Datasetpublic

liu-nlp/swedish-facts-v1

This is a benchmark for Sweden-related factual knowledge. Its questions are inspired by the hosts of the Swedish radio program Sommar I P1 as well as sports-related events in Sweden (e.g., events that are part of En Svensk Klassiker. Answers are designed to be minimal to enable simple string-based answer recall to approximate model performance as well as possible. Note that some samples in the dataset have multiple correct answers; if so, they are separated by commas. See the preprint for… See the full description on the dataset page: https://huggingface.co/datasets/liu-nlp/swedish-facts-v1.

sourceHugging Faceapache-2.0updated 10mo agoView on Hugging Face

4likes50downloads

Dataset Card

Answers are designed to be minimal to enable simple string-based answer recall to approximate model performance as well as possible. Note that some samples in the dataset have multiple correct answers; if so, they are separated by commas.

See the preprint for more information.

Citation

bibtex

@misc{kunz2025diagnosticbenchmarkswedenrelatedfactual,
      title={A Diagnostic Benchmark for Sweden-Related Factual Knowledge}, 
      author={Jenny Kunz},
      year={2025},
      eprint={2510.21360},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.21360}, 
}