Sarcasm SIGN: Interpreting Sarcasm with Sentiment Based Monolingual Machine Translation

摘要: Sarcasm is a form of speech in which speakers say the opposite what they truly mean order to convey strong sentiment. In other words, “Sarcasm giant chasm between I say, and person who doesn’t get it.”. this paper we present novel task sarcasm interpretation, defined as generation non-sarcastic utterance conveying same message original sarcastic one. We introduce dataset 3000 tweets, each interpreted by five human judges. Addressing monolingual machine translation (MT), experiment with MT algorithms evaluation measures. then SIGN: an based interpretation algorithm that targets sentiment defining element textual sarcasm. show while scores n-gram automatic measures are similar for all models, SIGN’s interpretations scored higher humans adequacy polarity. conclude discussion on future research directions our new task.