The training process is as follows.
We prepare a Chinese-English parallel corpus.
We used a Chinese-English rule-based machine translation. Thus, we obtain "ENGLISH" sentences from Chinese sentences. These "ENGLISH" sentences are pairs of English sentences.
We make an "ENGLISH"-English phrase table using training-phrase-model.perl[10].
We make an
-gram model from English sentences using SRILM
[6].
Fig. 1 shows the flow chart of the training process.