README.md

metadata

language:
  - en
license: apache-2.0

Acknowledgements

The PUZZLEQA is scraped from NPR Sunday Puzzle Official Website and NPR Puzzle Synopsis, made by a group of fans by running a mailing list that distributed questions and answers for each week’s puzzle. The authors of the dataset cleaned the data and made some multiple choice based on the question and answers.

Creation

The Multiple Choice Dataset is generated from PUZZLEQA dataset using the following algorithm.

1. Read the fr_big_exp.tsv.tsv file
2. Group rule-question-answer triples in a given Sunday together (so the rules of each question will be the same)
3. For each question, randomly select three other answers from answers on the same Sunday. Shuffle 3 selected answers with the correct answer for the given question to obtain 4 choices for this question. \
4. identify the correct answer for the given question as the "gold" answer.

Recent.tsv is the dataset based on the NPR PUZZLE in 2023.

Citation

@inproceedings{zhao2023solving, title={Solving and Generating NPR Sunday Puzzles with Large Language Models}, author={Jingmiao Zhao and Carolyn Jane Anderson}, year={2023}, eprint={2306.12255}, archivePrefix={arXiv}, primaryClass={cs.CL} }