Ontology-Based Understanding of Natural Language Queries Using Nested Conceptual Graphs

In a question answering system, users always prefer entering queries in natural language and not being constrained by a rigorous grammar. This paper proposes a syntax-free method for natural language query understanding that is robust to ill-formed questions. Nested conceptual graphs are defined as a formal target language to represent not only simple queries, but also connective, superlative, and counting queries. The method exploits knowledge of an ontology to recognize entities and determine their relations in a query. With smooth mapping to and from natural language, conceptual graphs simplify conversion rules from natural language queries and can be easily converted to other formal query languages. Experimental results of the method on the QA track datasets of TREC 2002 and TREC 2007 are presented and discussed.