Phonological Pun-derstanding

Proceedings of NAACL-HLT 2016, pages 654–663, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Phonological Pun-derstanding Aaron Jaech and Rik Koncel-Kedziorski and Mari Ostendorf University of Washington Seattle, WA {ajaech,kedzior,ostendor}@uw.edu Abstract Many puns create humor through the rela- tionship between a pun and its phonologically similar target. For example, in “Don’t take ge- ologists for granite” the word “granite” is a pun with the target “granted”. The recovery of the target in the mind of the listener is essen- tial to the success of the pun. This work intro- duces a new model for automatic target recov- ery and provides the first empirical test for this task. The model draws upon techniques for automatic speech recognition using weighted finite-state transducers, and leverages auto- matically learned phone edit probabilities that give insight into how people perceive sounds and into what makes a good pun. The model is evaluated on a small corpus where it is able to automatically recover a large fraction of the pun targets. 1 Introduction From the high culture of Shakespeare’s plays (Tanaka, 1992), to the depths of the YouTube com- ments section, from advertising slogans (Keller, 2009) to conversations with nerdy parents, puns are a versatile rhetorical device and their understand- ing is essential to any comprehensive approach to computational humor. Humor has been described as “one of the most interesting and puzzling research areas in the field of natural language understanding” (Yang et al., 2015). Puns, in particular, offer an in- teresting subject for study since their humor derives from wordplay and double-meaning. An important class of puns, known as parono- masic puns, are those where one entity, the pun, is phonologically similar to another, the target (Joseph, 2008). Consider an example from Crosbie (1977): “Sign by gate to nudist colony: Come in. We are Never Clothed.” Here, “clothed” is the pun and “closed” is the tar- get. Paronomasic puns are distinguished from ho- mographic puns such as “Two silkworms had a race. They ended up in a tie.” which puns on the two definitions of the word “tie”. When the pun and target are homophonic this is called a perfect pun, and when nearly homophonic an imperfect pun (Zwicky and Zwicky, 1986) (or a heterophonic pun (Hempelmann, 2003)). The fo- cus of this work is to propose and evaluate a model for target recovery of both perfect and imperfect paronomasic puns, assuming that the location of the pun word or word sequence. Ritchie (2005) classifies puns in terms of whether they are self-contained, i.e., based on general knowl- edge and humorous in a variety of circumstances, or contextually integrated, i.e. relying on a specific context such as a visual context, knowledge of a re- cent event or discussion. Many puns of this type are associated with cartoons or images, e.g. a cartoon with pies and cakes in the street having the caption “The streets were oddly desserted” (desserted/deserted). Contextually integrated puns lose their humor out of context because the pun is difficult …

Content truncated to 500 words