Synthetic multilingual hallucination datasets and classifiers show higher hallucination rates for the 0.6B Qwen3 model (up to 60%) and for lower-resource languages like Icelandic compared with larger models.
InProceedings of the 60th Annual Meeting of the Association for Computa- tional Linguistics, pages 3214–3252
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
A multilingual hallucination benchmark: MultiWikiQHalluA
Synthetic multilingual hallucination datasets and classifiers show higher hallucination rates for the 0.6B Qwen3 model (up to 60%) and for lower-resource languages like Icelandic compared with larger models.