cs_restaurants

  • Description:

Czech data-to-text dataset in the restaurant domain. The input meaning representations contain a dialogue act type (inform, confirm etc.), slots (food, area, etc.) and their values. It originated as a translation of the English San Francisco Restaurants dataset by Wen et al. (2015).

Split Examples
'test' 842
'train' 3,569
'validation' 781
  • Feature structure:
FeaturesDict({
    'delex_input_text': FeaturesDict({
        'table': Sequence({
            'column_header': string,
            'content': string,
            'row_number': int16,
        }),
    }),
    'delex_target_text': string,
    'input_text': FeaturesDict({
        'table': Sequence({
            'column_header': string,
            'content': string,
            'row_number': int16,
        }),
    }),
    'target_text': string,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
delex_input_text FeaturesDict
delex_input_text/table Sequence
delex_input_text/table/column_header Tensor string
delex_input_text/table/content Tensor string
delex_input_text/table/row_number Tensor int16
delex_target_text Tensor string
input_text FeaturesDict
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{dusek_neural_2019,
        author = {Dušek, Ondřej and Jurčíček, Filip},
        title = {Neural {Generation} for {Czech}: {Data} and {Baselines} },
        shorttitle = {Neural {Generation} for {Czech} },
        url = {https://www.aclweb.org/anthology/W19-8670/},
        urldate = {2019-10-18},
        booktitle = {Proceedings of the 12th {International} {Conference} on {Natural} {Language} {Generation} ({INLG} 2019)},
        month = oct,
        address = {Tokyo, Japan},
        year = {2019},
        pages = {563--574},
        abstract = {We present the first dataset targeted at end-to-end NLG in Czech in the restaurant domain, along with several strong baseline models using the sequence-to-sequence approach. While non-English NLG is under-explored in general, Czech, as a morphologically rich language, makes the task even harder: Since Czech requires inflecting named entities, delexicalization or copy mechanisms do not work out-of-the-box and lexicalizing the generated outputs is non-trivial. In our experiments, we present two different approaches to this this problem: (1) using a neural language model to select the correct inflected form while lexicalizing, (2) a two-step generation setup: our sequence-to-sequence model generates an interleaved sequence of lemmas and morphological tags, which are then inflected by a morphological generator.},
}