We propose a pre-reordering approach for Japanese-to-Chinese statistical machine translation (SMT). The approach uses dependency structure and manually designed reordering rules to arrange morphemes of Japanese sentences into Chinese-like word order, before a baseline phrase-based (PB) SMT system applied. Experimental results on the ASPEC-JC data show that the improvement of the proposed pre-reordering approach is slight on BLEU and mediocre on RIBES, compared with the organizer's baseline PB SMT system. The approach also shows improvement in human evaluation. We observe the word order does not differ much in the two languages, though Japanese is a subject-object-verb (SOV) language and Chinese is an SVO language.
展开▼