Nature and its Classification A Metaphysics of Science Conference On the Semantics of Natural Kinds: In Defence of the Essentialist Line TUOMAS E. TAHKO (Durham University) tuomas.tahko@durham.ac.uk http://www.dur.ac.uk/tuomas.tahko/ ABSTRACT Natural kind terms are important both for metaphysics and for semantics, but also notoriously hard to define accurately. In this paper I will look at, arguably, the best and most widely accepted solution to this problem, due to Putnam and Kripke (the focus will be on Putnam). I will show that, although this solution is very much on the right lines, it suffers from two serious drawbacks: it underestimates the capabilities of normal users of language, and fails to cash out all of its metaphysical implications. The central idea, towards which Putnam seems to hint but says very little of, is that our semantic framework concerning natural kinds reflects the categorical structure of reality, i.e. we classify natural kinds the way we do because the reality is organised accordingly. Putnam, however, thinks that only 'expert' speakers need to be aware of the 'essential nature' of natural kinds. This, as will be demonstrated, is unsatisfactory. The upshot of the paper is nevertheless positive, for with the appropriate revisions in place, it seems that we have a coherent and plausible story about the semantics of natural kind terms, as we can indeed ground them in ontology. 1
Natural kinds are both semantically and metaphysically a problematic class: they seem to elude simple, and sometimes even complicated definitions, yet intuitively they should have well-defined boundaries. The generally accepted story about the semantics of natural kinds is due to Putnam (1970, 1975b) and Kripke (1980) and I very much sympathise with the framework that they presented. However, there are philosophers who do not acknowledge this framework, and, more importantly, not even Putnam and Kripke seem to have realised its full metaphysical implications. Consequently, I think that it would be a good idea to look at the basics once again. Putnam introduced many of his ideas already in the paper 'Is Semantics Possible?' (1970). One of the first things that he points out is that although natural kinds such as lemons have 'defining characteristics', merely listing these characteristics can never be enough to define natural kinds, because there may be, for instance, abnormal members of the kind, such as green lemons (p. 140). Plausibly, the defining characteristics of natural kinds emerge because of some more fundamental features of the kind in question, what Putnam calls the 'essential nature' of the kind (ibid.). The problem that remains is that it is science which determines what these more fundamental features in fact are, and science is a fallible discipline. In other words, the most fundamental features of natural kinds are subject to revision. Now, clearly, the essential features themselves cannot change, but we might have gotten our story about them wrong. Thus we have no means to reliably fix our conceptual scheme according to the genuine essential features of natural kinds, yet we generally think that we use words like 'lemon' and 'tiger' accurately and that we do grasp the genuine essential features of the kind in question. This story is quite familiar to us and its upshot is as follows: 2
Even if cats turn out to be robots remotely controlled from Mars we will still call them 'cats' [...]. Not only will we still call them 'cats', they are cats [...]. But the fact that a term has several possible uses does not make it a disjunctive term; the mistake is in trying to represent the complex behaviour of a natural kind word in something as simple as an analytic definition. (Putnam 1970: 143.) This is the lesson of semantic externalism, but Putnam still needs to reconcile it with the ordinary usage of natural kind terms. The crucial idea here is to identify stereotypes (the characteristics of a normal member of a particular kind) with the correct natural kind. And here, new problems emerge. Consider 'aluminium' and a qualitatively indistinguishable metal 'molybdenum'. Putnam (1970: 150 ff) asks us to imagine a colony of Englishspeaking people on a spaceship, travelling towards a distant planet. None of them can recall the atomic weight or any other defining characteristics of aluminium or molybdenum. They have both these metals with them and they guess which one is which, incorrectly, as it happens. What can be done to preserve the 'normal' meaning of aluminium? There is not really anything that can be done if we want to maintain a purely conventionalist account: apparently the convention has changed within this colony. However, Putnam suggests that with the help of a test the colonists could be guided towards the normal use of 'aluminium'. This test is supposed to fix the extension of 'aluminium', thus: 'Meaning indeed determines extension; but only because extension (fixed by some test or other) is, in some cases part of the meaning (p. 62).' I take it that the test in question is some kind of a scientific test. And there has to be a test, as clearly the stereotype by itself does not suffice to fix the extension of natural kinds. But Putnam continues: 'Nothing normally need to be said about the extension, however, since the hearer knows that he can always consult an expert if any question comes up' (ibid.). 3
This suggests that no one else apart from experts can grasp extensions, no one else except experts really know how to use language correctly; indeed, no one else can have knowledge of essences. This seems to leave us in quite an awkward situation, for an expert on aluminium is probably not an expert on cats or whales, and an ordinary speaker is presumably not an expert on anything. The fact that we can consult such experts hardly gives us much comfort, for it would be quite a task to find one whenever we want to make sure that we are using natural kind terms correctly. Naturally this is not what Putnam had in mind, just the possibility of doing this is sufficient for him. But it seems that it gives us unreasonable liberty: we do not check our stereotypes very often, in fact, how are we even able to know when they fail? The colony of English-speakers was unaware of its mistake, but if we think about it, perhaps it was not a very serious mistake. It is important to note here that everyone in the colony already knew that the kind aluminium actually exists. Furthermore, and this is what Putnam does not take into account: they all had grasped the essence of the kind aluminium (given that they knew what aluminium is). That is to say that everyone can grasp essences, all rational human beings are capable of doing this, not only experts. What is (mostly) the task of experts, is to verify (or falsify) our initial classificatory scheme concerning natural kinds. Often this requires a lot of work, but once the work is done, anyone who understands the notion 'natural kind' certainly understands what natural kind terms refer to, namely the deep-structure of the kind its essential features. Thus, in one sense, the members of the English-speaking colony were able to use the concept 'aluminium' correctly at all times, because they knew that it has been verified that the natural kind aluminium actually exists. They failed simply in ostension: they pointed to the wrong material. 4
To make the case a bit more substantial, suppose that no one in the colony (or, indeed, on Earth) has ever heard of molybdenum, and they think that all the aluminium-like metal they have encountered is aluminium, but some of it is in fact molybdenum. This would perhaps be a mistake of a more serious kind, but in no way more disastrous for the account at hand, because we know that the empirical story is always subject to revision. If it turns out that what we think was aluminium is sometimes molybdenum, then the class of natural kinds would simply have a new member: molybdenum. But just the fact that we sometimes treated molybdenum as if it was aluminium does not mean that we had not grasped the essence of aluminium. In fact, this is ontologically quite uninteresting, and linguistically too. A scientist might be enthusiastic over a finding like this, but it gives us no reason to modify our ontology or semantics (except, of course, in the sense that we would need a word for the other aluminium-like metal: molybdenum). Let me try to explain how this account differs from Putnam's. What Putnam emphasises, especially in 'The Meaning of Meaning ' (1975b), is the social aspect of language, that is, extension is partly determined socially. This is already implicit in the use of 'stereotypes', which were introduced in his 'Is Semantics Possible?'. The problem with stereotypes is that they tend to be inaccurate, and in fact contingent. For instance, we associate all kinds of stereotypes with water: the stuff that comes from the tap, rains from the sky and fills the lakes. However, none of this is hardly essential for water. Of course, what is usually considered to be an essential feature of water is that it is H2O; and being able to distinguish water from liquids with different chemical composition, say XYZ, is something that Putnam associates with expert speakers. This is of course right because only experts can actually verify that water is H2O; only they know it by first hand experience. Consequently, Putnam's case for the social aspect of language is based on the fact that expert speakers give us new information about the world. All this may sound fine, 5
but it underestimates the skills of normal speakers. Consider tigers. My dictionary describes tigers as very large solitary cats with a yellowbrown coat striped with black, native to the forests of Asia. Most people would be quite happy with this definition. But scientifically, as well as in terms of the deep structure of natural kinds, it is clearly not a satisfactory definition. We would rather have to say something about the genetics of tigers. It should also be mentioned that there are eight different tiger subspecies (of which three are extinct) and so on. Again, this is something that expert speakers would tell us. Indeed, it seems to me that we cannot talk about anything (or at least about any natural kinds) without the help of these so called experts, because in the end, natural kind terms always reflect the scientific framework. But I do not think that this means that only expert speakers know what they are talking about. It seems to me that the semantics of natural kinds follow a very simple pattern. Putnam has outlined this pattern, but his account does not adequately explain how the expert speakers differ from ordinary speakers. Like I noted above, I do not think that expert speakers have a privileged access to natural kinds. Non-experts might be satisfied with the dictionary definition for 'tiger', which is more or less a description of what they look like, but even the dictionary definition contains one crucial word: 'cat'. 'Cat' is of course another natural kind term, which connects tigers with a broader classificatory framework. What I want to say here is that every speaker, be it an expert or a normal speaker, relies on the same underlying structure, the same classificatory framework, when trying to put tigers in the right place. Putnam hints towards something like this when he talks about semantic markers: 6
Not only do such features as 'animal', 'living thing', 'artifact', 'day of the week', 'period of time', attach with enormous centrality to the words 'tiger', 'clam', 'chair', 'Tuesday', 'hour'; but they also form part of a widely used and important system of classification. The centrality guarantees that items classified under these headings virtually never have to be reclassified; thus these headings are the natural ones to use as category-indicators in a host of contexts. (Putnam 1975: 267-68.) Putnam derives the idea of semantic markers from Fodor and Katz, and integrates it with his own idea of stereotypes. For Putnam, this is only a small clarification, but for my purposes, this is the central part of the theory. When someone asks what a 'tiger' is, I think that the question is really 'to which locker do tigers belong?'. Some stereotypes associated with tigers, such as 'carnivore', are in fact very heavily loaded with categorical information. For when I turn to my dictionary, the word 'carnivore' is explained to be associated with mammals of the order Carnivora, which comprises the cats, dogs, bears, hyenas, weasels, civets, raccoons and mongooses. So, the reference-fixing of natural kind terms clearly includes two stages: (1) the 'classification', i.e. to which 'locker' the natural kind term could belong to (2) the scientific account which verifies the connection between the most plausible potential 'locker' and the deep structure of the natural kind When we refer to tigers, we always aim to refer to the deep structure, the actual 'locker' that the natural kind 'tiger' belongs to. 1 The scientific explanation associated with that 'locker' gives us the details and makes sure that our initial classification corresponds with the actual categorical structure of reality. The latter part is, as has been noted, subject to revision. Accordingly, something like tigers turning out to be robots would not be disastrous for the picture. Indeed, the word 'tiger' would, at least first, still refer also to 1 Putnam (1990: 62-63) has shown some sympathy towards this sort of view. 7
the potential cat-like animal, although eventually this convention might change. Nevertheless, the revisability of the empirical verification is built in the framework. The upshot of this account is that the semantics of natural kind terms are fundamentally linked with the ontology of natural kinds. This might seem obvious, but it is far from it: in general, language is not a very reliable guide to ontology. Of course, even in this case the relationship between the semantics of natural kind terms and the categorical structure of reality which natural kinds are a part of is not from language to ontology, but rather from ontology to language. That is, we classify things into natural kinds because nature is in fact organised according to a certain categorical structure. This does not mean that our conceptual framework accurately corresponds with the actual categorical structure, but what is guaranteed, due to the self-correctiveness of science, is that slowly our framework approaches the actual structure of reality. Putnam's story about these matters is very much on the right lines, but I hope to have made it clear how we should develop on it. Firstly, both normal and expert speakers rely on the very same classificatory framework, and secondly, the most important features of this framework can be reduced to matters of ontology. With these revisions in place, I am happy to conclude that our understanding of the semantics of natural kind terms is in good shape. 2 2 I do not have the space here to address objections such as Mellor (1977), but I do not think that they pose serious problems for the revised view. Furthermore, in my talk I hope to have some time to discuss the implications of this view for the relationship between semantics and metaphysics. 8
References Kripke, S. (1980) Naming and Necessity (Cambridge, Mass.: Harvard University Press). Mellor, D. H. (1977) 'Natural Kinds', British Journal for the Philosophy of Science 28, pp. 299-312. reprinted in Pessin and Goldberg 1996: 69-80. Putnam, H. (1970) 'Is Semantics Possible?', H. E. Kiefer and M. K. Munitz (eds.), Language, Belief, and Metaphysics, (New York: State University of New York Press), pp. 50-63, reprinted in Putnam 1975a: 139-52. Putnam, H. (1975a) Mind, Language and Reality: Philosophical Paper, Vol. 2 (Cambridge: Cambridge University Press). Putnam, H. (1975b) 'The Meaning of Meaning ', K. Gunderson (ed.), Language, Mind and Knowledge, Minnesota Studies in the Philosophy if Science, vol. VII, (Minneapolis: University of Minnesota Press), pp. 131-93, reprinted in Putnam 1975a: 215-71. Putnam, H. (1990) 'Is Water Necessarily H2O?', in Conant, J. (ed.), Realism with a Human Face (Cambridge: Harvard University Press), pp. 54-79. 9