A Survey of Race, Racism, and Anti-Racism in NLP

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 1905–1925 August 1–6, 2021. ©2021 Association for Computational Linguistics 1905 A Survey of Race, Racism, and Anti-Racism in NLP Anjalie Field Carnegie Mellon University anjalief@cs.cmu.edu Su Lin Blodgett Microsoft Research sulin.blodgett@microsoft.com Zeerak Waseem University of Sheffield z.w.butt@sheffield.ac.uk Yulia Tsvetkov University of Washington yuliats@cs.washington.edu Abstract Despite inextricable ties between race and lan- guage, little work has considered race in NLP research and development. In this work, we survey 79 papers from the ACL anthology that mention race. These papers reveal various types of race-related bias in all stages of NLP model development, highlighting the need for proactive consideration of how NLP systems can uphold racial hierarchies. However, per- sistent gaps in research on race and NLP re- main: race has been siloed as a niche topic and remains ignored in many NLP tasks; most work operationalizes race as a fixed single- dimensional variable with a ground-truth label, which risks reinforcing differences produced by historical racism; and the voices of histor- ically marginalized people are nearly absent in NLP literature. By identifying where and how NLP literature has and has not considered race, especially in comparison to related fields, our work calls for inclusion and racial justice in NLP research practices. 1 Introduction Race and language are tied in complicated ways. Raciolinguistics scholars have studied how they are mutually constructed: historically, colonial pow- ers construct linguistic and racial hierarchies to justify violence, and currently, beliefs about the inferiority of racialized people’s language practices continue to justify social and economic exclusion (Rosa and Flores, 2017).1 Furthermore, language is the primary means through which stereotypes and prejudices are communicated and perpetuated (Hamilton and Trolier, 1986; Bar-Tal et al., 2013). However, questions of race and racial bias have been minimally explored in NLP literature. 1We use racialization to refer the process of “ascribing and prescribing a racial category or classification to an individual or group of people ...based on racial attributes including but not limited to cultural and social history, physical features, and skin color” (Hudley, 2017). While researchers and activists have increasingly drawn attention to racism in computer science and academia, frequently-cited examples of racial bias in AI are often drawn from disciplines other than NLP, such as computer vision (facial recognition) (Buolamwini and Gebru, 2018) or machine learn- ing (recidivism risk prediction) (Angwin et al., 2016). Even the presence of racial biases in search engines like Google (Sweeney, 2013; Noble, 2018) has prompted little investigation in the ACL com- munity. Work on NLP and race remains sparse, particularly in contrast to concerns about gender bias, which have led to surveys, workshops, and shared tasks (Sun et al., 2019; Webster et al., 2019). In this work, we conduct a comprehensive sur- vey of how NLP literature and research practices engage with race. We first examine 79 papers from the ACL Anthology that mention the words ‘race’, ‘racial’, or ‘racism’ and highlight …

Content truncated to 500 words