---
language:
- en
- sw
- luy
multilinguality:
- multilingual
task_categories:
- translation
task_ids:
- translation
tags:
- luhya
- african-languages
- low-resource
- multilingual-translation
- kenya
size_categories:
- 10K<n<100K
source_datasets:
- original
license: cc-by-4.0
---

# Luhya Multilingual Translation Dataset

## Dataset Description

This dataset contains 26,205 translation pairs for Luhya dialects, English, and Swahili. It was created to support machine translation research for Luhya, a Bantu language family spoken in Kenya.

### Languages and Dialects

**Languages:**
- en
- luy
- sw

**Luhya Dialects:**
- Bukusu
- Luwanga
- Marachi
- Maragoli
- Tsotso

### Dataset Statistics

| Split | Examples |
|-------|----------|
| Train | 20,963 |
| Validation | 2,621 |
| Test | 2,621 |
| **Total** | **26,205** |

### Data Sources

- Bible
- Dictionary
- Proverbs
- Translations

## Dataset Structure

### Data Fields

- `source_text`: Source text for translation
- `target_text`: Target translation (includes dialect tokens)
- `source_lang`: Source language code
- `target_lang`: Target language code  
- `dialect`: Specific Luhya dialect (if applicable)
- `domain`: Source domain (bible, dictionary, proverbs, etc.)

### Example

```json
{
  "source_text": "Hello, how are you?",
  "target_text": "<luy_bukusu> Murachi mulahi?",
  "source_lang": "en",
  "target_lang": "luy", 
  "dialect": "Bukusu",
  "domain": "greetings"
}
```

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("mamakobe/luhya-multilingual-dataset")

# Access splits
train_data = dataset['train']
val_data = dataset['validation'] 
test_data = dataset['test']
```

## Ethical Considerations

### Community Involvement
- Dataset created in collaboration with Luhya language communities
- Sources include traditional texts and community contributions
- Respects cultural and linguistic heritage

### Intended Use
- Machine translation research
- Language preservation efforts
- Educational applications
- Cultural documentation

### Limitations
- May not represent all Luhya dialects equally
- Translation quality varies by domain
- Some entries may contain cultural context not captured in direct translation

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{luhya_multilingual_2024,
  title={Luhya Multilingual Translation Dataset},
  author={mamakobe},
  year={2024},
  url={https://huggingface.co/datasets/mamakobe/luhya-multilingual-dataset}
}
```

## License

This dataset is released under CC-BY-4.0 license. Please respect the cultural heritage of the Luhya communities when using this data.

---

Dataset created: 2025-08-30