Datasets:

Modalities:
Text
Formats:
parquet
Languages:
Swedish
Size:
< 1K
Libraries:
Datasets
pandas
piqa-sv / README.md
jekunz's picture
Update README.md
e8e2fb0 verified
metadata
dataset_info:
  features:
    - name: prompt
      dtype: string
    - name: correct
      dtype: string
    - name: wrong
      dtype: string
  splits:
    - name: train
      num_bytes: 35150
      num_examples: 100
  download_size: 26343
  dataset_size: 35150
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
language:
  - sv

Draft of Swedish PIQA contribution to the MRL workshop.

The annotator of this dataset drew inspiration from a diverse set of web sources that cover everyday physical activities. These included sports-related Wikipedia pages and categories, as well as various Swedish sites such as villalivet.se (focused on house and garden life), naturvårdsverket.se (the Swedish Environmental Protection Agency), and krisinformation.se (a government portal for crisis information). From these sources, we derived questions spanning a broad range of topics: sports and outdoor activities, household and gardening tasks, cooking and baking, physical activities associated with traditional festivities, and traffic-related scenarios. The primary annotator wrote questions and correct and wrong answers, while a second annotator verified them. The dataset contains 100 samples, each consisting of a question ('prompt'), a correct answer ('correct') and a wrong answer ('wrong').