EXTRACTING SEMANTIC INFORMATION FROM ON-LINE SOURCES

In order to achieve broad-coverage semantic capabilities, natural language applications require large lexical databases. This paper describes the work of the Lexical Systems Group at IBM, whose goal is to create such a database, by using (semi-)automatic n1ethods and tools, to extract semantic information from machine-readable dictionaries and thesauri: type-setting tapes are turned into lexical databases representing meaning in hierarchical tree-structures; polysemous terms are disambiguated and indexed with their intended senses; and semantic relations arc extracted from the dictionary. The paper concludes with some applications and open problcn1s.