Elamur Research⌕
☰
home / papers / 2026 / Lexical Variation in English–Italian News Translation: A Comparative Study of Google Translate and ChatGPT

Lexical Variation in English–Italian News Translation: A Comparative Study of Google Translate and ChatGPT

Paper · 2026 · Aurora Trapella, Lieve Macken and Alessandra Molino

Summary

The paper asks whether a neural MT system and a general-purpose LLM differ in how lexically varied their Italian output is when translating English news. The English–Italian pair has received little attention in research on machine translationese (MTese), and Italian news translation still scores poorly in WMT evaluations. The authors built a source corpus of roughly 340,000 words of pre-AI British broadsheet journalism, drawn from the Dutch Parallel Corpus (DPC) and the EMA corpus. They then had Google Translate and ChatGPT-5.2 translate it via API. Concordance lines for three frequent cognitive verbs (know, think, believe) were annotated by the Italian equivalent chosen, and the results were compared across systems and across two genres: news reports and opinion articles. Both systems draw on largely the same solutions but distribute them differently. Google Translate uses more equivalents and more word-class shifts. ChatGPT concentrates on fewer lexical options and restructures more at phrase and clause level. Opinion pieces trigger more variation than reports in both systems. The work informs MTese research, translator and post-editor training, and AI literacy for journalists and lay users.

Key points

  • Source corpus: 339,049 words in total, split into news reports (182,622) and opinion articles (156,427).
    • DPC texts date from 2003–2008 and EMA texts from 2002–2004.
    • Outlets include The Guardian, The Independent, Financial Times, The Daily Telegraph and The Spectator.
  • Processing and translation setup:
    • Preprocessing used spaCy (en_core_web_lg for English, it_core_news_lg for Italian) for tokenization, lemmatization, POS tagging and NER.
    • Translations were generated in March 2026 by calling Google Translate's API at no cost and ChatGPT-5.2's API under a paid licence.
    • The scripts used a 2-second delay between requests, up to three retries, and paragraph-based chunking.
  • Verb-to-verb equivalents: Google Translate produced 16 distinct equivalents and GPT 14, in both genres.
    • Google Translate showed more nominal, adjectival and adverbial renderings.
    • GPT used nominalization only once overall but favoured reformulations such as a suo avviso and si direbbe.
  • Effect of genre and verb:
    • News reports leaned on a few reporting verbs (sapere, credere, ritenere, pensare), with high overlap between the two systems.
    • Opinion articles showed wider lexical and structural variety.
    • Think varied the most, while believe stayed stable.
  • Limitations and next steps:
    • The study is limited to three verbs, may be sensitive to prompt and API settings, and measures variation only quantitatively.
    • Planned work adds more NMT and LLM systems and a comparable corpus of original Italian news.

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 · 16,667 worksArtificial Intelligence · 16,194 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,263 works

Cite

Cite as Aurora Trapella, Lieve Macken and Alessandra Molino. 2026. Lexical Variation in English–Italian News Translation: A Comparative Study of Google Translate and ChatGPT. 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{trapella2026lexical,
  title = {{Lexical Variation in English–Italian News Translation: A Comparative Study of Google Translate and ChatGPT}},
  author = {Aurora Trapella, Lieve Macken and Alessandra Molino},
  booktitle = {Proceedings of the First Workshop on Style in GenAI-Translated Content (StyGenAI)},
  year = {2026},
  url = {https://aclanthology.org/2026.stygenai-1.7/}
}

References

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

  1. Biber, D., Johansson, S., Leech, G., Conrad, S., and Finegan, E. Longman Grammar of Spoken and Written English. find
  2. Bowker, L. (2024). Risks for lay users in machine translation and machine translation literacy. In The Social Impact of Automating Translation (pp. 60-76). find
  3. Routledge. Bullard, G. (2023). Smart ways journalists can exploit artificial intelligence. Nieman Reports. https://niemanreports.org/artificial-intelligence-newsrooms/ (niemanreports.org). Last accessed:2026-05-26. ↗
  4. Cagnan, P. (2025). Uso dell’intelligenza artificiale nelle redazioni italiane. Ordine dei Giornalisti. https://www.odg.it/uso-dellintelligenza-artificiale-nelle-redazioni-italiane/59863. Last accessed:2026-05-26. ↗
  5. Caswell, David (2019). Structured Journalism and the Semantic Units of News. Digital Journalism, 7(8), 1134-1156. find
  6. Cools, H., and Diakopoulos, N. (2024). Uses of Generative AI in the Newsroom: Mapping Journalists’ Perceptions of Perils and Possibilities. Journalism Practice, 1–19. https://doi.org/10.1080/17512786.2024.2394558 — [doi:10.1080/17512786.2024.2394558](https://doi.org/10.1080/17512786.2024.2394558) ↗
  7. Daems, J. (2022). Dutch literary translators' use and perceived usefulness of technology: The role of awareness and attitude. In Using technologies for creative-text translation (pp. 40-65). Routledge. doi: 10.4324/9781003094159-3 — [doi:10.4324/9781003094159-3](https://doi.org/10.4324/9781003094159-3) find
  8. Daems, J., De Clercq, O., and Macken, L. (2017). Translationese and post-editese: How comparable is comparable quality? Linguistica Antverpiensia, New Series: Themes in Translation Studies, 16, 89– 103. doi: 10.52034/lanstts.v16i0.434. — [doi:10.52034/lanstts.v16i0.434](https://doi.org/10.52034/lanstts.v16i0.434) find
  9. De Clercq, O., De Sutter, G., Loock, R., Cappelle, B., and Plevoets, K. (2021). Uncovering machine translationese using corpus analysis techniques to distinguish between original and machine-translated French. Translation Quarterly, 101, 21–45. ⟨hal-03406287⟩. find
  10. Espinosa-Giménez, M., and Forcada, M. L. (2024). Lexical impoverishment in neural machine translation: automatic and human study of Spanish-English noun translation. Tradumàtica tecnologies de la traducció, (22), 162-180. doi: 10.5565/rev/tradumatica.428. — [doi:10.5565/rev/tradumatica.428](https://doi.org/10.5565/rev/tradumatica.428) find
  11. Hermand, M. H. (2014). The Euroregional discourse. Increasing evidence towards legitimizing an institutional space. Mots. Les langages du politique, (106). doi: 10.4000/mots.21839. — [doi:10.4000/mots.21839](https://doi.org/10.4000/mots.21839) find
  12. Honnibal, M., Montani, I., Van Landeghem, S., and Boyd, A. (2020). spaCy: Industrial-strength Natural Language Processing in Python. https://doi.org/10.5281/zenodo.1212303 ↗
  13. Jiang, Y., and Niu, J. (2022). A corpus-based search for machine translationese in terms of discourse coherence. Across Languages and Cultures, 23(2), 148-166. doi: 10.1556/084.2022.00182. — [doi:10.1556/084.2022.00182](https://doi.org/10.1556/084.2022.00182) find
  14. Kocmi, T., Artemova, E., Avramidis, E., Bawden, R., Bojar, O., Dranch, K., Dvorkovich, A., Dukanov, S., Fishel, M., Freitag, M., Gowda, T., Grundkiewicz, R., Haddow, B., Karpinska, M., Koehn, P., Lakougna, H., Lundin, J., Monz, C., Murray, K., Nagata, M., Perrella, S., Proietti, L., Popel, M., Popov… — Findings of the WMT25 General Machine Translation Shared Task: Time to Stop Evaluating on Easy Test Sets
  15. Kong, D., and Macken, L. (2025a). Can Peter Pan Survive MT? A Stylometric Study of LLMs, NMTs, and HTs in Children’s Literature Translation. find
  16. Kong, D., and Macken, L. (2025b). Decoding Machine Translationese in English-Chinese News: LLMs vs. NMTs. In 20th Machine Translation Summit. find
  17. Loock, R. (2020). No more rage against the machine: How the corpus-based identification of machine-translationese can lead to student empowerment. Journal of Specialised Translation, (34), 150–170. ⟨halshs-02913980⟩. find
  18. Macken, L., De Clercq, O., Paulussen, H. (2011). “Dutch Parallel Corpus: A Balanced Copyright-Cleared Parallel Corpus.” META 56 (2): 374–90. find
  19. Molino, A. (2011). A Corpus-Based Study of Stance Adverbials in Learner Writing. find
  20. Roma: Aracne Editrice.A Murru, M. F., and Carlo, S. (2024). AI e newsmaking: Un’indagine esplorativa nelle redazioni nazionali e locali italiane. Mediascapes Journal, 23(1), 184– 198. https://rosa.uniroma1.it/rosa03/mediascapes/article/view/18714 Murtaza, G., and Qasmi, N. Q. (2013). Style and styl… — [doi:10.4324/9781003150640](https://doi.org/10.4324/9781003150640) ↗
  21. Niu, J., and Jiang, Y. (2024). Does simplification hold true for machine translations? A corpus-based analysis of lexical diversity in text varieties across genres. Humanities and Social Sciences Communications, 11(1), 1-10. doi: 10.1057/s41599-024-02986-7. — [doi:10.1057/s41599-024-02986-7](https://doi.org/10.1057/s41599-024-02986-7) find
  22. Roush, Chris. (2023). What Reuters is telling its journalists about using artificial intelligence. Talk-inbiz. https://talkingbiznews.com/?s=What+Reuters+is+telling+its+journalists+about+using+artificial+intelligence. Last accessed:2026-05-26. ↗
  23. The New York Times Trust Team. (2024). How The New York Times uses AI in journalism. The New York Times. https://www.nytimes.com/2024/10/07/reader-center/how-newyork-times-uses-ai-journalism.html. Last accessed:2026-05-26. ↗
  24. Vanmassenhove, E., Shterionov, D., and Gwilliam, M. (2021). Machine translationese: Effects of algorithmic bias on linguistic complexity in machine translation. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume (pp. 2203–2213)…
  25. Vanmassenhove, E., Shterionov, D., and Way, A. (2019). Lost in Translation: Loss and Decay of Linguistic Richness in Machine Translation. In Proceedings of Machine Translation Summit XVII Volume 1: Research Track (pp. 222-232). European Association for Machine Translation.
  26. Webster, R., Fonteyne, M., Tezcan, A., Macken, L., and Daems, J. (2020). Gutenberg Goes Neural: Comparing Features of Dutch Human Translations with Raw Neural Machine Translation Outputs in a Corpus of English Literary Classics. Informatics, 7(3), 32. doi:10.3390/informatics7030032. — [doi:10.3390/informatics7030032](https://doi.org/10.3390/informatics7030032) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/trapella-2026-lexical-variation-english-italian-news/