Narrative trajectories, dynamic time warping, and computational comparison
Jon Chun builds computational instruments for comparing stories as trajectories. The lineage runs from ensemble sentiment analysis and dynamic time warping between unequal-length arcs, through sentence-level trajectory comparison and explainable AI, to multimodal coherence and cross-language comparison with Katherine Elkins.
The problem
Two novels rarely tell the same story at the same pace or the same length. One runs 1,400 sentiment measurements; another runs 13,000. Before you can ask whether two stories have the same shape, you need a way to put arcs of unequal length on comparable footing, and a distance you can actually compute between them. That is the problem this line of work solves.
The pipeline
- Ensemble sentiment analysis. Dozens of sentiment models are passed over the same narrative and weighed against one another. Smoothing and model selection are treated as interpretive choices, not neutral preprocessing.
- Standardization and smoothing. Arcs are standardized and smoothed with a 10% moving average.
- LTTB downsampling. Largest-Triangle-Three-Buckets reduces every arc to a common number of points while preserving its peaks, valleys, and endpoints.
- Dynamic time warping. DTW computes the distance between two whole arcs while absorbing the temporal shifts and stretches that separate otherwise similar story shapes. Two stories that share a shape but not a tempo are recognized as similar.
- Distance matrices and clustering. The resulting DTW distances drive hierarchical clustering, so arcs can be grouped by shape rather than inspected one at a time.
Scope of the claim
Dynamic time warping is a general time-series technique that long predates this work. The precise claim is narrower and checkable: SentimentArcs (2021) introduced dynamic time warping to diachronic sentiment analysis of narrative, pairing it with LTTB downsampling and using the resulting distance matrix to cluster arcs of unequal length. Dates carry the priority; the Reception page documents the uptake.
Sentence-level trajectories and comparability
The 2023 paper with Katherine Elkins brought explainable AI to narrative, a domain where XAI had been largely confined to image classifiers. It supplied a sentence-level story-trajectory methodology together with the agglomeration and comparability solutions that let narratives of unequal length be compared on shared coordinates. Publication page →
Across modalities
MultiSentimentArcs (2024) extends trajectory comparison from text to film, measuring whether the arc a film's dialogue traces and the arc its images trace are in fact the same arc. It makes cross-modal coherence a quantity that can be measured rather than asserted. Publication page →
Across languages and cultures
The same trajectory infrastructure supports the question of what persists when a story moves between languages, translations, and cultures. That interpretive program is Katherine Elkins's, and it is where this technical work is put to use. Narrative translation, cultural transmission, and computational philology →
Papers
- SentimentArcs (2021). The ensemble pipeline and DTW. Publication page · arXiv:2110.09454
- eXplainable AI with GPT4 (2023). Sentence-level trajectories and comparability. Publication page · doi:10.1007/s42803-023-00069-8
- MultiSentimentArcs (2024). Cross-modal coherence in film. Publication page · doi:10.3389/fcomp.2024.1444549
Code and reproducibility
The open-source implementation has been maintained since 2019 at github.com/jon-chun/sentiment_arcs. It is the technical foundation of The Shapes of Stories (Cambridge University Press, 2022).