Ted Pedersen: 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. Existing methods that resolve ambiguity in word meaning 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 long-term goal of this research is to develop techniques that will liberate word sense disambiguation from these knowledge acquisition bottlenecks and thereby simplify their integration into natural language processing systems.

I will describe recent progress in meeting three specific objectives that move us closer to this goal: 1) the development of methods to discover the most relevant contextual features for determining the sense of an ambiguous word; 2) the development of disambiguation algorithms that learn from "just a few" manually created examples; and 3) the development of unsupervised methods that allow any set of word meanings to serve as the target of the disambiguation process.

Back to Clippers page


For questions or comments regarding this page, please contact: Detmar Meurers
Last modified: Mon Apr 22 10:32:46 EDT 2002