Live commentary in a football video game generated by an AI

Czaplicki, Michał (2023)

This paper is about an innovative approach to creating live football commentary for a video game using Artificial Intelligence (AI) technology. It is a possible solution to the increasing cost and time of computer game production nowadays. I propose using Generative Pre-trained Transformer-3.5 (GPT-3.5) to generate live commentary that dynamically updates with the game state. GPT-3.5 gets as input pre-written football commentary templates filled with game state variables, such as the score, time, player actions, etc. Next, it will be asked to rewrite them as a football commentary. I evaluated the system to assess the accuracy of the generated commentary and enjoyability for the user. My approach could be extended to other sports and video games, making it a valuable asset for the game development industry. The results show that such commentary is overall enjoyable for the players despite some limitations.
Czaplicki_BA_EEMCS.pdf