---
language:
- ku
- kmr
license: cc-by-2.0
size_categories:
- 10K<n<100K
task_categories:
- text-generation
- translation
tags:
- kurdish
- kurmanji
- nlp
- orthography
- grammar
---



# Cleaned Kurmanji Kurdish Sentences Dataset (Hawar Standard)

## Dataset Description
This dataset contains over 10,000 highly curated and grammatically corrected Kurmanji Kurdish sentences. While the original raw sentences were sourced from the open-source Tatoeba project, they have undergone extensive and meticulous editorial correction to meet the strict standards of the Hawar orthography and authentic Kurdish grammar (Celadet Alî Bedirxan rules).

### The Problem with Raw Data
Kurdish (Kurmanji) NLP resources often suffer from severe issues:
*   Machine-translated collocations (e.g., using "bersiv kirin" instead of the authentic "bersiv dan").
*   Incorrect casing, missing characters, and wrong use of circumflexes (ê, î, û, ç, ş).
*   Syntactical errors and wrong grammatical cases (Tewang/Ergativity).

### The Solution: Editorial Cleaning
Every sentence in this dataset has been processed to ensure maximum fluency, correct ergativity, native collocations, and perfect Hawar orthography. 
This makes the dataset highly valuable for:
*   **Text-to-Speech (TTS) & Automatic Speech Recognition (ASR):** Perfect pronunciation requires perfect text.
*   **Large Language Models (LLMs):** Fine-tuning models to write authentic, native-sounding Kurmanji.
*   **Language Learning Apps:** Providing flawless examples for students (e.g., Anki flashcards).

## Dataset Structure
*   `id`: The original Tatoeba sentence ID (for reference).
*   `original_text`: The raw, uncorrected text from Tatoeba.
*   `Korrektur_Kurmanci`: The final, expertly corrected and standardized sentence.

## License & Attribution
The original raw sentences are sourced from [Tatoeba](https://tatoeba.org/) and are released under the **CC-BY 2.0 (Creative Commons Attribution)** license. 

The extensive grammatical corrections, orthographic standardization, and editorial cleaning were performed by **Mehmet Cemal Celik**, a professional translator, writer, and linguist. This derivative work remains fully open and usable for both academic and commercial projects under the same CC-BY 2.0 terms.

## Author & Support
This dataset was cleaned and prepared by **Mehmet Cemal Celik**.

If you use this dataset in your research, commercial AI models, or applications, and would like to support the creation of more high-quality Kurdish NLP resources, consider supporting my work or reaching out for freelance translation, localization, and data-curation projects.

*   **Support my work:** [PayPal_ mc.chelik@gmail.com, https://www.patreon.com/cw/u85250978]
*   **Contact:** [www.amedcekojiyan.com, amedcj@gmail.com]