Reference

8. Naturalizing reference

From a naturalistic perspective, reference must ultimately be explained in scientifically acceptable terms. Attempted explanations have appealed to one or more of three causal relations between words and the world: historical, reliable and teleological.

Historical–causal theory of reference. Kripke, Donnellan and Putnam (see §4) did not claim to be naturalizing reference, but their theories – together with the role that perception of an object may play in determining the reference of demonstratives and referential descriptions (see §§5–6) – suggest the idea that reference might be explained naturalistically in historical–causal terms: a token refers to the object that played the appropriate role in causing it. But this idea, developed by Devitt, faces the qua-problem. In virtue of what is ‘Aristotle’, say, perceptually grounded in a ‘whole object’ and not a time-slice or undetached part of the object, each of which is equally present and causally efficacious? The problem is more pressing for natural-kind words. ‘Horse’ is grounded in a few horses. But those objects are not only horses, they are mammals, vertebrates and so on; they are members of very many natural kinds. Indeed, any horse is a member of indefinitely many non-natural kinds: it may be a pet, an investment and so on. In virtue of what is ‘horse’ grounded in an object qua horse, rather than qua mammal, pet, or whatever? So in virtue of what does it refer, as a result of such groundings, to all and only horses rather than all and only mammals, pets, or whatever?

Reliabilist theory of reference. Under the influence particularly of Fred Dretske (1981) and Jerry Fodor (1990), ‘reliabilist’, ‘informational’, or ‘indicator’ theories have been popular. The basic idea is that a token refers to objects of a certain sort because tokens of that type are reliably correlated with the presence of those objects; the tokens are ‘caused by’ those objects. The token ‘carries the information’ that a certain situation holds in much the same way that tree rings carry information about the age of a tree. There is a problem. How can the theory allow for error? Occasionally we see a muddy zebra and wrongly think ‘horse’. So, some zebras are among the things that would cause tokens of ‘horse’. What ‘horse’ is reliably correlated with is really the presence of horses, muddy zebras, the odd cow in bad light and so forth. So according to reliabilism, it should refer to horses, muddy zebras, the odd cow and so on (with the result that it was not wrong to think ‘horse’ after all). The problem is that many things that a token of a certain type does not refer to, including some denizens of Twin Earth, would cause a token of that type (see Semantics, informational).

Teleological theory of reference. Most fully developed by Ruth Millikan (1984), teleological theories explain the reference of a token in terms of its function, where that function is explained causally along Darwinian lines: a token’s function is what tokens of that type do that explains why they exist. This theory deals neatly with the problem of error because something – for example, sperm – can have a function which it does not reliably perform. An immediate consequence of the theory is that a token of a type that has not evolved will lack a referent. So the ‘thoughts’ and ‘utterances’ of an exact replica of Russell created by some cosmic accident would have no reference. This strikes many as implausible but is accepted by the theory’s proponents. To complete the theory it must be shown that tokens – even a belief like ‘computers make writing easier’ which could not plausibly be taken as innate – have a function in the required biological sense and that this function does indeed relate the token to its referent. Millikan has attempted this formidable task (see Semantics, teleological).