We analyzed the outputs of our statistical machine translation. The results are presented next.
Single sentences provided very good results. They had few or no errors. Good examples are ``B'' and ``E'' in table 4. We think it is the effect of a large number of Japanese-English parallel sentences.
Long sentences like complex or compound sentence are a little difficult to translate. Even though, these sentence have some errors, we think they provided acceptable translation results. Good examples are ``A'' and ``C'' and ``H'' in table 4. We think this is the effect of the long phrase table.
Some words were not translated and processed as unknown words. Good example sentences were ``F'' and ``G'' in table 4. "¥³¥Ð¥ä¥·|UNK |UNK |UNK" is regarded as unknown words.
Almost of all these words are a person's name or a place-name. Thus, if we add a procedure to the English alphabet, we will obtain better good results.
Some sentences were completely wrong. We must survey why they occurred. Good example is ``J'' in table 4.