IISERB Brains at SemEval-2022 Task 6: A Deep-learning Framework to Identify Intended Sarcasm in English

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 938 - 944 July 14-15, 2022 ©2022 Association for Computational Linguistics IISERB Brains at SemEval-2022 Task 6: A Deep-learning Framework to Identify Intended Sarcasm in English Tanuj Singh Shekhawat∗1, Manoj Kumar∗2, Udaybhan Rathore∗3, Aditya Joshi4, Jasabanta Patro5 1235 Indian Institute of Science Education and Research, Bhopal 4 SEEK, Australia {1tanuj19, 2manoj19, 3udaybhan19, 5jpatro}@iiserb.ac.in, 4ajoshi@seek.com.au Abstract This paper describes the system and the re- sulted models submitted by our team "IISERB Brains" to SemEval-2022 Task 6 competition. We participated in the three sub-tasks for En- glish language datasets. Our submitted models use BERT-based classifiers along with data aug- mentation. We ranked 19th out of 43 teams for sub-task A, 8th rank out of 22 teams for sub-task B, and 13th rank out of 16 teams for sub-task C. In this paper, we describe de- tails of our submissions, related evaluation and additional experiments conducted post the ter- mination of the shared task. Our code and used additional resources are present in GitHub1 for reproducibility. Authors with equal contribu- tions are marked by *. 1 Introduction Sarcasm in spoken or written form is a type of verbal irony that indicates the difference between the literal and intended meanings of an utterance (Joshi et al., 2017). While many people use it as a bitter remark to mock or ridicule a target(Patro et al., 2019), some also use it as a joke to amuse others(Joshi et al., 2015a). Sarcasm often used together or interchangeably with other ironic cat- egories, is considered an essential component of human communication. A large portion of the web and social media text is sarcastic, which creates a challenge for traditional natural language pro- cessing (NLP) tasks like sentiment classification, opinion mining, harassment detection, author pro- filing etc. Since these systems are deployed widely across various industries, administration, data an- alysts etc, designing a robust sarcasm detecting component would help the downstream tasks sub- stantially. The SemEval-2022 task 6(Abu Farha et al., 2022) identifies some of the challenges per- sisting till now, particularly in English and Arabic 1https://github.com/manojmahan/iSarca smEval-Intended-Sarcasm-Detection-In-Eng lish-main texts. Our team participated in the tasks floated for the English language under the name "IISERB Brains". The organizers have floated three sub- tasks: (i)Sub-task A to detect whether a given text is sarcastic, (ii)Sub-task B to identify which ironic category the sarcastic tweet belongs to, and (iii) Sub-task C to given two pieces of text, identify the sarcastic one. In this paper, we report the details of our par- ticipating systems and their performance on the evaluation data. The paper also contains details of additional experiments that were a part of our par- ticipation in the shared task. The paper is organised as follows. Section 2 describes past work related to the sub-tasks, while section 3 presents the dataset and task details. Following that, Section 4 reports the details of our systems followed by experiments and results in Section 5. Finally, in section 6, we conclude the paper …

Content truncated to 500 words