| This dissertation discusses the utilization of working memory during multiple sentence processing. In the first three chapters, we review working memory research and its relation to sentence processing. In later chapters, we present evidence that supports modifications to the model of working memory developed by Baddeley and Hitch (1974) and updated by Baddeley (2003). We developed a novel sentence comprehension task to measure accuracy in comprehending groups of three sentences. The task is to indicate if the third sentence has the same meaning as the first two sentences combined. In experiment 1, the sentence congruency task was compared with a semantic acceptability task using a self-paced moving window paradigm. In experiments 2--8, we used a rapid serial visual presentation paradigm to present sentences. The results demonstrate that subjects expect that sentences will be congruent, since they were more accurate at judging congruent than incongruent sentence trials. The results also demonstrate that subjects use at least two separate, but parallel pathways during sentence processing. To study the syntactic pathway, we manipulated the syntactic form of the sentences. Accuracy improved and reaction time decreased when the syntactic form of the first two sentences matched the syntactic form of the third sentence. Few consistent differences were found on performance between right-branching and center-embedded third sentences, and the interaction of syntactic form, animacy, and relative clause branching point are discussed as possible explanations for these results. A second pathway used during sentence processing was examined by manipulating the semantic reversibility of the sentences. An interaction was found between syntactic form and semantic reversibility, with subjects demonstrating lower accuracy detecting incongruities in semantically irreversible compared to semantically reversible right-branching or center embedded sentences, but lower accuracy on semantically reversible compared to semantically irreversible passive relative clause sentences. We propose that competition between processing in both pathways for overlapping working memory resources can explain these results, and discuss limitations in working memory previously unspecified in working memory literature. Finally, we report the results of an FMRI experiment using the sentence congruency task. |
