PBS - AN ECONOMICAL NATURAL-LANGUAGE QUERY INTERPRETER

A natural language interface, named PBS, to a bibliographic informatio n retrieval system, is described, which receives natural language sear ch statements from an end-user and identifies search facets and implie d Boolean logic operators. The program is intended for use in conjunct ion with an intermediary program (OAK) for online database searching, operating in a user's personal computer. Such a program must be fast a nd make minimal demands on memory, which places severe restrictions on storage of dictionaries and thesauri needed for full natural language interpretation. This initial version deals with a single domain, bibl iographic files, almost all of which use the same attributes, such as subject, author, or journal name. The objective of PBS is to identity terms in a natural language query, classify them by attribute, recogni ze the boolean logic stated or implied, conflate terms that have minor spelling variations, and present the terms to the intermediary progra m, organized by facet or term category. In testing, the main difficult y resulted from false identification of journal names. Yet, in light o f the editorial capabilities of OAK, which allow a user to correct a q uery before it is given to an IRS, PBS can be regarded as a workable n atural language interface that can alleviate some of the problems inhe rent in native command languages.