International Workshop on Spoken Dialogue Systems, 2023.
Abstract Dialog systems could benefit from the ability to detect instantaneous user dissatisfaction and other negative micro-sentiments. While this ability has been developed for some specific corpora, the question of generality has not previously been examined. In this paper we investigate whether the prosodic expression of negative microsentiment is similar across three diverse corpora: of commercial dialogs, cooperative gameplay, and casual conversations. We find very weak cross-domain performance and major differences in the patterns found, suggesting that no simple model can handle negative microsentiment in general.
Resources for the Watergirl corpus