UWAV at SemEval-2017 Task 7: Automated feature-based system for locating puns

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 449–452, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics UWAV at SemEval-2017 Task 7: Automated feature-based system for locating puns Ankit Vadehra University of Waterloo avadehra@uwaterloo.ca Abstract In this paper we describe our system cre- ated for SemEval-2017 Task 7: Detection and Interpretation of English Puns(Miller et al., 2017). We tackle subtask 1, pun detection, by leveraging features selected from sentences to design a classifier that can disambiguate between the presence or absence of a pun. We address subtask 2, pun location, by utilizing a decision flow structure that uses presence or absence of certain features to decide the next action. The results obtained by our system are en- couraging, considering the simplicity of the system. We consider this system as a precursor for deeper exploration on ef- ficient feature selection for pun detection. 1 Introduction Puns are ambiguous word pairs in language that play on the different meaning of the word (pol- ysemy), or utilize similarly pronounced sounds (phonology) often for a humorous effect(Miller and Turkovi´c, 2016). They are widely used in written and spoken literature, intended as jokes. The task tries to detect whether a sentence is a pun or not (Subtask 1). If a pun is detected, we try to detect which word in the sentence was meant as a pun (Subtask2). For example, the sentence, “I’d tell you a chemistry joke, but, I know I wouldn’t get a reaction.” is a play on the word reaction, which is a common phrase. But, including it after chemistry, which deals with chemical reactions; gives it a humorous intent. Another example, the sentence, “I used to meditate a lot, but now I only do it every now and zen.” plays on the similar pro- nunciation of the words zen and then. Puns often require knowledge about syntactic similarity be- tween words and phrases like meditation and zen, chemistry and reaction, etc. Recently, a definite trend is noticeable in de- veloping a system that can automatically detect puns efficiently so that the humorous nature of the pun can be captured. Pun detection is of high im- portance for language modeling, machine transla- tion and sentiment analysis task so that the actual meaning of terms can be understood. This allows effectively utilizing the intended meaning rather than just the words themselves. The absence of any unified global word knowl- edge base makes detections of puns difficult since automatic selection of intended meaning is hard. There exist many Word Sense Disambiguation (WSD) approaches, but they are inefficient in cap- turing the word play(Miller and Gurevych, 2015). This makes pun detection a very interesting area to work on. Due to the ambiguous nature of the words used, sentences containing puns are often wrongly clas- sified. Due to an absence of a globally used pun dataset, all work in this area utilize self- accumulated and tagged datasets. This makes it difficult to compare the performance of different approaches. We …

Content truncated to 500 words