온라인 회의 환경에서 실시간 음성 번역 성능 분석: FAR 모델 기반 EventCat 사례 연구
Evaluating real-time voice translation in online meetings: A FAR Model analysis of EventCat performance
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study investigates the quality of real-time voice translation without post-editing in an online meeting context, focusing on the AI interpretation engine, EventCat. While recent advancements in neural machine translation have led to widespread expectations of near-perfect automated interpretation, few empirical studies have examined its performance in real-time spoken scenarios. Unlike most prior research that primarily analyzed post-edited content, this study evaluated translation quality in a live-like setting by inviting the AI interpreter to a Zoom meeting and streaming actual speeches from formal events as input. Translation quality was evaluated through Pedersen’s (2017) FAR model, a framework that measures functional equivalence, acceptability, and readability. Results indicate that unedited machine translation (MT) output still does not meet the quality standards required in high-stakes situations, particularly in maintaining semantic coherence and contextual fidelity. This research contributes to the growing body of literature on real-time MT evaluation by extending the scope to spontaneous speech in virtual conferencing environments and providing empirical evidence on the limitations and potential of AI interpretation tools in professional settings.
Abstract
This study investigates the quality of real-time voice translation without post-editing in an online meeting context, focusing on the AI interpretation engine, EventCat. While recent advancements in neural machine translation have led to widespread expectations of near-perfect automated interpretation, few empirical studies have examined its performance in real-time spoken scenarios. Unlike most prior research that primarily analyzed post-edited content, this study evaluated translation quality in a live-like setting by inviting the AI interpreter to a Zoom meeting and streaming actual speeches from formal events as input. Translation quality was evaluated through Pedersen’s (2017) FAR model, a framework that measures functional equivalence, acceptability, and readability. Results indicate that unedited machine translation (MT) output still does not meet the quality standards required in high-stakes situations, particularly in maintaining semantic coherence and contextual fidelity. This research contributes to the growing body of literature on real-time MT evaluation by extending the scope to spontaneous speech in virtual conferencing environments and providing empirical evidence on the limitations and potential of AI interpretation tools in professional settings.