README.md

---
license:
- cc-by-sa-4.0
- odc-by
- cc0-1.0
- other
language:
- kmr
- ckb
- diq
multilinguality: multilingual
task_categories:
- text-generation
- fill-mask
pretty_name: Kurdish National Corpus v1.1 (Kurmanji-Native)
size_categories:
- 1M<n<10M
dataset_info:
  features:
  - name: id
    dtype: string
  - name: text
    dtype: string
  - name: title
    dtype: string
  - name: language
    dtype: string
  - name: dialect
    dtype: string
  - name: source
    dtype: string
  - name: url
    dtype: string
  - name: license
    dtype: string
  - name: author
    dtype: string
  - name: date
    dtype: string
  - name: num_chars
    dtype: int64
  - name: num_words
    dtype: int64
  - name: lid_label
    dtype: string
  - name: lid_prob
    dtype: float64
  - name: quality_score
    dtype: float64
  - name: subject
    dtype: string
  - name: level
    dtype: string
  - name: grade
    dtype: string
  - name: pages
    dtype: int64
  splits:
  - name: train
    num_bytes: 7681828199
    num_examples: 2734857
  download_size: 7681828199
  dataset_size: 7681828199
configs:
- config_name: default
  data_files:
  - split: train
    path: corpus_v2_open_release.jsonl
---

# The Largest Documented Kurmanji and Multi-Dialect Kurdish Dataset — v1.1

A large-scale, multi-dialect Kurdish text corpus prioritizing **native Kurmanji
(Northern Kurdish)** fluency, with substantial Sorani (Central Kurdish) and a
Zazaki baseline. Built for language modeling, tokenizer training, and
general-purpose Kurdish NLP.

**This release contains only openly-licensed or presumptively-free
redistributable content.** A parallel research-tier subset (copyrighted
commercial news/broadcast scrapes and a small set of copyrighted private
books — see *Licensing* below) exists locally for internal training use
but is deliberately excluded from this public release.

## Dataset Statistics (v1.1, open-release)

| Dialect | Documents | Words | Doc share | Word share |
|---|---|---|---|---|
| Kurmanji (kmr) | 2,121,482 | 603,988,734 | 77.6% | 73.0% |
| Sorani (ckb) | 595,877 | 219,089,076 | 21.8% | 26.5% |
| Zazaki (diq) | 17,498 | 4,764,337 | 0.6% | 0.6% |
| **Total** | **2,734,857** | **827,842,147** | | |

v1.1 adds ~371,500 Kurmanji documents (+167.1M words) over v1.0, almost
entirely from two new sources (see *Sources* below): `kurdish-web`, a
CC-BY-4.0 Kurdish/Zaza web-crawl corpus (Kurmanji portion only — this
project deliberately ingests Kurmanji-only from new multi-dialect sources
by default, leaving that source's Sorani/Zazaki configs uncollected), and
`kurdish-theses`, a small CC-BY-4.0 academic-thesis corpus.
Sorani and Zazaki totals are byte-for-byte unchanged from v1.0, confirming
this update is Kurmanji-only as intended.

### Exact token counts (verified, reproducible)

The word counts above are whitespace-token counts computed at extraction
time, not subword/model tokens. For an exact figure, every document in
this exact release file was tokenized with OpenAI's `tiktoken`
(`cl100k_base` encoding) on 2026-07-17:

| Dialect | Tokens (cl100k_base) | Share |
|---|---|---|
| Kurmanji (kmr) | 1,589,729,406 | 53.5% |
| Sorani (ckb) | 1,367,695,358 | 46.1% |
| Zazaki (diq) | 12,436,166 | 0.4% |
| **Total** | **2,969,860,930** | |

Sorani's Arabic script is markedly less token-efficient under generic BPE
vocabularies than Kurmanji's Latin script (roughly 7 tokens/word vs. ~2.8),
so its token share overstates its real proportion relative to the word
counts above; a custom Kurdish tokenizer would close most of this gap.

This count is exact and reproducible, not an estimate — run
`python scripts/count_exact_tokens.py data/final/corpus_v2_open_release.jsonl`
yourself against the released file to verify it. It is also, as far as we
know, the only published Kurdish corpus that discloses a per-dialect token
breakdown at all: other large aggregate Kurdish corpora advertise a single
multidialectal total token count without stating which tokenizer produced
it or how it splits across dialects, which makes independent verification
or a fair Kurmanji-specific comparison impossible from their documentation
alone.

## Sources

| Source | Documents | License |
|---|---|---|
| FineWeb-2 (HuggingFaceFW) | 876,654 | ODC-BY-1.0 |
| Kurdish Web Corpus (Kurmanji portion only; internal source tag `kurdish-web`) | 392,721 | CC-BY-4.0 — created by HF user [muzaffercky](https://huggingface.co/datasets/muzaffercky/kurdish-web) |
| Wiki-Ferheng (wiktextract of Kurdish Wiktionary) | 453,390 | CC-BY-SA-4.0 + GFDL |
| Kurdish Wiktionary (ku/ckb, direct XML extraction) | 374,565 | CC-BY-SA-4.0 |
| MADLAD-400 | 198,696 | ODC-BY-1.0 |
| HPLT 2.0 | 167,389 | CC0-1.0 |
| CC-100 | 136,305 | Common Crawl ToU |
| Kurdish Wikipedia (ku/ckb/diq) | 98,619 | CC-BY-SA-4.0 |
| OPUS CCAligned | 27,460 | CC-derived (Common Crawl terms) |
| Tatoeba | 8,228 | CC-BY-2.0-FR |
| Kurdish Academic Theses (internal source tag `kurdish-theses`) | 373 | CC-BY-4.0 — created by HF user [muzaffercky](https://huggingface.co/datasets/muzaffercky/kurdish-kurmanji-theses), declared for their derived/processed text; the underlying theses' upstream terms from YÖK Tez Merkezi (Turkey's national thesis repository) are not independently verified by this project — same category of caveat as the Rojava curriculum below |
| Kurdish Wikiquote | 152 | CC-BY-SA-4.0 |
| Internet Archive (public-domain-verified books) | 127 | Public Domain Mark / CC0 / age-verified PD |
| Rojava school curriculum (PDFs) | 99 | Presumptively free educational material, distributed by the Rojava Autonomous Administration for schools' use — *per the data provider's representation, not independently verified by this project against a specific copyright framework* (three Arabic-language subject textbooks from this same batch were identified and removed — see *Data Cleaning* below) |
| VOA Kurdish (Kurmanji + Sorani) | 72 | Public Domain (U.S. government work, 17 U.S.C. §105) |
| Kurdish Wikibooks | 7 | CC-BY-SA-4.0 |

The *Sources* and *Licensing* sections above are self-contained for this
release's purposes; they summarize a longer internal source-verification
process (per-source license checks, volume estimates, and the engineering
history behind each collection phase) that isn't included in this
distribution.

## Licensing — why some content is *not* in this release

This project maintains a two-tier policy: **open** (the license permits
redistribution, included here) and **research** (copyrighted material
collected for internal model training only, never redistributed). The
following sources exist in the project's full working corpus but are
**excluded from this public release**:

| Source | Docs excluded | Why |
|---|---|---|
| Ronahi TV news/broadcast (Kurmanji) | 48,768 | Copyrighted broadcast content|
| Kurdistan24 news (Kurmanji) | 165 | Copyrighted commercial news |
| Rudaw news (Kurmanji) | 93 | Copyrighted commercial news |
| Kurdistan Parliament legislation | 177 | Government text, presumptively PD but not independently verified against Iraqi/KRG statute |
| Local PDF books (incl. one political work) | 3 | Copyrighted, non-openly-licensed |

Note on the Rojava school curriculum specifically: this source is **included**
in the release (see *Sources* above), on the basis that it is distributed
freely by the Rojava Autonomous Administration for educational use. That
representation has not been independently verified by this project against
a specific copyright statute or license text — the same category of
uncertainty as the Kurdistan Parliament legislation above, which is *kept*
research-tier for exactly this reason. If you plan to build on this subset
specifically, you may want to do your own diligence before treating it as
unambiguously public domain.

If you need the research-tier subset for your own (non-redistributed)
training, it is available in the project's working directory; it is not
part of this Hugging Face release and should not be re-uploaded elsewhere
without independently clearing the licensing questions above.

## Data Cleaning & Quality Pipeline

1. **Collection** — per-source extractors (Wikimedia XML dumps, HPLT/MADLAD/
   FineWeb-2/CC-100 web-crawl shards, OPUS parallel corpora, targeted PDF/HTML
   scrapers) into a common JSONL schema.
2. **Exact deduplication** — hash-based, at merge time.
3. **Near-duplicate removal** — MinHash/LSH (Jaccard ≥0.8) across the whole
   prose corpus, keeper priority favoring more-curated sources.
4. **Language identification** — GlotLID (fastText), removing documents
   whose predicted language isn't a Kurdish variant above threshold.
   Dictionary/curriculum sources are exempted from *removal* but still
   fully *classified*, which enabled the fix below.
5. **Gopher-style quality filtering** — min/max length, symbol ratio,
   repetition (line/paragraph/n-gram), stopword presence; dictionary and
   curriculum sources are exempted from the length/repetition battery
   (short definitions and formulas would otherwise be wrongly rejected)
   but still checked for basic non-corruption.
6. **Surgical language sanitation (this release's key fix)** — during PDF
   ingestion of the Rojava school curriculum, three Arabic-language subject
   textbooks (an elementary Arabic-language course, titled "عربي" / "عربي
   مكون كردي") slipped through alongside ~100 genuinely Kurdish curriculum
   PDFs. **A first attempt to remove them** with a hand-built Arabic
   character/stopword density heuristic backfired badly: because several
   Arabic letters (`ة ث ص ض ط ظ ذ ي`) also appear incidentally in ordinary
   Kurdish text — quoting an Arabic-script proper noun in parentheses (a
   standard Kurdish-Wikipedia convention, e.g. *"Trablûs (bi erebî:
   طرابلس)"*), or a dictionary correctly documenting an Arabic loanword's
   etymology — the density filter **quarantined 17,574 documents when only
   3 were actually contaminated**, with 92% of the false positives coming
   from already-verified-clean FineWeb-2 Sorani content. **The fix**: every
   document's GlotLID classification (`lid_label`) was already computed and
   stored during step 4, just never used for this purpose. Filtering
   specifically on `source == "rojava-curriculum" AND lid_label ==
   "arz_Arab"` identified **exactly the 3 contaminated documents, with zero
   false positives** — confirmed by manual inspection of all three
   (subject fields read "عربي" / "عربي مكون كردي") and of the full 102-document
   curriculum set (the rest carry Kurdish or classification-noise labels,
   never a genuine Arabic one). This is the recommended lesson for anyone
   extending this corpus: **prefer a model-based whole-document signal
   already computed by your pipeline over a hand-built character/keyword
   heuristic** for language-purity checks — the latter cannot distinguish
   "this document is in language X" from "this document merely mentions
   language X." This quarantine step is re-applied on every pipeline
   rebuild (new sources go through the same merge → near-dedup → LID →
   quality → surgical-sanitation sequence); the v1.1 rebuild reconfirmed
   the identical 3 documents and zero false positives.

## Intended Use

Pretraining and fine-tuning language models, embeddings, tokenizers, and
other NLP systems for Kurdish (Kurmanji-priority, with Sorani and Zazaki
support). Not vetted for toxic/harmful content beyond the quality filters
above; downstream users training public-facing systems should apply their
own safety filtering.

## Citation

If you use this dataset, please cite the source repository and, where
applicable, the upstream datasets listed in *Sources* above (particularly
FineWeb-2, MADLAD-400, HPLT, and the Kurdish Wikimedia projects, each of
which has its own citation guidance).