Dynamic Coalition Structure Detection in Natural Language-based Interactions

TL;DR

This work tackles the challenge of predicting dynamic coalition structures in language-based, sequential multi-agent interactions, using Diplomacy as a testbed. It introduces a two-stage framework that first extracts candidate agreements from private dialogues and then evaluates their rationalizability via subjective hypergame beliefs, yielding edge weights for an evolving coalition graph $C_t=(N,E,\mathsf{Agmt},\mathsf{wt})$. Key contributions include a robust agreement-detection pipeline that fuses large language models with domain-specific intent models, and a hypergame-based rationalizability metric computed through order-sampling value functions to reflect both players' incentives and beliefs. Experiments on WebDiplomacy data show that the proposed approach outperforms Nash-equilibrium-based baselines in predicting which coalitions will be honored, providing a practical pathway to understanding coalition dynamics in realistic, language-driven negotiations. The method's generality suggests applicability to other multi-agent dialogue environments where incomplete information and evolving agreements govern strategic coordination.

Abstract

In strategic multi-agent sequential interactions, detecting dynamic coalition structures is crucial for understanding how self-interested agents coordinate to influence outcomes. However, natural-language-based interactions introduce unique challenges to coalition detection due to ambiguity over intents and difficulty in modeling players' subjective perspectives. We propose a new method that leverages recent advancements in large language models and game theory to predict dynamic multilateral coalition formation in Diplomacy, a strategic multi-agent game where agents negotiate coalitions using natural language. The method consists of two stages. The first stage extracts the set of agreements discussed by two agents in their private dialogue, by combining a parsing-based filtering function with a fine-tuned language model trained to predict player intents. In the second stage, we define a new metric using the concept of subjective rationalizability from hypergame theory to evaluate the expected value of an agreement for each player. We then compute this metric for each agreement identified in the first stage by assessing the strategic value of the agreement for both players and taking into account the subjective belief of one player that the second player would honor the agreement. We demonstrate that our method effectively detects potential coalition structures in online Diplomacy gameplay by assigning high values to agreements likely to be honored and low values to those likely to be violated. The proposed method provides foundational insights into coalition formation in multi-agent environments with language-based negotiation and offers key directions for future research on the analysis of complex natural language-based interactions between agents.

Figures (2)

• Figure 1: Proposed two-stage approach for learning coalition structures from natural language interactions in Diplomacy games. Stage 1 extracts agreements from pairwise dialogues to form an unweighted coalition structure. Stage 2 applies hypergame theory to assess the rationalizability of agreements for each player separately, which are then integrated into a weighted coalition structure representing the likelihood that an external observer believes agreements will be honored.
• Figure 2: An overview of our agreement detection framework. In this case, we are analyzing whether Italy and Austria have come to an agreement over Italy's unit F ION, and determine that an agreement has been reached for Italy to move this unit to the Eastern Mediterranean Sea (EAS).

Theorems & Definitions (3)

• Definition 1: Agreement
• Definition 2
• Definition 3: Subjective Rationalizability