A technique for managing the lexicon in a natural language interface to a changing data base

A difficulty in designing a Natural Language (NL) interface to a Data Base (DB) management system is insuring that DB updates do not obsolete the NL components. Of particular concern is the lexicon (the list of words that the system can process), mainly because NL queries can contain terms that appear as values in the DB, and hence are subject to change as the contents of the DB changes. For example, to process the question "Does John Jones still work for the company?", it is necessary to identify the string "John Jones" as a (potental) value in the "EMPLOYEE-NAME" field (assuming a suitable DB). If such names must appear in the lexicon for the system to process the query, then the lexicon will go out of date as the company's personnel shifts. Using the DB itself as an extension of the lexicon is equally problematic: the system will be unable to parse and provide a negative answer to such a question if the name does not appear at all in the DB. The approach suggested here is to infer a plausible field from the context of the query and semantic information about the domain that is implicitly encoded in the structure of the DB. This technique has been implemented in CO-OP, a NL DB query system that provides cooperative responses and operates with a typical CODASYL DB system. CO-OP treats the problem of selecting a plausible field as a special case of resolving semantic ambiguities. Examples drawn from the implementation are presented.