---
dataset_info:
  features:
  - name: question
    dtype: string
  - name: answer
    dtype: string
  - name: wrong_answers
    dtype: string
  - name: source_file
    dtype: string
  - name: question_number
    dtype: string
  splits:
  - name: train
    num_bytes: 915108
    num_examples: 1401
  download_size: 497465
  dataset_size: 915108
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
task_categories:
- question-answering
language:
- 'no'
tags:
- medical
---

# Norwegian Medical Question Answering Dataset (NorMedQA)

**Dataset Card for NorMedQA**

## Dataset Summary

NorMedQA is a benchmark dataset comprising 1,401 medical question-and-answer pairs in Norwegian (Bokmål and Nynorsk). Designed to evaluate large language models (LLMs) on medical knowledge retrieval and reasoning within the Norwegian context, the dataset is structured in JSON format. Each entry includes the source document name, question number (where available), question text, reference answer text, and additional distractor answers. The content originates from publicly available sources containing medical exam questions and has undergone cleaning and preprocessing.

## Supported Tasks and Leaderboards

- **Medical Question Answering**: Assessing LLMs' ability to answer medical questions accurately.
- **Language Understanding in Norwegian**: Evaluating comprehension and reasoning in Norwegian medical contexts.

## Languages

- Norwegian Bokmål
- Norwegian Nynorsk

## Dataset Structure

### Data Instances

Each data instance is a JSON object with the following fields:

- `question`: The medical question in Norwegian.
- `answer`: The correct answer text in Norwegian.
- `wrong_answers`: Distractor (incorrect) answers in Norwegian. Can be multiple, split by ;
- `source_file`: Name of the source document.
- `question_number`: Identifier for the question (if available).

### Data Splits

The dataset is provided as a single split:

- `train`: 1,401 examples

## Dataset Creation

### Curation Rationale

The dataset was curated to provide a resource for evaluating LLMs on medical question answering tasks in Norwegian, addressing the need for domain-specific evaluation benchmarks in less-resourced languages.

### Source Data

Questions and answers were sourced from publicly available Norwegian medical exam materials. The data underwent cleaning and preprocessing to ensure quality and consistency.

### Annotation Process

The dataset entries were derived from existing exam materials, with cleaning and formatting applied to structure the data appropriately. Additional distractor answers were generated or identified to support multiple-choice evaluation.

## Usage

NorMedQA can be used to evaluate the performance of LLMs on medical question answering tasks in Norwegian. It is compatible with evaluation frameworks like `lm-evaluation-harness`.

## Citation

If you use NorMedQA in your research, please cite the following:

Riegler, M. A. (2025). Norwegian Medical Question Answering Dataset - NorMedQA [Data set].

## License

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

## Additional Information

- **GitHub Repository**: [https://github.com/kelkalot/normedqa](https://github.com/kelkalot/normedqa)
- **Hugging Face Dataset Page**: [https://huggingface.co/datasets/SimulaMet/NorMedQA](https://huggingface.co/datasets/SimulaMet/NorMedQA)