Meet dataChess and CARLSy: Towards the Explanation of Chess Plays

ACL ARR 2025 February Submission1160 Authors

Abstract: We introduce dataChess, a curated dataset of annotated chess games, and CARLSy, a language model designed to explain the quality of chess moves through natural language. By fine-tuning a Large Language Model, we developed a specialized model for chess commentary generation that leverages recent advancements in Natural Language Processing. Our evaluation - both automatic and through a human study - demonstrates that our model produces commentary on par with state-of-the-art systems. However, it also reveals key challenges that can be addressed to enhance quality and reduce variability in the generated explanations.

Paper Type: Short

Research Area: NLP Applications

Research Area Keywords: educational applications, corpus creation

Contribution Types: NLP engineering experiment

Languages Studied: English

Submission Number: 1160