Coreference processing and levels of analysis in object-relative constructions; Demonstration of antecedent reactivation with the cross-modal priming paradigm

This paper is concerned with two related issues in sentence processing-one methodological and one theoretical. Methodologically, it provides an unconfounded test of the ability of the cross-modal lexical priming task, when used appropriately, to provide detailed evidence about the time-course of antecedent reactivation during sentence processing. Theoretically, it provides a study of the nature of the representation that is examined when a reference-seeking element is linked to its antecedent during the processing of object-relative clause constructions. In these studies, subjects heard sentences which contained a lexical ambiguity placed in a strong biasing context. In one study this ambiguous word was the moved or fronted object of the verb in an object-relative construction. A cross-modal lexical priming (CMLP) naming task was used to determine whether one or more of the meanings of the ambiguity are activated at three temporally distinct points during the sentence: (1) immediately after the lexical ambiguity (Study 1); (2) a later point that was 700 milliseconds before the offset of the main verb (Study 2); (3) immediately after this main verb (at the gap in this filler-gap construction) (Study 2). The probes in the CMLP task were controlled for potential confounds. The results demonstrate the following: At Test Point 1, all meanings of the ambiguity were activated; at Test Point 2, neither meaning of the ambiguity was (still) activated; at Test Point 3, only a single (context-relevant) meaning of the ambiguity was reactivated. It is concluded that an underlying (deep; non-surface-level) memorial representation of the sentence is examined during the process of linking an antecedent to a structural position requiring a referent, and that the CMLP task provides an unbiased measure of this reactivation. Further, it is concluded that this effect cannot be accounted for under a compound cue (Ratcliff & McKoon, 1994) explanation.The authors gratefully acknowledge the contributions of an anonymous reviewer for helpful suggestions and changes in this paper, and the support of a fellowship to the first author from grant T32 DC 00041 and NIH grant ROI DC00494 (to the second author) for this work.