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

TL;DR

This paper explores how language models can infer spatial layouts from unstructured narratives, enabling automated scene understanding for media applications.

Contribution

It introduces a novel evaluation suite and a training method combining rejection sampling and reinforcement learning for spatial reasoning from text.

Findings

Improved character attribution accuracy

Enhanced spatial plausibility in generated layouts

Better alignment with human preferences

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.