Idiom Savant at Semeval-2017 Task 7: Detection and Interpretation of English Puns

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 103–108, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics Idiom Savant at Semeval-2017 Task 7: Detection and Interpretation of English Puns Samuel Doogan Aniruddha Ghosh Hanyang Chen Tony Veale Department of Computer Science and Informatics University College Dublin Dublin, Ireland {samuel.doogan, aniruddha.ghosh, hanyang.chen}@ucdconnect.ie tony.veale@ucd.ie Abstract This paper describes our system, entitled Idiom Savant, for the 7th Task of the Se- meval 2017 workshop, “Detection and in- terpretation of English Puns”. Our system consists of two probabilistic models for each type of puns using Google n-grams and Word2Vec. Our system achieved f- score of 0.84, 0.663, and 0.07 in ho- mographic puns and 0.8439, 0.6631, and 0.0806 in heterographic puns in task 1, task 2, and task 3 respectively. 1 Introduction A pun is a form of wordplay, which is often pro- filed by exploiting polysemy of a word or by replacing a phonetically similar sounding word for an intended humorous effect. From Shake- speare’s works to modern advertisement catch- phrases (Tanaka, 1992), puns have been widely used as a humorous and rhetorical device. For a polysemous word, the non-literal meaning is addressed when contextual information has low accordance with it’s primary or most prominent meaning (Giora, 1997). A pun can be seen as a democratic form of literal and non-literal meaning. In using puns, the author alternates an idiomatic expression to a certain extent or provides enough context for a polysemous word to evoke non-literal meaning without attenuating literal meaning com- pletely (Giora, 2002). Task 7 of the 2017 SemEval workshop (Miller et al., 2017) involves three subtasks. The first sub- task requires the system to classify a given context into two binary categories: puns and non-puns. The second subtask concerns itself with finding the word producing the punning effect in a given context. The third and final subtask involves an- notating puns with the dual senses with which the punning effect is being driven. In a written context, puns are classified into 2 categories. Homographic puns shown in exam- ple 1, exploits polysemy of the language by us- ing a word or phrase which has multiple coher- ent meanings given its context; And heterographic puns shown in example 2, humorous effect is often induced by adding incongruity by replacing a pho- netically similar word which is semantically dis- tant from the context. (1) Tires are fixed for a flat rate. (2) A dentist hates having a bad day at the orifice. The rest of the paper is organized as fol- lows. Section 2 give a general description of our approach. Section 3 and 4 illustrate the de- tailed methodologies used for detecting and locat- ing Heterographic and Homographic puns respec- tively. In section 5, we provided an analysis of the system along with experimental results and finally section 6 contains some closing remarks and con- clusion. 2 General Approach We argue that the detection of heterographic puns rests on two assumptions. Firstly, the word …

Content truncated to 500 words