Natural Questions Simplified for Reading Comprehension

LLukas22's modified version of the Natural Questions dataset for question answering tasks, preprocessed into a SQuAD-like format. Each sample provides a shortened Wikipedia passage containing the answer as context.

Use Cases

• Train a reading comprehension model using the 'context' and 'question' fields in a SQuAD-like format.
• Fine-tune a question-answering system on Wikipedia-derived 'context' passages paired with 'question' and 'answer' fields.
• Benchmark extractive QA models on English text from Wikipedia articles.

Strengths

• Derived from the established Natural Questions (NQ) dataset, a standard for QA research.
• Preprocessed into a SQuAD-like format, a widely adopted structure for QA tasks.

Limitations

• The dataset is a modified version; the exact preprocessing steps and potential information loss from shortening contexts are not detailed.
• Key metadata like row count, column definitions, and license are unknown, limiting reproducibility and use assessment.

Provenance

Source

Modified from the original Natural Questions (NQ) dataset.

Collection Method

Original Wikipedia-derived questions and answers, with contexts shortened to the passage containing the answer.

Time Range

null

Freshness

Last updated on 2023-04 30.

Geography

null