CDA视角下对比分析人工翻译与谷歌翻译对政治话语的译文质量

发布时间：2018-04-24 13:49

本文选题：谷歌神经机器翻译 + 批评话语分析　；参考：《北京外国语大学》2017年硕士论文

【摘要】：相比于过去基于短语的机器翻译,近年来发展迅速的端到端神经机器翻译大幅提升了译文的准确度以及可读性。2016年谷歌公司推出的谷歌神经机器翻译系统(GNMT)更是取得了突破性进展:通过使用拥有八个解码层、八个编码层的深度LSTM网络以及连接解码器与编码器的attention机制,GNMT缩小了机器译文与人工译文的质量差距,甚至在不进行直接训练的情况下,通过间接数据得到的翻译模型便可以实现"零数据翻译"(zero-shottranslation)。然而单从字句意思对等的角度评价翻译质量其自身局限性。批评话语分析研究者认为,语言、权力关系以及意识形态三者之间相互作用;语言不仅受到社会因素影响,同时也能反作用于社会。因此笔者在本文中从批评话语分析角度出发,从话语的层面来测评人工译文与机器译文的翻译质量。笔者借用批判话语分析先驱费尔克劳的三维分析模型,以2015年、2016年、2017年三年外交部部长王毅在外交部两会记者会期间就"中国宏观外交政策"和"南海问题"所作回复作为原文样本;按照费尔克劳所述,通过"描述"、"解读"以及"解释"三个步骤对原文进行批评话语分析。笔者通过定性、定量对比分析外交部孙宁的现场翻译与Google Translate对原文样本的翻译,评价两者对于原文信息传递以及实现说话人话语目的的程度。通过对比分析,笔者发现就本文研究对象而言,孙宁译文在词语翻译的准确度以及语篇逻辑衔接上优于谷歌翻译的译文质量。
[Abstract]:Compared to the phrase-based machine translation in the past, The rapid development of end-to-end neural machine translation in recent years has greatly improved the accuracy and readability of the translation. In 2016, Google's Google Neural Machine Translation system (GNMTS) made a breakthrough: by using eight decode layers, The depth LSTM network of the eight coding layers and the attention mechanism of connecting the decoder to the encoder reduce the quality gap between the machine translation and the manual translation, even without direct training. The zero-shot translation can be realized by indirect data translation model. However, it is limited to evaluate translation quality from the perspective of meaning equivalence. Critical discourse analysts believe that language, power relations and ideology interact with each other; language is not only influenced by social factors, but also reacts on society. Therefore, from the perspective of critical discourse analysis, the author evaluates the translation quality of artificial and machine translations from the perspective of discourse. The author borrows the three-dimensional model of critical discourse analysis pioneer, Fairclaw, Taking the responses on "China's macro foreign policy" and "the South China Sea issue" made by Wang Yi, Foreign Minister Wang Yi, at a press conference of the two sessions of the Ministry of Foreign Affairs in 2015, 2016, 2017 as the original sample; according to Fairclaw, The critical discourse of the original text is analyzed through three steps: description, interpretation and interpretation. The author makes a qualitative and quantitative comparative analysis of Sun Ning's on-the-spot translation and Google Translate's translation of the original text samples, and evaluates the extent to which they can convey the original information and achieve the target of the speaker's utterance. Through comparative analysis, the author finds that, as far as the target of this study is concerned, Sun Ning's translation is superior to that of Google's in terms of the accuracy of the translation of words and the logical cohesion of the text.
【学位授予单位】：北京外国语大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：H085;H315.9

【相似文献】