The Fracture and Leap Cycle: Quantifying Narrative Surprise and Structural Resilience in Flash Fiction

This study investigates the structural and semantic dynamics of compressed narratives by analysing a filtered corpus of 2,636 Korean flash fiction stories. By leveraging sentence-level trajectories of surprisal, coherence, and semantic shift, the research models how information flow is managed under extreme length constraints. Clustering of normalised surprisal curves identifies five distinct structural archetypes, which are further categorised into two primary regimes: a “timing-driven” group focused on the strategic placement of climactic shocks, and an “attenuated” group defined by diffuse or bimodal information structures. An analysis of 3,889 significant surprisal events uncovers a robust “fracture and leap” cycle—where informational peaks induce localised coherence collapse while facilitating thematic reorientation—as a fundamental mechanic of narrative impact. The results demonstrate that the severity of this trade-off and the subsequent recovery dynamics vary systematically across these archetypes, distinguishing highly elastic “fast-reset” frameworks from those emphasising atmospheric persistence. These findings characterise flash fiction as a finely calibrated system of cognitive load management and provide a language-agnostic, transferable computational framework for formalist narrative analysis.