README.md

---
dataset_info:
  features:
  - name: translation
    struct:
    - name: ede
      dtype: string
    - name: vi
      dtype: string
  splits:
  - name: train
    num_bytes: 2441249.3201497775
    num_examples: 15092
  - name: validation
    num_bytes: 161757.83992511116
    num_examples: 1000
  - name: test
    num_bytes: 161757.83992511116
    num_examples: 1000
  download_size: 1628478
  dataset_size: 2764765.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: cc-by-nc-4.0
task_categories:
- translation
language:
- vi
size_categories:
- 10K<n<100K
---

# Rhade-Vietnamese Machine Translation Dataset

## Overview

This dataset provides a parallel corpus for Rhade (Ede) to Vietnamese machine translation. Rhade (also written *Ê Đê* or *Ede*) is an Austronesian language spoken primarily by the Ede ethnic minority in the Central Highlands of Vietnam. It is a low-resource language with very limited NLP resources, making this dataset a meaningful contribution to the multilingual NLP community.

The corpus consists of **17,097 sentence pairs** spanning religious, cultural, and general-domain text, curated and aligned from multiple publicly available Vietnamese-Ede bilingual sources, including independently collected Rhade data from HDOK ([@YPHIC](https://huggingface.co/YPHIC)).

---

## Dataset Statistics

| Attribute           | Value                                                     |
|---------------------|-----------------------------------------------------------|
| Language Pair       | Rhade (Ede) to Vietnamese (`ede`-`vi`)                   |
| Total Pairs         | 17,097                                                    |
| Splits              | `train` (17,097 rows)                                     |
| Format              | Parquet (auto-converted); JSON dict                       |
| Size (raw)          | 1.63 MB                                                   |
| Size (Parquet)      | 1.63 MB                                                   |
| Domain              | Religious text, general prose                             |
| Script              | Latin-based orthography (Ede), Vietnamese with diacritics |

---

## Data Format

Each record is a dictionary with two keys:

```json
{
  "translation": {
    "ede": "Bhung kahan ai tiê kriăng snăk.",
    "vi": "Các chiến sĩ rất chăm chỉ."
  }
}
```

| Field               | Description                    |
|---------------------|--------------------------------|
| `translation.ede`   | Source sentence in Rhade (Ede) |
| `translation.vi`    | Target sentence in Vietnamese  |

---

## Data Sources

The corpus was collected and aligned from the following publicly accessible sources:

| Source | Description | URL |
|--------|-------------|-----|
| VOV4 - Voice of Vietnam (Ede) | Vietnamese national broadcaster content in Ede language | https://vov4.vov.vn/ede |
| Revised Vietnamese Version Bible (RVV11) | Bilingual Bible corpus (Ede-Vietnamese), RVV 2011 edition | https://kinhthanh.httlvn.org/?v=RVV11 |
| HDOK / YPHIC (Hugging Face) | Independently collected Rhade text data and model resources | https://huggingface.co/YPHIC |

> All data has been collected in accordance with the publicly accessible nature of the source materials. Users are responsible for ensuring compliance with the terms of use of the original sources for their intended applications.

---

## Usage

### Loading with Hugging Face Datasets

```python
from datasets import load_dataset

dataset = load_dataset("NIRVLab/rhade-vietnamese-mt")
print(dataset["train"][0])
# Output:
# {'translation': {'ede': 'Bhung kahan ai tiê kriăng snăk.', 'vi': 'Các chiến sĩ rất chăm chỉ.'}}
```

### Fine-tuning a Translation Model

```python
from datasets import load_dataset
from transformers import MarianMTModel, MarianTokenizer

dataset = load_dataset("NIRVLab/rhade-vietnamese-mt")

for example in dataset["train"]:
    src = example["translation"]["ede"]
    tgt = example["translation"]["vi"]
    # ... your training pipeline here
```

---

## Intended Uses and Limitations

### Intended Uses

- Low-resource machine translation research (Ede to/from Vietnamese)
- Multilingual and cross-lingual NLP benchmarking
- Transfer learning and multilingual model evaluation
- Linguistic documentation and computational analysis of Rhade

### Limitations

- The corpus is domain-skewed toward religious and broadcast journalism text, which may limit generalization to other domains.
- Rhade is a low-resource language with limited pre-existing NLP tools; tokenization and morphological analysis may require custom solutions.

---

## License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. It is made available strictly for non-commercial purposes, including academic research, education, and linguistic study. Any commercial use, redistribution for profit, or incorporation into commercial products or services is explicitly prohibited. If you use this dataset, you must give appropriate credit to the original author and may not apply legal terms or technological measures that restrict others from using it under the same terms.

---

## Citation

If you use this dataset in your research, please cite the following:

```bibtex
@dataset{tran2025rhade,
  author    = {Tran, Nhan D.},
  title     = {{Rhade--Vietnamese Machine Translation Dataset}},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/NIRVLab/rhade-vietnamese-mt},
  note      = {A parallel corpus for low-resource Rhade (Ede)--Vietnamese
               machine translation, collected from VOV4 (Voice of Vietnam),
               the Revised Vietnamese Version Bible (RVV11), and Rhade text
               data sourced from HDOK (https://huggingface.co/YPHIC).},
}
```

This dataset incorporates Rhade text data originally collected by HDOK. If you make direct use of those resources, please also cite their work:

```bibtex
@misc{hdok2025rhade,
  author       = {HDOK},
  title        = {{Rhade AI API and rhade\_ttt\_model\_v01}},
  year         = {2025},
  howpublished = {Hugging Face},
  url          = {https://huggingface.co/YPHIC},
  note         = {Rhade language model and translation API for low-resource NLP,
                  available at https://huggingface.co/spaces/YPHIC/rhade-ai-api.},
}
```

> Please also consider citing the original data sources (VOV4 and HTTLVN) if you use content derived from those corpora.

---

## Author and Affiliation

**Nhan D. Tran**
Network for Intelligent Research Vietnam (NIRVLab)

For questions, issues, or contributions, please open a discussion on the Hugging Face dataset page at https://huggingface.co/datasets/NIRVLab/rhade-vietnamese-mt/discussions.

---

## Contributing

Contributions to expand the corpus — including additional domains, improved sentence alignment, or dedicated test sets — are warmly welcomed. Please open a discussion on the dataset page or contact the author directly.

---

## Acknowledgements

We thank the Voice of Vietnam (VOV4) and the Evangelical Church of Vietnam (HTTLVN) for making bilingual Ede-Vietnamese content publicly available. We also thank HDOK ([@YPHIC](https://huggingface.co/YPHIC), [yphichdok](https://github.com/yphichdok)) for independently collecting and openly sharing Rhade language data and model resources, part of which was incorporated into this corpus. This work contributes to the preservation and computational study of endangered and minority languages in Vietnam.