Communication Design for Autonomous Meta-Analysis: A Study of Multi-Agent LLM Communication

ACL ARR 2026 January Submission8700 Authors

Keywords: Multi-agent systems, LLM communication, Communication topology, Interaction protocol, Autonomous meta-analysis, Explainability

Abstract: Communication is a central but underexamined design choice in multi-agent language systems. While prior work relies on fixed interaction patterns, it remains unclear how communication structure itself shapes reliability, efficiency, and explainability. We present a communication-centric analysis that isolates communication design within a fixed document-grounded meta-analysis pipeline. By varying communication topology, interaction protocol, and message constraints only during quality-control stages, we enable controlled comparison across task-compatible designs. Using task-grounded metrics, we show that communication structure induces fundamental trade-offs. Highly connected interaction maximizes error correction but incurs high coordination cost and diffuses responsibility, whereas structured topologies such as committee-based and hierarchical verification achieve \emph{competitive} reliability at significantly lower cost Our results demonstrate that communication structure is a primary determinant of multi-agent reasoning behavior and should be treated as a methodological choice rather than an implementation detail.

Paper Type: Long

Research Area: AI/LLM Agents

Research Area Keywords: multi-agent systems, agent communication, agent coordination and negotiation, agent evaluation

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Approaches low compute settings-efficiency

Languages Studied: English

Submission Number: 8700