# RESDSQL: Decoupling Schema Linking and Skeleton Parsing for Text-to-SQL

**Authors:** Haoyang Li, Jing Zhang, Cuiping Li, Hong Chen  
**arXiv:** [2302.05965](https://arxiv.org/abs/2302.05965)  
**Date:** 2023-02-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2302.05965v3/

## Key points

- The paper presents a novel approach called RESDSQL to improve the accuracy of Text-to-SQL systems.
- Existing approaches face challenges in parsing complex SQL queries involving multiple schema items and logic operators.
- RESDSQL proposes a ranking-enhanced encoding and skeleton-aware decoding framework.
- The encoder-decoder model is modified to decouple the schema linking and skeleton parsing processes.
- Only the most relevant schema items are used, reducing effort for schema linking during SQL parsing.
- The decoder first generates the skeleton (SQL keywords) and then constructs the actual SQL query.
- This sequential generation process improves the accuracy of SQL parsing.
- Experimental results on Spider and its variants show promising performance and robustness for RESDSQL.
- RESDSQL offers a new perspective on tackling challenges in Text-to-SQL systems by decoupling schema linking and skeleton parsing processes.
- The availability of their code on GitHub allows for further exploration and implementation.

## Summary

The paper titled "RESDSQL: Decoupling Schema Linking and Skeleton Parsing for Text-to-SQL" presents a novel approach to improve the accuracy of Text-to-SQL systems. These systems aim to convert natural language queries into SQL queries, enabling users to interact with databases more intuitively. The authors highlight that existing approaches such as pre-trained language models face challenges in parsing complex SQL queries that involve multiple schema items (tables and columns) and logic operators. To address this issue, the authors propose a ranking-enhanced encoding and skeleton-aware decoding framework called RESDSQL. In this framework, the encoder-decoder model is modified to decouple the schema linking and skeleton parsing processes. Instead of injecting the entire set of unordered schema items into the encoder, only the most relevant ones are used which reduces the effort required for schema linking during SQL parsing. Furthermore, the decoder in RESDSQL first generates the skeleton (SQL keywords) and then constructs the actual SQL query. This sequential generation process implicitly constrains the SQL parsing which improves its accuracy. To evaluate their proposed framework, the authors conducted experiments on Spider and its three robustness variants: Spider-DK, Spider-Syn, and Spider-Realistic. The experimental results demonstrate that RESDSQL achieves promising performance and robustness in comparison to existing approaches. In conclusion, RESDSQL offers a new perspective on tackling challenges in Text-to-SQL systems by decoupling schema linking and skeleton parsing processes. Its ranking-enhanced encoding and skeleton aware decoding framework shows improved accuracy in parsing complex SQL queries involving multiple schema items and logic operators. The availability of their code on GitHub allows for further exploration and implementation of their approach.