Overview

Emergent communication studies how agents with limited or different observations invent signals to coordinate, describe shared worlds, and establish conventions without a fully specified language. Researchers use signaling and referential games, multi-agent reinforcement learning, evolutionary simulations, and interactions among humans or language-model agents; they evaluate task success, compositionality, transfer, and interpretability. Methods such as iterated learning, vector quantization, information bottlenecks, and prompt optimization shape messages and test their robustness under bandwidth limits. A central tension is that protocols can work well within a pair yet fail across agents, seeds, or settings, and more communication is not always useful when agents already share information through their environment.

Latest papers

CardsList

32 papers2 papers in the last four weeks, down 60% on the four weeks before. 0.0% of all new papers.

  1. VQEL: Enabling Self-Play in Emergent Language Games via Agent-Internal Vector Quantization

    Mar 6, 2025Mohammad Mahdi Samiei Paqaleh, Mehdi Jamalkhah, Mahdieh Soleymani BaghshahEmergent Communication

  2. ETHER: Aligning Emergent Communication for Hindsight Experience Replay

    Jul 28, 2023Kevin Yandoka Denamganaï, Daniel Hernandez, Ozan Vardal +2Reinforcement LearningExperience Replay