mlsum.json

{
  "overview": {
    "where": {
      "has-leaderboard": "no",
      "leaderboard-url": "N/A",
      "leaderboard-description": "N/A",
      "website": "N/A",
      "data-url": "[Gitlab](https://gitlab.lip6.fr/scialom/mlsum_data/-/tree/master/MLSUM)",
      "paper-url": "[ACL Anthology](https://aclanthology.org/2020.emnlp-main.647/)",
      "paper-bibtext": "```\n@inproceedings{scialom-etal-2020-mlsum,\n    title = \"{MLSUM}: The Multilingual Summarization Corpus\",\n    author = \"Scialom, Thomas  and\n      Dray, Paul-Alexis  and\n      Lamprier, Sylvain  and\n      Piwowarski, Benjamin  and\n      Staiano, Jacopo\",\n    booktitle = \"Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)\",\n    month = nov,\n    year = \"2020\",\n    address = \"Online\",\n    publisher = \"Association for Computational Linguistics\",\n    url = \"https://aclanthology.org/2020.emnlp-main.647\",\n    doi = \"10.18653/v1/2020.emnlp-main.647\",\n    pages = \"8051--8067\",\n    abstract = \"We present MLSUM, the first large-scale MultiLingual SUMmarization dataset. Obtained from online newspapers, it contains 1.5M+ article/summary pairs in five different languages {--} namely, French, German, Spanish, Russian, Turkish. Together with English news articles from the popular CNN/Daily mail dataset, the collected data form a large scale multilingual dataset which can enable new research directions for the text summarization community. We report cross-lingual comparative analyses based on state-of-the-art systems. These highlight existing biases which motivate the use of a multi-lingual dataset.\",\n}\n```",
      "contact-email": "{thomas,paul-alexis,jacopo}@recital.ai, {sylvain.lamprier,benjamin.piwowarski}@lip6.fr",
      "contact-name": "Thomas Scialom"
    },
    "languages": {
      "is-multilingual": "yes",
      "license": "other: Other license",
      "task-other": "N/A",
      "language-names": [
        "German",
        "Spanish, Castilian"
      ],
      "language-dialects": "There is only one dialect per language, Hochdeutsch for German and Castilian Spanish for Spanish. ",
      "language-speakers": "The German articles are crawled from S\u00fcddeutsche Zeitung and the Spanish ones from El Pais.",
      "intended-use": "The intended use of this dataset is to augment existing datasets for English news summarization with additional languages. ",
      "license-other": "Restricted to non-commercial research purposes.",
      "task": "Summarization",
      "communicative": "The speaker is required to produce a high quality summary of news articles in the same language as the input article."
    },
    "credit": {
      "organization-type": [
        "other"
      ],
      "organization-names": "CNRS, Sorbonne Universit\u00e9, reciTAL",
      "creators": "Thomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano",
      "funding": "Funding information is not specified.",
      "gem-added-by": "The original data card was written by Pedro Henrique Martins (Instituto de Telecomunica\u00e7\u00f5es) and Sebastian Gehrmann (Google Research) extended and updated it to the v2 format. The COVID challenge set was created by Laura Perez-Beltrachini (University of Edinburgh).  Data cleaning was done by Juan Diego Rodriguez (UT  Austin)."
    },
    "structure": {
      "data-fields": "The data fields are:\n\n- `text`: the source article (`string`).\n- `summary`: the output summary (`string`).\n- `topic`: the topic of the article (`string`).\n- `url`: the article's url (`string`).\n- `title`: the article's title (`string`).\n- `date`: the article's date (`string`).",
      "structure-description": "The structure follows previously released datasets. The `topic` and `title` fields were added to enable additional tasks like title generation and topic detection.",
      "structure-labels": "They are human written highlights or summaries scraped from the same website. ",
      "structure-example": "```\n{\n 'date': '00/01/2010',\n 'gem_id': 'mlsum_de-train-2',\n 'gem_parent_id': 'mlsum_de-train-2',\n 'references': [],\n 'target': 'Oskar Lafontaine gibt den Parteivorsitz der Linken ab - und seine Kollegen streiten, wer ihn beerben soll. sueddeutsche.de stellt die derzeit aussichtsreichsten Anw\u00e4rter f\u00fcr F\u00fchrungsaufgaben vor. Mit Vote.',\n 'text': 'Wenn an diesem Montag die Landesvorsitzenden der Linken \u00fcber die Nachfolger der derzeitigen Chefs Lothar Bisky und Oskar Lafontaine sowie des Bundesgesch\u00e4ftsf\u00fchrers Dietmar Bartsch beraten, geht es nicht nur darum, wer die Partei f\u00fchren soll. Es geht auch um die k\u00fcnftige Ausrichtung und St\u00e4rke einer Partei, die vor allem von Lafontaine zusammengehalten worden war. Ihm war es schlie\u00dflich vor f\u00fcnf Jahren gelungen, aus der ostdeutschen PDS und der westedeutschen WASG eine Partei zu formen. Eine Partei allerdings, die zerrissen ist in Ost und West, in Regierungswillige und ewige Oppositionelle, in Realos und Ideologen, in gem\u00e4\u00dfigte und radikale Linke. Wir stellen m\u00f6gliche Kandidaten vor. Stimmen Sie ab: Wen halten Sie f\u00fcr geeignet und wen f\u00fcr unf\u00e4hig? Kampf um Lafontaines Erbe: Gregor Gysi Sollte \u00fcberhaupt jemand die Partei alleine f\u00fchren, wie es sich viele Ostdeutsche w\u00fcnschen, k\u00e4me daf\u00fcr wohl nur der 62-j\u00e4hrige Gregor Gysi in Betracht. Er ist nach Lafontaine einer der bekanntesten Politiker der Linken und derzeit Fraktionsvorsitzender der Partei im Bundestag. Allerdings ist der ehemalige PDS-Vorsitzende und Rechtsanwalt nach drei Herzinfarkten gesundheitlich angeschlagen. Wahrscheinlich w\u00e4re deshalb, dass er die zerstrittene Partei nur \u00fcbergangsweise f\u00fchrt. Doch noch ist nicht klar, ob eine Person allein die Partei f\u00fchren soll oder eine Doppelspitze. Viele Linke w\u00fcnschen sich ein Duo aus einem westdeutschen und einem ostdeutschen Politiker, Mann und Frau. Foto: Getty Images',\n 'title': 'Personaldebatte bei der Linken - Wer kommt nach Lafontaine?',\n 'topic': 'politik',\n 'url': 'https://www.sueddeutsche.de/politik/personaldebatte-bei-der-linken-wer-kommt-nach-lafontaine-1.70041'\n}\n```",
      "structure-splits": "The statistics of the original dataset are:\n\n|\t               |  Dataset  | Train   | Validation  |  Test  | Mean article length | Mean summary length |  \n| :---        \t   | :----:   | :---:   |   :---:      |  :---:  |       :---:          |      :---:           |\n| German      \t   | 242,982  | 220,887 |11,394     |10,701 |570.6 (words)  | 30.36 (words)  |\n| Spanish          | 290,645  | 266,367 |10,358     |13,920 |800.5 (words)  |20.71 (words)  |\n\nThe statistics of the cleaned version of the dataset are:\n\n|\t               |  Dataset  | Train   | Validation  |  Test  |\n| :---        \t   | :----:   | :---:   |   :---:      |  :---:  |\n| German      \t   | 242,835  | 220,887 |11,392     |10,695 |\n| Spanish          | 283,228  |259,886  |9,977     |13,365 |\n\nThe COVID challenge sets have 5058 (de) and 1938 (es) examples. ",
      "structure-splits-criteria": "The training set contains data from 2010 to 2018. Data from 2019 (~10% of the dataset) is used for validation (up to May) and testing(May-December 2019). ",
      "structure-outlier": "Some topics are less represented within the dataset (e.g., Financial news in German and Television in Spanish). \n"
    },
    "what": {
      "dataset": "MLSum is a multilingual summarization dataset crawled from different news websites. The GEM version supports the German and Spanish subset alongside specifically collected challenge sets for COVID-related articles to test out-of-domain generalization."
    }
  },
  "curation": {
    "original": {
      "is-aggregated": "yes",
      "aggregated-sources": "www.lemonde.fr\nwww.sueddeutsche.de\nwww.elpais.com\nwww.mk.ru\nwww.internethaber.com",
      "rationale": "The rationale was to create a multilingual news summarization dataset that mirrors the format of popular English datasets like XSum or CNN/DM. ",
      "communicative": "The speaker is required to produce a high quality summary of news articles in the same language as the input article."
    },
    "language": {
      "found": [
        "Multiple websites"
      ],
      "crowdsourced": [],
      "created": "N/A",
      "machine-generated": "N/A",
      "validated": "not validated",
      "is-filtered": "algorithmically",
      "filtered-criteria": "In the original dataset, only one filter was applied: all the articles shorter than 50 words or summaries shorter than 10 words are discarded. \n\nThe GEM version additionally applies langID filter to ensure that articles are in the correct language. ",
      "obtained": [
        "Found"
      ],
      "producers-description": "The language producers are professional journalists. ",
      "topics": "4/5 of the original languages report their topics (except Turkish) and the distributions differ between sources. The dominant topics in German are Politik, Sport, Wirtschaft (economy). The dominant topics in Spanish are actualidad (current news) and opinion. French and Russian are different as well but we omit these languages in the GEM version. ",
      "pre-processed": "N/A"
    },
    "annotations": {
      "origin": "none",
      "rater-number": "N/A",
      "rater-qualifications": "N/A",
      "rater-training-num": "N/A",
      "rater-test-num": "N/A",
      "rater-annotation-service-bool": "no",
      "rater-annotation-service": [],
      "values": "N/A",
      "quality-control": [],
      "quality-control-details": "N/A"
    },
    "consent": {
      "has-consent": "no",
      "consent-policy": "N/A",
      "consent-other": "N/A",
      "no-consent-justification": "The copyright remains with the original data creators and the usage permission is restricted to non-commercial uses. "
    },
    "pii": {
      "has-pii": "yes/very likely",
      "no-pii-justification": "N/A",
      "is-pii-identified": "no identification",
      "pii-identified-method": "N/A",
      "is-pii-replaced": "N/A",
      "pii-replaced-method": "N/A",
      "pii-categories": [
        "sensitive information",
        "generic PII"
      ]
    },
    "maintenance": {
      "has-maintenance": "no",
      "description": "N/A",
      "contact": "N/A",
      "contestation-mechanism": "N/A",
      "contestation-link": "N/A",
      "contestation-description": "N/A"
    }
  },
  "gem": {
    "rationale": {
      "sole-task-dataset": "yes",
      "sole-language-task-dataset": "yes",
      "distinction-description": "In our configuration, the dataset is fully non-English. ",
      "contribution": "As the first large-scale multilingual summarization dataset, it enables evaluation of summarization models beyond English. ",
      "model-ability": "Content Selection, Content Planning, Realization"
    },
    "curation": {
      "has-additional-curation": "yes",
      "modification-types": [
        "data points removed",
        "data points added"
      ],
      "modification-description": "The modifications done to the original dataset are the following:\n\n- Selection of 2 languages (Spanish and German) out of the dataset 5 languages due to copyright restrictions.\n- Removal of duplicate articles.\n- Manually removal of article-summary pairs for which the summary is not related to the article.\n- Removal of article-summary pairs written in a different language (detected using the [langdetect](https://pypi.org/project/langdetect/) library).",
      "has-additional-splits": "yes",
      "additional-splits-description": "For both selected languages (German and Spanish), we compiled time-shifted test data in the form of new articles for the second semester of 2020 with Covid19-related keywords. We collected articles from the same German and Spanish outlets as the original MLSUM datasets (El Pais and S\u00fcddeutsche Zeitung). We used the scripts provided for the re-creation of the [MLSUM datasets](https://github.com/recitalAI/MLSUM). The new challenge test set for German contains 5058 instances and the Spanish one contains 1938.\n\nWe additionally sample 500 training and validation points as additional challenge sets to measure overfitting. ",
      "additional-splits-capacicites": "Generalization to unseen topics. "
    },
    "starting": {
      "research-pointers": "N/A",
      "technical-terms": "N/A"
    }
  },
  "results": {
    "results": {
      "other-metrics-definitions": "Novelty: Number of generated n-grams not included in the source articles.",
      "has-previous-results": "yes",
      "current-evaluation": "The GEM benchmark results (https://gem-benchmark.com/results) report a wide range of metrics include lexical overlap metrics but also semantic ones like BLEURT and BERT-Score.",
      "previous-results": "N/A",
      "model-abilities": "Content Selection, Content Planning, Realization",
      "metrics": [
        "METEOR",
        "ROUGE",
        "Other: Other Metrics"
      ],
      "original-evaluation": "ROUGE and METEOR both measure n-gram overlap with a focus on recall and are standard summarization metrics. Novelty is often reported alongside them to characterize how much a model diverges from its inputs. "
    }
  },
  "context": {
    "previous": {
      "is-deployed": "no",
      "described-risks": "N/A",
      "changes-from-observation": "N/A"
    },
    "underserved": {
      "helps-underserved": "no",
      "underserved-description": "N/A"
    },
    "biases": {
      "has-biases": "no",
      "bias-analyses": "N/A"
    }
  }
}