A Survey on Text-to-SQL Parsing: From Rule-Based Foundations to Large Language Models

Transforming natural language into SQL queries represents a significant challenge in both natural language processing (NLP) and database research fields. The main target of this transformation is to allow users to interact with databases using their natural language. This extends the scope of data analysis to include persons who do not have expertise in computer science or data engineering. In recent years, the implementation of deep learning models has yielded significant results. Following this, the introduction of large language models (LLMs) such as GPT-4 and BERT has lifted the standard for this task, promoting them to new heights of performance and capacity. Our survey provides a detailed review of the evolutionary phases of Text-to-SQL parsing. Initially, we investigate the primary approaches in this field that use rule-based models. Following this, we present the several deep learning models proposed for this task. After that, we introduce several datasets commonly used in training and evaluating the proposed models. In the last section, we examine the role of large language models (LLMs) in revolutionizing Text-to-SQL tasks.