Data Sets

1. Training Data

   A total of 100,000 Japanese-English simple sentences were used for training data.

2. Development Data

   We used 3,000 sentences for development data for Japanese-English SMT. Of these 3,000 sentences, 375 matched the Japanese-English patterns. We therefore obtained 375 ``English'' sentences for the results. These 375 ``English'' sentences were used as development data for the ``English''-English translation.

3. Test Data

   We used 10,000 Japanese-English simple sentences as test sentences.