Elamur Research⌕
☰
home / papers / 2019 / It's All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution

It's All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution

Paper · 2019 · Rowan Hall Maudslay, Hila Gonen, Ryan Cotterell et al.

Summary

Word embeddings absorb gender stereotypes from their training text. The common fix, projection-based Word Embedding Debiasing (WED), removes a single "gender direction." However, it leaves indirect bias intact: words that are not explicitly gendered still group by gender. This paper compares WED with Counterfactual Data Augmentation (CDA), which retrains embeddings on a corpus where gendered words have been swapped. Both are tested on English Gigaword and Wikipedia.

The authors propose two changes to CDA:

  • Counterfactual Data Substitution (CDS): each document is swapped at random instead of being duplicated, so the corpus keeps natural statistics.
  • Names Intervention: first names are paired and swapped too. Pairs are built from US Social Security Administration data using bipartite matching that balances name frequency and gender-specificity.

Both approaches reduce direct bias and keep embedding quality. WED, however, appears to remove too much gender information, while the CDA variants handle legitimate gender analogies much better. Only CDA/CDS combined with the Names Intervention noticeably weakens indirect bias. This shows that bias does not fit the linear-subspace view.

Key points

  • On non-biased gender analogies, CDA variants beat projection methods by about 19% on average across both corpora. The best WED variant had error rates of 27.1% vs. 9.3% unmitigated on Gigaword, and 18.8% vs. 8.7% on Wikipedia.
  • With the Names Intervention, cluster purity of the most biased words drops by about 49% on average. nCDS was significantly best at reducing how often these words could be reclassified by gender (p < 0.01), though the rate stayed high (88.9% on Wikipedia).
  • Name pairs cover the 2,500 most frequent names per gender in the SSA dataset. They are matched with the Hungarian method on frequency and gender-specificity, and Named Entity Recognition separates names from common nouns such as Rose or Mark.
  • CDS substitutes with probability 0.5 per document and is faster to train than duplicating the corpus. On SimLex-999 and sentiment classification, all methods stay close to the unmitigated baseline.
  • Stated limitations:
    • the methods depend on fixed lists of gendered word pairs;
    • they assume a gender binary;
    • the name list is US-only.
    • Suggested future work includes extending the approach to racial bias.

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

🎓University of CambridgeCambridge · United Kingdom155 works · 115,562 citations🎓Bar-Ilan UniversityRamat Gan · Israel118 works · 21,077 citations

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 worksSocial Sciences · 744 worksSafety Research · 221 works

🧭 Topics 1Research subjects and languages the work is about.Bias, fairness & safety · 209 works

Cite

Cite as Rowan Hall Maudslay, Hila Gonen, Ryan Cotterell and Simone Teufel. 2019. It's All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d19-1530

@inproceedings{maudslay2019it,
  title = {{It's All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution}},
  author = {Rowan Hall Maudslay, Hila Gonen, Ryan Cotterell and Simone Teufel},
  booktitle = {Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
  year = {2019},
  doi = {10.18653/v1/d19-1530},
  url = {https://doi.org/10.18653/v1/d19-1530}
}

References

19 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David Bamman and Noah A. Smith. 2014. Unsupervised discovery of biographical structure from text. Transactions of the Association for Computational Linguistics, 2:363–376. find
  2. Tolga Bolukbasi, Kai-Wei Chang, James Y. Zou, Venkatesh Saligrama, and Adam Tauman Kalai. 2016. Man is to computer programmer as woman is to homemaker? Debiasing word embeddings. In Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016,…
  3. 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
  4. 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…
  5. Eduardo Graells-Garrido, Mounia Lalmas, and Filippo Menczer. 2015. First women, second sex: Gender bias in Wikipedia. In Proceedings of the 26th ACM Conference on Hypertext &\#38; Social Media, HT ’15, pages 165–174, New York, NY, USA. ACM. find
  6. Anthony G. Greenwald, Debbie E. McGhee, and Jordan L. K. Schwartz. 1998. Measuring individual differences in implicit cognition: The Implicit Association Test. Journal of Personality and Social Psychology, 74(6):1464–1480. find
  7. Felix Hill, Roi Reichart, and Anna Korhonen. 2015. SimLex-999: Evaluating semantic models with (genuine) similarity estimation. American Journal of Computational Linguistics, 41(4):665–695.
  8. Harold W. Kuhn. 1955. The Hungarian method for the assignment problem. Naval Research Logistics Quarterly, 2(1-2):83–97. find
  9. Jey Han Lau and Timothy Baldwin. 2016. An empirical evaluation of doc2vec with practical insights into document embedding generation. In Proceedings of the 1st Workshop on Representation Learning for NLP, pages 78–86, Berlin, Germany. Association for Computational Linguistics. find
  10. Quoc Le and Tomas Mikolov. 2014. Distributed representations of sentences and documents. In Proceedings of the 31st International Conference on Machine Learning, volume 32 of Proceedings of Machine Learning Research, pages 1188–1196, Bejing, China. PMLR.
  11. Kaiji Lu, Piotr Mardziel, Fangjing Wu, Preetam Amancharla, and Anupam Datta. 2018. Gender bias in neural natural language processing. CoRR, abs/1807.11714.
  12. Laurens van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. Journal of Machine Learning Research, 9:2579–2605. find
  13. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S. Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. In C. J. C. Burges, L. Bottou, M. Welling, Z. Ghahramani, and K. Q. Weinberger, editors, Advances in Neural Information Processing Systems 26, p…
  14. Courtney Napoles, Matthew Gormley, and Benjamin Van Durme. 2012. Annotated gigaword. In AKBC-WEKEX Workshop at NAACL 2012.
  15. Brian A. Nosek, Mahzarin R. Banaji, and Anthony G Greenwald. 2002. Harvesting implicit group attitudes and beliefs from a demonstration web site. Group Dynamics: Theory, Research, and Practice, 6 1:101–115. find
  16. Joseph Reagle and Lauren Rhue. 2011. Gender bias in Wikipedia and Britannica. International Journal of Communication, 5(0). find
  17. Andrew Rosenberg and Julia Hirschberg. 2007. V-measure: A conditional entropy-based external cluster evaluation measure. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL), pages 410– 420, Prague,…
  18. Rachel Rudinger, Jason Naradowsky, Brian Leonard, and Benjamin Van Durme. 2018. Gender bias in coreference resolution. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 8–1…
  19. 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/maudslay-2019-it-s-all-name-mitigating/