Machine Translation

, Volume 24, Issue 2, pp 103–121

Monte Carlo techniques for phrase-based translation

Authors

    • University of Edinburgh
  • Barry Haddow
    • University of Edinburgh
  • Philipp Koehn
    • University of Edinburgh
  • Adam Lopez
    • University of Edinburgh
  • Chris Dyer
    • University of Maryland
  • Phil Blunsom
    • Oxford University Computing Laboratory
Article

DOI: 10.1007/s10590-010-9080-7

Cite this article as:
Arun, A., Haddow, B., Koehn, P. et al. Machine Translation (2010) 24: 103. doi:10.1007/s10590-010-9080-7

Abstract

Recent advances in statistical machine translation have used approximate beam search for NP-complete inference within probabilistic translation models. We present an alternative approach of sampling from the posterior distribution defined by a translation model. We define a novel Gibbs sampler for sampling translations given a source sentence and show that it effectively explores this posterior distribution. In doing so we overcome the limitations of heuristic beam search and obtain theoretically sound solutions to inference problems such as finding the maximum probability translation and minimum risk training and decoding.

Keywords

Statistical machine translationGibbs samplingMachine learningMCMC

Copyright information

© Springer Science+Business Media B.V. 2010