Measuring Fairness with Biased Rulers: A Survey on Quantifying Biases in Pretrained Language Models

By Pieter Delobelle, E. Tokpo, Toon Calders, Bettina Berendt. 2021

27 citations recorded by the source corpus.

An increasing awareness of biased patterns in natural language processing resources, like BERT, has motivated many metrics to quantify `bias' and `fairness'. But comparing the results of different metrics and the works that evaluate with such metrics remains difficult, if not outright impossible. We survey the existing literature on fairness metrics for pretrained language models and experimentally evaluate compatibility, including both biases in language models as in their downstream tasks. We do this by a mixture of traditional literature survey and correlation analysis, as well as by running empirical evaluations. We find that many metrics are not compatible and highly depend on (i) templates, (ii) attribute and target seeds and (iii) the choice of embeddings. These results indicate that fairness or bias evaluation remains challenging for contextualized language models, if not at least highly subjective. To improve future comparisons and fairness evaluations, we recommend avoiding embedding-based metrics and focusing on fairness evaluations in downstream tasks.

Subfields: Artificial Intelligence, Safety Research, 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 Safety Research

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

“I’m sorry to hear that”: Finding New Biases in Language Models with a Holistic Descriptor Dataset (2022, background): “…love or favoritism towards the descriptor in question, perhaps reflecting the polarizing nature of the descriptors; by contrast, templates reflecting curiosity of or identity with specific descriptors have less variance, perhaps because they contain fewer content words (Delobelle et al., 2021).”
Perturbation Augmentation for Fairer NLP (2022, background): “Many of the standard fairness metrics in NLP have flaws (Blodgett et al., 2021), and often different fairness metrics fail to agree (Delobelle et al., 2021; Goldfarb-Tarrant et al., 2021; Cao et al., 2022).”
"I'm sorry to hear that": finding bias in language models with a holistic descriptor dataset (2022, background): “…love or favoritism towards the descriptor in question, perhaps reﬂecting the polarizing nature of the descriptors; by contrast, template types reﬂecting curiosity of or identity with speciﬁc descriptors have less variance, perhaps because they contain fewer content words (Delobelle et al., 2021).”