Deriving Character Logic from Storyline as Codified Decision Trees
This addresses the need for more reliable and consistent behavioral profiles in role-playing agents, representing a novel method for a known bottleneck in agent grounding.
The paper tackles the problem of brittle behavior in role-playing agents by proposing Codified Decision Trees (CDT), a framework that induces executable and interpretable decision structures from narrative data, resulting in substantial performance improvements over human-written profiles and prior methods across 85 characters and 16 artifacts.
Role-playing (RP) agents rely on behavioral profiles to act consistently across diverse narrative contexts, yet existing profiles are largely unstructured, non-executable, and weakly validated, leading to brittle agent behavior. We propose Codified Decision Trees (CDT), a data-driven framework that induces an executable and interpretable decision structure from large-scale narrative data. CDT represents behavioral profiles as a tree of conditional rules, where internal nodes correspond to validated scene conditions and leaves encode grounded behavioral statements, enabling deterministic retrieval of context-appropriate rules at execution time. The tree is learned by iteratively inducing candidate scene-action rules, validating them against data, and refining them through hierarchical specialization, yielding profiles that support transparent inspection and principled updates. Across multiple benchmarks, CDT substantially outperforms human-written profiles and prior profile induction methods on $85$ characters across $16$ artifacts, indicating that codified and validated behavioral representations lead to more reliable agent grounding.