Proto-Elamite

Summary

About 1,700 clay tablets survive from southwestern Iran, most of them from Susa, written somewhere between 3300 and 2900 BC. The script is one of the world’s earliest. Its numerical systems are understood, most of them shared with or borrowed from Mesopotamian proto-cuneiform, and a few other signs have been identified. The rest is opaque.

The script is conventionally called Proto-Elamite, but the language it records is unidentified, and the name does not establish that it is Elamite. What stands in the way of reading it is structural: no known bilingual text, no later script that demonstrably descends from it, and a gap of more than eight centuries between it and Old Elamite.

What is documented

The corpus. Robert Englund counted "just over 1,600 pieces", with around 10,000 lines of text, in 2004. In November 2026 the Cuneiform Digital Library Initiative catalogue listed 1,755 artifacts under the period. A 2026 Cambridge volume gives about 1,700 tablets. For comparison, Englund puts the contemporary proto-cuneiform corpus from Babylonia at more than 6,000 tablets and fragments and 38,000 lines.

The sign count, and why the usual figure is wrong. Mecquenem’s 1949 list ran to more than 5,500 signs, which Englund attributes largely to variants. Meriggi cut it to fewer than 400 entries. Englund’s own dataset held over 1,900 discrete attested forms across 26,320 occurrences, clustering around roughly 500 basic shapes, with more than a thousand forms occurring only once and about three hundred only twice. He estimated the real system at 600 to 900 signs. A 2019 computational signlist resolves 1,623 non-numeric signs into 287 basic signs, 1,087 variants and 249 complex graphemes, alongside 49 numeric signs.

What is actually readable. The numerical systems. Englund describes a decimal system with no proto-cuneiform counterpart, a sexagesimal system "presumably loaned directly from Mesopotamia", a bisexagesimal system that differs only slightly from its Mesopotamian equivalent, and grain-capacity systems very nearly the same as Mesopotamia’s. Numerals dominate the texts: in the 2019 analysis dataset an average text holds 27 readable signs, of which only ten are not numerals. Englund identifies a few non-numerical signs, among them M288, a general sign for a measure of grain, and M388, a male dependent labourer. He calls the texts administrative.

What the specialists say about the prospects. Englund calls it "a largely undeciphered script of an entirely unknown dead language", and judged in 2004 that "The prospects of discovering script characteristics that could lead to a decipherment of proto-Elamite are not great". The 2026 Cambridge treatment states that it "remains only very partly deciphered".

Why it is hard, specifically. None of the sources the Registry consulted reports a bilingual text, and no later script securely descends from Proto-Elamite. Englund points to "the span of over 800 years unaccounted for between proto- and Old Elamite", which interrupts the palaeographic tradition. Most of its ideograms are, in his words, "of a substantially more abstracted form" than their proto-cuneiform counterparts, whose pictographic character often helps in reading them. And the publication history is poor: Englund describes Meriggi’s 1971 to 1974 sign list as "laced with incorrect identifications", with seemingly all of its signs published as mirror images, and notes how little attention was paid to accurate hand copies of the texts themselves.

What has changed. The Cuneiform Digital Library Initiative now hosts the entire corpus in transcription, with digital images or hand copies of almost every text. In 2012 about 1,100 tablets held in the Louvre were imaged with a reflectance transformation imaging dome, which lights a surface from many angles to bring out worn impressions. Jacob Dahl, who led that work, said: "you cannot decipher a writing system without having reliable images." Contemporary reporting put the position at 80 to 90 per cent of signs undeciphered.

The texts are short. In the 2019 analysis dataset of 1,399 texts, the longest holds 724 readable signs, 198 of them non-numeric. The mean text holds 27, of which ten are non-numeric.

Leading explanations

It is an administrative accounting script, partly readable (established). The texts are administrative records. The numerical systems and a few other signs are understood; the bulk of the non-numerical signs are not. None of the sources the Registry consulted disputes this.

The underlying language is unknown (widely accepted). The name is conventional. Englund calls the script "a largely undeciphered script of an entirely unknown dead language".

Linear Elamite descends from it (Desset and colleagues, 2022; disputed). If the later Linear Elamite script, used between about 2300 and 1800 BC, developed from Proto-Elamite, then a Linear Elamite decipherment should feed phonetic values backwards. Jacob Dahl disputes the descent. He argues that Linear Elamite is "not an organically developed writing system": late third-millennium scribes created it from recovered Proto-Elamite tablets, choosing the shapes of their signs from the most frequently occurring ones. He has also said he is not sure the 2022 decipherment succeeded. This is the live counter-argument.

Computational clustering will narrow the problem (Born and colleagues, 2019; hypothesis, untested as a route to decipherment). Unsupervised clustering can collapse variant forms onto basic signs and detect structure in complex graphemes. The authors present the techniques "as an aid to manual decipherment", and note that traditional efforts "have not yet succeeded in identifying a linguistic context".

Current status

Unsolved. Proto-Elamite remains unread as a language-bearing script. Its numerical systems and a few other signs are understood; the language underlying it is unidentified. The obstacles are unchanged: no known bilingual, no securely descended later script, the gap of more than eight centuries that Englund identified, heavily abstracted signs, and a legacy of unreliable sign lists and hand copies. What has changed is infrastructure rather than understanding. Three things would change this record: a bilingual or closely parallel text; a securely demonstrated descent from Proto-Elamite to Linear Elamite that transfers phonetic values backwards and survives peer review, with Dahl’s objection answered; or a published, independently reproduced set of phonetic readings yielding coherent grammar across multiple unseen tablets, rather than isolated name identifications.

Last reviewed: November 2026. Records are provisional. Where the evidence changes, the entry changes. Found an error? Tell us.