honourjesus/hausa-pq-speech-validated

Hausa WAEC PQ Speech Dataset (Validated)

Dataset Description

This dataset contains 1,361 multiple-choice WAEC past questions translated from English into Hausa, with human-validated Hausa translations. The data is designed to support speech synthesis, machine translation evaluation, and low-resource NLP research for Hausa — one of the most widely spoken languages in West Africa.

Languages

• Source: English (en)
• Target: Hausa (ha)

---

Dataset Structure

Data Splits

Split	Size
train	1,361

Note: A single train split is provided. Users are encouraged to create their own train/validation/test splits for downstream tasks.

Data Fields

Field	Type	Description
question_english	string	Original multiple-choice question in English (A/B/C/D format)
question_hausa_raw	string	Initial machine/human translation of the question into Hausa
question_hausa_with_options	string	Hausa question with options explicitly prefixed with "Option A/B/C/D"
question_hausa_validated	string (nullable)	Expert-validated Hausa translation with "Option" labels — available for 899 of 1,361 rows
answer_english	string (nullable)	Correct answer label in English (e.g. B. four)
answer_hausa	string (nullable)	Correct answer label in Hausa (e.g. Option B. huɗu)

Validation Coverage

Status	Count
Expert-validated Hausa question	899 (66%)
Not yet validated (raw+options available)	462 (34%)

---

Dataset Creation

Motivation

WAEC Examination in Nigerian secondary schools is predominantly conducted in English, creating barriers for learners whose dominant language is Hausa. This dataset was created to:

1. Enable text-to-speech (TTS) and automatic speech recognition (ASR) research in Hausa
2. Support machine translation evaluation (English → Hausa) using chrF, BLEU, and AfricoMET metrics
3. Provide a low-resource NLP benchmark for Hausa agricultural terminology

Source Data

Questions were sourced from WAEC past question papers (PQ), originally in English. Translations were produced and then validated by Hausa-language experts.

Annotation Process

1. Raw translation: Initial translation of English questions and answer options into Hausa
2. Option formatting: Answer options reformatted with explicit "Option A/B/C/D" prefixes for TTS clarity
3. Expert validation: Native Hausa-speaking domain experts reviewed and corrected translations, particularly for agricultural and scientific terminology

Known Issues / Limitations

• 462 rows (~34%) do not yet have expert-validated translations; question_hausa_validated is null for these rows
• 51 rows have missing answer_english; 1 row has missing answer_hausa
• 2 rows have missing question_english
• Hausa text uses standard orthographic conventions including ɗ, ƙ, ɓ, ƴ

---

Uses

Direct Use

• Speech synthesis (TTS): The validated Hausa questions can directly be used as utterances for building Hausa TTS corpora
• ASR training data: Paired English–Hausa text for cross-lingual acoustic model training
• MT Model Finetuning: Paired English–Hausa text for cross-lingual model fine-tuning
• MT evaluation: Use question_english → question_hausa_validated pairs for evaluating English-to-Hausa MT systems with chrF, BLEU, and AfricoMET

Out-of-Scope Use

This dataset should not be used for general-domain Hausa NLP without acknowledging its educational domain specificity.

---

Acknowledgements

Special thanks to the Hausa language validation team Tonative for their careful review of the translated questions and answer options. Their native Language expertise and knowledge of scientific and educational terminology were invaluable to this work.

---

Citation

If you use this dataset, please cite:

@dataset{hausa_pq_speech_2024,
  title        = {Hausa WAEC PQ Speech Dataset (Validated)},
  author       = {[Honour-Jesus Bezaleel]},
  year         = {2026},
  publisher    = {HuggingFace},
  url          = {https://huggingface.co/datasets/honourjesus/hausa-pq-speech-validated},
  license      = {CC-BY-4.0}
}

---

License

Creative Commons Attribution 4.0 International (CC-BY-4.0)