Learning Gender-Neutral Word Embeddings

By Jieyu Zhao, Yichao Zhou, Zeyu Li, Wei Wang, Kai-Wei Chang. 2018

461 citations recorded by the source corpus.

Word embedding models have become a fundamental component in a wide range of Natural Language Processing (NLP) applications. However, embeddings trained on human-generated corpora have been demonstrated to inherit strong gender stereotypes that reflect social constructs. To address this concern, in this paper, we propose a novel training procedure for learning gender-neutral word embeddings. Our approach aims to preserve gender information in certain dimensions of word vectors while compelling other dimensions to be free of gender influence. Based on the proposed method, we generate a Gender-Neutral variant of GloVe (GN-GloVe). Quantitative and qualitative experiments demonstrate that GN-GloVe successfully isolates gender information without sacrificing the functionality of the embedding model.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Representation & Self-Supervised Learning

In Safety, Robustness & Interpretability

Citations in the corpus

Language (Technology) is Power: A Critical Survey of “Bias” in NLP (2020, background): “…developed for a breadth of tasks including language modeling (Lu et al., 2018; Bordia and Bowman, 2019), coreference resolution (Rudinger et al., 2018; Zhao et al., 2018a), machine translation (Vanmassenhove et al., 2018; Stanovsky et al., 2019), sentiment analysis (Kiritchenko and Mohammad,…”
Reducing Sentiment Bias in Language Models via Counterfactual Evaluation (2019, methodology): “For word embeddings, occupational gender bias has been identiﬁed and addressed by measuring projections onto linear gender-related subspaces of word representations (Bolukbasi et al., 2016; Lemoine et al., 2018; Zhao et al., 2018b; Bordia & Bowman, 2019).”
Reducing Gender Bias in Word-Level Language Models with a Gender-Equalizing Loss Function (2019, methodology): “Additionally, Zhao et al. (2018) demonstrated that coreference resolution systems exhibit gender bias.”
Fair Is Better than Sensational: Man Is to Doctor as Woman Is to Doctor (2019, background): “man is to doctor as woman is to nurse, or man is to computer programmer as woman is to homemaker. Attempts at reducing bias, either via postprocessing (Bolukbasi et al. 2016) or directly in training (Zhao et al. 2018) have nevertheless left two outstanding issues: bias is still encoded implicitly (G”
Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them (2019, methodology): “, 2016b) and as part of the training procedure (Zhao et al., 2018).”
Societal Biases in Language Generation: Progress and Challenges (2021, methodology): “First, we obtain the lists of female words and male words from Zhao et al. (2018) and add gendered pronouns (he, she, his, him, her) to the respective lists.”
Toward Gender-Inclusive Coreference Resolution (2019, neutral): “In comparison to [Zhao et al. 2018] and [Rudinger et al. 2018] (as well as in contrast to our MAP dataset), we focused on naturally occurring data, but sampled specifically to surface more gender-related phenomena than may be found in, say, the Wall Street Journal.”
On Measuring and Mitigating Biased Inferences of Word Embeddings (2019, background): “There are also retraining-based mechanisms (e.g., (Zhao et al. 2018)), but given that building word embeddings can be prohibitively expensive, we focus on the much simpler post-hoc modiﬁcations.”
Towards Understanding and Mitigating Social Biases in Language Models (2021, methodology): “Corresponding methods for debiasing these embeddings for both binary (Bolukbasi et al., 2016; Zhao et al., 2018) and multiclass (Manzini et al., 2019) attributes across gender, race, and religion have been devised.”
Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models (2021, methodology): “Indeed, they mostly focus on measuring the assignment of stereotypical pronouns to professions (Zhao et al., 2018a).”
Queens Are Powerful Too: Mitigating Gender Bias in Dialogue Generation (2019, methodology): “Particular attention has been paid to uncovering, analyzing, and removing gender biases in word embeddings (Basta et al., 2019; Kaneko and Bollegala, 2019; Zhao et al., 2019, 2018b; Bolukbasi et al., 2016).”
BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage (2022, methodology): “Bias in gendered word frequency We count the number of female and male gendered words in the BB3 deployment using the list compiled by Zhao et al. (2018).”
Measuring Bias in Contextualized Word Representations (2019, background): “Zhao et al. (2018) discussed systems are biased when the model has strong association between gender and certain professions.”
Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias (2020, methodology): “For consistency with the Winobias analysis, we make the simplifying assumption that the gender stereotypicality of the participant is the opposite of that of the occupation.”
Self-Diagnosis and Self-Debiasing: A Proposal for Reducing Corpus-Based Bias in NLP (2021, methodology): “For static word embeddings, various algorithms for debiasing have been proposed (Bolukbasi et al., 2016; Zhao et al., 2018; Ravfogel et al., 2020; Go-nen and Goldberg, 2019), many of them being based on predeﬁned word lists or other external resources.”
Identifying and Reducing Gender Bias in Word-Level Language Models (2019, methodology): “Font and Costa-Juss`a (2019) study the impact of gender debiasing techniques by Bolukbasi et al. (2016) and Zhao et al. (2018) in machine translation.”
Gender Bias in Contextualized Word Embeddings (2019, methodology): “We explore two different strategies: (1) a training-time data augmentation technique (Zhao et al., 2018a), where we augment the corpus for training the coreference system with its gender-swapped variant (female entities are swapped to male entities and vice versa) and, afterwards, re-train the…”
Mitigating Gender Bias in Natural Language Processing: Literature Review (2019, methodology): “Probing Concept Size Winogender Schemas (Rudinger et al., 2018) Coreference Resolution Occupation 720 English Sentences WinoBias (Zhao et al., 2018a) Coreference Resolution Occupation 3,160 English Sentences GAP (Webster et al., 2018) Coreference Resolution Names 4,454 English Contexts EEC (Kiritche”