jaddoughman/Gender-Bias-Datasets-Lexicons

Language has a profound impact on our thoughts, perceptions, and conceptions of gender roles. Gender-inclusive language is, therefore, a key tool to promote social inclusion and contribute to achieving gender equality. Consequently, detecting and mitigating gender bias in texts is instrumental in halting its propagation and societal implications. However, there is a lack of gender bias datasets and lexicons for automating the detection of gender bias using supervised and unsupervised machine learning (ML) and natural language processing (NLP) techniques. Therefore, the main contribution of this work is to publicly provide labeled datasets and exhaustive lexicons by collecting, annotating, and augmenting relevant sentences to facilitate the detection of gender bias in English text. Towards this end, we present an updated version of our previously proposed taxonomy by re-formalizing its structure, adding a new bias type, and mapping each bias subtype to an appropriate detection methodology. The released datasets and lexicons span multiple bias subtypes including: Generic He, Generic She, Explicit Marking of Sex, and Gendered Neologisms. We leveraged the use of word embedding models to further augment the collected lexicons. The underlying motivation of our work is to enable the technical community to combat gender bias in text and halt its propagation using ML and NLP techniques.

Gender-Bias-Datasets-Lexicons

Abstract

Language has a profound impact on our thoughts, perceptions, and conceptions of gender roles. Gender-inclusive language is, therefore, a key tool to promote social inclusion and contribute to achieving gender equality. Consequently, detecting and mitigating gender bias in texts is instrumental in halting its propagation and societal implications. However, there is a lack of gender bias datasets and lexicons for automating the detection of gender bias using supervised and unsupervised machine learning (ML) and natural language processing (NLP) techniques. Therefore, the main contribution of this work is to publicly provide labeled datasets and exhaustive lexicons by collecting, annotating, and augmenting relevant sentences to facilitate the detection of gender bias in English text. Towards this end, we present an updated version of our previously proposed taxonomy by re-formalizing its structure, adding a new bias type, and mapping each bias subtype to an appropriate detection methodology. The released datasets and lexicons span multiple bias subtypes including: Generic He, Generic She, Explicit Marking of Sex, and Gendered Neologisms. We leveraged the use of word embedding models to further augment the collected lexicons. The underlying motivation of our work is to enable the technical community to combat gender bias in text and halt its propagation using ML and NLP techniques.

Gender Bias Taxonomy

Type	Subtype	Example	Implication
Generic Pronouns	Generic He	The client should receive his invoice in two weeks.	Biased Mental Imagery
Generic Pronouns	Generic She	A nurse should ensure that she gets adequate rest.	Biased Mental Imagery
Generic Pronouns	Gendered Generic Man	Good teachers know how to man the classroom.	Biased Mental Imagery
Sexism	Hostile Sexism	Women are incompetent at work.	Aggressive Behavior
Sexism	Benevolent Sexism	They’re probably surprised at how smart you are, for a girl.	Representational Harms
Occupational Bias	Gendered Division of Labor	Professors are men and elementary teachers are women.	Labor Force Participation
Occupational Bias	Gendered Roles & Duties	I’ll have my girl get you a cup of coffee.	Labor Force Participation
Exclusionary Bias	Explicit Marking of Sex	Chairman, Businessman, Manpower, Cameraman...	Representational Harms
Exclusionary Bias	Gender-based Neologisms	Man-bread, Man-sip...	Representational Harms
Exclusionary Bias	Gendered Word Ordering	“Men and Women", “Brothers and Sisters"...	Representational Harms
Semantics	Metaphors	“Cookie": lovely woman.	Bias Propagation
Semantics	Gendered Attributes	An unmarried male (bachelor) is a “personal choice”. An unmarried female (spinster) is derogatorily an “old maid".	Bias Propagation
Semantics	Old Sayings	A woman’s tongue three inches long can kill a man six feet high.	Bias Propagation

Citation

@article{doughman2022gender,
  title={Gender Bias in Text: Labeled Datasets and Lexicons},
  author={Doughman, Jad and Khreich, Wael},
  journal={arXiv preprint arXiv:2201.08675},
  year={2022}
}