The Five Ws of Multi-Agent Communication: Who Talks to Whom, When, What, and Why - A Survey from MARL to Emergent Language and LLMs

Multi-agent sequential decision-making underlies many real-world systems, from autonomous vehicles and robotics to collaborative AI assistants. In dynamic and partially observable environments, effective communication is crucial for reducing uncertainty and enabling coordination. While research in multi-agent communication (MA-Comm) spans diverse methods and paradigms, its central challenges can often be understood through the guiding lens of the Five Ws of communication: who talks to whom, when to speak, what to convey, and why communication is beneficial. These questions provide an intuitive thread across different approaches, even when not used as explicit section divisions. Progress in this field has been rapid. Within the Multi-Agent Reinforcement Learning (MARL) framework, early work emphasized static, hand-designed protocols, while later approaches introduced trainable, end-to-end communication models optimized with deep learning. This shift sparked interest in emergent language, where agents develop symbolic or structured messaging strategies through interaction. More recently, large language models (LLMs) have opened new possibilities, enabling natural language as a medium for reasoning, planning, and collaboration in more open-ended environments. Despite this momentum, there is still no dedicated survey that brings together these different lines of work. Most existing reviews focus narrowly on MARL, without fully addressing how communication is evolving from simple message passing to symbolic reasoning and language use. This paper aims to fill that gap. We provide a structured survey of MA-Comm, spanning traditional MARL approaches and emergent language studies. In light of growing interest in agentic and embodied AI, we also examine how LLMs are reshaping communication in both MARL contexts and broader multi-agent ecosystems. By using the Five Ws as a conceptual lens, our goal is to clarify the landscape, highlight key trends, and provide a foundation for future research at the intersection of communication, coordination, and learning in multi-agent systems.