人机耦合视域下的机器翻译质量评价方法

Quality Evaluation Methods of Machine Translation from the Perspective of Human-Machine Coupling

在传统的翻译质量评价研究中一直存在“普适指标”和“功能指标”的争议。普适指标追求为所有翻译作品建立客观、统一的评价标准,而功能指标则认为由于不同的翻译功能和目的,翻译质量也应当有不同的评价方式。面对多种多样的机器翻译评价方式,文章试图按照人工评价和机器自动评价两大范式对现有评价方式进行分类。其中,人工评价包括心灵主义方法、错误类型标注和翻译使用者评价方法;机器自动评价包括错误类型标注和对照人工方法。通过对前人评价方法进行归类,文章期待梳理现有机器翻译质量评价标准,为后续质量研究的方法选择提供参考。

In traditional translation quality evaluation research,there has always been the controversy between"universal indicators'and"functional indicators".Universal indicators pursue the establishment of objective and unified evaluation standards for all translation works,while functional indicators believe that due to different translation functions and purposes,translation quality should also have dif-ferent evaluation criteria.Faced with a variety of machine translation evaluation methods,this article attempts to classify existing evaluation methods according to the two paradigms of human evaluation and machine automatic evaluation.Human evaluation includes mentalism method,error type annotation and user evaluation method;machine automatic evaluation includes error type annotation and human-target method.By classifying previous evaluation methods,this article hopes to sort out the existing machine translation quality evaluation stand-ards and provide a reference for method selection in subsequent quality research.