Aspects of the automatic generation of SQL statements in a natural language query interface

摘要

In Natural Language Interfaces to databases it is currently a widely accepted technique to use an intermediate semantic representation of the NL sentences which is more or less based on First Order Predicate Logic, eventually augmented by some specific functions inherent to the database query language like, e.g. the aggregation functions of SQL. However, in order not to restrict the full power of NL one also has to augment the SQL language in some way, because SQL has some restrictions and shortcomings, of which the lack of a loop capability for handling ordinals (e.g. the 3rd lowest …, the 5th highest …) is the most serious.In this paper such an augmentation (which we call SQL +) is presented, and it is shown that—observing some simple rules—a NL sentence can be mapped from its internal representation to SQL + in a straightforward and uniform manner. Though these SQL + expressions may not always be optimal from a performance point of view, they have some advantages which we consider to be more useful.