syntactic analysis. The only means for feedback of this sort is through the routes provided by the interfaces. More
generally, there is no general-purpose“contextual”mechanism that permits context to affect syntax directly.
Thehypothesisthat therules ofgrammar establishthepossibleroutesof“informationflow”inprocessingcanonlygo
so far, of course. It can make predictions aboutrelativetiming: what structures must be in place before others can be
established. For instance, it predicts that semantics cannotconstrain phonology until after lexical look-up. For another
typical case, in thesentenceSandy was kissed by Chris, itis impossible for the perceiverto determine thatSandyis Patient
of the event (its semantic role) until it is determined thatSandyis subject of a passive (its syntactic role). This predicts
that there is a necessary time lag between syntactic and conceptual integration, a time lag observed in the numerous
experiments designed to test the separationof syntactic from semanticprocessing. However, the logic of the grammar
makes no predictions aboutabsolutetiming. That is a matter for experimental work to decide.
The competence model may also make predictions about what phenomena go together in what component of the
grammar. To the extent that the phenomena of some particular component (say syntactic integration) fall together
experimentally in terms of timing, processing load, speech errors, brain wave (ERP) signature, and/or brain
localization, and they contrast in these respects with phenomena of other components, we can align the competence
model with processing evidence (see Hagoort, Brown, and Osterhout 1999) and also the discussion of aspectual
coercion in section 12.2). Ideally, we may even be able to use evidence from processing to help decide in what
componentofthegrammar a particular phenomenonbelongs—which,as seeninprevious chapters, isoftena pressing
issue for the competence theory.
Further constraints on the interaction of components may arise from the way the processor is implemented in the
brain. Forster (1979) and Frazier (1987), for example, argue that in language perception there is no semantic feedback
at all before syntactic processing is complete—a constraint not predicted by the competence theory. Similarly, Levelt
(1989) argues that in production, the semantics of lexical items are selected completely before phonological material
becomes accessible—again a constraintbeyondthat predictedby the competence theory. On theother hand, Marslen-
Wilson and Tyler (1987), Tanenhaus et al. (1995), and many others argue thereissemantic feedback on syntax in the
course of perception, and Dell (1986; Dell et al. 1997) argues that phonological effects may actually be implicated in
semantic selection.
The competence model cannot settle these disputes. The freely incremental possibility looks simpler, but the model is
consistent with both possibilities. On the other hand, as seen above, the competence model does tell us that even free