Okwu — Open African-Language Data

About this project

African-language data exists, but too much of it is unlicensed, unattributed, or unusable. Okwu is the open layer that fixes that: a validated corpus, the tooling that keeps it clean, and the governance that keeps it accountable.

A validated corpus

Every pair reviewed by a qualified speaker who confirms the source and translation match in both directions.

Open tooling

Schema, validation checks, and release process are public and run on every change — verifiable, not asserted.

Community governance

Contribution, review, and release follow published rules, guided by the CARE and FAIR principles.

Example sentences

Open sentence browser

Set	Source	English translation
yor-eng	Báwo ní	Hi
swa-eng	Jina langu	My name
pcm-eng	How you de	Hi

Known limitations

Content is everyday/beginner register — greetings, conversation, common vocabulary.
Includes short phrases and single-word vocabulary entries alongside full sentences.

Use it

The corpus is on Hugging Face, ready to load into any ML pipeline. Pick a language pair — yor-eng, swa-eng, or pcm-eng — or take all three.

from datasets import load_dataset

ds = load_dataset(
  "Okwu/african-language-parallel-corpus",
  "swa-eng",
)
# 7,126 validated Swahili–English pairs

Licensing

Data

CDLA-Permissive-2.0
Reuse and redistribute with attribution.

Code & tooling

Apache-2.0
Validation and release tooling, open to build on.

Documentation

CC BY 4.0
Schema, guides, and data cards.

Use it, cite it, or help it grow.

The corpus is live and free to use today. Contributions — corrections and new data — go through an open review process.