Human Evaluation Results

We carried out the ABX test [13] on the proposed method and the baseline(Moses) for each of the Japanese translation sentences. The ABX test is a human evaluation. And this evaluation was carried out only one person. It involves a count of the sentences using the following criteria.

• Proposed ○: The proposed method's translation was better than the baseline (Moses).
• Proposed ×: The proposed method's translation was worse than the baseline (Moses).
• No difference: There was no difference in translation quality between the proposed method and the baseline(Moses).
• Same: Both outputs were completely the same.