Dossier · reading the world's oldest writing
How to read cuneiform.
And what the tablets actually say.
It is not a language. It is not an alphabet. And it is not, for the most part, literature. Here is what three thousand years of wedges in clay really contain — counted across this corpus — and how to follow a real tablet, line by line.
Short answers first
What is cuneiform?
Cuneiform is a writing system, not a language. The name comes from the Latin cuneus, “wedge”: scribes pressed a cut reed stylus into wet clay, and every sign is built from combinations of those wedge-shaped impressions. It was used in Mesopotamia from roughly 3300 BCE until the first century CE — more than three thousand years, which makes it the longest-serving script humanity has ever used.
What language is cuneiform written in?
Many. Cuneiform is a script that different peoples borrowed for entirely unrelated languages, the way the Latin alphabet now writes English, Turkish and Vietnamese. It was used for Sumerian (a language isolate), Akkadian and its Babylonian and Assyrian dialects, and also Eblaite, Elamite, Hurrian, Hittite, Luwian, Urartian and Old Persian. Asking what language cuneiform is is like asking what language the alphabet is.
Is cuneiform an alphabet?
Mesopotamian cuneiform is not an alphabet. It is logo-syllabic: a sign can stand for a whole word (a logogram, like GU₄ for “ox”), for a syllable (like -tu), or for a silent category marker placed beside a word (a determinative). The repertoire runs to roughly six hundred signs, though a working scribe used a few hundred. Two late offshoots did become alphabetic — Ugaritic, a true alphabet written in wedges, and Old Persian, a much reduced semi-syllabary of about thirty-six signs.
What do cuneiform tablets actually say?
Overwhelmingly, they are accounts. Of the 76,923 tablets in this corpus that carry a genre classification, 46,852 — 60.9% — are administrative: receipts, ration lists, livestock counts, delivery records. Literary texts are 4,208 tablets, or 5.5%. Legal documents are 6.5%, letters 7.4%, royal inscriptions 7.8%. For every tablet of literature there are roughly eleven of bookkeeping. Writing was invented to count things, and for most of its history that is mainly what it did.
What is the oldest cuneiform tablet?
The earliest tablets, from Uruk in southern Iraq around 3300 BCE, are written in proto-cuneiform: numerals plus pictographic signs, recording quantities of grain, beer, livestock and labour. They are administrative, not narrative. Crucially, they do not yet write speech — the signs record things and amounts, not sentences, and the leap to recording actual language comes a few centuries later.
How was cuneiform deciphered?
Through Old Persian, the simplest of the cuneiform scripts. Georg Friedrich Grotefend guessed royal names in Persepolis inscriptions in 1802; Henry Rawlinson copied the trilingual Behistun inscription from a cliff face in Iran between 1835 and 1847, giving scholars an Old Persian text beside its Akkadian version. Edward Hincks worked out the syllabic and logographic structure. In 1857 the Royal Asiatic Society settled the doubts by sending the same unpublished Assyrian inscription to four scholars independently; their translations agreed, and Assyriology became a discipline.
Can you learn to read cuneiform today?
Yes, and the first step is transliteration rather than the wedges themselves. Assyriologists write each sign in Latin letters with strict conventions: lowercase for a Sumerian reading, CAPITALS when the intended reading is uncertain, subscript numbers to distinguish signs that sound alike, square brackets for what is broken away. Learning to read those conventions lets you follow the structure of a tablet — as the worked example on this page shows — long before you can recognise the signs on the clay.
Method.Every tablet below is a real record from this site's corpus, with its photograph, its source and a link to its page. The transliterations are quoted exactly as they appear on those pages — they come from CDLI and ORACC editions, not from us. Translations are attributed where they are scholarly; where our engine supplies the reading it is labelled as such and commits the engine, not the editors. The counts in section II were computed over the 102,246 catalogued tablets on this site on 1 September 2026.
I · the first misconception
Cuneiform is a script, not a language.
Almost every question that begins “what does cuneiform say” contains a category error. Cuneiform is not something you translate; it is something you read out of, the way you read Latin letters out of a page before knowing whether the language is Spanish or Swahili. The wedges are a delivery system. What they deliver changed repeatedly over three millennia.
The script was devised for Sumerian, a language with no known relatives anywhere on earth. Within a few centuries it had been adapted for Akkadian, a Semitic language distantly related to Hebrew and Arabic, and Akkadian scribes kept writing Sumerian words inside Akkadian sentences the way English keeps e.g. and etc. — fossils of a prestige language nobody spoke any more. Then it travelled: Eblaite in Syria, Elamite in Iran, Hurrian, Hittite and Luwian in Anatolia, Urartian in the Armenian highlands, Old Persian for the Achaemenid kings.
That last one matters for a reason we will come back to. Old Persian cuneiform is not the sprawling six-hundred-sign system of Babylonia but a stripped-down set of about thirty-six signs — and because it was simple, it was the crack through which the whole script was eventually broken open.
The span
The earliest tablets in this corpus are dated around 3300 BCE. The last datable cuneiform documents anywhere are astronomical almanacs written in Babylon in the first century CE. Between those two points lie more than three thousand years — a longer working life than the Latin alphabet has yet managed, and the script outlived every language it was invented to serve.
II · what is actually on the tablets
For every tablet of literature, about eleven of bookkeeping.
The popular image of cuneiform is Gilgamesh, creation myths, law codes on black stone. The reality of the excavated record is a filing cabinet. When we group the tablets in this corpus by the genre their cataloguers assigned them, one category swallows everything else.
Of the 102,246 catalogued tablets here, 76,923 carry a genre classification. Among those:
Administrative
Receipts, ration lists, livestock counts, deliveries, inventories.
46,852
60.9%
Royal / monumental
Building inscriptions, annals, dedications, boundary stones.
6,037
7.8%
Letters
Correspondence — merchants, officials, kings, families.
5,730
7.4%
Legal
Contracts, sales, loans, adoptions, court decisions.
4,981
6.5%
Literary
Epics, hymns, laments, wisdom texts, myths.
4,208
5.5%
Lexical
Word lists and sign lists — the scribal schoolbooks.
3,385
4.4%
Omen
Divination: livers, births, dreams, eclipses, oil on water.
2,491
3.2%
School
Exercise tablets — pupils copying, and getting it wrong.
1,336
1.7%
Mathematical
Tables, geometry, problem texts in base 60.
908
1.2%
Bars are scaled to the largest category. A further 23,948 tablets carry no genre in their catalogue record and 1,375 are marked “uncertain”; both are excluded from the percentages above rather than silently folded into one of the buckets.
Sixty-one percent accounting against five and a half percent literature is not an accident of what survived. Clay survives indiscriminately; it is what was written. Writing was invented by administrators to track commodities, and for most of its history that is overwhelmingly what it did. The epics are real, and they are extraordinary, but they sit on top of a mountain of receipts — and the receipts are why we can reconstruct wages, diets, harvest yields and the price of a donkey, which no epic would ever have told us.
III · the mechanics
One sign, three possible jobs.
A cuneiform sign can be doing any of three different things, and part of reading is working out which. It can be a logogram, standing for a whole word: the sign GU₄ means “ox”, full stop. It can be a syllabogram, standing for a sound with no meaning of its own — si, qu, ru, tu — so words can be spelled out phonetically. Or it can be a determinative: a silent classifier written beside a word to say what kind of thing it is, never pronounced, the ancient equivalent of a file-type extension.
The system is not tidy, and one line from a Neo-Assyrian hymn in this corpus shows all of it happening at once. The scribe writes the word for the great temple-tower of Uruk as É.si-qu-ru-tu — the logogram É (“house”) followed by the word siqqurratu, “ziggurat”, spelled out syllable by syllable. A picture-word and a sounded-out word, welded together in a single noun.
The same tablet writes Babylon as KÁ.DINGIR.RA.KI. Read the parts: KÁ is “gate”, DINGIR is “god”, RA marks the genitive, and KI is a determinative for place names that is never spoken aloud. Gate-of-the-god — which is exactly what the Akkadian name Bāb-ilim, and behind it our word Babylon, actually means. The etymology is not a scholarly reconstruction; it is sitting there in the spelling.
British Museum, London (P336796) — photo via CDLI
Psalm in Praise of Uruk
SAA 3, 009 · Nineveh · 7th century BCE · literary · high confidence
⸢UNUG⸣.KI ÁG KI.MIN ÁG É.AN.NA É ḫa-mu-ti-ía / KI.MIN ÁG É.si-qu-ru-tu É la-le-ía / KI.MIN ÁG KÁ.DINGIR.RA.KI É AD-ia!
“Uruk; I love ditto, I love Eanna, my nuptial chamber! Ditto; I love (its) ziggurat, house of my pleasure! Ditto; I love Babylon, my father's house!”
Translation: Livingstone, Court Poetry and Literary Miscellanea, SAA 3 (1989), via ORACC/SAAo — CC BY-SA 3.0. Note É.AN.NA, “house of heaven”, the temple of Inana at Uruk: another logogram compound hiding in plain sight.
Read the tabletIV · the notation
Before the wedges, learn the brackets.
Nobody learns cuneiform by starting with clay. They start with transliteration — the standardised way Assyriologists write each sign in Latin letters, preserving every uncertainty of the original. It looks forbidding and is almost entirely conventional. Nine rules cover most of what you will meet on this site.
gu4, u8
Lowercase roman: a Sumerian reading the editor is confident about.
SAL, ZATU678
Capitals: the sign is identified but its intended reading here is not — so the editor names the sign rather than guessing a word.
gu4 · sila4 · gesz2
Subscript numbers distinguish signs that sound alike. The 4 in gu4 is not pronounced; it says “the fourth sign read gu”, the way a dictionary numbers homonyms.
1(N14), 2(u@c)
A numeral, with the counting system it is written in named in brackets. Mesopotamia used different numeral signs for different commodities.
[...]
Broken away. The clay is missing and nothing can be read.
x
A sign is visible but unidentifiable.
sign#
The sign is damaged but still readable.
sign?
The editor is not certain of this reading.
{d}, .KI
A determinative — silent. {d} marks a divine name, KI a place name, giš a wooden object. Never pronounced, always informative.
Those conventions are why a tablet page on this site can look like code. They are not decoration: they let a reader see exactly where the evidence stops and the editor's judgement begins — which is the whole ethical point of publishing a transliteration beside a translation.
V · a worked example
Reading a livestock account from 2400 BCE.
Here is an ordinary tablet — the kind that makes up sixty-one percent of the corpus. It comes from Girsu in southern Iraq, it is about four and a half thousand years old, and it is a tally of animals. Read line by line, it is completely transparent, and the numbers check out in front of you.
State Hermitage Museum, St Petersburg (P221946) — photo via CDLI
Nik 1, 177 — a count of ewes, oxen and donkeys
Girsu (mod. Tello) · c. 2400 BCE · Early Dynastic · Sumerian
1
1(gesz'u@c) 1(gesz2@c) u8 sila4-bi-ta
600 + 60 = 660 ewes, excluding their lambs. geš’u is the sign for 600, geš₂ the sign for 60; u8 is “ewe”, sila4 “lamb”, and -bi-ta means “from/apart from them”.
2
2(u@c) 4(asz@c) gu4 ab2-ta
20 + 4 = 24 oxen, excluding the cows. gu4 is “ox”, ab2 “cow”.
3
1(u@c) 6(asz@c) ansze
10 + 6 = 16 donkeys.
4
gu2-bi
“Its total.”
5–6
na-e-a / lugal-mu
Two personal names — the men accountable. lugal means “king” and is a common element in Sumerian names; lugal-mu is “my king”.
7
du11-ga-na
“By his word” — the declaration under which the count stands.
8
dub-sar
“Scribe” — literally tablet-writer: dub is “tablet”, sar is “to write”.
9
mu-gi4-a
The scribe “returned/repeated” it — the record was verified or reissued.
Notice what just happened. You read a four-and-a-half-thousand-year-old document. The arithmetic in line 1 — a sign for 600 plus a sign for 60 — resolves to the 660 in the translation, and you can check it yourself. That is the ordinary experience of cuneiform: not revelation, but bookkeeping you can audit across forty-five centuries.
VI · before writing wrote speech
The oldest tablets are receipts that cannot form a sentence.
Push back another nine hundred years, to Uruk around 3300 BCE, and the script is not yet able to do what it did in Girsu. Proto-cuneiform records quantities and commodities — but it does not encode grammar. There are no verbs, no word order carrying meaning, no way to write “the king gave barley to the temple” as opposed to the reverse. It lists. That is all it can do.
Vorderasiatisches Museum, Berlin (P000813) — photo via CDLI
ATU 5, pl. 013, W 6710,a
Uruk (mod. Warka) · c. 3300 BCE · proto-cuneiform · administrative
1
1(N14) 3(N01) , ZATU678 ERIM~a
A quantity in the N-system numerals, a commodity sign not yet identified (ZATU678 is a catalogue number, not a word), and ERIM — male workers.
4
5(N01) , SAL
Five of SAL — the sign for “woman/female”. Five women, or five female animals; the tablet does not say, and cannot.
7
4(N14) 8(N01) , [no commodity]
A total, with nothing named. The scribe knew what was being counted; the tablet was never meant to tell us.
This is the genuinely strange fact at the origin of writing: the first system was not a way of recording language. It was a way of recording amounts, and it worked perfectly well without being able to say anything. Speech was fitted into it afterwards, when scribes began using signs for their sound rather than their sense — the rebus leap that turned a bookkeeping notation into a script that could carry poetry.
The corpus lets you stand at that boundary. The Kish tablet sits just before the line, with signs used systematically but not yet phonetically; the proto-cuneiform account tablet from Uruk IV is the mature form of the accounting system that came first.
VII · the schoolbooks
They learned it from lists — and the lists survived.
Three thousand three hundred and eighty-five tablets in this corpus are lexical: word lists and sign lists, copied by pupils and masters for two and a half millennia. They are the reason cuneiform can be read at all today. A civilisation that spends that much clay teaching itself its own writing system leaves behind, in effect, its own dictionaries.
Schøyen Collection, Oslo (P006209) — photo via CDLI
An archaic list of titles and offices
Uruk period · c. 3100 BCE · lexical · low confidence
1
[1(N01)] , [...] SUKKAL#
SUKKAL — vizier, minister. Damaged but readable.
2
1(N01) , GAL~a GA~a
GAL, “great”, followed by GA, milk or dairy — a title built on “chief of…”.
5
1(N01) , GAL~a TUR3~a
“Great” + livestock pen: the head of the cattle enclosure.
The structure is the point: an entry number and a compound headed by GAL, “great”, repeated down the column. This is a list of ranks, organised so it can be memorised. Lists like this were copied, almost unchanged, for a thousand years after the language they describe had shifted underneath them.
Read the tabletThe same schools produced the mathematics. A scribe who could compute the side of a field from its area was doing, in base 60, the work that made the accounts possible — and left the working on the tablet.
Hilprecht Collection, University of Jena (P020479) — photo via CDLI
TMH 05, 065 — a field problem
Nippur · c. 2400 BCE · mathematical · high confidence
“67½ nindan, the long side (length); the short side (width): (so that the area is) 1 iku of field. (Result:) 1 nindan, 5 cubits, 2 šu-du-a (spans), 3 fingers and ⅓ finger.”
A length is given, a target area is fixed, and the required width is computed — down to fractions of a finger. The units descend nindan → cubit → span → finger, and the arithmetic is sexagesimal: the base-60 system that still divides our hours and our circles.
Read the tabletVIII · the breaking of the code
A cliff in Iran, and four scholars in a sealed room.
For most of recorded history nobody could read any of this. The script died with the last Babylonian astronomers, and the wedges became decoration on the ruins of a forgotten world. Recovery took the better part of a century and turned on the simplest version of the script rather than the richest.
In 1802 Georg Friedrich Grotefend, a schoolteacher in Göttingen, reasoned his way into Old Persian inscriptions from Persepolis by guessing that certain repeated groups were royal names and titles — X, great king, son of Y. He was partly right, which was enough. The decisive material came from Behistun in western Iran, where Darius I had cut a vast trilingual inscription into a cliff face six dozen metres above the road. Between 1835 and 1847 Henry Rawlinson had himself lowered down it, repeatedly, to copy the text. It gave scholars the same royal proclamation in Old Persian, Elamite and Akkadian — a Rosetta stone for Mesopotamia.
Old Persian fell first because it is nearly alphabetic. Akkadian was far harder: Edward Hincks established that signs could stand for syllables and for whole words at once, and that a single sign might have several values — the untidiness described in section III, which looked at the time like evidence that the decipherment was wishful thinking.
So in 1857 the Royal Asiatic Society ran an experiment. It sent a newly excavated, unpublished inscription of Tiglath-Pileser I to four scholars — Rawlinson, Hincks, William Henry Fox Talbot and Jules Oppert — who worked separately and returned sealed translations. The four agreed. Assyriology stopped being a controversy and became a field, and every translation on this site descends from that afternoon.
IX · where to start
Read one tablet properly rather than a hundred badly.
The catalogue on this site holds 102,246 tablets, 96,685 of them with a photograph of the original and 34,884 with a transliteration you can follow sign by sign. That is far too much to browse and exactly enough to read one thing carefully. Three routes in:
The theme page gathers the tablets where the script is itself the subject: lexical lists, school exercises, the literary texts that the schools transmitted.
The same story told as a chapter rather than a catalogue, from the first accounting marks to the epics.
Where the script begins: 1,781 tablets from the centuries when writing was still only able to count.
And if you want to see what the script was eventually capable of carrying, the two other dossiers on this site follow single stories across the corpus — a flood narrative older than Genesis, and what the astronomical tablets really say about gods descending from the sky.
The wedges were invented to count sheep. Everything else humanity has written down came afterwards, and rode in on the accounting.