README.md

metadata

dataset_info:
  features:
    - name: scenario
      dtype: string
    - name: correct_proverb
      dtype: string
    - name: incorrect_proverb
      dtype: string
  splits:
    - name: train
      num_bytes: 7299
      num_examples: 32
    - name: test
      num_bytes: 23286
      num_examples: 98
  download_size: 27103
  dataset_size: 30585
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

Dutch proverbs

This a collection of 130 brief scenario's in Dutch, combined with a proverb (spreekwoord) that matches the scenario (correct spreekwoord) and a proverb that does not match the scenario (incorrect spreekwoord).

Creation

The dataset was created and reviewed manually by a team of NFI employees (Netherlands Forensic Institute).

The creation process was as follows:

1. A proverb was selected
2. A scenario was created to match the proverb
3. An alternative, incorrect proverbs was selected

The alternative proverbs were not selected randomly. They were crafted to share similar words or a similar theme with the scenario, yet are still clearly the incorrect proverb for the situation.

Use

These proverbs can be used to test the ability of Large Language Models to match the right proverb to a scenario.