---
language:
  - ckb
  - kmr
  - diq
  - hac
license: cc-by-4.0
task_categories:
  - text-generation
  - fill-mask
pretty_name: Kurdish Corpus
size_categories:
  - 100M<n<1B
---

# Kurdish Corpus

A large-scale multi-source Kurdish language dataset for training language models.

## Dataset Statistics

- **Total Documents**: 1,797,686
- **Total Tokens**: 625,716,980
- **Shards**: 4
- **Built**: 2026-05-02

### By Language

| Language | Documents |
|----------|-----------|
| Sorani (ckb) | 1,274,425 |
| Kurmanji (kmr) | 478,540 |
| Zazaki (diq) | 34,069 |
| Hawrami (hac) | 10,652 |

### By Source Type

| Source Type | Documents |
|-------------|-----------|
| News | 1,443,750 |
| Web | 229,705 |
| Wikipedia | 124,231 |

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("kurdish-ai/kurdish-corpus")

# Access the data
for example in dataset["train"]:
    print(example["text"][:100])

# Filter by language
ckb_data = dataset.filter(lambda x: x["language"] == "ckb")

# Filter by source type
news_data = dataset.filter(lambda x: x["source_type"] == "news")

# Filter by quality
high_quality = dataset.filter(lambda x: x["quality_score"] >= 80)
```

## Schema

| Field | Type | Description |
|-------|------|-------------|
| text | string | Document text |
| source | string | Source identifier |
| source_type | string | Category (news, wikipedia) |
| language | string | Language code (ckb, kmr, diq, hac) |
| quality_score | int | Quality score 0-100 |
| word_count | int | Word count |
| token_count | int | Estimated token count |
| url | string | Source URL |

## License

CC-BY-4.0

## Citation

```bibtex
@dataset{kurdish_corpus_2026,
  title={Kurdish Corpus},
  author={Shahin Wahab},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/kurdish-ai/kurdish-corpus}
}
```