Sociolectal Analysis of Pretrained Language Models

Sheng Zhang, Xin Zhang, Weiming Zhang, Anders Søgaard


Abstract
Using data from English cloze tests, in which subjects also self-reported their gender, age, education, and race, we examine performance differences of pretrained language models across demographic groups, defined by these (protected) attributes. We demonstrate wide performance gaps across demographic groups and show that pretrained language models systematically disfavor young non-white male speakers; i.e., not only do pretrained language models learn social biases (stereotypical associations) – pretrained language models also learn sociolectal biases, learning to speak more like some than like others. We show, however, that, with the exception of BERT models, larger pretrained language models reduce some the performance gaps between majority and minority groups.
Anthology ID:
2021.emnlp-main.375
Volume:
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
Month:
November
Year:
2021
Address:
Online and Punta Cana, Dominican Republic
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
4581–4588
Language:
URL:
https://aclanthology.org/2021.emnlp-main.375
DOI:
Bibkey:
Copy Citation:
PDF:
https://aclanthology.org/2021.emnlp-main.375.pdf