University of Primorska
Faculty of Mathematics, Natural Sciences and Information Technologies
The rapid development of large language models (LLMs) has created a growing need for reliable methods to distinguish AI-generated text from human-written text. Existing detection approaches include statistical methods, stylometric analysis, perplexity-based techniques, transformer-based classifiers, and large neural models trained specifically for AI-text detection.
Although many of these methods achieve high accuracy on benchmark datasets, their performance often decreases when applied to new domains, new language models, paraphrased texts, or modified content. Another important limitation is language bias. Most existing detectors are developed and evaluated mainly for English, while their performance is often much lower for other languages, especially for morphologically rich and low-resource languages.
This talk investigates a simple question: how do existing detectors deal with under-represented (and in the era of LLMs all non-English languages belong to this group) and morphologically rich languages.
By focusing on simple statistical properties of embedding representations, the proposed method offers a more language-independent alternative for multilingual AI-text detection, particularly for languages that are underrepresented in current research.
Presenter: prof. dr. Jernej Vičič
Jernej is a professor at the Faculty of Mathematics, Natural Sciences and Information Technologies of the University of Primorska. He completed his studies in Computer Science, where he developed research interests in language technologies, machine learning, distributed systems, and network analysis. His current work focuses on artificial intelligence, natural language processing, distributed computing, and blockchain-based systems. He collaborates on several national and international research projects related to AI, data analysis, and advanced digital technologies.
The presentation will take place as part of the Monday Computer Science Seminar
on Monday, 5.10.2026, starting at 16:00 in the lecture room FAMNIT-VP3.
