This paper describes the NAIST statistical machine translation system for the IWSLT2012 Evaluation Campaign. We participated in all TED Talk tasks, for a total of 11 language-pairs. For all tasks, we use the Moses phrase-based decoder and its experiment management system as a common base for building translation systems. The focus of our work is on performing a comprehensive comparison of a multitude of existing techniques for the TED task, exploring issues such as out-of-domain data filtering, minimum Bayes risk decoding, MERT vs. PRO tuning, word alignment combination, and morphology.

Many syntactic machine translation decoders, including Moses, cdec, and Joshua, implement bottom-up dynamic programming to integrate N-gram language model probabilities into hypothesis scoring. These decoders concatenate hypotheses according to grammar rules, yielding larger hypotheses and eventually complete translations. When hypotheses are concatenated, the language model score is adjusted to account for boundary-crossing n-grams. Words on the boundary of each hypothesis are encoded in state, consisting of left state (the first few words) and right state (the last few words). We speed concatenation by encoding left state using data structure pointers in lieu of vocabulary indices and by avoiding unnecessary queries. To increase the decoder’s opportunities to recombine hypothesis, we minimize the number of words encoded by left state. This has the effect of reducing search errors made by the decoder. The resulting gain in model score is smaller than for right state minimization, which we explain by observing a relationship between state minimization and language model probability. With a fixed cube pruning pop limit, we show a 3-6% reduction in CPU time and improved model scores. Reducing the pop limit to the point where model scores tie the baseline yields a net 11% reduction in CPU time.