Based on FIPM (Football Interaction and Process Model [1]) - Representation of actions and game progression based on a formal notation using a series of facts, statements and rules with probabilistic and possibilistic arbitration to generate commentary.
Five Layers - Position and Motion, Action, Situation, Tactical, Assessment.
Commentary Generator
• Factual Model - Comment on simple actions taking place: Passing, Shooting, Offside, Fouls etc.
• Analytical Model - Comment on tactical implications of actions: importance of a Penalty, Red Card decision, Substitutions etc.
• Predictive Model - Comment on perceived opportunities, speculative commentary on e.g. possible substitutions or shooting opportunities etc.
• Emotional Model - Convey the broad range of emotions: Excitement, Disappointment, Anger and Disgust through voice accentuation and prosody based on action context and agent personality.
• Background Knowledge Model - Comment on Team or Player-specific background knowledge contained in a database, such as the number of games a player has played without scoring etc.
• Discourse Model - Simulate the conversational aspect of two interactive agents engaging in dialogue, such as interruption etc.
The above commentary models would be based on a dissection of the real-life counterparts in order to faithfully imitate the commentary skills of a professional football commentator with respect to perceived personality and presentation style. This research will work towards the advancement of human-like real-time commentary generation agents for complex and dynamic multi-agent systems.
References
[1] Beetz et al., "Computerized Real-Time Analysis of Football Games", IEEE CS and IEEE ComSoc, 2005.
