Enhancement of Fine-Grained Part-of-Speech Tagging for Nepali Text using BiLSTM-CRF and Word2Vec

Abstract

Part-of-speech (POS) tagging is an essential and foundational task in numerous Natural Language Processing (NLP) applications, including machine translation, sentiment analysis, text-to-speech conversion, speech recognition, text summarization, question answering, information retrieval, word sense disambiguation, and Named Entity Recognition. POS tagging entails assigning the correct tag to each token in the corpus, considering its context and the language's syntax. An optimal POS tagger plays a crucial role in computational linguistics. Its importance cannot be emphasized enough because inaccuracies in tagging can greatly affect the performance of complex natural language processing systems. In this work, a deep learning algorithm, BiLSTM with CRF, has been implemented for fine-grained Nepali POS tagging. The Gensim’s Word2Vec word embedding has been trained on sentences and has been used as the embedding layer while creating the model. Additionally, BiLSTM-CRF with Word2Vec has been compared to the well-known models GRU, LSTM, and BiLSTM. BiLSTM-CRF with the Word2Vec word embedding performed the best and achieved a new state of the art F1 score of 99.81% for fine grained Nepali text POS tagging on the Nepali Monolingual Written Corpus.