On Measuring and Mitigating Biased Inferences of Word Embeddings

Sunipa Dev, Tao Li, Jeff Phillips, Vivek Srikumar
8/25/2019
cs.CLcs.LG

Official code: sunipa/On-Measuring-and-Mitigating-Biased-Inferences-of-Word-Embeddings (8 stars, last commit 2021-03-17).

Abstract

Word embeddings carry stereotypical connotations from the text they are trained on, which can lead to invalid inferences in downstream models that rely on them. We use this observation to design a mechanism for measuring stereotypes using the task of natural language inference. We demonstrate a reduction in invalid inferences via bias mitigation strategies on static word embeddings (GloVe). Further, we show that for gender bias, these techniques extend to contextualized embeddings when applied selectively only to the static components of contextualized embeddings (ELMo, BERT).

View on arXivView PDF

Code Implementations(1)

sunipa/On-Measuring-and-Mitigating-Biased-Inferences-of-Word-EmbeddingsOfficial100%

Debiasing both contextual and context free embeddings and developing a downstream metric for bias measurement

83PythonNov 18, 20195 years ago
bertcontextual-embeddingsdebiasdebias-word-embeddingelmo+1 more

Similar Papers

Mitigating Gender Bias in Contextual Word Embeddings

89% match2024

Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them

87% match1 repo402019

Understanding the Origins of Bias in Word Embeddings

87% match1 repo222018

Detecting and Mitigating Indirect Stereotypes in Word Embeddings

86% match2023

Profiling Bias in LLMs: Stereotype Dimensions in Contextual Word Embeddings

85% match1 repo12024

Will the Prince Get True Love's Kiss? On the Model Sensitivity to Gender Perturbation over Fairytale Texts

84% match2023

Cite this paper

@article{dev2019measuring,
  title  = {On Measuring and Mitigating Biased Inferences of Word Embeddings},
  author = {Sunipa Dev and Tao Li and Jeff Phillips and Vivek Srikumar},
  year   = {2019},
  eprint = {1908.09369},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1908.09369}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1908.09369.svg)](https://codeofpaper.com/paper/1908.09369)

Discussion