DeliChess: A Multi-party Dialogue Dataset for Deliberation in Chess Puzzle Solving
Researchers have introduced DeliChess, a new dataset designed to study collaborative reasoning in multi-party dialogues. The dataset comprises 107 dialogues where groups solve chess puzzles, first individually and then collectively after discussion. Analysis indicates that group deliberation improves puzzle-solving accuracy, although the impact of probing utterances on performance is variable. AI
IMPACT Provides a new benchmark for evaluating collaborative reasoning and dialogue systems in complex strategic domains.