Reordering constraints for phrase-based statistical machine translation

Abstract

In statistical machine translation, the generation of a translation hypothesis is computationally expensive. If arbitrary reorderings are permitted, the search problem is NP-hard. On the other hand, if we restrict the possible reorderings in an appropriate way, we obtain a polynomial-time search algorithm. We investigate different reordering constraints for phrase-based statistical machine translation, namely the IBM constraints and the ITG constraints. We present efficient dynamic programming algorithms for both constraints. We evaluate the constraints with respect to translation quality on two Japanese-English tasks. We show that the reordering constraints improve translation quality compared to an unconstrained search that permits arbitrary phrase reorderings. The ITG constraints preform best on both tasks and yield statistically significant improvements compared to the unconstrained search.

Inversion Transduction Grammar Constraints for Mining Parallel Sentences from Quasi-Comparable Corpora

“…Zhang and Gildea (2005) [6] show that lexicalized ITGs can further improve alignment accuracy. With regard to translation rather than alignment accuracy, Zens et al (2004) [7] show that decoding under ITG constraints yields significantly lower word error rates and BLEU scores than the IBM constraints. Chiang (2005) [8] obtains significant BLEU score improvements via unsupervised induction of hierarchical phrasal bracketing ITGs.…”

Word sense disambiguation vs. statistical machine translation

“…For example, the WSD model's predictions could be employed in a Bracketing ITG translation model such as Wu (1996) or Zens et al (2004), or alternatively they could be incorporated as features for reranking in a maximum-entropy SMT model (Och and Ney, 2002), instead of using them to constrain the sentence translation hypotheses as done here. However, the preceding discussion argues that it is doubtful that this would produce significantly different results, since the inherent problem from the "language model effect" would largely remain, causing sentence translations that include the WSD's preferred lexical choices to be discounted.…”

Local phrase reordering models for statistical machine translation

“…Other researchers (Vogel, 2003;Zens and Ney, 2003;Zens et al, 2004) have reported performance gains in translation by allowing deviations from monotone word and phrase order. In these cases, reordering is not governed by an explicit probabilistic model over reordered phrases; a language model is employed to select the translation hypothesis.…”