Table of Contents

Dataset Description

Dataset and Task Summary

MLSum is a large-scale multiLingual summarization dataset. It is buillt from online news outlets and contains over 1.5M article-summary pairs in five different languages: French, German, Spanish, Russian, Turkish.

Why is this dataset part of GEM?

This dataset is part of the GEM benchmark for the task of summarization, alongside Wikilingua and Xsum, and acts as a large-scale, high-quality resource for cross-lingual summarization.


MLSum contains article-summary pairs in 5 languages. In GEM the languages that will be benchmarked are Spanish (ISO 639-1: es) and German (ISO 639-1: de).

Meta Information

Dataset Curators

This dataset was developed by a team of researchers from reciTAL and the Sorbonne Université: Thomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano.

Licensing Information

Usage of dataset is restricted to non-commercial research purposes only. Copyright belongs to the original copyright holders.

Citation Information

Please cite the following paper:

    title = "{MLSUM}: The Multilingual Summarization Corpus",
    author = {Scialom, Thomas  and Dray, Paul-Alexis  and Lamprier, Sylvain  and Piwowarski, Benjamin  and Staiano, Jacopo},
    booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
    year = {2020}


This dataset has no corresponding public leaderboard.

Dataset Structure

Data Instances

A train data example:

    "text": "This is a text",
	"summary": "A text",
    "topic": "football",
	"url": "",
	"title": "A sample",
    "date": "01/01/2001"

A validation or test data example:

    "text": "This is a text",
    "topic": "football",
	"url": "",
	"title": "A sample",
    "date": "01/01/2001"

Data Fields

The data fields are:

  • text: the source article (string).
  • summary: the output summary (string).
  • topic: the topic of the article (string).
  • url: the article's url (string).
  • title: the article's title (string).
  • date: the article's date (string).

Data Statistics

The statistics of the original dataset are:

Dataset Train Validation Test Mean article length Mean summary length
German 242,982 220,887 11,394 10,701 570.6 (words) 30.36 (words)
Spanish 290,645 266,367 10,358 13,920 800.5 (words) 20.71 (words)

The statistics of the cleaned version of the dataset are:

Dataset Train Validation Test
German 242,835 220,887 11,392 10,695
Spanish 283,228 259,886 9,977 13,365

Dataset Creation

The authors extracted gold-standard document-summary pairs, by considering news articles as the source document and their paired highlights/description as the summary.

Curation Rationale

The dataset was created in order to enable training and evaluation of summarization models in different languages. It allows to understand whether a given model is as fitted for a specific language as for any other.

Communicative Goal

The speaker is required to produce high quality summaries of articles.

Source Data

The article-summary pairs were extracted from the online version of the following newspapers: Le Monde4 (French), Süddeutsche Zeitung (German), El Pais (Spanish), Moskovskij Komsomolets (Russian), and Internet Haber (Turkish).

Initial Data Collection and Normalization

The authors gathered archived articles from 2010 to 2019. To avoid articles containing mostly audiovisual content, the authors discarded all article-summary pairs for which the articles were shorter than 50 words or summaries were shorter than 10 words.

Who are the source language producers?

No information is provided in the paper.


Any additional annotations are not collected for this dataset.

Annotation process


Who are the annotators?


Personal and Sensitive Information


Changes to the Original Dataset for GEM

The modifications done to the original dataset are the following:

  • Selection of 2 languages (Spanish and German) out of the dataset 5 languages.
  • Removal of duplicate items.
  • Manually removal of article-summary pairs for which the summary is not related to the article.
  • Removal of article-summary pairs written in a different language (detected using the langdetect library).

Special test sets

Data shift

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 Süddeutsche Zeitung). We used the scripts provided for the re-creation of the MLSUM datasets. The new challenge test set for German contains 5058 instances and the Spanish one contains 1938.

Considerations for Using the Data

Social Impact of the Dataset


Impact on Underserved Communities


Discussion of Biases


Other Known Limitations


Getting started with in-depth research on the task

Download the dataset using the Huggingface API.