March 11, 2022
While SQL is easy and widely used, it’s still inaccessible to laypeople. Since many users without coding backgrounds also need to perform data processing or retrieval, translating natural language to SQL (NL2SQL) has received extensive attention, especially with the recent success of deep learning technologies. Due to the complexity of this task, users may want to give feedback when the model makes mistakes. Interactive models help improve the model’s overall accuracy and usability. Nonetheless, current interaction methods are limited. In particular, users often lack the flexibility to proactively give feedback. In this talk, I will review related work of interactive SQL synthesis and then introduce our approach. Our tool generates structured explanations in natural language, serving as an interface. While the explanations make users understand how the system works, users can give feedback to the system through modifying them.