%0 Book Section %1 Lu2020 %A Lu, Kaiji %A Mardziel, Piotr %A Wu, Fangjing %A Amancharla, Preetam %A Datta, Anupam %B Logic, Language, and Security: Essays Dedicated to Andre Scedrov on the Occasion of His 65th Birthday %C Cham %D 2020 %E Nigam, Vivek %E Ban Kirigin, Tajana %E Talcott, Carolyn %E Guttman, Joshua %E Kuznetsov, Stepan %E Thau Loo, Boon %E Okada, Mitsuhiro %I Springer International Publishing %K bert_performance bias gender instituteclustering %P 189--202 %R 10.1007/978-3-030-62077-6_14 %T Gender Bias in Neural Natural Language Processing %U https://doi.org/10.1007/978-3-030-62077-6_14 %X We examine whether neural natural language processing (NLP) systems reflect historical biases in training data. We define a general benchmark to quantify gender bias in a variety of neural NLP tasks. Our empirical evaluation with state-of-the-art neural coreference resolution and textbook RNN-based language models trained on benchmark data sets finds significant gender bias in how models view occupations. We then mitigate bias with counterfactual data augmentation (CDA): a generic methodology for corpus augmentation via causal interventions that breaks associations between gendered and gender-neutral words. We empirically show that CDA effectively decreases gender bias while preserving accuracy. We also explore the space of mitigation strategies with CDA, a prior approach to word embedding debiasing (WED), and their compositions. We show that CDA outperforms WED, drastically so when word embeddings are trained. For pre-trained embeddings, the two methods can be effectively composed. We also find that as training proceeds on the original data set with gradient descent the gender bias grows as the loss reduces, indicating that the optimization encourages bias; CDA mitigates this behavior. %@ 978-3-030-62077-6

@inbook{Lu2020, abstract = {We examine whether neural natural language processing (NLP) systems reflect historical biases in training data. We define a general benchmark to quantify gender bias in a variety of neural NLP tasks. Our empirical evaluation with state-of-the-art neural coreference resolution and textbook RNN-based language models trained on benchmark data sets finds significant gender bias in how models view occupations. We then mitigate bias with counterfactual data augmentation (CDA): a generic methodology for corpus augmentation via causal interventions that breaks associations between gendered and gender-neutral words. We empirically show that CDA effectively decreases gender bias while preserving accuracy. We also explore the space of mitigation strategies with CDA, a prior approach to word embedding debiasing (WED), and their compositions. We show that CDA outperforms WED, drastically so when word embeddings are trained. For pre-trained embeddings, the two methods can be effectively composed. We also find that as training proceeds on the original data set with gradient descent the gender bias grows as the loss reduces, indicating that the optimization encourages bias; CDA mitigates this behavior.}, added-at = {2021-01-20T02:45:39.000+0100}, address = {Cham}, author = {Lu, Kaiji and Mardziel, Piotr and Wu, Fangjing and Amancharla, Preetam and Datta, Anupam}, biburl = {https://www.bibsonomy.org/bibtex/2cc8ec585f16c688abae6af846986e4f5/parismic}, booktitle = {Logic, Language, and Security: Essays Dedicated to Andre Scedrov on the Occasion of His 65th Birthday}, description = {Gender Bias in Neural Natural Language Processing | SpringerLink}, doi = {10.1007/978-3-030-62077-6_14}, editor = {Nigam, Vivek and Ban Kirigin, Tajana and Talcott, Carolyn and Guttman, Joshua and Kuznetsov, Stepan and Thau Loo, Boon and Okada, Mitsuhiro}, interhash = {cd15802c3527469a491c5b93fc493cfa}, intrahash = {cc8ec585f16c688abae6af846986e4f5}, isbn = {978-3-030-62077-6}, keywords = {bert_performance bias gender instituteclustering}, pages = {189--202}, publisher = {Springer International Publishing}, timestamp = {2021-01-20T02:45:39.000+0100}, title = {Gender Bias in Neural Natural Language Processing}, url = {https://doi.org/10.1007/978-3-030-62077-6_14}, year = 2020 }