shiima/aya-processed-central-kurdish-paraphrase

Aya Processed Central Kurdish Paraphrase Identification Dataset

Dataset Description

This dataset contains 49,401 preprocessed Central Kurdish paraphrase identification samples in conversational format, derived from the Aya Collection Language Split.

Languages

• Central Kurdish (ckb)

Dataset Structure

The dataset contains the following columns:

• id: Unique identifier for each sample
• prompt: Conversational format as JSON array containing the input-paraphrase pair

Prompt Format

Each prompt is a JSON array with the following structure:

[
  {
    "from": "user",
    "value": "preprocessed input text in Central Kurdish"
  },
  {
    "from": "assistant",
    "value": "preprocessed paraphrase in Central Kurdish"
  }
]

Data Processing

All text has been preprocessed using the asosoft library with the following pipeline:

1. Text normalization
2. Punctuation normalization
3. Line trimming

Example Usage

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("shiima/aya-processed-central-kurdish-paraphrase")

# View a sample
sample = dataset['train'][0]
print(f"ID: {sample['id']}")

# Parse the conversation
conversation = json.loads(sample['prompt'])
for turn in conversation:
    print(f"{turn['from']}: {turn['value']}")

# Output example:
# user: ئەم ڕستەیە بنووسە
# assistant: ئەم دێڕە بنووسە

# Iterate through samples
for sample in dataset['train']:
    conversation = json.loads(sample['prompt'])
    original = conversation[0]['value']
    paraphrase = conversation[1]['value']
    print(f"Original: {original}")
    print(f"Paraphrase: {paraphrase}")
    break

Use Cases

This dataset is suitable for:

• Fine-tuning paraphrase generation models for Central Kurdish
• Training models to identify paraphrases
• Text rewriting and style transfer
• Data augmentation for NLP tasks
• Conversational AI development

Source Dataset

This dataset is derived from:

• Original Dataset: Aya Collection Language Split
• Filter: Only 'paraphrase-identification' task type samples
• Language: Central Kurdish split

Statistics

• Total Samples: 49,401
• Task Type: Paraphrase Identification
• Language: Central Kurdish (ckb)
• Script: Arabic script
• Format: Conversational JSON format
• Columns: 2 (id, prompt)

Citation

If you use this dataset, please cite the original Aya Collection:

@misc{aya_collection,
  title={Aya Collection},
  author={Cohere For AI},
  year={2024},
  url={https://huggingface.co/datasets/CohereForAI/aya_collection}
}

License

This dataset inherits the license from the original Aya Collection dataset.

Acknowledgments

• Original dataset creators: Cohere For AI
• Preprocessing library: asosoft by Kurdish NLP

Contact

For questions or issues, please open an issue on the dataset repository.