Salvato in:
Dettagli Bibliografici
Autori principali: Liu, Guangliang, Chen, Bocheng, Zi, Han, Zhang, Xitong, Johnson, Kristen Marie
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2509.21456
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915627069865984
author Liu, Guangliang
Chen, Bocheng
Zi, Han
Zhang, Xitong
Johnson, Kristen Marie
author_facet Liu, Guangliang
Chen, Bocheng
Zi, Han
Zhang, Xitong
Johnson, Kristen Marie
contents Moral alignment has emerged as a widely adopted approach for regulating the behavior of pretrained language models (PLMs), typically through fine-tuning on curated datasets. Gender stereotype mitigation is a representational task within the broader application of moral alignment. However, this process often comes at the cost of degraded downstream task performance. Prior studies commonly aim to achieve a performance trade-off by encouraging PLMs to selectively forget only stereotypical knowledge through carefully designed fairness objective, while preserving their language modeling capability (overall forgetting). In this short paper, we investigate whether the performance trade-off can be achieved through the lens of forgetting and the fairness objective. Our analysis shows that the large datasets needed for satisfactory fairness highlight the limitations of current fairness objectives in achieving an effective trade-off: (1) downstream task performance is strongly correlated with overall forgetting; (2) selective forgetting reduces stereotypes, but overall forgetting increases. and (3) general solutions for alleviating forgetting are ineffective at reducing the overall forgetting and fail to improve downstream task performance.
format Preprint
id arxiv_https___arxiv_org_abs_2509_21456
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Diagnosing the Performance Trade-off in Moral Alignment: A Case Study on Gender Stereotypes
Liu, Guangliang
Chen, Bocheng
Zi, Han
Zhang, Xitong
Johnson, Kristen Marie
Computation and Language
Moral alignment has emerged as a widely adopted approach for regulating the behavior of pretrained language models (PLMs), typically through fine-tuning on curated datasets. Gender stereotype mitigation is a representational task within the broader application of moral alignment. However, this process often comes at the cost of degraded downstream task performance. Prior studies commonly aim to achieve a performance trade-off by encouraging PLMs to selectively forget only stereotypical knowledge through carefully designed fairness objective, while preserving their language modeling capability (overall forgetting). In this short paper, we investigate whether the performance trade-off can be achieved through the lens of forgetting and the fairness objective. Our analysis shows that the large datasets needed for satisfactory fairness highlight the limitations of current fairness objectives in achieving an effective trade-off: (1) downstream task performance is strongly correlated with overall forgetting; (2) selective forgetting reduces stereotypes, but overall forgetting increases. and (3) general solutions for alleviating forgetting are ineffective at reducing the overall forgetting and fail to improve downstream task performance.
title Diagnosing the Performance Trade-off in Moral Alignment: A Case Study on Gender Stereotypes
topic Computation and Language
url https://arxiv.org/abs/2509.21456