Breaking the Assistant Mold: Modeling Behavioral Variation in LLM Based Procedural Character Generation

2026-01-06

arXiv.org

AI Summary

This work addresses the prevalent limitations of current large language models in role generation—namely, their tendency toward monolithic moral stances and assistant-like response patterns, which constrain behavioral diversity and dramatic tension. To mitigate these issues, the authors propose PersonaWeaver, a novel framework that systematically identifies and alleviates both positive moral bias and assistant bias in character generation. By decoupling world-building attributes (e.g., identity, demographics) from behavior-shaping dimensions (e.g., moral stance, interaction style), PersonaWeaver enables controllable diversification across first-order behavioral traits and second-order linguistic features such as tone, punctuation, and utterance length. Experimental results demonstrate that this approach significantly enhances the diversity and realism of generated characters in terms of moral positioning, response strategies, and stylistic expression, thereby substantially improving their dramatic expressiveness in virtual settings.

Technology Category

• Humans and AI: Game Design — Virtual Humans, NPCs and Autonomous Characters
• Natural Language Processing: Generation
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Procedural content generation has enabled vast virtual worlds through levels, maps, and quests, but large-scale character generation remains underexplored. We identify two alignment-induced biases in existing methods: a positive moral bias, where characters uniformly adopt agreeable stances (e.g. always saying lying is bad), and a helpful assistant bias, where characters invariably answer questions directly (e.g. never refusing or deflecting). While such tendencies suit instruction-following systems, they suppress dramatic tension and yield predictable characters, stemming from maximum likelihood training and assistant fine-tuning. To address this, we introduce PersonaWeaver, a framework that disentangles world-building (roles, demographics) from behavioral-building (moral stances, interactional styles), yielding characters with more diverse reactions and moral stances, as well as second-order diversity in stylistic markers like length, tone, and punctuation. Code: https://github.com/mqraitem/Persona-Weaver

Problem

• procedural character generation
• alignment bias
• moral bias
• helpful assistant bias
• behavioral diversity

Innovation

• procedural character generation
• behavioral variation
• alignment bias
• PersonaWeaver
• large language models