Frequently asked questions

Govori is a family of modern learner's dictionaries. It's built around the things paper dictionaries never had room for and most apps don't bother with: complete grammar — every case, every tense, with the accent marked — search that recognises the inflected form you actually heard, and entries you can read on a phone in a shop queue.

It's fully offline: the whole dictionary ships inside the app, so lookups are instant and work in a basement, on a plane, or with a dead SIM. No ads and no account. Anonymous statistics are collected, searches included, but there's no account to attach them to — so what they add up to is trends, not people. The privacy policy says exactly what's gathered.

Anyone who learns the language or has to live in it. Expats getting through paperwork, students, immigrants, business travellers, heritage speakers who grew up hearing the language but never studied its grammar, translators, teachers.

It's built for both directions of the problem: understanding what you read or heard, and producing a form correctly yourself — which is where the grammar tables earn their place.

You don't have to. But if you use the language in real life, or you're learning it, you might want to: full grammar for every entry, search that forgives diacritics, scripts and word forms, no ads, and it all works offline.

The honest test: look up a word you struggled with this week. If the entry tells you something you couldn't get in ten seconds elsewhere, the app is for you.

I lived in Serbia as an immigrant. I was fascinated by the language, and frustrated by how little modern support it had. Wonderful sources exist — Wiktionary, the standard reference dictionaries, PDFs that circulate among learners — but using them meant juggling five tabs and a scanned book, and none of them was built for the moment you actually need a word.

At some point I got tired of that and decided to build a dictionary with a genuinely good experience for the way we live now: offline, fast, smart about search, complete on grammar, free of ads.

I believe that knowing the local language is the key to belonging anywhere in the world. This is my attempt to close part of that gap — and a small homage to the languages and dialects of the Balkans.

Govori is a Serbian word you meet on day one. It's both the imperative — "speak!" — and the third person: "he speaks", "she speaks". A dictionary should be an invitation, not a monument, so that seemed right.

The Slovene app is called Lahko — "easily", and also the word that turns a verb into "you can": lahko govoriš, "you can speak". Same idea, other language.

Hi — I'm Sergei, an immigrant from Russia. I spent two years in Serbia learning the language, and I now live in Slovenia, where I'm building the Slovene app for exactly the same reason.

I'm a software engineer, not a trained lexicographer, and that shapes how the dictionary is made: I don't invent language facts. Everything in it traces back to a published reference dictionary or a documented source, and the app can tell you which one.

In its early days Govori was meant to be a nice interface over Wiktionary — a free, genuinely impressive dictionary that already covers a lot of the language. I learned quickly that this doesn't work. Wiktionary's coverage is uneven: strong in places, thin exactly where everyday words live, and often silent about which regional standard a form belongs to. It's a great skeleton and a poor final answer.

So Govori uses its own approach to choosing words, writing entries and checking them. Wiktionary is still the starting skeleton, it's credited in the app, and its licence travels with it — but the dictionary is no longer a copy of it. Everything that matters is verified against published reference dictionaries and language portals: Matica srpska and Hrvatski jezični portal for BCMS, Fran and the Sloleks database for Slovene, among others.

Two steps: decide which words to add, then write and verify them.

The word list comes from three places:

1. Frequency rankings from language corpora — the words people actually use, most-used first. This is what keeps the project finite.
2. Simulated real-life situations — going to a doctor, renting a flat, registering a residence. These surface words that corpora under-represent but learners need immediately.
3. You — what people search for and don't find, plus direct reports when someone knows a word exists and it isn't there yet.

Then the writing. This is where language models do the heavy lifting: reviewing existing entries, drafting new ones, checking the accents and paradigm cells. They are not allowed to answer from memory. Every claim has to be transcribed from a named reference that actually prints it, and each entry stores a record of what was checked — including the references that were consulted and did not have the word. A claim with nothing behind it never reaches the app.

It isn't magic and it isn't perfect. It is, however, checkable — which is more than I could say for my own guesses.

Each standard is judged only by its own authorities, and none of them outranks another. For each one there's an ordered list of reference works: the first book that actually prints the thing in question settles it, and a book that simply doesn't have the word counts as silence rather than a verdict — that absence is recorded too, and the question moves down the list.

Which list applies depends on what's being decided. Form, accent and paradigms are separate questions with separate authorities:

• Serbian — on form, Matica srpska's Pravopis srpskoga jezika, then Šipka's Pravopisni rečnik (2010), then Matica's Rečnik srpskoga jezika (2011). On accent the Pravopis drops out of the list entirely: it prints no accent marks at all, so accent goes to Šipka first and Matica after.
• Croatian — on form, the Institute's Hrvatski pravopis (2013), then the Hrvatski jezični portal. On accent HJP stands alone, because that Pravopis covers orthography only.
• Montenegrin — Pravopis crnogorskoga jezika (2009), then Gramatika crnogorskoga jezika (2010).
• Bosnian — Halilović's Pravopis bosanskoga jezika (2017).

Meanings work a little differently. The big general dictionaries lead — Matica for Serbian, HJP for Croatian, which is also the richest source of definitions there is — and the layers those two leave out on purpose are covered by specialist works: a dictionary of Serbian slang, Škaljić's study of the Turkish, Arabic and Persian loan stratum, a lexicon of colloquial Sarajevo speech. Paradigm questions no dictionary answers go to the grammars, such as Piper and Klajn's Normativna gramatika.

The most telling part is what happens at the edges. Neither the Montenegrin nor the Bosnian codification marks accent — Halilović states outright that it's outside his scope — so for those two standards the dictionary claims no accent at all. It would be easy to quietly borrow the Serbian or Croatian mark and nobody would notice; instead the app says nothing, because an honest gap is worth more than a confident guess.

And when two standards disagree, that isn't a conflict to resolve. It's two answers, each correct in its own norm, and both are kept and labelled with whose it is.

It isn't a replacement for any of them. It's built on top of them, links out to them, and would be far worse without them.

They're different kinds of object:

• Matica srpska and the other standard reference dictionaries are the authority on the Serbian norm. They're also books: you can't search them by an inflected form you heard on the street, and they don't hand you a full conjugation on demand.
• Hrvatski jezični portal is an excellent free portal for the Croatian norm — online, Croatian-facing, and not built for someone whose starting language is English or Russian.
• The classic bilingual dictionaries (Stanković, Tolstoy and others) are superb works of scholarship, frozen at the year of their edition and shaped by paper.
• Most dictionary apps are word lists with a translation and an ad banner.

Govori's specific job is to be the learner's layer: every form, accented, in one place, searchable by whatever you actually heard, offline, with the sources named — and where the standards disagree, saying who says what instead of silently picking one.

Out of the argument, deliberately. I'm a guest in these languages, and it isn't my quarrel to adjudicate.

What I do instead is practical. There's one shared body of evidence behind the BCMS apps, and each individual form records which standards prescribe it, merely tolerate it, or reject it — judged against that standard's own authority, not my taste. Separate apps exist so that the standard you use is the default you see: someone in Podgorica shouldn't have to mentally filter Croatian endings out of every table, and someone in Zagreb shouldn't get Belgrade's.

Nothing is deleted for belonging to a neighbour. Where the standards diverge, the other form stays visible and labelled, because a learner deserves to know that a choice exists — that's information, not politics.

A lot of it is, permanently: search, meanings in English and Russian, usage examples, favourites and history. No ads anywhere in it. Searches are reported anonymously — that's how the "people looked for this and found nothing" list gets built — but with no account, no login and no profile, what comes back is trends rather than anyone's reading history.

What's paid is the depth — the grammar tables. Those are the part that takes years: a paradigm isn't scraped, it's verified cell by cell against reference works, and thousands of them have been corrected by hand.

I have no investors, I don't sell data, and I refuse to sell your attention to an ad network. That leaves one honest option: the people who get the most out of the dictionary pay for it, and their money pays for the next ten thousand words.

Everything is unlocked for 14 days after you install — the full app, no card, no account. After that, lookups, meanings and examples stay free forever, and the grammar tables become part of Govori Plus.

Plus comes in three shapes, and the app shows the current price in your own currency:

• Pečenje — one payment, yours forever. The best value and the one I'd pick.
• Sarma — a yearly subscription.
• Kafa — monthly, cancel whenever you like.

The food names are decorative; every option always shows exactly what it bills and when. No manipulation, no nagging reminders, and no push to start a subscription — the kind you later forget to cancel. The dictionary is deliberately built so that the free version and the paid one are both a fully working tool. And one purchase covers everything that ships afterwards.

Honestly, I'd rather sell you the app once — that's why the one-time purchase leads the list: over any real span of time it's the best value.

But not everyone is sure they'll stay with Serbian for long, or that the dictionary will turn out to be that useful to them. Paying for "forever" up front makes little sense then: a month or a year costs considerably less than the one-time purchase, and it's an unhurried way to find out whether this is your tool. If it turns out to be, the one-time purchase isn't going anywhere.

There's no tip jar. If you want to support the project, buy the one-time purchase — that is the donation, and it gets you the whole dictionary. If you already own it, telling one person who's learning the language does more for it than money.

All of it is planned, and I won't promise dates.

Audio is the most requested feature by a wide margin, and the design is done: pronunciation for core vocabulary bundled with the app for everyone, fuller coverage and slower playback as part of Plus.

Learning is coming as word lists you build yourself — save words from any entry, then practise them. Practising will stay free; that's not a thing I want to put behind a paywall. Alongside it, themed vocabulary for real situations, built from the same "what do you actually need at the bank" thinking that shapes the word list.

Case government — which case a verb takes, and with which preposition: bojati se + genitive, brinuti o + locative. This is the most common way a sentence comes out wrong while every single word in it is right, and settling it is exactly a dictionary's job. Verbs come first, and the English and Russian sides ship together. Prepositions already show their case in the iPhone app; for verbs the information is in place for roughly 260 of some 6,800, so what's left is mostly coverage — verb by verb against the Hrvatski jezični portal, Matica's dictionary, the Croatian valency database e-Glava, and Petrović and Dudić's dictionary of verb government.

English is the language most learners arrive with, so it was never a question.

Russian is there for two reasons. The honest one: it's my own first language, and I was building the dictionary I needed. The better one: Russian speakers are one of the largest groups of newcomers to Serbia and Montenegro, and there was nothing modern serving them.

There's also a linguistic reason that turned out to matter more than I expected. Russian is not a second-rate English here. It's a Slavic language, so it already has the categories Serbian needs — verbal aspect, cases that govern each other, reflexive verbs, participles. A Russian gloss can name a perfective pair in one word where English needs a whole clause and still loses the distinction. If you speak a Slavic language, that's a shortcut English simply can't give you.

Which is why the Russian side isn't a translation of the English side. It's written independently, against the Serbian word itself, and the two sometimes divide a word's meanings differently — because sometimes they genuinely should.

Next up: French is first in line, driven mostly by heritage speakers in France who'd rather learn through their own language than through English. German and Italian are the ones asked for after that. The plan is that you'll download only the language you use, so adding languages never inflates anyone else's app.

Nothing here is promised on a date, and the order isn't fixed. If you want a particular language, tell me — that's literally how the list gets sorted.

Slovene already exists as Lahko — a separate app with its own dictionary, built from Slovene sources rather than adapted from Serbian. Montenegrin, Croatian and Bosnian editions of Govori are in preparation, each defaulting to its own standard for the same reasons as above.

Beyond that I make no promises. Every language in this family is a year of somebody's life, and I'd rather do four properly than twelve badly.

Yes, please, and the most useful help is the smallest:

• Report a missing word or a wrong form from inside the app. These go straight into the work queue, and "I'm sure this word exists" reports are genuinely valuable.
• Join the Telegram group to argue with me about accents and hear what's coming.
• Teachers, translators, lexicographers, corpus people: write to support@govoridictionaries.com. If you teach the language and want it in your classroom, tell me — I'd like to make that easy.
• Leave a review. For an app with no marketing budget, that's oxygen.