Automatic Resolution of Semantic Ambiguity in Natural Language

Most words in natural language have multiple possible meanings. This simple fact causes no end of difficulties for computer systems that seek to understand and generate natural language. The semantic ambiguity of words impacts natural language subtasks such as prepositional phrase attachment and pronoun reference resolution, as well as large-scale applications such as machine translation and information retrieval. Automatic methods that resolve ambiguity in word meaning have the potential to advance the state-of-the-art in natural language processing as a whole, but most approaches to word sense disambiguation have proven difficult to deploy on a wide scale because they are dependent on the availability of specialized sources of knowledge that do not exist across a range of domains. The PI's goal in this project is to develop techniques that will ease and ultimately eliminate knowledge acquisition bottlenecks for word sense disambiguation. He will achieve this by pursuing three specific objectives: 1) develop methods that automatically identify the most relevant contextual features for determining the sense of any ambiguous word; 2) develop disambiguation algorithms that learn from 'just a few' manually created examples; and 3) develop unsupervised methods that allow any set of word meanings to serve as the target of the disambiguation process. The combined effect of meeting these objectives will be to liberate word sense disambiguation from dependence on particular knowledge sources and thereby simplify their integration into natural language processing systems