CS-UM6P at SemEval-2022 Task 6: Transformer-based Models for Intended Sarcasm Detection in English and Arabic

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 844 - 850 July 14-15, 2022 ©2022 Association for Computational Linguistics CS-UM6P at SemEval-2022 Task 6: Transformer-based Models for Intended Sarcasm Detection in English and Arabic Abdelkader El Mahdaouy Abdellah El Mekki Kabil Essefar† Abderrahman Skiredj† Ismail Berrada† School of Computer Science, Mohammed VI Polytechnic University, Morocco abdelkader.elmahdaouy@um6p.ma, abdellah.elmekki@um6p.ma {firstname.lastname}@um6p.ma† Abstract Sarcasm is a form of figurative language where the intended meaning of a sentence differs from its literal meaning. This poses a serious chal- lenge to several Natural Language Processing (NLP) applications such as Sentiment Analysis, Opinion Mining, and Author Profiling. In this paper, we present our participating system to the intended sarcasm detection task in English and Arabic languages. Our system1 consists of three deep learning-based models leveraging two existing pre-trained language models for Arabic and English. We have participated in all sub-tasks. Our official submissions achieve the best performance on sub-task A for Arabic language and rank second in sub-task B. For sub-task C, our system is ranked 7th and 11th on Arabic and English datasets, respectively. 1 Introduction Sarcasm is an important aspect of human natural language. It is characterized by the occurrence of a discrepancy between the intended and the lit- eral meaning of utterance (Wilson, 2006). The prevalence of this phenomenon may jeopardize the performance of many NLP applications, such as Sentiment Analysis, Opinion Mining, and Emo- tion Detection, among others (Maynard and Green- wood, 2014; Rosenthal et al., 2014; Van Hee et al., 2018). Indeed, sarcasm detection has been the sub- ject of many systematic investigation, where sev- eral shared tasks have been organized and a number of datasets have been introduced (Van Hee et al., 2018; Ghanem et al., 2019; Ghosh et al., 2020; Abu Farha et al., 2021). The existing datasets are either labeled by a human annotator or using dis- tant supervision signals such as the presence of a set of predefined hashtags (Oprea and Magdy, 2020). However, these labeling methods might be sub-optimal for intended sarcasm detection as 1The source code of our system is available at https: //github.com/AbdelkaderMH/iSarcasmEval the author’s sarcastic intent may differ from an annotator’s perceived meaning (Oprea and Magdy, 2019, 2020). Besides, most existing research works have focused on English language (Van Hee et al., 2018; Ghosh et al., 2020), while few studies have been introduced for other languages such as Arabic (Ghanem et al., 2019; Abu Farha et al., 2021). To overcome the aforementioned limitations, Abu Farha et al. (2022) have organized the iSar- casmEval shared task for intended sarcasm detec- tion in English and Arabic languages. In contrast to previous research work, the introduced dataset, for intended sarcasm detection, is labeled by the authors themselves. The authors are then asked to provide non-sarcastic rephrases that covey the same intended meaning of their sarcastic texts. Fur- ther, the iSarcasmEval’s organizers have relied on linguistic experts to categorize sarcastic texts into sarcasm, irony, satire, understatement, overstate- ment, and rhetorical questions (Leggitt and Gibbs, 2000). …

Content truncated to 500 words