EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Emergent Symbolic Communication for Coordinated Reasoning in Partially Observable Large Language Model Agent Systems

Wen Tang, Gary Hung

INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES · 03.08.2026 · 6 (2) · 13-25

DOI: 10.66238/ijrss.v6i2.118 ↗

Kurz gefasst

Fachartikel in der Zeitschrift INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES, online erschienen am 03.08.2026. Verfasst von Wen Tang und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The deployment of large language models as autonomous agents has demonstrated remarkable capabilities in natural language understanding, reasoning, and tool use. However, when multiple such agents operate in partially observable environments, their ability to coordinate effectively is often bottlenecked by the inefficiencies of standard natural language communication and the cognitive load of processing high-dimensional contextual information. This paper investigates the phenomenon of emergent symbolic communication among large language model agents tasked with cooperative problem-solving under strict partial observability. We propose a novel cognitive architecture that incentivizes agents to compress complex environmental states and reasoning trajectories into discrete, compositional symbolic representations. Through extensive empirical analysis across decentralized decision-making tasks, we demonstrate that agents can autonomously develop a highly efficient, context-dependent symbolic vocabulary that significantly outperforms standard natural language protocols in both transmission efficiency and coordination success rates. Furthermore, we provide a rigorous theoretical framework grounded in information theory to model the symbolic compression process. Our findings reveal that the emergent language exhibits structural properties akin to natural languages, such as Zipfian distribution and compositionality, while being heavily optimized for the specific constraints of the environment. This research advances the understanding of multi-agent coordination, offering a scalable methodology for deploying robust, bandwidth-efficient agent swarms in complex, real-world scenarios.

Zitieren

APA

Tang, W., & Hung, G. (2026). Emergent Symbolic Communication for Coordinated Reasoning in Partially Observable Large Language Model Agent Systems. INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES, 6(2), 13-25. https://doi.org/10.66238/ijrss.v6i2.118

Vancouver

Tang W, Hung G. Emergent Symbolic Communication for Coordinated Reasoning in Partially Observable Large Language Model Agent Systems. INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES. 2026;6(2):13-25. doi:10.66238/ijrss.v6i2.118

BibTeX
@article{Tang2026,
  author = {Tang, Wen and Hung, Gary},
  title = {Emergent Symbolic Communication for Coordinated Reasoning in Partially Observable Large Language Model Agent Systems},
  journal = {INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES},
  year = {2026},
  volume = {6},
  number = {2},
  pages = {13--25},
  doi = {10.66238/ijrss.v6i2.118}
}
Kostenlos starten →