Lexical Influences on the Perception of Sarcasm

Proceedings of the Workshop on Computational Approaches to Figurative Language, pages 1–4, Rochester, NY, April 26, 2007. c⃝2007 Association for Computational Linguistics Lexical Influences on the Perception of Sarcasm Roger J. Kreuz Gina M. Caucci Department of Psychology Department of Psychology The University of Memphis The University of Memphis Memphis, TN 38152 Memphis, TN 38152 rkreuz@memphis.edu gcaucci@memphis.edu Abstract Speakers and listeners make use of a vari- ety of pragmatic factors to produce and identify sarcastic statements. It is also possible that lexical factors play a role, al- though this possibility has not been inves- tigated previously. College students were asked to read excerpts from published works that originally contained the phrase said sarcastically, although the word sar- castically was deleted. The participants rated the characters’ statements in these excerpts as more likely to be sarcastic than those from similar excerpts that did not originally contain the word sarcasti- cally. The use of interjections, such as gee or gosh, predicted a significant amount of the variance in the participants’ ratings of sarcastic intent. This outcome suggests that sarcastic statements may be more formulaic than previously realized. It also suggests that computer software could be written to recognize such lexical factors, greatly increasing the likelihood that non- literal intent could be correctly interpreted by such programs, even if they are unable to identify the pragmatic components of nonliteral language. 1 Introduction It has long been assumed that verbal irony, or sar- casm, is principally a pragmatic phenomenon, and many studies from psycholinguistics have demon- strated the social, contextual, and interpersonal factors that affect its use and interpretation (for reviews, see Gibbs, 1994, 2003; Giora, 2003). An example of such a pragmatic factor is com- mon ground (Clark, 1996). The more familiar two people are with one other, the more likely it is that they will employ sarcasm (Kreuz, 1996). When interlocutors in a story share common ground, ex- perimental participants read sarcastic statements more quickly, and are more certain of the sarcastic intent, than when the interlocutors share little common ground (Kreuz and Link, 2002). These results can be explained in terms of a principle of inferability: speakers will only employ sarcasm if they are reasonably certain that their hearers will interpret it correctly (Kreuz, 1996). Such results have led to pessimistic forecasts concerning the likelihood that computer programs would ever be able to understand nonliteral lan- guage (e.g., Dews and Winner, 1997). If the use of such language relies solely on pragmatic factors, it would indeed be a considerable challenge to create software that could detect and interpret it. One difficulty with this conclusion is that most psycholinguistic studies of sarcasm have used ex- perimenter-generated materials instead of actual utterances. For example, sarcastic statements are often hyperbolic (Kreuz and Roberts, 1995), and so researchers have typically employed extreme con- structions, such as What perfectly lovely weather! as sarcastic commentary on a sudden downpour (Kreuz and Glucksberg, 1989). Such research, however, has unintentionally con- founded the pragmatic and the lexical aspects of …

Content truncated to 500 words