Staying In Character: Perspective-Bounded Memory For Book-Based Role-Playing Agents

2026-06-24

AI Summary

Long-form narrative role-playing often suffers from factual boundary violations—where characters access information beyond their perspective—and monotonous linguistic style. This work proposes REVERIEMEM, a three-tier memory architecture that introduces, for the first time, a perspective-constrained mechanism to explicitly model first-person scene memories, semantic facts annotated with visibility labels, and context-dependent behavioral-linguistic patterns. By rigorously distinguishing between perceptible and imperceptible information from the character’s viewpoint, the framework enforces knowledge fidelity while dynamically preserving stylistic diversity. Implemented atop large language models, REVERIEMEM achieves a 34.6-percentage-point improvement in knowledge boundary adherence on the KBF-QA benchmark and attains a 79% win rate on the five-dimensional BOOKWORLD narrative evaluation, significantly outperforming existing approaches.

Abstract

Recent LLM role-playing systems build character agents from novels by extracting characters, scenes, and relations. Yet long-narrative role-playing suffers from two failures: Factual Overreach, where shared retrieval or parametric memory lets a character use facts outside its perspective, and Stylistic Monotony, where profile descriptions flatten a character into a fixed voice. To address these failures, we propose REVERIEMEM, a three-layer memory architecture for book-based character agents. The episodic layer stores first-person scene memories; the semantic layer stores visibility-tagged facts; and the personality layer stores situation-dependent speech and behaviour patterns. For evaluation, we construct KBF-QA, a 4,386-question benchmark over eight novels for testing knowledge boundaries. REVERIEMEM improves Knowledge Boundary Fidelity by 34.6 percentage points over the strongest prior method. On BOOKWORLD's five-dimension pairwise narrative protocol, REVERIEMEM achieves a ~ 79% win rate, suggesting that perspective-bounded memory improves both boundary fidelity and character-grounded narrative generation.

Problem

• Factual Overreach
• Stylistic Monotony
• role-playing agents
• perspective-bounded memory
• character consistency

Innovation

• perspective-bounded memory
• REVERIEMEM
• character role-playing agents
• knowledge boundary fidelity
• narrative generation