Datasets:
GEM
/

Tasks:
Other
Modalities:
Text
Languages:
English
ArXiv:
Libraries:
Datasets
License:
squad_v2 / README.md
Abinaya Mahendiran
Updated README
18eda8f
|
raw
history blame
8.59 kB
metadata
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 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: a string feature.
  • gem_id: a string feature.
  • title: a string feature.
  • context: a string feature.
  • question: a string feature.
  • answers: a dictionary feature containing:
    • text: a string feature.
    • answer_start: a int32 feature.

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

More Information Needed

Who are the source language producers?

More Information Needed

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

More Information Needed

Considerations for Using the Data

Social Impact of Dataset

More Information Needed

Discussion of Biases

More Information Needed

Other Known Limitations

More Information Needed

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.