UniTuebingenCL at SemEval-2020 Task 7: Humor Detection in News Headlines

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1060–1065 Barcelona, Spain (Online), December 12, 2020. 1060 UniTuebingenCL at SemEval-2020 Task 7: Humor Detection in News Headlines Charlotte Sophie Ammer University of Tuebingen, Germany Department of Linguistics charlottesophie72074@gmail.com Lea Hannah Gr¨uner University of Tuebingen, Germany Department of Linguistics gauner.erle@gmail.com Abstract This paper describes the work done by the team UniTuebingenCL for the SemEval 2020 Task 7: “Assessing the Funniness of Edited News Headlines”. We participated in both sub-tasks: sub-task A, given the original and the edited headline, predicting the mean funniness of the edited headline; and sub-task B, given the original headline and two edited versions, predicting which edited version is the funnier of the two. A Ridge Regression model using Elmo and Glove embeddings as well as Truncated Singular Value Decomposition was used as the final model. A long short term memory model recurrent network (LSTM) served as another approach for assessing the funniness of a headline. For the first sub-task, we experimented with the extraction of multiple features to achieve lower Root Mean Squared Error. The lowest Root Mean Squared Error achieved was 0.575 for sub-task A, and the highest Accuracy was 0.618 for sub-task B. 1 Introduction With the rise of social media and social platforms such as Twitter and Facebook, the automatic interpreta- tion of shorter amounts of text becomes increasingly more significant. The task of “reading between the lines” is so far often constricted to detecting offensive language or slurs. However, detecting funniness in written content also motivates research due to several reasons. Discovering humorous content can help understand text on a deeper level and get its intention. It not only helps comprehending the author, but also the readers of the text and why they favor it. Humor can be viewed as a natural language phenomenon and therefore linguistically interesting, but it also proves a lot of potential for modern psychology. Using humor can be a way to gain intimacy, reduce stress, and can improve feelings and overall wellbeing (Crawford and Caltabiano, 2011). It is often linked to soft skills like character strength and can therefore also be an asset for human resources, resulting in a change of hiring processes. Humor is a communication skill that requires high levels of world knowledge and discourse insight. It is often provoked by an unexpected change of events, and has a lot to do with language usage and ambiguity. When used appropriately, humor can enrich human communication and serve as a tool to establish relationships. Humor has a sociological function and can therefore also be abused. Humor recognition can help detect such abusive humor, but also it may improve human-computer interaction. The task of automatic humor recognition is quite challenging because for a human to create or detect humor, world knowledge plays an important role. This is especially difficult in tasks where evaluation takes place without context dialogue to consider. Humor can also come in many variations, such as humor using anecdotes, self-deprecation, vulgarity and many more. …

Content truncated to 500 words