omanyasa/bantu-lang-id

English-Shona Language Identification Dataset

A multilingual dataset for language identification supporting English, Shona, Ndebele, Tonga, and Chewa languages.

📖 Documentation

Comprehensive documentation is available in the docs/ folder:

• Dataset Card - Complete dataset description and statistics
• Usage Examples - Code examples and tutorials
• Contributing Guide - Guidelines for contributing

🚀 Quick Start

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("omanyasa/english-shona-langid")

# Access data
train_data = dataset['train']
print(f"Training samples: {len(train_data)}")
print(f"Sample: {train_data[0]['text'][:50]}...")
print(f"Language: {train_data[0]['label']}")

🌍 Supported Languages

Language	Code	Version
English	en	v1, v2
Shona	sn	v1, v2
Ndebele	nd	v2
Tonga	toi	v2
Chewa	ny	v2

📊 Dataset Versions

• v1 (main branch): English and Shona only
• v2 (v2-dataset-clean branch): All 5 languages

🤝 Contributing

We welcome contributions! See the Contributing Guide for details.

� MLOps Integration

This dataset includes MLOps-ready components:

CI/CD Pipeline

• Automated Testing: GitHub Actions for data quality validation
• Benchmarking: Baseline model performance tracking
• Continuous Integration: Automated dataset validation on changes

Reproducibility

• Environment Management: environment.yml for conda environments
• Data Validation: Comprehensive test suite in tests/
• Version Control: Git LFS for large files, semantic versioning

Monitoring & Quality

• Data Quality Checks: Automated validation for text integrity
• Performance Benchmarks: Baseline model tracking
• Schema Validation: Feature and format consistency checks

Quick Setup for MLOps

# Clone with MLOps setup
git clone https://huggingface.co/datasets/omanyasa/english-shona-langid
cd english-shona-langid

# Set up environment
conda env create -f environment.yml
conda activate english-shona-langid

# Run tests
pytest tests/

# Run benchmarks
python scripts/benchmark.py

📊 Performance Benchmarks

Latest baseline performance (v2 dataset):

• Logistic Regression: 85.2% accuracy
• DistilBERT: 92.1% accuracy
• XGBoost: 89.7% accuracy

See benchmarks/ for detailed results and methodology

�📄 Citation

@dataset{english_shona_langid,
  title={English-Shona Language Identification Dataset},
  author={Omanyasa, [Your Name]},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/omanyasa/english-shona-langid}}
}

📜 License

This dataset is released under the Creative Commons Attribution 4.0 International License. You are free to share, adapt, and use the dataset for any purpose, provided you give appropriate credit.

Dataset Structure

data/
├── shona_en_lang_train.txt    # Training set (135 MB)
├── shona_en_lang_valid.txt    # Validation set (16.9 MB)
└── shona_en_lang_test.txt     # Test set (16.9 MB)

Data Format

Each line in the dataset files follows the FastText format:

__label__en [English sentence]
__label__shona [Shona sentence]

Example:

__label__en It was about three o 'clock before we could take a break .
__label__shona Ndakapindura kuti , " Hungu , Munyengetero waShe . "

Dataset Statistics

• Total sentences: 763,546
• Languages: English (en), Shona (shona)
• Training set: ~600,000 sentences
• Validation set: ~80,000 sentences
• Test set: ~80,000 sentences
• License: CC BY 4.0

Usage

For FastText Training

import fasttext

# Train a supervised language identification model
model = fasttext.train_supervised(
    input="data/shona_en_lang_train.txt",
    epoch=25,
    lr=1.0,
    wordNgrams=2,
    dim=100,
    loss='softmax'
)

# Test the model
result = model.test("data/shona_en_lang_test.txt")
print(f"Accuracy: {result[1]:.2f}%")

# Predict language of new text
text = "Hello, how are you?"
prediction = model.predict(text)
print(f"Language: {prediction[0][0].replace('__label__', '')}")

For Other ML Frameworks

The dataset can be easily adapted for scikit-learn, PyTorch, TensorFlow, or other machine learning frameworks by parsing the label and text components.

Publishing to Hugging Face

The repository includes a script for publishing the dataset to Hugging Face Hub:

from huggingface_publisher import publish_dataset

# Publish dataset
publish_dataset(
    hf_token="your_hf_token",
    repo_id="omanyasa/english-shona-langid",
    dataset_files=[
        "data/shona_en_lang_train.txt",
        "data/shona_en_lang_valid.txt", 
        "data/shona_en_lang_test.txt"
    ],
    readme_file="README.md"
)

Dependencies

See requirements.txt for the complete list of dependencies. Key packages include:

• huggingface_hub>=1.1.5 - For dataset publishing
• fasttext - For model training (optional)

License

This dataset is released under the Creative Commons Attribution 4.0 International License. You are free to share, adapt, and use the dataset for any purpose, provided you give appropriate credit.

Citation

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

English–Shona Language ID Dataset. (2025). 
Available at: https://huggingface.co/datasets/omanyasa/english-shona-langid

Contributing

Contributions to improve the dataset are welcome. Please ensure any additions maintain the quality and format consistency of the existing data.

Acknowledgments

This dataset was created to support language identification research and applications involving English and Shona languages, particularly for African NLP initiatives.