Elamur Research⌕
☰
home / papers / 2026 / Enhancing Multilingual RAG Systems with Debiased Language Preference-Guided Query Fusion

Enhancing Multilingual RAG Systems with Debiased Language Preference-Guided Query Fusion

Paper · 2026 · Jeonghyun Park, Byeongjeong Kim, Seojin Hwang et al.

Summary

Multilingual RAG (mRAG) systems appear to favor English, so practitioners often translate non-English queries into English before retrieval ("English pivoting"). The authors argue that this apparent preference comes mostly from how benchmarks are built, not from the models themselves. Using an MKQA subset aligned with KILT NQ, they show three effects. Gold evidence sits mainly in English Wikipedia. English documents flood retrieval results. Culturally local questions often have answers only in local-language pages.

They propose DeLP, a calibrated metric. It uses ridge regression to predict observed preference from structural factors such as corpus size, gold availability and cultural prior, and treats the remaining residual as the model's genuine preference. Under DeLP, the English advantage largely disappears. What remains is a preference for the query and document to share a language (monolingual alignment).

Based on this, they build DELTA. It rewrites each query once, adding a global English paraphrase plus bridge cues (titles, aliases, locale). DELTA changes neither the corpus nor the retriever, and it beats English pivoting and other mRAG baselines. This matters for fair evaluation and for serving non-English users.

Key points

  • Setup: 2.7K MKQA queries overlapping KILT NQ, which supplies gold Wikipedia passage IDs. Retrieval and reranking use BGE-m3 (top-50 candidates, top-5 passed as context). Generators are Qwen3-235B, DeepSeek-v3.1 and Gemini-2.5-Flash. The knowledge source is English Wikipedia plus the query-language Wikipedia.
  • Gold availability: about 73% of gold evidence is located in English Wikipedia, versus roughly 0.5–1.4% for each other language (ar, de, ja, ko, th, zh).
  • Cultural prior: GPT-4o-mini tagged each query's main locale language. For several of these languages, around 20% of locale-tied queries have gold pages only in the local Wikipedia.
  • Cue ablation (fixed evidence, generation accuracy): the average rises from 63.13 with the original query to 71.62 with the global cue, and to 72.89 with all cues combined.
  • Retrieval gains and cost: DELTA recovers 1,235 queries whose gold pages had been missed, with a median best rank of 5 and a 66.2% Top-10 rate. Because it generates a single fused query and translates no documents, its latency can be lower than that of English translation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Chung-Ang UniversitySeoul · South Korea3 works · 12 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksInformation Systems · 1,437 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as Jeonghyun Park, Byeongjeong Kim, Seojin Hwang and Hwanhee Lee. 2026. Enhancing Multilingual RAG Systems with Debiased Language Preference-Guided Query Fusion. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1353

@inproceedings{park2026enhancing,
  title = {{Enhancing Multilingual RAG Systems with Debiased Language Preference-Guided Query Fusion}},
  author = {Jeonghyun Park, Byeongjeong Kim, Seojin Hwang and Hwanhee Lee},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1353},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1353}
}

References

21 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akari Asai, Jungo Kasai, Jonathan H Clark, Kenton Lee, Eunsol Choi, and Hannaneh Hajishirzi. 2021. Xor qa: Cross-lingual open-retrieval question answering. In Proceedings of the 2021 conference of the North American chapter of the association for computational linguistics: human language technologie…
  2. Nadezhda Chirkova, David Rau, Hervé Déjean, Thibault Formal, Stéphane Clinchant, and Vassilina Nikoulina. 2024. Retrieval-augmented generation in multilingual settings. In Proceedings of the 1st Workshop on Towards Knowledgeable Language Models (KnowLLM 2024), pages 177–188, Bangkok, Thailand. Assoc… find
  3. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  4. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  5. Dayeon Ki, Marine Carpuat, Paul McNamee, Daniel Khashabi, Eugene Yang, Dawn Lawrie, and Kevin Duh. 2025. Linguistic nepotism: Trading-off quality for language preference in multilingual rag. arXiv preprint arXiv:2509.13930. 27124 9 find
  6. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Proceedings of the 34th Internation…
  7. Bo Li, Zhenghua Xu, and Rui Xie. 2025. Language drift in multilingual retrieval-augmented generation: Characterization and decoding-time mitigation. arXiv preprint arXiv:2511.09984. find
  8. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  9. Shayne Longpre, Yi Lu, and Joachim Daiber. 2021. MKQA: A linguistically diverse benchmark for multilingual open domain question answering. Transactions of the Association for Computational Linguistics, 9:1389–1406.
  10. Hoyeon Moon, Byeolhee Kim, and Nikhil Verma. 2025. Quality-aware translation tagging in multilingual rag system. In Proceedings of the 5th Workshop on Multilingual Representation Learning (MRL 2025), pages 161–177. find
  11. Jeonghyun Park and Hwanhee Lee. 2025. Investigating language preference of multilingual RAG systems. In Findings of the Association for Computational Linguistics: ACL 2025, pages 5647–5675, Vienna, Austria. Association for Computational Linguistics. find
  12. Bhavani Shankar, Preethi Jyothi, and Pushpak Bhattacharyya. 2024. In-context mixing (ICM): Code-mixed prompts for multilingual LLMs. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4162–4176, Bangkok, Thailand. Association for… find
  13. Nikhil Sharma, Kenton Murray, and Ziang Xiao. 2024. Faux polyglot: A study on information disparity in multilingual large language models. Preprint, arXiv:2407.05502. find
  14. Liang Wang, Nan Yang, and Furu Wei. 2023. Query2doc: Query expansion with large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 9414–9423.
  15. Suhang Wu, Jialong Tang, Baosong Yang, Ante Wang, Kaidi Jia, Jiawei Yu, Junfeng Yao, and Jinsong Su. 2024. Not all languages are equal: Insights into multilingual retrieval-augmented generation. Preprint, arXiv:2410.21970. — [arXiv:2410.21970](https://arxiv.org/abs/2410.21970) find
  16. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  17. Xiang Zhang, Senyu Li, Bradley Hauer, Ning Shi, and Grzegorz Kondrak. 2023. Don’t trust chatgpt when your question is not in english: a study of multilingual abilities and types of llms. arXiv preprint arXiv:2305.16339. find
  18. Jirui Qi, Raquel Fernández, and Arianna Bisazza. 2025. On the consistency of multilingual context utilization in retrieval-augmented generation. In Proceedings of the 5th Workshop on Multilingual Representation Learning (MRL 2025), pages 199–225, Suzhuo, China. Association for Computational Linguist… find
  19. Leonardo Ranaldi, Barry Haddow, and Alexandra Birch. 2025. Multilingual retrieval-augmented generation for knowledge-intensive task. arXiv preprint arXiv:2504.03616.
  20. David Rau, Hervé Déjean, Nadezhda Chirkova, Thibault Formal, Shuai Wang, Stéphane Clinchant, and Vassilina Nikoulina. 2024. Bergen: A benchmarking library for retrieval-augmented generation. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 7640–7663. find
  21. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/park-2026-enhancing-multilingual-rag-systems-debiased/