Given that Fodor accepts the relational nature of intentional attitudes, his first step is to try to clear the field of those who deny such a nature. Some authors, in fact, sustain the position (which Fodor calls the hypothesis of fusion) for which intentional attitudes are really "monadic" mental states. According to this view, "Simon believes that the Morning Star is Venus" expresses a mental state characterized by an individual "Simon" and a unary predicate "believes-that-the-Morning-Star-is-Venus" in which the verb of believing is semantically fused with the object of belief. Fodor presents several arguments, of which I will cite just two, against this thesis:
1) There are a very large number of sentences of the form a believes complement. If all these sentences are atomic, how could human beings possibly learn English? (Fodor attributes this to Davidson, 1965).
2) Different propositional attitudes frequently converge on the same content: e.g., one can both fear and believe that it will rain on Tuesday. But, according to the fusion thesis, except for the reference to John, "John fears that it will rain on Tuesday" has nothing in common with "John believes that it will rain on Tuesday." In particular, the fact that the form of the words "it will rain on Tuesday" occurs in both sentences is just an accident.
The position expressed by Carnap in Meaning and Necessity, taken up by Hartry Field in 1978 and Gilbert Harman in 1982, is interesting because it seeks to maintain the intuition that mental states are relational, without the necessity of any ontological commitment with respect to such conceptually troublesome entities as mental representations. Fodor presents several arguments which he believes justifies its rejection. First, he suggests that Carnap’s theory is fundamentally behavioristic. Carnap, that is, has one theory regarding the objects of intentional attitudes and a second theory regarding the character of the relation of individuals with such objects. The second theory is behavioristic because, for Carnap, "to believe such and such is to be disposed ...in specifiable conditions... to proffer occurrences of the correspondent of the sentences which attribute beliefs. But, patently, beliefs are not behavioural dispositions and, a fortiori, are not dispositions to proffer anything." In simpler terms, since behaviourism is presumably false, then there must be something wrong in some part of Carnap’s explanation of propositional attitudes.
But Fodor also has some more interesting and persuasive criticisms to level at the Carnapian view. His second argument is to the effect that Carnap’s theory can most naturally be read as a theory which considers the type-identity of the correspondents of sentences which attribute beliefs as necessary and sufficient conditions for the type-identity of the beliefs actually attributed since Carnap was primarily concerned with the opacity of beliefs. His approach, therefore, was based on a strategy of inheriting the opacity of beliefs from the opacity of citations. The problem is that this strategy fails in every case in which the identity conditions of beliefs are different from the identity conditions for sentences. For example, it is likely that the sentences "Phil believes that Alice shot Fred" and "Phil believes that Fred was shot by Alice" attribute type-identical beliefs. But they are clearly not type-identical sentences. Fodor points out that a way around this problem would be to admit that objects of belief are, essentially, systems of translations of sentences. But this results in circularity. One way of characterizing the relation of translation between sentences is by reference to the communicative intentions of speakers-hearers. But we cannot then "identify the translations in reference to the intentions while, simultaneously, individuating the propositional attitudes (including intentions) in reference to the translations. Along the same lines, one can believe that it is raining without being able to speak English. This implies, once again, that objects of belief must be systems of translation and we end up in something like the circle described above.
Fodor’s solution is to abandon the idea that propositional attitudes are relations between individuals and sentences of natural languages while not abandoning the idea that they are relations between individuals and sentences tout court. He proposes that the relation is between individuals and internal sentences, or systems of mental representations.
The fact that thoughts are not representations, nevertheless, does not mean that they have a nature similar to objects in the external world. In order to account for the peculiarity of their nature, it is necessary to appeal to a "third realm" outside of both the mental and the physical. But this leaves open the epistemological problem of how it is possible for an individual human mind to gain access to such abstract objects of the third realm. As Fodor puts it:
The difficulty illustrated by Fodor is of both an empirical and a conceptual nature. What is at issue is the question of the psychological plausibility of theories: the point, in other words, is that it doesn’t seem possible to determine the nature of mental content only in the abstract, without taking into consideration the conditions which render that content empirically plausible (specifying, for example, how a physical system must be constituted in order for it to be able to instantiate them). Given this empirical limit to conceptual speculation, that which is necessary is some mechanism capable of elaborating mental content. The contact with hypothetical entities of the third realm by way of an unspecified technique of grasping seems to exclude this possibility for Fodor.