Multi-Strategy Construction-Specific Parsing for Flexible Data Base Query and Update

The advantages of a multi-strategy, construction-specific approach to parsing in applied natural language processing are explained through an examination of two pilot parsers we have constructed. Our approach exploits domain semantics and prior knowledge of expected constructions, using multiple parsing strategies each optimized to recognize different construction types. It is shown that a multi strategy approach leads to robust, flexible, and efficient parsing of both grammatical and ungrammatical input in limited-domain, task oriented, natural language interfaces. We also describe plans to construct a single, practical, multi-strategy parsing system that combines the best aspects of the two simpler parsers already implemented into a more complex, embedded-constituent control structure. Finally, we discuss some issues in data base access and update, and show that a construction-specific approach, coupled with a case structured data base description, offers a promising approach to a unified, interactive data base query and update system.