未編輯自動語音辨識字幕對中譯英同步口譯之影響

Abstract

本研究旨在探討未經編輯之自動語音辨識（ASR）字幕於中進英同步口譯（SI）任務中的影響，其中又聚焦於口譯員對ASR系統之理解、認知負荷的變化，以及在處理數字、日期與專有名詞等口譯難點時之表現。 本研究邀請十位經過正規訓練之專業口譯員，分別於有ASR輔助與無ASR輔助之情境下進行兩段同步口譯。透過口譯輸出分析、NASA-TLX認知負荷量表、眼動追蹤數據，以及回顧性訪談等多重方法進行資料蒐集與分析。量化結果顯示，在ASR輔助情境下，口譯員於數字及日期之翻譯準確度顯著提升。NASA-TLX結果則進一步顯示，ASR輔助有效降低口譯員主觀之認知負荷。而眼動追蹤數據則揭示，於ASR輔助條件下，口譯員之視覺焦點更多集中於ASR字幕區，並能適時利用ASR字幕解決口譯難點，惟亦需面對ASR潛在的轉錄錯誤。 質性分析方面，ASR輔助對口譯員於理解與確認數字及專有名詞時具有顯著助益。然而，當ASR字幕出現錯誤或延遲時，亦可能造成額外干擾，迫使口譯員分散注意力進行監控與修正。 綜上所述，本研究指出未經編輯之ASR字幕對同步口譯任務具有雙面效應：一方面，ASR輔助能提升口譯員於特定難點（如數字與專有名詞）之譯文準確性；另一方面，ASR錯誤或延遲輸出則增加了口譯員注意力分散之風險。本研究的發現進一步深化了對ASR在中文至英文同步口譯應用中效益與限制的理解，亦強調提升ASR準確性之必要性，以期更有效地輔助口譯工作。

This study investigates the effects of unedited automatic speech recognition (ASR) support on simultaneous interpreting (SI) from Mandarin Chinese to English, focusing on interpreters' perceptions, cognitive load, and performance when encountering problem triggers such as numbers, dates, and proper names. Given the challenges interpreters face in real-time processing and managing cognitive overload, ASR has emerged as a potential tool to alleviate these demands. The study involved ten institutionally trained interpreters performing two SI tasks—one with ASR support and one without. The study collected data from interpreting outputs, NASA-TLX workload scores, eye-tracking software, and retrospective interviews. Quantitative analysis of performance accuracy showed significant improvements in rendering numbers and dates with ASR support. The NASA-TLX results revealed a statistically significant reduction in perceived cognitive workload when ASR was available. Eye-tracking data confirmed increased visual attention on the ASR display, with interpreters leveraging ASR for problem triggers while mitigating its limitations. Qualitative analysis highlighted that ASR support was helpful for comprehension and confirmation of numbers and names but also introduced distractions due to transcription errors or delays. The study concludes that unedited ASR has a dual effect on interpreter performance: it enhances output accuracy for problem triggers while demanding additional attention for error monitoring. These findings contribute to understanding the practical benefits and limitations of ASR integration in Mandarin Chinese-English SI and underscore the need for improved ASR accuracy to fully support interpreters.