Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings

Online texts -- across genres, registers, domains, and styles -- are riddled with human stereotypes, expressed in overt or subtle ways. Word embeddings, trained on these texts, perpetuate and amplify these stereotypes, and propagate biases to machine learning models that use word embeddings as featu...

Full description

Saved in:
Bibliographic Details
Published inarXiv.org
Main Authors Manzini, Thomas, Yao Chong Lim, Tsvetkov, Yulia, Black, Alan W
Format Paper
LanguageEnglish
Published Ithaca Cornell University Library, arXiv.org 02.07.2019
Subjects
Crime
Domains
Machine learning
Police
Religion
Stereotypes
Texts
Online AccessGet full text

Cover

More Information
Summary:Online texts -- across genres, registers, domains, and styles -- are riddled with human stereotypes, expressed in overt or subtle ways. Word embeddings, trained on these texts, perpetuate and amplify these stereotypes, and propagate biases to machine learning models that use word embeddings as features. In this work, we propose a method to debias word embeddings in multiclass settings such as race and religion, extending the work of (Bolukbasi et al., 2016) from the binary setting, such as binary gender. Next, we propose a novel methodology for the evaluation of multiclass debiasing. We demonstrate that our multiclass debiasing is robust and maintains the efficacy in standard NLP tasks.
ISSN:2331-8422