README.md

metadata

dataset_info:
  - config_name: default
    features:
      - name: question
        dtype: string
      - name: index
        dtype: int64
      - name: option-A
        dtype: string
      - name: option-B
        dtype: string
      - name: option-C
        dtype: string
      - name: correct
        dtype: string
      - name: origin
        dtype: string
    splits:
      - name: train
        num_bytes: 103251
        num_examples: 605
    download_size: 43667
    dataset_size: 103251
  - config_name: raw
    features:
      - name: question
        dtype: string
      - name: index
        dtype: int64
      - name: option-A
        dtype: string
      - name: option-B
        dtype: string
      - name: option-C
        dtype: string
      - name: correct
        dtype: string
      - name: origin
        dtype: string
    splits:
      - name: train
        num_bytes: 103906
        num_examples: 605
    download_size: 45297
    dataset_size: 103906
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
  - config_name: raw
    data_files:
      - split: train
        path: raw/train-*
task_categories:
  - question-answering
language:
  - da
pretty_name: Citizenship Tests Da.
size_categories:
  - n<1K

Dataset Card for "citizenship-test-da"

This dataset contains scraped questions and answers from Danish citizen tests (Danish: indfødsretsprøver og medborgerskabsprøver) from Juni 2019 to May 2023 from PDF's produced by ''Styrelsen for International Rekruttering og Integration'' (SIRI).

The dataset is released as an appendix to the thesis ''Are GLLMs Danoliterate? Benchmarking Generative NLP in Danish'' and permission by SIRI for this specific purpose.

The PDF's are available on SIRI's website. The default configuration has been semi-automatically cleaned to remove PDF artifacts using the Alvenir 3gram DSL language model. The examples were not deduplicated.

Further details can be found in Section 4.2.1 in the thesis.

• Produced by: Søren Vejlgaard Holm under supervision of Lars Kai Hansen and Martin Carsten Nielsen.
• Usable for: Question Answering Evaluation.
• Contact: Søren Vejlgaard Holm at swiho@dtu.dk or swh@alvenir.ai.