Hugging Face's logo

Datasets:
goaicorp
/
dyu-fr-parallel

Tasks:
Translation
Modalities:
Text
Formats:
parquet
Languages:
Dyula
French
Size:
10K - 100K
Tags:
dioula
dyula
burkina-faso
cote-d-ivoire
low-resource
parallel-corpus
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

Request access to dyu-fr-parallel

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

dyu-fr-parallel is shared under CC BY-SA 4.0. The data comes from the Koumankan4Dyula corpus by UVCI and Data354 (AI4D Africa); please credit them. Please read the card, its terms and licenses before requesting access, and make sure you understand them.

Log in or Sign Up to review the conditions and access this dataset content.

dyu-fr-parallel

Dioula-French parallel text for machine translation.
Language: Dioula and French 10,884 pairs License: CC BY-SA 4.0 GO AI Corp

dyu-fr-parallel holds the Dioula-French pairs GO AI Corp uses to train its Dioula translation models. For now they all come from one public corpus, Koumankan4Dyula; each new source will get its own subset. For speech, see dyu-speech.

Access is gated and requests are reviewed by hand. Collecting and cleaning Dioula parallel text takes a lot of time and resources, and we do it mostly out of passion. Please read this card, its terms and licenses before requesting access, and make sure you understand them. Respecting them keeps this work useful to the community and helps move our local languages forward. The license is CC BY-SA 4.0: no further terms are added.

Résumé en français. Paires dioula-français que GO AI Corp utilise pour entraîner ses modèles de traduction du dioula. Pour l'instant, elles viennent toutes d'un corpus public, Koumankan4Dyula (UVCI et Data354, AI4D Africa), sans doublons ni phrases de test FLORES+ ; chaque nouvelle source aura son propre sous-ensemble. Pour la parole, voir dyu-speech. Accès sur demande, examinée à la main : lisez cette carte, ses conditions et ses licences avant de demander l'accès. Licence CC BY-SA 4.0, sans conditions supplémentaires ; citez UVCI et Data354 et partagez vos adaptations sous la même licence.

Overview

Subset Content Source License Pairs
koumankan Common Voice French sentences and their Dioula translations Koumankan4Dyula, UVCI and Data354 CC BY-SA 4.0 10,884

default loads every subset, and every pair appears once.

Quick start 

from datasets import load_dataset

ds = load_dataset("goaicorp/dyu-fr-parallel", split="train")

Log in first (hf auth login) with an account that has access.

Columns

Column Description
id Stable id (hash of the normalized pair)
dyu, fr Dioula and French text
source Subset the pair belongs to
koumankan_ids Koumankan row ids of this pair (several when the same pair was recorded more than once)
koumankan_splits Koumankan split(s) the pair came from: train, dev, test

Sources and credits

All rows come from Koumankan4Dyula by the Université Virtuelle de Côte d'Ivoire (UVCI) and Data354, built in the Koumankan project under the AI4D Africa programme (IDRC, SIDA, ACTS) and released under CC BY-SA 4.0: uvci/Koumankan_mt_dyu_fr (text) and uvci/koumankan4dyula (speech). The French sentences come from Common Voice (CC0); a linguist translated them into Dioula. Source contacts: ismael21.kone@uvci.edu.ci, info@data354.com.

How it was built

  1. All of Koumankan (train, dev and test, 10,929 rows) is training data. Koumankan's splits share text: 809 train rows have the same French or Dioula as a dev or test row, because several translators translated the same short Common Voice sentence. We do not evaluate translation on it.
  2. Test French restored. The text release hides the French of its 1,393 test rows (written "0"); the speech release has it. Rows were joined by Common Voice id; train and dev French agree 100% between the two releases.
  3. FLORES+ removed. GO AI's earlier copy (goai-dioula-translation) had the 1,012 FLORES+ devtest sentences appended to train. We dropped them, and no remaining row repeats or closely paraphrases a FLORES+ sentence in Dioula, French or English.
  4. Duplicates merged: 45 exact repeats, ids kept in koumankan_ids.

Evaluation

Evaluate on FLORES+ dyu_Latn (dev, devtest); none of its sentences is in this dataset. Koumankan dev and test are training data here, so do not use them to evaluate a model trained on this dataset.

Limits

  • One source only: short Common Voice sentences (read-speech prompts), mostly everyday and general topics.
  • Dioula spelling varies between translators; no normalization was applied.
  • Many French sentences have several Dioula translations. This helps training, but the corpus has fewer distinct French sentences than pairs.

Access

Fill in the form on this page. Requests are reviewed by hand.

License

CC BY-SA 4.0, as the source. Credit UVCI and Data354 (Koumankan4Dyula, AI4D Africa) and GO AI Corp, and share adaptations under the same license.

Citation 

@misc{goaicorp2026dioulafrenchparallel,
  title        = {dyu-fr-parallel: Dioula-French parallel text for machine translation},
  author       = {Nyantudre, Wendyellé Abubakrh Alban and Bandaogo, Aristide},
  year         = {2026},
  howpublished = {Hugging Face Datasets (gated access), GO AI Corp},
  url          = {https://huggingface.co/datasets/goaicorp/dyu-fr-parallel}
}

@misc{uvci2024koumankan4dyula,
  title        = {Koumankan4Dyula: A Speech-To-Text Translation Corpus For Dyula Language},
  author       = {{Université Virtuelle de Côte d'Ivoire} and {Data354}},
  howpublished = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/uvci/koumankan4dyula}
}

Contacts

Downloads last month
47
Total file size:
888 kB

Models trained or fine-tuned on goaicorp/dyu-fr-parallel