Dataset Card for LR-Sum

LR-Sum is a automatic summarization dataset of newswire text with a focus on less resourced languages with a cc-by 4.0 license.

Dataset Details

Dataset Description

LR-Sum is a permissively-licensed dataset created with the goal of enabling further research in automatic summarization for less-resourced languages. LR-Sum contains human-written summaries for 39 languages, many of which are less-resourced. The data is based on the collection of the Multilingual Open Text corpus where the source data is public domain newswire collected from from Voice of America websites. LR-Sum is released under a Creative Commons license (CC BY 4.0), making it one of the most openly-licensed multilingual summarization datasets.

Curated by: BLT Lab: Chester Palen-Michel and Constantine Lignos
Shared by: Chester Palen-Michel
Language(s) (NLP): Albanian, Amharic, Armenian, Azerbaijani, Bengali, Bosnian, Burmese, Chinese, English, French, Georgian, Greek, Haitian Creole, Hausa, Indonesian, Khmer, Kinyarwanda, Korean, Kurdish, Lao, Macedonian, Northern Ndebele, Pashto, Persian, Portuguese, Russian, Serbian, Shona, Somali, Spanish, Swahili, Thai, Tibetan, Tigrinya, Turkish, Ukrainian, Urdu, Uzbek, Vietnamese
License: CC-BY 4.0

Dataset Sources [optional]

Multilingual Open Text v1.6 which is a collection of newswire text from Voice of America (VOA).

Paper: https://aclanthology.org/2023.findings-acl.427/
Repository: https://github.com/bltlab/lr-sum

Uses

The dataset is intended for research in automatic summarization in various languages, especially for less resourced languages.

Direct Use

The data can be used for training text generation models to generate short summaries of news articles in many languages. Automatic evaluation of automatic summarization is another use case, though we encourage also conducting human evaluation of any model trained for summarization.

Out-of-Scope Use

This dataset only includes newswire text, so models trained on the data may not be effective for out of domain summarization.

Dataset Structure

Each field is a string:

{
  'id': Article unique id
  'url': URL for the news article
  'title': The title of the news article
  'summary': The summary of the article
  'text': The full text of the news article not including title
}

Dataset Creation

Curation Rationale

Research in automatic summarization for less resourced languages.

Source Data

Voice of America (VOA)

Data Collection and Processing

See our paper for details on collection and processing.

Who are the source data producers?

Voice of America (VOA)

Annotation process

The summaries are found in news article meta data. More detail about the curation process can be found in our paper.

Who are the annotators?

The summaries are found in the news article meta data. The authors of the summaries are authors and staff for VOA.

Personal and Sensitive Information

The only sensative personal information would be information already published in news articles on VOA. See VOA's mission and values

Bias, Risks, and Limitations

The content in this dataset is newswire. See VOA's mission and values for more detail about the journalistic integrity and policy.

Recommendations

The data is newswire text. Training text generation models on this dataset will have similar risks and limitations to other text generation models including hallucinations and potentially inaccurate statements. For some languages that have fewer examples, issues with text generation models are likely to be more pronounced. The dataset is primarily released for research despite having a permissive license. We encourage users to thoroughly test and evaluate any models trained using this data before putting them into production environments.

Citation

If you make use of this dataset, please cite our paper using this bibtex:

BibTeX:

@inproceedings{palen-michel-lignos-2023-lr,
    title = "{LR}-Sum: Summarization for Less-Resourced Languages",
    author = "Palen-Michel, Chester  and
      Lignos, Constantine",
    editor = "Rogers, Anna  and
      Boyd-Graber, Jordan  and
      Okazaki, Naoaki",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.findings-acl.427",
    doi = "10.18653/v1/2023.findings-acl.427",
    pages = "6829--6844",
    abstract = "We introduce LR-Sum, a new permissively-licensed dataset created with the goal of enabling further research in automatic summarization for less-resourced languages.LR-Sum contains human-written summaries for 40 languages, many of which are less-resourced. We describe our process for extracting and filtering the dataset from the Multilingual Open Text corpus (Palen-Michel et al., 2022).The source data is public domain newswire collected from from Voice of America websites, and LR-Sum is released under a Creative Commons license (CC BY 4.0), making it one of the most openly-licensed multilingual summarization datasets. We describe abstractive and extractive summarization experiments to establish baselines and discuss the limitations of this dataset.",
}

Dataset Card Authors

Chester Palen-Michel @cpalenmichel

Dataset Card Contact

Chester Palen-Michel @cpalenmichel