Distinguishing Sarcasm From Literal Language: Evidence From Books and Blogging

Sarcasm production and comprehension have been traditionally described in terms of pragmatic factors. Lexical cues have received less attention, but they may be important potential indicators. A major obstacle to examining such features is determining sarcastic intent. One solution is to analyze statements explicitly marked as being sarcastic. This study examined Twitter postings marked with #sarcasm as well as dialog from Google Books containing the phrase “said sarcastically.” We used word counting and part-of-speech tagging to compare specific lexical features of the explicitly-marked sarcastic statements to statements by the same author not marked as sarcastic. Our results broadly support the Lexical Cues Hypothesis—certain word-level cues, such as interjections and positive affect terms, are stereotypic of sarcasm. A model incorporating these features performed comparably to human raters in making sarcastic versus nonsarcastic judgments. This finding shows promise for future work toward automatically identifying sarcasm in text.