Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings
This addresses the issue of perpetuating stereotypes in machine learning models for NLP applications, though it is incremental as it builds on prior binary debiasing work.
The paper tackles the problem of multiclass bias in word embeddings, such as race and religion, by extending binary debiasing methods and proposing a novel evaluation methodology, showing that their approach is robust and maintains performance in standard NLP tasks.
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.