Clustering WordNet Senses Utilizing Modified and Novel Similarity Metrics CS 229 Final Project Report

Introduction We approach the problem of clustering senses in Princeton's WordNet (Fellbaum 1998), a manually created dictionary/thesaurus which attempts to model the structure underlying human concepts. A synset, the fundamental unit in WordNet, is represented by a group of synonyms and a gloss definition, and is connected through a variety of semantic links, such as hypernyms (type-of) or meronyms (part-of), to other synsets. A particular word is associated with one or more synsets, each representing a particular sense of the word. While this electronic database provides an inventory with which to do Word Sense Disambiguation, the fine-grainedness of the senses – which sometimes even humans have trouble distinguishing between has posed a problem in achieving reasonable performance.