Abstract
Current research on dialogue systems focuses on dialogues between two interlocutors, building dialogue agents either for task completion or for user engagement.
To encourage the development of multi-party dialogue systems in the context of group problem-solving we introduce a novel dataset where a group of participants are solving a cognitive task. The dataset consists of over 200 dialogues and 6500 utterances. Furthermore, to facilitate better conversation analysis and bot development we propose a novel annotation schema that captures deliberation cues and will release 50 dialogues annotated with it.
