According to Wallis, philosophers of mind agree that a successful theory of representation must "describe conditions for representation in nonintentional and nonsemantic terms." Once beliefs are included as representations, there is no such agreement. Fodor's position is tendentious -- and therefore interesting -- precisely because he argues (a) that beliefs are representations (in a broad sense of the term) and (b) that the relation beliefs bear to their truth conditions can be characterized in nonsemantic terms. Fodor's theory may be flawed, but not for the reasons outline in Wallis' paper.
2. Fodor is not committed to applying his account of truth conditions for beliefs to all representations, however. And it would be implausible for him to do so. For Fodor is attempting to provide a "naturalistic" account of what Grice (1957) called "non-natural" meaning. If all representations satisfied Fodor's proposed sufficient condition, Fodor would be committed to the view that no representations have only what Grice called "natural" meaning. But at least many philosophers-- Fodor among them-- have held that there is an important difference between the internal states of frogs (etc.) and "fully" intentional entities like beliefs, desires, etc. Fodor's asymmetric dependence theory is supposed to capture this difference. But his theory is fully compatible with some other account (e.g., that of Dretske (1981)) of Grice's "natural meaning" as it applies to entities (other than beliefs) that cognitive scientists often characterize as representational. Thus, much of Wallis' criticism strikes me as misdirected.
3. More interesting is Wallis's barn example (5.5-5.8). Fodor's theory is supposed to apply, for example, to the belief I might express by saying, "Yonder lies a barn." Let us call tokens of this belief type "B-tokens." Details aside, Fodor's idea is that the false depends on the true, but not conversely. And Fodor is indeed committed to saying that breaking the barn/B-token connection will break the facsimile/B-token connection but breaking the facsimile/B-token connection need not break the barn/B-token connection. Wallis's example, however, at least as stated, presents no difficulty for Fodor.
4. First, one cannot KNOW (pace Wallis) that (all) large red buildings having extended rectangular bodies are barns, and I doubt that anyone even BELIEVES this, although it may be that agents are disposed to produce B-tokens in the presence of large red buildings having an extended rectangular body. (What agents BELIEVE, I take it, depends not merely on how they are disposed to react to stimuli, but also on whether they are disposed to "change their mind" in the light of new stimuli.) Wallis claims that breaking the facsimile/B-token connection by breaking the large-red-(etc.)/B-token connection will also break the barn/B-token connection. But this will be so only if the agent's sole route from barns to B-tokens is via the agent's large-red-(etc.)/B-token connection. Fodor's intuition (and mine), however, is that such an impoverished conception of a barn would not yield beliefs that "non-naturally" mean that yonder lies a barn. And given an agent with genuine barn-beliefs, there will be many ways to break the facsimile/B-token connection.
5. Suppose agents come to believe (because they look closely) that the building in question is only an inch deep. This will break the large-red-(etc.)/B-token connection, at least for the object at hand, without breaking the barn/B-token connection. But agents with genuine barn-beliefs can still come to form a (true) B-token because they are presented with a large-red-(etc.) object. Breaking such agents' barn/B-token connection would break their facsimile/B-token connection, but not conversely. For breaking their barn/B-token connection requires breaking their large-red-(etc.)/B-token connection, but not conversely.
6. Fodor claims that if instantiations of a property Q can cause tokens of the representational type B, where B-tokens have Gricean "non-natural" meaning, then either (i) the instantiation of Q is a truth condition for B-tokens or (ii) instantiations of Q can cause B-tokens only because instantiations of P can cause B-tokens, but not conversely, where the instantiation of P is a truth condition for B-tokens. Wallis's cases fall into two classes: those involving representational states that do not have non-natural meaning, and those not in conflict with (ii) above. Fodor's theory may be flawed -- see, e.g., the articles in Loewer and Rey (1991) -- but not, I think, for the reasons Wallis suggests.
Dretske, F. (1981) Knowledge and the Flow of Information. Cambridge MA: MIT Press.
Fodor, J. (1990) A Theory of Content and Other Essays. Cambridge MA: MIT Press.
Grice, P. (1957) "Meaning," Philosophical Review 66: 377-88.
Loewer, B. and Rey, G. (eds.) (1991) Meaning in Mind: Fodor and His Critics. Cambridge: Blackwell.
Wallis, C. (1992) Asymmetric Dependence and Mental Representation. PSYCOLOQUY 3(70) fodor-representation.1