Elamur Research
☰
home / papers / 2018 / Learning Gender-Neutral Word Embeddings

Learning Gender-Neutral Word Embeddings

Written for the Research library (claude-opus-5-5) from the paper text. Original — aclanthology.org/D18-1521.pdf

Summary

Word vectors trained on human-written text pick up gender stereotypes. Occupation words such as "programmer" or "doctor" drift toward one gender, and that bias can carry into systems like coreference resolvers or resume screeners. An earlier remedy by Bolukbasi et al. (2016), used here as the Hard-GloVe baseline, removes bias after training. It has two drawbacks: it relies on a separate classifier to decide which words should be neutral, and it erases gender information entirely. This paper instead handles bias during training. GN-GloVe extends the GloVe objective with two extra terms. The first concentrates the gender signal of male- and female-definition seed words (taken from WordNet) in a single reserved dimension. The second keeps the remaining dimensions of gender-neutral words orthogonal to a gender direction that is re-estimated at each epoch. The model is trained on a 2017 English Wikipedia dump. It almost always identifies true gender pairs in an analogy test and keeps word-similarity quality, losing only a little on analogies. It also narrows the stereotype gap in coreference. The method works for any language with a gender seed list, and the code is public.

Key points

  • Method: The embedding is split into a neutral part w(a) and a gendered part w(g), with k = 1. The method has two variants of the gender-separation loss (L1 and L2), and λd = λe = 0.8.
  • SemBias analogy test (440 instances): GN-GloVe chooses the gender-definition pair 97.7% of the time, compared with 80.2% for GloVe and 84.1% for Hard-GloVe. On a 40-instance subset built from pairs not used as seeds, the scores are 75%, 57.5% and 25%.
  • Standard benchmarks: Similarity scores rise, for example RW from 37.3 to 40.0 and MEN-TR-3k from 72.2 to 74.5. Analogy accuracy drops slightly, with Google going from 70.8 to 68.9 and MSR from 45.8 to 43.7.
  • Coreference (Lee et al., 2017 end-to-end model): OntoNotes 5.0 F1 stays close to the baseline (66.2 vs. 66.5). On WinoBias, the gap between PRO and ANTI subsets shrinks from 30.2 to 20.5, and to 16.1 when only w(a) is used. Hard-GloVe reaches 15.7.
  • Resource: Code and data are released at github.com/uclanlp/gn_glove.

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: Natural language processing

Topics: Coreference resolution

Cited by

167 papers of our library cite it (of the 140673 whose bibliographies we read); 17 of them have a page here, 150 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d18-1521

@inproceedings{zhao2018learning,
  title = {{Learning Gender-Neutral Word Embeddings}},
  author = {Zhao, Jieyu and Zhou, Yichao and Li, Zeyu and Wang, Wei and Chang, Kai-Wei},
  booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing},
  year = {2018},
  doi = {10.18653/v1/d18-1521},
  url = {https://doi.org/10.18653/v1/d18-1521}
}

References

38 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sanjeev Arora, Yuanzhi Li, Yingyu Liang, Tengyu Ma, and Andrej Risteski. 2018. Linear algebraic structure of word senses, with applications to polysemy. Transactions of the Association of Computational Linguistics, 6:483–495. find
  2. Sudie E Back, Rebecca L Payne, Annie N Simpson, and Kathleen T Brady. 2010. Gender and prescription opioids: Findings from the national survey on drug use and health. Addictive behaviors, 35(11):1001–1007. find
  3. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473. find our card
  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 Advances in Neural Information Processing Systems, pages 4349–4357. find our card
  5. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (EMNLP). Association for Computational Linguistics. find
  6. Elia Bruni, Gemma Boleda, Marco Baroni, and Nam-Khanh Tran. 2012. Distributional semantics in technicolor. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers-Volume 1, pages 136–145. find
  7. 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
  8. Manaal Faruqui, Jesse Dodge, Sujay K Jauhar, Chris Dyer, Eduard Hovy, and Noah A Smith. 2014. Retrofitting word vectors to semantic lexicons. arXiv preprint arXiv:1411.4166. find
  9. Lev Finkelstein, Evgeniy Gabrilovich, Yossi Matias, Ehud Rivlin, Zach Solan, Gadi Wolfman, and Eytan Ruppin. 2001. Placing search in context: The concept revisited. In Proceedings of the 10th international conference on World Wide Web, pages 406– 414. find
  10. Nikhil Garg, Londa Schiebinger, Dan Jurafsky, and James Zou. 2017. Word embeddings quantify 100 years of gender and ethnic stereotypes. arXiv preprint arXiv:1711.08412. — [arXiv:1711.08412](https://arxiv.org/abs/1711.08412) find
  11. Nikhil Garg, Londa Schiebinger, Dan Jurafsky, and James Zou. 2018. Word embeddings quantify 100 years of gender and ethnic stereotypes. Proceedings of the National Academy of Sciences, 115(16):E3635–E3644. find
  12. Guy Halawi, Gideon Dror, Evgeniy Gabrilovich, and Yehuda Koren. 2012. Large-scale learning of word relatedness with constraints. In Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 1406– 1414. ACM. find
  13. David A Jurgens, Peter D Turney, Saif M Mohammad, and Keith J Holyoak. 2012. Semeval-2012 task 2: Measuring degrees of relational similarity. In Proceedings of the First Joint Conference on Lexical and Computational Semantics-Volume 1: Proceedings of the main conference and the shared task, and Volu… find
  14. Matthew Kay, Cynthia Matuszek, and Sean A Munson. 2015. Unequal representation and gender stereotypes in image search results for occupations. In Human Factors in Computing Systems, pages 3819– 3828. find
  15. Kenton Lee, Luheng He, Mike Lewis, and Luke Zettlemoyer. 2017. End-to-end neural coreference resolution. EMNLP. find our card
  16. Omer Levy, Yoav Goldberg, and Ido Dagan. 2015. Improving distributional similarity with lessons learned from word embeddings. Transactions of the Association for Computational Linguistics, 3:211–225. find
  17. Thang Luong, Richard Socher, and Christopher Manning. 2013. Better word representations with recursive neural networks for morphology. In Proceedings of the Seventeenth Conference on Computational Natural Language Learning, pages 104–113. find our card
  18. Anna C McFadden, George E Marsh, Barrie Jo Price, and Yunhan Hwang. 1992. A study of race and gender bias in the punishment of school children. Education and treatment of children, pages 140–146. find
  19. Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013a. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781. find
  20. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013b. Distributed representations of words and phrases and their compositionality. In Advances in neural information processing systems, pages 3111–3119. find
  21. Tomas Mikolov, Wen-tau Yih, and Geoffrey Zweig. 2013c. Linguistic regularities in continuous space word representations. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 746–751. find
  22. George Miller and Christiane Fellbaum. 1998. Wordnet: An electronic lexical database. find
  23. Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. Glove: Global vectors for word representation. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532–1543. find
  24. Matthew E Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep contextualized word representations. arXiv preprint arXiv:1802.05365. find
  25. Kira Radinsky, Eugene Agichtein, Evgeniy Gabrilovich, and Shaul Markovitch. 2011. A word at a time: computing word relatedness using temporal semantic analysis. In Proceedings of the 20th international conference on World wide web, pages 337–346. find
  26. Daojian Zeng, Kang Liu, Yubo Chen, and Jun Zhao. 2015. Distant supervision for relation extraction via piecewise convolutional neural networks. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1753– 1762. find
  27. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2017. Men also like shopping: Reducing gender bias amplification using corpus-level constraints. arXiv preprint arXiv:1707.09457. find our card
  28. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2018. Gender bias in coreference resolution: Evaluation and debiasing methods. NAACL. find our card
  29. Herbert Rubenstein and John B Goodenough. 1965. Contextual correlates of synonymy. Communications of the ACM, 8(10):627–633. find
  30. Rachel Rudinger, Chandler May, and Benjamin Van Durme. 2017. Social bias in elicited natural language inferences. In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, pages 74–79. find our card
  31. Rachel Rudinger, Jason Naradowsky, Brian Leonard, and Benjamin Van Durme. 2018. Gender bias in coreference resolution. In NAACL. find our card
  32. Cicero dos Santos and Maira Gatti. 2014. Deep convolutional neural networks for sentiment analysis of short texts. In Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 69–78. find
  33. Simon Šuster, Ivan Titov, and Gertjan Van Noord. 2016. Bilingual learning of multi-sense embeddings with discrete autoencoders. arXiv preprint arXiv:1603.09128. find
  34. Latanya Sweeney. 2013. Discrimination in online ad delivery. Queue, 11(3):10. find
  35. Shyam Upadhyay, Kai-Wei Chang, Matt Taddy, Adam Kalai, and James Zou. 2017. Beyond bilingual: Multi-sense word embeddings using multilingual context. In Proceedings of the 2nd Workshop on Representation Learning for NLP, pages 101–110. find
  36. Byron C Wallace and Michael J Paul. 2016. jerk or judgemental? patient perceptions of male versus female physicians in online reviews. find
  37. Ralph Weischedel, Sameer Pradhan, Lance Ramshaw, Jeff Kaufman, Michelle Franchini, Mohammed El-Bachouti, Nianwen Xue, Martha Palmer, Jena D Hwang, Claire Bonial, et al. 2012. Ontonotes release 5.0. find
  38. Sirui Yao and Bert Huang. 2017. Beyond parity: Fairness objectives for collaborative filtering. In Advances in Neural Information Processing Systems, pages 2925–2934. find