Automatically detecting verbal irony (roughly, sarcasm) is a challenging task because ironists say something other than - and often opposite to - what they actually mean. Discerning ironic intent exclusively from the words and syntax comprising texts (e.g., tweets, forum posts) is therefore not always possible: additional contextual information about the speaker and/or the topic at hand is often necessary. We introduce a new corpus that provides empirical evidence for this claim. We show that annota-tors frequently require context to make judgements concerning ironic intent, and that machine learning approaches tend to misclassify those same comments for which annotators required additional context.
展开▼