2023/11/03 by Lucas Druart, Druart, Lucas, Léo Jacqmin +7
Computer Science · #Speech and dialogue systems #AI in Service Interactions #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2311.04922
In Task-Oriented Dialogue (TOD) systems, correctly updating the system's\nunderstanding of the user's needs is key to a smooth interaction. Traditionally\nTOD systems are composed of several modules that interact with one another.\nWhile each of these components is the focus of active research communities,\ntheir behavior in interaction can be overlooked. This paper proposes a\ncomprehensive analysis of the errors of state of the art systems in complex\nsettings such as Dialogue State Tracking which highly depends on the dialogue\ncontext. Based on spoken MultiWoz, we identify that errors on non-categorical\nslots' values are essential to address in order to bridge the gap between\nspoken and chat-based dialogue systems. We explore potential solutions to\nimprove transcriptions and help dialogue state tracking generative models\ncorrect such errors.\n