CoRAL: a Context-aware Croatian Abusive Language Dataset

In light of unprecedented increases in the popularity of the internet and social media, comment moderation has never been a more relevant task. Semi-automated comment moderation systems greatly aid human moderatorsby either automatically classifying the examples or allowing the moderators to prioritizewhich comments to consider first. However,the concept of inappropriate content is oftensubjective, and such content can be conveyedin many subtle and indirect ways. In this work,we propose CoRAL1 a language and culturally aware Croatian Abusive dataset covering phenomena of implicitness and relianceon local and global context. We show experimentally that current models degrade whencomments are not explicit and further degradewhen language skill and context knowledgeare required to interpret the comment.