摘要

由於人工智慧（AI）的發展，使得神經機器翻譯（TAN）應用的興起，其可及性與廣泛使用，不僅在台灣法語系學生之間相當普及，在翻譯專業領域亦然，這對翻譯課程的教學產生了重大影響。翻譯課程本已在語言學習與職業技能培養之間搖擺，如今這一新變局更加複雜了教學方法的選擇，使得課程設計急需重新思考。
本研究旨在比較不同的翻譯教學方法，分析其優勢與劣勢，同時強調在教學中規範性地使用神經機器翻譯之必要性。為此，我們在台灣的高等教育環境中設計並實施了兩種類型的翻譯習題（其一允許使用神經機器翻譯，其二禁止使用神經機器翻譯），並引入了一項後編輯（post-editing）習題，其靈感來自比利時魯汶天主教大學（UC Louvain）所發展的MTPEAS分類法。
在AI 所帶來的新興思考範疇，相比之下傳統的中法翻譯教學習題是否已展現出過時的跡象 ？此外，若在課堂中廣泛採用入機器翻譯-譯後編輯類型的習題，是否能以更恰當的方式整合TAN的應用，從而減少其帶來的弊病？

關鍵詞： 翻譯、神經機器翻譯、後編輯、入機器翻譯-譯後編輯、人工智慧

Abstract

The rapid rise of neural machine translation (NMT) applications, driven by advances in artificial intelligence (AI), accessibility and widespread use by students in French departments at Taiwanese universities, as well as in the professional translation sector, has profoundly disrupted the teaching of translation courses. These courses were already fluctuating between language learning and the acquisition of professional translation skills. This new context further complicates the pedagogical approach to translation teaching, calling for a thorough rethinking. This study therefore aims to compare different approaches to translation teaching, to highlight their respective advantages and limitations, while emphasizing the need to develop a guided and pedagogically sound use of NMT. To this end, two types of translation exercises were conducted: one with access to NMT tools and one without, along with the introduction of a post-editing exercise in a Taiwanese university context, inspired by the MTPEAS typology developed by UC Louvain. In light of the emergence of this new AI-driven paradigm, to what extent do traditional translation exercises reveal obsolete characteristics? Moreover, could the broader implementation of post-editing exercises enable an intelligent integration of NMT into translation courses while reducing its drawbacks?

Key words: translation; NMT; post-editing; MTPE; AI


DOI:10.30404/FLS.202512_(42).0004