uavocab.org is open: my Ukrainian word list, rebuilt for learners

The study is built to be checked. It shows the coverage curve, the method, a SHA-256 checksum for every file in every release, DOIs on Zenodo, and checks that re-run in your browser. That is the right page for someone deciding whether to trust the list. It is the wrong page for someone who wants to know what говорити means and what to learn after it.

uavocab.org is the reader's edition. It takes nothing from the study except its committed data: the word pages, the levels, the decks and the course are built from the same files the study releases, and no number on a guide is typed by hand. A test rebuilds the word pages and fails when they fall out of step with the study, so when the list improves, both doors change together. The research page on the new site is built the same way, from the study's own registries, so it cannot drift from the study either.

The plan called it ukvocab.org. To most readers "uk" means the United Kingdom, so the site is uavocab.org. The Python package (pip install ukvocab) and the Zenodo records keep the old name.

Every word with a page carries a level from A1 to C1, and the levels come from PULS, the validated Ukrainian vocabulary profile built by a team at the Ukrainian Catholic University in Lviv from 21 textbooks for learners and checked against two large corpora. The site uses a dated snapshot of it and says so.

The order is worth following because words are used so unevenly. The 860 words rated A1 are 37.95 percent of running text in modern Ukrainian. Each level after that buys less, and about a quarter of running text is words with no validated level at all, mostly rarer words along with names and numbers. The bar below is uavocab.org's own front-page chart, brought into this post.

A word page opens on the word with its stress marked, because Ukrainian print never marks it (how the stress column was built). A play button plays a recording of it where one exists, with the recording's credit. Below that come its part of speech, its level and frequency rank, its share of running text, its meanings, and the words near it in frequency. Here is дякувати, to thank. Shared, a word page shows a card with the word, its stress mark and its meaning, so the link previews the word itself.

The card beside the meanings is the part that matters most. It names the exact lesson in the course that teaches the word (дякувати is A1, unit 4, lesson 6) and links to that lesson and to the level's Anki deck. A word list that cannot tell you where to practice a word is a reference. This one is meant to be used.

Not every word gets a page. A word needs a validated level and at least one dictionary meaning, and for now only the 3,000 most frequent of those have one. A thin page is worse than none, so the cap rises once search engines show they are reading the first 3,000.

One small fix took longer than it should have. Roboto Flex, the variable cut of the site's typeface, draws the combining stress mark one letter late on Cyrillic, so кни́га looked stressed on the г. The site sets plain Roboto instead, and a test keeps it that way.

Learn Ukrainian, the free course, used to live on this site. On October 10 it moved to uavocab.org, and the old links redirect there. It teaches the validated words in frequency order: 44 units and 326 lessons from the alphabet to C1, twenty words to a lesson after the alphabet, spaced review that schedules every word from how well you knew it, and a checkpoint at the end of each unit. No account is needed, since progress stays in your browser, and signing in with your email keeps it in sync.

Moving a course whose progress lives in the browser means carrying the progress across too. A browser keeps each site's saved data away from every other site's, so jakelawrence.xyz/learn/handoff carries a learner's progress across once, inside the link to the new site, where it merges in without lowering anything they had already learned.

If you already study with Anki, the decks page has one deck per level, 5,909 cards in all, with the stressed word on the front and its meaning on the back, beside its level and a link to its lesson. Decks with recordings are assembled in your own browser from the recordings' public storage, so no server keeps a copy. With Anki open on a computer, the page can add the cards to it directly through the AnkiConnect add-on, skipping any card you already have, and the course can read your Anki reviews back into its own progress.

The levels are licensed CC BY-NC-SA 4.0, the PULS authors' choice, and that license travels with anything built on them. So the site is non-commercial by construction, with no ads and no paid tier. A test scans the site's code for ad and payment hooks and fails if one appears. It would have been free anyway; the license makes that a rule rather than a promise.

uavocab.org tells you which words matter most and drills them. It will not hold a conversation with you or explain why a case ending changed. If you can get a teacher or a course with speaking practice, start there and use this alongside it.

Three limits are worth knowing. The levels are a snapshot the PULS authors are still revising. The stress marks are still waiting on their native-speaker check. And recordings thin out as words get rarer: every card in the first-100 deck has one, but under a third of the C1 deck does.

Share cards for the level and guide pages, which only word pages have today. A Ukrainian-language edition of the site, for which the course's own Ukrainian strings are already written. And more word pages, once the first 3,000 are being read. If you are learning Ukrainian, start with the first 100 words; if you build tools for learners, the study's downloads are open.