README.md

metadata

license: cc-by-sa-3.0
language:
  - de
  - en
tags:
  - sentence-transformers
pretty_name: Natural Questions (German)
task_categories:
  - feature-extraction
  - sentence-similarity
size_categories:
  - 100K<n<1M
dataset_info:
  features:
    - name: query_de
      dtype: string
    - name: answer_de
      dtype: string
    - name: query
      dtype: string
    - name: answer
      dtype: string
  splits:
    - name: train
      num_bytes: 140561522
      num_examples: 100231
  download_size: 91774972
  dataset_size: 140561522
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Natural Questions (German)

This dataset is a German-translated version of the original Natural Questions question-answer dataset by Google. It has been translated from English into German and extended with additional columns to include both the original and the translated content.

This dataset can be used directly with Sentence Transformers to train multilingual embedding models or to fine-tune models on German QA tasks.

Dataset Structure

• Columns:

  • "query": Original English question
  • "answer": Original English answer passage
  • "question_de": Translated German question
  • "answer_de": Translated German answer passage

• Column types: all str

• Example:

  {
    'query': 'the si unit of the electric field is',
    'answer': 'Electric field An electric field is a field that surrounds electric charges...',
    'question_de': 'Die SI-Einheit des elektrischen Feldes ist',
    'answer_de': 'Ein elektrisches Feld ist ein Feld, das elektrische Ladungen umgibt...'
  }
  

Collection Strategy

This dataset is based on the pair subset from the embedding-training-data collection. The German translations were generated automatically and have not been manually verified. Use with caution in high-stakes applications.

Use Cases

The dataset is ideal for:

• Training or evaluating multilingual embedding models
• Cross-lingual retrieval and similarity tasks
• Question answering and sentence similarity in German

Licensing and Attribution

The dataset is based on content from the Natural Questions dataset by Google, translated under fair use. Please cite the original Natural Questions dataset in any work using this resource.