Universal Database Natural Language Interface Design Methods

Natural language interface for database(NLIDB), which is a newly kind of intelligentinterface, allows the user to interact with database system in natural language. NLIDB is thespecificapplicationofnaturallanguageprocessingtechniqueindatabasefield.The traditional methods of NLI designing include pattern-matching, syntax analysis,semantic grammar, and intermediate representation languages. Each of these methods hasadvantage and disadvantage, but none of them can construct a robust and transportable NLIsystematlowcost.To meet this, we think out lots of new methods, with which we can build a robust, naturallanguage-independent,domain-independentanddatasource-independentsystemquickly.First, we give a new definition to"meaning"according to the understanding-pattern ofhuman being. Basedonthat,wefindoutasemanticcoherencemethod,whichissimple,efficient,robust and independent of special natural language. To meet the requirement ofdomain-independent, we introduced ontology methodology to semantic coherenceand present asystemic ontology semantic coherence approach. We have completed the formalization work ofthisapproach,designedthealgorithmandmadeanevaluationofit.Second, we present a new kind concept ---- virtual ontology data source to solve theproblem of data source-independent. Virtual ontology data source shields the detail of low-leveldatasource,andprovides aconsistentaccessinterfacefortopperusers.Wediscussedthemethodof transforming relation database to virtual ontology data source (which is also called databasedeep annotation). We designed the database schema description language and mapping rulesdescriptionlanguage,andgavearesolventforrelationmodeling.At last, we bring out a framework of general natural language interface, and implement adatabasenaturallanguageinterface.Makinguseofaspecificexample,wemadeanevaluationofthissystem.