Semantic degradation of scientific text in AI editing
Language and Text · 30.06.2026 · 13 (2) · 213-225
DOI: 10.17759/langt.2026130216 ↗Kurz gefasst
Fachartikel in der Zeitschrift Language and Text, online erschienen am 30.06.2026. Verfasst von A.N. Semiletova. Die Arbeit nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
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.
Zitieren
APA
Semiletova, A. (2026). Semantic degradation of scientific text in AI editing. Language and Text, 13(2), 213-225. https://doi.org/10.17759/langt.2026130216
Vancouver
Semiletova A. Semantic degradation of scientific text in AI editing. Language and Text. 2026;13(2):213-225. doi:10.17759/langt.2026130216
BibTeX
@article{Semiletova2026,
author = {Semiletova, A.N.},
title = {Semantic degradation of scientific text in AI editing},
journal = {Language and Text},
year = {2026},
volume = {13},
number = {2},
pages = {213--225},
doi = {10.17759/langt.2026130216}
}
EUVIMED