pretty_name: SQuAD2.0
annotations_creators:
- crowdsourced
language_creators:
- crowdsourced
languages:
- en
licenses:
- cc-by-sa-4.0
multilinguality:
- monolingual
size_categories:
- 100K<n<1M
source_datasets:
- original
task_categories:
- question-answering
- question-generation
task_ids:
- open-domain-qa
- extractive-qa
paperswithcode_id: squad
Dataset Card for "squad_v2"
Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- Homepage: https://rajpurkar.github.io/SQuAD-explorer/
- Repository: https://rajpurkar.github.io/SQuAD-explorer/
- Paper: SQuAD: 100,000+ Questions for Machine Comprehension of Text
- Point of Contact: Google Group or robinjia
- Size of downloaded dataset files: 44.34 MB
- Size of the generated dataset: 122.57 MB
- Total amount of disk used: 166.91 MB
Dataset Summary
Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable.
SQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones. To do well on SQuAD2.0, systems must not only answer questions when possible, but also determine when no answer is supported by the paragraph and abstain from answering.
Supported Tasks and Leaderboards
SQuAD2.0 tests the ability of a system to not only answer reading comprehension questions, but also abstain when presented with a question that cannot be answered based on the provided paragraph. Leaderboard is present on the Homepage.
Languages
English (en)
Dataset Structure
We show detailed information for up to 5 configurations of the dataset.
Data Instances
squad_v2
- Size of downloaded dataset files: 146.21 MB
- Size of the generated dataset: 122.70 MB
- Total amount of disk used: 268.90 MB
An example of 'validation' looks as follows.
This example was too long and was cropped:
{
"gem_id": "gem-squad_v2-validation-1",
"id": "56ddde6b9a695914005b9629",
"answers": {
"answer_start": [94, 87, 94, 94],
"text": ["10th and 11th centuries", "in the 10th and 11th centuries", "10th and 11th centuries", "10th and 11th centuries"]
},
"context": "\"The Normans (Norman: Nourmands; French: Normands; Latin: Normanni) were the people who in the 10th and 11th centuries gave thei...",
"question": "When were the Normans in Normandy?",
"title": "Normans"
}
Data Fields
The data fields are the same among all splits.
squad_v2
id: astringfeature.gem_id: astringfeature.title: astringfeature.context: astringfeature.question: astringfeature.answers: a dictionary feature containing:text: astringfeature.answer_start: aint32feature.
Data Splits
The original SQuAD2.0 dataset has only training and dev (validation) splits. The train split is further divided into test split and added as part of the GEM datasets.
| name | train | validation | test |
|---|---|---|---|
| squad_v2 | 90403 | 11873 | 39916 |
Dataset Creation
Curation Rationale
The dataset is curated in three stages:
- Curating passages,
- Crowdsourcing question-answers on those passages,
- Obtaining additional answers
As part of SQuAD1.1, 10000 high-quality articles from English Wikipedia is extracted using Project Nayuki’s Wikipedia’s internal PageRanks, from which 536 articles are sampled uniformly at random. From each of these articles, individual paragraphs are extracted, stripping away images, figures, tables, and discarding paragraphs shorter than 500 characters.
SQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones.
Source Data
Initial Data Collection and Normalization
Who are the source language producers?
Annotations
Annotation process
The Daemo platform (Gaikwad et al., 2015), with Amazon Mechanical Turk as its backend is used for annotation.
- On each paragraph, crowdworkers were tasked with asking and answering up to 5 questions on the content of that paragraph and were asked spend 4 minutes on every paragraph. Questions need to be entered in a text box and answers need to be highlighted in the paragraph.
- To get an indication of human performance on SQuAD and to make the evaluation more robust, at least 2 additional answers for each question is obtained in the development and test sets.
- In the secondary answer generation task, each crowdworker was shown only the questions along with the paragraphs of an article, and asked to select the shortest span in the paragraph that answered the question. If a question was not answerable by a span in the paragraph, workers were asked to submit the question without marking an answer
Who are the annotators?
Crowdworkers from the United States or Canada with a 97% HIT acceptance rate, a minimum of 1000 HITs, were employed to create questions.
Personal and Sensitive Information
Considerations for Using the Data
Social Impact of Dataset
Discussion of Biases
Other Known Limitations
Additional Information
Dataset Curators
The authors of SQuAD dataset would like to thank Durim Morina and Professor Michael Bernstein for their help in crowdsourcing the collection of the dataset, both in terms of funding and technical support of the Daemo platform.
Licensing Information
The dataset is distributed under the CC BY-SA 4.0 license.
Citation Information
@article{2016arXiv160605250R,
author = {{Rajpurkar}, Pranav and {Zhang}, Jian and {Lopyrev},
Konstantin and {Liang}, Percy},
title = "{SQuAD: 100,000+ Questions for Machine Comprehension of Text}",
journal = {arXiv e-prints},
year = 2016,
eid = {arXiv:1606.05250},
pages = {arXiv:1606.05250},
archivePrefix = {arXiv},
eprint = {1606.05250},
}
Contributions
Thanks to @AbinayaM02 for adding this dataset to GEM. All the details are obtained from the cited paper.