A Framework for Computational Narrative Generation

1. Introduction: Navigating Narratives with Mental Shortcuts

In complex problem-solving, particularly under uncertainty, human cognition often deviates from exhaustive logical deduction. Instead, it employs mental shortcuts, or heuristics, to arrive at timely and effective conclusions. 1 The concept of “bounded rationality,” introduced by Herbert Simon, posits that while individuals strive for rational choices, they are constrained by cognitive limitations and available information. 2, 3 This led to the “heuristics and biases” research program by Daniel Kahneman and Amos Tversky, which identified specific cognitive shortcuts people use to simplify the judgment process. 1, 4 While heuristics enable rapid decision-making, they can also introduce predictable biases, leading to conclusions that are sufficient rather than optimal. 5

In parallel, the field of narratology studies the fundamental components of narrative structure. 6 A narrative is composed of a story (the sequence of events) and a discourse (the telling of those events). 7 The pivotal moments of change that propel the story forward are often referred to as story beats or plot points. These events are the core building blocks of a compelling narrative, creating dramatic development and engaging the audience.

The “story beat heuristic model” emerges at the intersection of these fields, representing a computational approach to generating and managing narrative structures. This model describes how an Artificial Intelligence (AI) system can use heuristic principles—analogous to human cognitive shortcuts—to construct and sequence these critical narrative moments. The central challenge lies in balancing the efficiency of heuristics with the need for narrative coherence and impact. This requires designing AI systems that can be “fast and frugal” in generating plot points while ensuring the structural and artistic integrity of the resulting story. 2

2. Understanding Heuristics: The Art of “Good Enough” Decisions

Heuristics are indispensable mental tools for making decisions in complex or time-sensitive situations. They function as cognitive “rules of thumb” that permit individuals to make judgments quickly and with reduced mental effort, bypassing more demanding logical analysis. 1, 5 This idea was formalized in Herbert Simon’s theory of “bounded rationality,” which recognizes the inherent cognitive limits of human decision-makers. 3 The work of Daniel Kahneman and Amos Tversky further illuminated this area by identifying specific heuristics and the systematic biases they can produce. 4

The core justification for using heuristics is their efficiency in situations where time or information is limited. 2 This is particularly relevant in “large world” scenarios characterized by significant uncertainty. Key types of heuristics include:

• Anchoring and Adjustment: This involves relying on an initial piece of information (the “anchor”) and adjusting from that point to make a decision. The initial anchor, even if arbitrary, can disproportionately influence the final judgment. 1

• Availability Heuristic: This shortcut involves judging the likelihood of an event based on the ease with which instances come to mind. Events that are more easily recalled are perceived as more frequent or probable, which can distort judgment. 1

• Representativeness Heuristic: This involves making judgments by comparing a situation or individual to a pre-existing mental prototype or stereotype. This can lead to stereotyping and errors in judging probability. 1

While heuristics provide speed and efficiency, this comes at the cost of potential inaccuracies. 5 They can produce “good enough” outcomes but may overlook superior long-term solutions. 4 This trade-off between speed and precision is a foundational concept. The “fast and frugal” view, proposed by researchers like Gerd Gigerenzer, argues that in many real-world contexts, these simple heuristics can be surprisingly effective and “ecologically rational.” 2 This principle extends to AI systems, which also operate under constraints like computational resources and must adopt adaptive strategies for creative processes.

Table 1: Core Cognitive Heuristics and Their Characteristics

3. Deconstructing Narrative Events: The Rhythmic Pulse of Story

A compelling story is built upon a sequence of significant events that advance the plot. In narratology, the study of narrative structure, a fundamental distinction is made between the story (the chronological sequence of events, or fabula) and the discourse (the way those events are presented, or syuzhet). 7 The term “story beat” is a practical descriptor for a key narrative event—a moment of meaningful change that propels the story forward and serves as the smallest unit of plot progression. These beats are the pivotal points that shape dramatic development and sustain audience engagement.

Effective storytelling relies on the careful arrangement of these narrative events to create a cohesive and engrossing experience. By marking shifts in tone, action, or character emotion, beats transform a simple sequence of occurrences into a structured narrative. They can be categorized by their narrative function:

• Expository Events: Introduce characters, setting, and background information.

• Actions: A character performs a decisive act that moves the plot forward.

• Revelations: A character or the audience discovers critical information that alters the narrative direction.

• Transitions: Bridge scenes, time, or locations to maintain narrative flow.

• Escalations: Events that increase tension, conflict, or stakes.

• Reversals: Unexpected developments that create surprise and alter the narrative trajectory.

The modularity of these events provides a granular level of control for an AI shaping a narrative. This is particularly valuable for managing the complexity of story generation and is a core principle in interactive narratives, where user actions can trigger the next “beat” without derailing the overarching plot. 8 The chaining of these dramatic moments creates a “narrative grammar,” an implicit set of rules governing their sequence and interaction. For an AI, identifying and encoding these rules—for instance, that a revelation often follows an expository event and precedes a character’s reaction—is crucial for generating coherent and compelling narratives.

Table 2: Types of Narrative Events and Their Impact

4. The Story Beat Heuristic Model: Where AI Meets Storytelling

The “story beat heuristic model” is a specialized application of heuristic principles in AI systems designed for creating and managing sequences of narrative events. 10 Instead of conducting an exhaustive search of all possible narrative paths, these models use heuristics to focus on probable and effective sequences, greatly improving the efficiency of story generation. 11 This approach falls within the broader field of Computational Narratology, which integrates data-driven methods with narratological theories to model and generate narrative structures. 12

The design of these models is informed by theories from both narratology and cognitive science:

• Narratology: This discipline provides a formal understanding of narrative, which computational models can operationalize by translating abstract principles into functional algorithms. 13 For example, a model might implement theories of plot or character to guide story construction, ensuring the generated narrative is coherent and adheres to established conventions. 11

• Cognitive Accounts: Some models are based on theories of human creative writing. The Engagement-Reflection (E-R) model, for example, describes creative writing as a cycle of generating ideas (Engagement) and then evaluating and modifying them (Reflection). 11 This offers a framework for an AI to iteratively develop a story, applying heuristics during the “Reflection” phase to refine generated story beats.

Heuristic search is central to how these AI systems plan coherent story sequences. In narrative planning, heuristics are essential for navigating the vast space of possible story developments efficiently. 10 They guide the system toward actions that satisfy specific narrative criteria, such as ensuring characters act believably according to their goals and beliefs. 10 A common heuristic technique is “relaxed planning,” which involves solving a simplified version of the problem to estimate the distance to the goal, thereby guiding the search for the next optimal beat. 10

A significant advantage of this approach is explainability. Because these systems often rely on explicit rules derived from narrative theory, their decision-making process is more interpretable than that of “black-box” models. 10 This transparency is valuable for fostering effective human-AI collaboration in creative writing and for understanding the machine’s output. 14

5. AI Systems Utilizing Heuristics for Narrative Generation

Several AI systems leverage heuristic principles to construct stories, each with a different focus, from plot generation to stylistic imitation.

• MEXICA: This system is based on the “Engagement-Reflection” cognitive model of creative writing. It generates narratives as sequences of actions aiming for novelty and coherence, using a knowledge base of character conflicts and emotional links. Heuristics are applied during the “Reflection” phase to evaluate and refine the story, and to guide the plot if the system reaches an impasse. 11

• Curveship: This system computationally models narratological theories from scholars like Gérard Genette. Rather than modeling creativity itself, Curveship uses a set of parameters called “spin” to control stylistic elements like the narrator’s role, the chronological ordering of events, and the omission of events (ellipsis) to imitate various literary styles. 11

• Sabre: This narrative planning system models the intentions of both the author and the virtual characters. It uses heuristic search to generate plans that ensure every character’s action is explainable by that character’s goals and beliefs, thus prioritizing narrative believability. 10

It is critical to distinguish these traditional heuristic-based systems from Large Language Models (LLMs).

• Underlying Mechanism: Traditional systems like MEXICA operate as explicit models of storytelling, using structured knowledge of plot and narrative theory. 11 LLMs are primarily statistical predictors that learn to continue word sequences from vast training data, without an inherent, symbolic understanding of narrative structure. 10

• Control: The parameters in heuristic-based systems directly control specific narrative elements like story pace or narrator style. 11 LLM parameters, such as “temperature,” control the statistical properties of the generated text, offering less direct narrative control.

• Purpose and Reproducibility: Heuristic models are often developed as research tools for humanistic inquiry and tend to be open and reproducible. 11 Many powerful LLMs are proprietary, developed for commercial applications, and are frequently updated, making experimental results difficult to replicate. 11

An emerging trend involves using LLMs as heuristic guides within classical planning frameworks. In this hybrid approach, an LLM can suggest potential next actions, providing a heuristic estimate of the distance to a narrative goal. 10 This suggests a complementary future where rule-based heuristic models provide structural grounding and explainability, while LLMs offer linguistic fluency and broad associative knowledge.

Table 3: Comparison of Heuristic-Based AI Narrative Systems

6. Strengths and Limitations of Heuristic-Based Narrative AI

The use of heuristics in AI narrative generation presents a distinct profile of advantages and challenges.

Strengths:

• Efficiency: Heuristics simplify complex narrative problems, enabling rapid story generation, which is valuable when computational resources are limited. 2

• Structured Coherence: By relying on explicit rules from narrative theory, these systems generate stories with strong internal consistency and character believability. 10

• Explicit Control: Developers can exert precise control over narrative elements like style, pacing, and emotional arcs through adjustable parameters. 11

• Suitability for Interaction: Heuristic planning is well-suited for interactive narratives, where the system must adapt to user choices while maintaining a coherent plot. 8

• Explainability: The rule-based nature of many heuristic models allows for greater transparency, making it easier to understand why a narrative choice was made, which aids debugging and human-AI collaboration. 10, 14

Limitations:

• Sub-optimal Outcomes: Heuristics may produce “good enough” rather than optimal solutions, potentially overlooking more creative or nuanced narrative paths. 5

• Adaptability: Heuristics based on fixed rules may struggle to adapt to novel or dynamic narrative contexts compared to more flexible, data-driven models.

• Scalability: While more scalable than exhaustive search, classical planning approaches can still be computationally intensive for generating very long and complex narratives. 10

• Potential for Bias: If the underlying heuristics or the data they are derived from contain flaws or biases, the AI system risks perpetuating these biases in its generated stories.

• Coherence vs. Agency: In interactive narratives, there is a core challenge in balancing a coherent, pre-planned story with meaningful user agency, as user actions can introduce inconsistencies that disrupt the planned narrative. 8

The challenge of balancing coherence and agency highlights a deeper limitation: heuristic models for interactive narratives require sophisticated mechanisms to repair coherence or dynamically adapt their rules in response to unexpected player choices, adding significant design complexity. Furthermore, the risk of reinforcing systemic biases underscores a critical ethical consideration. The design of these systems demands not only technical proficiency but also a careful ethical evaluation of the values embedded within the heuristics themselves.

Table 4: Strengths and Limitations of Heuristic-Based AI in Narrative Generation

7. Conclusion: The Evolving Landscape of Heuristic-Driven Storytelling AI

The story beat heuristic model is a significant paradigm in computational creativity, leveraging the efficiency of cognitive shortcuts to generate and manage the core events of a narrative. This approach provides a pathway for creating structured, coherent, and interactive stories grounded in established narrative and cognitive theories.

The field is evolving toward hybrid approaches that integrate traditional heuristic planning with the capabilities of Large Language Models (LLMs). 10 In these models, LLMs can act as powerful “heuristic guides,” suggesting potential plot developments to a more structured planning system. 10 This suggests a future where these paradigms are not competitive but complementary. Heuristic models can provide the structural backbone and explainability for a narrative, while LLMs offer linguistic fluency and broad associative knowledge. The story beat heuristic model is thus poised to evolve into a meta-model that orchestrates different AI capabilities to achieve optimal results.

Future research will also focus on enhancing human-machine collaboration. Frameworks that enable human writers to actively critique and revise AI-generated content point toward a new creative paradigm. 14 In this model, AI is not an autonomous author but an intelligent co-creator. The heuristics become tools that empower human writers to direct, refine, and collaborate with the AI, making the creative process more efficient and innovative. This evolution promises increasingly sophisticated and engaging narratives forged through a synergistic relationship between human ingenuity and artificial intelligence.

Works Cited

1. Verywell Mind. (2024). Heuristics: Definition, Examples, and How They Work.

2. Gilovich, T., & Griffin, D. (2010). Judgmental Heuristics: A Historical Overview. In The Oxford Handbook of Social Cognition. Oxford University Press.

3. Böger, T., & Gutwald, R. (2020). A Brief History of Heuristics: How Did Research on Heuristics Evolve? HHL Working Paper No. 189.

4. Barberis, N. C. (2012). Psychologists at the Gate: A Review of Daniel Kahneman’s Thinking, Fast and Slow. Journal of Economic Literature, 50(4), 1039-1051.

5. Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases. Science, 185(4157), 1124-1131.

6. EBSCO Research Starters. Narratology (literary theory).

7. StudySmarter. Narratology: Definition, Principles & Theory.

8. Culler, J. (2001). The Pursuit of Signs: Semiotics, Literature, Deconstruction. Cornell University Press.

9. Peinado, F., & Gervás, P. (2023). Computational Models for Understanding Narrative. Bergen Language and Linguistics Studies, 13.

10. Riedl, M. O., & Bulitko, V. (2013). Interactive Narrative: An Intelligent Systems Approach. AI Magazine, 34(1), 67-77.

11. Senanayake, U., et al. (2025). Language Models as Narrative Planning Heuristics. Department of Computer Science, University of Kentucky.

12. Cambridge University Press. (2023). Call for Papers: Computational Narratology. Journal of Cultural Analytics.

13. Gervás, P. (2018). From Narratology to Computational Story Composition and Back. IOS Press Ebooks.

14. Rahman, A., et al. (2024). Collective Critics for Creative Story Generation. arXiv preprint arXiv:2410.02428.