ACL-SQL: Generating SQL Queries from Natural Language

Relational databases are prevalent in handling large and complex databases; however, writing the SQL queries can be a tedious task for databases. Recent studies for automating the task of translating natural language text to SQL query are not robust to queries having multi-table dependencies. Also, considering the lack of such research in scientific domain, we introduce ACL-SQL, a dataset with complex queries depending on up to five tables and benchmark the dataset on a simple approach. Evaluation shows that our approach is reasonably precise and can be adopted for practical applications. The dataset and codes are available at https://github.com/rohitshantarampatil/sql-nlp.