Language agents learn to mimic literary characters beyond dialogue
ACTOR: Action-Centered Temporal Organization for Literary Role-Playing Agents
Computation and Language
Summary
Stories often have characters that do more than just talk—they act and change over time. The authors created a system called ACTOR that reads novels and learns the actions and thoughts of characters in order, not just their conversations. This helps computers better understand and mimic how characters behave throughout a story. Their approach also teaches models to tell when a character's actions don't match their personality, improving the accuracy of story-based AI.
What this means in practice
- •For game developers: Create story-driven game characters that exhibit realistic, evolving behaviors beyond just dialogue lines.$Commercial implications: Enables development of immersive role-playing game characters with dynamic, personality-consistent actions that enhance player engagement.
- •For interactive fiction creators: Design AI characters that respond with consistent actions and thoughts over story time to increase narrative depth and believability.
Authors
Yuxiang Gao, Tianjie Ju, Zheng Wu, Zongru Wu, Hao Fei
Abstract
Large language agents for literary role-playing are typically modeled around dialogue generation, leaving characters' non-dialogue behaviors and their dynamic evolution over story time underrepresented. In this paper, we propose ACTOR (Action-Centered Temporal Organization for Literary Role-Playing Agents), an automated framework for constructing such behavior trajectories from raw novel texts. ACTOR extracts dialogue and persona-relevant non-dialogue behaviors, represents each behavior using a Thought-Action-Observation structure, and organizes behavior trajectories according to the internal temporal order of the story. We build a behavior dataset from 143 diverse literary works. For each source behavior, we generate counterfactual behaviors that are contextually plausible yet inconsistent with the character's persona. We find that existing models struggle to reliably infer character-consistent actions from historical behaviors and evolving character states. Using these behavior pairs, we train models with Regularized Preference Optimization (RPO), which augments DPO with a negative log-likelihood loss on the source behavior to jointly optimize behavior preference and generation. We further design a four-way next-behavior selection task to evaluate the model's ability to faithfully mimic character actions. Across 17,560 evaluation instances, the model trained on ACTOR's preference pairs achieves substantially improved accuracy. This result demonstrates that structured, temporally organized supervision enables models to leverage character profiles and prior behaviors to infer subsequent actions, greatly advancing literary role-playing beyond mere dialogue imitation. Code and data are available at https://github.com/SilentStar233/ACTOR.