Mapping natural language into SQL in a NLIDB

Abstract

Since the 1970's, there has been growing interest in understanding and answering human language questions. Despite this, little progress has been made in developing an interface that any untrained user can use to query very large databases using natural language. In this research, the design of a novel system is discussed. Tree-like structures are built for every question and each query, and a tree kernel, which represents trees in terms of their substructures, is used to define feature spaces. A machine learning algorithm is proposed that takes pairs of trees as training input and derives the unknown final SQL query by matching propositional and relational substructures. © 2008 Springer-Verlag Berlin Heidelberg.