Researchers propose the task of chess strategy verbalization to describe the underlying strategies behind chess engine move suggestions in natural language, addressing the difficulty human players face in comprehending these plans.
The work introduces a pipeline for verbalizing strategies and an evaluation framework for objectively assessing the generated descriptions. Experiments demonstrate that natural language serves as a promising and interpretable medium for communicating strategic information to both human and LLM players.
The study highlights key insights, including the importance of evaluating strategies beyond the main line, the limitations of pure concept-based descriptions, and the constraints of relying on LLMs rather than humans for evaluation.