NOESIS II: Predicting Responses, Identifying Success, and Managing Complexity in Task-Oriented Dialogue

Abstract

Real-world conversation often involves more than two participants and complex conversation structures, but most datasets for dialogue research simplify the task to make it more tractable. This shared task built on prior tasks for goal-oriented dialogue, moving towards more realistic settings. Seventeen teams participated in the primary task, predicting the next utterance in a multi-party conversation, and several teams participated in supplementary tasks. All of the datasets have been publicly released, providing a standard benchmark for future work in this space.

Publication
8th Edition of the Dialog System Technology Challenges at AAAI 2019
Jonathan K. Kummerfeld
Jonathan K. Kummerfeld
Postdoctoral Research Fellow

Postdoc working on Natural Language Processing.

Related