探討融入譯後編輯訓練提升外文系學生之翻譯成效

Investigating the Role of Post-Editing Training in Improving the Translation Performance of English Majors

隨著人工智慧的快速發展，機器翻譯技術也迅速進步。Google翻譯透過導入神經網絡技術，大幅提升了翻譯的準確度，尤其在正式文本中，準確率可高達60%至70%。然而，Google翻譯在處理跨語言文本時，仍面臨許多限制，必須仰賴人工校稿與修正。本研究旨在探討外文系學生經過機器翻譯後編輯訓練的翻譯成效及看法，以及如何引導學生察覺錯誤並學習自我修改。研究對象為14位來自臺灣南部某國立大學修習英語寫作課程的外文系學生。研究者以「注意假設」為理論框架，設計為期8週的訓練課程，介紹學生辨識五類常見機器翻譯錯誤，並引導學生修正錯誤以提升翻譯準確度。本研究採混合式研究法，量化研究工具涵蓋前、後測問卷及學生的文本成績；質化研究工具則包含學生翻譯的文本及半結構式訪談。研究結果顯示，學生文本一及文本二的分數有顯著的差異，也就是文本二的平均分數較文本一為高。大部分的學生經過機器翻譯後編輯的訓練，都能夠辨識及修正機器翻譯的五類錯誤。此外，多數學生對此訓練方式表示肯定，認為有助於提升翻譯品質。根據研究結果，本研究提出實用的教學及研究建議。

With the emergence of artificial intelligence, machine translation (MT) has rapidly advanced, including neural network-based models such as, Google Translate (GT). These improvements have led to translations that can achieve accuracy rates as high as 60% to 70% in formal texts. However, GT still lacks the ability to deal with inter lingual texts and requires human’s proofreading and editing. Addressing this gap, this study aims to investigate the translation performance and perceptions of English majors after receiving post-editing training on noticing MT errors and learning to correct these errors by themselves. A total of 14 English majors taking an English composition course at a national university in southern Taiwan participated in this study over eight weeks. Drawing on the noticing hypothesis, the training introduced students to identify five common types of MT errors and guide them to correct these errors to improve translation accuracy. This study adopted a mixed-method research approach. Quantitative data sets included pre- and post- questionnaires and students’ text scores; qualitative data sources included students’ translated texts and semi-structured interviews. The results showed a significant difference in the scores between Text 1 and Text 2, with the average scores of Text 2 being higher than that of Text 1. This indicates that most students are able to notice and correct the five types of MT errors after receiving training on post-editing MT outcomes. Most students responded positively to this training and believed that it helped improve translation performance. Based on the research results, this study provides practical teaching and research recommendations.