shuaichenchang / odis-text-to-sql Goto Github PK

Dear Authors,

Thank you very much for your work. We have noticed in the 'Synthetic data generation' step of 'In-domain Synthetic Demonstration Creation,' you utilized the SHiP method from the paper

Importance of synthesizing high-quality data for text-to-sql parsing(https://aclanthology.org/2023.findings-acl.86.pdf)

but we have been unable to find the open-source code for this paper. Could you possibly provide a link to any accessible code repositories you are aware of, or share your own implementation of this method? Thank you very much!
Best regards,

Hello. I see here that you mentioned "In the demonstration examples, we replace the annotated NLQs with these synthetic NLQs while keeping the SQL queries unchanged. This setup allows the LLMs to be exposed to the in-domain SQL distribution while remaining unaware of the annotated in-domain NLQs or other inputs beyond its own generated NLQs".

I don't quite understand what "synthetic NLQs" and "annotated NLQs" are, can you explain them to me?

It is better to explain with some examples, so that I can understand all your research more easily. Thank you