What Georgians actually say

If you have ever tried to learn Georgian vocabulary in a sensible order, you have used a frequency list. Maybe not knowingly. It is underneath the Anki decks, the apps, the “500 most common words” posts.

Here is the thing nobody mentions: every one of those lists is built from written Georgian. News articles. Wikipedia. Parliamentary reports. That is the text that exists in bulk, so that is what gets counted.

Which means the most common Georgian words, according to the standard lists, include op’ozitsia (opposition) at number 130 and saarchevno (electoral) at 145.

Meanwhile shemidzlia, “I can”, sits at number 9,599.

I have been learning Georgian since 2024, entirely through the courses we built. I use “I can” constantly. I have never once needed to say “electoral”.

So we built the list that should have existed. It is free, it has 10,000 words in it, and you can read it or download it here.

What we actually counted

Nobody has recorded and transcribed spoken Georgian at any real scale. That corpus does not exist. So we took the two registers that sit closest to it:

Film and TV dialogue. Subtitles are written to be spoken aloud. They are full of the words people actually use on each other: yes, no, wait, I think, come on.

Conversational forum writing. Not blog posts, not articles. People arguing about football, asking for advice, telling stories. It is typed, but it is talk.

That gave us 68 films of dialogue and 184,209 forum posts, around 5.7 million words of Georgian in total. Then we counted, lemmatized (so ts’ignebshi and ts’ignis both count toward ts’igni, book), and ranked.

The result is the 10,000 most common words in spoken Georgian, free, under a CC BY license. Take it, use it, build on it.

What Georgians actually say

Here is the gap. Left column: where a word ranks in speech. Right column: where the standard written list puts it.

Look at that right-hand column. Nine thousand five hundred. Nine thousand six hundred. These words are sitting at the very bottom of the written list, which is another way of saying: if you learn Georgian from a standard frequency list, you will learn nine thousand other words before you learn how to say “probably”.

And it runs the other way too. Here are words the written list puts near the top:

That is not a language. That is a newspaper.

None of this is anyone’s fault, exactly. It is just what happens when you count the text that happens to be lying around and call the result “the most common Georgian words”. The words are common. In newspapers.

The bit where it gets worse

While building this, we went to check our numbers against the existing standard. There is one Georgian frequency list that circulates everywhere, in Anki decks and language apps and academic papers. It is derived from a public subtitle corpus.

Its top-ranked Georgian words are not Georgian.

They are Bulgarian. At some point in the pipeline, Bulgarian subtitle text got read through the wrong character encoding, and out came letter sequences that look Georgian and mean absolutely nothing. Nobody checked.

You can prove it. The mangling is a straight letter-for-letter substitution, so you can solve it like a cipher, from the data, without knowing the key in advance. We did. Here is what the top of the “Georgian” list actually says:

The top nine entries on the most-copied Georgian word list are the top nine Bulgarian words, in order. Of its top 200 “Georgian” words, 184 are real Bulgarian words.

It is not a Georgian frequency list with some errors in it. It is the Bulgarian frequency list wearing Georgian letters.

Then we went to the source corpus to see how deep it went. Of the 351 Georgian subtitle files in it, 282 are not Georgian. Eighty percent of that archive is mangled Bulgarian.

This has been copied into decks and apps and study lists for years. If you have ever drilled a Georgian “most common words” deck, there is a decent chance you have memorized Bulgarian.

We threw out all 282 files and kept the 68 real ones. That is why our dialogue corpus is smaller than it might have been, and I would take 68 real films over 351 files where four out of five are garbage.

What surprised me

The particles. Georgian is full of little words that carry the entire mood of a sentence, and written lists throw them out on principle because they are not “content words”.

Kho (right?), nu (don’t, or a hesitation noise), aba (well then), ho (yeah), kholme (the habitual marker, “usually”), metki (the quotative, “I said”). Khoda, which is roughly “and so”, and which every Georgian I know uses about once a minute.

There are 230 of these in our top 10,000, and they are some of the most useful words on the list. They are how you sound like a person instead of a textbook. We kept every one.

The other thing: ro. It is the lazy spoken form of rom (that), the way people actually type it and say it, and on the forum it is the fourteenth most common word in the language. Fourteenth. It is not in any textbook I have seen.

What we found in our own work

We checked our new list against our existing flashcard decks, which were built from the written frequency data like everyone else’s.

We are missing 4,804 of these words. Not obscure ones. Maints (still), aghar (not anymore), albat (probably), kholme (usually), khoda (and so).

That is a real gap in our own product, and we found it by building this. We are going to fix it.

Take the list

The full list lives here. Free, CC BY 4.0. Top 10,000 words with English glosses, transliteration, part of speech, and separate frequencies for the two registers so you can weight them yourself. There is a full methodology too, including a section on the mistakes we made and how we caught them, because we just spent a section of this post criticizing someone else’s data and it would be pretty rich to hide our own.

If you are learning Georgian: start at the top and work down. The first few hundred words on this list will get you further in a conversation than the first few thousand on any other.

If you build language tools: take it, it is yours. Just link back.