We report on efforts to build large-scale translation systems for eight European language pairs. We achieve most gains from the use of larger training corpora and basic modeling, but also show promising results from integrating more linguistic annotation.

We present a new translation model that include undecorated hierarchical-style phrase rules, decorated source-syntax rules, and partially decorated rules. Results show an increase in translation performance of up to 0.8 % BLEU for German–English translation when trained on the news-commentary corpus, using syntactic annotation from a source language parser. We also experimented with annotation from shallow taggers and found this increased performance by 0.5 % BLEU. 1

We present two methods that merge ideas from statistical machine translation (SMT) and translation memories (TM). We use a TM to retrieve matches for source segments, and replace the mismatched parts with instructions to an SMT system to fill in the gap. We show that for fuzzy matches of over 70%, one method outperforms both SMT and TM baselines. 1

In this paper we apply lightly-supervised training to a hierarchical phrase-based statistical machine translation system. We employ bitexts that have been built by automatically translating large amounts of monolingual data as additional parallel training corpora. We explore different ways of using this additional data to improve our system. Our results show that integrating a second translation model with only non-hierarchical phrases extracted from the automatically generated bitexts is a reasonable approach. The translation performance matches the result we achieve with a joint extraction on all training bitexts while the system is kept smaller due to a considerably lower overall number of phrases. 1

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. 1.

We present a translation model enriched with shallow syntactic and semantic information about the source language. Base-phrase labels and semantic role labels are incorporated into an hierarchical model by creating shallow semantic “trees”. Results show an increase in performance of up to 6 % in BLEU scores for English-Spanish translation over a standard phrase-based SMT baseline. 1