Can professional interpreters truly convey the speaker’s sentiment? Exploring the potential of a computational approach

Abstract This study investigates whether sentiment analysis, a natural language processing technique, can be used to examine accuracy in interpreting. The data were obtained from a parallel bidirectional corpus of original speeches delivered at the United Nations and their simultaneous renditions provided by professional interpreters. Specifically, this study explores how much sentiment can be conveyed across languages via accurate renditions, how interpreting direction affects the conveyance of sentiment, and how sentiment analysis may help with accuracy assessment. The results show that the sentiment orientation and distribution expressed in the source text can be largely projected into the target text via accurate renditions. This finding confirms the validity of using translational language to create cross-lingual sentiment analysis tools. It also reveals the potential of integrating sentiment analysis into automated interpreting quality assessment frameworks. In addition, this study shows that the amount of sentiment conveyed in each direction seems to vary, suggesting that directionality has an impact on the emotional tone being communicated by the interpreters.