Sign In

Communications of the ACM

ACM TechNews

Researchers Improve Method for Removing Gender Bias in NLP

two circuitry faces in profile, illustration

Credit: Unsplash

Researchers at the University of Alberta developed a methodology that eliminates gender bias from text while retaining other critical contextual data in natural language processing models.

A process called word embedding converts words into numbers that researchers can plot on a graph and visualize their relationships to one another, in order to better measure gender bias and determine if it was eliminated.

Associate professor Bei Jiang said debiasing also often reduces or removes semantic information that could be important in future tasks involving the word embeddings.

The new method preserves semantic information, and also outperformed leading debiasing techniques in various tasks that were assessed according to word embedding.

From University of Alberta
View Full Article


Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account