When it comes to language, ML has for a long time only approached the problem of language understanding, and was striving to do so using supervised learning to find statistical regularities on large amounts of data. This narrow focus was leaving aside language emergence (within a population of autonomous agents, using reinforcement learning (RL), without relying on pre-recorded natural language data) and language grounding (in some other modalities, e.g. sight, with more complex inputs than text).

Thus, such approaches were failing to capture the functional and interactive aspects of both natural and artificial languages. Traditional ML approaches were ignoring the how and why of language use: to communicate and facilitate cooperation between autonomous agents. In contrast, research in Emergent Communication (EC) studies learning to communicate by interacting with other agents to solve collaborative tasks in complex and diverse environments.

Thanks to deep RL’s ability to handle complex data, EC is no longer confined to being studied by linguistics and theoretical AI practitioners, it can now be studied practically in complex multi-agent scenarios. Thus, a recent resurgence and standardization effort manifested itself in the form of four previous successful workshops (2017-2020). Those workshops have gathered the community to discuss how, when, to what end, and how fast communication emerges, producing research later published at top ML venues (e.g., ICLR, ICML, AAAI).

Following those breakthroughs, research directions rapidly branched out to create New Frontiers where applications in diverse domains are leveraging EC. We therefore believe that EC has significant potential to impact a wide range of disciplines both within AI (e.g. MARL, visual-question answering, explainability, robotics) and beyond (e.g. social linguistics, cognitive science, philosophy of language).