BuzzSaw at SemEval-2017 Task 7: Global vs. Local Context for Interpreting and Locating Homographic English Puns with Sense Embeddings

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 444–448, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics BuzzSaw at SemEval-2017 Task 7: Global vs. Local Context for Interpreting and Locating Homographic English Puns with Sense Embeddings Dieke Oele CLCG University of Groningen The Netherlands d.oele@rug.nl Kilian Evang CLCG University of Groningen The Netherlands k.evang@rug.nl Abstract This paper describes our system partici- pating in the SemEval-2017 Task 7, for the subtasks of homographic pun location and homographic pun interpretation. For pun interpretation, we use a knowledge- based Word Sense Disambiguation (WSD) method based on sense embeddings. Pun- based jokes can be divided into two parts, each containing information about the two distinct senses of the pun. To exploit this structure we split the context that is input to the WSD system into two local con- texts and find the best sense for each of them. We use the output of pun interpre- tation for pun location. As we expect the two meanings of a pun to be very dissim- ilar, we compute sense embedding cosine distances for each sense-pair and select the word that has the highest distance. We describe experiments on different methods of splitting the context and compare our method to several baselines. We find ev- idence supporting our hypotheses and ob- tain competitive results for pun interpreta- tion. 1 Introduction A pun is a word used in a context to evoke two or more distinct senses for humorous effect. For example, in the 1987 movie “The Running Man”, Arnold Schwarzenegger’s character cuts his en- emy Buzzsaw in half with a chainsaw, then an- nounces: “He had to split.” The verb split is the pun here, evoking two senses in the context: that of leaving, and that of disintegrating into two parts. Recognizing and appreciating puns requires so- phisticated feats of intelligence currently unique to humans. A recently proposed set of artificial intel- ligence tasks (Miller et al., 2017) challenges com- puters to try their hand at it: pun detection (tell whether or not a text contains a pun), pun location (given a text with a pun, tell which word is the pun) and pun interpretation (given a pun in context, tell which senses it evokes). Pun interpretation is closely related to the task of Word Sense Disambiguation. A typical WSD system chooses that sense of a word which fits best in the context the word appears in. A pun interpretation system, however, should return not one but two different senses of a word. Miller and Turkovi´c (2016) suggest a straightforward ex- tension of the WSD approach to pun interpreta- tion: choose the best scoring sense and second- best scoring sense for the word in its context. However, this approach does not take into account the specific structure of pun-based jokes. In most cases, such jokes can be divided into two parts, where in the first part cues for one sense are con- centrated, and in the second part, cues for another …

Content truncated to 500 words