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

5267 It’s All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution Rowan Hall Maudslay1 Hila Gonen2 Ryan Cotterell1 Simone Teufel1 1 Department of Computer Science and Technology, University of Cambridge 2 Department of Computer Science, Bar-Ilan University {rh635,rdc42,sht25}@cam.ac.uk hilagnn@gmail.com Abstract This paper treats gender bias latent in word em- beddings. Previous mitigation attempts rely on the operationalisation of gender bias as a projection over a linear subspace. An alter- native approach is Counterfactual Data Aug- mentation (CDA), in which a corpus is dupli- cated and augmented to remove bias, e.g. by swapping all inherently-gendered words in the copy. We perform an empirical comparison of these approaches on the English Gigaword and Wikipedia, and find that whilst both success- fully reduce direct bias and perform well in tasks which quantify embedding quality, CDA variants outperform projection-based methods at the task of drawing non-biased gender analo- gies by an average of 19% across both corpora. We propose two improvements to CDA: Coun- terfactual Data Substitution (CDS), a variant of CDA in which potentially biased text is ran- domly substituted to avoid duplication, and the Names Intervention, a novel name-pairing technique that vastly increases the number of words being treated. CDA/S with the Names Intervention is the only approach which is able to mitigate indirect gender bias: following de- biasing, previously biased words are signifi- cantly less clustered according to gender (clus- ter purity is reduced by 49%), thus improving on the state-of-the-art for bias mitigation. 1 Introduction Gender bias describes an inherent prejudice against a gender, captured both by individuals and larger social systems. Word embeddings, a popular machine-learnt semantic space, have been shown to retain gender bias present in corpora used to train them (Caliskan et al., 2017). This results in gender-stereotypical vector analogies `a la Mikolov et al. (2013), such as man:computer programmer :: woman:homemaker (Bolukbasi et al., 2016), and such bias has been shown to materialise in a variety of downstream tasks, e.g. coreference resolution (Rudinger et al., 2018; Zhao et al., 2018). By operationalising gender bias in word embed- dings as a linear subspace, Bolukbasi et al. (2016) are able to debias with simple techniques from lin- ear algebra. Their method successfully mitigates direct bias: man is no longer more similar to com- puter programmer in vector space than woman. However, the structure of gender bias in vector space remains largely intact, and the new vectors still evince indirect bias: associations which result from gender bias between not explicitly gendered words, for example a possible association between football and business resulting from their mutual association with explicitly masculine words (Go- nen and Goldberg, 2019). In this paper we continue the work of Gonen and Goldberg, and show that an- other paradigm for gender bias mitigation proposed by Lu et al. (2018), Counterfactual Data Augmen- tation (CDA), is also unable to mitigate indirect bias. We also show, using a new test we describe (non-biased gender analogies), that WED might be removing too much gender information, casting …

Content truncated to 500 words