NQ (Natural Questions) is a large-scale dataset of real user questions issued to Google search, and answers found from Wikipedia by annotators. NQ is designed for the training and evaluation of automatic question answering systems that can handle open-domain questions that require reasoning and commonsense knowledge

NQ is a benchmark for testing the natural language understanding and question answering ability of large language models (LLMs), such as BERT, GPT-3, and RoBERTa. It can be used to evaluate how well these models can understand natural language and answer questions that are naturally posed by humans in a search engine context. NQ is also intended to encourage the development of more robust and generalizable models that can handle various domains and tasks.

Here are some examples of questions from the NQ dataset, taken from different domains¹:

• Who is the author of the book the little prince?
  • Antoine de Saint-Exupéry
• What is the most common use of a firing pin?
  • to strike the primer of a cartridge and ignite the propellant
• Where is the notre dame cathedral located?
  • Paris, France
• How many people are in the international space station?
  • six
• What is the name of the virus that causes chickenpox?
  • varicella zoster virus

• The original paper by Kwiatkowski et al. that introduces the dataset and the methodology used to create it.
• The GitHub repository by Google Research that contains the dataset, the code, and the models for NQ.
• The Google AI website that provides the dataset download link and the leaderboard for the top-performing models.
• The Hugging Face website that provides the dataset download link and the documentation for loading and using the dataset.
• The Papers With Code website that lists the papers and code related to the NQ dataset and task.