This study proposes new methodological features to evaluate computer-assisted interpreting (CAI) tools in remote simultaneous interpreting by comparing performance with different tools when compiling glossaries and consulting them during tasks. The traditional informationprocessing paradigm is superseded by a situated cognition framework that examines multimodal multitasking and interaction in authentic settings. The research project involved a mixed-methods pretest-post-test design across three rounds of data collection (baseline and two post-test rounds). Chinese L1 and English L2 interpreting trainees (n = 22) were split into an InterpretBank group (experimental) and an Excel group (control). Term accuracy, speech fluency, cognitive effort, and overall quality were measured through both qualitative indicators and quantitative metrics. CAI tools were found to enhance interpreter performance, but their effectiveness depends on the usage environment, task complexity, and individual adaptation strategies. This study addresses two significant gaps: the need for more robust evaluation methods for CAI tools and the limited research on their use with distant language pairs. Our findings underscore the importance of keeping interpreters involved in tool development and evaluation processes. The methodology and results offer practical insights into CAI tool design, interpreter training, and professional practice in an increasingly technology-dependent field.
There is growing interest in using machine translation quality estimation (MTQE) metrics and models and large language models (LLMs) to assess information fidelity in interpreting quality automatically. However, studies have been limited to text-based assessments only. This study compared speech-to-speech (S2S) assessment and text-based assessment. The experiment began by segmenting audio recordings of simultaneous interpreting (SI) into one-minute intervals and isolating the source speech and the target interpretations in each segment. We used LLMs, BLASER, and speech embeddings and the last hidden states from HuBERT and Wav2Vec to assess interpreting quality at the speech level. In addition, we explored the use of automatic speech recognition (ASR) for transcribing segments, coupled with human verification and LLM along with MTQE models such as COMET and TransQuest for minute-level text-based assessment. The findings indicate the following: (1) LLMs cannot conduct the speech-based assessment of interpreting quality directly but demonstrate certain capabilities in text-based assessment when evaluating based on transcriptions, displaying a moderately high correlation with human ratings (Pearson r = 0.66); (2) in contrast, BLASER operates directly at the speech level and demonstrates a comparable correlation (r = 0.63) with human judgements, confirming its potential for speech-based quality assessment; (3) a combined metric integrating both S2S and text-based assessments, as proposed in this study, accounts for approximately 47% of the variance in human judgement scores, which highlights the potential of integrated metrics to enhance the development of machine-learning models for assessing interpreting quality. Such metrics offer an automated, cost-effective, and laboursaving method of evaluating SI by human beings and enable direct quality estimation in end-to-end speech-to-text (S2T) and S2S machine interpreting (MI) systems for continuous quality monitoring during training and deployment.
This study investigated the ways in which professional interpreters' visual attention is captured by real-time source text captions generated by automatic speech recognition (ASR) and realtime target text subtitles generated by automatic speech translation (AST)-both of which are presented in parallel during simultaneous interpreting (SI). Using eye-tracking data, we examined whether attention to ASR and AST varies according to the interpreting direction (L1-L2 vs L2-L1) and how it relates to interpreting accuracy. The findings indicate that the ASR captions captured more attention than the AST subtitles, especially in the L1-L2 direction. In the L2-L1 direction, greater attention to AST subtitles in the interpreters' L1 (Chinese) was linked to a greater level of accuracy. These findings were explained by the interaction of bottom-up and top-down factors. Insights into these attention patterns could possibly inform the design of tailored training sessions to equip interpreters better for SI in technology-assisted environments.
Transformative advancements in technologies such as Automatic Speech Recognition, Natural Language Processing, Deep Learning, and Generative AI have significantly accelerated the evolution of both Machine Interpreting (MI) and Computer-Assisted Interpreting (CAI), fundamentally reshaping the interpreting ecosystem. MI has progressed from being based on statistical machine-translation (MT) models to being augmented by large-language and multimodal models, while undergoing a transition from cascade to end-to-end systems. These developments have markedly enhanced MI's capacity to manage increasingly complex and diverse domains, linguistic features, and operational contexts. CAI, originally designed to streamline the preparatory processes for interpreting, has evolved to incorporate real-time functionalities that are integrated into the interpreting workflow, enabling the effective management of complex terminology and other problem-triggers. This introduction begins by providing a comprehensive overview of the evolution, current state, and future directions of MI and CAI, followed by an introduction to the seven contributions featured in this special issue. These studies encompass a diverse range of topics, including MI quality evaluation, a comparison of human-machine interpreting products and process, CAI tool assessment in remote simultaneous interpreting (SI), the implications of automatic speech recognition (ASR) for consecutive interpreting, the effect of multimodal inputs, and the user-machine interaction patterns with live captioning in SI. This introduction, along with the seven contributions in this issue, aims to advance the growing body of knowledge on the transformative impact of MI and CAI in reshaping interpreting practice, pedagogy, and research.
Computer-assisted simultaneous interpreting (CASI) is increasingly exposing interpreters to multimodal digital environments by means of added visual input. Yet little is known about how multimodal text and speaker inputs affect overall interpreting quality, cognitive load, and attention dynamics. Drawing on analyses of quality assessment, self-reported questionnaires, and eye-tracking data, this study compared CASI with conventional SI regarding the effects on overall interpreting quality, cognitive load, and attention dynamics. In it, 30 student interpreters performed simultaneous interpreting (SI) from English to Chinese across four input conditions: audio-only, audio-video, audio-text, and audio-video-text. The results indicate that the visual input provided by CASI tools significantly improved their interpreting quality and reduced their cognitive load; moreover, the audio-video-text CASI condition yielded the highest interpreting quality and a relatively low cognitive load. In addition, the student interpreters strategically adjusted their attention across different visual input conditions. When audio, video, and text inputs were available concurrently, they prioritized textual input over both auditory cues and speaker video and demonstrated distinct patterns of strategic coordination. In this article, the theoretical, practical, and pedagogical implications of these findings are discussed.
