Evaluating Debiasing Techniques for Intersectional Biases

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2492–2498 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2492 Evaluating Debiasing Techniques for Intersectional Biases Shivashankar Subramanian Xudong Han Timothy Baldwin Trevor Cohn Lea Frermann School of Computing and Information Systems University of Melbourne, Australia shivashankarrs@gmail.com xudongh1@student.unimelb.edu.au {tbaldwin,t.cohn,lfrermann}@unimelb.edu.au Abstract Bias is pervasive in NLP models, motivating the development of automatic debiasing tech- niques. Evaluation of NLP debiasing methods has largely been limited to binary attributes in isolation, e.g., debiasing with respect to bi- nary gender or race, however many corpora involve multiple such attributes, possibly with higher cardinality. In this paper we argue that a truly fair model must consider ‘gerryman- dering’ groups which comprise not only sin- gle attributes, but also intersectional groups. We evaluate a form of bias-constrained model which is new to NLP, as well an extension of the iterative nullspace projection technique which can handle multiple protected attributes. 1 Introduction Text data reflects the social and cultural biases in the world, and NLP models and applications trained on such data have been shown to reproduce and amplify those biases. Discrimination has been identified across diverse sensitive attributes includ- ing gender, disability, race, and religion (Caliskan et al., 2017; May et al., 2019; Garimella et al., 2019; Nangia et al., 2020; Li et al., 2020). While early work focused on debiasing typically binarized protected attributes in isolation (e.g., age, gender, or race; Caliskan et al. (2017)), more recent work has adopted a more realistic scenario with multiple sensitive attributes (Li et al., 2018) or attributes covering several classes (Manzini et al., 2019). In the context of multiple protected attributes, gerrymandering refers to the phenomenon where an attempt to make a model fairer towards some group results in increased unfairness towards an- other group (Buolamwini and Gebru, 2018; Kearns et al., 2018; Yang et al., 2020). Notably, algorithms can be fair towards independent groups, but not towards all intersectional groups. Despite this, de- biasing approaches within NLP have so far been evaluated only using independent group fairness white person (W) person of colour (¬W) � � A B male (M) C D female (F) Figure 1: Group intersection and gerrymandering: A=white male, B=male person of colour, C=white fe- male, D=female person of colour. when modelling datasets with multiple attributes, disregarding intersectional subgroups defined by combinations of sensitive attributes (see Figure 1). The primary goal of this work is to evaluate in- dependent and intersectional identity debiasing ap- proaches in relation to fairness gerrymandering for text classification tasks. To this end, we evalu- ate bias-constrained models (Cotter et al., 2019b) and iterative nullspace projection (INLP; Ravfogel et al. (2020)), a post-hoc debiasing method which we extend to handle intersectional groups. The constrained model jointly optimizes model perfor- mance and model fairness, while INLP seeks to learn a hidden representation which is independent of the protected attributes. INLP does not consider the trade-off between accuracy and fairness, but rather it iteratively maximizes fairness in an uncon- …

Content truncated to 500 words