Gaps and garden-paths: Studies on the architecture and computational machinery of the human sentence processor.

摘要

This work presents two experiments that investigate two fundamental questions regarding the nature of the human sentence processor: (1) Is the architecture of the processing system interactive or non-interactive in nature? and (2) Are computations carried out in parallel, or in a serial fashion? Both studies looked at how syntactic ambiguity is resolved in an effort to address these issues.;In the first experiment, a cross-modal lexical priming paradigm was used to examine the nature of the architecture of the processing system (i.e., interactive vs. non-interactive). The core finding was that the syntactic processing routine responsible for locating wh-gaps is insensitive to semantic considerations. Specifically, it was found that wh-phrases are reactivated in direct object position in cases where the wh-phrase is plausible as a direct object as well as when the wh-phrase is implausible as a direct object. And this effect was observed despite the fact that the matrix verbs in the sentences were three-place predicates (e.g., remind) which, based on the results of an experiment reported by Tanenhaus et al. (1989), appear to reveal interactive effects. This finding extends the evidence in support of a non-interactive characterization of the architecture of the sentence processor.;The second experiment used an all-visual, on-line, lexical priming technique in order to consider the question of parallel versus serial processing. Syntactic attachment ambiguities were investigated. The core finding was that both of the structural possibilities associated with the ambiguous (italicized) portion of sentences such as, "They told the boy that the teacher was having trouble with to leave the room," were shown to have processing consequences simultaneously. The processing consequences of one analysis was realized as a garden-path effect (as in the above sentence), while the other analysis was evidenced by a reactivation effect. This finding supports the hypothesis that computations are performed in parallel.;As a prelude to these experimental investigations, an attempt is made to highlight some issues surrounding the nature of the so-called reactivation effect--the phenomenon that is exploited in the present experiments.