語音辨識系統提升逐步口譯數字精確度之探究

Abstract

本研究旨在探討自動語音辨識工具（ASR, Automatic Speech Recognition）是否能協助口譯員在逐步口譯中提高處理數字的正確率。本研究共邀請十位受試者進行實驗，使用 iPhone 內建的語音轉錄功能，並採質性與量化研究。受試者分成 A、B 兩組，針對兩段同樣的文本（段落一及段落二）進行口譯。A 組翻譯段落一時有 ASR 的即時語音轉錄稿，段落二則無；B 組翻譯段落一時無 ASR ，段落二則搭配 ASR 工具。整體口譯結束後對受試者進行訪談，收集質性數據以了解口譯員對於 ASR 工具的看法。實驗量性數據使用成對樣本 t 檢定分析，結果顯示 A 組在有 ASR 幫助的情況下，參與者的數字準確度有顯著提升，但 B 組卻無。但深入分析兩組表現後，發現 B 組有兩位受試者並無參考 ASR 工具。將其排除後，剩下的八位受試者中有七位在 ASR 輔助的情況下皆表現較佳。另外，根據研究者的觀察以及受試者的訪談，對 ASR 工具是否熟悉以及譯者本身對於數字在英中之間的語際轉換技巧都會影響 ASR 能夠發揮的效用。實驗數據搭配質性訪談顯示，ASR 工具確實有助於提升數字精確度，但 ASR 工具也有其優缺。優點為有助於處理語速過快的語料、提供口譯員安心感；缺點為令人分心、介面不友善、影響專業形象。口譯員使用工具的方式並無定論，但大部分受試者表示未來想要使用 ASR 工具進行口譯，也認為在口譯訓練中能略為介紹 ASR 工具。綜合來看，ASR 工具在具有數字的逐步口譯中具潛在輔助價值。本研究對於口譯訓練及口譯實務具有一定價值，可作為口譯教師或口譯員實務上的參考依據。This study aims to examine whether Automatic Speech Recognition (ASR) tools can assist interpreters in improving number accuracy during consecutive interpreting. A total of ten participants were recruited for the experiment. The ASR tool was the built-in Dictation feature on the iPhone. The study adopted a mixed-methods design, combining both quantitative and qualitative approaches. Participants were divided into two groups: Group A and Group B. Both groups interpreted two same segments. Group A was supported with ASR transcripts for Segment 1 but not for Segment 2, whereas Group B completed Segment 1 without support and Segment 2 with the ASR tool. After completing both interpreting tasks, participants were interviewed regarding their perceptions of the ASR tool. The data collected was then subject to paired-samples t-tests. Results showed that Group A showed higher accuracy in number rendition when supported by ASR, while Group B did not show noticeable difference. However, further analysis showed that two participants in Group B did not consult ASR. Excluding the two participants, a total of eight participants consulted ASR and seven of them showed better performance. In addition, based on observational data and the retrospective interview, interpreters’ familiarity with ASR and their number conversion skills between English and Chinese affects the effectiveness of ASR tools. Combined with quantitative data, the qualitative data revealed that the ASR tool helped enhance numerical accuracy and there are advantages and disadvantages of using ASR. For benefits, ASR helps interpreters to tackle fast speech rate and gives users a sense of reassurance. As for the drawbacks, ASR can be distracting, has an unfriendly interface, and undermines professional image. How interpreters use the tool varies, but most participants expressed interest in using ASR tools in future interpreting tasks and suggested that basic introduction of such a tool could be involved in interpreting programs. Overall, the findings indicate that ASR holds potential as a supportive aid in consecutive interpreting tasks that have numbers. This study offers valuable implications for both interpreter training and professional practice, and may serve as a reference during training or interpreting classes.