Branching Narratives: Character Decision Points Detection
Alexey Tikhonov
TL;DR
The paper defines the Character Decision Points Detection (CHADPOD) task to identify moments in narratives where characters make consequential choices. It introduces a CHADPOD dataset derived from CYOA game graphs, comprising 1,462 binary decision-point tasks with prefix/postfix text segments, and validates model performance using DeBERTa, BERT, ALBERT, and zero-shot LLMs, achieving up to 89% accuracy with strong baselines. A comparative analysis against Turning Points shows semantic differences between CHADPOD and Freytag-style pivots, underscoring the role of character agency. A Text Segmentation Study demonstrates practical use by segmenting classic literature around branching moments, while discussions outline future directions such as broader datasets and finer-grained classifications of decision points.
Abstract
This paper presents the Character Decision Points Detection (CHADPOD) task, a task of identification of points within narratives where characters make decisions that may significantly influence the story's direction. We propose a novel dataset based on CYOA-like games graphs to be used as a benchmark for such a task. We provide a comparative analysis of different models' performance on this task, including a couple of LLMs and several MLMs as baselines, achieving up to 89% accuracy. This underscores the complexity of narrative analysis, showing the challenges associated with understanding character-driven story dynamics. Additionally, we show how such a model can be applied to the existing text to produce linear segments divided by potential branching points, demonstrating the practical application of our findings in narrative analysis.