zimankit · زمانکیت

Website · Try it in your browser · Guide · Download · Changelog

Kurdish language skills and wordlist builder for Central Kurdish (Sorani, ckb, Arabic script) and Northern Kurdish (Kurmanji, kmr, Hawar Latin script).

ziman means "language" in both varieties.

Install

zimankit needs Python 3.11 or newer and has no dependencies.

pip install zimankit

On Debian, Ubuntu or Kali, the latest release also has a package:

sudo apt install ./python3-zimankit_*_all.deb

The release also has the full source as .zip and .tar.gz, and SHA256SUMS. Optional extras: pip install "zimankit[data]" reads .mdx and .zim files, and pip install "zimankit[kaggle]" adds the Parquet export. The guide covers every command and the Python API.

Quick start

uv venv --python 3.13 .venv && uv pip install --python .venv/bin/python mdict-utils libzim pytest
source .venv/bin/activate

python -m zimankit analyze کتێبەکەم jinekê
python -m zimankit ipa پیاوەکە کتێبەکەم jinekê dîtin      # pjaːwækæ k(ɯ)teːbækæm ʒɪnɛkeː diːtɯn
python -m zimankit conjugate کردن
echo "Min tu dîtim." | python -m zimankit ergativity       # agreement_ok: false
python -m zimankit parse "Min pirtûka nû li dibistanê xwend."   # head and relation of each word
python -m zimankit lid "Ma yew roje şima ra vat ke ziwanê ma zaf rind o"   # zza
python -m pytest -q

Build the wordlists from your own data:

cp config/sources.example.toml config/sources.local.toml   # set the paths (file is git-ignored)
./scripts/build_wordlists.sh

# Kaggle-ready dataset: flat columns, CSV + Parquet + SQLite + data card -> output/kaggle/
uv pip install --python .venv/bin/python pyarrow
python run_pipeline.py --stages export

Documentation

Start at the documentation index.

Learning Kurdish? The tutorials are two complete courses, Sorani and Kurmanji, with 23 lessons each from the alphabet to ergativity, tagged 🟢 beginner / 🟡 intermediate / 🔴 advanced. Every conjugation, declension, IPA and transliteration table in them is generated by zimankit (python scripts/gen_tutorials.py), and a test fails if a lesson drifts from the code.

Also: agent skills · cross-language examples · engineering brief · contributing and releases · changelog · website source.

Layout

zimankit/
├── run_pipeline.py          lexicon → dictionaries → corpus → build → rules → export
├── config/                  pipeline.toml · sources.example.toml · sources.local.toml (ignored)
├── zimankit/
│   ├── tools/               mdx_extractor.py · zim_cleaner.py
│   ├── shared_utils/        normalizer · tokenizer · lid · phonology_core · text_helpers · unicode_tables · schema
│   ├── skills_ckb/          orthography · phonology_ipa · morphology · verbs · config.toml
│   ├── skills_kmr/          orthography · phonology_ipa · morphology · verbs · syntax · translit · config.toml
│   ├── lexicon/             wiki_lexicon · corpus · harvest · builder · export
│   ├── data/                authored reference tables (core verbs, function words)
│   ├── dependency.py        sentence structure: head and relation of each word
│   ├── rules.py             grammar_rules_<lang>.json exporter
│   └── cli.py               python -m zimankit …
├── output/{ckb,kmr}/        generated wordlists and reports
├── output/kaggle/           flat Kaggle dataset (export stage)
├── docs/                    developer chapters · tutorials/{ckb,kmr} (generated tables)
├── assets/brand/           logo, icons, social, ads, print, screenshots (scripts/make_brand.py)
├── site/                   website source (scripts/build_site.sh → _site/, deployed from main)
├── .github/workflows/      CI · release (on vX.Y.Z tags) · website
├── examples/  skills/  scripts/  tests/

Output schema

grapheme, lemma, pos, ipa, morph_meta, translit, frequency, sources, confidence — see chapter 16.

کتێبەکەم,کتێب,NOUN,k(ɯ)teːbækæm,"{""definite"":""def"",""enclitic"":""1sg"",""number"":""sg"",""seg"":""کتێب+ەکە+م""}",kitêbekem,137,"encyclopedia,news,web",0.31
jinekê,jin,NOUN,ʒɪnɛkeː,"{""case"":""obl"",""definite"":""ind"",""gender"":""f"",""number"":""sg"",""seg"":""jin+ekê""}",ژنەکێ,530,"encyclopedia,lexicon-form,news",0.754

The logo

The logo is a sign of deq, the traditional hand-poked tattoo of Kurdish women: the figure read as the tree of life, a symbol of fertility, growth and life force. Its circle stands for the cycle of life, its line and V shapes for a sprout or a tree, and its dots for protection against the evil eye. At its heart shines Roj, the 21-rayed sun of the flag of Kurdistan; roj / ڕۆژ means both "sun" and "day". Together they stand for Kurdish (ziman, "language") as a living language.

The full story, with the history and meaning of deq, sources, colours, usage, and every logo, icon, social, ad and print file, is in assets/brand/.

Licence

The code is licensed under the GNU General Public License v3.0 or later.

Generated data is licensed separately. The large generated files under output/ are git-ignored (regenerate them with the pipeline; 300-row samples are kept). Wordlists derived from openly licensed sources may carry share-alike obligations when redistributed — check the licences of the data you configure. The published Kaggle dataset and the sample CSVs in output/ are CC BY-SA 4.0.