A functional role theory of content identifies content with the role of a representation in cognition. In the essay, I develop a causal role theory of content. But it is possible to identify content with other sorts of relations; in particular, several advocates of a functional role view have chosen to identify a representation's content with its *inferential role*.(31) However, as I will now argue, this is a bad choice--the notion of an inference is, in fact, not available to the naturalist.
According to an inferential role account, a (mental) term's meaning for an individual is identical to its role in the totality of inferences that the individual is disposed to draw, or accept or reject. The result is a naturalistic conception of meaning which reduces the content of terms to the roles of terms in an individual's patterns of inference-making.
How plausible is this account? We may begin by noting two general drawbacks which may lead us to question the choice of inferences as role-specifiers. First, this view implies that only a relatively select group of beings are capable of having representations, namely those that are capable of drawing inferences. If 'inference' is understood in a strict sense, then this may imply that only those beings with relatively sophisticated language and reasoning skills are capable of having representations. Which is to say that the inferential role view may imply that among known living creatures, only humans are capable of representation. Moreover, it may seems that there are some humans (e.g., the retarded) who are extremely limited in their inferential abilities but who are not without representational abilities (e.g., in perception.) Now, the question of which beings are capable of representation, and for that matter inference, is certainly an open issue. But the fact is that the inferential role view appears constrain the scope of representation issue without any real justification (i.e., why should it be the case that only beings capable of inference are capable of representation?) By contrast, nothing in the causal role view limits the class of beings that are capable of representation; in particular, the view does not imply that a being's internal/mental states must be linguistic in order to achieve content.
A second general drawback regarding the view that only inference-behavior constitutes content is that it is plausible to think that the meaning of certain terms is somehow critically dependent upon non-language-like mental states, viz. phenomenal or qualitative or imagistic states. For example, one standard account of the meaning of 'red' is that, roughly, for an object to be red is for it to cause normal human observers to have a red image under normal viewing conditions. Again, under anything like a strict interpretations of 'inference,' such states will not be included in the inferential role of (mental) terms,(32) thus ruling out such accounts of meaning, without any particular justification. Again, this is not a problem for the causal role view; in fact, the causal role view readily assigns content to non-linguistic representational states, e.g. images, as long as the causal roles are appropriate.
Some may find these plausible reasons for avoiding an inferential view, while others with a linguistic bias may welcome these implications. However, there is a further, and perhaps unmeetable problem for inferential role semantics--the account itself appears to be circular. To see this, consider the question of how the class of inferences is to be determined for a given speaker. At its most basic, the notion of "inference" is that of a transition from one set of statements to another, where that transition has some warrant that makes it truth preserving--i.e., the truth of the premises provide some assurance or guarantee of the conclusion's truth. It may seem, then, that the inferential role semanticist should define the meaning of a terms as the term's role in the class of statement-transition patterns that are truth-preserving according to some sort of warrant.(33) But to be truth-preserving is for the premises and conclusion to have truth conditions such that the conclusion is true if the premises are. But inferential role semantics seeks to *naturalize* semantic assignments, which obviously involve truth conditions (perhaps they *are* just truth conditions.) So this account is flatly circular, since in order to analyze truth conditions, the inferential role theorist must pre-suppose a characterization of truth conditions. In a word, the notion of inference as we have characterized it (and as it is usually understood) is *intensional,* so inferential role theory can't appeal to this notion in a naturalistic explanation of the intensional.
It is easy to overlook this fact, since it is common practice to assume that inferences can be syntactically characterized--and indeed they can--using the syntactic notion of a sentence. Moreover, in contemporary computational cognitive psychology, it is standardly assumed that the processes underlying inference-drawing can be syntactically explained in terms of symbol (e.g., sentence) transformations. But it is one thing to be able to explain or characterize the elements of a concept's extension, it is another to be able to delineate the extension itself. And it is fairly clear that the latter cannot be done syntactically. Any two sentences can be paired together, yet, presumably, it's not true that all combinations of sentences count as inferences. As for cognitive processes, note that the mere fact that one sentence is regularly produced from another will not make this transition an inference process. For instance, through random association, Skinnerian conditioning, or the like, I may be disposed to affirm "the Blues is king" whenever the color blue is mentioned, yet, surely, I am not drawing an inference. And similar considerations show that making a sentence transition process rule-governed need not make it inferential. Again, it may be a rule of my cognitive functioning that whenever blue is mentioned then I affirm "the Blues is king," but without it having, or my intending to it to have, any sort of (inference-like) *warrant.* Since it thus appears that the notion of "inference" is not legitimately available to the meaning naturalist, inferential roles cannot be the right sort of relations to identify content with in a naturalistic functional role theory of content.(34)