Using domain specific language and sequence to sequence models as a hybrid framework for a natural language interface to a database solution

Abstract

The aim of this project is to provide a new approach to solving the problem of converting natural language into a language capable of querying a database or data repository. This problem has been around for a while, in the 1970's the US Navy developed a solution called LADDER and since then there have been an array of solutions, approaches and tweaks that have kept the research community busy. The introduction of electronic assistants into the smart phone in 2010 has given new impetus to this problem. With the increasingly pervasive nature of data and its ever expanding use to answer questions within business science, medicine extracting data is becoming more important. The idea behind this project is to make data more democratised by allowing access to it without the need for specialist languages. The performance and reliability of converting natural language into structured query language can be problematic in handling nuances that are prevalent in natural language. Relational databases are not designed to understand language nuance. This project introduces the following components as part of a holistic approach to improving the conversion of a natural language statement into a language capable of querying a data repository.

● The idea proposed in this project combines the use of sequence to sequence models in conjunction with the natural language part of speech technologies and domain specific languages to convert natural language queries into SQL. The approach being proposed by this chapter is to use natural language processing to perform an initial shallow pass of the incoming query and then use Google's Tensor Flow to refine the query with the use of a sequence to sequence model.

● This thesis is also proposing to use a Domain Specific Language (DSL) as part of the conversion process. The use of the DSL has the potential to allow the natural language query to be translated into more than just an SQL statement, but any query language such as NoSQL or XQuery.