CSP4SDG: Constraint and Information-Theory Based Role Identification in Social Deduction Games with LLM-Enhanced Inference

2025-11-09

AI Summary

In social deduction games (e.g., Avalon, Mafia), players conceal their identities and actively deceive others, posing dual challenges for role inference: unobservable identities and untrustworthy utterances. To address this, we propose a real-time probabilistic reasoning framework that integrates constraint satisfaction (CSP) with information theory. It defines four language-agnostic constraint types—distinguishing hard logical constraints from weighted soft constraints—and employs an information-gain-driven entropy reduction mechanism to dynamically update hypothesis weights. A closed-form scoring rule ensures convergence of truthful statements to classical logic semantics. The framework requires no LLM fine-tuning and enables interpretable, low-latency inference. Evaluated on three public benchmarks, it significantly outperforms pure LLM baselines. When integrated as a lightweight reasoning augmentation module, it substantially improves large language models’ role identification accuracy—demonstrating the effectiveness and competitiveness of structured symbolic reasoning in social deduction tasks.

Technology Category

• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Knowledge Representation and Reasoning: Reasoning with Beliefs
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

In Social Deduction Games (SDGs) such as Avalon, Mafia, and Werewolf, players conceal their identities and deliberately mislead others, making hidden-role inference a central and demanding task. Accurate role identification, which forms the basis of an agent's belief state, is therefore the keystone for both human and AI performance. We introduce CSP4SDG, a probabilistic, constraint-satisfaction framework that analyses gameplay objectively. Game events and dialogue are mapped to four linguistically-agnostic constraint classes-evidence, phenomena, assertions, and hypotheses. Hard constraints prune impossible role assignments, while weighted soft constraints score the remainder; information-gain weighting links each hypothesis to its expected value under entropy reduction, and a simple closed-form scoring rule guarantees that truthful assertions converge to classical hard logic with minimum error. The resulting posterior over roles is fully interpretable and updates in real time. Experiments on three public datasets show that CSP4SDG (i) outperforms LLM-based baselines in every inference scenario, and (ii) boosts LLMs when supplied as an auxiliary"reasoning tool."Our study validates that principled probabilistic reasoning with information theory is a scalable alternative-or complement-to heavy-weight neural models for SDGs.

Problem

• Identifies hidden player roles in social deduction games using constraints
• Analyzes game events and dialogue through linguistically-agnostic constraint classes
• Provides interpretable probabilistic reasoning as alternative to neural models

Innovation

• Constraint-satisfaction framework analyzes gameplay objectively
• Information-gain weighting links hypotheses to entropy reduction
• Closed-form scoring ensures truthful assertions converge with minimal error