Elamur Research⌕
☰
home / papers / 2026 / Bias Mitigation in Hiring-Related NLP: Interactions Between Masking, Rewriting, and Adversarial Debiasing

Bias Mitigation in Hiring-Related NLP: Interactions Between Masking, Rewriting, and Adversarial Debiasing

Paper · 2026 · Alexandre Puttick and Rami El-Wazzi

Summary

NLP fairness research for recruitment usually tests one debiasing technique at a time, mostly on English data and apart from real use cases. This study examines what happens when data-level and model-level interventions are combined in a Norwegian, hiring-like scenario. The authors collected NTNU faculty bios and used STEM vs. non-STEM classification as a stand-in for a recruitment decision. They compared three interventions on a fine-tuned NbAiLab/nb-bert-base classifier: masking of names and gendered pronouns, case-based rewriting with GenWriter, and adversarial debiasing. Each configuration was scored on task accuracy, gender gaps in accuracy and equality of odds, WEAT/LPBS intrinsic tests, and how well gender can be recovered from hidden representations. Using all three methods together pushed gender recoverability close to chance without hurting classification quality, and in some settings it improved. Fairness gaps and intrinsic bias scores, however, shifted inconsistently. The authors therefore argue that mitigation in hiring NLP should be evaluated on several dimensions and grounded in the actual downstream task.

Key points

  • Dataset: 1,357 NTNU bios (736 male, 621 female; 832 STEM, 525 non-STEM) covering professors, associate professors and PhD candidates. Departments were hand-assigned to STEM (27 of 51) or non-STEM.
  • Gender labels: Labels are binary only, because there was just one non-binary example. They were inferred by an NbAiLab/nb-sbert-base classifier trained on 100 hand-labelled bios, which reached about 98% accuracy in 5-fold cross-validation.
  • GenWriter setup: The sentence classifier was trained on 222 human-labelled and 900 GPT 5.1-labelled sentences (88% agreement with humans) plus 200 synthetic Non-Professional samples. It reached 81.9 ± 1.53% cross-validation accuracy.
  • Adversarial setup: A single 3-layer MLP discriminator reads the final-layer [CLS] vector. Training alternates one main-task epoch with ten adversary epochs (Adam, learning rate 1e-5).
  • Findings and limits: Lower leakage did not consistently coincide with lower WEAT/LPBS bias. The study is limited by the proxy task, data from a single institution, binary gender only, and encoder-only models (no LLMs).

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

🎓Bern University of Applied SciencesBern · Switzerland4 works · 154 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksSocial Sciences · 1,772 worksSafety Research · 543 worksBusiness, Management and Accounting · 32 worksOrganizational Behavior and Human Resource Management · 12 works

🧭 Topics 1Research subjects and languages the work is about.Bias, fairness & safety · 517 works

Cite

Cite as Alexandre Puttick and Rami El-Wazzi. 2026. Bias Mitigation in Hiring-Related NLP: Interactions Between Masking, Rewriting, and Adversarial Debiasing. Proceedings of the 6th International Conference on Natural Language Processing for the Digital Humanities.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.nlp4dh-1.19

@inproceedings{puttick2026bias,
  title = {{Bias Mitigation in Hiring-Related NLP: Interactions Between Masking, Rewriting, and Adversarial Debiasing}},
  author = {Alexandre Puttick and Rami El-Wazzi},
  booktitle = {Proceedings of the 6th International Conference on Natural Language Processing for the Digital Humanities},
  year = {2026},
  doi = {10.18653/v1/2026.nlp4dh-1.19},
  url = {https://doi.org/10.18653/v1/2026.nlp4dh-1.19}
}

References

18 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 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.
  2. Aylin Caliskan, Joanna J Bryson, and Arvind Narayanan. 2017. Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334):183–186. find
  3. Pieter Delobelle, Ewoenam Kwaku Tokpo, Toon Calders, and Bettina Berendt. 2022. Measuring fairness with biased rulers: A comparative study on bias metrics for pre-trained language models. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguis… find
  4. Sina Fazelpour and Zachary C Lipton. 2020. Algorithmic fairness from a non-ideal perspective. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, pages 57–63. find
  5. Shira Mitchell, Eric Potash, Solon Barocas, Alexander D’Amour, and Kristian Lum. 2021. Algorithmic fairness: Choices, assumptions, and definitions. Annual review of statistics and its application, 8(1):141–163. find
  6. Seraphina Goldfarb-Tarrant, Rebecca Marchant, Ricardo Muñoz Sánchez, Mugdha Pandya, and Adam Lopez. 2021. Intrinsic bias metrics do not correlate with application bias. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Confere…
  7. Shweta Soundararajan and Sarah Jane Delany. 2025. Genwriter: Reducing gender cues in biographies through text rewriting. In Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP), pages 347–357. find
  8. Hila Gonen and Yoav Goldberg. 2019. Lipstick on a pig: Debiasing methods cover up systematic gender biases in word embeddings but do not remove them. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol…
  9. Xudong Han, Timothy Baldwin, and Trevor Cohn. 2021. Diverse adversaries for mitigating bias in training. arXiv preprint arXiv:2101.10001. find
  10. Xudong Han, Timothy Baldwin, and Trevor Cohn. 2022. Towards equal opportunity fairness through adversarial learning. arXiv preprint arXiv:2203.06317. — [arXiv:2203.06317](https://arxiv.org/abs/2203.06317) find
  11. Moritz Hardt, Eric Price, and Nati Srebro. 2016. Equality of opportunity in supervised learning. Advances in neural information processing systems, 29.
  12. Hoda Heidari, Michele Loi, Krishna P Gummadi, and Andreas Krause. 2019. A moral framework for understanding fair ml through economic models of equality of opportunity. In Proceedings of the conference on fairness, accountability, and transparency, pages 181–190. find
  13. Ewoenam Kwaku Tokpo, Pieter Delobelle, Bettina Berendt, and Toon Calders. 2023. How far can it go? on intrinsic gender bias mitigation for text classification. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 3418–3433. find
  14. Schrasing Tong, Minseok Jung, Ilaria Liccardi, and Lalana Kagal. 2026. Measuring perceptions of fairness in ai systems: The effects of infra-marginality. arXiv preprint arXiv:2603.05889. — [arXiv:2603.05889](https://arxiv.org/abs/2603.05889) find
  15. Meike Zehlike, Francesco Bonchi, Carlos Castillo, Sara Hajian, Mohamed Megahed, and Ricardo Baeza-Yates. 2017. Fa* ir: A fair top-k ranking algorithm. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, pages 1569– 1578. find
  16. Brian Hu Zhang, Blake Lemoine, and Margaret Mitchell. 2018. Mitigating unwanted biases with adversarial learning. In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, pages 335– 340. find
  17. Ran Zmigrod, Sabrina J Mielke, Hanna Wallach, and Ryan Cotterell. 2019. Counterfactual data augmentation for mitigating gender stereotypes in languages with rich morphology. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1651–1661.
  18. Keita Kurita, Nidhi Vyas, Ayush Pareek, Alan W Black, and Yulia Tsvetkov. 2019. Measuring bias in contextualized word representations. arXiv preprint arXiv:1906.07337.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/puttick-2026-bias-mitigation-hiring-related-nlp/