“Part of Speech Tagging – A Corpus Based Approach”

Abstract

POS tagging, an ideal way to augment a corpus is an imperative abstraction for text mining. However with an increase in the amount of linguistic errors and distinctive fashion of language ambiguities, the data filtered by POS tagging is noisier. In this paper, probabilistic tagging and tagging based on Markov models are combined to estimate the association probabilities. Based on this combined approach, error estimation model is defined. Comparison study is made on different corpus available in NLTK such as Crubadan, Brown and INSPEC. The results obtained by the proposed methodologies show a drastic increase in the accuracy rate of about 98% when compared to the existing algorithms which shows an average of 96% accurate. The performance measure is plotted to calculate the error ratio across the maximum-likelihood estimation.