Developing stochastic versions of ``unification-based'' and other linguistically-realistic theories of grammar is difficult because they do not naturally decompose into tree-structured dependencies. Abney's (1997) Markov Random Field approach provides a general framework for describing such non-tree-structured dependencies, but the estimator he proposes seems impractical. We describe an estimator based on pseudo-likelihood, and use it to develop a stochastic version of Lexical Functional Grammar, which we test on two small hand-parsed corpora.
Joint work with S. Geman, S. Canon, Z. Chi and S. Riezler.
For the colloquium series schedule, see the UMD Computational Linguistics Colloquium Series web page at http://umiacs.umd.edu/~resnik/cl_colloquium/. If you are interested in meeting with the speaker, please contact Mari Broman Olsen (molsen@umiacs.umd.edu) or Philip Resnik (resnik@umiacs.umd.edu).