Detecting Derogatory Compounds – An Unsupervised Approach

Proceedings of NAACL-HLT 2019, pages 2076–2081 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2076 Detecting Derogatory Compounds – An Unsupervised Approach Michael Wiegand∗0, Maximilian Wolf∗, Josef Ruppenhofer†‡ ∗Spoken Language Systems, Saarland University, Saarbr¨ucken, Germany †Leibniz ScienceCampus, Heidelberg/Mannheim, Germany ‡Institute for German Language, Mannheim, Germany michael.wiegand@lsv.uni-saarland.de maximilian.wolf@lsv.uni-saarland.de ruppenhofer@ids-mannheim.de Abstract We examine the new task of detecting derogatory compounds (e.g. curry muncher). Derogatory compounds are much more diffi- cult to detect than derogatory unigrams (e.g. idiot) since they are more sparsely represented in lexical resources previously found effec- tive for this task (e.g. Wiktionary). We pro- pose an unsupervised classification approach that incorporates linguistic properties of com- pounds. It mostly depends on a simple distri- butional representation. We compare our ap- proach against previously established methods proposed for extracting derogatory unigrams. 1 Introduction Abusive or offensive language is commonly de- fined as hurtful, derogatory or obscene utterances made by one person to another person.1 Examples are (1)-(3). 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) Just want to slap the stupid out of these bimbos!!! (3) Go lick a pig you arab muslim piece of scum. Due to the rise of user-generated web content, in particular on social media networks, the amount of abusive language is also steadily growing. NLP methods are required to focus human review ef- forts towards the most relevant microposts. A substantial amount of abusive utterances comprises derogatory words (e.g. bimbo or scum). Automatic extraction methods of such words are required since new derogatory words constantly enter language. Wiegand et al. (2018a) extracted a 0Present affiliation: berg/Mannheim, GermanyLeibniz ScienceCampus, Heidel- 1http://thelawdictionary.org/ large list of such expressions and demonstrated its importance for text classification. In this work, we focus on a subtype of deroga- tory terms, namely derogatory compounds (e.g. booze hound, curry muncher, fault finder). Distin- guishing such multi-word expressions from non- derogatory ones (e.g. fox hound, mile muncher, branch finder) is more difficult than classifying unigrams since they are only sparsely represented in general-purpose lexical resources which have previously been found an effective source from which to learn abusive language, such as Wik- tionary.2 For example, while 97% of the deroga- tory unigrams of the gold standard lexicon in Wie- gand et al. (2018a) are contained in Wiktionary, less than 17% of the derogatory compounds used as our gold standard in this work can be found. Despite their sparsity in lexical resources derogatory compounds are a frequent phe- nomenon, particularly in German data, which is why we study this task on that language. On the German benchmark corpus for abusive language detection, the GermEval corpus (Wiegand et al., 2018b), we found that of the abusive microposts in the test set that include at least one derogatory expression, 39% contain a derogatory compound. In our work, …

Content truncated to 500 words