AI evidence extraction

GigaChat rhetorical potential for transforming metadiscoursive patterns in Russian academic writing

Olga Boginskaya · Russian language studies · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.22363/2618-8163-2026-24-1-56-70

Methodology & findings

Study design

Corpus-based empirical study with computational analysis.

Main result

The study found that "GigaChat удалось передать свойственные современному русскоязычному научно-техническому дискурсу категоричность, убедительность и объективность" (GigaChat managed to convey the categorical character, persuasiveness and objectivity inherent in contemporary Russian-language scientific and technical discourse). Additionally, "ИИ повысил уровень оценочности, добавив экспрессивные элементы, которые подчеркивают значимость отдельных аспектов исследования" (AI increased the level of evaluativeness by adding expressive elements that emphasize the significance of individual aspects of research), and "общее количество метадискурсивных маркеров увеличилось с 7,5 до 10 на 1000 слов" (the total number of metadiscursive markers increased from 7.5 to 10 per 1000 words).

Research paradigm

Empirical-interpretative (corpus analysis with qualitative interpretation of AI system outputs)

Author conclusions

The authors conclude that "модель GigaChat, следуя разработанному промпту, значительно улучшила метадискурсивную организацию текста в соответствии с нормами академического письма, принятыми в технических науках. Тем не менее, начинающие авторы, использующие ИИ-технологии для написания научных текстов, должны критически оценивать ИИ-продукты" (the GigaChat model, following the developed prompt, significantly improved the metadiscursive organization of text in accordance with academic writing norms accepted in technical sciences. Nevertheless, beginning authors using AI technologies for writing scientific texts should critically evaluate AI products). They further state that "Оптимальное применение ИИ в обучении академическому письму предполагает его использование как вспомогательного инструмента в сочетании с постоянным контролем со стороны человека за соответствием текста конвенциям, принятым в той или иной дисциплине" (optimal application of AI in teaching academic writing assumes its use as a supplementary tool in combination with constant human monitoring of the text's compliance with conventions accepted in a particular discipline).

Risk of bias

Small sample size (40 abstracts from master's students); Selection bias: only abstracts of 150+ words from highly-cited Russian scientists (Hirsch index ≥10) were included in reference corpus; Discipline-specific bias: study limited to engineering texts, may not generalize to other fields; Single AI model tested (GigaChat); no comparison with other AI systems; Prompt design bias: structured prompt specifically crafted for technical discourse norms may overfit to engineering conventions; No inter-rater reliability reported for manual annotation of metadiscursive markers; Selection bias: Only abstracts of 150+ words were included in the reference corpus; only papers by single authors with Hirsch index ≥10 were selected; Confounding: The prompt design explicitly directed the AI toward specific metadiscursive patterns, which may not reflect the AI's default behavior; Limited sample diversity: Only 40 abstracts from graduate students in engineering; reference corpus limited to 6 specific Russian journals in engineering; Publication bias in reference corpus: Only published abstracts from established journals with high-impact authors were used as standard; Selection bias in corpus composition: reference corpus (K3) limited to single-author papers with author h-index ≥10, potentially excluding typical academic writing; Sample limitation: 40 abstracts from engineering students may not be representative of all disciplines or student populations; Potential prompt bias: the structured prompt used for GigaChat included specific examples that may have influenced the AI's output toward particular rhetorical patterns; Temporal limitation: reference corpus limited to 2020-2025 period; Language-specific bias: findings specific to Russian language; generalizability to other languages unknown

Limitations

The authors note that "начинающие авторы, использующие ИИ-технологии для написания научных текстов, должны критически оценивать ИИ-продукты, обращая внимание на принятый в той или иной области научного знания стиль изложения" (beginning authors using AI technologies for writing scientific texts must critically evaluate AI products, paying attention to the style adopted in particular areas of scientific knowledge)
Additionally, they state that "хотя ИИ и способен создавать связные тексты, он испытывает определенные трудности в передаче риторических аспектов научной аргументации" (although AI is capable of creating coherent texts, it experiences certain difficulties in conveying the rhetorical aspects of scientific argumentation)
Further, they recommend that "для формирования более репрезентативного эталона рекомендуется проведение последующих исследований с увеличением выборки и ее стратификацией как по журналам, так и по конкретным научным специальностям" (to create a more representative standard, it is recommended to conduct subsequent studies with increased sample size and stratification both by journals and by specific scientific specialties).

Open questions raised

Limited research on AI's influence on Russian-language academic metadiscourse (most prior studies focus on English)
Need for studies on adaptation of language models to disciplinary norms
Development of post-editing algorithms to help authors correct AI-generated texts according to metadiscursive conventions
Investigation of how metadiscursive practices vary by narrow scientific specialty
Expansion of research beyond engineering to other academic disciplines
Need for larger sample sizes and stratification by journals and specific scientific specialties

Data: Reference corpus of 100 abstracts from Russian engineering journals: Aviatsionnye materialy i tekhnologii, Fizicheskaya mezomekanika, Trudy VIAM, Vestnik mashinostroeniya, Dvigatelestroyeniye, Vestnik MAI (published 2020-2025, indexed in RSCI database); Reference corpus: 100 abstracts from Russian engineering journals (Авиационные материалы и технологии, Физическая мезомеханика, Труды ВИАМ, Вестник машиностроения, Двигателестроение, Вестник МАИ) published 2020-2025, indexed in RSCI database. Not explicitly made available; specific dataset URLs not provided.; Reference corpus: 100 abstracts from Russian engineering journals (Авиационные материалы и технологии, Физическая мезомеханика, Труды ВИАМ, Вестник машиностроения, Двигателестроение, Вестник МАИ), available through RSCI databaseExtracted from: pdfAgreement 65%

Explore related topics

AI Manuscript Writing Academic Writing Empirical Study