This dataset is shared under CC BY-NC 4.0, which means you are free to use, share, and adapt it for non-commercial research and educational purposes with attribution. You can read the full license at https://creativecommons.org/licenses/by-nc/4.0/.

Ghana-QA: Multilingual Ghanaian Question-Answer Pairs Dataset

Dataset Summary

Ghana-QA is a large-scale multilingual question-answering dataset derived from Ghanaian news articles. The dataset covers English and three major Ghanaian languages: Twi (Akan), Ewe, and Ga.

The dataset is organized into dedicated subsets (configurations) so you can easily load individual languages or the full dataset.

• Total QA Pairs: 5,125,796
• Subsets / Configurations: english, twi, ewe, ga, and all (default)
• Source: Ghanaian news articles (Citi Newsroom, MyJoyOnline, etc.)
• Project: Part of the GhanaNLP initiative

Available Subsets

Subset Name	Language	Code (lang)	Number of QA Pairs	Description
english	English	eng	1,578,136	QA pairs in English from Ghanaian news
twi	Twi (Akan)	twi	1,182,530	Parallel QA pairs translated into Twi
ewe	Ewe	ewe	1,182,565	Parallel QA pairs translated into Ewe
ga	Ga	ga	1,182,565	Parallel QA pairs translated into Ga
all (default)	Multilingual	—	5,125,796	All 4 language subsets combined

How to Use

Load individual language subsets or the entire dataset using the Hugging Face datasets library:

1. Load Specific Language Subsets

from datasets import load_dataset

# English subset
english_qa = load_dataset("ghananlpcommunity/ghana-qa", "english")
print(f"English samples: {len(english_qa['train'])}")
print(english_qa['train'][0])

# Twi subset
twi_qa = load_dataset("ghananlpcommunity/ghana-qa", "twi")
print(f"Twi samples: {len(twi_qa['train'])}")
print(twi_qa['train'][0])

# Ewe subset
ewe_qa = load_dataset("ghananlpcommunity/ghana-qa", "ewe")
print(f"Ewe samples: {len(ewe_qa['train'])}")
print(ewe_qa['train'][0])

# Ga subset
ga_qa = load_dataset("ghananlpcommunity/ghana-qa", "ga")
print(f"Ga samples: {len(ga_qa['train'])}")
print(ga_qa['train'][0])

2. Load the Combined Dataset (Default)

from datasets import load_dataset

# Loads all languages combined
dataset = load_dataset("ghananlpcommunity/ghana-qa")
print(f"Total samples: {len(dataset['train'])}")

3. Streaming Mode (No Full Download Required)

from datasets import load_dataset

# Stream examples on the fly
dataset = load_dataset("ghananlpcommunity/ghana-qa", "english", streaming=True)
for sample in dataset['train'].take(5):
    print("Q:", sample['question'])
    print("A:", sample['answer'])

Data Fields

Each example contains the following fields:

• question (string): The question text in the respective language.
• answer (string): The concise, factual answer to the question in the respective language.
• lang (string): Language identifier code (eng, twi, ewe, or ga).

Sample Data Instances

English (english):

{
  "question": "What happens if a political party member is suspended?",
  "answer": "Suspended party members lose voting rights until they complete reinstatement procedures. Amnesty may apply, but membership forfeiture resets status.",
  "lang": "eng"
}

Twi (twi):

{
  "question": "Sɛ wogyae amammui kuw no muni bi a, dɛn na ɛba?",
  "answer": "Amannɔne a wɔagyae wɔn adwuma no hwere hokwan a wɔwɔ sɛ wɔtow aba no kosi sɛ wobewie nhyehyɛe a wɔde bɛsan akɔ adwuma mu no. Amnesty betumi abisa, nanso asɔremma a wɔfa wɔn ho wɔn ho no resets tebea.",
  "lang": "twi"
}

Ewe (ewe):

{
  "question": "Nukae adzɔ ne wodzudzɔ dunyaheha me tɔ aɖe ƒe dɔwɔwɔ?",
  "answer": "Akpa siwo wodzudzɔ la bu akɔdada ƒe gomenɔamesiwo vaseɖe esime wowu ɖoɖo siwo wowɔ be woagbugbɔ wo axɔ ɖe dɔa me nu. Atsɔtsɔtsɔke ate ŋu awɔ dɔ, gake hamevinyenye ƒe nusiwo woxɔ le ame si ƒe ɖoƒea gbugbɔa ɖoƒea.",
  "lang": "ewe"
}

Ga (ga):

{
  "question": "Mɛni baa kɛji akɛ ajie maŋkwramɔŋ kuu lɛ mlinyo ko?",
  "answer": "Kuu lɛ mli bii ni agbeɔ naa lɛ laajeɔ hegbɛi ni akɛwoɔ oshikifɔɔ mli lɛ kɛyashi be mli ni amɛgbeɔ gbɛjianɔtoo ni akɛbaaku sɛɛ aba lɛ naa. Amnesty baanyɛ abi, shi kuu lɛ mli bii ni ajie amɛ lɛ haa shihilɛ lɛ yaa hiɛ ekoŋŋ.",
  "lang": "ga"
}

Dataset Creation

1. Source Articles: Curated Ghanaian news articles covering politics, economy, health, education, sports, and culture from Ghanaian news portals.
2. Q&A Generation: High-quality QA pairs generated using LLM prompts designed to produce mutually exclusive queries and concise, accurate answers grounded strictly in the news text.
3. Translation: Parallel translations into Twi (Akan), Ewe, and Ga produced with high-quality translation pipelines tailored for Ghanaian languages.
4. Validation & Deduplication: Rigorous validation, removal of incomplete/malformed pairs, cleaning of formatting artifacts, and deduplication.

Citation Information

If you use this dataset in your research or application, please cite:

@dataset{Ghana-QA_2025,
  title = {Ghana-QA: Multilingual Ghanaian Question-Answer Pairs},
  author = {Mich-Seth Owusu and Priscilla Lartey and Gerhardt Datsomor},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/ghananlpcommunity/ghana-qa}
}

License

This dataset is released under CC BY-NC 4.0 — free to use, share, and adapt for non-commercial research and educational purposes with attribution.

Acknowledgments

Special thanks to the primary contributors of the Ghana-QA project: Mich-Seth Owusu, Priscilla Lartey, Gerhardt Datsomor, and all volunteers from the GhanaNLP community. This work is part of the GhanaNLP initiative to advance natural language processing for Ghanaian and African languages.