• Yogender Kumar
• Dr Arun K Khosla
• Dr Geeta Sikka

Recent advancements in neural network-based language representation have facilitated the transfer of learned internal states from trained models to various downstream natural language processing tasks, including Part of Speech (POS) Tagging and question answering. It has demonstrated the notable improvements achieved by leveraging pre-trained language models, particularly when labeled data is limited. We collected our dataset from the publicly accessible Indian Languages Corpora Initiative (ILCI) phase-II project, which encompasses 28,733 sentences from six diverse text domains, namely science and technology, religion, health, entertainment, sports, and agriculture. The dataset has been meticulously annotated using the Bureau of Indian Standards (BIS) tagset, which includes a comprehensive set of 34 grammatical categories. It consists of 25,859 sentences for training, 1,437 sentences for validation, and 1,437 sentences for the test set. To best of our knowledge this is a first attempt to make a part of speech tagger using transformer architecture for low resource Punjabi language. With only 28733 sentences we have developed a system that achieved an F1 score of 84.46% on unseen data for the POS tagging task from six different domains.

Part of speech tagging, Transformers, IndicBERT, BiLSTM, Natural language processing

"Gurmukhi Punjabi Part of speech tagging using IndicBERT-BiLSTM architecture", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.10, Issue 7, page no.a595-a603, July-2023, Available :http://www.jetir.org/papers/JETIR2307073.pdf

"Gurmukhi Punjabi Part of speech tagging using IndicBERT-BiLSTM architecture", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.10, Issue 7, page no. ppa595-a603, July-2023, Available at : http://www.jetir.org/papers/JETIR2307073.pdf