Putting the dialogue into dialogue modelling - Invited talk

Current dialogue systems in use (such as Siri, Alexa etc) and in development (LaMDA) are built using massive amounts of text data, which often includes no dialogic interactions whatsoever. Even those which claim to use dialogue in their training data tend to use impoverished text-based dialogues from e.g. online forums. This means that the training data and the situations of use of spoken dialogue systems are different in interesting and relevant ways. I will discuss some of the impacts of this and potential solutions.
Research areas:
Type of Publication:
In Proceedings
Book title:
Proceedings of the Young Researchers Roundtable on Spoken Dialogue Systems
Invited talk
Hits: 677