We calculated the 4-gram model using ngram-count in the Stanford Research Institute Language Model (SRILM) toolkit[6], and used the smoothing parameter as " -ukndiscount -interpolate". With the 19972 parallel sentences, we obtained the followings. For 5#5-gram, we had 8346 lines. For 6#6-gram, we had 49685 lines. For 7#7-gram, we had 17241 lines. For 8#8-gram, we had 14651 lines.