Lawyers are Dishonest? Quantifying Representational Harms in Commonsense Knowledge Resources

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5016–5033 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 5016 Lawyers are Dishonest? Quantifying Representational Harms in Commonsense Knowledge Resources Ninareh Mehrabi∗ Pei Zhou∗ Fred Morstatter Jay Pujara Xiang Ren Aram Galstyan University of Southern California and Information Sciences Institute {ninarehm,peiz,morstatt,jpujara,xiangren,galstyan}@usc.edu Abstract 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 bi- ases, it is important to ensure that such biases are not conflated with the notion of common- sense. Here we focus on two widely used CSKBs, ConceptNet and GenericsKB, and es- tablish the presence of bias in the form of two types of representational harms, overgeneral- ization of polarized perceptions and represen- tation disparity across different demographic groups in both CSKBs. Next, we find similar representational harms for downstream mod- els that use ConceptNet. Finally, we propose a filtering-based approach for mitigating such harms, and observe that our filtered-based ap- proach 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. 1 Introduction Commonsense knowledge is important for a wide range of natural language processing (NLP) tasks as a way to incorporate information about every- day situations necessary for human language under- standing. Numerous models have included knowl- edge resources such as ConceptNet (Speer et al., 2017) for question answering (Lin et al., 2019), sarcasm generation (Chakrabarty et al., 2020), and dialogue response generation (Zhou et al., 2018, 2021), among others. However, commonsense knowledge resources are mostly human-generated, either crowdsourced from the public (Speer et al., 2017; Sap et al., 2019) or crawled from massive web corpora (Bhakthavatsalam et al., 2020). For example, ConceptNet originated from the Open ∗equal contribution Source Examples ConceptNet (lady, UsedFor, fxxk) (lawyer, RelatedTo, dishonest) (church, UsedFor, brain washing) GenericsKB (Chinese people are very reclusive group of people) (Lawyers are registered menace to society) Table 1: Biased cases in ConceptNet and GenericsKB. Mind Common Sense project that collects com- monsense statements online from web users (Singh et al., 2002)1 and GenericsKB consists of crawled text from public websites. One issue with this ap- proach is that the crowdsourcing workers and web page writers may conflate their own prejudices with the notion of commonsense. For instance, we have found that querying for some target words such as “church” as shown in Table 1 in ConceptNet, results in biased triples. The potentially biased nature of commonsense knowledge bases (CSKB), given their increasing popularity, raises the urgent need to quantify biases both in the knowledge resources and in the down- stream models that use these resources. We present the first study on measuring bias in two large CSKBs, namely ConceptNet (Speer et al., 2017), the most widely used knowledge graph in com- monsense reasoning tasks, and GenericsKB (Bhak- thavatsalam et al., 2020), which …

Content truncated to 500 words