针对词一级的混淆网络(CN)构建的对齐方法和骨架翻译选择的问题进行了研究。将增量的对齐方法引入到原有的翻译错误率(TER)对齐框架中,并对加入候选翻译的顺序进行了不同的尝试;同时选用两种最小贝叶斯风险(MBR)策略来选择骨架翻译;为了改进TER中同义词的对齐,进行了同义词匹配。在实验中,把候选翻译的顺序、选择骨架翻译方法、是否引入同义词这三种策略进行了组合。匹配数据显示,可以确定一种组合策略来最大地提高系统融合性能。
The key problems of alignment and selection of skeleton translauon m ueslgn o~ a w, (CN) were investigated. An incremental alignment method was introduced into the translation error rate (TER) alignment framework, and two minimum Bayes risk (MBR) strategies were applied to skeleton selection. In addition, the synonym matching method was used to improve the alignment of synonym in the TER-based alignment process. The experimental results show that the performance of system combination can be improved using a best combination strategy which combines candidate order, skeleton selection and synonym matching.