Linguistic Nepotism: Trading-off Quality for Language Preference in Multilingual RAG

Dayeon Ki, Marine Carpuat, Paul Mcnamee, Daniel Khashabi, Eugene Yang, Dawn Lawrie, Kevin Duh
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:57072-57109, 2026.

Abstract

Multilingual Retrieval-Augmented Generation (mRAG) systems enable language models to answer knowledge-intensive queries with citation-supported responses across languages. Despite their growing use, an open questions is whether the mixture of different document languages impacts generation and citation behavior in unintended ways. To investigate this, we introduce a controlled methodology using model internals to measure language preference while holding other factors such as document relevance constant. Across eight languages and six open-weight models, we find that models preferentially cite English sources when queries are in English, with this bias amplified for lower-resource languages and for documents positioned mid-context. More crucially, we find that models sometimes trade-off document relevance for language preference, indicating that citation choices are not always driven by informativeness alone. Our findings shed light on how language models leverage multilingual context and influence citation behavior.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-ki26a, title = {Linguistic Nepotism: Trading-off Quality for Language Preference in Multilingual {RAG}}, author = {Ki, Dayeon and Carpuat, Marine and Mcnamee, Paul and Khashabi, Daniel and Yang, Eugene and Lawrie, Dawn and Duh, Kevin}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {57072--57109}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/ki26a/ki26a.pdf}, url = {https://proceedings.mlr.press/v306/ki26a.html}, abstract = {Multilingual Retrieval-Augmented Generation (mRAG) systems enable language models to answer knowledge-intensive queries with citation-supported responses across languages. Despite their growing use, an open questions is whether the mixture of different document languages impacts generation and citation behavior in unintended ways. To investigate this, we introduce a controlled methodology using model internals to measure language preference while holding other factors such as document relevance constant. Across eight languages and six open-weight models, we find that models preferentially cite English sources when queries are in English, with this bias amplified for lower-resource languages and for documents positioned mid-context. More crucially, we find that models sometimes trade-off document relevance for language preference, indicating that citation choices are not always driven by informativeness alone. Our findings shed light on how language models leverage multilingual context and influence citation behavior.} }
Endnote
%0 Conference Paper %T Linguistic Nepotism: Trading-off Quality for Language Preference in Multilingual RAG %A Dayeon Ki %A Marine Carpuat %A Paul Mcnamee %A Daniel Khashabi %A Eugene Yang %A Dawn Lawrie %A Kevin Duh %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-ki26a %I PMLR %P 57072--57109 %U https://proceedings.mlr.press/v306/ki26a.html %V 306 %X Multilingual Retrieval-Augmented Generation (mRAG) systems enable language models to answer knowledge-intensive queries with citation-supported responses across languages. Despite their growing use, an open questions is whether the mixture of different document languages impacts generation and citation behavior in unintended ways. To investigate this, we introduce a controlled methodology using model internals to measure language preference while holding other factors such as document relevance constant. Across eight languages and six open-weight models, we find that models preferentially cite English sources when queries are in English, with this bias amplified for lower-resource languages and for documents positioned mid-context. More crucially, we find that models sometimes trade-off document relevance for language preference, indicating that citation choices are not always driven by informativeness alone. Our findings shed light on how language models leverage multilingual context and influence citation behavior.
APA
Ki, D., Carpuat, M., Mcnamee, P., Khashabi, D., Yang, E., Lawrie, D. & Duh, K.. (2026). Linguistic Nepotism: Trading-off Quality for Language Preference in Multilingual RAG. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:57072-57109 Available from https://proceedings.mlr.press/v306/ki26a.html.

Related Material