Representational pattern similarity of electrical brain activity reveals rapid and specific prediction during language comprehension

Abstract

Predicting upcoming stimuli and events is a critical function of the brain, and understanding the mechanisms of prediction has thus become a central topic in neuroscientific research. Language provides a fertile testing ground for examining predictive mechanisms, as comprehenders use context to predict different features of upcoming words. Although there is a substantive body of research on prediction in language, many aspects of the mechanisms of prediction remain elusive, in part due to a lack of methodological tools to probe prediction formation in the moment. To elucidate what features are neurally pre-activated and when, we used representational similarity analysis (RSA) on data from a sentence reading task (Federmeier et al., 2007). We compared EEG activity patterns elicited by expected and unexpected sentence final words to patterns from the preceding words of the sentence, in both strongly and weakly constraining sentences. Pattern similarity with the final word was increased in an early time window (suggestive of visual feature activation) following the presentation of the pre-final word, and this increase was modulated by both expectancy and constraint (greatest for strongly constrained expected words). This was not seen at earlier words, suggesting that predictions are precisely timed. Additionally, pre-final word activity - the predicted representation - had negative similarity with later final word activity, but only for strongly expected words. Together, these findings shed light on the mechanisms of prediction in the brain: features of upcoming stimuli are rapidly pre-activated following related cues, but the predicted information may receive reduced subsequent processing upon confirmation.