Lawyers are Dishonest? Quantifying Representational Harms in Commonsense Knowledge Resources

By Ninareh Mehrabi, Pei Zhou, Fred Morstatter, Jay Pujara, Xiang Ren, Aram Galstyan. 2021

47 citations recorded by the source corpus.

Warning: this paper contains content that may be offensive or upsetting. Commonsense knowledge bases (CSKB) are increasingly used for various natural language processing tasks. Since CSKBs are mostly human-generated and may reflect societal biases, it is important to ensure that such biases are not conflated with the notion of commonsense. Here we focus on two widely used CSKBs, ConceptNet and GenericsKB, and establish the presence of bias in the form of two types of representational harms, overgeneralization of polarized perceptions and representation disparity across different demographic groups in both CSKBs. Next, we find similar representational harms for downstream models that use ConceptNet. Finally, we propose a filtering-based approach for mitigating such harms, and observe that our filtered-based approach can reduce the issues in both resources and models but leads to a performance drop, leaving room for future work to build fairer and stronger commonsense models.

Subfields: Artificial Intelligence, Knowledge Graphs & Representation, AI Ethics, Fairness & Society

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 Knowledge Graphs & Representation

In AI Ethics, Fairness & Society

Citations in the corpus

A Survey on Bias and Fairness in Machine Learning (2019, background): “In addition, authors in [105] write about the existing representational biases in different knowledge bases that are widely used in Natural Language Processing (NLP) applications for different commonsense reasoning tasks.”
Commonsense-Focused Dialogues for Response Generation: An Empirical Study (2021, methodology): “Our work uses ConceptNet to ﬁlter for commonsense-focused dialogues, but Mehrabi et al. (2021) have found representational harms in common sense resources.”
What do Bias Measures Measure? (2021, background): “In addition, Mehrabi et al. (2021) and Rudinger et al. (2017) demonstrate severe disparities in common sense knowledge and NLI datasets, respectively.”
Explaining Toxic Text via Knowledge Enhanced Text Generation (2022, background): “On the other hand, while our source (ConceptNET) for explicit knowledge may be biased (Mehrabi et al., 2021), retrieved stereotypes are often mixed with general, unbiased facts.”
On Measures of Biases and Harms in NLP (2021, background): “In addition, Mehrabi et al. (2021) and Rudinger et al. (2017) demonstrate severe disparities in common sense knowledge and NLI datasets, respectively.”
Deep Bidirectional Language-Knowledge Graph Pretraining (2022, unknown): “Second, the ConceptNet knowledge graph (Speer et al. 2017) used in this work has been shown to encode stereotypes (Mehrabi et al. 2021), rather than completely clean commonsense knowledge.”
Perturbation Augmentation for Fairer NLP (2022, background): “Training models on data with representational issues can lead to unfair or poor treatment of particular demographic groups (Baro-cas et al., 2017; Mehrabi et al., 2021), a problem that is particularly egregious for historically marginalized groups, including people of color (Field et al., 2021),…”