About

A dictionary and grammar of Karachay, the Turkic language of the Karachay people of the Caucasus, built around an engine that generates every form of a word rather than storing them.

Why this exists

Karachay has good printed dictionaries, and they are hard to use. They are out of print, they are in Russian, they disagree with each other, and none of them shows a learner what a word does when it is put in a sentence. We wanted one place where a word can be looked up in any spelling and in any of its forms, and where what comes back can be trusted.

The aim is the best Karachay–English dictionary there is: accurate enough that a speaker trusts it, complete enough that a learner does not hit a wall.

Who it is for

Karachay speakers who want to check a word or read one in its own language; learners and heritage speakers working out why an ending changed; and readers of Karachay texts, songs and poetry tracing a form on the page back to its word. The English is the working language of the definitions, and the Karachay is always shown in the script the reader has chosen.

How a word gets here

Nothing is published because a dictionary says so. The printed sources are evidence to weigh against each other: they disagree, they carry Soviet-era edits, and the scanned ones carry scanning damage. A proposal is triangulated across them, checked against a corpus of Karachay literature, and put in a review queue. It becomes an entry only when Davud, a native speaker, accepts it. Where the evidence is thin, the entry waits. A held word costs nothing; a wrong one stays on the site until somebody notices.

Today the site holds 9,090 entries drawing on 24 cited works, listed on the sources page. Definitions are our own English, written from those sources; their wording is not reproduced.

What the machine does, and what it does not

The form tables are generated. One engine reads a word's Cyrillic spelling, applies the sound and suffix rules, and builds every conjugation, case and possessive; the search reads the same output, which is why a form met in a real text leads back to its entry. That work is rule-following, and it is tested against a golden master of every form the engine can build.

Automated tools also read the sources, propose senses and rank drafts by how often the corpus uses them. None of that publishes anything. Meaning is where machines are least reliable, so nothing publishes until a speaker accepts it. Where a form is generated but unattested, or a word's stem is unsettled, the entry says so rather than printing a confident guess.

Helping

Corrections are the most useful thing you can send: a wrong definition, a form no one says, a missing sense, a word we do not have. Every entry page has a link for telling us about that entry, and it feeds the same review queue the rest of the work goes through. Attestations help too — where you heard a word, who says it, and what they meant by it.

Using the data, and citing it

The definitions and the generated tables are ours to share, and you are welcome to quote, teach from, and link to them. The source dictionaries are not ours: they are cited, never reproduced. If you use the site in published work, please cite it, and please name the data version — the dictionary changes as entries are reviewed.

Karachay dictionary (karachay.com), data version 0c59a01d, retrieved 16 September 2026.

The conventions the entries use are explained on abbreviations and conventions, and the spelling on spelling and scripts.