Computational Linguistics Colloquium

Thursday, 16 January 2014, 16:15
Conference Room, Building C7.4

Exploring the brain’s response to word information

The amount of information conveyed by each word of a sentence can be quantified by probabilistic models of the language. As is well known by now, such information values predict word–reading times: More informative words take longer to read. However, there has been only very little neurophysiological research into word–information effects. In this talk, I will present new evidence from exploratory fMRI and EEG studies intended to reveal how the brain responds to word information. Participants were presented with excerpts from novels while their brain activity was recorded, and the resulting BOLD or ERP responses were related to the words’ information values using large–scale regression analyses. Results suggest that the human sentence processing systems predicts the form of upcoming words. Moreover, a comparison between the predictive values of information estimates under different probabilistic language models provides evidence that a sentence’s sequential structure is more important than its hierarchical structure for generating expectations about upcoming words.

If you would like to meet with the speaker, please contact Vera Demberg.