michsethowusu/twi-english-parallel-synthetic-50m

Twi-English Parallel Dataset

Dataset Description

This dataset contains a large-scale parallel corpus of Twi-English sentence pairs, featuring synthetically generated Twi sentences with corresponding English paraphrases. The dataset is designed to support machine translation, cross-lingual understanding, and other NLP tasks involving the Twi language (a dialect of Akan spoken in Ghana).

Dataset Summary

• Total Size: 47,924,398 parallel sentence pairs
• Language Pair: Twi (tw) ↔ English (en)
• Generation Method: Synthetic generation with dictionary-based paraphrasing

Languages

• Twi (tw): A dialect of Akan, primarily spoken in Ghana. Part of the Niger-Congo language family.
• English (en): Used as the target language for translation tasks.

Dataset Structure

Data Fields

• twi_sentence: Synthetically generated sentence in Twi language
• english_parahrase: Corresponding English paraphrase created using Twi-to-English dictionary mapping

Dataset Size

Total Examples
47,924,398

Example Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("michsethowusu/twi-english-parallel-synthetic-50m")

# Access the data
print("Twi:", dataset[0]['twi_sentence'])
print("English:", dataset[0]['english_parahrase'])

# Iterate through examples
for example in dataset:
    twi_text = example['twi_sentence']
    english_text = example['english_parahrase']
    # Your processing here...

Dataset Creation

Source Data

The Twi sentences in this dataset were synthetically generated, and English paraphrases were created through dictionary-based translation methods. This approach enables the creation of large-scale parallel corpora for low-resource languages like Twi.

Data Collection Process

1. Synthetic Generation: Twi sentences were generated using computational methods
2. Dictionary Mapping: English paraphrases were created using Twi-to-English dictionary resources
3. Quality Control: Data cleaning and filtering applied to ensure quality

Annotations

The dataset does not contain manual annotations but relies on algorithmic generation and dictionary-based translation.

Considerations for Use

Social Impact and Biases

• This dataset uses synthetic generation methods, which may not capture the full linguistic diversity and cultural nuances of natural Twi speech
• Dictionary-based paraphrasing may introduce systematic biases or miss contextual meanings
• Users should be aware that synthetic data may not generalize perfectly to real-world Twi usage

Limitations

• Synthetic Nature: Generated sentences may not reflect natural language patterns
• Dictionary Constraints: English paraphrases are limited by dictionary coverage and may lack contextual accuracy
• Domain Coverage: May not cover all domains or registers of Twi language use

Recommended Use Cases

• Machine Translation: Training MT systems for Twi-English translation
• Cross-lingual NLP: Developing cross-lingual models involving Twi
• Low-resource Language Research: Studying synthetic data approaches for under-resourced languages
• Educational Tools: Creating language learning applications for Twi

Licensing and Attribution

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Citation

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

@dataset{twi_english_parallel,
  title={Twi-English Parallel Dataset},
  author={[Your Name/Organization]},
  year={2025},
  url={https://huggingface.co/datasets/michsethowusu/twi-english-parallel-synthetic-50m},
  note={Large-scale synthetic parallel corpus for Twi-English translation}
}

Additional Information

Dataset Curators

[Add your information here]

Contact Information

For questions or issues regarding this dataset, please [add contact information].

---

Disclaimer: This dataset contains synthetically generated content. While efforts have been made to ensure quality, users should validate the data for their specific use cases.