[Paper Review] Multigenre AI-powered Story Composition

Edirlei Soares de Lima, Margot M. E. Neggers+1 more|arXiv (Cornell University)|May 6, 2024

Artificial Intelligence in Games4 citations

TL;DR

This paper introduces a multigenre AI-powered story composition framework that enforces thematic consistency by guiding narrative generation through five fundamental genres—comedy, romance, tragedy, satire, and mystery. Using a two-phase AI-driven process involving example retrieval and most specific generalization, the authors develop genre patterns that inform the PatternTeller prototype, enabling interactive, user-guided story creation with consistent thematic structure across diverse narrative types.

ABSTRACT

This paper shows how to construct genre patterns, whose purpose is to guide interactive story composition in a way that enforces thematic consistency. To start the discussion we argue, based on previous seminal works, for the existence of five fundamental genres, namely comedy, romance - in the sense of epic plots, flourishing since the twelfth century -, tragedy, satire, and mystery. To construct the patterns, a simple two-phase process is employed: first retrieving examples that match our genre characterizations, and then applying a form of most specific generalization to the groups of examples in order to find their commonalities. In both phases, AI agents are instrumental, with our PatternTeller prototype being called to operate the story composition process, offering the opportunity to generate stories from a given premise of the user, to be developed under the guidance of the chosen pattern and trying to accommodate the user's suggestions along the composition stages.

Motivation & Objective

• To establish a manageable set of five fundamental narrative genres that can guide thematic consistency in interactive story composition.
• To develop a systematic method for constructing genre patterns that encode structural and thematic elements of each genre.
• To implement an AI-powered prototype (PatternTeller) that uses these patterns to generate stories from user-provided premises while accommodating user input.
• To ensure narrative coherence and thematic fidelity in AI-generated stories by anchoring them in well-defined genre structures.
• To bridge classical narrative theory with modern LLM-based generation through a formalized, reusable pattern system.

Proposed method

• Employing a two-phase process: first retrieving story examples matching each genre’s characterization, then applying most specific generalization (MSG) to extract common structural and thematic features.
• Using AI agents to automate both example retrieval and MSG computation, ensuring scalability and consistency in pattern construction.
• Defining genre patterns based on synthesis of canonical narrative structures from literary theory (e.g., Frye’s genres and Ryan’s plot types) and validated through example analysis.
• Designing the PatternTeller prototype as an LLM-driven interface that accepts user premises and composes stories under the guidance of selected genre patterns.
• Integrating user suggestions iteratively during story composition to maintain interactivity and personalization while preserving genre integrity.
• Leveraging semiotic relations and narrative schemata to ensure that generated stories align with genre-specific expectations in plot, character, and tone.

Figure 1: The user interface of PatternTeller being used to extract the pattern for the mystery genre (Pattern 1 ) using the following stories as input: “Murder on the Orient Express” by Agatha Christie (1934), “The Da Vinci Code” by Dan Brown (2003), and “Sherlock Holmes” series by Sir Arthur Conan

Experimental results

Research questions

• RQ1 Can a minimal set of five fundamental genres—comedy, romance, tragedy, satire, and mystery—serve as a robust foundation for guiding thematic consistency in AI-powered story composition?
• RQ2 How can AI agents be effectively used to construct genre patterns through example retrieval and most specific generalization?
• RQ3 To what extent can genre patterns ensure thematic and structural coherence in interactive, user-guided story generation using LLMs?
• RQ4 How does the integration of user suggestions affect narrative coherence when guided by genre-specific patterns?
• RQ5 Can the proposed method generalize across diverse narrative types, including emerging genres like dystopian satire in video games?

Key findings

• The five selected genres—comedy, romance, tragedy, satire, and mystery—form a comprehensive and practically useful foundation for guiding thematic consistency in interactive story composition.
• The two-phase method of example retrieval followed by most specific generalization successfully identifies and formalizes recurring structural and thematic patterns across genre-specific story collections.
• The PatternTeller prototype demonstrates effective generation of thematically consistent stories from user-provided premises, with dynamic adaptation to user suggestions during composition.
• AI agents significantly enhance the scalability and reproducibility of genre pattern construction, enabling systematic reuse across narrative projects.
• The integration of genre patterns into LLM-based storytelling improves narrative coherence and reduces thematic drift, especially in complex or hybrid narrative forms.
• The approach supports emerging narrative trends such as dystopian satire, validating the relevance and adaptability of the selected genre framework.

Figure 2: The first events generated for the story “ Merlin’s Heiress: The Tragic Symphony of Eira’s Ascension ” in the storyboard style of PatternTeller.