Nurse is Closer to Woman than Surgeon? Mitigating Gender-Biased Proximities in Word Embeddings

Abstract

Word embeddings are the standard model for semantic and syntactic representations of words. Unfortunately, these models have been shown to exhibit undesirable word associations resulting from gender, racial, and religious biases. Existing post-processing methods for debiasing word embeddings are unable to mitigate gender bias hidden in the spatial arrangement of word vectors. In this paper, we propose RAN-Debias, a novel gender debiasing methodology that not only eliminates the bias present in a word vector but also alters the spatial distribution of its neighboring vectors, achieving a bias-free setting while maintaining minimal semantic offset. We also propose a new bias evaluation metric, Gender-based Illicit Proximity Estimate (GIPE), which measures the extent of undue proximity in word vectors resulting from the presence of gender-based predilections. Experiments based on a suite of evaluation metrics show that RAN-Debias significantly outperforms the state-of-the-art in reducing proximity bias (GIPE) by at least 42.02%. It also reduces direct bias, adding minimal semantic disturbance, and achieves the best performance in a downstream application task (coreference resolution).

Fair Embedding Engine: A Library for Analyzing and Mitigating Gender Bias in Word Embeddings

“…FEE, on the other hand, provides holistic functionality by equipping a suite of evaluation and debiasing methods, along with a flexible design to assist researchers in developing new solutions. FEE currently offers three debiasing methods as a part of its debiasing module: HardDebias (Bolukbasi et al, 2016b), HSRDebias (Yang and Feng, 2020), and RANDebias (Kumar et al, 2020). The bias metrics module consist of the following: SemBias (Zhao et al, 2018b), direct and indirect bias (Bolukbasi et al, 2016a), Gender-basied Illicit Proximity Estimate (GIPE) and Proximity bias (Kumar et al, 2020), Percent Male Neighbours (PMN) (Gonen and Goldberg, 2019) and Word Embedding Association Test (WEAT) (Caliskan et al, 2017b).…”

“…Motivation: Visualizations provide useful insights into the behaviour of a set of data points. Many prior debiasing methods (Bolukbasi et al, 2016a;Kumar et al, 2020) have strongly motivated their work by illustrating certain undesirable associations prevalent in standard word embeddings. Thus, through FEE we also provide off the shelf visualization capabilities that might help users to build reliable intuitions and uncover hidden biases in their models.…”

“…(Zhao et al, 2018b;Lu et al, 2018;Bordia and Bowman, 2019). Along with the development of debiasing procedures, numerous metrics to evaluate the efficacy of each debiasing procedure have also been proposed (Zhao et al, 2018b;Bolukbasi et al, 2016a;Kumar et al, 2020). Although the domain has largely benefited from the contributions of different researchers, the domain still lacks open source software projects that unify such diverse but fundamentally similar methods in an organized and standard manner.…”

Dynamically Disentangling Social Bias from Task-Oriented Representations with Adversarial Attack

“…In this paper, we focus on social bias, such as gender bias which is the preference or prejudice towards one gender over the other (Moss-Racusin et al, 2012), race bias and age bias. From the perspective of the debiasing target, previous debiasing works can be approximately classified into two types, word embedding (Bolukbasi et al, 2016;Caliskan et al, 2017;Zhao et al, 2018;Manzini et al, 2019;Wang et al, 2020;Kumar et al, 2020) and sentence embedding (Xu et al, 2017;Elazar and Goldberg, 2018a;Zhang et al, 2018;Ravfogel et al, 2020). The former aims to reduce the gender bias in word embedding, either as a post-processing step (Bolukbasi et al, 2016) or as part of the training procedure (Zhao et al, 2018).…”

Proceedings of the 3rd Workshop on Gender Bias in Natural Language Processing

“…Early methods Zhao et al, 2018b) focused on eliminating direct bias from the embedding space, quantified as associations between gender-neutral words and an explicit gender vocabulary. In response to an influential critique paper by , the current trend is to focus on eliminating indirect bias from the embedding space, quantified either by gender-induced proximity among embeddings (Kumar et al, 2020) or by residual gender cues that could be learned by a classifier (Ravfogel et al, 2020;Davis et al, 2020).…”