Christopher Allison

• While there is now ample evidence of prediction in language processing, the overwhelming majority of this evidence comes from “prediction encouraging set-ups”. Thus, while many theories of language processing posit an inherent role of prediction in language processing, the reliance on prediction encouraging set-ups and the findings that some manipulations drastically reduce prediction (e.g. concurrent phonological demand) or that some groups (e.g., non-native speakers, illiterate adults, children, older adults) show reduced prediction have led some to question the ubiquity of predictive language processing. Two standard parts of the “prediction encouraging set-up” are slow speech rates and a minimally demanding processing context/environment. The three studies included in this dissertation manipulate both of those, specifically by using faster speech rates and a concurrent nonlinguistic working memory demand, during the also more demanding task of non-native predictive language processing. I report evidence that simple semantic predictions in non-native English speakers are generally robust to both increases in speech rate and concurrent working memory demands. However, either gaze behavior or predictions that require language/vision interactions as well as predictions that require combining multiple representations may be reduced as demand increases.