Thirumurai: A Large Dataset of Tamil Shaivite Poems and Classification of Tamil Pann

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6556–6562 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6556 Thirumurai: A Large Dataset of Tamil Shaivite Poems and Classification of Tamil Pann Shankar Mahadevan1, Rahul Ponnusamy2, Prasanna Kumar Kumaresan2, Prabakaran Chandran3, Ruba Priyadharshini4, Sangeetha Sivanesan5, Bharathi Raja Chakravarthi6 1Thiagarajar College of Engineering, Madurai, India 2Indian Institute of Information Technology and Management - Kerala, India 3Mu Sigma Inc., 4ULTRA Arts and Science College, Madurai, India 5National Institute of Technology, Tiruchirappalli, India 6Insight SFI Research Centre for Data Analytics, National University of Ireland Galway shankarmahadevan12901@gmail.com, {rahul.mi20, prasanna.mi20}@iiitmk.ac.in, prabakaran.chandran98@gmail.com, rubapriyadharshini.a@gmail.com, sangeetha@nitt.edu, bharathi.raja@insight-centre.org Abstract Thirumurai, also known as Panniru Thirumurai, is a collection of Tamil Shaivite poems dating back to the Hindu revival period between the 6th and the 10th century. These poems are par excellence, in both literary and musical terms. They have been com- posed based on the ancient, now non-existent Tamil Pann system and can be set to music. We present a large dataset containing all the Thirumurai poems and also attempt to classify the Pann and author of each poem using transformer based architectures. Our work is the first of its kind in dealing with ancient Tamil text datasets, which are severely under-resourced. We explore sev- eral Deep Learning-based techniques for solving this challenge effectively and provide essential insights into the problem and how to address it. Keywords: Thirumurai, Pann Classification, Tamil NLP 1. Introduction NLP has advanced by leaps and bounds since the introduction of Deep Learning, particularly in the English language. This is mainly attributable to the vast amount of data for English that is available. Every second, massive volumes of data are generated from a variety of sources all across the internet. Compa- nies with access to such large amounts of data are contributing significantly to the NLP community by training models with ever-increasing parameters, such as the trillion parameter Switch transformer (Fedus et al., 2021). Because Deep Learning is a data-driven learning approach, a large data corpus is required to build good NLP systems. As the number of parameters in NLP models grows beyond trillions, they become increasingly resource and power hungry. Such models are trained on petabytes of data, such as the Common Crawl data, which has over 12 years of internet data. Due to the size of the data they are being trained on, this not only improves the accuracy of the models, but it also allows them to generalize to new, previously unknown tasks. There exists different kinds of datasets for different tasks like the Stanford Question Answering Dataset (SQuAD) (Rajpurkar et al., 2018) for question answering, Amazon Product Dataset (142.8 million reviews spanning May 1996 – July 2014) for sentiment analysis, WordNet database (Miller, 1995) for text classification, dataset of research papers from Arxiv which has been used to generate abstracts for research papers automatically and many other datasets like this. These types of datasets have aided us in preventing hate speech, obtaining …

Content truncated to 500 words