Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LinguaGame: A Linguistically Grounded Game-Theoretic Paradigm for Multi-Agent Dialogue Generation

Yuxiao Ye, Yiming Zhang, Yiran Ma, Huiyuan Xie, Huining Zhu, Zhiyuan Liu

TL;DR

LinguaGame reframes multi-agent dialogue as a signalling game grounded in linguistic reasoning to improve communication efficiency without task-specific coupling. It introduces a training-free equilibrium approximation that runs at inference time to adjust utterance selection, enabling plug-and-play integration with existing LLM-based MAS. The framework, validated in simulated courtroom and debate scenarios with expert judgments, shows consistent gains in utterance coherence, conciseness, and argumentative quality driven by improved articulation rather than task optimization. The work highlights the potential of pragmatic reasoning over communicative goals as a general approach to enhancing multi-agent dialogue across domains, while acknowledging domain-specific limitations and responsible-AI considerations.

Abstract

Large Language Models (LLMs) have enabled Multi-Agent Systems (MASs) where agents interact through natural language to solve complex tasks or simulate multi-party dialogues. Recent work on LLM-based MASs has mainly focused on architecture design, such as role assignment and workflow orchestration. In contrast, this paper targets the interaction process itself, aiming to improve agents' communication efficiency by helping them convey their intended meaning more effectively through language. To this end, we propose LinguaGame, a linguistically-grounded game-theoretic paradigm for multi-agent dialogue generation. Our approach models dialogue as a signalling game over communicative intents and strategies, solved with a training-free equilibrium approximation algorithm for inference-time decision adjustment. Unlike prior game-theoretic MASs, whose game designs are often tightly coupled with task-specific objectives, our framework relies on linguistically informed reasoning with minimal task-specific coupling. Specifically, it treats dialogue as intentional and strategic communication, requiring agents to infer what others aim to achieve (intents) and how they pursue those goals (strategies). We evaluate our framework in simulated courtroom proceedings and debates, with human expert assessments showing significant gains in communication efficiency.

LinguaGame: A Linguistically Grounded Game-Theoretic Paradigm for Multi-Agent Dialogue Generation

TL;DR

LinguaGame reframes multi-agent dialogue as a signalling game grounded in linguistic reasoning to improve communication efficiency without task-specific coupling. It introduces a training-free equilibrium approximation that runs at inference time to adjust utterance selection, enabling plug-and-play integration with existing LLM-based MAS. The framework, validated in simulated courtroom and debate scenarios with expert judgments, shows consistent gains in utterance coherence, conciseness, and argumentative quality driven by improved articulation rather than task optimization. The work highlights the potential of pragmatic reasoning over communicative goals as a general approach to enhancing multi-agent dialogue across domains, while acknowledging domain-specific limitations and responsible-AI considerations.

Abstract

Large Language Models (LLMs) have enabled Multi-Agent Systems (MASs) where agents interact through natural language to solve complex tasks or simulate multi-party dialogues. Recent work on LLM-based MASs has mainly focused on architecture design, such as role assignment and workflow orchestration. In contrast, this paper targets the interaction process itself, aiming to improve agents' communication efficiency by helping them convey their intended meaning more effectively through language. To this end, we propose LinguaGame, a linguistically-grounded game-theoretic paradigm for multi-agent dialogue generation. Our approach models dialogue as a signalling game over communicative intents and strategies, solved with a training-free equilibrium approximation algorithm for inference-time decision adjustment. Unlike prior game-theoretic MASs, whose game designs are often tightly coupled with task-specific objectives, our framework relies on linguistically informed reasoning with minimal task-specific coupling. Specifically, it treats dialogue as intentional and strategic communication, requiring agents to infer what others aim to achieve (intents) and how they pursue those goals (strategies). We evaluate our framework in simulated courtroom proceedings and debates, with human expert assessments showing significant gains in communication efficiency.
Paper Structure (32 sections, 10 equations, 3 figures, 2 tables)

This paper contains 32 sections, 10 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM-based MAS
  4. Game-theoretic MAS
  5. MAS Behaviour Improvement
  6. Method
  7. MAS Structure
  8. MAS for Courtroom Proceedings
  9. MAS for Debates
  10. Game Formulation
  11. Signalling Game
  12. Intent-strategy as Private Information
  13. Game-MAS Integration
  14. Equilibrium Approximation
  15. Utility Functions
  16. ...and 17 more sections

Figures (3)

  • Figure 1: Integration of the signalling game into the MAS in LinguaGame. The sender (the agent in blue) generates candidate utterances, which are interpreted by the receiver (the agent in red). The utterance with the highest probability under the updated sender policy is appended to the dialogue context. All intermediate information produced during the gaming process (in the grey box) is discarded once the game concludes.
  • Figure 2: Distribution of intents and strategies for courtroom proceedings with LGMAS.
  • Figure 3: Distribution of intents and strategies for debates with LGMAS.