next up previous
次へ: Pattern-Based Machine Translation 上へ: Statistical Pattern-Based Machine Translation 戻る: Concept of Two-Stage Machine

Related Work

A two-stage MT system has already been proposed [5], [10], [11], [12]. L.Dugast, et al[11] and M.Simard, et al[12] applied SYSTRAN and SMT for French-English translation. Their concept was that SMT works as a post process for SYSTRAN. From the results of these study, these system are more effective than SYSTRAN or SMT. For M.Simard's research [12], the BLEU sore was 25.98 for SMT and was 28.80 for SYSTRAN + SMT in English-French translation. Also, the BLEU sore was 25.17 for SMT and was 26.79 for SYSTRAN + SMT in French-English translation. On the other hand, Ehara[10] reported on the same system for Japanese-English translation for a patent task. The BLEU score was 0.2821 for SMT and was 0.2921 for RBMT + SMT. Ehara's RBMT system was a commercial Japanese-English system. For these systems, SMT was used in the post process for RBMT, which means that SMT used as language adaptation.



Jin'ichi Murakami 平成22年12月20日