N-Hance at SemEval-2017 Task 7: A Computational Approach using Word Association for Puns

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 436–439, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics N-Hance at SemEval-2017 Task 7: A Computational Approach using Word Association for Puns ¨Ozge Sevgili ˙Izmir Institute of Technology, Computer Engineering, Urla, ˙Izmir, Turkey ozgesevgili@iyte.edu.tr Nima Ghotbi ˙Izmir Institute of Technology, Computer Engineering, Urla, ˙Izmir, Turkey nimaghotbi @iyte.edu.tr Selma Tekir ˙Izmir Institute of Technology, Computer Engineering, Urla, ˙Izmir, Turkey selmatekir@iyte.edu.tr Abstract This paper presents a system developed for SemEval-2017 Task 7, Detection and In- terpretation of English Puns consisting of three subtasks; pun detection, pun loca- tion, and pun interpretation, respectively. The system stands on recognizing a dis- tinctive word which has a high association with the pun in the given sentence. The in- tended humorous meaning of pun is iden- tified through the use of this word. Our official results confirm the potential of this approach. 1 Introduction Word Sense Disambiguation (WSD) (Navigli, 2009) is the task of determining the sense of a word in a specific context computationally. In the general case, a polysemous word’s a single, specific sense is meant in a given context. How- ever, a type of wordplay called pun suggests two or more meanings, by exploiting multiple mean- ings of words, or of similar-sounding words. Un- derstanding the linguistic realization of puns and automatically detecting and disambiguating them are important for computational linguistics. Ma- chine learning-based approaches are too expen- sive and impractical for this area as these humour constructs employ lexical-semantic anomalies and they are hard to be found in regular training sets. Thus, knowledge-based and unsupervised meth- ods are prevalent. There are three tasks in the detection and interpretation of English puns: pun detection-the given context contains a pun or not, pun identification-location of the pun word, and pun disambiguation-identifying the two meanings referred by the pun in the given context. Current approaches to pun interpretation rely on Lesk (1986), which considers the highest overlap between the context and gloss in order to return the target sense (two senses) for the pun. Miller and Gurevych (2015) performs automatic pun disam- biguation in a comprehensive experimental setup for the first time using three Lesk variants along with Random and MFS (Most Frequent Sense) baselines. The general conclusion is that pun dis- ambiguation results are poorer when compared with traditional WSD and traditional WSD must be extended with pun-specific features to increase accuracy in this area. Our consideration for the task of detection and interpretation of English puns is mainly based on the assumption that pun containing contexts in- clude a distinctive word that can be paired with the pun. This word has a central role in the word- play such that the intended humorous meaning of pun is in connection with this word. This claim implies that this word’s association with a pun is considerably greater than the other words’ as- sociation scores with the pun. As the threshold to be used here is determined using all the sen- …

Content truncated to 500 words