mbaye930/wolof-arabic-parallel-corpus

MudawanSn: A Gold-Standard Wolof--Arabic Parallel Corpus for Machine Translation

A publicly available parallel corpus for the Wolof–Arabic language pair, a gold-standard resource containing 1,271 sentence-aligned pairs. The corpus consists of manual translations from Wolof into Modern Standard Arabic (MSA). The source texts are drawn from the MasakhaNER corpus, covering politics, society, religion, and sports in Senegalese news discourse.

Dataset Structure

The dataset contains three splits:

• —`train`: 1,080 sentence pairs (from train.tsv)
• —`validation`: 95 sentence pairs (from dev.tsv)
• —`test`: 96 sentence pairs (from test.tsv)

Data Fields

• —id: Unique identifier for the parallel sentence pair.
• —wolof: The sentence in Wolof.
• —arabic: The manual translation in Modern Standard Arabic.
• —french_source: The original French news text from which the Wolof source sentence was translated.
• —source_dataset: The original source dataset.
• —source_type: Category of the text source.

How to Load the Dataset

You can easily load this dataset using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("mbaye930/wolof-arabic-parallel-corpus")

# Access the splits
train_data = dataset["train"]
val_data = dataset["validation"]
test_data = dataset["test"]

# Print a sample
print(train_data[0])

Dataset Summary & Quality Control

• —Translation Process: The parallel corpus was translated manually from Wolof to Arabic. Crucially, the translations were produced directly from the Wolof source texts by a native Wolof speaker and Arabic scholar, avoiding pivot-language translation errors.
• —Alignment & Verification: Cross-lingual semantic verification was conducted using a LASER3 semi-automated loop, verifying that the cosine similarity scores of the pairs were high and filtering out misaligned or erroneous source texts. The average LASER3 cosine similarity is 0.75, indicating high semantic alignment.

Licensing

The dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

Citation

If you use this corpus or the baseline code in your research, please cite our paper:

@inproceedings{mbaye&diop2026gold,
  title     = {MudawanSn: A Gold-Standard Wolof--Arabic Parallel Corpus for Machine Translation},
  author    = {Mbaye, Mouhamed and Diop, Thierno },
  year      = {2026}
}