubitably molybdenum and thus quite different from the extension of 'aluminium' as laymen use the term here. My Fregean waterworks, chlorinated or not, will not wash with this tale. Our laymen know that aluminium isn't molybdenum, even if they can't tell"what the difference is. So we can't make our term 'aluminium' apply also to Twin Earth molybdenum as we made 'water' apply also
1

The sense of 'chlorine' might of course have changed in the wake of this discovery, to make the term apply only to the more common isotope; just as the sense of 'water' might have changed to exclude XYZ or DzO. But that is not what happened, and anyway not what Putnam needs. If changes of belief about the microstructure of kinds do produce changes of extension in kind terms, that rather recommends a Fregean view.

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

304 D. H. Mellor to XYZ. The reason of course is that there are high priests as well as laymen, experts who can tell the difference. Our experts fix the extensions of 'aluminium' and 'molybdenum' for laymen here; experts on Twin Earth fix them the other way round for laymen there. There is, as Putnam ([19751, p. 227) puts it, a "division of linguistic labor". Very well. It need not be my beliefs that fix the reference or extension of terms which I can use quite well in my limited way. So I defer to experts, whose job it is to say what such a term really applies to. The reference or extension in any possible world of the term as we use it may nevertheless still be some Fregean function of our experts' beliefs. In Australia, for example, to be "back of Bourke" is to be way out in the outback. That Bourke is at an edge of Australian civilisation is all I know about that place, certainly not enough to enable me to tell Bourke from several other places. Yet I can still refer to Bourke, as I have just done, by taking it to be whatever place would best fit more expert geographers' beliefs about it. Our knowledge of most things and places must be like this; certainly most of our knowledge of natural kinds. So no doubt the labour of reference is divided, as Putnam says, but it may be a Fregean labour for all that (Dummett [19731, PP. 138-9; [i974], PP- 530-i). 4 Fregeans can cope with Putnam's Twin Earth tales. Frege is still in the ring; so how does Putnam fare on points? His rival theory gives the extension of natural kind terms in two stages. First, archetypes in this world, paradigm specimens of the kind. Then anything, in any possible world, that has a suitable "same-kind" relation to the archetypes. It is for science to tell us what the same-kind relation is for any category of kinds (for H20, Putnam proposes, implausibly enough, a same-liquid relation). Generally, Putnam assumes that the relation will specify a shared microstructure. But whatever shared properties Putnam's same-kind relation picks out will be essential properties of the kind, since the relation is assumed to be an equivalence relation that holds across all possible worlds (Putnam [19751, p. 232), not just in this one. So not just actual specimens of the kind share the specified properties: nothing could be of the kind and not share them with the kind's archetypes. Putnam's necessity is metaphysical, not epistemic. A natural kind may be known long before its essential microstructure is known. Putnam's theory is radically anti-Fregean: given archetypes, a kind term's extension is fixed by its same-kind relation, regardless of what anyone believes. Putnam's argument for kind essences credits kinds both with archetypes and with cross-world equivalence relations holding between all things of the kind. Real natural kinds need have neither. Take archetypes. Their

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Natural Kinds 305 role is to fix the kind term's extension without recourse to its Fregean sense. Putnam takes it that they must therefore be in this world, so that they can be picked out ostensively. Thus pointing out Lake Michigan as an archetypal sample of water (Putnam's example!) fixes the extension of 'water' regardless of its sense (just as naming the lake 'Lake Michigan' is supposed to fix the reference of that expression regardless of its sense). Then the rest of the extension of 'water', in any possible world, is just what stands in the relevant same-kind relation to this archetype. That is why Putnam ([19751, P. 234) says that kind terms are 'indexical', like 'I', 'now', 'here' and 'this'. Ostensive reference, to just this archetype in this world, is thus essential to the mechanism of Putnam's essentialist theory. He must show, therefore, that our use of kind terms actually incorporates it. An extension of Kripke's causal theory of naming offers to show this. The theory has some irrelevantly contentious aspects, about how users of names pass them on'; all we need here is that roughly, some archetype must be causally "upwind" of any use of a natural kind term. Thus our uses of 'water' and 'aluminium' are supposed to derive causally from our (or our experts') causal acquaintance with archetypal specimens of H20 and aluminium respectively; and that is supposed to be why H20 and aluminium are what we refer to by those terms. The corresponding Twin Earth uses derived causally from archetypes of XYZ and molybdenum; which is why those are the kinds they refer to by the same terms. That is the theory. Unfortunately, archetypes do not constrain our use of natural kind terms in that way. True, botanists designate type specimens of plant species, and geneticists designate cultures to exemplify gene-types (Jardine [1975]). But these specimens are causally downwind of the usage they are supposed to constrain. They are chosen to fit botanical and genetic knowledge, not the other way round. They are certainly not the specimens whose classification caused the corresponding kind terms to be used in the first place; they may well indeed not even be of the same kind as those specimens. Hence, as Jardine [1975] and Zymach ([1976], pp. 123-4)
1 'When the name is "passed from link to link", the receiver of the name must, I think,

intend when he learns it to use it with the same reference as the man from whom he heard it' (Kripke [1972], p. 302). Intention isn't enough, however: 'I am asked to name a capital city and I say 'Kingston is the capital of Jamaica'; ... [I] said something strictly and literally true even though it turns out that the man from whom I picked up this scrap of information was actually referring to Kingston-upon-Thames and making a racist observation' (G. Evans [I973], p. 194). 'We are left with this: that a name refers to an object if there exists a chain of communication, stretching back to the introduction of the name as standing for that object, at each stage of which there was a successful intention to preserve its reference. This proposition is indisputably true; but hardly illuminating' (Dummett [I1973], p. 151). See also J. E. J. Altham [1973], PP.
209-25.

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

306

D. H. Mellor

have observed, our most authoritative specimens of a kind might on Putnam's account not even be of that kind. So some archetypal natural kinds have the wrong archetypes; others have none at all. Consider elements high in the periodic table, that do not occur in nature and have never been made. We have names for them, but there may never be archetypes to constrain our use of the names. Even if specimens eventually appear, the discovery, creation or synthesis of previously unknown fundamental particles, elements and compounds can surely be predicted. The term 'neutrino' applied to just the same particles when it was used to predict their existence as it has applied to since their discovery. Ostensive reference (say to a bubble chamber photograph) could not have fixed its extension then; why suppose exactly the same extension is fixed that way now? Even if we were to grant Putnam his archetypes, however, his essentialism would still fail to follow. No reason is given why particular properties must be common to all things in all possible worlds that are of the same kind as the archetypes. Suppose that all samples of water in fact share ten "important" properties, but that water could lack any one of them, so that only the disjunction of all conjunctions of nine of them is essential. In this case of course sameness of kind is not the equivalence relation
Putnam ([i9751, p. 231) says it is, since it is not transitive; two merely

possible samples of water could differ in two of the ten properties. But Putnam's account doesn't in fact provide transitivity, since what makes things water in other possible worlds is their likeness to archetypes in this world, not their likeness to each other. To claim that the relation is an equivalence relation, so that archetypes have to share the same properties with all possible samples of the kind, is just gratuitously to assume the essentialist conclusion. 5 Putnam's account of the extension of kind terms both fails to be true and fails to entail essentialism. I turn now to Kripke's derivation of essentialism from the necessary self-identity of natural kinds. In evaluating his argument it is essential to keep the reference of a kind term clearly distinguished from its extension; to which end I hereafter distinguish the supposedly singular term 'water' from the corresponding predicate '... is water', and likewise for other kind terms. Now Kripke ([1972], p. 349) admits that a causal mechanism is not needed to secure the reference of natural kind terms. 'Neutrino' could be introduced, as it was, by theoretical description and still be applied "rigidly" in Kripke's sense. That is, we can still consider the consequences of just that kind of particle failing in this or that respect to satisfy the theoretical descriptions that in

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Natural Kinds 307 fact served to identify it. 'Neutrino', so understood, is for Kripke a nonFregean rigid designator, since its reference in other possible worlds is not constrained to satisfy these theoretical descriptions, which must be supposed to provide its Fregean sense. For this to work, of course, there must actually be neutrinos, near enough as specified. They needn't be in the observed past, to serve as archetypes; but they do need to be somewhere in this world, past, present or future. Otherwise the requisite uniqueness of reference would not be secured. Many different kinds of particle will satisfy our theoretical descriptions of neutrinos in various possible worlds; and nothing but the reality of one of these will single it out as the unique referent of the term 'neutrino'. Were there in fact no neutrinos, the term could for Kripke no more designate a natural kind than 'unicorn' can (Kripke [1972], p. 763)However, there are neutrinos, just as there is H20. Let us therefore grant for the moment that 'H20', like 'neutrino', can be made a nonFregean rigid designator by theoretical description and Kripkean fiat. Grant also that 'water' is such a designator, in this case perhaps even because our use of the term derives causally from Putnamesque archetypes like Lake Michigan. What more is needed for 'water' and 'H20' to designate the same kind? And how is sameness of kind related to coextensiveness in the corresponding predicates '... is water' and '... is H20', which is what matters for essentialism? We must in fact tackle the latter question first in order to answer the former. As referents of singular terms, kinds are obscure entities, not notably clearer than properties or attributes in their criteria of identity. If the necessary self-identity of kinds is to have any implications for essentialism, these criteria will have to be spelled out in terms of predicate extensions. One such criterion presents itself at once. For water to be the same natural kind as H20, it seems at least to be necesssary for '. . . is water' and '. . . is HO20' to be coextensive in all possible worlds; otherwise some possible world would contain something whose membership of the kind depended on what the kind was called, which seems implausible. But then it looks as if 'Water is H20', construed as an identity statement linking rigid designators, already and trivially entails that all samples of water in all possible worlds are also samples of H20. Far from the necessity of this identity establishing that being H20 is an essential property of water, that is just what must be the case for the identity claim to be true at all. It is indeed not clear what more the metaphysical necessity of 'Water is HO20'could consist in, since we have already used up our possible worlds in saying what makes it true. No doubt it is necessary if true; but are only because if in this world '... is water' and '... is HO20' coextensive-

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

308

D. H. Mellor

in-all-worlds, so they are in all other worlds.' Because the identity criteria of kinds have thus to be given in terms of their extensions, there is no useful inference to essentialism from the necessity of identity, even if the kind terms involved are admitted to be rigid designators; if anything, the inference will be the other way. One might try arguing, however, that the requirements of natural kind identity have been pitched too high. Perhaps what is needed is coextensiveness, not in all possible worlds, but in this world and in those nearest to it. After all, science is supposed to give us essences; yet the most scientists can show us in fact is lawful coextensiveness between '. . . is water' and '. .. is H20'. That is, we suppose they can show that not only are all samples of water samples of H20, but that if anything were a sample of water it would be a sample of H20. Now that need not be a claim about all possible worlds, since the consequent of a true subjunctive conditional need not be true in all possible worlds in which the antecedent is true; it need only be true in those worlds which are sufficiently like ours (Lewis [1973]). Waiving the difficulties of characterising lawfulness in terms of truth in nearby worlds without begging the question, suppose for the sake of argument that the lawfulness of 'All and only samples of water are samples of H20', so characterised, suffices for the identity of water and H20. Will the rigidity of 'water' and 'H20' now secure necessity for this identity and thus coextensiveness in all worlds, however remote, for '... is water' and '... is H20'? To see that it will not do the latter, we must appreciate how loose the connection is between the reference of 'water' (say) and the extension of '... is water' in various possible worlds. Obviously there could be much more water than there is, or much less. That is to say, there are possible worlds with samples of water that don't exist in this world; and others in which some real watery individuals (like Lake Michigan) don't exist at all. Consequently, in yet other possible worlds, there is water all right, although no individual sample of it is identical with any of the real samples that we have. In each of these worlds the singular term 'water' has the same reference (namely, of course, water-whatever that is), while the extension of the predicate '. .. is water' may differ totally from what it is here or in other possible worlds. Now we are at present supposing, for the sake of Kripke's argument, that coextensiveness between '... is water' and '. . . is H20' in all worlds is not required for the identity of water and H20, only coextensiveness in
I ignore as incredible accounts of metaphysical (as opposed to epistemic) necessity in which this does not follow, i.e. in which the accessibility relation between possible worlds is not transitive (Hughes and Cresswell [1968]); but see below.

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Natural Kinds 309 this and some suitable class of nearby worlds. We now see that in other worlds the extension of '. . . is water' may be quite different from its extensions hereabouts; and so may that of '. .. is H20'. And whatever the necessity of identity may do to secure that 'water' and 'H20' have the same reference in these other worlds, it will do nothing to secure that '. . . is water' and '. .. is H20' have the same extensions, which is what essentialism needs. 6 Even if 'water', 'H20' and similar kind terms were rigid designators, therefore, essentialism would get no help from the supposed necessity of identities like 'Water is H20', however the reference of the singular terms involved is related to the extension of the corresponding predicates. But there is anyway no good reason to admit that these terms are nonFregean rigid designators. Kripke implies ([1971], pp. 146-8) that without rigid designators we would need Lewis' [1973] "counterparts" in order to state counterfactuals about things (and kinds). I share Kripke's distaste for counterparts, but this is a false dichotomy. Fregean names may designate the same things or kinds in many possible worlds, namely in all those that can be specified by Fregeanly intelligible counterfactuals about them. On a "cluster" version of the description theory (e.g. Searle [1958]), taking account of Putnam's division of linguistic labour, such counterfactuals can suppose the lack of almost any property of the thing, or of things of the kind, and of all such properties attributed by any one nonexpert speaker. That is quite enough to give a specious appearance of nonFregean rigidity to what may in fact be Fregean names. Of course it is trivially true that 'water' applies to water whatever intelligible counterfactual supposition we make about it. The question is whether something like a Fregean sense of 'water' limits the range of such suppositions. I see no reason to deny that it does, and equally none to assert that anything about water itself makes the antecedents of any such counterfactuals necessarily false. 7 We have seen that essentialism can be extracted neither from Putnam's archetypes, nor from a merely stipulated rigidity of reference via modish truisms about identity. The existence of essences in Kripke's and Putnam's theories is no more than a gratuitous assumption on their part. Its appeal lies solely in that of the stock exemplars of essential properties; and that appeal, I shall argue in conclusion, is specious. In biological species, for example, there is a distinct dearth of suitable properties shared even in this world. Capacity to interbreed with fertile offspring is the obvious candidate, but even that is well known to lack

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

3 o D. H. Mellor the transitivity Putnam's same-kind relation would need in order to yield essentialism. It usually doesn't hold between all and only members of the same species here, never mind in other possible worlds. Nor is coming of a common stock plausibly essential to a species, which could easily have been cross-bred from independent mutations (pace Dummett [1973], pp. 143-5). (I fear this suggestion derives partly from the idea (Kripke [1972], pp. 312-4) that an offspring's parents are essential to it; of which idea it perhaps suffices to remark that 'If John hadn't been a Kennedy he wouldn't have been shot' is a plainly intelligible contingent statement (cf. Dummett [1973], P. I32).) Elements and chemical compounds offer more scope for essentialism, since all their specimens at least share some properties in this world. Of properties supposed to be shared in all possible worlds, we have seen that microstructure provides the stock exemplars; it is worth asking why. Scientists commonly employ a principle of "microreduction",1 i.e. roughly the principle that properties of things should be explained in terms of the properties and relations of their spatial parts. Many properties can indeed be so explained; and the assumption that they can underlies standard techniques for studying things in convenient (e.g. laboratory) isolation from their normal surroundings. Microreductive theories are thus relatively easily testable, which is a well-known Popperian virtue in science. It is therefore both good method to look for microreductive explanation of kinds of things and an importantly pervasive fact that they can be found. Where, moreover, such explanation is both comprehensive and deductive, we may be able to replace reference to things of a kind with reference to their parts. So it is, roughly, that reference to gas samples was made eliminable, by the classical kinetic theory, in favour of reference to gas particles. Suppose we now adopt a Quinean view of ontology, and admit only kinds of things that need to be referred to in stating what we know. Since kinetic theory makes reference to gas samples redundant, they disappear from our ontology. Similarly, let us suppose, with water and gold. Microreductive theories make reference to anything more than H20 molecules and gold atoms redundant. There is, we might say, nothing to water and gold but the particles of which they are composed. And this, I reckon, is the source of the idea that water would have to be H2O in any possible world, because H20 is all there is to water. The inference is specious. This way of removing items from our ontology requires deducibility. If all the macroscopic properties of a kind are not
1

Schlesinger [1963], ch. 2. The next four paragraphs condense an argument given more fully in my [1973], PP. Iio-iz.

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

Natural Kinds 311 deducible from its microstructure, reference to things of the kind is still required. And if they are deducible, then they occur in any possible world the microstructure occurs in. So if the microstructure is essential for this reason, so are all the macroscopic properties it explains. So-called 'essential' properties are thus really no more essential than any other shared properties of a kind. They are just properties ascribed by the primitive predicates in a comprehensive deductive theory of the kind. That is presumably what makes them what Putnam ([19751, P. 232) calls 'the important physical properties' of the kind; and the principle of microreduction will no doubt ensure that such properties will most commonly concern the kind's microstructure. Our anti-Fregean essentialists must now suppose a property's "importance", so construed, to be a feature of the world independent of our beliefs and theories. If we knew everything, perhaps that would be so, as Ramsey ([1978], ch. 6A) thought in 1928. But as Ramsey ([1978], ch. 6B) observed in 1929, we don't know everything. Even microreductive theories cannot be construed as claiming completeness merely by virtue of claiming truth. There may after all be endlessly smaller scale detail to the world than any such theory describes. Its presently primitive predicates may cease to be primitive at any time, when it is itself explained by some deeper theory, without at all impugning either its truth or its explanatory value. Water may still be H20, however much we subsequently learn of the nuclear structure of hydrogen and oxygen atoms.

8 I have tried in this paper to dispose of some of modern essentialism's newer and more seductive arguments. Putnam's Twin Earth tales do not, as he supposes, dispose of Fregean alternatives to essentialist theory. His own account of the extension of natural kind terms is false of nearly all natural kinds and would not yield essentialism even if it were true. Kripke's theory of the reference of kind terms likewise fails to yield essentialism as a product of the necessary self-identity of natural kinds. The stock candidates for essential properties, moreover, are either not even shared in this world by all things of the kind, or their status is evidently more a feature of our theories than of the world itself. In short, our essentialists' premises are false, their arguments invalid, and the plausibility of their conclusions specious. Their essences can go back in their Aristotelian bottles, where they belong. University of Cambridge

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions

312

D. H. Mellor

REFERENCES J. E. J. [1973]: 'The causal theory of names II', Aristotelian Society Supplementary Volume, 47, pp. 209-25. DUMMETT, M. [1973]: Frege: Philosophy of Language, London: Duckworth. M. DUMMETT, [I1974]: 'Postscript', Synthese, 27, pp. 523-34. EVANs, G. [1973]: 'The causal theory of names I', Aristotelian Society Supplementary Volume, 47, pp. 187-2zo8. FEYERABEND, P. K. [1962]: 'Explanation, reduction and empiricism', in Feigl, H. and Maxwell, G. (eds.): Minnesota Studies in the Philosophy of Science Vol. 3, pp. 28-97, Minneapolis: University of Minnesota Press. FEYERABEND, P. K. [1970]: 'Consolations for the specialist', in Lakatos, I. and Musgrave, A. (eds.): Criticism and the Growth of Knowledge, pp. 197-230. Cambridge: Cambridge University Press. to HUGHES, G. E. and CRESSWELL, M. J. [1968]: An Introduction Modal Logic. London: Methuen. JARDINE, N. [1975]: 'Natural kinds, rigid designation and the growth of science', Paper delivered to Cambridge Moral Sciences Club. KRIPKE, S. [1971]: 'Identity and necessity', in Munitz, M. K. (ed.): Identity and Individuation, pp. I35-64. New York: New York University Press. KRIPKE, S. [1972]: 'Naming and necessity', in Davidson, D. and Harman, G. (eds.). Semantics of Natural Language, pp. 253-355; 763-9. Dordrecht: Reidel. KUHN,T. S. [1962]: The Structure of Scientific Revolutions. Chicago: University of Chicago Press. KUHN, T. S. [1970]: 'Reflections on my critics', in Lakatos, I. and Musgrave, A. (eds.); Criticism and the Growth of Knowledge, pp. 231-78. Cambridge: Cambridge University Press. LEWIs, D. K. [1973]: Counterfactuals. Oxford: Blackwell. D. MELLOR, H. [1973]: 'Materialism and phenomenal qualities II', Aristotelian Society Supplementary Volume, 47, pp. 107-19. PUTNAM,H. [1975]: 'The meaning of "meaning"', Mind, Language and Reality, pp. 215-71. Cambridge: Cambridge University Press. QUINE,W. V. 0. [1960]: Word and Object. Cambridge, Mass: M.I.T. Press. RAMSEY, F. P. [I 1978]: Foundations: Essays in Philosophy, Logic, Mathematics and Economics. London: Routledge and Kegan Paul. SCHLESINGER, G. [1963]: Method in the Physical Sciences. London: Routledge. SEARLE, J. R. [1958]: 'Proper names', Mind, 67, pp. 166-73. E. ZYMACH, M. [1976]: 'Putnam's theory on the reference of substance terms', Journal of Philosophy, 73, pp. 116-27.
ALTHAM,

This content downloaded on Fri, 1 Mar 2013 16:21:24 PM All use subject to JSTOR Terms and Conditions