| Many Natural Language Processing applications need to establish relations between words. Some of these relations are provided in lexical resources like WordNet and its extensions. The research in this thesis focuses on finding the semantic connectivity between pairs of words. The approach is to identify semantic paths between concepts called lexical chains. This becomes possible when the glosses in WordNet are transformed into semantic relations and the gloss concepts are disambiguated. This extended WordNet becomes a large semantic graph on which connectivity between concepts can be traced. The research presented here addresses the problems of filtering out meaningless paths, ranking the remaining paths, labeling paths with appropriate semantic names, and identifying the sources of errors. Lexical chains can be viewed as queering mechanisms on knowledge bases and we used them in measuring the semantic similarity and relatedness of words. |
