Development of a Virtual View for Processing Complex Natural Language Queries

Abstract

Natural language interfaces to databases (NLIDBs) allow inexperienced users to formulate questions in natural language (NL) to DB management systems. A survey of the literature on NLIDBs shows that the prevalence of the interfaces developed to date are limited by two aspects: it is difficult to port them from one database (DB) to another, and they fail on some questions, either because they misinterpret the questions or they are unable to respond even when the answer exists. Among the problems that occur in NL questions, there is a problem that may arise when querying complex DBs: semantically implied entities. This problem is related to the semantic meaning of a query, when upon referring to an entity (DB table), another entity (or entities) is semantically implied to which the first entity is related, and usually the user ignores the relationship between the two (or more) entities. This chapter describes a method for using a virtual view in a NLIDB, which allows processing queries that involve this problem more efficiently than a previous version of the interface that uses materialized views. Specifically, experimental results show that the new version reduces by 23% the time for translating questions from NL to SQL.