Adaptation of Reordering Models for Statistical Machine Translation

Boxing Chen, George Foster and Roland Kuhn

Previous research on domain adaptation (DA) for statistical machine translation (SMT) has mainly focused on the translation model (TM) and the language model (LM). To the best of our knowledge, there is no previous work on reordering model (RM) adaptation for phrase-based SMT. In this paper, we demonstrate that mixture model adaptation of a lexicalized RM can significantly improve SMT performance, even when the system already contains a domain-adapted TM and LM. We find that, surprisingly, different training corpora can vary widely in their reordering characteristics for particular phrase pairs. Furthermore, particular training corpora may be highly suitable for training the TM or the LM, but unsuitable for training the RM, or vice versa, so mixture weights for these models should be estimated separately. An additional contribution of the paper is to propose two improvements to mixture model adaptation: smoothing the in-domain sample, and weighting instances by document frequency. Applied to mixture RMs in our experiments, these techniques (especially smoothing) yield significant performance improvements.