A Corpus-based Analysis of Semantic Similarity in Machine Translation and Human Interpreting

The incorporation of machine translation (MT) in interpreting activities has drawn increasing research attention. However, MT has often been examined as an assistant to interpreters within the framework of computer- or AI-assisted interpreting. Its independent performance in rendering original meaning in interpreting context, particularly in comparison to interpreter performance, has been relatively unclear. Within the Chinese-English language pair, this study examines the performance of MT systems and simultaneous interpreters in conveying semantic meaning, measuring it through semantic similarity using LaBSE, COMET, and partial human assessment. The findings suggest that though MT systems may deliver sufficient semantic meaning, they have yet to reach human parity in conveying implicit semantic cohesion and nuanced language use, which thus requires further fine-tuning.