Development of LLM-based system for hallucination spans detection in the multilingual outputs of LLMs

Karkani, Dimitra; Καρκάνη, Δήμητρα

dc.contributor.author	Karkani, Dimitra	en
dc.contributor.author	Καρκάνη, Δήμητρα	el
dc.date.accessioned	2026-02-12T08:27:30Z
dc.date.available	2026-02-12T08:27:30Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/63399
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.31094
dc.description	Εθνικό Μετσόβιο Πολυτεχνείο--Μεταπτυχιακή Εργασία. Διεπιστημονικό-Διατμηματικό Πρόγραμμα Μεταπτυχιακών Σπουδών (Δ.Π.Μ.Σ.) "Επιστήμη Δεδομένων και Μηχανική Μάθηση"	el
dc.rights	Default License
dc.subject	Hallucination	en
dc.subject	LLM	en
dc.subject	Detection	en
dc.subject	Multilingualism	en
dc.subject	Prompting	en
dc.title	Development of LLM-based system for hallucination spans detection in the multilingual outputs of LLMs	en
heal.type	masterThesis
heal.classification	Machine Learning	en
heal.language	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2025-06
heal.abstract	In recent years, Large Language Models are gaining increasing popularity due to their remarkable performance across different natural language processing tasks and domains.However, they often suffer from certain problems that limit their usage and raise questions concerning their reliability. One of this problems is the generation of Hallucinations, i.e. outputs that are semantically or factually inconsistent with their inputs or real-world knowledge. Therefore a lot of research is conducted around their causes and possible solutions. This thesis describes the theoretical analysis and development of an LLM-based system for hallucination span detection in multilingual outputs of LLMs. It arises from the shared task of the SemEVAL-2025 Mu-SHROOM that aims to motivate the study of the underexplored area of multilingual hallucination detection. An efficient, training-free LLM prompting strategy is proposed that enhances detection by translating multilingual text spans into English. The consistency of this approach highlights the effectiveness of our translation strategy for hallucination detection, demonstrating its applicability regardless of the source language.	en
heal.advisorName	Στάμου, Γιώργος	el
heal.committeeMemberName	Σταφυλοπάτης, Ανδρεας-Γεώργιος	el
heal.committeeMemberName	Βουλόδημος, Αθανάσιος	el
heal.committeeMemberName	Στάμου, Γιώργος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών	el
heal.academicPublisherID	ntua
heal.numberOfPages	109 σ.	el
heal.fullTextAvailability	false