Unsupervised Hindi Word Sense Disambiguation based on Network Agglomeration

Word sense disambiguation (WSD) is an essential task in computational linguistics for language understanding applications such as information retrieval, question answering, machine translation, text summarization etc. In this paper we propose an unsupervised WSD method for a Hindi sentence based on network agglomeration. First we create the sentence graph G for the given sentence. This sentence graph collectively represents all the interpretations of the sentence. Now from this sentence graph G we create the interpretation graph G' subset of G for each of the interpretation of the sentence. To identify the desired interpretation we compute network agglomeration for all the interpretation graphs. Thus the relevant interpretation having highest value of network agglomeration is identified. The results on the standard sense tagged corpus show better performance for the proposed method than the previous approaches.