Identifying Implicitly Abusive Remarks about Identity Groups using a Linguistically Informed Approach

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5600 - 5612 July 10-15, 2022 ©2022 Association for Computational Linguistics Identifying Implicitly Abusive Remarks about Identity Groups using a Linguistically Informed Approach Michael Wiegand Digital Age Research Center (D!ARC) Alpen-Adria-Universität Klagenfurt AT-9020 Klagenfurt, Austria michael.wiegand@aau.at Elisabeth Eder Institut für Germanistik Alpen-Adria-Universität Klagenfurt AT-9020 Klagenfurt, Austria elisabeth.eder@aau.at Josef Ruppenhofer Leibniz Institute for the German Language D-68161 Mannheim, Germany ruppenhofer@ids-mannheim.de Abstract We address the task of distinguishing implicitly abusive sentences on identity groups (Muslims terrorize the world daily) from other group- related negative polar sentences (Muslims de- spise terrorism). Implicitly abusive language are utterances not conveyed by abusive words (e.g. bimbo or scum). So far, the detection of such utterances could not be properly addressed since existing datasets displaying a high degree of implicit abuse are fairly biased. Following the recently proposed strategy to solve implicit abuse by separately addressing its different sub- types, we present a new focused and less biased dataset that consists of the subtype of atomic negative sentences about identity groups. For that task, we model components that each ad- dress one facet of such implicit abuse, i.e. de- piction as perpetrators, aspectual classification and non-conformist views. The approach gen- eralizes across different identity groups and languages. 1 Introduction Abusive language is commonly defined as hurtful, derogatory or obscene utterances made by one per- son to another person.1 Examples are (1)-(2). In the literature, closely related terms include hate speech (Waseem and Hovy, 2016) or cyber bullying (Zhong et al., 2016). While there may be nuanced differences in meaning, they are all compatible with the general definition above. (1) stop editing this, you dumbass. (2) Go lick a pig you arab muslim piece of scum. Due to the rise of user-generated web content, the amount of abusive language is growing. NLP 1http://thelawdictionary.org methods are required to focus human review ef- forts towards the most relevant microposts. Though there has been much work on abusive language de- tection in general, there has been little work focus- ing on implicit forms of abusive language (3)-(4) (Waseem et al., 2017). By implicit we understand abusive language that is not conveyed by (unam- biguously) abusive words (e.g. bimbo, scum). (3) Did Stevie Wonder choose these models? (4) You inspire my inner serial killer. Detailed analyses of the output of existing classi- fiers have also revealed that currently only explicit abuse can be reliably detected (van Aken et al., 2018; Wiegand et al., 2019, 2021b). In this paper, we define implicit abuse as those abusive utterances that lack any abusive word ac- cording to the largest lexicon of abusive words available, i.e. the lexicon by Wiegand et al. (2018). In particular, datasets focusing on abuse towards identity groups (Jews, gay people etc.) contain a high degree of implicit abuse. For example, accord- ing to Wiegand et al. (2021b), on the dataset from Waseem and Hovy (2016), 56% of the abusive in- stances …

Content truncated to 500 words