Publications / In Search of a Translator
In Search of a Translator: Using AI to Evaluate What’s Lost in Translation
Frontiers in Computer Science 6, published 13 August 2024.
Bibliographic record
- Author
- Katherine Elkins
- Title
- In Search of a Translator: Using AI to Evaluate What’s Lost in Translation
- Journal
- Frontiers in Computer Science
- Volume
- 6
- Published
- 13 August 2024
- DOI
- 10.3389/fcomp.2024.1444021
- Access
- Open access
- Last updated
- 14 July 2026
Summary
This article compares a literary original with multiple translations across whole-narrative time. It asks which dimensions of style, emotional arc, pacing, and reading experience persist, which are transformed or flattened, and what AI-assisted comparison can reveal that sentence-by-sentence evaluation misses.
The method combines diachronic stylometry, lexical richness, readability, and multilingual emotional-arc modeling, running from VADER through Mistral. It compares Proust’s Du côté de chez Swann with the Moncrieff, Enright, and Davis translations, treating each version as a trajectory over narrative time rather than as a sequence of sentences to be scored one by one.
The findings: translations flatten the original’s diachronic lexical richness and reproduce its emotional arc only incompletely; Davis comes closest on the measures examined; and larger context-sensitive models best capture the nuance of the French original. The larger point is methodological. Literary translation can be evaluated computationally at the scale of the whole narrative, as trajectories over narrative time, without replacing literary judgment.
What this work contributed
- Compared an original and its translations as whole-narrative trajectories rather than sentence by sentence.
- Showed that translation measurably flattens diachronic lexical richness and only partly reproduces the emotional arc.
- Established that AI-assisted comparison can surface features of a translation that close reading alone can miss, without displacing literary judgment.
- Provided a precedent for AI evaluation of literary translation at the scale of the book.
Relationship to earlier and later work
It applies the methodology of The Shapes of Stories (2022) to translation, and depends on the trajectory-comparison instruments built with Jon Chun, which make narratives of unequal length comparable.
Together with The Shapes of Cinderella (2025) it forms one research program on what persists and what changes when stories cross languages, cultures, and time. See Computational Narrative, Translation, and Cultural Transmission for the full lineage, and Scholarly Reception for the citation record.
How to cite
Chicago
Elkins, Katherine. “In Search of a Translator: Using AI to Evaluate What’s Lost in Translation.” Frontiers in Computer Science 6 (2024). https://doi.org/10.3389/fcomp.2024.1444021.
MLA
Elkins, Katherine. “In Search of a Translator: Using AI to Evaluate What’s Lost in Translation.” Frontiers in Computer Science, vol. 6, 2024, doi:10.3389/fcomp.2024.1444021.
APA
Elkins, K. (2024). In search of a translator: Using AI to evaluate what’s lost in translation. Frontiers in Computer Science, 6. https://doi.org/10.3389/fcomp.2024.1444021
BibTeX
@article{elkins2024translator,
author = {Elkins, Katherine},
title = {In Search of a Translator: Using AI to Evaluate What's Lost in Translation},
journal = {Frontiers in Computer Science},
volume = {6},
year = {2024},
doi = {10.3389/fcomp.2024.1444021}
}