README.md

---
license: cc-by-4.0
task_categories:
- question-answering
- text-generation
language:
- en
- adh
- teo
- nyn
- lg
tags:
- uganda
- agriculture
- multilingual
- japadhola
- ateso
- runyankore
- luganda
- qa
- buair
pretty_name: BUAIR Uganda Multilingual Q&A
size_categories:
- 10K<n<100K
configs:
- config_name: english
  data_files: english/train-*
- config_name: japadhola
  data_files: japadhola/train-*
- config_name: ateso
  data_files: ateso/train-*
- config_name: runyankore
  data_files: runyankore/train-*
- config_name: luganda
  data_files: luganda/train-*
- config_name: all
  data_files: all/train-*
---

# BUAIR Uganda Multilingual Q&A

Parallel **question–answer** dataset for Ugandan languages, curated by the **BUAIR Voice** initiative at Busitema University.

Each language config contains the same 5,000 agriculture / rural-livelihood Q&A pairs, with English as the shared source and high-quality translations into Japadhola, Ateso, Runyankore, and Luganda.

## Languages

| Config | Language | ISO 639-3 | Pairs |
|--------|----------|-----------|------:|
| `english` | English | `en` | 5,000 |
| `japadhola` | Japadhola (Adhola) | `adh` | 5,000 |
| `ateso` | Ateso | `teo` | 5,000 |
| `runyankore` | Runyankore | `nyn` | 5,000 |
| `luganda` | Luganda | `lg` | 5,000 |
| `all` | All languages stacked | — | 25,000 |

## Dataset fields

| Column | Type | Description |
|--------|------|-------------|
| `id` | int | Pair id within a language (1–5,000) |
| `question` | string | Question text in that language |
| `answer` | string | Answer text in that language |
| `language` | string | ISO 639-3 code (`en`, `adh`, `teo`, `nyn`, `lg`) |
| `language_name` | string | Human-readable language name |

The `all` config also includes `global_id` (1–25,000) across languages.

## Loading examples

```python
from datasets import load_dataset

# One language
en = load_dataset("BUAIR/Uganda-Multilingual-QA", "english", split="train")
adh = load_dataset("BUAIR/Uganda-Multilingual-QA", "japadhola", split="train")

print(en[0]["question"])
print(adh[0]["question"])

# All languages together
all_langs = load_dataset("BUAIR/Uganda-Multilingual-QA", "all", split="train")
print(all_langs[0])
```

## Parallel alignment

Rows are **aligned by `id`** across language configs: English row `id=42` corresponds to the Japadhola / Ateso / Runyankore / Luganda translations of the same Q&A pair.

## Domain

Questions and answers focus on **Ugandan agriculture and community knowledge** (crop spacing, varieties, drought tolerance, rural practices, etc.).

## Source

Built from the BUAIR multilingual Q&A workbook (`Multiligual-QA.xlsx`), with sheets:

- Japadhola
- Ateso
- Runyankore
- Luganda

English `Question` / `Answer` columns are identical across sheets; language sheets provide the translated question and answer columns.

## Ethical use

Please credit **BUAIR Voice / Busitema University** and the language communities when using this dataset. Prefer the published language codes and do not invent speaker or translator identities that are not in the release.

## License

[Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/)

## Citation

```bibtex
@dataset{buair_uganda_multilingual_qa_2026,
  title        = {BUAIR Uganda Multilingual Q&A},
  author       = {BUAIR Voice Project, Busitema University},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/BUAIR/Uganda-Multilingual-QA}},
  note         = {Parallel Q&A in English, Japadhola, Ateso, Runyankore, and Luganda.}
}
```

## Contact

- **Project:** BUAIR Voice — Busitema University AI Research
- **Hub:** [BUAIR](https://huggingface.co/BUAIR)
- **Dataset card updated:** 2026-08-06