niksss at SemEval-2022 Task 6: Are Traditionally Pre-Trained Contextual Embeddings Enough for Detecting Intended Sarcasm ?

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 907 - 911 July 14-15, 2022 ©2022 Association for Computational Linguistics niksss at SemEval-2022 Task 6: Are Traditionally Pre-Trained Contextual Embeddings Enough for Detecting Intended Sarcasm ? Nikhil Singh Manipal University Jaipur nikhil3198@gmail.com Abstract This paper presents the 10th and 11th place sys- tem for Subtask A - English and Subtask A - Arabic respectively of the SemEval 2022 - Task 6. The purpose of the Subtask A was to classify a given text sequence into sarcastic and non- sarcastic. We also breifly cover our method for Subtask B which performed subpar when compared with most of the submissions on the official leaderboard . All of the developed solu- tions used a transformers based language model for encoding the text sequences with necessary changes of the pretrained weights and classifier according to the language and subtask at hand . 1 Introduction According to (Yaghoobian et al., 2021), "Sarcasm detection is the task of identifying irony contain- ing utterances in sentiment-bearing texts". Even though sarcastic humor is present throughout social media, it is hard for even humans to comprehend it certainly as the reader doesn’t always perceive it the same way the speaker intended with intri- cate socio-psychological and cultural references. With the way, current models for text representation are trained like Word2Vec (Mikolov et al., 2013) and Glove (Pennington et al., 2014), which have the same representation for a particular word irre- spective of the context they fail to incorporate the meaning of sarcastic texts. Even though the recent state-of-the-art language models provide contex- tual embeddings for words, it still fails to tell the sarcastic humor from normal as we explain later in the paper. There have been a lot of attempts at computa- tionally automating sarcasm detection in the liter- ature. Discernibly, the task of sarcasm detection can be classified into content and context-based methods. With features like Structural, morphosyn- tactic and semantic ambiguity features (Reyes et al., 2012),User mentions (replies), emoticons, N-grams, dictionary- and, sentiment-lexicon-based features(González-Ibáñez et al., 2011)) and fea- tures based on word embedding similarity (Joshi et al., 2016) coming inside content based method. With the surge of seq2seq based model such as BERT (Bidirectional Encoder Representations from Transformers (Devlin et al., 2018), RoBERTa (Liu et al., 2019), and XLNet (Yang et al., 2019), etc. have been heavily employed for sarcasm de- tection in the literature. (Potamias et al., 2020) proposed an R-CNN-RoBERTa which is a hybrid model leveraging RoBERTa’s contextual embed- ding into a recurrent convolutional neural network. (Dadu and Pant, 2020) created an ensemble of RoBERTa and ALBERT (Lan et al., 2019). Seeing the success of pre-trained language model, we de- cided to leverage the pre-trained contextual embed- dings and transformers toward sarcasm detection. While some of the previous textual sarcasm de- tection datasets involved annotation via finding some predefined criteria, such as including specific tags (e.g. sarcasm, irony) (Ptácek et al., 2014) asso- ciated with the text. Other datasets involved manual labeling …

Content truncated to 500 words