Dataset Card for bAbi QA

Dataset Summary

The (20) QA bAbI tasks are a set of proxy tasks that evaluate reading comprehension via question answering. Our tasks measure understanding in several ways: whether a system is able to answer questions via chaining facts, simple induction, deduction and many more. The tasks are designed to be prerequisites for any system that aims to be capable of conversing with a human. The aim is to classify these tasks into skill sets,so that researchers can identify (and then rectify) the failings of their systems.

Supported Tasks and Leaderboards

The dataset supports a set of 20 proxy story-based question answering tasks for various "types" in English and Hindi. The tasks are:

task_no	task_name
qa1	single-supporting-fact
qa2	two-supporting-facts
qa3	three-supporting-facts
qa4	two-arg-relations
qa5	three-arg-relations
qa6	yes-no-questions
qa7	counting
qa8	lists-sets
qa9	simple-negation
qa10	indefinite-knowledge
qa11	basic-coreference
qa12	conjunction
qa13	compound-coreference
qa14	time-reasoning
qa15	basic-deduction
qa16	basic-induction
qa17	positional-reasoning
qa18	size-reasoning
qa19	path-finding
qa20	agents-motivations

The "types" are are:

• en

  • the tasks in English, readable by humans.

• hn

  • the tasks in Hindi, readable by humans.

• shuffled

  • the same tasks with shuffled letters so they are not readable by humans, and for existing parsers and taggers cannot be used in a straight-forward fashion to leverage extra resources-- in this case the learner is more forced to rely on the given training data. This mimics a learner being first presented a language and having to learn from scratch.

• en-10k , shuffled-10k and hn-10k

  • the same tasks in the three formats, but with 10,000 training examples, rather than 1000 training examples.

• en-valid and en-valid-10k

  • are the same as en and en10k except the train sets have been conveniently split into train and valid portions (90% and 10% split).

To get a particular dataset, use load_dataset('babi_qa',type=f'{type}',task_no=f'{task_no}') where type is one of the types, and task_no is one of the task numbers. For example, load_dataset('babi_qa', type='en', task_no='qa1') .

Languages

Dataset Structure

Data Instances

An instance from the en-qa1 config's train split:

{'story': {'answer': ['', '', 'bathroom', '', '', 'hallway', '', '', 'hallway', '', '', 'office', '', '', 'bathroom'], 'id': ['1', '2', '3', '4', '5', '6', '7', '8', '9', '10', '11', '12', '13', '14', '15'], 'supporting_ids': [[], [], ['1'], [], [], ['4'], [], [], ['4'], [], [], ['11'], [], [], ['8']], 'text': ['Mary moved to the bathroom.', 'John went to the hallway.', 'Where is Mary?', 'Daniel went back to the hallway.', 'Sandra moved to the garden.', 'Where is Daniel?', 'John moved to the office.', 'Sandra journeyed to the bathroom.', 'Where is Daniel?', 'Mary moved to the hallway.', 'Daniel travelled to the office.', 'Where is Daniel?', 'John went back to the garden.', 'John moved to the bedroom.', 'Where is Sandra?'], 'type': [0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1]}}

Data Fields

• story : a dictionary feature containing:
  • id : a string feature, which denotes the line number in the example.
  • type : a classification label, with possible values including context , question , denoting whether the text is context or a question.
  • text : a string feature the text present, whether it is a question or context.
  • supporting_ids : a list of string features containing the line numbers of the lines in the example which support the answer.
  • answer : a string feature containing the answer to the question, or an empty string if the type s is not question .

Data Splits

The splits and corresponding sizes are:

train	test	validation
en-qa1	200	200	-
en-qa2	200	200	-
en-qa3	200	200	-
en-qa4	1000	1000	-
en-qa5	200	200	-
en-qa6	200	200	-
en-qa7	200	200	-
en-qa8	200	200	-
en-qa9	200	200	-
en-qa10	200	200	-
en-qa11	200	200	-
en-qa12	200	200	-
en-qa13	200	200	-
en-qa14	200	200	-
en-qa15	250	250	-
en-qa16	1000	1000	-
en-qa17	125	125	-
en-qa18	198	199	-
en-qa19	1000	1000	-
en-qa20	94	93	-
en-10k-qa1	2000	200	-
en-10k-qa2	2000	200	-
en-10k-qa3	2000	200	-
en-10k-qa4	10000	1000	-
en-10k-qa5	2000	200	-
en-10k-qa6	2000	200	-
en-10k-qa7	2000	200	-
en-10k-qa8	2000	200	-
en-10k-qa9	2000	200	-
en-10k-qa10	2000	200	-
en-10k-qa11	2000	200	-
en-10k-qa12	2000	200	-
en-10k-qa13	2000	200	-
en-10k-qa14	2000	200	-
en-10k-qa15	2500	250	-
en-10k-qa16	10000	1000	-
en-10k-qa17	1250	125	-
en-10k-qa18	1978	199	-
en-10k-qa19	10000	1000	-
en-10k-qa20	933	93	-
en-valid-qa1	180	200	20
en-valid-qa2	180	200	20
en-valid-qa3	180	200	20
en-valid-qa4	900	1000	100
en-valid-qa5	180	200	20
en-valid-qa6	180	200	20
en-valid-qa7	180	200	20
en-valid-qa8	180	200	20
en-valid-qa9	180	200	20
en-valid-qa10	180	200	20
en-valid-qa11	180	200	20
en-valid-qa12	180	200	20
en-valid-qa13	180	200	20
en-valid-qa14	180	200	20
en-valid-qa15	225	250	25
en-valid-qa16	900	1000	100
en-valid-qa17	113	125	12
en-valid-qa18	179	199	19
en-valid-qa19	900	1000	100
en-valid-qa20	85	93	9
en-valid-10k-qa1	1800	200	200
en-valid-10k-qa2	1800	200	200
en-valid-10k-qa3	1800	200	200
en-valid-10k-qa4	9000	1000	1000
en-valid-10k-qa5	1800	200	200
en-valid-10k-qa6	1800	200	200
en-valid-10k-qa7	1800	200	200
en-valid-10k-qa8	1800	200	200
en-valid-10k-qa9	1800	200	200
en-valid-10k-qa10	1800	200	200
en-valid-10k-qa11	1800	200	200
en-valid-10k-qa12	1800	200	200
en-valid-10k-qa13	1800	200	200
en-valid-10k-qa14	1800	200	200
en-valid-10k-qa15	2250	250	250
en-valid-10k-qa16	9000	1000	1000
en-valid-10k-qa17	1125	125	125
en-valid-10k-qa18	1781	199	197
en-valid-10k-qa19	9000	1000	1000
en-valid-10k-qa20	840	93	93
hn-qa1	200	200	-
hn-qa2	200	200	-
hn-qa3	167	167	-
hn-qa4	1000	1000	-
hn-qa5	200	200	-
hn-qa6	200	200	-
hn-qa7	200	200	-
hn-qa8	200	200	-
hn-qa9	200	200	-
hn-qa10	200	200	-
hn-qa11	200	200	-
hn-qa12	200	200	-
hn-qa13	125	125	-
hn-qa14	200	200	-
hn-qa15	250	250	-
hn-qa16	1000	1000	-
hn-qa17	125	125	-
hn-qa18	198	198	-
hn-qa19	1000	1000	-
hn-qa20	93	94	-
hn-10k-qa1	2000	200	-
hn-10k-qa2	2000	200	-
hn-10k-qa3	1667	167	-
hn-10k-qa4	10000	1000	-
hn-10k-qa5	2000	200	-
hn-10k-qa6	2000	200	-
hn-10k-qa7	2000	200	-
hn-10k-qa8	2000	200	-
hn-10k-qa9	2000	200	-
hn-10k-qa10	2000	200	-
hn-10k-qa11	2000	200	-
hn-10k-qa12	2000	200	-
hn-10k-qa13	1250	125	-
hn-10k-qa14	2000	200	-
hn-10k-qa15	2500	250	-
hn-10k-qa16	10000	1000	-
hn-10k-qa17	1250	125	-
hn-10k-qa18	1977	198	-
hn-10k-qa19	10000	1000	-
hn-10k-qa20	934	94	-
shuffled-qa1	200	200	-
shuffled-qa2	200	200	-
shuffled-qa3	200	200	-
shuffled-qa4	1000	1000	-
shuffled-qa5	200	200	-
shuffled-qa6	200	200	-
shuffled-qa7	200	200	-
shuffled-qa8	200	200	-
shuffled-qa9	200	200	-
shuffled-qa10	200	200	-
shuffled-qa11	200	200	-
shuffled-qa12	200	200	-
shuffled-qa13	200	200	-
shuffled-qa14	200	200	-
shuffled-qa15	250	250	-
shuffled-qa16	1000	1000	-
shuffled-qa17	125	125	-
shuffled-qa18	198	199	-
shuffled-qa19	1000	1000	-
shuffled-qa20	94	93	-
shuffled-10k-qa1	2000	200	-
shuffled-10k-qa2	2000	200	-
shuffled-10k-qa3	2000	200	-
shuffled-10k-qa4	10000	1000	-
shuffled-10k-qa5	2000	200	-
shuffled-10k-qa6	2000	200	-
shuffled-10k-qa7	2000	200	-
shuffled-10k-qa8	2000	200	-
shuffled-10k-qa9	2000	200	-
shuffled-10k-qa10	2000	200	-
shuffled-10k-qa11	2000	200	-
shuffled-10k-qa12	2000	200	-
shuffled-10k-qa13	2000	200	-
shuffled-10k-qa14	2000	200	-
shuffled-10k-qa15	2500	250	-
shuffled-10k-qa16	10000	1000	-
shuffled-10k-qa17	1250	125	-
shuffled-10k-qa18	1978	199	-
shuffled-10k-qa19	10000	1000	-
shuffled-10k-qa20	933	93	-

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Initial Data Collection and Normalization

Code to generate tasks is available on github

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

[More Information Needed]

Who are the annotators?

[More Information Needed]

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

Jesse Dodge and Andreea Gane and Xiang Zhang and Antoine Bordes and Sumit Chopra and Alexander Miller and Arthur Szlam and Jason Weston, at Facebook Research.

Licensing Information

Creative Commons Attribution 3.0 License

Citation Information

@misc{dodge2016evaluating,
      title={Evaluating Prerequisite Qualities for Learning End-to-End Dialog Systems}, 
      author={Jesse Dodge and Andreea Gane and Xiang Zhang and Antoine Bordes and Sumit Chopra and Alexander Miller and Arthur Szlam and Jason Weston},
      year={2016},
      eprint={1511.06931},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Contributions

Thanks to @gchhablani for adding this dataset.