Nepali POS Tagging Using Deep Learning Approaches

Deep Learning approaches are being extensively used in Part of Speech (POS) tagging. POS tagging is one of the important step in Natural Language Processing (NLP) including Machine Translation, Retrieval of Information, developing question answering system, word sense disambiguation, text summarization, Named Entity Recognition, text to speech conversion and classification. The efficiency of POS tagging heavily rely on syntactic, contextual information and morphology of the language. POS tagging in Nepali Language is very difficult as it is morphologically rich. This research paper focuses on implementing and comparing various deep learning approaches for POS tagging in Nepali Language. Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM) and Bidirectional LSTM were implemented in tagged Nepali corpus. The result of Bidirectional LSTM (Bi-LSTM) was better than other approaches. Keywords: POS, NLP, RNN, GRU, LSTM, Bi-LSTM, Nepali corpus