Human Evaluation

We survayed all sentences for A-rank, B-rank, and C-rank. Also we selected 20 sentences at random for D-rank. The results of the ABX evaluation are listed in Table 11.

Table 11: Results of Human Evaluation

Rank	Proposed $\bigcirc$	RBMT $\bigcirc$	No difference	Same
A rank	8	5	10	0
B rank	3	2	8	0
C rank	2	4	2	0
D rank	1	10	9	0

Table 11 indicates that the proposed method was superior to the rule-based translation for the A and B ranks. In comparison, its results were split for the C rank, and it was inferior to the rule-based translation for the D rank. This shows the effectiveness of the proposed method for the A and B ranks.