译文质量的自动评价对机器翻译研究具有十分重要的意义。但现有方法主要是针对书面语翻译,没有考虑到口语翻译的特征。因此,本文提出了一种面向口语的新型的自动评价方法,通过定义信息段、标注权重和设计多种匹配策略等方法,使自动评价结果与人工打分更为接近,同时也提高了评价过程对不同输出译文的适应能力。各项实验表明,该算法对译文质量变化具有较高的敏感度,而且可以对输出译文质量作出与手工评判较为接近的评价结果。

The automatic evaluation of output quality for machine translation systems is a difficult problem. Most research in this section is for writing language ,which is quit different from oral language. Therefore ,this paper provides a new automatic evaluation algorithm for speech translation system. It defines the block according to the word position ,tags the information weight and designs several matching methods. All these make the automatic evaluation results more closely to the score of human. And the automatic evaluating can also process more different sentences. Several examinations are given at the end of the paper. They show the algorithm has a good sensitivity for different output quality and its evaluation results is as same as the human’s feeling.