Text-to-Stage: Spatial Layouts from Long-form Narratives

2026-03-18

AI Summary

This study addresses the challenge of automatically generating stage play layouts—including scenes, character positions, movement trajectories, and room types—from long-form narrative texts that lack explicit spatial or relational cues. To this end, it introduces theatrical blocking theory into spatial reasoning with large language models for the first time, proposing a novel training framework that integrates rejection-based supervised fine-tuning, Best-of-N sampling, and GRPO reinforcement learning. A rule-based, verifiable reward system grounded in dramaturgical principles is developed to guide model optimization. Evaluated on canonical English literary corpora, the approach significantly improves character assignment accuracy, spatial plausibility, and action economy, outperforming existing baselines in both automatic metrics and human preference evaluations.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Natural Language Processing: Generation
• Humans and AI: Game Design — Procedural Content Generation & Storytelling

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

In this work, we probe the ability of a language model to demonstrate spatial reasoning from unstructured text, mimicking human capabilities and automating a process that benefits many downstream media applications. Concretely, we study the narrative-to-play task: inferring stage-play layouts (scenes, speaker positions, movements, and room types) from text that lacks explicit spatial, positional, or relational cues. We then introduce a dramaturgy-inspired deterministic evaluation suite and, finally, a training and inference recipe that combines rejection SFT using Best-of-N sampling with RL from verifiable rewards via GRPO. Experiments on a text-only corpus of classical English literature demonstrate improvements over vanilla models across multiple metrics (character attribution, spatial plausibility, and movement economy), as well as alignment with an LLM-as-a-judge and subjective human preferences.

Problem

• spatial reasoning
• narrative-to-play
• stage layout
• unstructured text
• text-to-stage

Innovation

• spatial reasoning
• narrative-to-play
• rejection SFT
• GRPO
• stage layout generation