Information Extraction From Text Through Application of Advanced Processing Techniques

The objective of this Phase I effort is to provide information seekers with rich, contextual, time-stamped information about important people, companies, organizations, countries, inventions, or any other significant proper named entity. Capitalizing on the latest developments in discourse analysis and natural language processing, as well as utilizing the TextWise DR-LINK text retrieval system, this proposal will develop a chronological information extraction system (CHESS), capable of exploiting the rich trail of archived information that exists in many data bases. CHESS will take advantage of the common practice among writers of including information-rich linguistic constructions in close proximity to related proper names. By recognizing the proper names in a text, then locating and recording the associated linguistic construction, it is possible to construct complex, in-depth historiographies of events and their specific relation to a given proper name that can span several decades. This proposal will develop an extraction system for three types of linguistic construction -- appositional, relative clause, and copula sentence -- and will develop a system for mapping proper names and associated linguistic constructions to conceptual categories to produce concept-relation-concept triples. The result will be a system ideally suited for trend analysis, historical issues, domain-dependent scenario analysis, and biography.