Cognitive Neuroscience of Language Lab
University of Colorado
Real time human language comprehension: Predictive processing in an unpredictable world
In order to comprehend human language in real time, readers and listeners make predictions about the upcoming linguistic input. Predictions enable comprehenders to prepare in advance for rapidly unfolding, pervasively ambiguous linguistic inputs. But predictions will often be wrong in a noisy, unpredictable world, and this can require language comprehenders to engage in a sometimes costly revision process. I will present findings from my research, using neurophysiological (EEG) and behavioral measures, that inform a developing theory of real-time, predictive language comprehension. One branch of this work is about establishing whether predictions occur and what predictions are: do comprehenders actually pre-activate representations before they are needed or are predictions some other sort of computation? Relatedly, what is the information content of predictions? Do we predict the physical forms of specific words, or more abstract categories like noun and verb? A second, equally important branch of the work is about uncovering the mechanisms that enable the revision of incorrect predictions, which is crucial to successful interpretation.