Method

საათში is not სააცჰი

The international standard for writing Georgian in Latin letters promises that the Georgian can always be recovered. For two letter sequences, tsh and dzh, it cannot, and the obvious way to read them back gets everyday words wrong. What the gap is, who it catches, and how we think software should read them

13 November 2026 7 min read Stephen Masters

Georgian says “in” with a suffix, rather than a separate word. საათი (saati) is an hour, and საათში (saatshi) is “in an hour”. საფრანგეთი (saprangeti) is France, and საფრანგეთში (saprangetshi) is “in France”. The suffix is -ში (-shi), the noun drops its final ი (i) to take it, and so any noun whose stem ends in თ (t) puts a თ directly before a შ (sh). That is a lot of nouns: the names of many countries end in -ეთი (-eti), for a start.

The Latin spellings here, and throughout, are those of the current international standard, ISO 9984:2026. Its introduction says that “the system is entirely reversible”: from the Latin, the Georgian can always be recovered. Try it with saatshi. The letters tsh can be cut as t and sh, which gives back საათში, or as ts and h. The standard does not say which, and the cut a program reaches for first gives სააცჰი: also saatshi in Latin letters, and not a word.

Anyone who can read and write Georgian will almost certainly pick the correct Georgian word based on the context of a sentence, but for any automated transliteration or for a cataloguer who is not fluent, this task is not so obvious.

We described this gap, among much else, in our account of the 2026 standards. This is a more focussed story for anyone with romanised Georgian in hand, and a need to get the correct script back.

How the standard keeps letters apart

ISO 9984:2026 is built on the romanisation Georgia itself uses on road signs and in passports, which spells several single Georgian letters with two Latin ones: შ is sh, ჟ is zh, ც is ts, ძ is dz, ხ is kh and ღ is gh.

That is comfortable to read and impossible to reverse, because the same two Latin letters can also be two Georgian letters that happen to sit side by side. So the standard adds one device: a middle dot (·, U+00B7) between two letters that would otherwise be read as one.

Georgian	Written	Without the dot it would read as
`ზ` then `ჰ`	z·h	zh, the letter `ჟ`
`თ` then `ს`	t·s	ts, the letter `ც`
`დ` then `ზ`	d·z	dz, the letter `ძ`

The standard lists the pairs that take the dot, six in all, and says of the dot that “this avoids any ambiguity and ensures the reversibility of the system”.

The two sequences the list misses

Every pair on the list is two single letters that together look like a two-letter spelling. The list has nothing for the case where three Latin letters can be cut in two places:

Latin	Cut after the first letter	Cut after the second
tsh	t + sh: `თ` then `შ`	ts + h: `ც` then `ჰ`
dzh	d + zh: `დ` then `ჟ`	dz + h: `ძ` then `ჰ`

None of those four pairs is on the list, so none gets a dot, and tsh and dzh each stand for two different spellings. We checked every sequence of two and of three Georgian letters, and these are the only ones that fail to read back.

We reported it through our national standards body. The reply agreed that both sequences should have been on the list, and said the list will be corrected when the standard is next revised. Until then, the standard as published is what conforming software writes.

Why it is not a corner case

For dzh it very nearly is one. For tsh it is not, because of that ending.

თ then შ is ordinary Georgian. Besides საათში and საფრანგეთში there is წუთში, in a minute, and it turns up inside compounds too: ხუთშაბათი is Thursday.
ც then ჰ appears only in borrowed names, most of them German: ალცჰაიმერი (Alzheimer), ერცჰერცოგი (archduke, from Erzherzog), პფორცჰაიმი (Pforzheim).
დ then ჟ is rare, and found in foreign names spelled out letter by letter from another alphabet. ძ then ჰ we could not find in use at all.

We counted all four in the titles of the Georgian Wikipedia, and the table is in the longer article. Words with თ then შ outnumber words with ც then ჰ several times over, and most of them are a noun with -ში.

Two kinds of reader are caught.

Someone who does not read Georgian. A cataloguer or researcher holding saprangetshi has no way to tell which letters were meant, and the standard they are relying on says there is nothing to decide.

Software. The usual way to read a romanisation back is to take the longest spelling that matches at each position. At the t of saatshi that rule prefers ts to t, takes ც, and is left with an h: ჰ. So the obvious implementation picks the reading that only borrowed names have, and gets the everyday word wrong every time. Ours did, until version 4.5.0.

Looking at the neighbouring letters does not rescue it: khutshabati (Thursday) and altshaimeri (Alzheimer) both continue tsha.

When writing, follow the standard as published. Don’t add dots of your own to tsh or dzh. Your output would differ from every other conforming implementation’s, and nobody reading it could know which rule you had applied.

When reading back, take the reading Georgian words have. Read tsh as თ then შ, and dzh as დ then ჟ. Every noun with -ში then comes back right. The cost is the borrowed names: altshaimeri comes back as ალთშაიმერი, not ალცჰაიმერი. That is a real loss, and we think it is the right one to accept. A wrong Alzheimer is one name in a text; a wrong -ში is every other sentence.

Let a person settle it with a dot. A reader that skips the middle dot wherever it stands can be told what was meant: alts·haimeri reads back as ალცჰაიმერი. That gives an editor a way to put right the few names that matter without changing any rule.

Keep the Georgian if you have it. A romanisation is for exchange and for people who cannot read the script. Where the original spelling exists, store it, and treat the Latin form as derived.

For finding, none of this matters. If you match on a folded key and not on the reconstructed script, both readings share a key, and altshaimeri still finds ალცჰაიმერი.

This is what tetrak-translit does from version 4.5.0:

from tetrak_translit import to_script, transliterate

transliterate("საათში", "ka/iso_9984")       # 'saatshi'
to_script("saatshi", "ka/iso_9984")           # 'საათში'

transliterate("ალცჰაიმერი", "ka/iso_9984")   # 'altshaimeri'
to_script("altshaimeri", "ka/iso_9984")       # 'ალთშაიმერი', not the original
to_script("alts·haimeri", "ka/iso_9984")      # 'ალცჰაიმერი'

When the correction comes

Each sequence needs a dot in only one of its two pairs, and the next revision of ISO 9984 will have to choose which. We hope it dots the borrowed names (ts·h, dz·h): ordinary text then stays exactly as the national system writes it, and the rule above for reading back remains correct.

If it dots the common pair instead, so that საათში is written saat·shi, an undotted tsh will from then on mean ც then ჰ, and the default has to flip. Text written under the 2026 edition will still be ambiguous, and nothing in the text says which edition wrote it. If you keep romanised Georgian for the long term, record the edition of the standard it was made under, beside the data.