Proposed Method

Conventional pattern-based machine translation is costly because the translation patterns are made manually. In return, the output is grammatical and tends to be a good translation. On the other hand, statistical machine translation is low cost because it uses only source and target sentence pairs that do not have to be manually related. However, statistical machine translation often outputs ungrammatical translation sentences. To overcome these problems, we focused on that the corresponding word pairs between the source language and the target language can be automatically obtained from SMT [5]. GIZA++[3] can get the source and target word pairs automatically from the source and target sentence pairs. Thus, we can make the source and target translation patterns using the automatically obtained source and target word pairs.
We implemented the program for automatically create English-Japanese statistical pattern-based machine translation. This program makes the English-Japanese translation patterns and the English-Japanese word pairs as well. We investigated the proposed English-Japanese statistical pattern-based machine translation and surveyed the standard SMT to make a comparison with the proposed method.