WIDE COVERAGE SYNTACTIC PROCESSING

FOR MANY LANGUAGE PROCESSING APPLICATIONS, WIDE COVERAGE OF THE LANGUAGE IS IMPORTANT. THAT IS, IT IS IMPORTANT FOR THEPARSER TO ASSIGN REASONABLE STRUCTURES TO MANY STRINGS OF THE LANGUAGE. WIDE COVERAGE HAS NOT BEEN ACHIEVED IN SPITE OF MANY PERSON-YEARS OF EFFORT. THE GRAMMARS OF WIDE COVERAGE PARSERS TEND TO BECOME LARGE, MORE DIFFICULT TO EXTEND, AND LESS EFFICIENT AS COVERAGE OF THE LANGUAGE GETS BETTER. THIS PROJECT WILL DEMONSTRATE THAT THIS BOTTLENECK CAN BE AVOIDED BY AN EFFICIENT PARSER THAT IS DESIGNED TO ACCEPT A SUPERSET OF THE LANGUAGE. THE PROBLEM THAT SUCH SYSTEMS HAVE ENCOUNTERED IN THE PAST COMES NOT FROM THE ACCEPTANCE OF UNGRAMMATICAL STRINGS, BUT FROM THE DETECTION OF SPURIOUS AMBIGUITY IN GRAMMATICAL STRINGS. THIS PROBLEM IS AVOIDED ON THE PRESENT APPROACH BY USING NEW AND MORE POWERFUL DISAMBIGUATION STRATEGIES BASED ON A MORE SYSTEMATIC REPRESENTATION OF SEMANTIC PROPERTIES AND COMMONSENSE KNOWLEDGE THAN HAS BEEN AVAILABLE BEFORE, TOGETHER WITH POWERFUL AUTOMATIC REASONING TECHNIQUES. PHASE I WILL SPECIFY THE DESIGN OF THIS SYSTEM IN DETAIL ANDUSE A CAREFUL ASSESSMENT OF PROTOTYPE PERFORMANCE TO ESTABLISH THE FEASIBILITY OF THE PROPOSED DESIGN FOR COMMERCIAL APPLICATIONS. IN PHASE II THE SPECIFIED DESIGN WILL BE BUILT AND TESTED IN TARGET APPLICATIONS.