The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9902

Title:Word Sense Disambiguation Using Common Sense Knowledge
Date: 1999
Abstract:Word sense disambiguation is the process of determining which sense of a polysemous word is intended in a given context. Cyc [note 1] is a very large knowledge base which offers the possibility of overcoming the difficulties involved in scaling knowledge driven disambiguation systems to go beyond the restricted domain. This possibility was investigated in two ways. First, Cyc was used to rule out certain possible senses for apolysemous word on the grounds that their use would violate constraints imposed by the surrounding text and Cyc's ``Common Sense'' definitions. Second, Cyc's knowledge base was regarded as a large semantic network [note 2]. It was assumed that the intended sense of a polysemous word has a closer semantic relationship to other concepts in the context than do the unintended senses. Hence, choosing that one of the word senses which is closest to the context in terms of distance across the network would be a good disambiguation method. It was concluded that the first method would require a very large investment of effort to be made general, although it would be the more effective of the two if this was done. It was also found that the structure of Cyc makes its use as a semantic network problematic. Thefailure of this approach seems to be due to lack of knowledge in our version of Cyc and inefficiencies in the search techniques. Varioussuggestions for future improvements are made.

[Search These Pages] [DAI Home Page] [Comment]