DSTC6 has the following three tracks:
Please go to the linked page to find the detail specification of each track.
End-to-End Goal Oriented Dialog Learning |
End-to-End Conversation Modeling |
Dialogue Breakdown
|
|
|
|
|
This task is aiming to build End-to-End dialog systems for goal-oriented applications. Goal-oriented dialog technology is an important research issue and End-to-End dialog learning has emerged as a primary research subject in the domain of conversation agent learning. It consists in learning a dialog policy from transactional dialogs of a given domain. In this task, the automatic system responses generated using a given task-oriented dialog data will be evaluated. |
Human-to-Human dialog data is a good resource to train conversation models to mimic human dialog behaviors. The target of the track is to generate responses of dialog systems automatically using End-to-End training of neural networks using Human-to-Human dialog corpus. The conversational models for (a) fully data-driven dialog and (b) knowledge grounded dialog will be tested in this track. |
The track is aiming to detect whether the system utterance causes dialogue breakdown indicating a situation in a dialog where users cannot proceed with the conversation in a given dialog context. The participants of the dialogue breakdown detection track will develop a dialogue breakdown detector that outputs a dialogue breakdown label. |