processor needs to know that theindividual denotedby thisNP plays a certainthematicrole (say Patient)in theaction
denoted by this verb. Similarly, consider a classic example of ambiguous PP attachment such asFran painted the sign on
the porch. ThesyntacticprocessorcansaythatthefinalPP maybeattachedtoeither theNPobjector theVP. But thisis
of no use to thesemanticprocessor,whichis concerned withwhether thespeaker is specifyingthelocation of thesign
or the location of the action of painting. In both these cases, it is the linking rules, implemented by the
syntax–semantics interface processor, that are needed to make the connection.
We see from these examples that syntactic structure can play no role in semantics except by being correlated with its
semantic consequences. In other words,“sending”syntactic information to the semantic processor actually entails a
process of correlation between one level of structure and another, a non-trivial process. It is not like sending a signal
down a wire or a liquid down a pipe. It is, rather, a computation in its own right, just the sort of computation that an
interface processor performs.^114
7.5.3 The“bi-domain specicity” of interface modules
So far I have spoken of an interface module as if it simply accesses its characteristic input structure and produces its
characteristic output structure. But we can be more precise.
For a simple case, consider the auditory system. Alvin Liberman and colleagues (e.g. Liberman 1996; Liberman and
Studdert-Kennedy 1977) have argued at lengththat the auditory-phonetic interface processor is a specialized modular
device that runs in parallel with ordinary auditory processing (e.g. church bell and thunder perception), starting with
the same auditory input but analyzing it differently. In addition, two other specialized devices, voice recognition and
auditory affect (tone of voice/emotion) perception, use the auditory-signal in still different ways. Each of these two is
known to be subject to differential dissociation due to brain damage (Etcoff 1986; 1989). That is, the
IMPLICATIONS FOR PROCESSING 223
(^114) This of course applies with double force to the eye-tracking experiments discussed in section 7.2. What the human eye tells the human brain is useless to the syntactic
processor unless it can be converted into syntactic for mat, using the standard interfaceprinciples fro mvision into conceptual structure and fro mconceptual structure into
syntax.Clifton and Ferreira (1987: 290) make the tentative hypothesis that“some representational vocabulary—for instance, thematic roles [an aspect of conceptual
structure—RJ]—may evenbeshared between themodulesofthegrammaticalsystemand thegeneral-purposesystemfor representingknowledgeand beliefs.”Whatshould
be emerging from the present discussion is that it is absolutelyessentialfor modules to share information, and that they do so through interface modules.