gigaword

ds = tfds.load('huggingface:gigaword')

Headline-generation on a corpus of article pairs from Gigaword consisting of
around 4 million articles. Use the 'org_data' provided by
https://github.com/microsoft/unilm/ which is identical to
https://github.com/harvardnlp/sent-summary but with better format.

There are two features:
  - document: article.
  - summary: headline.

{
    "document": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "summary": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}