Word embeddings quantify 100 years of gender and ethnic stereotypes
By Nikhil Garg, Londa Schiebinger, Dan Jurafsky, James Y. Zou. 2017
1,107 citations recorded by the source corpus.
Significance Word embeddings are a popular machine-learning method that represents each English word by a vector, such that the geometry between these vectors captures semantic relations between the corresponding words. We demonstrate that word embeddings can be used as a powerful tool to quantify historical trends and social change. As specific applications, we develop metrics based on word embeddings to characterize how gender stereotypes and attitudes toward ethnic minorities in the United States evolved during the 20th and 21st centuries starting from 1910. Our framework opens up a fruitful intersection between machine learning and quantitative social science. Word embeddings are a powerful machine-learning framework that represents each English word by a vector. The geometric relationship between these vectors captures meaningful semantic relationships between the corresponding words. In this paper, we develop a framework to demonstrate how the temporal dynamics of the embedding helps to quantify changes in stereotypes and attitudes toward women and ethnic minorities in the 20th and 21st centuries in the United States. We integrate word embeddings trained on 100 y of text data with the US Census to show that changes in the embedding track closely with demographic and occupation shifts over time. The embedding captures societal shifts—e.g., the women’s movement in the 1960s and Asian immigration into the United States—and also illuminates how specific adjectives and occupations became more closely associated with certain populations over time. Our framework for temporal analysis of word embedding opens up a fruitful intersection between machine learning and quantitative social science.
Subfields: Cultural Studies, General Social Sciences, Statistical and Nonlinear Physics, 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 Cultural Studies
- Lyle Ungar
- Rasmus Précenth
- Yulia Tsvetkov
- Michal Kosinski
- Isabelle Augenstein
- Karolina Stańczak
- João Sedoc
- Hanna M. Wallach
In General Social Sciences
- Yulia Tsvetkov
- Isabelle Augenstein
- Karolina Stańczak
- Rasmus Précenth
- Ruslan Salakhutdinov
- Michal Kosinski
- Lyle Ungar
- Hanna M. Wallach
In Statistical and Nonlinear Physics
- Rasmus Précenth
- Yulia Tsvetkov
- Michal Kosinski
- Lyle Ungar
- William Yang Wang
- Isabelle Augenstein
- Karolina Stańczak
- Hanna M. Wallach
In Natural Language Processing
- Mohit Bansal
- Yejin Choi
- Douwe Kiela
- Noah A. Smith
- Percy Liang
- Adina Williams
- Samuel R. Bowman
- Jason Aaron Edward Weston
In Safety, Robustness & Interpretability
- Adina Williams
- Yulia Tsvetkov
- Kai-Wei Chang
- Dirk Hovy
- Sunipa Dev
- Eric Michael Smith
- Douwe Kiela
- Isabelle Augenstein
Citations in the corpus
- Language (Technology) is Power: A Critical Survey of “Bias” in NLP (2020, methodology): “We also discarded all papers using “bias” in NLP systems to measure social “bias” in text or the real world (e.g., Garg et al., 2018).”
- Fair Is Better than Sensational: Man Is to Doctor as Woman Is to Doctor (2019, background): “On the contrary: Embeddings do encode human biases (Caliskan, Bryson, and Narayanan 2017; Garg et al. 2018; Kozlowski, Taddy, and Evans 2019; Gonen and Goldberg 2019), and we agree that this issue deserves the full attention of the field (Hovy and Spruit 2016).”
- Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them (2019, background): “race and gender) that naturally occur in the data used to train them (Caliskan et al., 2017; Garg et al., 2018).”
- Gender Bias in Coreference Resolution (2018, background): “A number of papers explore (gender) bias in English word embeddings: how they capture implicit human biases in modern (Caliskan et al., 2017) and historical (Garg et al., 2018) text, and methods for debiasing them (Bolukbasi et al., 2016).”
- Social Biases in NLP Models as Barriers for Persons with Disabilities (2020, background): “…studies have studied unintended biases in NLP models against other historically marginalized groups (Bolukbasi et al., 2016; Caliskan et al., 2017; Garg et al., 2017; Barocas et al., 2017; Garg et al., 2019; Dixon et al., 2018; Noble, 2018; Manzini et al., 2019; Sap et al., 2019; May et al.,…”
- Measuring Bias in Contextualized Word Representations (2019, background): “, 2014), have been shown to exhibit social biases present in human-generated training data (Bolukbasi et al., 2016; Caliskan et al., 2017; Garg et al., 2018; Manzini et al., 2019).”
- On Measuring Social Biases in Sentence Encoders (2019, background): “However, prominent word embeddings such as word2vec (Mikolov et al., 2013) and GloVe (Pennington et al., 2014) encode systematic biases against women and black people (Bolukbasi et al., 2016; Garg et al., 2018, i.a.), implicating many NLP systems in scaling up social injustice.”
- Learning Gender-Neutral Word Embeddings (2018, background): “A temporal analysis about word embeddings (Garg et al., 2018) captures changes in gender and ethnic stereotypes over time.”
- Racial Bias in Hate Speech and Abusive Language Detection Datasets (2019, background): “For example, in unsupervised learning, word embed-dings often contain biases (Bolukbasi et al., 2016; Caliskan et al., 2017; Garg et al., 2018) which persist even after attempts to remove them (Go-nen and Goldberg, 2019).”
- Mitigating Gender Bias in Natural Language Processing: Literature Review (2019, background): “Finally, the authors demonstrate a positive correlation between the strength of association of an occupation word embedding with the female gender and the percentage of females in that occupation in United States, with the percentages taken from Bureau of Labor Statistics labor force participation d”
- Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings (2019, methodology): “Past work has shown that it is possible to linearly separate multiple social classes based on components of word embeddings (Garg et al., 2018).”
- Data and its (dis)contents: A survey of dataset development and use in machine learning research (2020, background): “…in natural language processing datasets frequently reflect social biases and stereotypes relating to race, gender, (dis)ability, and more [Garg et al., 2018, Hutchinson et al., 2020] and correlations between gender and activities depicted in computer vision datasets have been shown to…”
- Multi-Dimensional Gender Bias Classification (2020, background): “In particular, NLP models often learn biases against others based on their gender (Bolukbasi et al., 2016; Hovy and Spruit, 2016; Caliskan et al., 2017; Rudinger et al., 2017; Garg et al., 2018; Gonen and Goldberg, 2019; Dinan et al., 2019a).”
- Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting (2019, background): “Recent work has studied the ways in which stereotypes and other human biases may be reflected in semantic representations such as word embeddings [10, 13, 21].”
- Lessons from archives: strategies for collecting sociocultural data in machine learning (2019, background): “For instance, [19] shows trained word embeddings replicating Asian stereotypes in language data from historical Google Books and the Corpus of Historical American English (COHA).”
- Challenges and frontiers in abusive content detection (2019, background): “This feeds into broader research which shows that computational methods can encode and reinforce social biases – even when they are meant to ameliorate them (Garg, Schiebinger, Jurafsky, & Zou, 2017).”
- Towards Accountability for Machine Learning Datasets: Practices from Software Engineering and Infrastructure (2020, background): “icism in recent years due to their tendency to perpetuate and amplify social inequality [13, 44]. Data is frequently identified as a key source of these failures through its role in “bias-laundering” [40, 51, 54, 118, 124]. For example, recent studies have uncovered widespread prevalence of undesira”
- Counterfactual Data Augmentation for Mitigating Gender Stereotypes in Languages with Rich Morphology (2019, background): “One type of societal bias that has received considerable attention from the NLP community is gender stereotyping (Garg et al., 2017; Rudinger et al., 2017; Sutton et al., 2018).”
- Interpretations are useful: penalizing explanations to align neural networks with prior knowledge (2019, background): “Moreover, DNNs are also known to exploit bias regarding gender, race, and other sensitive attributes present in training datasets (Garg et al., 2018; Obermeyer et al., 2019; Dressel & Farid, 2018).”