Codified Foreshadowing-Payoff Text Generation

TL;DR

CFPG reframes narrative coherence as explicit causal realization by encoding foreshadow–payoff relations as Foreshadow–Trigger–Payoff predicates and tracking them in a dynamic foreshadow pool $C_t$. The system uses a codified Select–Generate–Update loop to gate payoff realization, injecting explicit payoff constraints during generation and grounding progress via state updates. Built on BookSum, the authors automatically mine a large-scale Foreshadow–Payoff dataset that anchors setups and resolutions to sentence indices. Experiments show substantial improvements in payoff activation accuracy and narrative alignment over prompt-based baselines, and analyses quantify gains in causal saliency and online sensing, supporting the claim that explicit narrative mechanics enhance long-range coherence. These findings suggest a practical path toward moving LLM-based storytelling from fluent surface-level text toward true narrative competence.

Abstract

Foreshadowing and payoff are ubiquitous narrative devices through which authors introduce commitments early in a story and resolve them through concrete, observable outcomes. However, despite advances in story generation, large language models (LLMs) frequently fail to bridge these long-range narrative dependencies, often leaving "Chekhov's guns" unfired even when the necessary context is present. Existing evaluations largely overlook this structural failure, focusing on surface-level coherence rather than the logical fulfillment of narrative setups. In this paper, we introduce Codified Foreshadowing-Payoff Generation (CFPG), a novel framework that reframes narrative quality through the lens of payoff realization. Recognizing that LLMs struggle to intuitively grasp the "triggering mechanism" of a foreshadowed event, CFPG transforms narrative continuity into a set of executable causal predicates. By mining and encoding Foreshadow-Trigger-Payoff triples from the BookSum corpus, we provide structured supervision that ensures foreshadowed commitments are not only mentioned but also temporally and logically fulfilled. Experiments demonstrate that CFPG significantly outperforms standard prompting baselines in payoff accuracy and narrative alignment. Our findings suggest that explicitly codifying narrative mechanics is essential for moving LLMs from surface-level fluency to genuine narrative competence.

Figures (7)

• Figure 1: Illustration of Foreshadow–Trigger–Payoff decomposition using a narrative example from The Hound of the Baskervilles. The disappearance of the boot introduces an unresolved causal commitment, which remains dormant until a triggering narrative condition activates its resolution.
• Figure 2: Overview of the CFPG framework. CFPG maintains a codified causal state in the form of a foreshadow pool $C_t$, where each element is a structured $(F, T, P)$ triple. At each step t, an eligibility selection module deterministically selects a subset $S_t \subseteq C_t$ based on codified trigger constraints, which conditions the language model to generate the next continuation $y$. The generated text updates both the narrative prefix $X_{t+1}$ and the foreshadow pool $C_{t+1}$ via a codified state transition that resolves satisfied commitments and introduces new foreshadows. The right panel shows the simplified CFPG loop in pseudocode.
• Figure 3: A representative F-T-P triple extracted from the BookSum corpus. We anchor each element to specific narrative segments, ensuring that the latent causal link is verifiable and the payoff is temporally justified.
• Figure 4: Visualization of attention patterns during payoff generation. The heatmaps (left and center) compare the attention weights allocated to foreshadowing setup tokens under the vanilla prompting baseline and CFPG. The line plot (right) quantifies the resulting Causal Saliency Gain relative to the baseline, showing that CFPG consistently maintains significantly higher mean attention to the setup region throughout the generation process. Each row corresponds to a different narrative instance.
• Figure 5: Temporal decision dynamics of payoff detection for Qwen-2.5-7B-Instruct. CFPG shows reduced premature activation, a sharp decision transition at the gold payoff, and sustained post-resolution confidence compared to the baseline.