Inducing Synchronous Grammars with Slice Sampling
Phil Blunsom and Trevor Cohn
In: NAACL-HLT 2010, 1-6 Jun 2010, Los Angeles, USA.

Abstract

This paper describes an efficient sampler for synchronous grammar induction under a non-parametric Bayesian prior. Inspired by ideas from slice sampling, the proposed sampler is able to draw samples from the posterior distributions of models for which the standard dynamic programing based sampler proves intractable for non-trivial corpora. We compare our sampler to a previously proposed Gibbs sampler and demonstrate strong improvements in terms of both training log-likelihood and performance on an end-to-end translation evaluation.

[Edit]