短语统计机器翻译的句法调序模型

为了处理统计机器翻译中的长距离调序,在基于短语的统计翻译模型的基础上提出了句法调序模型.该模型按照短语切分来分割句法树结构,从而能够避免短语和句法结构的不一致性.在该模型中依据短语对齐和短语内词对齐确定句法树部分结构的调序顺序,依据各个节点上的调序概率计算子结构的调序概率,作为对数线性模型的特征函数.该模型的实验结果比经典的短语统计翻译模型的BLEU评分有明显提高.结果表明句法调序模型对于基于短语的统计机器翻译是有效的,能够较好地将句法知识和短语翻译过程结合起来.

Syntax-based reordering model for phrasal statistical machine translation

To deal with the long-distance reordering, a linguistically syntax-based reordering model was presented for phrasal statistical machine translation. In this model, the syntax structure was decomposed according to the phrase segmentation to avoid the inconsistence between phrase and syntax. The reordering sequence of the sub-structures of a parse tree was decided by the word and phrase alignments. The reordering probability of the sub-structure was calculated on the reordering probabilities of the inside nodes, which was defined as a feature function of the log-linear statistical translation model. Experimental results show that the BLEU scores of the translation results were significantly improved compared with a conventional statistical phrase-based model. Therefore, it is effective to introduce the linguistic syntax for phrase reordering. The presented reordering model is able to efficiently incorporate the syntax into the translation process of phrases.