Probabilistic models of language comprehension and production

February 4, 2015
12 noon - 1:30 pm
Barnum Hall, Room 008
Speaker: Roger Levy, University of California, San Diego
Host: Departments of Computer Science, Philosophy, and Psychology


Human language use is a central problem for the advancement of machine intelligence, and poses some of the deepest scientific challenges in accounting for the capabilities of the human mind. In this talk I describe several major advances we have recently made in this domain that have led to a state-of-the-art theory of language comprehension and production as rational, cooperative, goal-driven inference and action. These advances were made possible by combining leading ideas and techniques from computer science, psychology, and linguistics to define probabilistic models over detailed linguistic representations and testing their predictions through naturalistic data and controlled experiments.

In language comprehension, I describe a detailed expectation-based theory of real-time language understanding that unifies three topics central to the field - ambiguity resolution, prediction, and syntactic complexity - and that finds broad empirical support. I also describe a "noisy-channel" theory which generalizes the expectation-based theory by removing the assumption of modularity between the processes of individual word recognition and sentence-level comprehension. This theory accounts for critical outstanding puzzles for previous approaches, and when combined with reinforcement learning yield state-of-the-art models of human eye movement control in reading. This work on comprehension sets the stage for a theory in language production in which speakers tend toward an optimal distribution of information content throughout their utterances, whose predictions we confirm in statistical analysis of conversational speech.

Finally, I describe research in computational and experimental pragmatics, built on a highly general Bayesian framework of interleaved semantic composition and pragmatic inference where the cooperative goal is bringing the beliefs of the listener into as close an alignment as possible with those of the speaker while maintaining brevity. More generally, this program of research exemplifies how interdisciplinary work bridging multiple fields can lead to fundamental advances in cognitive science.

*Note: Different day/time/room than usual for this talk*