Combining Pos Taggers for Improved Accuracy to Create Telugu Annotated Texts for Information Retrieval

POS Tagging is the process of assigning a correct POS tag (can be a noun, verb, adjective, adverb, or other lexical category marker) to each word of the sentence. POS taggers are developed by modeling the morpho-syntactic structure of natural language text. We attempted to improve the accuracy of existing Telugu POS taggers by using an voting algorithm. The three Telugu Pos taggers viz., (1) Rule-based POS tagger (2) Brill Tagger (3) Maximum Entropy POS taggers are developed with an accuracy of 98.016%, 92.146%, and 87.818 respectively. An annotated corpus of 12000 words is used to train the last two taggers. An error analysis is made to find out the errors made by these three taggers and methods to improve the accuracy of these taggers are then examined. As a first step, a voting algorithm is proposed to build an ensemble Telugu POS tagger to get better results. This tagged output could be used for a variety of NLP (Natural Language Processing) applications, mainly used for word sense disambiguation (WSD) is retrieving Telugu documents.