ELSC Seminar Series

Predictive processing is thought to play a central role in many cognitive domains, including language. In sentence processing, it was shown that upcoming words are probabilistically pre-activated based on the context. In this talk, I will present a series of behavioral and event-related potentials experiments exploring the finer-grained mechanisms of lexical prediction, beyond activations.

Our findings suggest a strong form of prediction – “pre-updating” – which occurs in highly constraining contexts and entails actual integration of the predicted material into the sentence representation, before bottom-up evidence for its occurrence. We also found that the tendency to pre-update is variable between participants, and can be modulated throughout the experiment. Further, we show that when a strongly predicted word is not encountered in the input, it undergoes inhibition in order to allow for integration of the actual input. Finally, we found that simultaneously activated predictions interact, with their relative activations depending on their relatedness to one another. Based on our results, we propose a general model of lexical prediction in sentence processing.