A comparison of Google translate and Youdao translation : a case study of sentences Selected from Financial Times

• المؤلف الرئيسي: Zhang, Xue Qing
• مؤلفون آخرون: Helena Gao
• التنسيق: Theses and Dissertations
• اللغة: English
• منشور في: 2019
• الموضوعات: Humanities::Language
• الوصول للمادة أونلاين: http://hdl.handle.net/10356/78898
• المؤسسة: Nanyang Technological University

With the rising demand for cross-cultural communication arising from globalisation, emerging technologies in the IT industry have made machine translation tools the most accessible in meeting one’s daily translation needs. In light of this, this paper aims to conduct a comparison of two mainstream online MT tools, namely Google Translate and Youdao Translation, through determining which tool provides higher English-Chinese translation quality in terms of translating informative content. To achieve this, 90 sample sentences originally in English with official Chinese translation of each sentence were selected from Financial Times, global business newspaper. Translation results from Google and Youdao were first evaluated by an automatic evaluation metric, BLEU, to determine which tool was likely to have translation outputs that are more consistent with the human translated version. For a further comparison, sample sets were reviewed qualitatively and quantitatively based on a linguistic classification framework with regards to the emergence of translation errors. This error analysis part was also designed to provide MT developers cues about the common mistakes committed by MTs from a linguistic perspective. Results obtained from both approaches revealed a higher probability of Youdao in providing quality translations as compared that of Google. In particular, Youdao outperformed Google by an average of 22.9% BLEU score while the total number of linguistic errors produced by Youdao was 48 fewer than that produced by Google. Meanwhile, there was no statistical evidence indicating a negative correlation between sentence length of input content and the output quality of the MT. It was also found that lexical error was the most common type of translation error generated by both MT tools.