WORD SENSE DISAMBIGUATION USING UNSUPERVISED APPROACH APPLIED ON PUNJABI LANGUAGE

Abstract

Research is being carried out for machines to be able to better decipher an ambiguous word. The majority of work done in Punjabi, a regional language of India and one among the 10 most spoken languages of the world, is limited to supervised techniques. The primary limitation of which is that it requires a sense-tagged corpus and in case an ambiguous word is not listed, the technique will not be able to decipher its meaning. The unsupervised techniques on the other hand uses an untagged corpus and from the instances available with it, it deciphers the correct meaning of the unlisted ambiguous word. This could be either done by clustering instances of the target word or clustering the context of the target word. Using the metrics like precision, recall and f-measure, the results obtained from token based unsupervised methodology were analyzed and compared with decision tree, which showed better outcome than decision tree.