A robust bias mitigation procedure based on the Stereotype Content model

Eddie Ungless, Amy Rafferty, Hrichika Nag, Björn Ross

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

The Stereotype Content model (SCM) states that we tend to perceive minority groups as cold, incompetent or both. In this paper we adapt existing work to demonstrate that the Stereotype Content model holds for contextualised word embeddings, then use these results to evaluate a fine-tuning process designed to drive a language model away from stereotyped portrayals of minority groups. We find the SCM terms are better able to capture bias than demographic agnostic terms related to pleasantness. Further, we were able to reduce the presence of stereotypes in the model through a simple fine-tuning procedure that required minimal human and computer resources, without harming downstream performance. We present this work as a prototype of a debiasing procedure that aims to remove the need for a priori knowledge of the specifics of bias in the model.
Original languageEnglish
Title of host publicationProceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science
EditorsDavid Bamman, Dirk Hovy, David Jurgens, Katherine Keith, Brendan O'Connor, Svitlana Volkova
PublisherAssociation for Computational Linguistics (ACL)
Pages207-217
Number of pages11
ISBN (Print)978-1-959429-20-3
Publication statusPublished - 2 Feb 2023
EventThe 5th Workshop on Natural Language Processing and Computational Social Science - Abu Dhabi, United Arab Emirates
Duration: 7 Dec 20227 Dec 2022
Conference number: 5
https://sites.google.com/site/nlpandcss/home/nlp-css-at-emnlp-2022

Workshop

WorkshopThe 5th Workshop on Natural Language Processing and Computational Social Science
Abbreviated titleNLP+CSS 2022
Country/TerritoryUnited Arab Emirates
CityAbu Dhabi
Period7/12/227/12/22
Internet address

Fingerprint

Dive into the research topics of 'A robust bias mitigation procedure based on the Stereotype Content model'. Together they form a unique fingerprint.

Cite this