Computational Linguistics Colloquium

Friday June 29, 14:15, Seminar Room, Building 17
NOTE UNUSUAL DAY AND TIME

Stochastic ``Unification-based'' Grammars

Stochastic ``unification-based'' grammars are an attempt to develop sound statistical estimation techniques for richly structured ``unification based'' grammar formalisms such as LFG and HPSG. The introduction reviews Probabilistic Context-Free Grammars (PCFGs) and explains why the techniques that work so well for PCFGs fail in the presence of non-context-free dependencies, motivating the log linear models used here. The body of the talk concentrates on estimating these models, and explains the conditional likelihood or pseudo-likelihood used in our experiments. Depending on time and interest, the relationship of these log linear models to Optimality Theory may be described, as well as the possible application of this approach to machine translation.

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