GameCommBench Introduces Unified Evaluation for AI-Generated Game Commentary

Summary

AI-generated game commentary requires multimodal perception, strategic reasoning, and contextual knowledge, but existing studies use fragmented game settings, modalities, and evaluation protocols. The authors introduce GameCommBench, a unified benchmark covering board games, sports, and esports. Its commentary is aligned with heterogeneous game contexts and annotated by commentary type. They also propose Type-Aware Commentary Evaluation (TACE), a structured framework intended to assess different functional forms of commentary rather than relying only on overlap-based or broad holistic scores. The study validates TACE for reliability and agreement with human judgments, then applies it to representative AI commentators. Results show non-uniform capability profiles across systems and identify live observation and strategic analysis as major bottlenecks. GameCommBench and TACE are presented as a basis for more comparable and interpretable evaluation of AI-generated game commentary.