Skloňuj
A web app for drilling Czech declension: nouns, adjectives and personal pronouns in all seven cases, singular and plural, with an explanation after every answer. Built for learners from A1 to B2 and for classes working through Krok za krokem.
"Skloňuj" is the Czech imperative for "decline (a noun)." In other words, "decline it!".
Live at sklonuj.com.
Features
Practice
- Four kinds of question: type the form, name the case, fill in a sentence, and a multi-step question that walks from the noun's pattern to its case to the form (with an adjective step for adjective–noun agreement)
- About 3,200 nouns, 60 adjectives and the personal pronouns, graded A1 to B2, in some 420 sentence frames that are matched to words by meaning, so the app asks for "Bydlím v Praze", not "Bydlím v polévce"
- An explanation every time: why the sentence needs that case, which of the 14 noun patterns the word follows, and the ending rule, with the ending in bold
- Diacritics-aware grading: "zenu" for "ženu" counts as a near miss, not a wrong answer; where Czech has two correct forms (pánovi / pánu) both are accepted
- Spaced repetition by pattern: each pattern × case × number is scheduled on its own, so what you get wrong comes back sooner and what you know is left alone; a missed question is also asked again a few questions later
- Less typing: a chip fills in the dictionary form so you only change the ending, and an accent bar follows the cursor
- Recorded Czech audio for every form (about 26,000 recordings), with the browser's voice as a fallback
Levels and textbooks
- CEFR levels A1–B2 decide which cases, words and plurals appear
- Krok za krokem chapters: practise with the cases and vocabulary of one lesson of KzK 1 (24 lessons) or KzK 2 (20 lessons)
Reference
- Lookup for 18,000+ nouns with full tables and English glosses, from MorphoDiTa
- Guides to the cases, the 14 patterns, the pronouns and the prepositions
Progress and classes
- Progress per case and pattern, a list of recent mistakes to review, streaks and achievements; saved on the device, and synced if you sign in
- A weekly leaderboard with generated Czech nicknames instead of real names
- Classes: teachers create a class, set assignments with a due date, and see each student's results
Data Sources
Declension data is generated by MorphoDiTa, a morphological analyzer built on the MorfFlex CZ 2.1 dictionary, developed at the Institute of Formal and Applied Linguistics (UFAL), Faculty of Mathematics and Physics, Charles University, Prague.
English translations are sourced from Wiktionary via kaikki.org.
For full citations and licensing details, see DATA_SOURCES.md.
Tech Stack
Getting Started
pnpm install
pnpm dev
Copy .env.example to .env and fill it in: sign-in, progress sync and classes need a Supabase project.
To run type checks, linting and tests:
pnpm check
pnpm lint
pnpm test
To format the codebase:
pnpm format
Adding Words
The word banks in src/lib/data/ are generated, not edited by hand. Declensions come from MorphoDiTa and translations from Wiktionary:
python3 scripts/add-word.py <lemma> [<lemma> ...] # adds to the lemma list and metadata
python3 scripts/build_word_bank_morphodita.py --merge --only "a,b,c" # builds just those words into the bank
pnpm tts:generate # records audio for the new forms
A wrong form is corrected in scripts/form_overrides.json and the word rebuilt, never patched in the JSON. Sentence frames (sentence_templates.json and its adjective and pronoun counterparts) are plain JSON and can be added to directly. CLAUDE.md has the full workflow, including proper nouns, nouns that decline like adjectives, and the Krok za krokem chapter lists.
To rebuild the lookup dictionary:
python3 scripts/build-dictionary.py
Audio
Pronunciation comes from pre-generated MP3s (edge-tts, cs-CZ-AntoninNeural) served from Cloudflare R2, with the Web Speech API as fallback. After a bank change:
pnpm tts:generate # resumable; only new forms are recorded
bash scripts/upload_audio_to_r2.sh sklonuj-audio # uploads what changed since the last upload
See the docstring in scripts/generate_tts.py for flags and troubleshooting.
License
Sklonuj is free for non-commercial use. It is dual-licensed by content type:
- Source code (everything not listed below) — PolyForm Noncommercial 1.0.0. A software licence that permits use, modification and redistribution for any non-commercial purpose.
- Linguistic data and audio (
src/lib/data/**, static/audio/**, the word lists in scripts/) — CC BY-NC-SA 4.0. This is required by the upstream MorfFlex CZ licence and cannot be relaxed.
Commercial use of either part requires separate permission (and, for the data, a commercial licence from ÚFAL for MorfFlex CZ). See DATA_SOURCES.md for upstream attribution.
