SSN_NLP_MLRG at SemEval-2022 Task 4: Ensemble Learning strategies to detect Patronizing and Condescending Language

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 400 - 404 July 14-15, 2022 ©2022 Association for Computational Linguistics SSN_NLP_MLRG at SemEval-2022 Task 4: Ensemble Learning strategies to detect Patronizing and Condescending Language Kalaivani Adaikkan and Durairaj Thenmozhi Department of Computer Science and Engineering Sri Sivasubramaniya Nadar College of Engineering, Chennai, India kalaiwind@gmail.com, theni_d@ssn.edu.in Abstract In this paper, we describe our efforts at Se- mEval 2022 Shared Task 4 on Patronizing and Condescending Language (PCL) Detection. This is the first shared task to detect PCL which is to identify and categorize PCL language to- wards vulnerable communities. The shared task consists of two subtasks: Patronizing and Condescending language detection (Subtask A) which is the binary task classification and iden- tifying the PCL categories that express the con- descension (Subtask B) which is the multi-label text classification. For PCL language detection, We proposed the ensemble strategies of a sys- tem combination of BERT, Roberta, Distilbert, Roberta large, Albert achieved the official re- sults for Subtask A with a macro f1 score of 0.5172 on the test set which is improved by baseline score. For PCL Category identifica- tion, We proposed a multi-label classification model to ensemble the various Bert-based mod- els and the official results for Subtask B with a macro f1 score of 0.2117 on the test set which is improved by baseline score. 1 Introduction Social media is a wide platform and it grows rapidly. People can communicate with each other and express their opinions easily without any hesita- tion on social media. Patronizing and Condescend- ing Language (PCL) 1 is language use shows a superior attitude rise towards vulnerable communi- ties in the social media. This effect is unconscious and the intention is trying to help communities like an individual, group of people in need by raising awareness, moving the user audience to action, and standing for the rights. However, these dominant attitudes can lead to discrimination and the user audience is unaware of this diminishing treatment due to its subtlety. Moreover, online social media publications reached more audiences in day-to-day life and we noticed that diminishing treatment of 1https://www.merriam-webster.com/dictionary/patronizing vulnerable groups leads to greater inequalities. so, PCL can potentially be very harmful, as it feeds stereotypes, routinizes discrimination, and drives to greater exclusion. Detecting PCL language and its categorization of PCL language on social media have gained a lot of interest recently. The detection of PCL is still an emergent area of study in NLP. To the best of my knowledge, this is the first shared task 2 to detect the PCL and their categories from the vulnerable communities. The challenge is to detect that PCL is difficult both for humans and NLP systems, due to its subtle nature, and its subjectivity reasoning required to understand this kind of language. Se- mEval 2022 task 4 presents the problems of detect- ing PCL and its categorizes of PCL which express the condescending language in English tweets to the NLP community. The PCL shared task consists …

Content truncated to 500 words