Elamur Research⌕
☰
home / papers / 2026 / Evaluating the Effect of Retrieval Augmentation on Social Biases

Evaluating the Effect of Retrieval Augmentation on Social Biases

Paper · 2026 · Tianhui Zhang, Yi Zhou and Danushka Bollegala

Summary

RAG is widely used to ground LLM outputs in external, up-to-date knowledge. It is less clear whether retrieved content changes the social biases a system expresses. The authors build a controlled evaluation of a standard, decomposable RAG pipeline that isolates the document collection, the retriever and the generator. Queries come from QA-style bias benchmarks: BBQ for English, JBBQ for Japanese and CBBQ for Chinese. Each has ambiguous and disambiguated contexts. The retrieval corpora are built from public, bias-annotated datasets with controlled stereotypical or anti-stereotypical polarity across gender, race, age and religion. Bias is measured with accuracy and a Diff-bias score adapted from Jin et al. (2024). The main finding is that stereotypes in retrieved documents are often strongly amplified in model answers, even for generators that show little bias without retrieval. This holds across all three languages and also affects Self-RAG, while the choice of retriever matters comparatively little. Because RAG is often seen as a reliability fix, these results argue for auditing corpora and full pipelines for bias before deployment.

Key points

  • The study covers 16 generator LLMs, 3 languages (English, Japanese, Chinese) and 4 bias categories (gender, race, age, religion), using BBQ, JBBQ and CBBQ.
  • Bias amplification from stereotype-laden retrieved documents appears consistently. Sparse retrieval is somewhat more bias-sensitive than dense retrievers. Bias does not rise steadily with the number of retrieved documents, because relevance drops as more are added.
  • The metrics are accuracy on ambiguous contexts (correct answer is always "UNKNOWN") and on disambiguated contexts, plus Diff-bias scores ranging from −1 to 1. Original BBQ metrics could not be used because the Chinese and Japanese benchmarks lack the required metadata.
  • DPO fine-tuning of Qwen2.5-7B-Inst., Llama3-8B-Inst. and Qwen-14B-Inst on GenderAlign (8000 preference-paired dialogues) lowered Diff-bias in every setting, most clearly on ambiguous questions. Prompt-based in-context debiasing worked less well once long, biased retrieved contexts were added.
  • Stated limitations: only QA was evaluated, only three languages were covered, and corpus coverage per bias type is uneven (e.g., only four datasets address age).

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

🏢AmazonSeattle · United States112 works · 40,565 citations🎓Cardiff UniversityCardiff · United Kingdom22 works · 2,444 citations🎓University of LiverpoolLiverpool · United Kingdom11 works · 546 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksSafety Research · 543 works

🧭 Topics 3Research subjects and languages the work is about.Evaluation & benchmarks · 1,220 worksInformation retrieval & search · 473 worksBias, fairness & safety · 517 works

Cite

Cite as Tianhui Zhang, Yi Zhou and Danushka Bollegala. 2026. Evaluating the Effect of Retrieval Augmentation on Social Biases. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-long.233

@inproceedings{zhang2026evaluating,
  title = {{Evaluating the Effect of Retrieval Augmentation on Social Biases}},
  author = {Tianhui Zhang, Yi Zhou and Danushka Bollegala},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-long.233},
  url = {https://doi.org/10.18653/v1/2026.eacl-long.233}
}

References

45 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. Self-rag: Learning to retrieve, generate, and critique through self-reflection.
  2. Soumya Barikeri, Anne Lauscher, Ivan Vulić, and Goran Glavaš. 2021. RedditBias: A real-world resource for bias evaluation and debiasing of conversational language models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conf… find
  3. Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna Wallach. 2020. Language (technology) is power: A critical survey of “bias” in nlp. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5454–5476.
  4. Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Proc. of NuerIPs, pages 4349–4357.
  5. Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, and Jonathan Larson. 2024. From local to global: A graph RAG approach to query-focused summarization. arXiv [cs.CL]. find
  6. Yunfan Gao, Yun Xiong, Meng Wang, and Haofen Wang. 2024. Modular rag: Transforming rag systems into lego-like reconfigurable frameworks. Preprint, arXiv:2407.21059. — [arXiv:2407.21059](https://arxiv.org/abs/2407.21059) find
  7. Rishav Hada, Agrima Seth, Harshita Diddee, and Kalika Bali. 2023. “fifty shades of bias”: Normative ratings of gender bias in GPT generated English text. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 1862– 1876, Singapore. Association for Computatio…
  8. Mengxuan Hu, Hongyi Wu, Zihan Guan, Ronghang Zhu, Dongliang Guo, Daiqing Qi, and Sheng Li. 2024. No free lunch: Retrieval-augmented generation undermines fairness in llms, even for vigilant users. Preprint, arXiv:2410.07589. find
  9. Yufei Huang and Deyi Xiong. 2024. CBBQ: A chinese bias benchmark dataset curated with human-AI collaboration for large language models. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 2917– 2929, Tori… find
  10. Hideki Isozaki and Tsutomu Hirao. 2003. Japanese zero pronoun resolution based on ranking rules and machine learning. In Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing, pages 184–191.
  11. Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. 2021. Unsupervised dense information retrieval with contrastive learning.
  12. Gautier Izacard and Edouard Grave. 2021. Leveraging passage retrieval with generative models for open domain question answering. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 874–880, Stroudsburg, PA, USA. Associati…
  13. Jiho Jin, Jiseon Kim, Nayeon Lee, Haneul Yoo, Alice Oh, and Hwaran Lee. 2024. KoBBQ: Korean bias benchmark for question answering. Transactions of the Association for Computational Linguistics, 12:507–524.
  14. Taeyoun Kim, Jacob Springer, Aditi Raghunathan, and Maarten Sap. 2025. Mitigating bias in rag: Controlling the embedder. arXiv preprint arXiv:2502.17390. find
  15. Satyapriya Krishna, Kalpesh Krishna, Anhad Mohananey, Steven Schwarcz, Adam Stambler, Shyam Upadhyay, and Manaal Faruqui. 2024. Fact, fetch, and reason: A unified evaluation of retrieval-augmented generation. arXiv [cs.CL]. find
  16. Shachi H Kumar, Saurav Sahay, Sahisnu Mazumder, Eda Okur, Ramesh Manuvinakurike, Nicole Beckage, Hsuan Su, Hung-yi Lee, and Lama Nachman. 2024. Decoding biases: Automated methods and llm judges for gender bias detection in language models. arXiv preprint arXiv:2408.03907. — [arXiv:2408.03907](https://arxiv.org/abs/2408.03907) find
  17. Philippe Laban, Alexander R Fabbri, Caiming Xiong, and Chien-Sheng Wu. 2024. Summary of a haystack: A challenge to long-context LLMs and RAG systems. arXiv [cs.CL].
  18. John Lalor, Yi Yang, Kendall Smith, Nicole Forsgren, and Ahmed Abbasi. 2022. Benchmarking intersectional biases in NLP. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3598–3609, Stroudsburg, PA,… find
  19. Patrick Lewis, Ethan Perez, Aleksandara Piktus, F Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Kuttler, M Lewis, Wen-Tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive NLP tasks. Adv. Neural Inf. Process. Syst., abs/2005.11…
  20. Jingling Li, Zeyu Tang, Xiaoyu Liu, Peter Spirtes, Kun Zhang, Liu Leqi, and Yang Liu. 2024a. Prompting fairness: Integrating causality to debias large language models. arXiv preprint arXiv:2403.08743. — [arXiv:2403.08743](https://arxiv.org/abs/2403.08743) find
  21. Jingling Li, Zeyu Tang, Xiaoyu Liu, Peter Spirtes, Kun Zhang, Liu Leqi, and Yang Liu. 2024b. Steering llms towards unbiased responses: A causality-guided debiasing framework. In ICLR 2024 Workshop on Reliable and Responsible Foundation Models. find
  22. Yingji Li, Mengnan Du, Rui Song, Xin Wang, Mingchen Sun, and Ying Wang. 2024c. Mitigating social biases of pre-trained language models via contrastive self-debiasing with double data augmentation. Artificial Intelligence, 332:104143. find
  23. Zichao Lin, Shuyan Guan, Wending Zhang, Huiyan Zhang, Yugang Li, and Huaping Zhang. 2024. Towards trustworthy llms: a review on debiasing and dehallucinating in large language models. Artificial Intelligence Review, 57(9):243. find
  24. Weicheng Ma, Brian Chiang, Tong Wu, Lili Wang, and Soroush Vosoughi. 2023. Intersectional stereotypes in large language models: Dataset and analysis. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 8589–8597, Stroudsburg, PA, USA. Association for Computational Linguis… find
  25. Moin Nadeem, Anna Bethke, and Siva Reddy. 2021. StereoSet: Measuring stereotypical bias in pretrained language models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long…
  26. Nikita Nangia, Clara Vania, Rasika Bhalerao, and Samuel R Bowman. 2020. CrowS-pairs: A challenge dataset for measuring social biases in masked language models. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1953–1967, Stroudsburg, PA, USA. As…
  27. Bo Ni, Zheyuan Liu, Leyao Wang, Yongjia Lei, Yuying Zhao, Xueqi Cheng, Qingkai Zeng, Luna Dong, Yinglong Xia, Krishnaram Kenthapadi, et al. 2025. Towards trustworthy retrieval augmented generation for large language models: A survey. CoRR. leaders: Gender stereotypes in machine translation and langu…
  28. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36:53728– 53741.
  29. Navid Rekabsaz and Markus Schedl. 2020. Do neural ranking models intensify gender bias? In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 2065–2068. find
  30. Mareike Riedel and Vanessa Rau. 2025. Religion and race: the need for an intersectional approach. Identities, pages 1–21. find
  31. Daisuke Oba, Masahiro Kaneko, and Danushka Bollegala. 2024. In-contextual gender bias suppression for large language models. In Findings of the Association for Computational Linguistics: EACL 2024, pages 1722–1742. find
  32. Jisu Shin, Hoyun Song, Huije Lee, Soyeong Jeong, and Jong C Park. 2024. Ask llms directly,“what shapes your bias?”: Measuring social bias in large language models. In Findings of the Association for Computational Linguistics ACL 2024, pages 16122–16143. find
  33. Alicia Parrish, Angelica Chen, Nikita Nangia, Vishakh Padmakumar, Jason Phang, Jana Thompson, Phu Mon Htut, and Samuel Bowman. 2022. BBQ: A hand-built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2086–2105, Dublin, Ireland. Asso…
  34. Sumit Soman and Sujoy Roychowdhury. 2024. Observations on building RAG systems for technical documents. In The Second Tiny Papers Track at ICLR 2024. find
  35. Ethan Perez, Sam Ringer, Kamile Lukosiute, Karina Nguyen, Edwin Chen, Scott Heiner, Craig Pettit, Catherine Olsson, Sandipan Kundu, Saurav Kadavath, Andy Jones, Anna Chen, Benjamin Mann, Brian Israel, Bryan Seethor, Cameron McKinnon, Christopher Olah, Da Yan, Daniela Amodei, Dario Amodei, Dawn Drain… — Discovering Language Model Behaviors with Model-Written Evaluations
  36. Matúš Pikuliak, Stefan Oresko, Andrea Hrckova, and Marian Simko. 2024. Women are beautiful, men are Yi Chern Tan and L. Elisa Celis. 2019. Assessing social and intersectional biases in contextualized word representations. In Advances in Neural Information Processing Systems, volume 32. Curran Associ… find
  37. Jinyang Wu, Feihu Che, Chuyuan Zhang, Jianhua Tao, Shuai Zhang, and Pengpeng Shao. 2024. Pandora’s box or aladdin’s lamp: A comprehensive analysis revealing the role of RAG noise in large language models. arXiv [cs.CL].
  38. Xuyang Wu, Shuowei Li, Hsin-Tai Wu, Zhiqiang Tao, and Yi Fang. 2025. Does RAG introduce unfairness in LLMs? evaluating fairness in retrieval-augmented generation systems. In Proceedings of the 31st International Conference on Computational Linguistics, pages 10021–10036, Abu Dhabi, UAE. Association… find
  39. Shi-Qi Yan, Jia-Chen Gu, Yun Zhu, and Zhen-Hua Ling. 2024. Corrective retrieval augmented generation. arXiv preprint arXiv:2401.15884.
  40. Hitomi Yanaka, Namgi Han, Ryoma Kumon, Jie Lu, Masashi Takeshita, Ryo Sekizawa, Taisei Kato, and Hiromi Arai. 2024. Analyzing social biases in japanese large language models. arXiv [cs.CL]. find
  41. Xiao Yang, Kai Sun, Hao Xin, Yushi Sun, Nikita Bhalla, Xiangsen Chen, Sajal Choudhary, Rongze Daniel Gui, Ziran Will Jiang, Ziyu Jiang, Lingkun Kong, Brian Moran, Jiaqi Wang, Yifan Ethan Xu, An Yan, Chenyu Yang, Eting Yuan, Hanwen Zha, Nan Tang, Lei Chen, Nicolas Scheffer, Yue Liu, Nirav Shah, Rakes… find
  42. Yue Yu, Wei Ping, Zihan Liu, Boxin Wang, Jiaxuan You, Chao Zhang, Mohammad Shoeybi, and Bryan Catanzaro. 2024. Rankrag: Unifying context ranking with retrieval-augmented generation in llms. Advances in Neural Information Processing Systems, 37:121156– 121184. find
  43. Tao Zhang, Ziqian Zeng, Yuxiang Xiao, Huiping Zhuang, Cen Chen, James Foulds, and Shimei Pan. 2024. Genderalign: An alignment dataset for mitigating gender bias in large language models. arXiv preprint arXiv:2406.13925. find
  44. Jiaxu Zhao, Meng Fang, Zijing Shi, Yitong Li, Ling Chen, and Mykola Pechenizkiy. 2023. CHBias: Bias evaluation and mitigation of Chinese conversational language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13538– 13… find
  45. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2018. Gender bias in coreference resolution: Evaluation and debiasing methods. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-evaluating-effect-retrieval-augmentation-social/