Learning Bias-reduced Word Embeddings Using Dictionary Definitions

An, H.; Liu, Xiaojiang; Zhang, Donald

doi:10.18653/v1/2022.findings-acl.90

Cited by 4 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bolukbasi et al (2016); Gonen and Goldberg (2019)). Most of the following work focused on technical subtleties about metrics, extensions to other languages or contexts, application to language models, evaluation on downstream tasks or automating the whole process, from assessment to mitigation (Guo and Caliskan, 2021;Guo et al, 2022;An et al, 2022;Kaneko and Bollegala, 2021).…”

Section: A Critical Perspective On Methods For Bias Assessmentmentioning

confidence: 99%

Bias assessment for experts in discrimination, not in computer science

Alonso Alemany,

Benotti,

Maina

et al. 2023

Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP)

View full text Add to dashboard Cite

Approaches to bias assessment usually require such technical skills that, by design, they leave discrimination experts out. In this paper we present EDIA, a tool that facilitates that experts in discrimination explore social biases in word embeddings and masked language models. Experts can then characterize those biases so that their presence can be assessed more systematically, and actions can be planned to address them. They can work interactively to assess the effects of different characterizations of bias in a given word embedding or language model, which helps to specify informal intuitions in concrete resources for systematic testing.

show abstract

Section: A Critical Perspective On Methods For Bias Assessmentmentioning

confidence: 99%

Bias assessment for experts in discrimination, not in computer science

Alonso Alemany,

Benotti,

Maina

et al. 2023

Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP)

View full text Add to dashboard Cite

show abstract

“…Linear projection [21] Hard debias [9] Interactive null space projection [32] Subspace orthogonal word embedding [1] Else Hard debias, gender swap, and bias fine-tuning [31] Iterative adversarial disentanglement [17] Dictionary definitions leverage based train-time debiasing [4] Sentence embedding…”

Section: Projectionmentioning

confidence: 99%

Adversarial Discriminator to Mitigate Gender Bias in Abusive Language Detection

Park,

Cho

2023

Frontiers in Artificial Intelligence and Applications

View full text Add to dashboard Cite

Abusive language detection models tend to have a gender bias problem in which the model is biased towards sentences containing identity words of specific gender groups. Previous studies to reduce bias, such as projection methods, tend to lose information in word vectors and sentence context, resulting in low detection accuracy. This paper proposes a novel method that mitigates gender bias while preserving original information by regularizing sentence embedding vectors based on information theory. Latent vectors generated by an autoencoder are debiased through dual regularization using a gender discriminator, an abuse classifier, and a decoder. While the gender discriminator labels are randomized, the discriminator confuses the gender feature, and the classifier retains the abuse information. Latent vectors are regularized through information theoretic adversarial optimization that disentangles and mitigates gender features. We show that the proposed method successfully orthogonalizes the direction of the correlated information and reduces the gender feature through calculation of subspaces and embedding vector visualization. Moreover, the proposed method maintains the highest accuracy among the four state-of-the-art bias mitigation methods and shows superior performance in reducing gender bias in four different Twitter datasets for abusive language detection.

show abstract

“…To quantify the separation of clusters, we conduct a binary classification of the SR vectors for each pair of name groups that are associated with different demographic attributes. This evaluation is similarly used by Gonen and Goldberg (2019); An et al (2022). For two name groups representing different social groups (e.g., AA female names and EA female names), we use the classical KMeans algorithm (K = 2) to cluster the SR vectors and make a binary prediction that indicates the membership of either cluster.…”

Section: Success Rate (Sr)mentioning

confidence: 99%

SODAPOP: Open-Ended Discovery of Social Biases in Social Commonsense Reasoning Models

An,

Li,

Zhao

et al. 2023

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

A common limitation of diagnostic tests for detecting social biases in NLP models is that they may only detect stereotypic associations that are pre-specified by the designer of the test. Since enumerating all possible problematic associations is infeasible, it is likely these tests fail to detect biases that are present in a model but not pre-specified by the designer. To address this limitation, we propose SODAPOP 1 (SOcial bias Discovery from Answers about PeOPle), an approach for automatic social bias discovery in social commonsense question-answering. The SODAPOP pipeline generates modified instances from the Social IQa dataset (Sap et al., 2019b) by ( 1) substituting names associated with different demographic groups, and (2) generating many distractor answers from a masked language model. By using a social commonsense model to score the generated distractors, we are able to uncover the model's stereotypic associations between demographic groups and an open set of words. We also test SODAPOP on debiased models and show the limitations of multiple state-of-the-art debiasing algorithms.

show abstract

Learning Bias-reduced Word Embeddings Using Dictionary Definitions

Cited by 4 publications

References 0 publications

Bias assessment for experts in discrimination, not in computer science

Bias assessment for experts in discrimination, not in computer science

Adversarial Discriminator to Mitigate Gender Bias in Abusive Language Detection

SODAPOP: Open-Ended Discovery of Social Biases in Social Commonsense Reasoning Models

Contact Info

Product

Resources

About