False Alarms: How Linguistic Features Of Non-Native Academic Writing Trigger False Positives In AI Detection Tools

Rania Za’rour

Assistant professor, Department of English Language and Literature, School of Foreign Languages, The University of Jordan r.zarour@ju.edu.jo

00962795798521

Abstract: The growing reliance on AI-generated content detectors in academic publishing has introduced new complexities for non-native English writers, particularly within the highly structured genre of academic papers. This study explores how common linguistic features among second-language (English) academic writers may evoke false positive results from AI detection tools.

Drawing on my experience as a non-native academic author, I systematically analyzed a personal corpus of four drafts from two types of research articles. Two of these manuscripts were highly quantitative statistical analyses, and the other two were linguistics papers including qualitative discussions alongside their academic arguments. All the papers received minimal grammatical checking, and AI sophisticated tools were not used in their production. After using several AI detection tools on these drafts, I noticed a distinct pattern regarding the instances of AI detection. First, every paper was flagged for containing AI-generated material. It seems that even minimal use of grammatical checks could evoke detection. Additionally, the statistical papers, which feature a strict structure and emphasize quantitative data, were flagged as AI-generated more often than the linguistics papers. The latter allowed for greater stylistic variation and qualitative nuance. Second language researchers often use formulaic expressions, formal tone, limited lexical diversity, and strict adherence to structural conventions to meet academic standards. These linguistic features probably trigger AI detectors as machine-generated. The outcomes indicate a bias in existing detection systems that may disadvantage non-native authors, emphasizing the necessity for more inclusive detection strategies that acknowledge non-native writing styles and the variety of academic Englishes.

Keywords: AI detection, second language writing, academic English, false positives, linguistic features, academic publishing​