同步口譯與機器即時字幕翻譯之觀感與偏好：以臺灣觀眾為例

Abstract

人工智慧科技的進步帶動了即時字幕翻譯的發展，Wordly AI和Zoom等平台已能提供即時字幕翻譯服務，使口譯員開始擔心職業會受到威脅。儘管目前已有不少研究比較機器與人類的翻譯能力，卻少有研究探討觀眾對聆聽人類口譯與閱讀機器即時字幕翻譯的觀感差異。 本研究招募29位臺灣聽眾並調查其對同步口譯與機器即時字幕翻譯之觀感與偏好，為往後更大規模的相關研究提供參考。研究參與者首先觀看一段有英文及俄文演講的影片，兩演講均提供同步口譯或機器即時字幕翻譯服務，影片及翻譯服務的安排並採用對抗平衡設計抵銷下列三項變數的影響：翻譯模式（聆聽同步口譯或閱讀機器即時字幕翻譯）、翻譯的來源語（英文或俄文）、演講的順序（英文先或俄文先）。影片結束後進行焦點團體訪談，探討參與者對同步口譯及機器即時字幕翻譯之觀感與偏好，以及背後原因。 結果顯示，參與者整體給予口譯員的表現高度評價，並偏好使用其服務，該觀點主要受到參與者對兩翻譯模式翻譯品質的主觀看法影響，參與者對來源語及模式的熟悉度則為形塑其觀感與偏好的次要影響因素。令人意外的是，信任感在本研究中並未造成顯著影響。理想即時翻譯模式方面，即使許多參與者表示同步口譯已是令人滿意的模式，但有更多參與者希望同步口譯內容能以字幕方式呈現，顯示參與者仍習慣閱讀字幕，且即時翻譯服務未來可朝向人機器合作發展。 本研究也發現，部分參與者無法容忍同步口譯當中出現短暫的時間差及少數不自然停頓和重複語句，並認為機器即時字幕翻譯軟體在精確度、客觀度、精力等方面均更勝口譯員，顯示其對口譯員抱有不切實際的期待，並對機器即時字幕翻譯軟體的能力抱有過度信心，背後原因可能是參與者使用同步口譯的經驗不足，或是對現代科技的知識有限。 綜上所述，本研究提供臺灣研究參與者對同步口譯與機器即時字幕翻譯觀感與偏好的概覽，並整理出兩翻譯模式在參與者眼中的優缺點及理想模式，為同步口譯及機器即時字幕翻譯的未來發展提供參考依據。結果顯示，口譯員和即時翻譯軟體不是敵人，兩者應攜手合作，為即時翻譯使用者提供最理想的解決方案。

Advancements in AI technology have enabled the development of live translated subtitling services, such as Wordly AI and Zoom’s live subtitles, sparking concerns among spoken language interpreters about the potential threat to their profession. While there has been substantial research comparing the translation abilities of machines and humans, little attention has been given to the perceptions of audience members when comparing listening to human interpreters and reading machine-generated live subtitles. This study reports the findings of the pilot phase of a larger investigation into the preferences and perceptions of 29 Taiwanese audience members regarding simultaneous interpreting (SI) and machine-generated live subtitles (MGLS). Participants in the study watched a video featuring an English and a Russian speech, which were either simultaneously interpreted into Chinese by a human interpreter or translated into Chinese subtitles using MGLS software. The participants were counterbalanced across three variables: the mode of translation provided (listening to SI or reading MGLS), the source language interpreted or subtitled (English or Russian), and the sequence of the two speeches (English or Russian first). Focus group interviews were then conducted to explore participants’ perceptions and preferences of MGLS and SI and the underlying reasons. The results indicate that participants generally rated interpreters’ performance highly and preferred their services. Their responses were mostly influenced by their perceived quality of the modes. While source language familiarity and mode familiarity also played a role in shaping their perceptions and preferences, trust was surprisingly not a significant factor. Regarding participants’ ideal live translation modes, while many thought SI was already satisfactory, more hoped that SI could be transcribed and shown as subtitles, indicating participants’ habit of reading subtitles and suggesting human-machine cooperation as a possible future development of live translation services. This study also discovered that some participants tended to have impractical expectations for interpreters and were overly optimistic about MGLS’s abilities, as they could not tolerate a few seconds of time lag and a few unnatural pauses and sentence repetitions in SI, and they thought MGLS was all the better in terms of accuracy, objectivity, and stamina, which may be due to their inexperience in using SI and limited knowledge about modern technology. In conclusion, this study provides an overall look into Taiwanese participants’ perceptions and preferences of SI and MGLS, highlighting the current perceived strengths and downsides of each mode, and indicating the participants’ ideal modes to serve as a reference for future SI and MGLS development. The results suggest that interpreters and live translation software are not rivals. Instead, they should work together to generate an optimal solution for live translation users.