Telugu named entity recognition using bert

Named entity recognition (NER) is a fundamental step for many Natural Language Processing tasks that aim to classify words into a predefined set of named entities (NE). For high-resource languages like English, many deep learning architectures have produced good results. However, the NER task has not yet achieved much progress for Telugu, a low resource Language. This paper performs the NER task on Telugu Language using Word2Vec, Glove, FastText, Contextual String embedding, and bidirectional encoder representations from transformers (BERT) embeddings generated using Telugu Wikipedia articles. These embeddings have been used as input to build deep learning models. We also investigated the effect of concatenating handcrafted features with the word embeddings on the deep learning model’s performance. Our experimental results demonstrate that embeddings generated from BERT added with handcrafted features have outperformed other word embedding models with an F1-Score 96.32%.