Bengali POS Tagging Using Bi-LSTM with Word Embedding and Character-Level Embedding

Part-of-speech tagging (POS) is an important and very fundamental process in natural language processing (NLP). POS tagging is required as a preprocessing task in many types of linguistic research such as named entity recognition (NER), word sense disambiguation, information extraction, natural language translation, and sentiment analysis. In this paper, we propose a practical Bengali POS tagger, which takes as input a text written in Bengali and gives a POS tagged output. In recent times, Bi-LSTM networks have been proven effective in sequential data processing but not very much tested on resource-poor and inflectional languages like Bengali. This paper addresses the issues of the POS tagging task for the Bengali language using Bi-LSTM with transfer learning by applying pre-trained word embedding information. The POS tagged output from our proposed model can be used directly for other applications of Bengali language processing as our proposed tagger can also handle out-of-vocabulary (OOV) words. Our experiment reveals that Bi-LSTM with transfer learning is effective for tagging Bengali documents.