MDR Cluster-Debias: A Nonlinear Word Embedding Debiasing Pipeline

Published in International Conference on Social Computing, Behavioral-Cultural Modeling, & Prediction and Behavior Representation in Modeling and Simulation , 2020

PDF Code.

Abstract:Existing methods for debiasing word embeddings often do so only superficially, in that words that are stereotypically associated with, e.g., a particular gender in the original embedding space can still be clustered together in the debiased space. However, there has yet to be a study that explores why this residual clustering exists, and how it might be addressed. The present work fills this gap. We identify two potential reasons for which residual bias exists and develop a new pipeline, MDR Cluster-Debias, to mitigate this bias. We explore the strengths and weaknesses of our method, finding that it significantly outperforms other existing debiasing approaches on a variety of upstream bias tests but achieves limited improvement on decreasing gender bias in a downstream task. This indicates that word embeddings encode gender bias in still other ways, not necessarily captured by upstream tests.

Recommended citation:

@misc{du2020mdr,
    title={MDR Cluster-Debias: A Nonlinear WordEmbedding Debiasing Pipeline},
    author={Yuhao Du and Kenneth Joseph},
    year={2020},
    eprint={2006.11642},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}