Computationally Recognizing Wordplay in Jokes

摘要: Computationally Recognizing Wordplay in Jokes Julia M. Taylor (tayloj8@email.uc.edu) Lawrence J. Mazlack (mazlack@uc.edu) Electrical & Computer Engineering and Science Department University of Cincinnati piece literature which the funniness culminates final sentence.” Most researchers agree that jokes can be broken into two parts, a setup punchline. The is first part joke, usually consisting most text, establishes certain expectations. punchline much shorter portion it causes some form conflict. It force another interpretation on violate an expectation, or both (Ritchie, 1998). As are relatively short, may possible to recog- nize them computationally. Computational recognition possible, but not easy. An “intelligent” joke recognizer requires world knowledge “understand” jokes. Abstract In artificial intelligence, have begun look at ap- proaches for computational humor. Although there appears no complete model recognizing verbally expressed humor, recognize based statistical language techniques. This in- vestigation humor recognition. considers restricted set all wordplay as component examines limited domain “Knock Knock” method uses Raskin's theory its theoretical foundation. original phrase complimentary different scripts overlap joke. algorithm deployed learns patterns text N-grams provides heuristic focus location where occur. generator produce utter- ance similar pronunciation given word, determines if utterance valid. Once discovered, de- termines found transforms Theories Humor Introduction Thinkers from ancient time Aristotle Plato present day strived discover define origins commonly, early definitions relied laughter: what makes people laugh humorous. Recent works separate laughter make own dis- tinct category response. Today almost many theories humor; cases, derived (Latta, 1999). Some say only definition covers aspects also impossible (Attardo, 1994). interesting subject study because difficult define, sense varies person person. same find something funny one day, next, depending person’s mood, has happened him her recently. These factors, among others, challenging. unaware complex steps involved recognition, consider these order approach ability human being. A common verbal, “verbally ex- pressed, humor” (Ritchie 2000). Verbally involves reading understanding texts. While understating meaning computer, not. One subclasses Hetzron (1991) defines “a short humorous Raskin’s (1985) Semantic Theory Verbal strongly influenced assumption every com- patible with scripts, those oppose each other punch line, therefore generating effect. Another Suls’ (1972) two-stage model, false expectation. following used process using • read, predictions conflict prediction, keep going If input conflicts prediction: o ending – PUZZLEMENT ending, try resolve: No rules Cognitive –HUMOR There been attempts generation 1996; Binsted, Lessard Levison, 1992; McDonough, 2001; McKay, 2002; Stock Strapparava, 2002) pun recognizers (Takizawa, et al. Yokogawa, Japanese. However, do appear any efforts. partly due absence unambiguous algorithm. cases Raskin Suls, does offer formal algorithm, second specify cognitive rule is, leaving major open interpretation. jokes, involving verbal play, class words sound, meanings. difference between meanings creates breaks