Neural evidence for representationally specific prediction in language processing

Abstract

Previous studies suggest that people generate predictions during language comprehension at multiple linguistic levels. It has been hypothesized that, under some circumstances, this can result in the pre-activation of specific lexico-semantic representations. We asked whether such representationally specific semantic pre-activation can be detected in the brain ahead of encountering bottom-up input. We measured MEG activity as participants read highly constraining sentences in which the final word could be predicted. We found that both spatial and temporal patterns of the brain activity prior to the onset of this word were more similar when the same words were predicted than when different words were predicted. This pre-activation was transient and engaged a left inferior and medial temporal region. These results suggest that unique spatial patterns of neural activity associated with the pre-activation of distributed semantic representations can be detected prior to the appearance of new sensory input, and that the left inferior and medial temporal regions may play a role in temporally binding such representations, giving rise to specific lexico-semantic predictions.