英汉机译错误分类及数据统计分析

作为一项历时5年探索机器翻译译后编辑自动化研究的重大阶段性成果,该项目采用汽车技术文献作为研究对象,通过对10万个英汉翻译句对做机器译文和人工标准译文的对比分析,发现机译典型错误并进行数据统计分析。介绍该研究所依据的机译错误分类标准,并以实例加以说明,得出的机译错误分析数据统计显示,机译错误率超过八成,其中词汇类错误占七成。本研究探索的机器翻译译后编辑新思路,对于降低机器翻译人工成本、 提高机译效率和质量具有较高的实际价值和理论意义。

Error Patterns and Statistical Analyses of an English-Chinese Machine Translation Corpus

This paper is based upon a five-year research project on a new attempt at post-editing of machine translation (MT).The research identified the MT error patterns of a text corpus of 100 000 sentence pairs of a car maintenance manual and did statistical analyses of the MT errors.Examples are used in this paper for illustrations of the error patterns.Our statistical analyses reveal that the MT errors reach as high as over 80%,and among the MT errors lexical errors stand at 70%.This research explored a cutting-edge attempt at MT post-editing and was of great significance in improving the MT quality and efficiency as well as lowering the MT cost.