Linguistic Markers of AI-Generated Texts in Uzbek: A Stylometric and Pragmatic Analysis of Human-Written and LLM-Generated Academic Discourse.

This article examines the linguistic markers of LLM-generatedUzbek academic texts from stylometric and pragmatic perspectives. The relevance ofthe study is determined by the rapid spread of AI-based text generation in scholarlycommunication, the growing importance of academic integrity and authorshipattribution, and the lack of research on this issue in Uzbek as a low-resource language.The aim of the study is to identify and describe the lexica-grammatical, syntactic andpragmatic features that distinguish human-written academic discourse from textsgenerated by large language models. The material comprises a comparative corpus ofhuman-authored and LLM-generated texts on identical topics. The analysis employsstylometric methods (lexical diversity, mean sentence and word length, word frequencydistribution), discourse-pragmatic analysis (hedging, metadiscourse markers, modality,expression of authorial stance) and statistical testing. The findings reveal distinctivelinguistic “fingerprints” of LLM-generated texts, including stylistic uniformity, lexicalrepetition, calqued syntactic structures and pragmatic neutrality. The scientific noveltylies in the systematic description of linguistic markers of AI-generated texts in Uzbek.The results can be applied to authorship attribution, academic integrity policies, thedevelopment of automatic detection tools for Uzbek, and philological education.