Adding intelligence to non-corpus based word sense disambiguation

Natural language processing applications invariably perform word sense disambiguation as one of its processing steps. The accuracy of sense disambiguation depends upon an efficient algorithm as well as a reliable knowledge-base in the form of annotated corpus and/or dictionaries in machine readable form. Algorithms working on corpus for sense disambiguation are generally employed as supervised machine learning systems. But such systems need ample training on the corpus before being applied on the actual data set. This paper discusses an unsupervised approach of a graph-based technique that solely works on a machine-readable dictionary as the knowledge source. This approach can improve the bottleneck problem that persists in corpus-based word sense disambiguation. The method described here attempts to make the algorithm more intelligent by considering various WordNet semantic relations and auto-filtration of content words before graph generation.