Quality Assessment Of C-E Neural Machine Translation

The End-to-End Neural Machine Translation(NMT)has greatly improved the fluency and readability of machine translation.Meanwhile,new problems follows:fluency might be achieved at the cost of faithfulness,and the traditional assessment approach of machine translation--BLEU might not be valid.This study uses President Xi Jinping's keynote speech at the opening of the Boao Forum for Asia Annual Conference 2018 as example,takes the perspective of Error Analysis,develops a new model of quality assessment for neural machine translation,makes word-for-word comparative analysis of the translation made by 3 mainstream NMT system—Google,Baidu and IFLYTEK—through reading literature and making both quantitative analysis and qualitative analysis.This thesis tries to analyze the errors made by neural machine translation in levels of substance errors,text errors and discourse errors.The study shows that most of the errors happened in areas of the recognition of the original text,the breaking down of the sentences,the lexical grammar,the logic,the cultural information and meanings behind the texts;neural machine translation should further enhance its translating ability,reading and understanding ability,and do better in breaking down sentences,choosing the right words,translating under"general context",The quality of neural machine translation is far worse than that of human translators and is not able to replace human translators in a short time.The study suggests that the future translation model should be a "3-Level Model",and human translators are advised to play a higher-level role.