Dataset Card for GEM

Dataset Summary

GEM is a benchmark environment for Natural Language Generation with a focus on its Evaluation, both through human annotations and automated Metrics.

GEM aims to:

• measure NLG progress across 13 datasets spanning many NLG tasks and languages.
• provide an in-depth analysis of data and models presented via data statements and challenge sets.
• develop standards for evaluation of generated text using both automated and human metrics.

It is our goal to regularly update GEM and to encourage toward more inclusive practices in dataset development by extending existing data or developing datasets for additional languages.

You can find more complete information in the dataset cards for each of the subsets:

• CommonGen
• Czech Restaurant
• DART
• E2E
• MLSum
• Schema-Guided Dialog
• WebNLG
• Wiki-Auto/ASSET/TURK
• WikiLingua
• XSum

The subsets are organized by task:

{
    "summarization": {
        "mlsum": ["mlsum_de", "mlsum_es"],
        "wiki_lingua": ["wiki_lingua_es_en", "wiki_lingua_ru_en", "wiki_lingua_tr_en", "wiki_lingua_vi_en"],
        "xsum": ["xsum"],
    },
    "struct2text": {
        "common_gen": ["common_gen"],
        "cs_restaurants": ["cs_restaurants"],
        "dart": ["dart"],
        "e2e": ["e2e_nlg"],
        "totto": ["totto"],
        "web_nlg": ["web_nlg_en", "web_nlg_ru"],
    },
    "simplification": {
        "wiki_auto_asset_turk": ["wiki_auto_asset_turk"],
    },
    "dialog": {
        "schema_guided_dialog": ["schema_guided_dialog"],
    },
}

Each example has one target per example in its training set, and a set of references (with one or more items) in its validation and test set.

Supported Tasks and Leaderboards

More Information Needed

Languages

More Information Needed

Dataset Structure

We show detailed information for up to 5 configurations of the dataset.

Data Instances

common_gen

• Size of downloaded dataset files: 1.76 MB
• Size of the generated dataset: 8.80 MB
• Total amount of disk used: 10.56 MB

An example of validation looks as follows.

{'concept_set_id': 0,
 'concepts': ['field', 'look', 'stand'],
 'gem_id': 'common_gen-validation-0',
 'references': ['The player stood in the field looking at the batter.',
                'The coach stands along the field, looking at the goalkeeper.',
                'I stood and looked across the field, peacefully.',
                'Someone stands, looking around the empty field.'],
 'target': 'The player stood in the field looking at the batter.'}

cs_restaurants

• Size of downloaded dataset files: 1.40 MB
• Size of the generated dataset: 1.25 MB
• Total amount of disk used: 2.64 MB

An example of validation looks as follows.

{'dialog_act': '?request(area)',
 'dialog_act_delexicalized': '?request(area)',
 'gem_id': 'cs_restaurants-validation-0',
 'references': ['Jakou lokalitu hledáte ?'],
 'target': 'Jakou lokalitu hledáte ?',
 'target_delexicalized': 'Jakou lokalitu hledáte ?'}

dart

• Size of downloaded dataset files: 28.01 MB
• Size of the generated dataset: 26.17 MB
• Total amount of disk used: 54.18 MB

An example of validation looks as follows.

{'dart_id': 0,
 'gem_id': 'dart-validation-0',
 'references': ['A school from Mars Hill, North Carolina, joined in 1973.'],
 'subtree_was_extended': True,
 'target': 'A school from Mars Hill, North Carolina, joined in 1973.',
 'target_sources': ['WikiSQL_decl_sents'],
 'tripleset': [['Mars Hill College', 'JOINED', '1973'], ['Mars Hill College', 'LOCATION', 'Mars Hill, North Carolina']]}

e2e_nlg

• Size of downloaded dataset files: 13.92 MB
• Size of the generated dataset: 11.58 MB
• Total amount of disk used: 25.50 MB

An example of validation looks as follows.

{'gem_id': 'e2e_nlg-validation-0',
 'meaning_representation': 'name[Alimentum], area[city centre], familyFriendly[no]',
 'references': ['There is a place in the city centre, Alimentum, that is not family-friendly.'],
 'target': 'There is a place in the city centre, Alimentum, that is not family-friendly.'}

mlsum_de

• Size of downloaded dataset files: 331.27 MB
• Size of the generated dataset: 907.00 MB
• Total amount of disk used: 1238.27 MB

An example of validation looks as follows.

{'date': '00/04/2019',
 'gem_id': 'mlsum_de-validation-0',
 'references': ['In einer Kleinstadt auf der Insel Usedom war eine junge Frau tot in ihrer Wohnung gefunden worden. Nun stehen zwei Bekannte unter Verdacht.'],
 'target': 'In einer Kleinstadt auf der Insel Usedom war eine junge Frau tot in ihrer Wohnung gefunden worden. Nun stehen zwei Bekannte unter Verdacht.',
 'text': 'Kerzen und Blumen stehen vor dem Eingang eines Hauses, in dem eine 18-jährige Frau tot aufgefunden wurde. In einer Kleinstadt auf der Insel Usedom war eine junge Frau tot in ...',
 'title': 'Tod von 18-Jähriger auf Usedom: Zwei Festnahmen',
 'topic': 'panorama',
 'url': 'https://www.sueddeutsche.de/panorama/usedom-frau-tot-festnahme-verdaechtige-1.4412256'}

mlsum_es

• Size of downloaded dataset files: 490.29 MB
• Size of the generated dataset: 1253.63 MB
• Total amount of disk used: 1743.92 MB

An example of validation looks as follows.

{'date': '05/01/2019',
 'gem_id': 'mlsum_es-validation-0',
 'references': ['El diseñador que dio carta de naturaleza al estilo genuinamente americano celebra el medio siglo de su marca entre grandes fastos y problemas financieros. Conectar con las nuevas generaciones es el regalo que precisa más que nunca'],
 'target': 'El diseñador que dio carta de naturaleza al estilo genuinamente americano celebra el medio siglo de su marca entre grandes fastos y problemas financieros. Conectar con las nuevas generaciones es el regalo que precisa más que nunca',
 'text': 'Un oso de peluche marcándose un heelflip de monopatín es todo lo que Ralph Lauren necesitaba esta Navidad. Estampado en un jersey de lana azul marino, supone la guinda que corona ...',
 'title': 'Ralph Lauren busca el secreto de la eterna juventud',
 'topic': 'elpais estilo',
 'url': 'http://elpais.com/elpais/2019/01/04/estilo/1546617396_933318.html'}

schema_guided_dialog

• Size of downloaded dataset files: 8.24 MB
• Size of the generated dataset: 43.66 MB
• Total amount of disk used: 51.91 MB

An example of validation looks as follows.

{'dialog_acts': [{'act': 2, 'slot': 'song_name', 'values': ['Carnivore']}, {'act': 2, 'slot': 'playback_device', 'values': ['TV']}],
 'dialog_id': '10_00054',
 'gem_id': 'schema_guided_dialog-validation-0',
 'prompt': 'Yes, I would.',
 'references': ['Please confirm the song Carnivore on tv.'],
 'target': 'Please confirm the song Carnivore on tv.',
 'turn_id': 15}

totto

• Size of downloaded dataset files: 179.03 MB
• Size of the generated dataset: 722.88 MB
• Total amount of disk used: 901.91 MB

An example of validation looks as follows.

{'example_id': '7391450717765563190',
 'gem_id': 'totto-validation-0',
 'highlighted_cells': [[3, 0], [3, 2], [3, 3]],
 'overlap_subset': 'True',
 'references': ['Daniel Henry Chamberlain was the 76th Governor of South Carolina from 1874.',
                'Daniel Henry Chamberlain was the 76th Governor of South Carolina, beginning in 1874.',
                'Daniel Henry Chamberlain was the 76th Governor of South Carolina who took office in 1874.'],
 'sentence_annotations': [{'final_sentence': 'Daniel Henry Chamberlain was the 76th Governor of South Carolina from 1874.',
                           'original_sentence': 'Daniel Henry Chamberlain (June 23, 1835 – April 13, 1907) was an American planter, lawyer, author and the 76th Governor of South Carolina '
                                                'from 1874 until 1877.',
                           'sentence_after_ambiguity': 'Daniel Henry Chamberlain was the 76th Governor of South Carolina from 1874.',
                           'sentence_after_deletion': 'Daniel Henry Chamberlain was the 76th Governor of South Carolina from 1874.'},
                          ...
                          ],
 'table': [[{'column_span': 1, 'is_header': True, 'row_span': 1, 'value': '#'},
            {'column_span': 2, 'is_header': True, 'row_span': 1, 'value': 'Governor'},
            {'column_span': 1, 'is_header': True, 'row_span': 1, 'value': 'Took Office'},
            {'column_span': 1, 'is_header': True, 'row_span': 1, 'value': 'Left Office'}],
           [{'column_span': 1, 'is_header': True, 'row_span': 1, 'value': '74'},
            {'column_span': 1, 'is_header': False, 'row_span': 1, 'value': '-'},
            {'column_span': 1, 'is_header': False, 'row_span': 1, 'value': 'Robert Kingston Scott'},
            {'column_span': 1, 'is_header': False, 'row_span': 1, 'value': 'July 6, 1868'}],
           ...
          ],
 'table_page_title': 'List of Governors of South Carolina',
 'table_section_text': 'Parties Democratic Republican',
 'table_section_title': 'Governors under the Constitution of 1868',
 'table_webpage_url': 'http://en.wikipedia.org/wiki/List_of_Governors_of_South_Carolina',
 'target': 'Daniel Henry Chamberlain was the 76th Governor of South Carolina from 1874.',
 'totto_id': 0}

web_nlg_en

• Size of downloaded dataset files: 12.35 MB
• Size of the generated dataset: 13.95 MB
• Total amount of disk used: 26.29 MB

An example of validation looks as follows.

{'category': 'Airport',
 'gem_id': 'web_nlg_en-validation-0',
 'input': ['Aarhus | leader | Jacob_Bundsgaard'],
 'references': ['The leader of Aarhus is Jacob Bundsgaard.'],
 'target': 'The leader of Aarhus is Jacob Bundsgaard.',
 'webnlg_id': 'dev/Airport/1/Id1'}

web_nlg_ru

• Size of downloaded dataset files: 7.28 MB
• Size of the generated dataset: 8.02 MB
• Total amount of disk used: 15.30 MB

An example of validation looks as follows.

{'category': 'Airport',
 'gem_id': 'web_nlg_ru-validation-0',
 'input': ['Punjab,_Pakistan | leaderTitle | Provincial_Assembly_of_the_Punjab'],
 'references': ['Пенджаб, Пакистан, возглавляется Провинциальной ассамблеей Пенджаба.', 'Пенджаб, Пакистан возглавляется Провинциальной ассамблеей Пенджаба.'],
 'target': 'Пенджаб, Пакистан, возглавляется Провинциальной ассамблеей Пенджаба.',
 'webnlg_id': 'dev/Airport/1/Id1'}

wiki_auto_asset_turk

• Size of downloaded dataset files: 121.37 MB
• Size of the generated dataset: 145.69 MB
• Total amount of disk used: 267.07 MB

An example of validation looks as follows.

{'gem_id': 'wiki_auto_asset_turk-validation-0',
 'references': ['The Gandalf Awards honor excellent writing in in fantasy literature.'],
 'source': 'The Gandalf Awards, honoring achievement in fantasy literature, were conferred by the World Science Fiction Society annually from 1974 to 1981.',
 'source_id': '350_691837-1-0-0',
 'target': 'The Gandalf Awards honor excellent writing in in fantasy literature.',
 'target_id': '350_691837-0-0-0'}

wiki_lingua_es_en

• Size of downloaded dataset files: 161.56 MB
• Size of the generated dataset: 274.28 MB
• Total amount of disk used: 435.84 MB

An example of validation looks as follows.

'references': ["Practice matted hair prevention from early in your cat's life. Make sure that your cat is grooming itself effectively. Keep a close eye on cats with long hair."],
'source': 'Muchas personas presentan problemas porque no cepillaron el pelaje de sus gatos en una etapa temprana de su vida, ya que no lo consideraban necesario. Sin embargo, a medida que...',
'target': "Practice matted hair prevention from early in your cat's life. Make sure that your cat is grooming itself effectively. Keep a close eye on cats with long hair."}

wiki_lingua_ru_en

• Size of downloaded dataset files: 161.56 MB
• Size of the generated dataset: 201.43 MB
• Total amount of disk used: 362.99 MB

An example of validation looks as follows.

{'gem_id': 'wiki_lingua_ru_en-val-0',
 'references': ['Get immediate medical care if you notice signs of a complication. Undergo diagnostic tests to check for gallstones and complications. Ask your doctor about your treatment '
                'options.'],
 'source': 'И хотя, скорее всего, вам не о чем волноваться, следует незамедлительно обратиться к врачу, если вы подозреваете, что у вас возникло осложнение желчекаменной болезни. Это ...',
 'target': 'Get immediate medical care if you notice signs of a complication. Undergo diagnostic tests to check for gallstones and complications. Ask your doctor about your treatment '
           'options.'}

wiki_lingua_tr_en

• Size of downloaded dataset files: 161.56 MB
• Size of the generated dataset: 9.87 MB
• Total amount of disk used: 171.42 MB

An example of validation looks as follows.

{'gem_id': 'wiki_lingua_tr_en-val-0',
 'references': ['Open Instagram. Go to the video you want to download. Tap ⋮. Tap Copy Link. Open  Google Chrome. Tap the address bar. Go to the SaveFromWeb site. Tap the "Paste Instagram Video" text box. Tap and hold the text box. Tap PASTE. Tap Download. Download the video. Find the video on your Android.'],
 'source': 'Instagram uygulamasının çok renkli kamera şeklindeki simgesine dokun. Daha önce giriş yaptıysan Instagram haber kaynağı açılır. Giriş yapmadıysan istendiğinde e-posta adresini ...',
 'target': 'Open Instagram. Go to the video you want to download. Tap ⋮. Tap Copy Link. Open  Google Chrome. Tap the address bar. Go to the SaveFromWeb site. Tap the "Paste Instagram Video" text box. Tap and hold the text box. Tap PASTE. Tap Download. Download the video. Find the video on your Android.'}

wiki_lingua_vi_en

• Size of downloaded dataset files: 161.56 MB
• Size of the generated dataset: 39.12 MB
• Total amount of disk used: 200.68 MB

An example of validation looks as follows.

{'gem_id': 'wiki_lingua_vi_en-val-0',
 'references': ['Select the right time of year for planting the tree. You will usually want to plant your tree when it is dormant, or not flowering, during cooler or colder times of year.'],
 'source': 'Bạn muốn cung cấp cho cây cơ hội tốt nhất để phát triển và sinh tồn. Trồng cây đúng thời điểm trong năm chính là yếu tố then chốt. Thời điểm sẽ thay đổi phụ thuộc vào loài cây ...',
 'target': 'Select the right time of year for planting the tree. You will usually want to plant your tree when it is dormant, or not flowering, during cooler or colder times of year.'}

xsum

• Size of downloaded dataset files: 243.08 MB
• Size of the generated dataset: 67.40 MB
• Total amount of disk used: 310.48 MB

An example of validation looks as follows.

{'document': 'Burberry reported pre-tax profits of £166m for the year to March. A year ago it made a loss of £16.1m, hit by charges at its Spanish operations.\n'
             'In the past year it has opened 21 new stores and closed nine. It plans to open 20-30 stores this year worldwide.\n'
             'The group has also focused on promoting the Burberry brand online...',
 'gem_id': 'xsum-validation-0',
 'references': ['Luxury fashion designer Burberry has returned to profit after opening new stores and spending more on online marketing'],
 'target': 'Luxury fashion designer Burberry has returned to profit after opening new stores and spending more on online marketing',
 'xsum_id': '10162122'}

Data Fields

The data fields are the same among all splits.

common_gen

• gem_id: a string feature.
• concept_set_id: a int32 feature.
• concepts: a list of string features.
• target: a string feature.
• references: a list of string features.

cs_restaurants

• gem_id: a string feature.
• dialog_act: a string feature.
• dialog_act_delexicalized: a string feature.
• target_delexicalized: a string feature.
• target: a string feature.
• references: a list of string features.

dart

• gem_id: a string feature.
• dart_id: a int32 feature.
• tripleset: a list of string features.
• subtree_was_extended: a bool feature.
• target_sources: a list of string features.
• target: a string feature.
• references: a list of string features.

e2e_nlg

• gem_id: a string feature.
• meaning_representation: a string feature.
• target: a string feature.
• references: a list of string features.

mlsum_de

• gem_id: a string feature.
• text: a string feature.
• topic: a string feature.
• url: a string feature.
• title: a string feature.
• date: a string feature.
• target: a string feature.
• references: a list of string features.

mlsum_es

• gem_id: a string feature.
• text: a string feature.
• topic: a string feature.
• url: a string feature.
• title: a string feature.
• date: a string feature.
• target: a string feature.
• references: a list of string features.

schema_guided_dialog

• gem_id: a string feature.
• act: a classification label, with possible values including AFFIRM (0), AFFIRM_INTENT (1), CONFIRM (2), GOODBYE (3), INFORM (4).
• slot: a string feature.
• values: a list of string features.
• dialog_id: a string feature.
• turn_id: a int32 feature.
• prompt: a string feature.
• target: a string feature.
• references: a list of string features.

totto

• gem_id: a string feature.
• totto_id: a int32 feature.
• table_page_title: a string feature.
• table_webpage_url: a string feature.
• table_section_title: a string feature.
• table_section_text: a string feature.
• column_span: a int32 feature.
• is_header: a bool feature.
• row_span: a int32 feature.
• value: a string feature.
• highlighted_cells: a list of int32 features.
• example_id: a string feature.
• original_sentence: a string feature.
• sentence_after_deletion: a string feature.
• sentence_after_ambiguity: a string feature.
• final_sentence: a string feature.
• overlap_subset: a string feature.
• target: a string feature.
• references: a list of string features.

web_nlg_en

• gem_id: a string feature.
• input: a list of string features.
• target: a string feature.
• references: a list of string features.
• category: a string feature.
• webnlg_id: a string feature.

web_nlg_ru

• gem_id: a string feature.
• input: a list of string features.
• target: a string feature.
• references: a list of string features.
• category: a string feature.
• webnlg_id: a string feature.

wiki_auto_asset_turk

• gem_id: a string feature.
• source_id: a string feature.
• target_id: a string feature.
• source: a string feature.
• target: a string feature.
• references: a list of string features.

wiki_lingua_es_en

• gem_id: a string feature.
• source: a string feature.
• target: a string feature.
• references: a list of string features.

wiki_lingua_ru_en

• gem_id: a string feature.
• source: a string feature.
• target: a string feature.
• references: a list of string features.

wiki_lingua_tr_en

• gem_id: a string feature.
• source: a string feature.
• target: a string feature.
• references: a list of string features.

wiki_lingua_vi_en

• gem_id: a string feature.
• source: a string feature.
• target: a string feature.
• references: a list of string features.

xsum

• gem_id: a string feature.
• xsum_id: a string feature.
• document: a string feature.
• target: a string feature.
• references: a list of string features.

Data Splits

common_gen

cs_restaurants

dart

e2e_nlg

mlsum_de

mlsum_es

schema_guided_dialog

totto

web_nlg_en

web_nlg_ru

wiki_auto_asset_turk

wiki_lingua_es_en

wiki_lingua_ru_en

wiki_lingua_tr_en

wiki_lingua_vi_en

xsum

Dataset Creation

Curation Rationale

More Information Needed

Source Data

Initial Data Collection and Normalization

More Information Needed

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

More Information Needed

Licensing Information

CC-BY-SA-4.0

Citation Information

@article{gem_benchmark,
  author    = {Sebastian Gehrmann and
               Tosin P. Adewumi and
               Karmanya Aggarwal and
               Pawan Sasanka Ammanamanchi and
               Aremu Anuoluwapo and
               Antoine Bosselut and
               Khyathi Raghavi Chandu and
               Miruna{-}Adriana Clinciu and
               Dipanjan Das and
               Kaustubh D. Dhole and
               Wanyu Du and
               Esin Durmus and
               Ondrej Dusek and
               Chris Emezue and
               Varun Gangal and
               Cristina Garbacea and
               Tatsunori Hashimoto and
               Yufang Hou and
               Yacine Jernite and
               Harsh Jhamtani and
               Yangfeng Ji and
               Shailza Jolly and
               Dhruv Kumar and
               Faisal Ladhak and
               Aman Madaan and
               Mounica Maddela and
               Khyati Mahajan and
               Saad Mahamood and
               Bodhisattwa Prasad Majumder and
               Pedro Henrique Martins and
               Angelina McMillan{-}Major and
               Simon Mille and
               Emiel van Miltenburg and
               Moin Nadeem and
               Shashi Narayan and
               Vitaly Nikolaev and
               Rubungo Andre Niyongabo and
               Salomey Osei and
               Ankur P. Parikh and
               Laura Perez{-}Beltrachini and
               Niranjan Ramesh Rao and
               Vikas Raunak and
               Juan Diego Rodriguez and
               Sashank Santhanam and
               Jo{\~{a}}o Sedoc and
               Thibault Sellam and
               Samira Shaikh and
               Anastasia Shimorina and
               Marco Antonio Sobrevilla Cabezudo and
               Hendrik Strobelt and
               Nishant Subramani and
               Wei Xu and
               Diyi Yang and
               Akhila Yerukola and
               Jiawei Zhou},
  title     = {The {GEM} Benchmark: Natural Language Generation, its Evaluation and
               Metrics},
  journal   = {CoRR},
  volume    = {abs/2102.01672},
  year      = {2021},
  url       = {https://arxiv.org/abs/2102.01672},
  archivePrefix = {arXiv},
  eprint    = {2102.01672}
}

Contributions

Thanks to @yjernite for adding this dataset.