Team LRL_NC at SemEval-2022 Task 4: Binary and Multi-label Classification of PCL using Fine-tuned Transformer-based Models

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 421 - 431 July 14-15, 2022 ©2022 Association for Computational Linguistics Team LRL_NC at SemEval-2022 Task 4: Binary and Multi-label Classification of PCL using Fine-tuned Transformer-based Models Kushagri Tandon, Niladri Chatterjee Indian Institute of Technology Delhi Hauz Khas, Delhi-110016, India {kushagri.tandon,niladri}@maths.iitd.ac.in Abstract Patronizing and condescending language (PCL) can find its way into many mediums of public discourse. Presence of PCL in text can pro- duce negative effects in the society. The chal- lenge presented by the task emerges from the subtleties of PCL and various data dependent constraints. Hence, developing techniques to detect PCL in text, before it is propagated is vital. The aim of this paper is twofold, a) to present systems that can be used to classify a text as containing PCL or not, and b) to present systems that assign the different categories of PCL present in text. The proposed systems are primarily rooted in transformer-based pre- trained language models. Among the models submitted for Subtask 1, the best F1-Score of 0.5436 was achieved by a deep learning based ensemble model. This system secured the rank 29 in the official task ranking. For Subtask 2, the best macro-average F1-Score of 0.339 was achieved by an ensemble model combining transformer-based neural architecture with gra- dient boosting label-balanced classifiers. This system secured the rank 21 in the official task ranking. Among subsequently carried out ex- periments a variation in architecture of a sys- tem for Subtask 2 achieved a macro-average F1-Score of 0.3527. 1 Introduction The aim of the current task, viz. Patronizing and Condescending Language Detection (Pérez- Almendros et al., 2022), is to identify presence of condescending and patronizing tones in text, par- ticularly by the media when referring to vulnerable communities. Preponderance of PCL in news and different social media texts, often targeted towards marginalised and under represented communities is a major social concern these days. It can feed stereotypes, tilt the scales of superiority towards a particular community, and fuel discriminatory behaviour. Thus the task of identifying PCL partic- ularly that directed to vulnerable communities is a crucial task. The task is based on the English language Don’t Patronize Me! dataset (Pérez-Almendros et al., 2020). It consists of two subtasks, binary classi- fication (Subtask 1) and multi-label classification (Subtask 2). In Subtask 1, given a paragraph the aim is to predict whether the paragraph consists of any form of patronizing and condescending lan- guage (PCL). In Subtask 2, the aim is to assign each paragraph a subset of labels which express different categories of PCL. The major challenges offered in such tasks emerge from various linguistic aspects, such as use of cryptic sentences, sarcasms used, polysemous nature of the English words among oth- ers. For the present task, further challenge emerges from the imbalance of dataset as discussed in Sec- tion 2. Experiments were conducted with various sys- tems and the best performing ones for both the subtasks are discussed in detail. The two systems …

Content truncated to 500 words