Open parallel-text resources for Yorùbá, Swahili and Nigerian Pidgin, together with the validation tooling and documented review process used to produce them.
Current release: v1.0 · Licence: Apache-2.0 (code) · CDLA-Permissive-2.0 (data) · CC BY 4.0 (docs) — see LICENSING.md
| Corpus | 10,363 validated pairs — Yorùbá 1,193 · Swahili 7,126 · Nigerian Pidgin 2,044 |
| Access | Hugging Face: Okwu/african-language-parallel-corpus |
| Citation | See CITATION.cff |
| Contributing | See CONTRIBUTING.md |
| Governance | See GOVERNANCE.md |
| Path | What it is |
|---|---|
schema/ |
The record schema, field definitions and visibility rules |
validation/ |
Automated checks run on every contribution |
eval/ |
Benchmark policy and held-out evaluation sets |
docs/ |
Getting started, schema reference, validation guide, limitations |
release/ |
Release runbook and manifest format |
fixtures/ |
Sample records used by tests and documentation |
Available now
- Public versioned corpus release
- Machine-readable record schema, enforced in CI
- Automated checks: schema, Unicode normalisation, duplicate detection
- Documented contribution and review process
Roadmap
- Language-identity checking
- Provenance and rights classification
- Benchmark leakage checking
- Web contribution and review interface (currently handled through issues and pull requests)
from datasets import load_dataset
ds = load_dataset("Okwu/african-language-parallel-corpus")See docs/getting-started.md for loading, filtering and citation guidance.
See docs/known-limitations.md. Read it before using this data in published work.