Elamur Research⌕
☰
home / papers / 2026 / Toying with Style: Can GenAI Mimic a Literary Translator's Voice?

Toying with Style: Can GenAI Mimic a Literary Translator's Voice?

Paper · 2026 · Beniamin Sopot and Dorothy Kenny

Summary

This pilot study asks whether general-purpose LLMs can be steered by prompting alone to translate in the personal style of a specific human literary translator, the Polish translator Jan Rybicki. The authors took a 1,000-word English novel extract, for which Rybicki had produced a translation that was never published, and had ChatGPT 5.4 and Claude Sonnet 4.6 render it into Polish. They used four prompts: a plain zero-shot request, a zero-shot request naming Rybicki's style, a one-shot prompt and a few-shot prompt. This gave eight machine outputs. Each output was compared with Rybicki's version using the stylo package, looking at sentence length, sentence length distribution, TTR and most frequent words, with Burrows's Delta and BCT clustering. No prompt moved either model measurably toward Rybicki's profile. The human text stayed an outlier, while each model clustered with its own outputs whatever the prompt. This suggests that prompting is a weak route to personalised, style-aware tools for computer-assisted literary translation (CALT).

Key points

  • Setup:
    • ChatGPT was accessed through Arena (gpt-5.4-mini-high) in incognito sessions.
    • Claude Sonnet 4.6 Adaptive was used on claude.ai with cross-chat memory switched off.
    • The few-shot prompt contained 3 of Rybicki's earlier translations and followed the prompt template of Tang et al. (2025).
  • Sentence length: Rybicki's mean was 15.38 words per sentence. The LLM outputs ranged from 17.87 (Claude, one-shot) to 19.68 (ChatGPT, few-shot).
  • Sentence count: The source text has 44 sentences.
    • Rybicki splits sentences more often, giving 58 sentences, against 47–50 in the AI versions.
    • He used 26 short sentences (≤10 words) against 19–22 for the models.
    • He used 8 long sentences (≥25 words) against 14–16 for the models.
  • Lexical diversity: TTR was the only measure where the outputs came close to Rybicki. His score was 0.63 and the models ranged from 0.627 to 0.672, with Claude consistently above ChatGPT.
  • Overall pattern: Differences between the outputs reflected each system's built-in behaviour more than the prompt used. This is in line with O'Sullivan (2025). The authors plan a qualitative follow-up examining features such as irony and other rhetorical devices.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 works

Cite

Cite as Beniamin Sopot and Dorothy Kenny. 2026. Toying with Style: Can GenAI Mimic a Literary Translator's Voice?. Proceedings of the First Workshop on Style in GenAI-Translated Content (StyGenAI).

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{sopot2026toying,
  title = {{Toying with Style: Can GenAI Mimic a Literary Translator's Voice?}},
  author = {Beniamin Sopot and Dorothy Kenny},
  booktitle = {Proceedings of the First Workshop on Style in GenAI-Translated Content (StyGenAI)},
  year = {2026},
  url = {https://aclanthology.org/2026.stygenai-1.6/}
}

References

24 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Chiang, Wei-Li, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Hao Zhang, Banghua Zhu, Michael Jordan, Joseph E. Gonzalez, and Ion Stoica. 2024. Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference. https://doi.org/10.48550/arXiv.2403.04132 ↗
  2. Eder, Maciej, Jan Rybicki, and Mike Kestemont. 2016. Stylometry with R: A Package for Computational Text Analysis, The R Journal, 8: 107–21. find
  3. Equity. 2026. Don’t Sign On Set: 8 Steps to protect yourself from AI exploitation. https://www.equity.org.uk/advice-and-support/know-your-rights/ai-toolkit/dont-sign-on-set-8-steps-to-protect-yourself-from-ai-exploitation (Accessed: 21 April 2026). ↗
  4. Hansen, Damien, and Emmanuelle Esperança-Rodier. 2022. Human-Adapted MT for Literary Texts: Reality or Fantasy? In Sheila Castilho, Rocío Caro Quintana, Maria Stasimioti, and Vilelmini Sosoni (eds) NeTTT 2022, 178–190, Rhodes, Greece. ⟨hal-04038025⟩ Hawkinson, Eric. 2025. The Em Dash as a Site of Co… — [doi:10.62883/filj2955](https://doi.org/10.62883/filj2955) find
  5. Herrmann, J. Berenike, Karina van Dalen-Oskam, and Christof Schöch. 2015. Revisiting Style, a Key Concept in Literary Studies. Journal of Literary Theory 9 (1): 25–52. find
  6. Karpinska, Marzena and Mohit Iyyer. 2023. Large language models effectively leverage document-level context for literary translation, but critical errors persist. https://doi.org/10.48550/arXiv.2304.03245 ↗
  7. Kenny, Dorothy. In preparation. Do it in Style! Human and automatic translation revisited. find
  8. Kenny, Dorothy. 2025. Literary machine translation: from taboo to controversy. In Stefan Baumgarten and Michael Tieber (eds) The Routledge Handbook of Translation Technology and Society. 381–395, Routledge, New York. 10.4324/9781003271314-33 — [doi:10.4324/9781003271314-33](https://doi.org/10.4324/9781003271314-33) find
  9. Kenny, Dorothy, and Marion Winters. 2024. Customisation, Personalisation and Style in Literary Machine Translation. In Translation, Interpreting and Technological Change: Innovations in Research, Practice and Training. Ursula Böser, Sharon Deane-Cox, and Marion Winters (eds). 59–79. Bloomsbury, Lond… — [doi:10.5040/9781350212978](https://doi.org/10.5040/9781350212978) find
  10. Lee, Changsoo. 2021. How do machine translators measure up to human literary translators in stylometric tests? Digital Scholarship in the Humanities 37(3): https://doi.org/10.1093/llc/fqab091 813–829. — [doi:10.1093/llc/fqab091](https://doi.org/10.1093/llc/fqab091) ↗
  11. O’Sullivan, James. 2025. Stylometric comparisons of human versus AI-generated creative writing. Humanities and Social Sciences Communications, 12(1), 1708. https://doi.org/10.1057/s41599-025-05986-3 — [doi:10.1057/s41599-025-05986-3](https://doi.org/10.1057/s41599-025-05986-3) ↗
  12. Przystalski, K., J. K Argasiński, I. Grabska-Gradzińska, and J. K Ochab. 2026. Stylometry recognizes human and LLM-generated texts in short samples. Expert Systems with Applications, 296, 129001. https://doi.org/10.1016/j.eswa.2025.129001 — [doi:10.1016/j.eswa.2025.129001](https://doi.org/10.1016/j.eswa.2025.129001) ↗
  13. Rybicki, Jan. 2006. Burrowing into Translation: Character Idiolects in Henryk Sienkiewicz’s Trilogy and Its Two English Translations. Literary and Linguistic Computing 21 (1):91–103. find
  14. Rybicki, Jan. 2012. The Great Mystery of the (Almost) Invisible Translator: Stylometry in Translation. In Quantitative Methods in Corpus-Based Translation Studies, edited by Michael Oakes and Meng Ji, 231–248. John Benjamins. find
  15. Rybicki, Jan. 2025. Can Machine Translation of Literary Texts Fool Stylometry? Digital Scholarship in the Humanities 40 (1): 268–276. find
  16. Rybicki, Jan, and Magda Heydel. 2013. The stylistics and stylometry of collaborative translation: Woolf’s Night and Day in Polish. Literary and Linguistic Computing, 28(4):708–717. https://doi.org/10.1093/llc/fqt027 — [doi:10.1093/llc/fqt027](https://doi.org/10.1093/llc/fqt027) ↗
  17. Similarweb. 2026. Top AI Chatbots and Tools Websites Ranking. Available at: https://www.similarweb.com/top-websites/ai-chatbots-and-tools/ (Accessed: 21 April 2026). ↗
  18. Tang, Minghua, Chen Bian, Liming Yang, and Xueling Zhong. 2025. Key-concept thinking prompting for improved reasoning in large language models. Neurocomputing, 656:130986, https://doi.org/10.1016/j.neucom.2025.130986. — [doi:10.1016/j.neucom.2025.130986](https://doi.org/10.1016/j.neucom.2025.130986) ↗
  19. Thai, Katherine, Marzena Karpinska, Kalpesh Krishna, William Ray, Moira. Inghilleri, John Wieting and Mohit Iyyer. 2022. Exploring Document-Level Literary Machine Translation with Parallel Paragraphs from World Literature, Proceedings of the 2022 Conference on Empirical Methods in Natural Language P… find
  20. Winters, Marion and Dorothy Kenny. 2024. Mark my keywords: a translator-specific exploration of style in literary machine translation. In Andrew Rothwell, Andy Way and Roy Youdale (eds) Computer-Assisted Literary Translation. 69—87, Routledge, New York. 10.4324/9781003357391-5 — [doi:10.4324/9781003357391-5](https://doi.org/10.4324/9781003357391-5) find
  21. Wolff, Rachel. 2026. What is the best LLM for translation? A comparison of top AI translation models. Lokalise Blog. Available at: https://lokalise.com/blog/what-is-the-best-llm-for-translation/ (Accessed: 21 April 2026). ↗
  22. Wu, Minghao, Jiahao Xu, Yulin Yuan, Gholamreza Haffari, Longyue Wan, Weihua Luo, Kaifu Zhang. 2025. (Perhaps) Beyond Human Translation: Harnessing Multi-Agent Collaboration for Translating Ultra-Long Literary Texts. Transactions of the Association for Computational Linguistics, vol. 13, pp. 901–922.…
  23. Yao, Xiaofang, Yong-Bin Kang and Anthony McCosker. 2025. Missing the human touch? A computational stylometric analysis of GPT-4 translations of online Chinese literature. Translation Spaces. https://doi-org.dcu.idm.oclc.org/10.1075/ts.24043.yao — [doi:10.1075/ts.24043.yao](https://doi.org/10.1075/ts.24043.yao) ↗
  24. Yirmibeşoğlu, Zeynep, Olgun Dursun, Harun Dallı, Mehmet Şahin, Ena Hodzik, Sabri Gürses, and Tunga Güngör. 2023. Incorporating human translator style into English-Turkish literary machine translation, arXiv preprint arXiv:2307.11457 *Fig. 1: 100 MFW Cluster Analysis* *Fig. 2: Bootstrap Consensus Tre… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sopot-2026-toying-style-can-genai-mimic/