SRHR at SemEval-2017 Task 6: Word Associations for Humour Recognition

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 401–406, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics SRHR at SemEval-2017 Task 6: Word Associations for Humour Recognition Andrew Cattle Xiaojuan Ma Hong Kong University of Science and Technology Department of Computer Science and Engineering Clear Water Bay, Hong Kong {acattle,mxj}@cse.ust.hk Abstract This paper explores the role of semantic relatedness features, such as word asso- ciations, in humour recognition. Specif- ically, we examine the task of infer- ring pairwise humour judgments in Twit- ter hashtag wars. We examine a va- riety of word association features de- rived from the University of Southern Florida Free Association Norms (USF) (Nelson et al., 2004) and the Edinburgh Associative Thesaurus (EAT) (Kiss et al., 1973) and find that word association- based features outperform Word2Vec sim- ilarity, a popular semantic relatedness measure. Our system achieves an ac- curacy of 56.42% using a combination of unigram perplexity, bigram perplex- ity, EATtweet-avg difference, USFmax forward, EATword-avg difference, USFword-avg difference, EATmin forward, USFtweet-max difference , and EATmin backward. 1 Introduction What makes something funny? Humour is very personal; what is funny to one person may not be funny to another. Yet, there are still certain works which seem to have widespread appeal, from co- median Louis C.K. to sitcom The Big Bang The- ory. What makes these works more humorous to the average person than similar ones? A good place to start might be with the show @midnight and their nightly Hashtag Wars segment. Each night viewers are given a prompt in the form of a Twitter hashtag and asked to tweet their funni- est responses. Given two such tweets, how can we decide which is funnier? This paper largely focuses on semantic relatedness-based features and their application in humour recognition. It is reasonable to assume that a punchline should be related to a setup; that is to say, a tweet’s relevance to its hashtag prompt should be apparent. Similarly, it is reasonable to assume that punchlines should have a certain amount of unexpectedness; in other words, funnier tweets should be harder to guess. As such, it follows that semantic relation strength in general should serve as a barometer for humour where weaker relations are less understandable and stronger relations are more obvious (Cattle and Ma, 2016). Moreover, we hypothesize that the interplay between this understandability and unexpectedness should provide an even more powerful indication of humour. 2 Previous Work Early work on computational humour focused more on humour generation in specific contexts, such as punning riddles (Binsted and Ritchie, 1994; Ritchie et al., 2007), humorous acronyms (Stock and Strapparava, 2003), or jokes in the form of “I like my X like I like my Y” (Petrovic and Matthews, 2013). Labutov and Lipson (2012) offered a slightly more generalized approach using Semantic Script Theory of Humour. Recently, humour recognition has gained in- creasing attention. Taylor and Mazlack (2004) presented a method for recognizing wordplay in “Knock Knock” jokes. Mihalcea and Strapparava (2005) …

Content truncated to 500 words