structure in terms of which rules of inference and/or meaning postulates are stated, and in terms of which judgments
of truth are determined—what I callhere“conceptual structure”and what Fodor wouldcall“narrow content”or“the
syntax of the Language of Thought.”
Notice, however, that the process of language perception is supposed to lead to belieffixation, i.e. determining the
truth value of the heard utterance. The difficulty is thatin order to befixed, a belief has to be formulated in terms of conceptual
structure.Alinguisticstructure is simply the wrong vehicle for belieffixation. Fodor's faculty-sized module is therefore
stranded without access to the cognitivestructures it needs to attain.^113 Structure-constrained modularity does not face
this problem: the transition from linguistic structure to conceptual structure is accomplished by an interface module
that is altogether parallel in character to the one that mediates between phonology and syntax. (The latter is not a
module for Fodor either: it is just some unexamined subcomponent within the larger Fodorian language perception
module.)
This gap in Fodor's account of the modularity of language is an instance of a more pervasive problem in theorizing
about cognitive processes, which arises fro mfailing to observe the distinction between integrative and interface
processors. The proble mis revealedin the co m mon practiceof drawing diagra ms of processing along the lines of (5),
(5) syntactic processor→semantic processorand saying things like“the output of the syntactic processor (or module) is sent to the semantics.”
Although this makes sense intuitively, it neglects an important step. A syntactic processor is concerned strictly with
elementsofsyntactictrees likeNs, VPs, complementizers, and casemarking. Its parse ofa sentence is stillmade out of
syntactic units. A semantic processor knows nothing about NPs and case marking; rather it knows about things like
conceptualized objects and events, and about claims to truth and falsity. Hence the syntactic informationthat a certain
NP is the object of a verb is useless to the semantic processor. The semantic
222 ARCHITECTURAL FOUNDATIONS
(^113) This point is made in different ways by Marslen-Wilson and Tyler (1987) and Jackendoff (1987). See Jackendoff (2000) for further discussion and examples.Many people
with who mI have discussed this point deny that Fodor really intends the language module to stop at syntax. However, Fodor's text is quite explicit. Forinstance, he has a
long excursus (1983: 79-82) on the lexical priming experiments we discussed in section 7.4.2. He goes to great lengths to deny thatbugprimesinsectby virtue of its
meaning—which for hi mis inaccessible to the language module. He appeals instead to“stupid”“interlexical associations”that are supposed to be somethingotherthan
meaning-based. Whilethis mightbe possiblefor pairs likesaltandpepperthat are oftenheard in juxtaposition, it is less likelyforbugandinsect. As far as I know, theliterature
has rightfully neglected this aspect of Fodor's proposal.