An Implementation of a Deep Learning NL prototype

This paper describes a Python implementation of a deep learning algorithm to interrogate a financial database using multiple natural languages. The input is a natural language question and the output is an SQL Select statement, which is then executed to get the correct answer. The algorithm splits the question into its individual words and then asks the supervisor to map it to its suitable type and then map its type to its suitable reference. Significant words in each question are mapped by a supervisor to the proper knowledge types, which are Quantifier, Column Name, Column Value or Operator. The Table name will be identified eventually from the Column Names. The system will store this knowledge to understand future questions. Non-significant words will not be mapped to any knowledge type, however this will be considered as a knowledge and will be stored as well. The supervisor has to be knowledgeable in the natural language used and in the database domain.