SQL translator using artificial neural networks

The paper presents a novel approach to convert natural language database requests into SQL. The SQL translator acts as a frontend to any information system and provides a natural language interface (NLI) to the end user. The translator is designed to understand everyday English requests and invoke appropriate database reporting tool for a valid query. Each valid query is mapped onto an appropriate SQL command through a neural net based converter/translator. A special class of recurrent neural network paradigm suggested by M.I. Jordan (1988) has been successfully used for this purpose. The query is read one word at a time and clamped at the input of the recurrent network. After recognising the valid sequences of words the output of the network stabilises and indicates the category of the SQL command. The biggest advantage of using a neural net approach is the trainable and adaptive behaviour of the translator. The translator can be trained to recognise queries in other languages and domain by designing a domain dependent training set and incorporating a dictionary of related words. The dictionary of commonly used words and synonyms required for this purpose is an ordinary DOS file that can be shared by multiple users in a network