Источник
ACL Workshop
Дата публикации
31.07.2025
Авторы
Елисей Рыков В. Олисов Максим Савкин Артем Важенцев Ксения Титова Александр Панченко Василий Коновалов Юлия Беликова
Поделиться

SmurfCat at SemEval- 2025 Task 3: Bridging External Knowledge and Model Uncertainty for Enhanced Hallucination Detection

Аннотация

The Multilingual shared-task on Hallucinationsand Related Observable Overgeneration Mistakesin the SemEval-2025 competition aimsto detect hallucination spans in the outputs ofinstruction-tuned LLMs in a multilingual context.In this paper, we address the detection ofspan hallucinations by applying an ensembleof approaches. In particular, we synthesizeda dataset and fine-tuned LLM to detect hallucinationspans. In addition, we combined thisapproach with a white-box method based on uncertaintyquantification techniques. Using ourcombined pipeline, we achieved 3rd place indetecting span hallucinations in Arabic, Catalan,Finnish, Italian, and ranked within the topten for the rest of the languages.

Присоединяйтесь к AIRI в соцсетях