SHSK0118/TransformerPivotTranslation

1. Dataset Summary

This dataset provides parallel corpora for pivot-based neural machine translation research. It contains Polish–Czech, Czech–German, and Polish–German parallel sentence pairs, derived from the JRC-Acquis corpus and preprocessed for reproducible experiments.

The dataset was created to support research on pivot translation, particularly investigating Czech as an intermediate language for Polish–German translation.

2. Languages

• Polish (pl)
• Czech (cs)
• German (de)

Language Pairs:

• pl–cs
• cs–de
• pl–de

3. Dataset Structure

Each language pair is split into training, validation, and test sets with the following ratio:

9 : 0.5 : 0.5

Directory structure:

./
  ├ JRC-Acquis.cs-de.*
  ├ JRC-Acquis.cs-pl.*
  └ JRC-Acquis.de-pl.*

Each file contains aligned parallel sentence pairs.

4. Data Source

The dataset is derived from the JRC-Acquis Parallel Corpus, which consists of European Union legislative texts.

The original corpus is widely used in multilingual NLP research.

5. Preprocessing

The following preprocessing steps were applied:

• Sentence alignment verification
• Basic normalization
• Train/validation/test splitting (9:0.5:0.5)

No subword segmentation (e.g., BPE or SentencePiece) is applied in this dataset version.

6. Intended Use

This dataset is intended for:

• Pivot-based neural machine translation research
• Low-resource translation experiments
• Comparative evaluation of pivot language selection
• Evaluation using BLEU, GLEU, and CHRF metrics

It is not specifically designed for commercial deployment.

7. Reproducibility

The dataset includes pre-split files identical to those used in the original experiments.

In the original study:

• 60,000 sentence pairs were used for training per translation model
• 10,000 sentences were used for testing
• Transformer-based models were applied

8. Evaluation Metrics

• BLEU
• GLEU
• CHRF

9. Ethical Considerations

The dataset is based on publicly available legislative texts from the European Union. No personally identifiable information (PII) is intentionally included.

Users should verify compliance with the original JRC-Acquis license before use.

10. Citation

If you use this dataset, please cite:

Tokunaga, N., Ninomiya, T., & Tamura, A. (2020).
Neural Machine Translation for Polish–German via Czech as a Pivot Language.

Dataset citation:

Tokunaga, N., Ninomiya, T., & Tamura, A. (2020).
TransformerPivotTranslation Dataset.
Hugging Face.
https://huggingface.co/datasets/SHSK0118/TransformerPivotTranslation

11. License

This dataset inherits the licensing conditions of the original JRC-Acquis corpus. Users must ensure compliance with the original terms of use.

12. Contact

For academic inquiries, please contact the dataset authors.