Deep belief network based part-of-speech tagger for Telugu language

Abstract

Indian languages have very less linguistic resources, though they have a large speaker base. They are very rich in morphology, making it very difficult to do sequential tagging or any type of language analysis. In natural language processing, parts-of-speech (POS) tagging is the basic tool with which it is possible to extract terminology using linguistic patterns. The main aim of this research is to do sequential tagging for Indian languages based on the unsupervised features and distributional information of a word with its neighboring words. The results of the machine learning algorithms depend on the data representation. Not all the data contribute to creation of the model, leading a few in vain and it depends on the descriptive factors of data disparity. Data representations are designed by using domain-specific knowledge but the aim of Artificial Intelligence is to reduce these domain-dependent representations, so that it can be applied to the domains which are new to one. Recently, deep learning algorithms have acquired a substantial interest in reducing the dimension of features or extracting the latent features. Recent development and applications of deep learning algorithms are giving impressive results in several areas mostly in image and text applications.