Grammatical Tagging for the Kannada Text Documents using Hybrid Bidirectional Long-Short Term Memory Model

Abstract

Kannada is one of the most spoken languages in India. Despite the large usage base, like other major Indian languages, there exist minimal linguistic resources for computing and processing. Rich morphology and agglutinative nature of this language pose a great challenge to even the most basic of natural language processing applications like lemmantization, parts of speech tagging, summarization etc. In this paper, we have discussed a deep learning based perspective} for the grammatical tagging by utilizing hybrid models of bidirectional long short term memory(BDLSTM) and linear chain conditional random fields(CCRF). A database of Kannada documents with 15500 manually tagged words is used for this task. Proposed hybrid model shows a promising result of 81.02%.