Multilingual large language models (LLMs) are increasingly used for factual question answering, yet their accuracy varies across languages in ways that are difficult to interpret. A central challenge is that many multilingual probing benchmarks conflate multiple factors: the language used to ask the question, the cultural-linguistic context of the entities being queried, and the popularity skew of entities. In our paper, we disentangle these factors by asking: (i) how strongly does the Language of the Question (LoQ) affect factual recall, (ii) does matching LoQ to an entity-associated Language of the Entity (LoE) improve performance, and (iii) do these effects persist when entity popularity is controlled. To this end, we introduce WILA-PopQA, a new Wikidata-grounded benchmark spanning 9 languages with matched popularity profiles, and probe 12 open-weight models of varying sizes and architectures under aligned and misaligned LoQ–LoE conditions. We evaluate models’ answers to 4 types of questions about entity biographical properties in all selected languages. Results show that LoQ is the dominant source of variation. LoQ–LoE alignment does not consistently yield the highest accuracy, and performance depends on the property being asked. These results suggest that prompt language is an actionable experimental factor for multilingual factual evaluation.
Iferroudjene, M., Poggel, L., Schimmenti, A., Yang, D., Shivashankar, K., Kalo, J., et al. (2026). A Wikidata-Based Framework to Measure Cross-Lingual Bias in Multilingual Large Language Models [10.63317/29b354pejyrt].
A Wikidata-Based Framework to Measure Cross-Lingual Bias in Multilingual Large Language Models
Schimmenti, AndreaCo-primo
;
2026
Abstract
Multilingual large language models (LLMs) are increasingly used for factual question answering, yet their accuracy varies across languages in ways that are difficult to interpret. A central challenge is that many multilingual probing benchmarks conflate multiple factors: the language used to ask the question, the cultural-linguistic context of the entities being queried, and the popularity skew of entities. In our paper, we disentangle these factors by asking: (i) how strongly does the Language of the Question (LoQ) affect factual recall, (ii) does matching LoQ to an entity-associated Language of the Entity (LoE) improve performance, and (iii) do these effects persist when entity popularity is controlled. To this end, we introduce WILA-PopQA, a new Wikidata-grounded benchmark spanning 9 languages with matched popularity profiles, and probe 12 open-weight models of varying sizes and architectures under aligned and misaligned LoQ–LoE conditions. We evaluate models’ answers to 4 types of questions about entity biographical properties in all selected languages. Results show that LoQ is the dominant source of variation. LoQ–LoE alignment does not consistently yield the highest accuracy, and performance depends on the property being asked. These results suggest that prompt language is an actionable experimental factor for multilingual factual evaluation.| File | Dimensione | Formato | |
|---|---|---|---|
|
unpaywall-bitstream-340667589.pdf
accesso aperto
Tipo:
Versione (PDF) editoriale / Version Of Record
Licenza:
Creative commons
Dimensione
4.55 MB
Formato
Adobe PDF
|
4.55 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



