運用即時字幕駕馭非母語口音：學生口譯員之研究

Abstract

自 COVID-19 疫情爆發以來，線上會議的需求激增，遠距口譯任務亦隨之大幅增加。儘管許多線上會議平台目前已具備內建口譯功能，但遠距會議的便利性也帶來了新的挑戰，特別是放大了非母語口音對口譯員的干擾。在遠距工作環境中，口譯員不再有機會於會前與講者交流並熟悉其口音。所幸，許多線上平台現已提供自動即時字幕功能。過去的研究證實，自動語音辨識能提升數字與專有名詞的譯文精確度，然而，內建自動即時字幕在協助口譯員處理重口音時的潛在助益，仍有待進一步探索。 本研究採混合研究法，結合口譯實驗、刺激回溯訪談以及問卷調查，旨在探討於遠距會議中，當講者具備明顯非英文母語口音時，使用線上會議平台的自動即時字幕是否能提升英譯中同步口譯的精準度。本研究招募了 8 位 A 語言為中文、B 語言為英文的受試者。受試者在 Google Meet 平台上進行了兩場各 10 分鐘的英譯中同步口譯任務，兩場演講皆由具明顯非母語口音的講者發表，其中一場開啟自動即時字幕，另一場則無。 研究者針對口譯產出進行客觀評分，以分析譯文的實際精準度是否與受試者的自我感知表現一致；同時，透過訪談與問卷探討受試者對即時字幕有效性與可用性的看法。研究結果顯示，使用自動即時字幕並未對客觀的口譯產出品質產生統計學上的顯著影響。然而，質性資料指出，受試者在任務過程中發展出了特定的認知與注意力分配策略，以應對非母語口音及動態字幕所帶來的雙重挑戰。本研究期待藉由實證成果，了解自動即時字幕在遠距口譯中如何協助口譯員處理明顯非母語口音，進而對口譯研究領域做出貢獻。

The number of online conferences has increased significantly since the outbreak of the COVID-19 pandemic, leading to a corresponding surge in remote interpreting assignments. While many online meeting platforms now provide built-in interpretation functions, the convenience of remote meetings has introduced new challenges—most notably, the exacerbated impact of non-native accents. In a remote setting, interpreters no longer have the opportunity to interact with speakers and familiarize themselves with their accents in advance. Fortunately, many online platforms offer automatic live captions. Although past studies indicate that automatic speech recognition increases rendition accuracy for numbers and proper nouns, the potential usefulness of built-in automatic live captions for interpreters navigating heavy accents remains largely unexplored. Adopting a mixed-methods approach, this study conducted an experiment, a stimulated retrospective interview, and a survey to examine whether the use of automatic live captions on online meeting platforms can enhance the accuracy of simultaneous interpreting (English to Mandarin Chinese) when the speaker exhibits a strong non-native accent. Eight participants, with Mandarin Chinese as their A language and English as their B language, interpreted two ten-minute speeches on Google Meet, one with automatic live captions and one without. The output was evaluated to determine if objective accuracy aligned with the participants’ perceived performance, while the interviews and surveys explored their perceptions of the captions’ effectiveness and usability. The results indicated that the adoption of automatic live captions did not have a statistically significant impact on the objective quality of the interpretation output. However, the qualitative data revealed that participants developed specific cognitive strategies to address the intersecting challenges posed by both the non-native accent and the multimodal demands of the automatic live captions. This study hopes to contribute to the field of interpreting by providing empirical evidence on whether and how automatic live captions can support interpreters in handling speeches delivered with strong non-native accents in remote settings.