Sarcasm Detection Using Grice\u27s Maxims

A model of sarcasm interpretation must account for all instances of sarcasm, including sarcastic questions, sarcastic utterances without context, etc. I illustrate how Pretense Theory (Clark & Gerrig, 1984) and Mention Theory (Sperber & Wilson, 1981) fall short and argue for a modified version of Grice\u27s meaning-inversion model of sarcasm (1975, 1978) which allows for implicatures and other non-truth-conditional components of an utterance to be included in the meaning-inversion mechanism, along the lines of Camp (2012). Under a Gricean framework, I propose a new set of sarcasm detection cues I call maxim-violation cues which includes hyperbole, self-contradiction, and manner violation. I provide empirical evidence that English speakers utilize these cues when detecting sarcasm on Twitter, an environment which precludes to a great extent the use of vocal, visual, and contextual cues