EMERGENCE OF GROUNDED, OPTIMALLY COMPOSITIONAL SPATIAL LANGUAGE AMONG HOMOGENEOUS AGENTS
Keywords: multi-agent reinforcement learning, Language Emergence, Cultural Evolution
TL;DR: Emergent communication in multi-agent population with human language characteristics
Abstract: A mechanism of effective communication is integral to human existence. An
essential aspect of a functional communication scheme among a rational human
population involves an efficient, adaptive, and coherent apparatus to convey one’s goal to others. Such an effective macro characteristic can
emerge in a finite population through adaptive learning via trial and error
at the individual (micro) level, with nearly consistent individual learning faculty and experience across the population. In this paper, we study and hypothesize
pertinent aspects of glossogenetics, specifically primal human communication mechanisms, through computational modeling. In particular, we model the
process as a language game within the fabric of a decentralized, multi-agent
deep reinforcement learning setting, where the agents with local learning and neural
cognitive faculties interact through a series of dialogues. Our homogeneous agents seek to achieve the principle of least effort and overcome the poverty of stimulus through efficient concept selection, guided feedback and mirror learning. In our examinations,
we observe the emergence of successful and structured communication among static and dynamic agent populations through consistent and continual learning.
Supplementary Material: zip
Primary Area: reinforcement learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11343
Loading