WAZOBIALABS/nigerian-pidgin-eval

Wazobia Labs — Nigerian Pidgin Evaluation Set

Status: In development — target release Q3 2026 Builder: Wazobia Labs License: CC-BY-4.0

---

What This Is

Every AI lab building for Nigerian Pidgin has the same unsolved problem: they cannot evaluate whether their model actually works.

There is no gold-standard benchmark. No culturally verified test set. No sarcasm corpus. No benchmark that captures the difference between forming and contempt, between hustle_fatigue and hustle_energy, between a sincere compliment and its sarcastic twin.

This evaluation set solves that.

---

What Will Be In This Dataset

The Nigerian Pidgin Evaluation Set is a separate gold-standard corpus drawn from and extending the main WAZOBIALABS/nigerian-pidgin-voice-text dataset.

Target Specifications

Specification	Target
Total entries	200 gold-standard entries
Emotion categories	Minimum 20 entries per category across all 13 original categories
Sarcasm pairs	40 complete pairs (sincere + sarcastic twin)
Female speaker representation	Minimum 50%
Health domain	Minimum 20 entries
Inter-annotator agreement	Verified by second native speaker annotator
Annotation quality	Every entry reviewed, not sampled

What Makes This Gold-Standard

Sarcasm pairs with sincere twins. The exact same Pidgin phrase annotated twice — once sincere, once sarcastic. AI models must distinguish them using context and cultural register, not just vocabulary.

Nigerian emotion taxonomy. Not positive/negative/neutral. The full 13-category taxonomy: forming, hustle_fatigue, market_energy, prayer_gratitude, betrayal, craving, and more — categories that cannot be built without lived Nigerian cultural knowledge.

Tonal disambiguation entries. Same phrase, different context, different meaning. "E don do" consoling someone crying vs commanding someone to stop fighting vs cutting off excessive talking. Your model must handle all three.

Female voice representation. Minimum 50% female-voiced entries — a hard architectural constraint, not a recommendation.

Health domain coverage. Maternal health, menstrual health, mental health, caregiving — registers completely absent from formal NLP corpora.

---

Why Evaluation Infrastructure Matters

When Meta, Google, Microsoft, and African AI labs build Nigerian Pidgin models, they face a fundamental problem: what does "works" mean?

WAXAL and similar initiatives have built ASR corpora. They can transcribe. But can they understand?

Can they tell that "You come early today oh" is a sincere compliment when said to someone who arrived unexpectedly on time — and a devastating sarcastic insult when said to someone two hours late?

Can they tell that "I dey mind my business" is contempt, not neutrality?

Can they score hustle_fatigue vs hustle_energy — two speakers both talking about the grind, one depleted and one fired up?

Without an evaluation set that tests these distinctions, there is no way to know. Wazobia Labs builds the evaluation set. You cannot skip the evaluation.

---

Pricing and Access

The evaluation set will be released under CC-BY-4.0 with a standard tier for academic and research use.

Enterprise licensing for production deployment and model benchmarking will be available separately with support, update guarantees, and documentation.

For early access and enterprise licensing enquiries: wazobialabs@gmail.com

---

Timeline

Milestone	Target
Alpha structure established	May 2026 ✅
100 gold-standard entries	June 2026
Inter-annotator review complete	July 2026
200 gold-standard entries	August 2026
Public release v1.0	Q3 2026

---

Foundation Dataset

This evaluation set is built on and extends:

WAZOBIALABS/nigerian-pidgin-voice-text v0.6 — 480 entries — 16 emotion categories — CC-BY-4.0

---

Citation

@dataset{wazobia_labs_pidgin_eval_2026,
  author    = {Okoye, Stephanie},
  title     = {Wazobia Labs Nigerian Pidgin Evaluation Set},
  year      = {2026},
  version   = {0.1.0-alpha},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/WAZOBIALABS/nigerian-pidgin-eval},
  license   = {CC-BY-4.0},
  note      = {In development — target release Q3 2026}
}

---

About Wazobia Labs

Wazobia Labs builds African language AI infrastructure that doesn't exist but should. We identify the specific, high-value gaps in African language data that existing datasets leave open — then build exactly those gaps with commercial licensing, production-grade quality, and the cultural specificity that real AI products need.

Contact: wazobialabs@gmail.com Hugging Face: WAZOBIALABS Founded: Lagos, Nigeria — 2025

Wa. Zo. Bia. The evaluation is coming.