The training model of our proposed machine translation system has three parts. The first process involves constructing an ``English''-English phrase table, the second process involves constructing a Japanese-English phrase table, and the third part involves constructing a language model ( -gram).