OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings

By Sunipa Dev, Tao Li, Jeff M. Phillips, Vivek Srikumar. 2020

60 citations recorded by the source corpus.

Language representations are known to carry stereotypical biases and, as a result, lead to biased predictions in downstream tasks. While existing methods are effective at mitigating biases by linear projection, such methods are too aggressive: they not only remove bias, but also erase valuable information from word embeddings. We develop new measures for evaluating specific information retention that demonstrate the tradeoff between bias removal and information retention. To address this challenge, we propose OSCaR (Orthogonal Subspace Correction and Rectification), a bias-mitigating method that focuses on disentangling biased associations between concepts instead of removing concepts wholesale. Our experiments on gender biases show that OSCaR is a well-balanced approach that ensures that semantic information is retained in the embeddings and bias is also effectively mitigated.

Subfields: Artificial Intelligence

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

Citations in the corpus

Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologies (2021, neutral): “A bulk of social bias studies on language models have focused on binary gender and the stereotypes associated with masculine and feminine attributes (Bolukbasi et al., 2016; Webster et al., 2018; Dev et al., 2020b).”
UNQOVERing Stereotypical Biases via Underspecified Questions (2020, methodology): “These biases are subsequently passed on to downstream tasks such as co-reference resolution (Rudinger et al., 2018; Zhao et al., 2018), textual entailment (Dev et al., 2020a), and translation (Stanovsky et al., 2019).”
On Measures of Biases and Harms in NLP (2021, background): “…et al., 2021), detect (Bolukbasi et al., 2016; Nangia et al., 2020), measure (Caliskan et al., 2017; Zhao et al., 2019; Webster et al., 2018; Li et al., 2020), and mitigate (Dev and Phillips, 2019; Ravfogel et al., 2020; Sun et al., 2019; Dev et al., 2021a) the social biases encoded by NLP models.”
The Tail Wagging the Dog: Dataset Construction Biases of Social Bias Benchmarks (2022, background): “…Nadeem et al., 2021) and expressed in specific tasks (Nangia et al., 2020; Kirk et al., 2021; Schramowski et al., 2022; Prabhumoye et al., 2021; Srinivasan and Bisk, 2021; Kirk et al., 2021; Parrish et al., 2021; Baldini et al., 2022; Czarnowska et al., 2021; Dev et al., 2021a; Zhao et al., 2021).”
Bias and Fairness in Large Language Models: A Survey (2023, background): “WEAT* (Dev et al., 2021) presents an alternative, where W1 and W2 are instead defintionally masculine and feminine words (e.”