102 J. A. FODOR

102 J. A. FODOR
property (call it "W") that mental representations have iff they express the
property of being a witch. So, then, instead of explaining the difference
between Seymor's witch hunting and his unicorn hunting by reference to
differences between the contents of the causally implicated mental representations,
we could explain it by reference to the difference between U and W.
And, of course, if mental operations are indeed computational, there is going
to have to be some formal difference between the mental representation(s)
that mediate unicorn hunting and the one(s) that mediate witch hunting,
assuming that the difference between hunting unicorns and hunting witches
could show up (even counterfactually) in distinct behaviors. For, as we have
seen, it is the burden of the computational theory of mental operations that
only formal differences between mental representations can issue in distinct
behaviors; mental representations that differ only in their semantic properties
must ipso facto be identical in their causal roles.
There is, however, really no reason at all to suppose that there are formal
doppelgangers of each feature of the contents of mental representations that
we need to advert to in our accounts of the intentional properties of behavior.
Positing such "type-to-type" correspondences between formal and semantic
properties of mental representations involves a much stronger assumption
than that each causally efficacious difference in content must correspond
to some formal difference or other. Clearly, we have no right to build our
theories of mind upon this stronger assumption since it seems entirely possible
that formally quite different mental representations should be, as it were,
synonymous. (Dennett has emphasized, correctly in my view, that we shall
have to take this possibility especially seriously if we want ascriptions of
propositional attitudes to be comparable across individuals; a fortiori if we
want them to be comparable across species.) Moreover, one might feel, even
if there were such coextensions between features of content and features
of form, it would nevertheless be semantic facts, and not the syntactic ones,
that really account for intentionality. If, for example, there is something
about a mental representation that makes the behavior it causes unicorn
hunting rather than witch hunting, surely it is not something about the
shape of the representation; it is something that has to do with what the
representation is (or purports to be) about.
So, I think that there really are only two options given the general
framework of RTM. Either we assume that there are no explanatorily indispensible
intentional properties of behavior (specifically, that there are no
counterfactual supporting generalizations that subsume behavior in virtue
of its intentional properties) or we assume that the notion of the content
of a mental representation is ineliminable at least insofar as macrolevel
psychological theorizing is concerned. I simply cannot take the first of these
options seriously since we have—or so it seems to me—no notion of behavioral
systematicity at all except the one that makes behavior systematic under
intentional description. So I shall simply take it for granted that you cannot
save the cognitive science program by going syntactic. Either mental representations
are going to honest-to-God represent, or we are going to have to
find an alternative to RTM.