A Multilingual Dataset of Racial Stereotypes in Social Media Conversational Threads

Findings of the Association for Computational Linguistics: EACL 2023, pages 686–696 May 2-6, 2023 ©2023 Association for Computational Linguistics A Multilingual Dataset of Racial Stereotypes in Social Media Conversational Threads Tom Bourgeade1,∗, Alessandra Teresa Cignarella2,∗, Simona Frenda2,∗, Mario Laurent1, Wolfgang S. Schmeisser-Nieto3, Farah Benamara1,4, Cristina Bosco2, Véronique Moriceau1, Viviana Patti2 and Mariona Taulé3 1. IRIT, Université de Toulouse, CNRS, Toulouse INP, UT3, Toulouse, France 2. Dipartimento di Informatica, Università degli Studi di Torino, Italy 3. CLiC research group, UBICS, Universitat de Barcelona, Spain 4. IPAL, CNRS-NUS-ASTAR, Singapore Warning: This paper contains exam- ples of potentially offensive content. Abstract In this paper, we focus on the topics of mis- information and racial hoaxes from a per- spective derived from both social psychol- ogy and computational linguistics. In partic- ular, we consider the specific case of anti- immigrant feeling as a first case study for ad- dressing racial stereotypes. We describe the first corpus-based study for multilingual racial stereotype identification in social media con- versational threads. Our contributions are: (i) a multilingual corpus of racial hoaxes, (ii) a set of common guidelines for the annota- tion of racial stereotypes in social media texts, and a multi-layered, fine-grained scheme, psy- chologically grounded on the work by Fiske et al., including not only stereotype presence, but also contextuality, implicitness, and forms of discredit, (iii) a multilingual dataset in Ital- ian, Spanish, and French annotated follow- ing the aforementioned guidelines, and cross- lingual comparative analyses taking into ac- count racial hoaxes and stereotypes in on- line discussions. The analysis and results show the usefulness of our methodology and re- sources, shedding light on how racial hoaxes are spread, and enable the identification of negative stereotypes that reinforce them. 1 Introduction Racial Hoaxes (RHs) are “a communicative act oriented to spread fallacious information against a social group” (Russell, 1998). As social media have become a dominant means of communica- tion, investigating them is crucial for tackling the spread of RHs. We approach this task combining * The first three authors contributed equally. psychological and computational linguistics meth- ods with a multilingual, cross-cultural perspective (Italian, Spanish, and French). In particular, RHs can contribute to the diffu- sion of stereotypes about people belonging to the outgroup, i.e., a social group with features that dif- fer from the ingroup (Rooduijn et al., 2021) and are, thus, more vulnerable. Even common, naive users are as likely to become spreaders of RHs as malicious users (Papapicco et al., 2022). In this paper, we cover a specific theme: anti-immigrant stereotypes. The discursive construction of immi- grants and refugees in user interaction on social media has been studied by Ekman (2019), who has shown how racial expressions and overt racism are becoming increasingly normalized, thus leading to prejudices and racial stereotypes and, eventually, even harmful acts. Overall, the attention to these topics is rela- tively new in the NLP community, and thus, there is still a meaningful lack of annotated resources for the development of automatic tools to detect stereotypes and related phenomena. …

Content truncated to 500 words