back to

Seminar: 2/17 - Srini Narayanan

11:00 to 12:30 PM      at:  

LOCATION CHANGE: Tolman 3105
International Computer Science Institute
"A Bayesian Model of Human Sentence Processing."

Language comprehension is a classic problem of reasoning under uncertainty. Language comes to us as a noisy, unsegmented, ambiguous mass of auditory waveforms or visual stimuli. We have to somehow combine this input with other knowledge to come up with appropriate interpretations and actions. How might humans address this problem of finding the best fitting interpretation under uncertainty? Evidence from behavioral and from imaging experiments suggests that language analysis involves bringing to bear syntactic, semantic, and pragmatic knowledge to select the best interpretation of the input. In previous work, we have proposed that that humans solve this problem of linguistic decision-making by acting as Bayesian probabilistic reasoners. I will describe our basic processing model, behavioral predictions about reading time, and recent results of testing these predictions on experimental data. Finally, I will situate the work in the context of a larger effort to build neurally plausible models of language acquisition and use, the NTL (http://www.icsi.berkeley.edu/NTL) project, and touch on an ambitious new effort within the NTL group to build a scalable construction grammar parser based on these principles.