大规模无参考译文质量自动评测技术的研究

Research on automatic translation quality evaluation technology without translation references for large-scale translations

传统的译文质量自动评测方法主要针对机器自动翻译结果并通过与参考译文的对比实现译文质量的评估。提出一种大规模无参考译文质量自动评测方法,该方法可以在无参考译文的情况下,从正确性、流畅性和忠实度等方面对人工译文进行质量检查并标注可能的错误点,即利用语言规则实现正确性的判断,利用语言模型实现流畅性的判断,利用词对齐结果和词典信息实现忠实度的判断。基于该方法实现的译文质量辅助检查系统在国家知识产权局百万专利翻译项目中的应用效果表明,使用该系统的测试组较未使用该系统的测试组的平均译文质量和终稿完成效率均有明显提升。

The traditional automatic translation quality evaluation method mainly focuses on the automatic translation results of machine,and compares with the translation references to evaluate the translation quality.This paper puts forward a multi-strategy based automatic translation quality evaluation method without translation references for large-scale translations.The method can check and identify errors in artificial translations for accuracy,fluency,adequacy,etc.without translation references.i.e.identifying the accuracy by language rule,the fluency by language model,and the adequacy by word alignment results and dictionary information.When the translation quality auxiliary examination system based on this method was applied to Million Patent Translation Project of the State Intellectual Property Office,compared with the test group not using this system,the test group using this system has higher translation quality and completion efficiency of final results.