Semantic degradation of scientific text in AI editing

Context and relevance. Generative language models are increasingly used for editing scientific texts. However, their use is associated with a specific type of distortion in which improvement of textual form is accompanied by a reduction in the accuracy of conveying the degree of justification of claims, the boundaries of generalization, and the author’s stance. Objective. To theoretically substantiate the concept of semantic degradation of scientific text under AI editing and to demonstrate how it manifests itself when a generative model processes a theoretically dense humanities fragment. Hypothesis. When editing complex scientific discourse, AI improves textual form while simultaneously shifting epistemic modality: reducing uncertainty, increasing categorical tone, smoothing conceptual openness, and weakening the author’s nuanced stance. Methods and materials. The study was conducted within a theoretical-analytical framework with elements of qualitative comparative analysis. The material consisted of a fragment from the chapter “Thought and Word” in L.S. Vygotsky’s Thinking and Speech. The original text and three AI-edited versions generated using DeepSeek were analyzed. The edited versions were compared with the original through qualitative comparative analysis. Results. The findings show that even under explicit instructions to preserve meaning, the language model systematically shifts the epistemic balance of the text. The distortion emerges through a redistribution of semantic emphasis rather than through factual errors. Conclusions. Semantic degradation occurs when stylistic improvement of a text is achieved at the cost of reducing its epistemic and conceptual precision. The risk of semantic degradation should be considered a systemic effect of the contemporary infrastructure of academic writing, in which generative AI amplifies the demand for rhetorically smooth and overly persuasive forms of knowledge.

1. Богинская, О.А. (2023). Лексико-синтаксические актуализаторы хеджирования в русском языке: опыт анализа отзывов о диссертации. Русистика, 21(1), 18–32. https://doi.org/10.22363/2618-8163-2023-21-1-18-32
   Boginskaya, O.A. (2023). Lexico-syntactic actualizers of hedging in the Russian language: An analysis of dissertation reviews. Rusistika, 21(1), 18–32. (In Russ.). https://doi.org/10.22363/2618-8163-2023-21-1-18-32
2. Викторова, Е.Ю. (2024). Хеджинг-стратегия в оценочном научном дискурсе: междисциплинарное исследование. Вестник Волгоградского государственного университета. Серия 2, Языкознание, 23(1), 5—17. https://doi.org/10.15688/jvolsu2024.1.1
   Viktorova, E.Yu. (2024). Transdisciplinary study of hedging strategy in evaluative academic discourse. Science Journal of Volgograd State University. Linguistics, 23(1), 5—17. (In Russ.). https://doi.org/10.15688/jvolsu2.2024.1.1
3. Горина, О.Г., Храброва, В.Е. (2017). Лингвистический хеджинг как коммуникативная структура (в русле корпусных исследований). Вестник Новосибирского государственного университета. Серия: Лингвистика и межкультурная коммуникация, 15(3), 44–53. https://doi.org/10.25205/1818-7935-2017-15-3-44-53
   Gorina, O.G., Khrabrova, V.E. (2017). Linguistic hedging as a communicative structure (within corpus-based research). Vestnik Novosibirskogo gosudarstvennogo universiteta. Seriya: Lingvistika i mezhkul'turnaya kommunikatsiya, 15(3), 44–53. (In Russ.). https://doi.org/10.25205/1818-7935-2017-15-3-44-53
4. Зашихина, И.М. (2024). Научные публикации и большие языковые модели: поймет ли нейросеть российскую науку? Научный редактор и издатель, 9(1 Suppl. 2), 2S31–2S https://doi.org/10.24069/SEP-24-11
   Zashikhina, I.M. (2024). Scientific publications and large language models: Will a neural network understand Russian science? Nauchnyi redaktor i izdatel', 9(1 Suppl. 2), 2S31–2S46. (In Russ.). https://doi.org/10.24069/SEP-24-11
5. Сахарова, А.В. (2020). Языковые средства выражения объективной эпистемической модальности в научном дискурсе. Научный диалог, (4), 151–163. https://doi.org/10.24224/2227-1295-2020-4-151-163
   Sakharova, A.V. (2020). Linguistic means of expressing objective epistemic modality in scientific discourse. Nauchnyi dialog, (4), 151–163. (In Russ.). https://doi.org/10.24224/2227-1295-2020-4-151-163
6. Amirjalili, F., Neysani, M., Nikbakht, A. (2024). Exploring the boundaries of authorship: A comparative analysis of AI-generated text and human academic writing in English literature. Frontiers in Education, 9, 1347421. https://doi.org/10.3389/feduc.2024.1347421
7. An, Y., Yu, J.H., James, S. (2025). Investigating the higher education institutions’ guidelines and policies regarding the use of generative AI in teaching, learning, research, and administration. International Journal of Educational Technology in Higher Education, 22, Article 10. https://doi.org/10.1186/s41239-025-00507-3
8. Bender, E.M., Gebru, T., McMillan-Major, A., Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21) (pp. 610–623). New York: ACM. https://doi.org/10.1145/3442188.3445922
9. Chelli, M., Shabani, M., Hsu, C., Bender, J. (2024). Accuracy of references generated by ChatGPT and Bard for systematic reviews. Journal of Medical Internet Research, 26, Article e53164. https://doi.org/10.2196/53164
10. Desaire, H., Chua, A.E., Isom, M., Jarosova, R., Hua, D. (2023). Distinguishing academic science writing from humans or ChatGPT with over 99% accuracy using off-the-shelf machine learning tools. Cell Reports Physical Science, 4(6), Article 101426. https://doi.org/10.1016/j.xcrp.2023.101426
11. Ganjavi, C., Eppler, M.B., Pekcan, A., Biedermann, B., Abreu, A., Collins, G.S., Gill, I.S., Cacciamani, G.E. (2024). Publishers’ and journals’ instructions to authors on use of generative artificial intelligence in academic and scientific publishing: Bibliometric analysis. BMJ, 384, Article e077192. https://doi.org/10.1136/bmj-2023-077192
12. Geng, M., Trotta, R. (2024). Is ChatGPT transforming academics’ writing style? In Proceedings of the 41st International Conference on Machine Learning (ICML), Workshop “The Next Generation of AI Safety”. OpenReview. https://openreview.net/pdf/723496dea6538f9253334c3e923b5b533bd667c6.pdf (viewed: 27.03.2026)
13. Hassan, A., Barber, S.J. (2021). The effects of repetition frequency on the illusory truth effect. Cognitive Research: Principles and Implications, 6, Article 38. https://doi.org/10.1186/s41235-021-00301-5
14. Hyland, K. (1998). Hedging in scientific research articles. Amsterdam: John Benjamins
15. Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y., Madotto, A., Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), Article 248. https://doi.org/10.1145/3571730
16. Lingard, L. (2020). The academic hedge: Modal tuning in research writing. Perspectives on Medical Education, 9(1), 5–8. https://doi.org/10.1007/s40037-019-00559-y
17. Maynez, J., Narayan, S., Bohnet, B., McDonald, R. (2020). On faithfulness and factuality in abstractive summarization. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020) (pp. 1906–1919). Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.173
18. Messeri, L., Crockett, M.J. (2024). Artificial intelligence and illusions of understanding in scientific research. Nature, 627, 49–58. https://doi.org/10.1038/s41586-024-07146-0
19. Millar, N., Batalo, B., Budgell, B. (2022). Trends in the use of promotional language (hype) in abstracts of successful National Institutes of Health grant applications, 1985–2020. JAMA Network Open, 5(8), Article e2228676. https://doi.org/10.1001/jamanetworkopen.2022.28676
20. Pagnoni, A., Balachandran, V., Tsvetkov, Y. (2021). Understanding factuality in abstractive summarization with FRANK: A benchmark for factuality metrics. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT 2021) (pp. 4812–4829). Association for Computational Linguistics. https://doi.org/10.18653/v1/2021.naacl-main.383
21. Peters, U., Chin-Yee, B. (2025). Generalization bias in large language model summarization of scientific research. Royal Society Open Science, 12, 241776. https://doi.org/10.1098/rsos.241776
22. Ryba, R., Doubleday, Z.A., Dry, M.J., Semmler, C., Connell, S.D. (2021). Better writing in scientific publications builds reader confidence and understanding. Frontiers in Psychology, 12, Article 714321. https://doi.org/10.3389/fpsyg.2021.714321
23. Sanz-Tejeda, A., Domínguez-Oller, J.C., Baldaquí-Escandell, J.M., Gómez-Díaz, R., García-Rodríguez, A. (2026). The impact of generative AI on academic reading and writing: A synthesis of recent evidence (2023–2025). Frontiers in Education, 10, 1711718. https://doi.org/10.3389/feduc.2025.1711718
24. Shanahan, M. (2022). Talking about large language models. arXiv preprint arXiv:2212.03551. https://doi.org/10.48550/arXiv.2212.03551
25. Stump, A., Voss, A., Rummel, J. (2024). The illusory certainty: Information repetition and impressions of truth enhance subjective confidence in validity judgments independently of the factual truth. Psychological Research, 88, 1288–1297. https://doi.org/10.1007/s00426-024-01956-7
26. Vinkers, C.H., Tijdink, J.K., Otte, W.M. (2015). Use of positive and negative words in scientific PubMed abstracts between 1974 and 2014: Retrospective analysis. BMJ, 351, h6467. https://doi.org/10.1136/bmj.h6467
27. Zhao, W. (2025). Reconstructing stance in EFL doctoral thesis writing through generative artificial intelligence. Humanities and Social Sciences Communications, 12, 1963. https://doi.org/10.1057/s41599-025-06249-x