I Feel Offended, Don’t Be Abusive! Implicit/Explicit Messages in Offensive and Abusive Language

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6193–6202 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6193 I Feel Offended, Don’t Be Abusive! Implicit/Explicit Messages in Offensive and Abusive Language Tommaso Caselli♣,Valerio Basile⋄, Jelena Mitrovi´c‡, Inga Kartoziya♣, Michael Granitzer‡ ♣University of Groningen, ⋄University of Turin, ‡University of Passau Groningen The Netherlands, Turin Italy, Passau Germany ⋄ valerio.basile@unito.it, ♣t.caselli@rug.nl,♣i.kartoziya.1@student.rug.nl ‡{jelena.mitrovic|michael.granitzer}@uni-passau.de Abstract Abusive language detection is an unsolved and challenging problem for the NLP community. Recent literature suggests various approaches to distinguish between different language phenomena (e.g., hate speech vs. cyberbullying vs. offensive language) and factors (degree of explicitness and target) that may help to classify different abusive language phenomena. There are data sets that annotate the target of abusive messages (i.e.OLID/OffensEval (Zampieri et al., 2019a)). However, there is a lack of data sets that take into account the degree of explicitness. In this paper, we propose annotation guidelines to distinguish between explicit and implicit abuse in English and apply them to OLID/OffensEval. The outcome is a newly created resource, AbuseEval v1.0, which aims to address some of the existing issues in the annotation of offensive and abusive language (e.g., explicitness of the message, presence of a target, need of context, and interaction across different phenomena). Keywords: Corpus Annotation, Social Media Processing, Statistical and Machine Learning Methods 1. Introduction Social media platforms have been promoted as (virtual) places where many people from different parts of the world can engage in productive discussions and share opin- ions (Jurgens et al., 2019). At the same time, quoting Um- berto Eco, “everyone who inhabits the planet, including crazy people and idiots, has the right to the public word [and], on Internet, your message has the same authority as the Nobel laureate [...].” 1 Alongside such flattening in the process of production, consumption and sharing of infor- mation, toxic and abusive behavior online has surged. Recently, there has been an increasing effort from the Nat- ural Language Processing (NLP) community to develop methods for the automatic detection of abusive language and related phenomena, as the volume of data is such that it has become impossible to manually track and monitor it (Nobata et al., 2016; Kennedy et al., 2017). This has taken different forms such as the creation of data sets or corpora in multiple languages (Waseem and Hovy, 2016a; Ross et al., 2017; Poletto et al., 2017; Founta et al., 2018; Ibrohim and Budi, 2019), 2 the promotion of evaluation campaigns (Kumar et al., 2018; Wiegand et al., 2018b; Bosco et al., 2018; Zampieri et al., 2019b; Basile et al., 2019), the organization of thematic workshops and confer- ences, 3 and the compilation of surveys (Schmidt and Wie- gand, 2017; Fortuna and Nunes, 2018). However, research in this area is still fragmented and the field has been flooded with different terminologies, per- spectives and understandings of this phenomenon, some- how lacking of consensus on shared definitions. Waseem et al. (2017) clearly point out how …

Content truncated to 500 words