Algorithmic Narrative Structures: How Large Language Models Mimic and Disrupt Classical Structuralist Literary Syntax

Abstract

With the advent of Large Language Models (LLMs) and their widespread and easy adoption, computational writing has become a central one in the fields of literary theory, computational linguistics, and digital humanities. However, the use of LLM-generated fiction in a narratological context still needs to be further theorized. This work explores the possibility of the classical structuralist narrative syntax being embedded in contemporary LLMs, and its temporary dis/simultaneous disruption. The research draws on the typology of Vladimir Propp's Morphology of the Folktale and Joseph Campbell's monomyth, to create a mixed-methods design using 500 AI-generated stories from four model families (GPT, Claude, Gemini, and Llama). Both quantitative and qualitative analyses, mapping Proppian functions and Hero's Journey stages, respectively, were conducted and quantified, with a qualitative analysis focused on anomalies in the structure such as semantic drift, repetitive loops, character irregularity, disjointed story lines and dislocated resolutions. The results indicate that the LLMs are very good at repeating the events of surface-level narrative elements: 87.6% of narratives included an initial lack or absentation, 84.2% included a call or mediation event, and 76.8% included a confrontation. Likewise, elements of Hero's Journey presented with high frequency like the ordinary world, the call to adventure, the threshold crossing and ordeal. But only 28.4% of all content was strict monomyth completion and 53.6% of narratives experienced the presence of one or more algorithmic interrupts. It is argued in this article that storytelling in LLM is not just a copy of some of the universal elements of narratives, nor an uninformed random variation on human-made forms. Rather, it results in a syntax of algorithmic narratives that is emergent, based on recognizable classical functions, but is susceptible to local (coherence) failure, recursive (re)combination, and premature closure. With this contribution, AI storytelling is declared a key space which should be extended beyond textual authoring and systems of human design.