A Survey on Gender Bias in Natural Language Processing

Abstract

Language can be used as a means of reproducing and enforcing harmful stereotypes and biases and has been analysed as such in numerous research. In this paper, we present a survey of 304 papers on gender bias in natural language processing. We analyse definitions of gender and its categories within social sciences and connect them to formal definitions of gender bias in NLP research. We survey lexica and datasets applied in research on gender bias and then compare and contrast approaches to detecting and mitigating gender bias. We find that research on gender bias suffers from four core limitations. 1) Most research treats gender as a binary variable neglecting its fluidity and continuity. 2) Most of the work has been conducted in monolingual setups for English or other high-resource languages. 3) Despite a myriad of papers on gender bias in NLP methods, we find that most of the newly developed algorithms do not test their models for bias and disregard possible ethical considerations of their work. 4) Finally, methodologies developed in this line of research are fundamentally flawed covering very limited definitions of gender bias and lacking evaluation baselines and pipelines. We see overcoming these limitations as a necessary development in future research.

Looking for a Handsome Carpenter! Debiasing GPT-3 Job Advertisements

“…The problem is then in defining this list of words. To avoid overfitting to one axis of gender bias, we construct a composite score based on pre-existing lists which have in turn been defined through experiments and empirical assessments (Schmader et al, 2007;Gaucher et al, 2011;Sap et al, 2017;Stanczak and Augenstein, 2021). The presence of words which are more likely to be associated with one gender does not directly result in biased outcomes.…”

“…n positive signifiers n words NCR VAD Lexicon This measure is based on a list of words rated on the emotional dimensions of valence, arousal, and dominance which has been used in gender bias research. In particular, weakness (low dominance), passiveness (low arousal or agency), and badness (valence) may be associated with a female stereotype (Stanczak and Augenstein, 2021). Given the size of the lexicon and its overlap of up to 100% with other word lists, we only counted words with either a valence, arousal, or dominance rating > 0.75 on a scale from 0 to 1.…”

How Conservative are Language Models? Adapting to the Introduction of Gender-Neutral Pronouns

“…Sun et al (2021) propose a rewriting task where data is transferred from gendered to gender-neutral pronouns to train more inclusive language models. Cao and Daumé III (2020) and Dev et al (2021) discuss the necessity of including non-binary pronouns into NLP research (see also Stanczak and Augenstein (2021)).…”

Indigenous Language Revitalization and the Dilemma of Gender Bias

“…Bias in NLP systems often goes without notice, it's often not even detected until after the systems are launched and used by consumers, which can have adverse effects on our society, such as when it shows false information to people which leads them to believe untrue things about society or them-selves; thereby changing their behavior for better or worse (Stanczak and Augenstein, 2021). The harm of bias in NLP has been understated by some people and overstated by others, who dismiss its relevance or refuse to engage with it altogether.…”