Unsupervised Discovery of Gendered Language through Latent-Variable Modeling

By Alexander Miserlis Hoyle, Lawrence Wolf-Sonkin, Hanna M. Wallach, Isabelle Augenstein, Ryan Cotterell. 2019

56 citations recorded by the source corpus.

Studying the ways in which language is gendered has long been an area of interest in sociolinguistics. Studies have explored, for example, the speech of male and female characters in film and the language used to describe male and female politicians. In this paper, we aim not to merely study this phenomenon qualitatively, but instead to quantify the degree to which the language used to describe men and women is different and, moreover, different in a positive or negative way. To that end, we introduce a generative latent-variable model that jointly represents adjective (or verb) choice, with its sentiment, given the natural gender of a head (or dependent) noun. We find that there are significant differences between descriptions of male and female nouns and that these differences align with common gender stereotypes: Positive adjectives used to describe women are more often related to their bodies than adjectives used to describe men.

Subfields: Artificial Intelligence, Gender Studies, Natural Language Processing, 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 Gender Studies

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

Based on billions of words on the internet, people = men (2022, background): “…was taken from prior work that, in addition to identifying the syntactic gender bias of each verb, indicated the valence (i.e., sentiment) of the verb as positive, negative, or neutral and indicated whether the verb’s gender bias occurred with arguments in the subject or object position ( 51 ).”
Queens Are Powerful Too: Mitigating Gender Bias in Dialogue Generation (2019, methodology): “Nouns and adjectives are binned into gendered bins via an aggregation of existing gendered word lists (Zhao et al., 2018b,a; Hoyle et al., 2019).”
Multi-Dimensional Gender Bias Classification (2020, background): “For example, adjectives which describe women have been shown to differ from those used to describe men in numerous situations (Trix and Psenka, 2003; Gaucher et al., 2011; Moon, 2014; Hoyle et al., 2019), as do verbs that take nouns referring to men”
Investigating Failures of Automatic Translation in the Case of Unambiguous Gender (2021, methodology): “Verbs were automatically discovered from Wikipedia using dependency parses to find verbs that take women or men preferentially highly as subjects or direct objects (Hoyle et al., 2019).”
Theories of “Gender” in NLP Bias Research (2022, background): “[62] Alexander Hoyle, Wolf-Sonkin, Hanna Wallach, Isabelle Augenstein, and Ryan Cotterell.”
Adhering, Steering, and Queering: Treatment of Gender in Natural Language Generation (2020, background): “Similar differences are identiﬁed in how people talk about men and women (e.g., [12,32,51,69]).”
Unsupervised Discovery of Implicit Gender Bias (2020, background): “Less-supervised approaches focus on corpuslevel analyses, such as associations between gendered terms and occupational stereotypes (Wagner et al., 2015; Bolukbasi et al., 2016; Fu et al., 2016; Joseph et al., 2017; Nakandala et al., 2017; Friedman et al., 2019; Chaloner and Maldonado, 2019; Hoyle et”
A Survey on Gender Bias in Natural Language Processing (2021, methodology): “Differences in depictions of men and women have been prolificly quantified using point-wise mutual information (PMI) [Hoyle et al. 2019; Rudinger et al. 2017; Stańczak et al. 2021].”
[RE] Bad Seeds: Evaluating Lexical Methods for Bias Measurement (2022, methodology): “We use the same seed sets as used in the experiments for the original paper (Bolukbasi et al., 2016) (Kozlowski et al., 2019) (Garg et al., 2018) (Caliskan et al., 2017) (Manzini et al., 2019) (Zhao et al., 2018) (Hoyle et al., 2019).”
Gender and Representation Bias in GPT-3 Generated Stories (2021, background): “Even so, depictions of women still foreground their physical appearances (Hoyle et al., 2019), and por-tray them as weak and less powerful (Fast et al., 2016b; Sap et al., 2017).”