Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5443–5453 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 5443 Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation Tianlu Wang1∗ Xi Victoria Lin2 Nazneen Fatema Rajani2 Bryan McCann2 Vicente Ordonez1 Caiming Xiong2 1University of Virginia {tw8cb, vicente}@virginia.edu 2Salesforce Research {xilin, nazneen.rajani, bmccann, cxiong}@salesforce.com Abstract Word embeddings derived from human- generated corpora inherit strong gender bias which can be further amplified by downstream models. Some commonly adopted debiasing approaches, including the seminal Hard Debias algorithm (Bolukbasi et al., 2016), apply post-processing procedures that project pre-trained word embeddings into a subspace orthogonal to an inferred gender subspace. We discover that semantic-agnostic corpus regularities such as word frequency captured by the word embeddings negatively impact the performance of these algorithms. We propose a simple but effective technique, Double-Hard Debias, which purifies the word embeddings against such corpus regularities prior to inferring and removing the gender subspace. Experiments on three bias miti- gation benchmarks show that our approach preserves the distributional semantics of the pre-trained word embeddings while reducing gender bias to a significantly larger degree than prior approaches. 1 Introduction Despite widespread use in natural language pro- cessing (NLP) tasks, word embeddings have been criticized for inheriting unintended gender bias from training corpora. Bolukbasi et al. (2016) high- lights that in word2vec embeddings trained on the Google News dataset (Mikolov et al., 2013a), “pro- grammer” is more closely associated with “man” and “homemaker” is more closely associated with “woman”. Such gender bias also propagates to downstream tasks. Studies have shown that coref- erence resolution systems exhibit gender bias in predictions due to the use of biased word embed- dings (Zhao et al., 2018a; Rudinger et al., 2018). Given the fact that pre-trained word embeddings ∗This research was conducted during the author’s intern- ship at Salesforce Research. have been integrated into a vast number of NLP models, it is important to debias word embeddings to prevent discrimination in NLP systems. To mitigate gender bias, prior work have proposed to remove the gender component from pre-trained word embeddings through post- processing (Bolukbasi et al., 2016), or to compress the gender information into a few dimensions of the embedding space using a modified training scheme (Zhao et al., 2018b; Kaneko and Bollegala, 2019). We focus on post-hoc gender bias mitigation for two reasons: 1) debiasing via a new training ap- proach is more computationally expensive; and 2) pre-trained biased word embeddings have already been extensively adopted in downstream NLP prod- ucts and post-hoc bias mitigation presumably leads to less changes in the model pipeline since it keeps the core components of the original embeddings. Existing post-processing algorithms, including the seminal Hard Debias (Bolukbasi et al., 2016), debias embeddings by removing the component that corresponds to a gender direction as defined by a list of gendered words. While Bolukbasi et al. (2016) demonstrates that such methods alleviate gender bias in word analogy tasks, Gonen and Gold- berg (2019) argue that the …

Content truncated to 500 words