metadata

license: cc-by-4.0
dataset_info:
  features:
    - name: sentence
      dtype: string
    - name: question
      dtype: string
    - name: options
      sequence: string
    - name: answer
      dtype: string
  splits:
    - name: train
      num_bytes: 11002
      num_examples: 80
    - name: test
      num_bytes: 15200
      num_examples: 100
    - name: validation
      num_bytes: 13089
      num_examples: 82
  download_size: 28710
  dataset_size: 39291
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*

The Modified Winograd Schema Challenge (MWSC)

Dataset Description

Homepage: http://decanlp.com
Repository: https://github.com/salesforce/decaNLP
Paper: The Natural Language Decathlon: Multitask Learning as Question Answering
Point of Contact: Bryan McCann, Nitish Shirish Keskar
Size of downloaded dataset files: 19.20 kB
Size of the generated dataset: 39.35 kB
Total amount of disk used: 58.55 kB

Dataset Summary

Examples taken from the Winograd Schema Challenge modified to ensure that answers are a single word from the context. This Modified Winograd Schema Challenge (MWSC) ensures that scores are neither inflated nor deflated by oddities in phrasing.

Dataset Structure

Data Instances

default

Size of downloaded dataset files: 0.02 MB
Size of the generated dataset: 0.04 MB
Total amount of disk used: 0.06 MB

An example looks as follows:

{
    "sentence": "The city councilmen refused the demonstrators a permit because they feared violence.",
    "question": "Who feared violence?",
    "options": [ "councilmen", "demonstrators" ],
    "answer": "councilmen"
}

Data Fields

The data fields are the same among all splits.

default

sentence: a string feature.
question: a string feature.
options: a list of string features.
answer: a string feature.

Data Splits

name	train	validation	test
default	80	82	100

Licensing Information

Our code for running decaNLP has been open sourced under BSD-3-Clause.

We chose to restrict decaNLP to datasets that were free and publicly accessible for research, but you should check their individual terms if you deviate from this use case.

From the Winograd Schema Challenge:

Both versions of the collections are licenced under a Creative Commons Attribution 4.0 International License.

Citation Information

If you use this in your work, please cite:

@inproceedings{10.5555/3031843.3031909,
  author = {Levesque, Hector J. and Davis, Ernest and Morgenstern, Leora},
  title = {The Winograd Schema Challenge},
  year = {2012},
  isbn = {9781577355601},
  publisher = {AAAI Press},
  abstract = {In this paper, we present an alternative to the Turing Test that has some conceptual and practical advantages. A Wino-grad schema is a pair of sentences that differ only in one or two words and that contain a referential ambiguity that is resolved in opposite directions in the two sentences. We have compiled a collection of Winograd schemas, designed so that the correct answer is obvious to the human reader, but cannot easily be found using selectional restrictions or statistical techniques over text corpora. A contestant in the Winograd Schema Challenge is presented with a collection of one sentence from each pair, and required to achieve human-level accuracy in choosing the correct disambiguation.},
  booktitle = {Proceedings of the Thirteenth International Conference on Principles of Knowledge Representation and Reasoning},
  pages = {552–561},
  numpages = {10},
  location = {Rome, Italy},
  series = {KR'12}
}

@article{McCann2018decaNLP,
  title={The Natural Language Decathlon: Multitask Learning as Question Answering},
  author={Bryan McCann and Nitish Shirish Keskar and Caiming Xiong and Richard Socher},
  journal={arXiv preprint arXiv:1806.08730},
  year={2018}
}

Contributions

Thanks to @thomwolf, @lewtun, @ghomasHudson, @lhoestq for adding this dataset.