Results of Human Evaluation

We randomly selected 100 sentences from the 1,143 output sentences that were matched with the Japanese-English patterns. Then we evaluated these 100 sentences. The results are listed in Table 17.

Table: Results of Human Evaluation

Proposed Moses	30 / 100
Proposed $<$ Moses	9 / 100
Proposed $\approx$ Moses	50 / 100
Proposed $=$ Moses	11 / 100

As the table indicates, the proposed method achieved better evaluation than Moses. The $p$ -value was exceeded for 0.95. This means that the proposed method is effective for human evaluation.