Survey Maps How Narrative Theory Shapes LLM Story Generation and Understanding

A new survey published in Artificial Intelligence Review examines how natural language processing research uses large language models to engage with concepts from narrative studies. The article, authored by David Yiyang Liu, Aditya Joshi and Paul Dawson, is open access and was shared early as a peer-reviewed, accepted version carrying a permanent DOI, subject to further edits before the final version of record.

The survey applies established distinctions from narratology to categorise ongoing efforts that connect narrative theories with large language model methods for automatic story generation and understanding. According to the authors, applications of narrative theories using these models deliver promising methods across both task families, and the review provides an overview of ongoing research efforts together with the broader narrative studies landscape.

Several findings emerge from the categorisation. The authors report that narrative texts used in this research come from diverse sources beyond just literature. They also identify theoretical synthesis and validation as potential outcomes of this line of work, suggesting that computational experiments can feed back into narrative scholarship rather than only drawing on it.

The survey finds an asymmetry between the two task families. Generation tasks lag behind understanding in several ways, including theoretical application, post-training methods, the exploration of non-fiction narratives, and the treatment of narrative levels beyond the classical fabula and discourse distinction. This gap indicates that producing stories with theoretically informed large language models remains less developed than analysing them.

For future directions, the authors argue against pursuing a single, generalised benchmark for narrative quality. Instead, they propose that progress can benefit from defining and improving theory-based metrics for individual narrative attributes, continuing large-scale theory-driven literary, social and cultural analysis, generating narratives in situated contexts, and continuing experiments in which model outputs can be used to validate or refine narrative theories.

The authors position the work as a contextual foundation for more systematic and theoretically informed narrative research in NLP. By mapping where computational narratology and large language model research currently intersect, the survey outlines both the concepts being borrowed from narrative studies and the areas where that engagement remains thin, offering researchers a structured view of an evolving interdisciplinary field.

Subject of Research: A survey of narrative theory-driven large language model methods for automatic story generation and understanding

Article Title: Narrative theory-driven LLM methods for automatic story generation and understanding: a survey

Article References: Liu, D. Y., Joshi, A., & Dawson, P. (2026). Narrative theory-driven LLM methods for automatic story generation and understanding: a survey. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11717-0

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11717-0

Keywords: large language models, narratology, story generation, story understanding, natural language processing, survey, narrative theory, computational narratology, theory-based metrics, Artificial Intelligence Review, Narrative, theory-driven