Elamur Research⌕
☰
home / papers / 2026 / Uncovering Currency Bias and Syntax Gap in Text Embedding Models

Uncovering Currency Bias and Syntax Gap in Text Embedding Models

Paper · 2026 · Saurav Sudevan, Harsh Pal, Yatin Katyal et al.

Summary

Bias research on text-embedding models has mostly covered gender, race and religion. How these models represent money-related tokens has received little attention. The authors argue that currency identifiers are not neutral, because they carry geographic and economic connotations that can leak into financial AI. They test 14 open-source embedding models along three axes:

  • Syntax Gap. A bidirectional retrieval probe, scored with MRR in both clean and distractor-heavy settings, checks whether a currency's full name, acronym and symbol end up close together.
  • Associative bias. A rank-based score called Difference of Median Ranks (DMR) checks whether currencies sit nearer to poverty or malice anchors than to wealth or lawfulness anchors.
  • Downstream effects. Counterfactual currency swaps in sentences measure output changes in sentiment analysis, STS and credit default prediction.

Notation alignment turns out to be uneven and model-dependent. Some currencies acquire negative "personas," and simply changing the currency can flip predictions. This matters for automated credit scoring and compliance in emerging markets.

Key points

  • Retrieval queries put the word "Currency" before acronyms and symbols to avoid polysemy (e.g., ALL). Ambiguous symbols such as ₩ were excluded.
  • The sentiment robustness test uses 248 test sentences with one currency mention each. Every mention is replaced by each dictionary currency as a name, an acronym and a symbol. Instability is measured with prediction entropy.
  • A baseline mitigation augments training data by rewriting "$" as "US$", "USD" or "United States Dollar." For multilingual-e5-large, this lowered the credit-default flip rate from 2.53% to 1.59% (about 37% relative), with little change in AUCPR.
  • Models examined include gte-multilingual-base, LaBSE, bge-m3, multilingual-e5-large, e5-large-instruct, Qwen3-Embedding-0.6B and snowflake-arctic-embed-m-v1.5.
  • The authors report that the effects are not mainly caused by tokenization or model capability. Proprietary models were not evaluated.

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

 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 1Research subjects and languages the work is about.Syntax & grammar formalisms · 214 works

Cite

Cite as Saurav Sudevan, Harsh Pal, Yatin Katyal and Sahil Manchanda. 2026. Uncovering Currency Bias and Syntax Gap in Text Embedding Models. 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.2118

@inproceedings{sudevan2026uncovering,
  title = {{Uncovering Currency Bias and Syntax Gap in Text Embedding Models}},
  author = {Saurav Sudevan, Harsh Pal, Yatin Katyal and Sahil Manchanda},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.2118},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.2118}
}

References

42 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kirti Bhagat, Kinshuk Vasisht, and Danish Pruthi. 2025. Richer output for richer countries: Uncovering geographical disparities in generated stories and travel recommendations. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 4645–4653. find
  2. 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. Advances in neural information processing systems, 29.
  3. Sean Cao, Charles CY Wang, and Yi Xiang. 2025. When llms go abroad: Foreign bias in ai financial predictions. Available at SSRN 5440116. find
  4. Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. 2024. Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation. arXiv preprint arXiv:2402.03216.
  5. Pengyu Cheng, Weituo Hao, Siyang Yuan, Shijing Si, and Lawrence Carin. 2021. Fairfil: Contrastive neural debiasing method for pretrained text encoders. arXiv preprint arXiv:2103.06413. — [arXiv:2103.06413](https://arxiv.org/abs/2103.06413) find
  6. Fangxiaoyu Feng, Yinfei Yang, Daniel Cer, Naveen Arivazhagan, and Wei Wang. 2022. Language-agnostic bert sentence embedding. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 878–891.
  7. FinLang. 2024. ‘FinLang/finance-embeddings-investopedia‘. urlhuggingface.co. Version 1.0. find
  8. Michael Günther, Jackmin Ong, Isabelle Mohr, Alaeddine Abdessalem, Tanguy Abel, Mohammad Kalim Akram, Susana Guzman, Georgios Mastrapas, Saba Sturua, Bo Wang, and 1 others. 2023. Jina embeddings 2: 8192-token general-purpose text embeddings for long documents. arXiv preprint arXiv:2310.19923.
  9. Shailja Gupta and Rajesh Ranjan. 2024. Evaluation of llms biases towards elite universities: a persona-based exploration. arXiv preprint arXiv:2407.12801. — [arXiv:2407.12801](https://arxiv.org/abs/2407.12801) find
  10. Allen H Huang, Hui Wang, and Yi Yang. 2023. Finbert: A large language model for extracting information from financial text. Contemporary Accounting Research, 40(2):806–841. find
  11. Nitish Jaipuria, Lorenzo Gatto, Zijun Kan, Shankey Poddar, Bill Cheung, Diksha Bansal, Ramanan Balakrishnan, Aviral Suri, and Jose Estevez. 2025. Case: An agentic ai framework for enhancing scam intelligence in digital payments. arXiv preprint arXiv:2508.19932. — [arXiv:2508.19932](https://arxiv.org/abs/2508.19932) find
  12. Robin Jia and Percy Liang. 2017. Adversarial examples for evaluating reading comprehension systems. arXiv preprint arXiv:1707.07328.
  13. Mahammed Kamruzzaman, Hieu Minh Nguyen, and Gene Louis Kim. 2024. " global is good, local is bad?": Understanding brand bias in llms. arXiv preprint arXiv:2406.13997. find
  14. Hadas Kotek, Rikker Dockum, and David Sun. 2023. Gender bias and stereotypes in large language models. In Proceedings of the ACM collective intelligence conference, pages 12–24. find
  15. Nir Kshetri. 2025. The rise of agentic ai in finance: Opportunities, risks, and human-centric integration. IT Professional, 27(4):19–24. find
  16. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and Sebastian Riedel. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing Systems (…
  17. Yingji Li, Mengnan Du, Rui Song, Xin Wang, and Ying Wang. 2023a. A survey on fairness in large language models. arXiv preprint arXiv:2308.10149. — [arXiv:2308.10149](https://arxiv.org/abs/2308.10149) find
  18. Zehan Li, Xin Zhang, Yanzhao Zhang, Dingkun Long, Pengjun Xie, and Meishan Zhang. 2023b. Towards general text embeddings with multi-stage contrastive learning. arXiv preprint arXiv:2308.03281.
  19. Mohammad Luqman, Himaanshu Gauba, Prabhat Kumar, Akshar Prabhu Desai, Ajay Yadav, Ritu Prajapati, and Pranjul Yadav. 2025. Agentic ai in finance: A comprehensive overview. find
  20. Jabez Magomere, Emanuele La Malfa, Manuel Tonneau, Ashkan Kazemi, and Scott A Hale. 2025. When claims evolve: Evaluating and enhancing the robustness of embedding models against misinformation edits. In Findings of the Association for Computational Linguistics: ACL 2025, pages 22374–22404. find
  21. Sahil Manchanda and Pannaga Shivaswamy. 2025. What is in a name? mitigating name bias in text embedding similarity via anonymization. In Findings of the Association for Computational Linguistics: ACL 2025, pages 17759–17781, Vienna, Austria. Association for Computational Linguistics. find
  22. Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, and Stefano Ermon. 2024. Large language models are geographically biased. arXiv preprint arXiv:2402.02680. find
  23. Chandler May, Alex Wang, Shikha Bordia, Samuel R Bowman, and Rachel Rudinger. 2019. On measuring social biases in sentence encoders. arXiv preprint arXiv:1903.10561.
  24. Luke Merrick, Danmei Xu, Gaurav Nuti, and Daniel Campos. 2024. Arctic-embed: Scalable, efficient, and accurate text embedding models. arXiv preprint arXiv:2405.05374. — [arXiv:2405.05374](https://arxiv.org/abs/2405.05374) find
  25. Tarek Naous, Michael J Ryan, Alan Ritter, and Wei Xu. 2023. Having beer after prayer? measuring cultural bias in large language models. arXiv preprint arXiv:2305.14456.
  26. Huy Nghiem, John Prindle, Jieyu Zhao, and Hal Daumé III. 2024. " you gotta be a doctor, lin": An investigation of name-based bias of large language models in employment recommendations. arXiv preprint arXiv:2406.12232. find
  27. Zhijie Nie, Zhangchi Feng, Mingxin Li, Cunwang Zhang, Yanzhao Zhang, Dingkun Long, and Richong Zhang. 2024. When text embedding meets large language model: A comprehensive survey. arXiv preprint arXiv:2412.09165. — [arXiv:2412.09165](https://arxiv.org/abs/2412.09165) find
  28. Dmitry Nikolaev and Sebastian Padó. 2023. Representation biases in sentence transformers. arXiv preprint arXiv:2301.13039. find
  29. Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, and 1 others. 2011. Scikit-learn: Machine learning in python. the Journal of machine Learning research, 12:2825–2830. find
  30. Vasyl Rakivnenko, Nestor Maslej, Jessica Cervi, and Volodymyr Zhukov. 2024. Bias in text embedding models. arXiv preprint arXiv:2406.12138. — [arXiv:2406.12138](https://arxiv.org/abs/2406.12138) find
  31. N Reimers. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084.
  32. Alpay Sabuncuoglu and Carsten Maple. 2025. Identifying representation bias in large language models used in financial sentiment analysis. In 2025 IEEE Symposium on Computational Intelligence for Financial Engineering and Economics (CiFer), pages 1–7. IEEE. find
  33. Pola Schwöbel, Jacek Golebiowski, Michele Donini, Cédric Archambeau, and Danish Pruthi. 2023. Geographical erasure in language generation. arXiv preprint arXiv:2310.14777. find
  34. Claude E. Shannon. 1948. A mathematical theory of communication. Bell System Technical Journal, 27(3):379–423. find
  35. Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. 2019. Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. In Proceedings of the 28th ACM international conference on information and knowledge management, pages 1441–1450. find
  36. Yixuan Tang and Yi Yang. 2025. Finmteb: Finance massive text embedding benchmark. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. find
  37. I Tenney. 2019. Bert rediscovers the classical nlp pipeline. arXiv preprint arXiv:1905.05950.
  38. Henrique Schechter Vera, Sahil Dua, Biao Zhang, Daniel Salz, Ryan Mullins, Sindhu Raghuram Panyam, Sara Smoot, Iftekhar Naim, Joe Zou, Feiyang Chen, and 1 others. 2025. Embeddinggemma: Powerful and lightweight text representations. arXiv preprint arXiv:2509.20354. — [arXiv:2509.20354](https://arxiv.org/abs/2509.20354) find
  39. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024. Multilingual e5 text embeddings: A technical report. arXiv preprint arXiv:2402.05672.
  40. Yuwei Yin, Yazheng Yang, Jian Yang, and Qi Liu. 2023. Finpt: Financial risk prediction with profile tuning on pretrained foundation models. arXiv preprint arXiv:2308.00065. — [arXiv:2308.00065](https://arxiv.org/abs/2308.00065) find
  41. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, and 1 others. 2025. find
  42. Qwen3 embedding: Advancing text embedding and reranking through foundation models. arXiv preprint arXiv:2506.05176.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sudevan-2026-uncovering-currency-bias-syntax-gap/