Understanding the Origins of Bias in Word Embeddings

TL;DR

This paper introduces a method to trace and understand the origins of bias in word embeddings by analyzing how changes in training data influence bias, aiding in bias mitigation.

Contribution

The authors develop a technique using influence functions to identify how perturbations in training data affect bias in word embeddings, providing insights into bias sources.

Findings

Influence function-based approximations are highly accurate.

Bias in embeddings can be traced back to specific training documents.

Removing certain documents reduces bias significantly.

Abstract

The power of machine learning systems not only promises great technical progress, but risks societal harm. As a recent example, researchers have shown that popular word embedding algorithms exhibit stereotypical biases, such as gender bias. The widespread use of these algorithms in machine learning systems, from automated translation services to curriculum vitae scanners, can amplify stereotypes in important contexts. Although methods have been developed to measure these biases and alter word embeddings to mitigate their biased representations, there is a lack of understanding in how word embedding bias depends on the training data. In this work, we develop a technique for understanding the origins of bias in word embeddings. Given a word embedding trained on a corpus, our method identifies how perturbing the corpus will affect the bias of the resulting embedding. This can be used to trace the origins of word embedding bias back to the original training documents. Using our method, one can investigate trends in the bias of the underlying corpus and identify subsets of documents whose removal would most reduce bias. We demonstrate our techniques on both a New York Times and Wikipedia corpus and find that our influence function-based approximations are very accurate.