UMUTeam at SemEval-2021 Task 7: Detecting and Rating Humor and Offense with Linguistic Features and Word Embeddings

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1096–1101 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 1096 UMUTeam at SemEval-2021 Task 7: Detecting and Rating Humor and Offense with Linguistic Features and Word Embeddings Jos´e Antonio Garc´ıa-D´ıaz Facultad de Inform´atica, Universidad de Murcia, Campus de Espinardo, 30100 Murcia, Spain joseantonio.garcia8@um.es Rafael Valencia-Garc´ıa Facultad de Inform´atica, Universidad de Murcia, Campus de Espinardo, 30100 Murcia, Spain valencia@um.es Abstract In writing, humor is mainly based on figura- tive language in which words and expressions change their conventional meaning to refer to something without saying it directly. This flip in the meaning of the words prevents Natu- ral Language Processing from revealing the real intention of a communication and, there- fore, reduces the effectiveness of tasks such as Sentiment Analysis or Emotion Detection. In this manuscript we describe the participa- tion of the UMUTeam in HaHackathon 2021, whose objective is to detect and rate humor- ous and controversial content. Our proposal is based on the combination of linguistic features with contextual and non-contextual word em- beddings. We participate in all the proposed subtasks achieving our best result in the con- troversial humor subtask. 1 Introduction In this manuscript we describe our participation in the shared task HaHackathon 2021 (Meaney et al., 2021), proposed in the Forum for Information Re- trieval Evaluation (IberEval’2021) whose objective is the identification and evaluation of humorous and offensive texts written in English. Humor al- lows you to present the reality by highlighting the comic and ridiculous side of life. Humor is hard to identify, even for humans (Vrticka et al., 2013). On the one hand, there are many forms of humor: from mild forms, such as jokes or puns, that result in better and safer social environments, to biting forms, like sarcasm, which is used as a rhetorical device. Humor can also have a constructive end, as happens in satire, where irony, double meaning, and sharp analogies are used to ridicule someone or something. On the other hand, the sharper the hu- mor, the more effort it takes to understand it. When humor is misunderstood, or when humor itself has a transgressive purpose, it can lead to controver- sies and confrontations. Furthermore, an added difficulty is that humor is highly subjective and context dependent, making it even more difficult to understand (Jiang et al., 2019). Nevertheless, the benefits of endowing to a machine basic notions about humor and figurative language understand- ing outweigh the challenges they pose, because they lead to natural language-based interfaces to feel more naturally, such as chat-bots and virtual assistants (Ritschel et al., 2019) and more reliable results in tasks such as opinion mining. We participate in all the proposed subtasks of HaHackathon’2021 with two systems that com- bines linguistic features (LF) extracted with a tool developed by our research group with (1) pre- trained word embeddings (PWE) from fastText (Mikolov et al., 2017) and GloVe (Pennington et al., 2014), and (2) contextual word embeddings from BERT (Devlin …

Content truncated to 500 words