Verbarie
Reading in French, on any e-reader.
Verbarie builds bilingual dictionaries for Kindle, Kobo, KOReader and Apple Dictionary. Tap any word in a French book — in the form it takes in the sentence — and get a real dictionary entry instead of a single word thrown at you.
French ↔ Russian is finished, in both directions. Sales open soon.
Buying
The dictionaries go on sale shortly.
I will announce the price and the date by email. The last proofing pass is still running: a dictionary this size should not ship before the checks agree, and I would rather not promise a date I cannot keep.
Leave your address and I will write once, when sales open. No newsletter, nothing else.
What you pay for
- — finished files for your e-reader, not raw data;
- — every format, and both directions of the pair;
- — updates as corrections land;
- — help with installation if something will not load.
What stays free
The dictionary is built from openly licensed sources, and the data itself stays under CC BY-SA 4.0. That means the file you buy is yours to copy and pass on: I am not selling the right to read it, I am selling the assembled, checked and packaged result. Blender, Ardour and O’Reilly books are sold the same way.
Why
The goal is access, not another dictionary.
A reader in a foreign language stops in the same place every time: an unknown word. Whether that stop lasts two seconds or ends the book depends entirely on what happens when you tap it.
For English, that problem was solved long ago. For most other pairs it was not. Good bilingual dictionaries exist, but they are locked behind a price, a device, a publisher, or a licence that forbids anyone from improving them — and for a great many language pairs they simply do not exist in a form an e-reader can use.
Verbarie is an attempt to close that gap for as many readers as possible, one pair at a time. French is always one of the two languages: it is the language I know well enough to be exacting about, and it is the second language of tens of millions of readers who have no decent dictionary into their own.
Every pair is built in both directions, so the same work serves two readers at once: someone reading French, and a French speaker reading the other language.
The problem
What usually goes wrong when you tap a word
The entry is a stub
Most e-reader dictionaries return a single translation with no part of speech, no gender, no register. For a polysemous word that is not help — it is a wrong answer with good manners.
The word is not in the form you found it
Books contain « eûmes » and «городами», not « avoir » and «город». A dictionary that only indexes lemmas fails on most of a running text.
It only works one way
A French–Russian dictionary rarely comes with a usable Russian–French one, so half the readers of the same pair are left out.
You cannot fix it
A closed dictionary with an error stays wrong forever. Here, anyone who finds a bad gloss can write in — and it gets fixed in the next release.
What you get
One tap, one full entry
n. m. · n. f.
Senses
- 1.n. m.книга; том
- 2.n. m.текст, произведение
- 3.n. f.фунт (0,5 кг)
- 4.n. f.фунт (денежная единица)
Usage example
Il a écrit un livre sur Paris.
Он написал книгу о Париже.
Set phrase
à livre ouvert
с листа, без подготовки
Inflected forms found in text
livreslivre
A shortened example of what appears on the device. Senses are grouped by part of speech, nouns carry gender, Russian headwords carry stress, and inflected forms resolve to the lemma.
Inside an entry
Everything a printed dictionary gives you, on the device
Senses grouped by part of speech
Noun senses under the noun, verb senses under the verb — not a flat list of comma-separated words.
Gender and grammar
Gender on French nouns; case, number and aspect information on the Russian side.
Conjugation and declension
A compact conjugation table for French verbs; genitive and plural for Russian nouns.
Pronunciation and stress
IPA on French headwords, stress marks on Russian ones — the thing a learner cannot guess.
Usage examples
A translated example sentence for most entries, so a sense is anchored to real use.
Idioms and set phrases
Expressions that mean nothing when translated word by word are given their own lines.
Inflected-form lookup
Hundreds of thousands of forms are indexed, so tapping a conjugated or declined word lands on the right entry.
Every major format
Kindle (MOBI), Kobo, KOReader (StarDict), Apple Dictionary and EPUB, built from one source.
Figures
The French ↔ Russian pair today
French → Russian
- 99 000
- headwords
- 460 000
- recognised word forms
- 98.1 %
- with IPA transcription
- 90.4 %
- with a usage example
- 9 713
- entries with two or more senses
Russian → French
- 147 000
- headwords
- 2 700 000
- recognised word forms
- 91.3 %
- with stress marks
- 91.9 %
- with a usage example
- 13 018
- entries with two or more senses
92.7 % of running text resolves to an entry
Frequency-weighted coverage of running French text, measured over the 50 000 most frequent word types.
Figures from the 2026-07 build.
Both directions
One build, two readers
A bilingual dictionary is usually shipped one way, because the reverse direction is a second project: you cannot simply flip the arrows and get a usable result — you need the other language’s morphology, its stress or tone, its own examples.
Verbarie treats the reverse direction as part of the same build. The French ↔ Russian pair ships as two dictionaries: one for a Russian speaker reading Balzac, one for a French speaker reading Chekhov, each with the grammar apparatus of the language being looked up.
That is why the next pairs matter beyond their own readers. A French ↔ Japanese dictionary is not only for francophones learning Japanese — it is also, and mainly, for Japanese readers who want to open a French novel.
Devices
Where it works
| Device | File | How it installs |
|---|---|---|
| Kindle | MOBI | Send to Kindle, then pick it as the dictionary for French or Russian. |
| Kobo | dicthtml zip | Drop into .kobo/custom-dict/ and restart. |
| KOReader | StarDict | Unpack into koreader/data/dict/ — works on any device running KOReader. |
| Apple Books / macOS | Apple Dictionary | Installs into the system Dictionary app. |
| Anything else | EPUB | A readable dictionary you can open like a book. |
Languages
French is fixed. The other side is open.
Every pair keeps French on one side and adds one language on the other. The pairs I most want to build next are Japanese, Chinese and Korean — three languages with enormous numbers of French learners and, in the other direction, very little for a reader who wants to open a French novel. But I am open to any language: if you want a pair that is not on this list, ask for it.
- French ↔ RussianOn sale soon
- French ↔ JapaneseLooking for volunteers
- French ↔ ChineseLooking for volunteers
- French ↔ KoreanLooking for volunteers
- Your languageAsk for it
Volunteers
I am looking for people to check new dictionaries.
The pipeline can produce a draft dictionary for a new pair — headwords, forms, grammar, machine-generated glosses — reasonably fast. What it cannot do is tell me whether those glosses are actually right in your language. That needs a person who speaks it.
You do not need to be a lexicographer or a linguist, and you do not need perfect French. You need to be a confident speaker of the other language, and honest when you are not sure — “I don’t know” is a useful answer here, a confidently wrong one is not.
Straight talk about money
The dictionaries are sold, and asking for unpaid work on a paid product without saying so would not be honest. So, out loud: anyone who checks a pair gets that pair free, for good, with every update, plus beta access and their name in the release. If you put serious hours into a pair, write to me and we will agree on payment separately. What the sales bring in is what makes the next pair happen at all.
What the work actually looks like
- 1You get a batch of entries — a word, its part of speech, and the proposed translations — as a plain table.
- 2You mark each one: right, wrong, or unsure. If it is wrong, you write what it should be.
- 3Batches are small on purpose. An hour of attention is worth more than a promised weekend that never comes.
- 4Anything marked unsure goes into a queue, not into the dictionary. Confidently wrong is worse than missing.
Especially wanted
- Japanese, Chinese and Korean speakers — my three target pairs.
- Speakers of any language who read in French and know what is missing from their own dictionaries.
- People who can point at open, freely licensed lexical data for their language: word lists, morphology, examples. Finding sources is half the work.
How it is made
Open sources, machine drafts, human checking
Every dictionary is built only from freely licensed data: Lexique for French frequency and word forms, Wiktionary for extra vocabulary, IPA and synonyms, Littré (1863, public domain) for literary and rare words, OpenCorpora for Russian morphology.
First glosses are generated with a language model, then put through an automated quality pipeline and corrected by hand where the checks fail or where a person reports an error. Translations are AI-generated and can be wrong; that is exactly why reporting an error is a first-class part of the project.