An Annotated Social Media Corpus for German

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6127–6135 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6127 An Annotated Social Media Corpus for German Eckhard Bick University of Southern Denmark Odense, Denmark eckhard.bick@mail.dk Abstract This paper presents the German Twitter section of a large (2 billion word) bilingual Social Media corpus for Hate Speech research, discussing the compilation, pseudonymization and grammatical annotation of the corpus, as well as special linguistic features and peculiarities encountered in the data. Among other things, compounding, accidental and intentional orthographic variation, gendering and the use of emoticons/emojis are addressed in a genre-specific fashion. We present the different layers of linguistic annotation (morphosyntactic, dependencies and semantic types) and explain how a general parser (GerGram) can be made to work on Social Media data, pointing out necessary adaptations and extensions. In an evaluation run on a random cross-section of tweets, the modified parser achieved F-scores of 97% for morphology (fine-grained POS) and 92% for syntax (labeled attachment score). Predictably, performance was twice as good in tweets with standard orthography than in tweets with spelling/casing irregularities or lack of sentence separation, the effect being more marked for morphology than for syntax. Keywords: Social Media corpus, Hate Speech, German Corpus Linguistics, Constraint Grammar, Syntactic parsing, Non-standard orthography, Emoji annotation 1. Introduction Hate Speech (HS) against ethnic, religious and national minorities is a growing concern in online discourse (e.g. Foxman & Wolf 2013), creating a conflict of interest in societies that want to advocate freedom of speech on the one hand, and to protect minorities against defamation on the other (Herz & Molnar 2012). Social networks, under political pressure to take action, have begun to filter for what they perceive as outright hate speech. Reliable data, actionable definitions and linguistic research is needed for such filtering to be effective without being disproportionate, but also in order to allow policy makers, educational institutions, journalists and other public influencers to understand and counteract the phenomenon of hate speech. The three-year research project behind the work described here is called XPEROHS (Baumgarten et al. 2019) and investigates the expression and perception of online hate speech with a particular focus on immigrant and refugee minorities in Denmark and Germany. In addition to experimental and questionnaire studies, data is collected from two major social networks, Twitter and Facebook. The material is used to examine and quantify linguistic patterns found in hateful discourse and to identify derogatory terms and outright slurs, as well as metaphors used in a demeaning and target-specific way. Finally the corpus is used to provide graded HS examples for the project's empirical work on HS perception. 2. The Corpus 2.1 Corpus Size and Sources The XPEROHS corpus is a monitor corpus, where posts and comments were collected continuously from Twitter (late 2017 - mid 2019) and Facebook (late 2017 - mid 2018) for both German and Danish, using the networks' query APIs. Other social networks were considered, but not used because …

Content truncated to 500 words