[RE] Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation

TL;DR

This paper introduces a method to reduce gender bias in word embeddings by tailoring them specifically for bias mitigation, aiming to improve fairness in NLP applications.

Contribution

It presents a novel approach called Double-Hard Debias that effectively diminishes gender bias in word embeddings without sacrificing their semantic quality.

Findings

Significant reduction in gender bias metrics in embeddings.

Preservation of semantic relationships post-debiasing.

Improved fairness in downstream NLP tasks.

Abstract

Despite widespread use in natural language processing (NLP) tasks, word embeddings have been criticized for inheriting unintended gender bias from training corpora. programmer is more closely associated with man and homemaker is more closely associated with woman. Such gender bias has also been shown to propagate in downstream tasks.