A comparative analysis of NLP models for Text-to-SQL generation

With the constant increase in the volume and complexity of data that is being generated, the ability to effectively interface with databases has emerged as a crucial leverage. Despite the recent emergence of a variety of query languages, SQL continues to be the most popular tool for accessing and modifying data in extensive database systems. However, SQL's rigid syntax and steep learning curve present a significant barrier to non-expert users. To overcome these difficulties, the high-level query interfaces using NLP techniques to convert natural language queries into SQL have been designed to make database systems more easily accessible. Overall, this survey offers a systematic comparison of the latest development of NLP-based Text-to-SQL models. Various techniques that we consider are Seq2Seq models, Transformer-based models, and Grammar-based models. Furthermore, we examine several evaluation measures like execution accuracy, logical form accuracy, and computational cost to compare the proposed models for the handling of large datasets of Spider and WikiSQL. This survey aims to guide future research by highlighting the strengths and limitations of existing approaches, while also identifying potential directions for improving NLP-driven SQL generation in large-scale database environments.