Translation templates were very important translation knowledge in transfer-based MT system. They could be learned automatically from bilingual corpus and were written manually by MT knowledge engineers. But there were lots of noises in template base which would decrease the translation quality. This paper presented a manually-editing tool that provided edition functions including automatic evaluation of templates, edition of templates, insertion of templates and deletion of templates. In this tool, MT knowledge engineers could check and edit translation templates according to automatic evaluation scores. This tool aimed at helping to optimize template base in MT system which would eliminate the conflicts between translation templates and the errors in translation templates. Experimental results indicated that after MT knowledge engineers edited translation templates in this tool, translation assessment score of open test corpus improved 3.85% under 3 g bleu metric.
Chun-Xiang Zhang, Xue-Yao Gao, Kun Liu, Zhi-Mao Lu, Da-Song Sun, Hong-Li Lv and Yong Liu, 2011. A Manually-editing Tool of Translation Templates Based on Automatic Evaluation. Information Technology Journal, 10: 1457-1461.