Computational Linguistics Colloquium

Thursday, 27 October 2011, 16:15
Conference Room, Building C7 4

Anticaption is making me look: Prediction, priming and computation in language understanding

Over the last several years, theories of human language processing have emerged that posit highly top-down architectures that actively and optimally forecast upcoming words. Simultaneously, there has been a resurgence of theories that assume modularity of initial processing and late integration of top-down information. I will describe two studies that address both trends. The first study uses eye tracking data to challenge optimality assumptions. Specifically, we find that some linguistic `anticipation` is less forward-looking that it appears in contemporary experimental paradigms. Much (though not all) anticipation may be explained by passive phenomena like priming rather than optimal forecasting, potentially reducing the computational complexity that must be attributed to human language comprehension. The other study uses event related potentials (ERPs) to re-evaluate a central finding that motivates modularity assumptions in some theories, and reveals that a component argued to reflect encapsulated syntactic processing (the ELAN) is sensitive to anticipation based on nonlinguistic expectations. These seemingly contrary results are consistent with current theories that posit dual (active and passive) processing mechanisms, as well as dynamical systems approaches such as Tabor`s self-organizing parser.

If you would like to meet with the speaker, please contact Matthew Crocker.