Gender Bias in Contextualized Word Embeddings

TL;DR

This paper investigates gender bias in ELMo's contextualized word embeddings, revealing data-driven biases and proposing mitigation strategies to reduce such biases in downstream tasks.

Contribution

The study provides a comprehensive analysis of gender bias in ELMo embeddings and introduces methods to effectively mitigate this bias.

Findings

ELMo training data contains more male than female entities.

ELMo embeddings encode gender information systematically.

Bias in ELMo affects coreference systems and can be mitigated.

Abstract

In this paper, we quantify, analyze and mitigate gender bias exhibited in ELMo's contextualized word vectors. First, we conduct several intrinsic analyses and find that (1) training data for ELMo contains significantly more male than female entities, (2) the trained ELMo embeddings systematically encode gender information and (3) ELMo unequally encodes gender information about male and female entities. Then, we show that a state-of-the-art coreference system that depends on ELMo inherits its bias and demonstrates significant bias on the WinoBias probing corpus. Finally, we explore two methods to mitigate such gender bias and show that the bias demonstrated on WinoBias can be eliminated.