Modularity in locating Wh-gaps

In this paper we report our finding that the parsing routine responsible for locating the deep-structure position of a wh-phrase operates independently of semantic/pragmatic knowledge, suggesting a modular organization of the human sentence processor. Specifically, wh-phrases that were semantically implausible as a direct object of the matrix verb were nonetheless reactivated in that position. Further, reactivation of the wh-phrase was observed despite the fact that the matrix verbs were three-place predicates (e.g.,remind) — a class of verbs which, some experiments suggest, allow interactive effects (e.g., Tanenhaus, Boland, Garnsey, & Carlson, 1989). We argue that this disparity in findings is due to the difference between the cross-modal lexical priming (CMLP) task used in the present study, and the tasks used by Tanenhaus et al.We gratefully acknowledge Penny Prather, Catherine Stern, and David Swinney for their technical assistance, and also Brian Hickok. This work has also benefitted from helpful discussions with Michael Tanenhaus and Stephen Crain, as well as from the comments of two anonymous reviewers.