Native-speaker eval data for N-ATLAS in Yoruba, Hausa, Igbo, and Pidgin
Hi Awarri / N-ATLAS team,
N-ATLAS launching at UNGA80 as an open, multilingual, multimodal model for Yoruba, Hausa, Igbo, and Nigerian
Pidgin, fine-tuned on hundreds of millions of tokens of instruction data with language-specific speech
models, and built explicitly to map and digitize Nigeria's linguistic heritage while building datasets, is
one of the most mission-aligned programs we have seen. The dataset-building part is precisely where we can
add a layer: native-speaker judgment of fluency, preference, and cultural correctness in Yoruba, Hausa,
Igbo, and Pidgin is the single hardest tier of human evaluation to source anywhere in the world.
That is what Boltwork runs. We are a human-data network of native speakers in long-tail languages, with a
live African-language panel, and quality built into the pipeline: hidden gold checks, multi-judge consensus
across distinct workers, per-worker trust scores, and a written rationale plus an error-category tag and
provenance on every single judgment, so the labels are auditable and reusable, not a bare score. Clean
JSONL out.
Because your mission is dataset-building, the natural first step is a small, free proof: send 100 to 300 of
your own items (preference pairs, output ratings, or fluency judgments) in any of your languages, and we
return labeled JSONL in about a week with full QC stats. On our pilot set we run ~92% gold accuracy and
~95% inter-annotator agreement. No cost, no commitment. The cleanest read is on whichever language your
current data is thinnest in.
One operational detail that lets us reach and retain native speakers across Nigeria: we pay them instantly
in Bitcoin over Lightning, which sidesteps the payout and KYC friction that usually breaks the unit
economics for local-language work.
Methodology and deliverable format: https://lightningfaucet.com/boltwork/data/. Reply and I will send a
redacted sample record.
Would your data team be open to a short conversation?
Paul Ferguson, Boltwork