YNU-HPCC at SemEval-2022 Task 6: Transformer-based Model for Intended Sarcasm Detection in English and Arabic

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 956 - 961 July 14-15, 2022 ©2022 Association for Computational Linguistics YNU-HPCC at SemEval-2022 Task 6: Transformer-based Model for Intended Sarcasm Detection in English and Arabic Guangmin Zheng, Jin Wang and Xuejie Zhang School of Information Science and Engineering Yunnan University Kunming, China Contact: gmzheng@mail.ynu.edu.cn, {wangjin, xjzhang}@ynu.edu.cn Abstract In this paper, we (a YNU-HPCC team) describe the system we built in the SemEval-2022 com- petition. As participants in Task 6 (titled “iSar- casmEval: Intended Sarcasm Detection In En- glish and Arabic”), we implement the sentiment system for all three subtasks in English and Arabic. All subtasks involve the detection of sarcasm (binary and multilabel classification) and the determination of the sarcastic text loca- tion (sentence pair classification). Our system primarily applies the sequence classification model of a bidirectional encoder representation from a transformer (BERT). The BERT is used to extract sentence information from both direc- tions for downstream classification tasks. A sin- gle basic model is used for single-sentence and sentence-pair binary classification tasks. For the multilabel task, the Label-Powerset method and binary cross-entropy loss function with weights are used. Our system exhibits com- petitive performance, obtaining 12/43 (21/32), 11/22, and 3/16 (8/13) rankings in the three official rankings for English (Arabic). 1 Introduction Satirical text is a rhetorical device for implicitly expressing emotions by using words that are con- trary to the actual intention to achieve a satirical or humorous linguistic effect. The true semantics of satirical texts cannot be directly inferred from the text vocabulary, and contradictions exist between their literal meaning and the true intention. There- fore, the detection of sarcasm and its sentiment discrimination are more challenging in natural lan- guage processing (NLP) problems. Task 6 in the SemEval-2022 competition is a sar- casm detection task that consists of three subtasks (Abu Farha et al., 2022). • Subtask A: Detect sarcastic meaning from a given tweet. • Subtask B: Identify the tweet as “no sarcasm” or one or more of the six given sarcastic speech categories. • Subtask C: Determine the position of the satir- ical text from the given tweets and their non- satirical restatements (0 means the first sen- tence is satirical, and 1 means the second sen- tence is satirical). In previous sarcasm detection tasks, researchers used supervised learning methods based on the support vector machines and logistic regression (González-Ibáñez et al., 2011) to study ironic and non-ironic tweets that directly express positive and negative views. However, these traditional machine learning methods cannot mine the deep semantic information hidden in the text. Relying only on the surface semantic information can easily result in an incorrect judgment regarding irony. In contrast, deep learning methods show excellent results in deep semantic mining. By using the context in- formation of the text to be detected (Bamman and Smith, 2015), we can further mine the behavior information of social users. We can achieve better performance by using a bidirectional recurrent neu- ral network to capture the …

Content truncated to 500 words