Mono Versus Multilingual BERT: A Case Study in Hindi and Marathi Named Entity Recognition

Named entity recognition (NER) is the process of recognizing and classifying important information (entities) in text. Proper nouns, such as a person’s name, an organization’s or location’s are examples entities. The NER one modules applications like human resources, customer support, search engines, content classification, academia. In this work, we consider for low-resource Indian languages Hindi Marathi. transformer-based models have been widely used tasks. We different variations BERT base-BERT, RoBERTa, AlBERT benchmark them on publicly available Marathi datasets. provide exhaustive comparison monolingual multilingual establish simple baselines currently missing literature. show that MahaRoBERTa model performs best whereas XLM-RoBERTa NER. also perform cross-language evaluation present mixed observations.