WLV at SemEval-2018 Task 3: Dissecting Tweets in Search of Irony

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 553–559 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics WLV at SemEval-2018 Task 3: Dissecting Tweets in Search of Irony Omid Rohanian, Shiva Taslimipoor, Richard Evans and Ruslan Mitkov Research Group in Computationa Linguistics University of Wolverhampton Wolverhampton, UK {omid.rohanian,shiva.taslimi,r.j.evans,r.mitkov}@wlv.ac.uk Abstract This paper describes the systems submitted to SemEval 2018 Task 3 “Irony detection in En- glish tweets” for both subtasks A and B. The first system leveraging a combination of senti- ment, distributional semantic, and text surface features is ranked third among 44 teams ac- cording to the official leaderboard of the sub- task A. The second system with slightly differ- ent representation of the features ranked ninth in subtask B. We present a method that en- tails decomposing tweets into separate parts. Searching for contrast within the constituents of a tweet is an integral part of our system. We embrace an extensive definition of con- trast which leads to a vast coverage in detect- ing ironic content. 1 Introduction In figurative language (also known as trope), there is a departure from literal use of words. In order to decode meaning, therefore, it is not enough to rely solely on the literal sense of individual words. Irony and sarcasm are two types of such language that exploit this technique in similar ways. They “both involve deliberately saying something that is incongruous or the opposite of what the speaker knows to be true” (Hanks, 2013). This is some- times formulated as a transgression of the Gricean maxim of quality (Grice, 1975)1. Under this assumption it follows that the viola- tion is only permissible thanks to shared knowl- edge between the speaker and the hearer. In order to achieve this goal, the speaker frames the mes- sage with some form of commentary or metames- sage that signals the ironic or sarcastic nature of the message. This is usually realised through negation of the original meaning (Haiman, 1998). Regardless of their similarities, irony and sar- casm are not technically the same as they might be 1“Do not say what you believe to be false.” employed for different purposes. It is widely ac- cepted that sarcasm involves some degree of ver- bal aggression and ridicule directed at the hearer, whilst irony can simply be used for humorous or emphatic effect. It has been shown that compu- tational processing of irony and sarcasm requires some knowledge of the context in which they ap- pear, sometimes including paralinguistic informa- tion (Wallace et al., 2014). Exploring ironicity has practical implications, since performance of sentiment analysis systems is directly affected by knowledge about irony and sarcasm (Pozzi et al., 2016). As part of the 12th workshop on semantic eval- uation (SemEval-2018), Shared Task 3 defines two subtasks with regards to irony detection in English tweets (Van Hee et al., 2018). Subtask A involves binary classification. The objective is to train a system that can label tweets as ironic or not. Sub- task B …

Content truncated to 500 words