I think it will help sharpen the focus of our discussion of the "language of thought" if we separate some issues.
 
    (1) Is there internal representation?  I.e., do we have internal states that have representational content?  Behaviorists and Wittgensteinians say no.  Dennett, surprisingly, says yes.  The Churchlands say, well, sort of, in a small way ("calibrational content").  Cognitive psychologists and remaining philosophers of mind say unequivocally yes.
 
    (2) Are there internal representations whose contents characteristically involve person-level concepts?  In particular, are there representations whose contents are also those of ordinary beliefs, desires and other propositional attitudes?  Here is where Dennett and the Churchlands get off the bus.

    (3) Assuming there are representations that so correspond to folk propositional attitudes, should we identify the attitudes with those representations?  I separate this from (2) because someone might find a reason to say no here, but I don't know of anyone who has.

    (4) Assuming the attitudes are identical with such representations, are the representations themselves semantically (a) unstructured, (b) structured in a sentence-like way broadly conceived, or (c, parenthetically) structured in some non-sentence-like way?  "Sentence-like" means, (i) having some smallest meaningful parts, some of which refer to things and properties in the world, and (ii) being compositional.  I have nothing particular in mind under (c); it is only a logical possibility.

     Of those questions, I think the most controversial should be (2)--not (4) as Fodor alleges in "Why There Still...."  (No doubt he said that (4) was the real issue because he thought he'd adequately argued in his previous works for "yes" answers to (1), (2) and (3).)  Of our arguments, which support "yes" to (2)?

    Argument 1 would be weak at best in support of (2).  (Imagine a Behaviorist opponent; s/he would again respond that all a belief's semantical properties are derived or inherited from the intentional properties of behavior.)  Argument 2 would not impress the Behaviorist either, for the same reason.  Argument 3 is more impressive, since cognitive psychologists often attribute person-level representations (not just Dennettian or Churchlandian ones) to human subjects.  (And remember the general arguments against both Analytical Behaviorism and behaviorism in psychology.)  Argument 4 still works, I believe; to counter it, the Behaviorist would have to come up with a system of outer, behavioral physical states of affairs that could physically realize mental-morphemes-plus-compositional-rules, and I see no prospect of doing that.

    But to return to (4):  Once it's come down to that specific question, there are at least three arguments for "yes" to (4).  The first is my Chomskyan argument 4.  The second is the related one I came up with in class:  Our colored placards that had code meanings had those meanings by explicit convention.  There are no such conventions, indeed no conventions of any kind, that assign meanings to neural states.  Nor (as before) is it plausible to suppose that every one of our unboundedly many possible judgment contents has its own designated unstructured brain state.

    Third (I only sort of mentioned this in class):  Consider inference.  Thoughts figure in inferences, deductive and inductive and others.  Inferences are valid or correct in virtue of the forms of the thoughts occurring in them, just as in the case of sentences.  One moves from "All F are G" and "X is an F" to "X is a G," or from "P or Q" and "Not P" to "Q."  That movement is an actual psychological process.  If the representations were unstructured and had no elements in common, that process would be inexplicable.  Or at least obvious patterns would be missed:  The subject nearly always infers things whose content is of the form "X is a G" from things whose contents are of the forms "All F are G" and "X is an F."  The best explanation of that is that the forms of the contents are coded or somehow reflected in the structures of the representations themselves; hence the representations have such structures.  What other explanation might there be?