The role of working memory in comprehension of doubly embedded relative clauses: a self-paced reading and eye tracking study

Language processing has been a focus of working memory research since Baddeley introduced his Model of Working Memory in the 1970’s. There has been continued discussion over whether the same working memory (WM) system that underlies verbally-mediated tasks relying on conscious, controlled, processing also provides the resources used in language processing. Recently, Caplan, DeDe, Waters, & Michaud (2011) found that increased reading times at only the most difficult point of the most difficult sentences presented in their study (sentences with doubly embedded relative clauses) correlated with improved comprehension. They hypothesized that this correlation occurs because at these points where normal parsing fails, individuals with high working memory capacities use ancillary comprehension mechanisms that rely on verbal working memory. Caplan and Waters (2013) proposed that use of verbal working memory for ancillary comprehension in sentence processing may appear behaviorally as improved comprehension with longer reading times in self-paced reading tasks and as regressive eye movements out of these points where parsing is thought to fail. This thesis attempted to replicate the above mentioned finding of Caplan et al. (2011). This study also added an eyetracking task to enable measurement of regressive eye movements and a measure of working memory to permit analysis of individual differences. Forty-eight healthy adults completed a working memory battery (alphabet span, subtract two span, and sentence span), a self-paced reading task, and an eye-tracking task. For the self-paced reading and eye tracking components, participants read sentences with doubly embedded relative clauses and parallel sentences with sentential complements. Linear mixed effects models found increased self-paced reading times and go-past times at the hardest point in the harder sentences (those with doubly embedded relative clauses) as working memory increased. These results support the hypothesis that ancillary comprehension mechanisms are used in sentence processing at points where comprehension is extremely difficult. In the attempted replication of the findings of Caplan et al. (2011), logistic mixed effects models showed increased accuracy as reading times increased at the hardest point in the harder sentences, and also as reading times increased at five of the other seven segments. Logistic mixed effects models showed no significant increase in regressions out of the hardest point in harder than in the easier sentences as working memory increased. These results can be taken as further evidence, using eye tracking methods combined with self-paced reading and measurement of working memory, that ancillary comprehension mechanisms may be used in sentence processing when the limits of the normal parser are exceeded