Referential Game

Referential games are interactive tasks where agents communicate to identify a target object or concept, advancing our understanding of language emergence, grounding, and cross-cultural communication. Current research focuses on improving agent performance through coupled comprehension and generation models, leveraging techniques like contrastive learning and reinforcement learning (e.g., PPO) to optimize communication strategies and adapt to diverse listener behaviors. These studies contribute to both theoretical advancements in artificial intelligence, particularly in areas like emergent communication and multi-agent collaboration, and practical applications such as human-computer interaction and cross-lingual communication systems.