Context-dependent phrasal translation lexicons for statistical machine translation
Marine Jacinthe Carpuat, Dekai Wu · 2007
Most current statistical machine translation (SMT) systems make very little use of contextual infor-mation to select a translation candidate for a given input language phrase. However, despite evidence that rich context features are useful in stand-alone translation disambiguation tasks, recent studies re-ported that incorporating context-rich approaches from Word Sense Disambiguation (WSD) meth-ods directly into classic word-based SMT sys-tems, surprisingly, did not yield the expected im-provements in translation quality. We argue here that, instead, it is necessary to design a context-dependent lexicon that is specifically matched to a given phrase-based SMT model, rather than sim-ply incorporating an independently built and tested WSD module. In this approach, the baseline SMT phrasal lexicon, which uses translation probabili-ties that are independent of context, is augmented with a context-dependent score, defined using in-sights from standalone translation disambiguation evaluations. This approach reliably improves per-formance on both IWSLT and NIST Chinese-English test sets, producing consistent gains on all eight of the most commonly used automated evaluation metrics. We analyze the behavior of the model along a number of dimensons, includ-ing an analysis confirming that the most important context features are not available in conventional phrase-based SMT models. 1