web_nlg

  • Description:

The data contains sets of 1 to 7 triples of the form subject-predicate-object extracted from (DBpedia)[https://wiki.dbpedia.org/] and natural language text that's a verbalisation of these triples. The test data spans 15 different domains where only 10 appear in the training data. The dataset follows a standarized table format.

Split Examples
'test_all' 4,928
'test_unseen' 2,433
'train' 18,102
'validation' 2,268
  • Feature structure:
FeaturesDict({
    'input_text': FeaturesDict({
        'context': string,
        'table': Sequence({
            'column_header': string,
            'content': string,
            'row_number': int16,
        }),
    }),
    'target_text': string,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
input_text FeaturesDict
input_text/context Tensor string
input_text/table Sequence
input_text/table/column_header Tensor string
input_text/table/content Tensor string
input_text/table/row_number Tensor int16
target_text Tensor string
  • Citation:
@inproceedings{gardent2017creating,
    title = ""Creating Training Corpora for {NLG} Micro-Planners"",
    author = ""Gardent, Claire  and
      Shimorina, Anastasia  and
      Narayan, Shashi  and
      Perez-Beltrachini, Laura"",
    booktitle = ""Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"",
    month = jul,
    year = ""2017"",
    address = ""Vancouver, Canada"",
    publisher = ""Association for Computational Linguistics"",
    doi = ""10.18653/v1/P17-1017"",
    pages = ""179--188"",
    url = ""https://www.aclweb.org/anthology/P17-1017.pdf""
}