Does automatic speech recognition help in consecutive interpreting that involves varying accents? Evidence from interpreting performance and cognitive load

Abstract

This study investigated the impact of AI-powered automatic speech recognition (ASR) technology on interpreting performance and cognitive load during consecutive interpreting (CI), particularly in the case of source speeches that feature varying accents. Multiple performance metrics – fidelity, fluency, target language (TL) quality and overall quality – were assessed. Fluency was measured by means of delivery rate, frequency and mean length of silent pauses, and the overall occurrence of disfluencies. Cognitive load was evaluated using subjective self-ratings and an objective measure of fundamental frequency (F0). Twenty-four advanced student interpreters conducted four CI tasks each – two with an unfamiliar accent and two without one – both with and without ASR assistance. The results indicate that the impact of ASR on CI is multifaceted. Whereas ASR improved the interpreting fidelity, it reduced the delivery rate and increased the frequency of silent pauses. This complexity is more pronounced when the source speech features an unfamiliar accent, which could lead to interpreters’ over-reliance on ASR and as a result compromise the TL quality. No significant effect was observed on the overall quality of the interpreting. Notably, ASR did not affect the interpreters’ cognitive load across all the phases of CI, regardless of the presence of unfamiliar accents in the source speeches.