Masahiro, Kaneko and Bollegala, D 
ORCID: 0000-0003-4476-7003
(2019)
Gender-preserving Debiasing for Pre-trained Word Embeddings.
In: Annual Meeting of the Association for Computational Linguistics, 2019-7-28 - 2019-8-2, Florence, Italy.
Abstract
Word embeddings learnt from massive text collections have demonstrated significant levels of discriminative biases such as gender, racial or ethnic biases, which in turn bias the down-stream NLP applications that use those word embeddings. Taking gender-bias as a working example, we propose a debiasing method that preserves non-discriminative gender-related information, while removing stereotypical discriminative gender biases from pre-trained word embeddings. Specifically, we consider four types of information: \emph{feminine}, \emph{masculine}, \emph{gender-neutral} and \emph{stereotypical}, which represent the relationship between gender vs. bias, and propose a debiasing method that (a) preserves the gender-related information in feminine and masculine words, (b) preserves the neutrality in gender-neutral words, and (c) removes the biases from stereotypical words. Experimental results on several previously proposed benchmark datasets show that our proposed method can debias pre-trained word embeddings better than existing SoTA methods proposed for debiasing word embeddings while preserving gender-related but non-discriminative information.
| Item Type: | Conference or Workshop Item (Unspecified) |
|---|---|
| Additional Information: | Accepted as a long paper to the 57th Annual Conference of the Association for Computational Linguistics (ACL-2019) |
| Uncontrolled Keywords: | cs.CL, cs.CL, cs.LG |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 02 Sep 2019 08:29 |
| Last Modified: | 19 Jan 2023 00:28 |
| Open Access URL: | https://www.aclweb.org/anthology/P19-1160/ |
| Related URLs: | |
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3053018 |
Altmetric
Altmetric
