# QURG: Question Rewriting Guided Context-Dependent Text-to-SQL Semantic Parsing

**Authors:** Linzheng Chai, Dongling Xiao, Jian Yang, Liqun Yang, Qian-Wen Zhang, Yunbo Cao, Zhoujun Li, Zhao Yan  
**arXiv:** [2305.06655](https://arxiv.org/abs/2305.06655)  
**Date:** 2023-05-11  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.06655v1/

## Key points

- The paper addresses the challenge of translating multi-turn natural language questions into SQL queries in a context-dependent manner.
- QURG is a novel approach that leverages question rewriting to enhance models' contextual understanding.
- The first step involves training a question rewriting model to complete the current question based on its context.
- This completion process generates a rewriting edit matrix, which captures the changes made to the original question.
- A two-stream matrix encoder is designed to jointly model the rewriting relations between the question and its context, as well as the schema linking relations between natural language and structured schema.
- Experimental results demonstrate that QURG significantly improves performance on two large-scale context-dependent datasets, SParC and CoSQL.
- QURG excels in handling challenging and long-turn questions.
- The proposed method outperforms existing techniques on various evaluation metrics and datasets, particularly showing its effectiveness in dealing with complex and lengthy questions.

## Summary

The paper titled "QURG: Question Rewriting Guided Context-Dependent Text-to-SQL Semantic Parsing" addresses the challenge of translating multi-turn natural language questions into SQL queries in a context-dependent manner. To tackle this problem, the authors propose QURG, a novel approach that leverages question rewriting to enhance models' contextual understanding. The first step involves training a question rewriting model to complete the current question based on its context. This completion process generates a rewriting edit matrix, which captures the changes made to the original question. To jointly model the rewriting relations between the question and its context, as well as the schema linking relations between natural language and structured schema, a two-stream matrix encoder is designed. Experimental results demonstrate that QURG significantly improves performance on two large-scale context-dependent datasets, namely SParC and CoSQL. Notably, it excels in handling challenging and long-turn questions. In summary, QURG presents an innovative approach for addressing context-dependency in Text-to-SQL semantic parsing by leveraging question rewriting guided by question context. The proposed method outperforms existing techniques on various evaluation metrics and datasets, particularly showing its effectiveness in dealing with complex and lengthy questions.