Automatic Correction of Syntax Errors in SuperSQL Queries

SuperSQL is an extended language of SQL. By structuring the output of relational databases, SuperSQL enables the user to generate various types of structured documents with various layouts which are not represented in SQL. There is a problem that the larger and more complicated the SuperSQL query is, the more difficult it is to detect errors and the more time is spent on debugging. In this study, we propose a system that automatically detects and corrects syntax errors in user queries. When a query parsing fails, the system reanalyzes the query and predicts a correction by using deep learning. To modify the query, we use recurrent neural network and attention mechanism. By presenting the predicted modifications to users, the burden of debugging can be reduced and the efficiency of user's work can be improved.