Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation

By Tianlu Wang, Xi Victoria Lin, Nazneen Rajani, Vicente Ordonez, Caimng Xiong. 2020

67 citations recorded by the source corpus.

Word embeddings derived from human-generated corpora inherit strong gender bias which can be further amplified by downstream models. Some commonly adopted debiasing approaches, including the seminal Hard Debias algorithm, apply post-processing procedures that project pre-trained word embeddings into a subspace orthogonal to an inferred gender subspace. We discover that semantic-agnostic corpus regularities such as word frequency captured by the word embeddings negatively impact the performance of these algorithms. We propose a simple but effective technique, Double Hard Debias, which purifies the word embeddings against such corpus regularities prior to inferring and removing the gender subspace. Experiments on three bias mitigation benchmarks show that our approach preserves the distributional semantics of the pre-trained word embeddings while reducing gender bias to a significantly larger degree than prior approaches.

Subfields: Artificial Intelligence, Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

Language (Technology) is Power: A Critical Survey of “Bias” in NLP (2020, neutral): “…(2019); Prates et al. (2019); Prost et al. (2019); Qian et al. (2019); Swinger et al. (2019); Zhao et al. (2019); Zhou et al. (2019); Ethayarajh (2020); Huang et al. (2020); Jia et al. (2020); Popović et al. (2020); Pryzant et al. (2020); Vig et al. (2020); Wang et al. (2020); Zhao et al. (2020).”
Queens Are Powerful Too: Mitigating Gender Bias in Dialogue Generation (2019, methodology): “…employ (i.e., CDA), and the other of which extends to sentences a word-embedding post-processing method (Bolukbasi et al., 2016) that has been shown to be ine ﬀ ective at removing gender bias (Gonen and Goldberg 2019, but see Wang et al. 2020 for a more recent, perhaps more e ﬀ ective attempt).”
The Birth of Bias: A case study on the evolution of gender bias in an English language model (2022, background): “…(Zhang et al., 2018; Zhao et al., 2018b; Liu and Avci, 2019), or on measuring and ﬁxing biases in word embeddings or internal states of language models (Bolukbasi et al., 2016; Ethayarajh et al., 2019; Wang et al., 2020; Basta et al., 2019; May et al., 2019; Kurita et al., 2019; Tan et al., 2021).”
Theories of “Gender” in NLP Bias Research (2022, background): “pdf [153] Tianlu Wang, Xi Victoria Lin, Nazneen Fatema Rajani, Bryan McCann, Vicente Ordonez, and Caiming Xiong.”
Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models (2024, unknown): “…et al., 2022; Orgad & Belinkov, 2023), training classifiers with more favorable inductive biases (Kirichenko et al., 2023; Zhang et al., 2022; Iskander et al., 2024), or editing out undesired concepts (Iskander et al., 2023; Belrose et al., 2023; Wang et al., 2020; Ravfogel et al., 2020; 2022a;b).”
Can Existing Methods Debias Languages Other than English? First Attempt to Analyze and Mitigate Japanese Word Embeddings (2020, methodology): “According to Wang et al. (2020) Table 7: Experimental results on the neighborhood metric in the case of “women” and “men”.”