A BERT BASED FRAMEWORK FOR NAMED ENTITY RECOGNITION IN THE KUMAUNI LANGUAGE

Abstract

Named Entity Recognition (NER) is a essential mission in Natural Language Processing (NLP) that entails figuring out and categorizing entities which includes names, places, and groups from a given text. While massive advancements had been achieved for broadly spoken languages, low-resource languages like Kumauni remain underexplored. This observe introduces a BERT-primarily based framework for Named Entity Recognition in the Kumauni language, leveraging the energy of pre-skilled transformer models. The proposed method addresses the challenges posed via limited annotated datasets and the linguistic complexity of Kumauni. Experimental effects exhibit that the BERT-primarily based version substantially outperforms traditional techniques including BiLSTM-CRF, CRF, and rule-based totally structures. The model achieves brand new precision, keep in mind, and F1-score, showcasing its effectiveness in managing low-useful resource languages.