MAximeSobrier/WiLI-2018-fixed

Dataset Card for wili_2018_fixed

Dataset Summary

This dataset includes many fixes to the WiLI-2018 dataset:

• fix mislabels
• sentences in multiple languages, with more than 1 prominent langaue, were removed
• duplicate lines were removed

test.csv: 168 lines removed, 1283 labels updated

train.csv: 127 lines removed, 1157 labels updated

Examples of labels updated:

• "Fischer, Wladimir, »The Role of Dositej Obradovic in the Construction of Serbian Identities During the 19th Century,« Spaces of Identity (1.3, 2001), 67–87.",94 (Slovak) -> 41 (English)
• "Jean Royer (réal.) et Jacques Bloch-Morange (itw), « Albert Ducrocq : la cybernétique », 1re partie, La page des sciences, Radiodiffusion-télévision française, 14 décembre 1961, 10 minutes 27 secondes [en linha].",176 (Occitan) -> 82 (French)
• The Wikimedia Foundation is conducting a request for comment on a proposed program that could provide legal assistance to users in specific support roles who are named in a legal complaint as a defendant because of those roles. We wanted to be sure that your community was aware of this discussion and would have a chance to participate in that discussion.,120 (Picard) -> 41 (English)

Examples of sentences with multiple prominent languages removed:

• "6. Hadj Yusuf AbdulRahman (editor), Kitab Rabi'a al-Qulub fi Dhikr Manaqib wa Fada'il Sayyidina as Sheikh Nur Hussein (""""The Springtime of Hearts in Memory of the Virtues and Merits of Our Lord the Sheikh Nur Hussein""""), Cairo , 1927",68
• "Na 2017 ọ na a depụtara na 2000 Outstanding Intellectuals of the 21st Century (10th mbipụta) na, Dictionary of International Biography (39th mbipụta) site na International Biographical Centre, Cambridge.",216

The dataset is no longer balanced.

WiLI-2018, the Wikipedia language identification benchmark dataset, contains 235000 paragraphs of 235 languages. The dataset is balanced and a train-test split is provided.

Supported Tasks and Leaderboards

[Needs More Information]

Languages

235 Different Languages

Dataset Structure

Data Instances

"Ti Turkia ket maysa a demokrata, sekular, unitario, batay-linteg a republika nga addaan ti taga-ugma a tinawtawid a kultura. Ti Turkia ket umadadu a naipatipon iti Laud babaen ti panagkameng kadagiti organisasion a kas ti Konsilo iti Europa, NATO, OECD, OSCE ken ti G-20 a dagiti kangrunaan nga ekonomia. Ti Turkia ket nangrugi a nakitulag ti napno a panagkameng iti Kappon ti Europa idi 2005, nga isu ket maysa idin a kumaduaan a kameng iti Europeano a Komunidad ti Ekonomia manipud idi 1963 ken nakadanon ti maysa a tulagan ti kappon ti aduana idi 1995. Ti Turkia ket nagtaraken iti asideg a kultural, politikal, ekonomiko ken industria a panakibiang iti Tengnga a Daya, dagiti Turko nga estado iti Tengnga nga Asia ken dagiti pagilian ti Aprika babaen ti panagkameng kadagiti organisasion a kas ti Turko a Konsilo, Nagsaupan nga Administrasion iti Turko nga Arte ken Kultura, Organisasion iti Islamiko a Panagtitinnulong ken ti Organisasion ti Ekonomiko a Panagtitinnulong.",207

Data Fields

sentence,label

labels.csv provides a mapping of some of the labels to the ISO 639-1 language code.

Data Splits

About 173000 lines of text each for train and test data.

Dataset Creation

Curation Rationale

• fix mislabels
• sentences in multiple languages, with more than 1 prominent langaue, were removed
• duplicate lines were removed

Source Data

Initial Data Collection and Normalization

Wikipedia dump from 2018

Who are the source language producers?

Wikipedia contributors

Annotations

Annotation process

[Needs More Information]

Who are the annotators?

[Needs More Information]

Personal and Sensitive Information

[Needs More Information]

Considerations for Using the Data

Social Impact of Dataset

[Needs More Information]

Discussion of Biases

[Needs More Information]

Other Known Limitations

[Needs More Information]

Additional Information

Dataset Curators

The dataset was initially created by Thomas Martin. Curated by Maxime Sobrier.

Licensing Information

ODC Open Database License v1.0

Citation Information

Curated dataset:

Citation to be added

Original dataset:

@dataset{thoma_martin_2018_841984,
  author       = {Thoma, Martin},
  title        = {{WiLI-2018 - Wikipedia Language Identification database}},
  month        = jan,
  year         = 2018,
  publisher    = {Zenodo},
  version      = {1.0.0},
  doi          = {10.5281/zenodo.841984},
  url          = {https://doi.org/10.5281/zenodo.841984}
}

Contributions

Thanks to @Shubhambindal2017 for adding this dataset.

Derived from MartinThoma/wili_2018 by Martin Thoma.