Dataset Card for SIB-200

Table of Contents

• Table of Contents
• Dataset Description
  • Dataset Summary
  • Supported Tasks and Leaderboards
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Social Impact of Dataset
  • Discussion of Biases
  • Other Known Limitations
• Additional Information
  • Dataset Curators
  • Licensing Information
  • Citation Information
  • Contributions

Dataset Description

• Homepage: homepage
• Repository: github
• Paper: paper
• Point of Contact: d.adelani@ucl.ac.uk

Dataset Summary

SIB-200 is the largest publicly available topic classification dataset based on Flores-200 covering 205 languages and dialects.

The train/validation/test sets are available for all the 205 languages.

Supported Tasks and Leaderboards

• topic classification: categorize wikipedia sentences into topics e.g science/technology, sports or politics.

Languages

There are 205 languages available :

Dataset Structure

Data Instances

The examples look like this for English:

from datasets import load_dataset
data = load_dataset('Davlan/sib200', 'eng_Latn') 

# Please, specify the language code

# A data point example is below:

{
'label': 0, 
'index_id': 1523, 
'text': 'Mutation adds new genetic variation, and selection removes it from the pool of expressed variation.'
}

Data Fields

• label: topic id
• index_id: sentence id in flores-200
• text: text

The topics correspond to this list:

"science/technology", "travel", "politics", "sports", "health", "entertainment", "geography"

Data Splits

For all languages, there are three splits.

The original splits were named train, dev and test and they correspond to the train, validation and test splits.

The splits have the following sizes :

Dataset Creation

Curation Rationale

The dataset was introduced to introduce new resources for 205 languages, many are under-served for natural language processing.

[More Information Needed]

Source Data

The source of the data is from the news domain, details can be found here ****

Initial Data Collection and Normalization

The articles were word-tokenized, information on the exact pre-processing pipeline is unavailable.

Who are the source language producers?

The source language was produced by journalists and writers employed by the news agency and newspaper mentioned above.

Annotations

Annotation process

Details can be found here **

Who are the annotators?

Annotators were recruited from Masakhane

Personal and Sensitive Information

The data is sourced from newspaper source and only contains mentions of public figures or individuals

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

Users should keep in mind that the dataset only contains news text, which might limit the applicability of the developed systems to other domains.

Additional Information

Dataset Curators

Licensing Information

The licensing status of the data is CC 4.0 Commercial

Citation Information

Provide the BibTex-formatted reference for the dataset. For example:

@misc{adelani2023sib200,
      title={SIB-200: A Simple, Inclusive, and Big Evaluation Dataset for Topic Classification in 200+ Languages and Dialects}, 
      author={David Ifeoluwa Adelani and Hannah Liu and Xiaoyu Shen and Nikita Vassilyev and Jesujoba O. Alabi and Yanke Mao and Haonan Gao and Annie En-Shiun Lee},
      year={2023},
      eprint={2309.07445},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Contributions

Thanks to @dadelani for adding this dataset.