評估 Google 翻譯表現 —以 4 種語言翻譯 100 篇英文論文為例

Abstract

隨著翻譯的需求與日俱增，各種翻譯軟體如雨後春筍般湧現。時逢論文抄襲爭議層出不窮，因此許多學校發布新的相關規定來因應，要求在提交論文前需通過指定的論文相似度檢測。若有人利用翻譯軟體將他人的論文經過多層翻譯後，當作自己的論文，則如何評估翻譯的表現也因此成為了一個重要的議題。在眾多案例中，中國蘇州大學社會學院的學生邵寶在其博士論文《清末留日學生與日本社會》中，在未引用或標注來源的情況下將日本學者酒井順一郎的著作翻譯至中文後直接使用，相似比例達到其論文的80%，而我們決定採用其中翻譯相似的概念，以翻譯軟體生成不同版本的譯本後，再透過論文比對系統來評估該翻譯軟體的表現。 在本論文中，我們會瞭解檢測翻譯表現這個領域的發展現況，並以100篇英文論文為我們的資料。我們將論文相似度比對系統Turnitin作為檢測Google翻譯軟體表現的工具，來分析牽涉到中文、日文、法文和德文的翻譯中不同的翻譯方式對論文相似度的影響，包括單層翻譯和雙層翻譯。結果顯示出翻譯方式對論文相似度具有顯著影響，其中牽涉到日文的翻譯表現較差，而牽涉到中文的翻譯表現較佳。

With the increasing demand for translation, various translation software has emerged rapidly. Amidst ongoing plagiarism controversies, many universities have implemented new regulations requiring theses to pass designated similarity checks before submission. If someone uses translation software to translate another person's paper multiple times and then presents it as their own, evaluating the performance of such translations becomes a critical issue. A notable case is that of Shao Bao, a student at the School of Sociology at Soochow University in China, who translated and used approximately 80% of Japanese scholar Sakai Junichiro's work "Qing Dynasty Chinese Students in Japan and Language and Culture Contact" in his doctoral dissertation without proper citation. Based on this case, we adopt the concept of translation similarity by using translation software to generate different versions of translations and then using a thesis similarity detection system to evaluate the accuracy of the translation software.

This study explores the current state of translation accuracy evaluation, using 100 English academic papers as the dataset. Turnitin, the similarity detection system is employed to assess the performance of Google Translate, analyzing the impact of different translation methods—single-layer and double-layer translations—on thesis similarity across Chinese, Japanese, French, and German translations. The results indicate that translation methods significantly affect thesis similarity, with translations involving Japanese performing worse and those involving Chinese performing better.