Machine Translation Evaluation Based On A Hybrid Strategy

The remarkable development of machine learning and cloud computing has greatly improved the quality of machine translation outputs,and machine translation engines have been widely accepted,which has brought great changes to the language service industry.Now the flourishing post-editing becomes the mainstream of the industry,and there is a gradual increase in the number of freelancers in language service industry.For these translators,selecting a low-cost and high-quality free machine translation engine is the key to performing post-editing at a lower cost.From the perspective of post-editing freelancers,taking the English-Chinese bilingual texts from Financial Times as samples,the author evaluated the output quality of three free machine translation engines,Google Translate,Baidu Translate,and Youdao Translate with a hybrid strategy,combining the automatic evaluation algorithm BLEU and the human evaluation model Multidimensional Quality Metrics(MQM).The research will provide MT performance evaluation results and a hybrid translation quality evaluation path for the reference of translators and researchers.The results show that Youdao Translate is the best for translating the texts from Financial Times among the three evaluated MT engines.In the automatic evaluation stage,40% of the outputs by Youdao Translate have reached 60% of the quality of reference translations(human translations),and in the human evaluation stage,the number of translation issues generated by Youdao Translate is far lower than that of Baidu Translate and Google Translate.In addition,the translation engine is less affected by the source column of the texts from Financial Times and the issue distributions are stable.However,when working on post-editing tasks with Youdao Translate,translators need to pay extra attention to “Mistranslation” and “Terminology” issues.