Debiasing Pretrained Text Encoders by Paying Attention to Paying Attention - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Debiasing Pretrained Text Encoders by Paying Attention to Paying Attention

Résumé

Natural Language Processing (NLP) models are found to exhibit discriminatory stereotypes across many social constructs, e.g. gender and race. In comparison to the progress made in reducing bias from static word embeddings, fairness in sentence-level text encoders received little consideration despite their wider applicability in contemporary NLP tasks. In this paper, we propose a debiasing method for pretrained text encoders that both reduces social stereotypes, and inflicts next to no semantic damage. Unlike previous studies that directly manipulate the embeddings, we suggest to dive deeper into the operation of these encoders, and pay more attention to the way they pay attention to different social groups. We find that stereotypes are also encoded in the attention layer. Then, we work on model debiasing by redistributing the attention scores of a text encoder such that it forgets any preference to historically advantaged groups, and attends to all social classes with the same intensity. Our experiments confirm that reducing bias from attention effectively mitigates it from the model's text representations.
Fichier principal
Vignette du fichier
EMNLP_2022___Debiasing_Pretrained_Text_Encoders_by_Paying_Attention_to_Paying_Attention.pdf (740.68 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03919992 , version 1 (24-01-2023)

Identifiants

  • HAL Id : hal-03919992 , version 1

Citer

Yacine Gaci, Boualem Benattallah, Fabio Casati, Khalid Benabdeslem. Debiasing Pretrained Text Encoders by Paying Attention to Paying Attention. 2022 Conference on Empirical Methods in Natural Language Processing, Dec 2022, Abu Dhabi, United Arab Emirates. pp.9582-9602. ⟨hal-03919992⟩
172 Consultations
98 Téléchargements

Partager

Gmail Facebook X LinkedIn More