Deep Learning Techniques for Part of Speech Tagging by Natural Language Processing

For many natural language processing applications, part of speech (POS) tagging remains as a preliminary task. Marathi, is observed as a popular language in India but it only has limited tools and corpus for NLP applications. An accurate POS tagger is essential for many NLP tasks like sentiment analysis, named entity recognition, dependency parsing, etc. This research work proposes a deep learning model and bidirectional long short-term memory (Bi-LSTM) model to perform POS tagging for Marathi text. We achieved an accuracy of 85% for the deep learning model and 97% for the Bi-LSTM model. Our contribution here is based on three folds - building a deep learning model, building the Bi-LSTM model, comparison with machine learning techniques for the same dataset.