Using Automatic Word Sense Discrimination to generate a Semantic Lexicon

Automatic word sense discrimination is the process of distinguishing the number of unique senses of a target word in a given corpus. This work approaches word sense discrimination as an unsupervised clustering problem on the context of the target word in web documents. Using the features from the computed clusters, the system constructs a new lexicon entry for the target word which includes the semantic and syntactic constraints for each discriminated sense. The lexicon entries are evaluated for precision and recall against sense inventories created from three human dictionaries. This work:

• Uses syntactic and semantic features not in the current word sense discrimination literature
• Uses a corpus not in the current word sense discrimination literature (web documents)
• Generates lexicon entries based on the results of word sense discrimination

MS Thesis Committee:

• Sergei Nirenburg
• Marjorie McShane
• Stephen Beale