融入双语最大名词组块的树-串统计机器翻译模型

Tree-to-string model integrated with bilingual maximal-length noun chunk

在统计机器翻译中融入语言学知识具有重要的理论研究和应用价值.在考察了具有丰富的句法和语义信息的双语最大名词组块后,提出和实现了在树-串统计翻译模型中融入双语最大名词短语的统计机器翻译框架.通过在汉-英测试集的实验证明:相比基线模型,采用所述框架的翻译模型的BLEU值提高了1.66%,而且翻译速度也得到了提高.

It has important theoretical and application value to promote the statistical machine translation by integrating meaningful linguistic knowledge effectively.After inspected structural characteristics of maximal-length noun chunks with rich syntactic and semantic information,we proposed a statistical machine translation model which integrated with bilingual maximal-length noun chunks for improving an existing tree-to-string machine translation system.Under this scenario,we experimented on a Chinese-English corpus and achieved an improvement of 1.66 BLEU percentage point over a non-adapted state-of-the-art tree-to-string baseline system,and had a significant improvement over the baseline method on decoding speed in practice.