deepmind/narrativeqa

Dataset Card for Narrative QA

Dataset Summary

NarrativeQA is an English-lanaguage dataset of stories and corresponding questions designed to test reading comprehension, especially on long documents.

Supported Tasks and Leaderboards

The dataset is used to test reading comprehension. There are 2 tasks proposed in the paper: "summaries only" and "stories only", depending on whether the human-generated summary or the full story text is used to answer the question.

Languages

English

Dataset Structure

Data Instances

A typical data point consists of a question and answer pair along with a summary/story which can be used to answer the question. Additional information such as the url, word count, wikipedia page, are also provided.

A typical example looks like this:

{
    "document": {
        "id": "23jncj2n3534563110",
        "kind": "movie",
        "url": "https://www.imsdb.com/Movie%20Scripts/Name%20of%20Movie.html",
        "file_size": 80473,
        "word_count": 41000,
        "start": "MOVIE screenplay by",
        "end": ". THE END",
        "summary": {
            "text": "Joe Bloggs begins his journey exploring...",
            "tokens": ["Joe", "Bloggs", "begins", "his", "journey", "exploring",...],
            "url": "http://en.wikipedia.org/wiki/Name_of_Movie",
            "title": "Name of Movie (film)"
        },
        "text": "MOVIE screenplay by John Doe\nSCENE 1..."
    },
    "question": {
        "text": "Where does Joe Bloggs live?",
        "tokens": ["Where", "does", "Joe", "Bloggs", "live", "?"],
    },
    "answers": [
        {"text": "At home", "tokens": ["At", "home"]},
        {"text": "His house", "tokens": ["His", "house"]}
    ]
}

Data Fields

• document.id - Unique ID for the story.
• document.kind - "movie" or "gutenberg" depending on the source of the story.
• document.url - The URL where the story was downloaded from.
• document.file_size - File size (in bytes) of the story.
• document.word_count - Number of tokens in the story.
• document.start - First 3 tokens of the story. Used for verifying the story hasn't been modified.
• document.end - Last 3 tokens of the story. Used for verifying the story hasn't been modified.
• document.summary.text - Text of the wikipedia summary of the story.
• document.summary.tokens - Tokenized version of document.summary.text.
• document.summary.url - Wikipedia URL of the summary.
• document.summary.title - Wikipedia Title of the summary.
• question - {"text":"...", "tokens":[...]} for the question about the story.
• answers - List of {"text":"...", "tokens":[...]} for valid answers for the question.

Data Splits

The data is split into training, valiudation, and test sets based on story (i.e. the same story cannot appear in more than one split):

Train	Valid	Test
32747	3461	10557

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Initial Data Collection and Normalization

Stories and movies scripts were downloaded from Project Gutenburg and a range of movie script repositories (mainly imsdb).

Who are the source language producers?

The language producers are authors of the stories and scripts as well as Amazon Turk workers for the questions.

Annotations

Annotation process

Amazon Turk Workers were provided with human written summaries of the stories (To make the annotation tractable and to lead annotators towards asking non-localized questions). Stories were matched with plot summaries from Wikipedia using titles and verified the matching with help from human annotators. The annotators were asked to determine if both the story and the summary refer to a movie or a book (as some books are made into movies), or if they are the same part in a series produced in the same year. Annotators on Amazon Mechanical Turk were instructed to write 10 question–answer pairs each based solely on a given summary. Annotators were instructed to imagine that they are writing questions to test students who have read the full stories but not the summaries. We required questions that are specific enough, given the length and complexity of the narratives, and to provide adiverse set of questions about characters, events, why this happened, and so on. Annotators were encouraged to use their own words and we prevented them from copying. We asked for answers that are grammatical, complete sentences, and explicitly allowed short answers (one word, or a few-word phrase, or ashort sentence) as we think that answering with a full sentence is frequently perceived as artificial when asking about factual information. Annotators were asked to avoid extra, unnecessary information in the question or the answer, and to avoid yes/no questions or questions about the author or the actors.

Who are the annotators?

Amazon Mechanical Turk workers.

Personal and Sensitive Information

None

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

[More Information Needed]

Licensing Information

The dataset is released under a Apache-2.0 License.

Citation Information

@article{kocisky-etal-2018-narrativeqa,
    title = "The {N}arrative{QA} Reading Comprehension Challenge",
    author = "Ko{\v{c}}isk{\'y}, Tom{\'a}{\v{s}}  and
      Schwarz, Jonathan  and
      Blunsom, Phil  and
      Dyer, Chris  and
      Hermann, Karl Moritz  and
      Melis, G{\'a}bor  and
      Grefenstette, Edward",
    editor = "Lee, Lillian  and
      Johnson, Mark  and
      Toutanova, Kristina  and
      Roark, Brian",
    journal = "Transactions of the Association for Computational Linguistics",
    volume = "6",
    year = "2018",
    address = "Cambridge, MA",
    publisher = "MIT Press",
    url = "https://aclanthology.org/Q18-1023",
    doi = "10.1162/tacl_a_00023",
    pages = "317--328",
    abstract = "Reading comprehension (RC){---}in contrast to information retrieval{---}requires integrating information and reasoning about events, entities, and their relations across a full document. Question answering is conventionally used to assess RC ability, in both artificial agents and children learning to read. However, existing RC datasets and tasks are dominated by questions that can be solved by selecting answers using superficial information (e.g., local context similarity or global term frequency); they thus fail to test for the essential integrative aspect of RC. To encourage progress on deeper comprehension of language, we present a new dataset and set of tasks in which the reader must answer questions about stories by reading entire books or movie scripts. These tasks are designed so that successfully answering their questions requires understanding the underlying narrative rather than relying on shallow pattern matching or salience. We show that although humans solve the tasks easily, standard RC models struggle on the tasks presented here. We provide an analysis of the dataset and the challenges it presents.",
}

Contributions

Thanks to @ghomasHudson for adding this dataset.