Not Just a Bunch of Words: The Relevance of Prosody in (Machine) Interpreting

Recent advances in artificial intelligence have stimulated interest in the potential of machine interpreting (MI) to complement – or eventually rival – human performance. Grounded in Relevance Theory, this study investigates the treatment of prominence in simultaneous interpreting of European Parliament speeches, comparing outputs from human interpreters and three MI systems. Using Praat, a corpus of 35 multilingual speeches is analysed in order to assess the alignment between source – and target-text prosody. Findings confirm that, while human interpreters variably reproduce prosodic cues under cognitive load, MI systems disregard them, leading to pragmatic impoverishment. The results highlight the need to reconceptualise interpreter expertise, with training emphasising pragmatic adequacy, and to inform future MI development through prosody-aware modelling.