Development of Part of Speech Tagger for Assamese Using HMM

Abstract

This article presents the work on the Part-of-Speech Tagger for Assamese based on Hidden Markov Model (HMM). Over the years, a lot of language processing tasks have been done for Western and South-Asian languages. However, very little work is done for Assamese language. So, with this point of view, the POS Tagger for Assamese using Stochastic Approach is being developed. Assamese is a free word-order, highly agglutinate and morphological rich language, thus developing POS Tagger with good accuracy will help in development of other NLP task for Assamese. For this work, an annotated corpus of 271,890 words with a BIS tagset consisting of 38 tag labels is used. The model is trained on 256,690 words and the remaining words are used in testing. The system obtained an accuracy of 89.21% and it is being compared with other existing stochastic models.