Marslen-Wilsonand Tyler 1987; Tanenhaus et al. 1995) has beenabout therelationbetween syntax and semantics—in
particular, whether or not syntactic parsing (inside the language module) is influenced by semantics (outside the
language module). The usual argument goes: such-and-such a property of meaning or context has a measurable
influence on syntactic parsing, therefore language perception is not informationally encapsulated, therefore modularity
is wrong. The usual reply (e.g. Forster 1979; Altmann 1987;Clifton and Ferreira 1987; Frazier 1987; Fodor et al. 1992)
goes: If you look at the phenomena more subtly and do the following experiments, youfind that up to a certain point
in time, in fact syntax isnotinfluenced by semantics, therefore modularity is right.
But notice that the argument concernsFodorianmodularity, where syntax is claimed to be insulated from semantics.
Fodorian modularity might be wrong without requiring a reversion to indiscriminately general-purpose processing. In
structure-constrained modularity, as we have seen, semantic effects on syntax are indeed possible, but they are
constrained by the need to pass through the semantics–syntax interface. In language perception, such feedback will
necessarily follow previous steps of syntactic integration, syntax–semantics interfacing,and semanticintegration. Thus
the logic of processing predicts, as observed, that there is an initial feedback-free period in syntactic parsing, but
thereafterfeedback is possible. Thatis, theobserved timing ofprocessingis a functionofwhat informationis available
when, not some extrinsic constraint.
7.5.2 Interface modules are how integrative modules talk to each other
Peopleoftensee mtothink ofa modular capacity as entirelyindependentof therest of thef- mind. This is part of what
underlies the widespread conception of a modular language capacity as an isolated “grammar box”(Chapter 4).
Domain specificity and informational encapsulation indeed seem to imply such a position. However, notice that an
entirely domain-specific and informationallyencapsulated module wouldbe functionallydisconnected from the rest of
the (f-)mind, and could therefore serve no purpose in the larger goal of helping the organism perceive and behave. So
there is a problem with such a caricature of modularity: how do informationally encapsulated modules“talk to each
other”? Structure-constrained modularity provides an answer: levels of structure communicate with each other
through interface modules.
This solution closes a crucial gap in Fodor's conceptionof modularity. Fodor proposes that theoutput of thelanguage
perception module is some language-specific“shallow”structure, perhaps a syntactic parse or a“logical form”—he is
not entirely specific. Heisspecific in denying that this output is a for mof cognitive