The Impact of Debiasing on the Performance of Language Models in Downstream Tasks is Underestimated

Kaneko, M, Bollegala, D ORCID: 0000-0003-4476-7003 and Okazaki, N (2023) The Impact of Debiasing on the Performance of Language Models in Downstream Tasks is Underestimated In: Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers), 2023-11 - 2023-11, Bali, Indonesia.

PDF IJCNLP_AACL2023__task_perfomance_of_debiased_lm.pdf - Author Accepted Manuscript Download (713kB) | Preview

Abstract

Pre-trained language models trained on large-scale data have learned serious levels of social biases. Consequently, various methods have been proposed to debias pre-trained models. Debiasing methods need to mitigate only discriminatory bias information from the pretrained models, while retaining information that is useful for the downstream tasks. In previous research, whether useful information is retained has been confirmed by the performance of downstream tasks in debiased pretrained models. On the other hand, it is not clear whether these benchmarks consist of data pertaining to social biases and are appropriate for investigating the impact of debiasing. For example in gender-related social biases, data containing female words (e.g. “she, female, woman”), male words (e.g. “he, male, man”), and stereotypical words (e.g. “nurse, doctor, professor”) are considered to be the most affected by debiasing. If there is not much data containing these words in a benchmark dataset for a target task, there is the possibility of erroneously evaluating the effects of debiasing. In this study, we compare the impact of debiasing on performance across multiple downstream tasks using a wide-range of benchmark datasets that containing female, male, and stereotypical words. Experiments show that the effects of debiasing are consistently underestimated across all tasks. Moreover, the effects of debiasing could be reliably evaluated by separately considering instances containing female, male, and stereotypical words than all of the instances in a benchmark dataset.

Item Type:	Conference Item (Unspecified)
Divisions:	Faculty of Science & Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User:	Symplectic Admin
Date Deposited:	09 Nov 2023 09:17
Last Modified:	23 May 2026 08:10
DOI:	10.18653/v1/2023.ijcnlp-short.4
Related Websites:	Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3176698
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.