ghananlpcommunity/ghana-farmer-qa-ga

Ghana Farmer Q&A -- English / Ga

151,427 parallel English-Ga question-answer pairs on Ghanaian smallholder farming, grounded in 15,174 recorded interviews with Ghanaian farmers.

This is the Ga edition of `ghananlpcommunity/ghana-farmer-qa`: the same farmer-grounded Q&A pairs, with each question and answer aligned to a Ga translation. Every row holds the English and the Ga side by side, plus the interview passage the pair came from.

Where the content comes from

1. 1.Field recordings. Ghanaian farmers were recorded discussing their practice in Twi -- cocoa spacing, pest pressure, soil fertility, harvest and pricing. These were transcribed and translated into English, published as `ghananlpcommunity/twi-english-agric`.
2. 2.Question generation. For each of the 15,174 interview passages, a model (Gemini 3.6 Flash) was asked to write ~10 standalone questions a farmer might put to an extension officer about the practices described in that passage, plus an explanatory answer for each. Each question had to stand on its own -- no follow-ups, no references to "the text above" -- so the pairs are usable outside their original context.
3. 3.Flattening. One row per Q&A pair, carrying its source passage forward.
4. 4.Translation. The English question and answer were machine-translated into Ga (Google Translate, gaa), with each output checked as non-empty and distinct from its source and retried on failure. Rows with untranslated residue were dropped, leaving 151,427 of 151,469.

Columns

Common topics include pest and disease control, soil and nutrient management, planting and spacing, crop growth and behaviour, root and tuber crops, and vegetable farming.

What it's useful for

Ga is a low-resource language with little parallel text available, and almost none in a practical domain like agriculture. That makes this useful for:

• —Training and evaluating English<->Ga translation, especially in-domain agricultural and advisory language that general corpora don't cover.
• —Building Ga-language farmer advisory tools -- chat assistants, IVR and voice systems, extension material -- where the question register matters as much as the content.
• —Multilingual QA: the same 151,427 questions exist in English, Ewe, Ga and Twi, so you can compare model behaviour across languages on identical content.
• —Terminology mining for agricultural vocabulary in Ga -- crop names, pests, practices -- which is thin in existing resources.
• —A post-editing base. Machine translation gets you coverage cheaply; human review of this text is the fastest route to a gold-standard Ga corpus. The English source sits in the same row to make review straightforward.

Limitations

• —The Ga text is unreviewed machine translation. Google Translate support for Ga is recent and uneven, so expect errors, and treat the Ga side as a starting point rather than a reference. English was translated via a Thai pivot, which adds its own loss of fidelity.
• —The English answers are model-written and not expert-reviewed. They are grounded in farmer interviews, but no agronomist verified them. Do not deploy as extension advice without review.
• —`category` is free text, with ~24,000 near-duplicate variants. Cluster or remap before using it as a label.
• —Interview coverage shapes topic coverage -- this is not a balanced survey of Ghanaian agriculture.

License

CC BY-NC 4.0, inherited from the source dataset `ghananlpcommunity/twi-english-agric`.