Gender Bias in Contextualized Word Embeddings
This addresses gender bias in NLP systems for fairness and equity, though it is incremental as it builds on existing bias analysis and mitigation work.
The paper quantified gender bias in ELMo's contextualized word embeddings, finding systematic encoding of gender information and male-female disparities, and showed that a coreference system using ELMo inherits this bias with significant effects on WinoBias. It then mitigated the bias using two methods, eliminating it on WinoBias.
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.