/EMNLP23-APEL

Some code and data for the paper: "Labeling Programs with Non-Programmers Indirectly via Active Examples: A Case Study with Text-to-SQL"

Primary LanguagePython

EMNLP23-APEL

Code and data for the paper: Non-Programmers Can Label Programs Indirectly via Active Examples: A Case Study with Text-to-SQL, to appear in EMNLP 2023. Joint work with Charlie Snell, Dan Klein, and Jason Eisner.

To install: run pip3 install -r requirements.txt

example_minimal_witness.py contains a simple example for runing the algorithm in Section 3 which maximize the information gain subject to a size constraint. In this algorithm, we first randomly generate large random databases, choose the most informative one, then reduce its size.

spiderdevfixes.csv contains all the original SPIDER annotations we corrected, along with the reason. The corresponding author of the SPIDER dataset endorses our correction.

Citation:

@InProceedings{Zhong-Snell-Klein-Eisner:2023:APEL,
  title     = {Non-Programmers Can Label Programs Indirectly via Active Examples: A Case Study with Text-to-SQL},
  author    = {Ruiqi Zhong and Charlie Snell and Dan Klein and Jason Eisner},
  booktitle = {Proceedings of EMNLP},
  address   = {},
  pages     = {},
  month     = {December},
  year      = {2023},
}