Gloria Virginia¹, Jeffrey Susilo¹, and Aditya Wikan Mahastama¹, Informatics Department, Universitas Kristen Duta Wacana, Indonesia

This study presents an automated football commentary system that integrates computer vision, rule-based reasoning, and natural language generation within a cognitively inspired framework. The modular design employs a YOLOv8 model for real-time object detection, K-Means clustering for team classification, and a rule-based reasoning module for event recognition based on spatial and temporal conditions such as ball possession and dead-ball states. Detected events are transformed into expressive, sportcaster-style commentary using a generative language model. Experiments on real football match videos and evaluations by 30 football enthusiasts using a five-point Likert scale demonstrated high detection precision, reliable team, identification and accurate recognition of key events. Beyond technical performance, the framework promotes inclusivity through potential extensions such as multilingual commentary, closed captioning, and audio description, supporting equitable access for diverse audience.

Cognitive AI, Computer Vision, Rule-Based Reasoning, Natural Language Generation, Automated Football Commentary