amanuelbyte/af-en-translation-dataset

Afrikaans-English Translation Dataset

A high-quality, curated parallel corpus for machine translation between English (en) and Afrikaans (af), created through extensive data processing and quality filtering.

📊 Dataset Overview

• Language Pair: English → Afrikaans (en-af)
• Total Sentence Pairs: 161,644
• Quality Threshold: QE ≥ 0.7
• Processing Date: September 22, 2025
• Repository: amanuelbyte/af-en-translation-dataset

🎯 Key Features

• Quality Assured: All sentence pairs meet minimum Quality Estimation (QE) score of 0.7
• Deduplicated: Extensive deduplication process removes redundant content
• Multi-Domain: Covers diverse domains including technical, religious, educational, and general content
• Clean Data: Filtered for language identification, semantic similarity, and rule-based quality checks

📈 Dataset Statistics

Data Volume

• Before Deduplication: 192,541 sentence pairs
• After Deduplication: 161,644 sentence pairs
• Deduplication Ratio: 16.1% reduction

Quality Metrics

• Average QE Score: 0.78
• Highest Quality Dataset: Tatoeba (QE: 0.861)
• Lowest Quality Dataset: KDE4 (QE: 0.723)

📚 Data Sources

This dataset combines 20 high-quality sources:

Source	Dataset	Pairs	QE Score	Domain
wikimedia	Wikipedia Articles	56,766	0.781	General
bible-uedin	Bible Translations	51,118	0.748	Religious
OpenSubtitles	Movie Subtitles	37,731	0.795	Entertainment
qed	Europarl Questions	10,412	0.773	Political
GNOME	Software Localization	6,169	0.761	Technical
spc	Scientific Papers	6,811	0.739	Academic
kde4	KDE Software	5,365	0.723	Technical
Pontoon-Translations	Mozilla Translations	5,606	0.785	Technical
Weblate-Translations	Weblate Platform	3,978	0.757	General
TED2020	TED Talks	1,882	0.783	Educational
ntrex	News Articles	1,806	0.777	News
Tatoeba	Community Translations	2,197	0.861	General
smol	Smol Dataset	812	0.789	General
lingvanex_test_references	Test References	748	0.840	General
ELRC-wikipedia_health	Health Wikipedia	371	0.808	Medical
ELRC_2922	European Language Resources	369	0.808	General
ELRC-3039-wikipedia_health	Health Wikipedia	371	0.808	Medical
ted_talks_iwslt	TED/IWSLT Talks	27	0.770	Educational
ELRC-2638-monumentos_2007	Cultural Heritage	2	0.848	Cultural
lingvanex_test_references	Test References	748	0.840	General

🛠️ Processing Pipeline

The dataset underwent rigorous processing:

1. Rule-based Filtering: Length constraints, character validation
2. Semantic Filtering: Sentence embedding similarity (threshold: 0.7)
3. Language Detection: FastText + AfroLID models for accurate language identification
4. Quality Estimation: Africomet QE model for translation quality scoring
5. Sentence-level QE Filtering: Only pairs with QE ≥ 0.7 retained
6. Deduplication: 4-stage process removing identical and similar content

Deduplication Process

1. Remove identical source-target pairs
2. Remove duplicate (source, target) combinations
3. Remove duplicate source sentences
4. Remove duplicate target sentences

💾 Dataset Format

The dataset is available in Hugging Face datasets format with the following structure:

from datasets import load_dataset

dataset = load_dataset("amanuelbyte/af-en-translation-dataset")
print(dataset)

# Dataset structure:
# DatasetDict({
#     "train": Dataset({
#         features: ['en', 'af'],
#         num_rows: 161644
#     })
# })

Columns

• en: English source text
• af: Afrikaans target text

📖 Usage Examples

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("amanuelbyte/af-en-translation-dataset")

# Access the training split
train_data = dataset["train"]

# Get a sample
sample = train_data[0]
print(f"English: {sample['en']}")
print(f"Afrikaans: {sample['af']}")

For Machine Translation Training

# Prepare data for training
source_texts = train_data["en"]
target_texts = train_data["af"]

# Statistics
print(f"Dataset size: {len(train_data)}")
print(f"Average source length: {sum(len(text.split()) for text in source_texts) / len(source_texts)".1f"}")
print(f"Average target length: {sum(len(text.split()) for text in target_texts) / len(target_texts)".1f"}")

🎓 Applications

This dataset is suitable for:

• Machine Translation: Training and fine-tuning MT models
• Language Modeling: Pretraining language models for Afrikaans
• Cross-lingual Transfer: Multilingual NLP tasks
• Quality Estimation: Research on translation quality
• Linguistic Studies: Analysis of Afrikaans-English language pairs

🔍 Quality Assessment

The dataset includes comprehensive quality metrics:

• Individual Dataset QE Scores: Range from 0.723 to 0.861
• Overall Average QE: 0.78
• Quality Distribution: All datasets exceed 0.7 threshold
• Consistency: Uniform quality across different domains

📄 License

This dataset is released under CC BY-SA 4.0 license. The individual source datasets maintain their original licenses.

🤝 Citation

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

@dataset{af-en-translation-dataset-2025,
  title={Afrikaans-English Translation Dataset},
  author={Amanuel Tewolde},
  year={2025},
  publisher={Hugging Face},
  journal={Hugging Face Hub},
  howpublished={\url{https://huggingface.co/amanuelbyte/af-en-translation-dataset}}
}

📞 Contact

For questions or issues regarding this dataset, please contact:

• Maintainer: Amanuel Tewolde
• Email: [email protected]
• Organization: Hugging Face Community

🔗 Related Resources

• AfroLID Model - Language identification
• Africomet QE - Quality estimation
• Processing Pipeline - Source code

---

Note: This dataset was created using automated processing pipelines with quality thresholds. While extensive filtering has been applied, users should perform their own quality assessment based on their specific use cases.