Feedback Selecting of Manually Acquired Rules Using Automatic Evaluation
Xian‐Hua Li, Yajuan Lü, Yao Meng, Qun Liu, Hao Yu · 2011
As the number of manually acquired rules in a patent translation system increases, conflicts between rules inevitably exist. Meanwhile, lacking the matched cases of manually acquired rules in real bilingual corpus, people may conceive rules which cause implausible impact on machine translation quality. In this paper, we propose a feedback selecting algorithm for manually acquired rules in patent translation using automatic evaluation, which picks out manually acquired rules that benefit machine translation quality. Experiments show that we achieve significant improvement in terms of BLEU (+5.23 points over baseline).