Gender Bias in Meta-Embeddings

TL;DR

This paper investigates how gender bias in source embeddings affects meta-embeddings, revealing amplification of biases and proposing a novel debiasing method to mitigate this issue.

Contribution

It is the first study to analyze gender bias amplification in meta-embeddings and introduces a new debiasing approach based on meta-embedding learning.

Findings

Meta-embeddings amplify gender biases compared to source embeddings.

Debiasing both sources and meta-embeddings is necessary to reduce bias.

A novel debiasing method using multiple debiasing techniques on a single source improves bias mitigation.

Abstract

Different methods have been proposed to develop meta-embeddings from a given set of source embeddings. However, the source embeddings can contain unfair gender-related biases, and how these influence the meta-embeddings has not been studied yet. We study the gender bias in meta-embeddings created under three different settings: (1) meta-embedding multiple sources without performing any debiasing (Multi-Source No-Debiasing), (2) meta-embedding multiple sources debiased by a single method (Multi-Source Single-Debiasing), and (3) meta-embedding a single source debiased by different methods (Single-Source Multi-Debiasing). Our experimental results show that meta-embedding amplifies the gender biases compared to input source embeddings. We find that debiasing not only the sources but also their meta-embedding is needed to mitigate those biases. Moreover, we propose a novel debiasing method based on meta-embedding learning where we use multiple debiasing methods on a single source embedding and then create a single unbiased meta-embedding.