← Research

Research / Narrative Translation and Cultural Transmission

Computational Narrative, Translation, and Cultural Transmission

What stays the same, and what changes, when a story moves across languages, translations, cultures, and media.

Katherine Elkins studies what remains recognizable and what changes when stories move across languages, translations, cultures, and media. With Jon Chun, she developed a program of computational narrative research that treats stories as measurable trajectories while returning every computational pattern to the passages, languages, and interpretive traditions that give it meaning.

The central question

A story that travels is changed by the journey. It is translated, retold, adapted, and rebuilt inside a different language and a different moral world, and yet it stays recognizable. The question this research asks is which structures survive that passage and which do not: which patterns persist, which are flattened, which are rebuilt, and which differences come from the language, the segmentation, the genre, the culture, or the analytical model itself.

Answering it requires two things at once. Stories have to be measurable, as trajectories that can be compared across works of unequal length and across languages that divide a text differently. And every measurement has to be returned to the passage that produced it, in the original language, where a human reader decides what it means.

A dated chronology

Each entry is dated to its first public record: an arXiv posting or a publication.

  1. 2019, Can Sentiment Analysis Reveal Structure in a “Plotless” Novel? Introduced middle reading and tested whether computational emotional-arc methods survive nonlinear narrative; joined quantitative trajectories to close reading. Elkins and Chun, arXiv 1910.01441.
  2. 2021, SentimentArcs. Established an ensemble pipeline for diachronic sentiment analysis and applied dynamic time warping to compare and cluster narrative arcs of unequal length. Chun, arXiv 2110.09454.
  3. 2022, The Shapes of Stories. Developed the methodology for narrative sentiment analysis and middle reading. Publication page →
  4. 2023, eXplainable AI with GPT-4 for Story Analysis and Generation. Developed sentence-level story trajectories, explainability, and comparability solutions for narratives represented on shared coordinates. Chun and Elkins, DOI.
  5. 2024, In Search of a Translator. Compared a literary original with multiple translations across whole-narrative time, asking what dimensions of style and emotional experience persist, flatten, or change. Publication page →
  6. 2024, MultiSentimentArcs. Extended narrative-trajectory comparison across modalities, measuring coherence between a film’s visual and dialogue arcs. Chun, technical record.
  7. 2025, Beyond Plot. Synthesized the methodological argument that emotional structure can reveal narrative organization not reducible to plot or character. Publication page →
  8. 2025, The Shapes of Cinderella. Compared original-language, clause-level trajectories across Chinese, French, and German variants and framed persistence and change as a problem of cultural transmission and computational philology. Publication page →

What the program established

Four methodological results carry across the publications. Narratives of unequal length can be compared as trajectories, because dynamic time warping measures distance between arcs while absorbing the shifts and stretches that make two tellings of the same story different lengths. Model selection, smoothing, and scale are interpretive choices rather than neutral preprocessing, and the story curve a researcher sees depends on them. Sentences are not a safe unit for cross-language comparison, because languages divide the same story into sentences differently; clauses are the defensible unit. And computational pattern-finding is only half a method: the other half is returning each peak and valley to the original-language passage and deciding what it means.

Translation and transmission in The Shapes of Stories

The Shapes of Stories is not only a history of Vonnegut, Reagan, and computational story shapes. It also treats translation and transmission, asking how narrative patterns persist, drift, or change as stories move across versions and cultural contexts. That question becomes the explicit subject of the later work, in Proust across three English translations and in Cinderella across Classical Chinese, French, and German.

The Shapes of Stories, publication page →

Key publications

Methods glossary

Emotional arc
The trajectory of sentiment across a narrative, read as a structure rather than a mood.
Narrative trajectory
A story represented as a time series, so that two stories can be placed on shared coordinates and compared.
Middle reading
Reading at more than one analytical distance: computational trajectories locate the structure, close reading of the passage decides what it means.
Dynamic time warping
A distance measure between two time series that tolerates shifts and stretches, which is what makes narratives of unequal length comparable.
Segmentation
The choice of unit. Languages divide a story into sentences differently, so sentence-level analysis can manufacture cross-language artifacts; clauses are the defensible unit.
Computational philology
Elkins’s term for a human and AI method that moves between computational modeling of narrative and granular, original-language interpretation, treating technical choices as interpretive decisions.

Related technical work

The trajectory-comparison instruments behind this research are built by Jon Chun, whose research pages document the ensemble sentiment pipeline, dynamic time warping between unequal-length arcs, sentence-level story trajectories, and the cross-modal extension in MultiSentimentArcs.

On this site, see also Sentiment Analysis, Narrative Intelligence, and Computational Philology, AI Creativity, Authorship, and Literary Translation, and the citation record in Scholarly Reception.