1 Rerendering Top Ontologies

Data mining and information extraction rely on a number of natural language tasks that require semantic typing; that is, the ability of an application to accurately determine the conceptual categories of syntactic constituents. Semantic typing serves tasks such as relation extraction by improving anaphora resolution and entity identification. Domain-specific semantic typing also benefits statistical categorization and disambiguation techniques that require generalizations across semantic classes to make up for the sparsity of data. This applies, for example, to the problem of prepositional attachment, as well as identification of semantic relations between constituents within nominal compounds (see, for example, related discussion in Rosario and Hearst (2001)). Semantic typing has other direct applications, such as query reformulation, the filtering of results according to semantic type restrictions. The set of categories used in semantic typing must be adequate enough to serve such tasks. In the biomedical domain, there are a number of efforts to develop specialized taxonomies and knowledge