The purpose of this contract is to develop methods to analyze and represent information in biomedical texts. This will require sophisticated natural language processing capabilities, involving lexical, syntactic, semantic and pragmatic analysis of these texts. The Natural Language Systems group of the Lister Hill Center for Biomedical Communications is pursuing this work as an intramural research project and seeks to collaborate with outside organizations presently conducting closely related research. The overall objective of this contract is to establish methods for testing the hypothesis that access to a bibliographic database, such as the National Library of Medicine's MEDLINE database, can be improved by automated analysis of the free test in the system. The project work will involve modification and extension of aspects of a natural language parser. The MEDLINE database has citation records for several million articles in biomedicine, representing several thousand journals. Each citation record includes the title, an author prepared abstract when available, author and journal names, and a set of Medical Subject Headings under which the article has been indexed by expert indexers. The free text in the system is found in the title and abstract fields of the citation records. Titles are normally complex noun phrases, while abstracts are composed of well-formed, although highly specialized, English sentences. The purpose of the work under this contract is to develop methods for parsing this text. The parsing procedure should result in a set of well-specified logical forms, representing the meaning of the phrases and sentences in the citation record.
