Conceptual Clustering of Korean Concordances Using Similarity between Morphemes

This paper describes a method for the conceptual clustering of Korean concordances. We present a method of computing conceptual similarity between concordances using the number of cooccurring morphemes and the similarities between morphemes. We use mutual information, the similarity between mutual information values and vector similarity to compute similarity between morphemes. When we try to cluster the concordances extracted from about 170,000 word size part of speech tagged corpus, the system correctly clusters 90.16% of concordances. The proposed method has shown that the information about similarity between morphemes can be used to disambiguate word senses.