panagoa/tatoeba_raw_fra_kbd

French-to-Kabardian Tatoeba Raw Machine Translations Dataset

This dataset contains translations of sentences from French to Kabardian, sourced from the Tatoeba project. It provides a substantial parallel corpus for machine translation and linguistic research involving the Kabardian language.

Dataset Description

Dataset Overview

This dataset contains over 14 million sentence pairs, with source sentences in French and their translations in Kabardian. Each entry includes metadata such as the original Tatoeba sentence ID, language codes, and similarity scores.

The dataset serves as a valuable resource for:

• Machine translation systems targeting Kabardian
• Linguistic research on low-resource languages
• Cross-lingual NLP applications
• Preservation and documentation of the Kabardian language

Dataset Structure

{
  'tatoeba_id': 12345,
  'source_lang': 'fra',
  'target_lang': 'kbd',
  'source': 'Example sentence in French',
  'target': 'Translation in Kabardian',
  'similarity': 0.95,
  'model': 'model-name',
  '__index_level_0__': 0
}

Dataset Creation

Source Data

The source data comes from the Tatoeba Project, which is a collection of sentences and translations in many languages. The French sentences were used as source text for machine translation into Kabardian.

Translation Process

This dataset is a raw machine translation output without any human validation or post-editing:

• All translations were generated using various fine-tuned versions of the NLLB-200 (No Language Left Behind) model
• No manual corrections or human review was performed on the translations
• The dataset represents direct output from the machine translation system
• The 'similarity' score refers to the model's internal confidence metric
• The 'model' field indicates which specific fine-tuned NLLB-200 version was used for each translation

Considerations for Using the Data

Social Impact and Biases

This dataset contributes to language preservation and technological inclusion for the Kabardian language community. However, as with any translation dataset:

• There may be cultural nuances that are difficult to translate accurately
• The distribution of topics may not represent the full breadth of the language
• Potential biases in the source content could be propagated through translations

Discussion of Limitations

As this is a raw machine translation dataset without human validation:

• Translation quality may vary significantly across the dataset
• There may be systematic errors common to neural machine translation systems
• The translations have not been verified by native Kabardian speakers
• The model may struggle with cultural concepts, idioms, and specialized terminology
• The translations should be considered as a starting point rather than a gold standard
• This dataset is most useful for further fine-tuning of translation models or as a basis for human post-editing

Additional Information

Dataset Curators

This dataset has been curated from the Tatoeba project's translations. For specific information about the curation process, please contact the dataset contributors.

Licensing Information

This dataset is licensed under the Creative Commons Attribution 2.0 Generic (CC BY 2.0) license, which allows for sharing and adaptation with appropriate attribution.

Citation Information

If you use this dataset in your research, please cite both the Tatoeba project and this dataset collection.

Contributions

Contributions to improve the dataset are welcome. Please refer to the Hugging Face dataset contribution guidelines.