The dialogue breakdown detection challenge: Task description, datasets, and evaluation metrics