A four-thousand-year-old script surrendered in 2022 to a corpus barely longer than this article — while far greater archives stay silent, and the reason overturns everything we thought we knew about reading the past.
The mound at Susa had been giving up its dead for eleven years when the strange tablets began to appear.
It was 1903, and the French mission on the plain of Khuzestan, in what is now south-western Iran, was working at industrial scale. Jacques de Morgan’s crews moved earth by the ton, cutting into a settlement that had been continuously occupied since before the invention of the wheel. Out of the trenches came Mesopotamian cuneiform, monumental sculpture, the great stele bearing the laws of Hammurabi. And among them, on a handful of clay tablets and stone objects, came rows of a script nobody had ever catalogued: elegant, angular, unmistakably deliberate, and entirely unreadable.
It was named Linear Elamite. It would keep its silence for one hundred and nineteen years.
Generations of philologists took their turn. They had every advantage the discipline could offer — Susa was one of the best-excavated sites on earth, the surrounding cultures were literate and well documented, and the tablets sat in the Louvre where anyone could study them. None of it helped. By 2018, after more than a century of concentrated effort, sound values had been proposed for precisely twelve signs. Twelve. The script was, by any reasonable measure, a dead end.
Then, in July 2022, a journal of Assyriology carried the announcement that Linear Elamite could be read.
The astonishing part is not that it happened. It is what it happened with. The entire surviving corpus of Linear Elamite amounts to roughly forty inscriptions — 1,863 individual signs, of which 1,810 remain legible. That is less text than fills the front page of a newspaper. It is a fraction of what epigraphers routinely describe as the bare minimum for statistical attack on an unknown writing system.
Now consider what has not been read. Proto-Elamite, from the same region and nine centuries older, survives in more than 1,600 clay tablets, some running to two hundred lines. Silent. The Indus Valley civilisation — whose cities housed tens of thousands, whose standardised weights and engineered drains have become shorthand for administrative sophistication — left between four and five thousand inscribed objects. Silent. Linear A, the script of Minoan Crete, survives in some 1,427 specimens carrying more than seven thousand signs, and has absorbed the attention of linguists, archaeologists and wartime cryptographers for over a century. Silent.
Set the numbers in a column and something is badly wrong with the story we tell about lost writing. The script with forty inscriptions gave way. The scripts with hundreds and thousands did not.
Whatever unlocks a dead writing system, it is not the amount of it that survives.
FIGURE 1 — HERO IMAGE. Editorial archaeological photography. A single inscribed silver Elamite beaker on a matte charcoal plinth under soft museum conservation lighting, incised linear signs raised into relief by low raking light from camera left. Eye-level angle, 85 mm lens, shallow depth of field with the far rim falling gently out of focus, background receding to near-black. Cool metallic palette against warm earth undertones, contemplative and austere atmosphere. Ultra-realistic, premium magazine quality, no people, no text, no watermark, no logo. 16:9, 8K.
The Arithmetic That Refuses to Add Up
The assumption buried in almost everything written about undeciphered scripts is that they are undeciphered because too little of them survives, and that one spectacular excavation would settle the matter. It is the logic behind every headline promising that artificial intelligence, given enough data, will finish the job. It is also the logic of the Rosetta Stone: find the bilingual, break the code.
Neither premise holds, and the evidence against them has been sitting in plain view.
Consider the two Elamite scripts. They come from the same region, the same broad cultural sphere, the same excavations. Proto-Elamite has forty times the surviving material of Linear Elamite. It has been studied longer. It is better published, and in recent years it has been photographed with reflectance imaging and made freely available online. By every metric of data availability it should have fallen first.
It has not fallen at all — and the reason is a matter of genre rather than volume. The Proto-Elamite tablets are accounts. They record grain, livestock, labour and rations, and more than half of the sign tokens in the published corpus are numerals, drawn from several parallel counting systems whose values shift depending on what is being measured. There is grammar in there somewhere, and the Oxford Assyriologist Jacob Dahl has argued that certain signs function syllabically, which would make these the earliest syllabic texts known anywhere. But a ledger is not a sentence. Bookkeeping is the one genre of writing that can be produced in enormous quantity while carrying almost no language at all.
This is the trap that shadows the entire subject. Writing was very largely invented in order to count things. Accounting is what gets pressed into clay, and clay is what survives the millennia; letters, contracts, poetry and law arrive later and tend to be committed to materials that rot. The forms of evidence that preserve a script best are precisely the forms that carry the least speech. The archive that saved Proto-Elamite is the same archive that renders it mute.
So if not quantity — what?
FIGURE 2 — SECTION IMAGE. Museum-quality still-life photography. An overhead view of three Proto-Elamite accounting tablets arranged on a neutral grey conservation surface, their impressed numerical notations and incised signs picked out by diffuse overhead light with a single soft directional source at forty-five degrees. Directly overhead camera angle, 50 mm lens, deep focus across all three objects, edges falling into soft shadow. Restrained palette of buff clay, cool grey and charcoal. Clinical, archival atmosphere. Ultra-realistic archaeological realism, no people, no text, no watermark, no logo. 3:2, 8K.
What Actually Broke the Code
The breakthrough came from objects that nobody had thought to treat as the key.
In 2018 the French philologist François Desset published nine inscribed vessels of silver and copper, known as gunagi beakers, held in private and museum collections. They carried Linear Elamite text. Working through the sign sequences, Desset and his collaborators noticed something they already recognised — not a word, not a grammatical pattern, but a set of names.
They were the names of rulers of the Sukkalmah dynasty, the house that governed Elam in the twentieth century BCE: Ebarat, Šilhaha. These were not mysterious figures. They appear in cuneiform documents written by literate neighbours, in a script that has been readable since the nineteenth century. Their pronunciation was already known. And here they were, spelled out inside an unreadable text, at positions where the phrasing of royal inscriptions suggested a king’s name ought to sit.
From those fixed points, sound values propagated outward across the signary like a crossword solved from a single confirmed answer. The 2022 paper reported the system as syllabic and largely legible. The classification remains contested — Jacob Dahl disputes that the script is purely syllabic, and the German scholar Michael Mäder has argued that roughly thirty per cent of the characters are logographic — but the decipherment itself is broadly accepted.
Not mass, not bilingualism, but a name whose pronunciation was already known, sitting inside a text nobody could read.
The mechanism is not luck. It is structural, and it follows from what proper names are.
Ordinary vocabulary can hide behind logograms. A sign meaning “ox” tells a modern reader what the scribe was talking about while revealing nothing whatever about the sound he made when he said it — which is why Chinese characters can be read aloud in mutually unintelligible dialects. Names have no such refuge. When a scribe must record a foreign king, a distant city or a personal name with no established sign behind it, the writing system is forced to spell. Names drag phonetic values into daylight. They are the point at which any script, however word-based its habits, is obliged to become a record of speech.
And names have a second property that matters just as much: they can be checked from outside. A neighbouring civilisation’s chronicle, a later king list, a place name that survived onto a modern map — each provides an independent register against which a proposed reading either matches or fails. Without that external check, a decipherment cannot be tested, only asserted.
FIGURE 3 — SECTION IMAGE. Macro photography, editorial science style. Extreme close-up of incised linear script on a tarnished silver vessel wall, the frame filled by four or five signs, tool marks and micro-abrasions fully resolved. Raking side light from camera left throwing each stroke into hard shadow, 100 mm macro lens, razor-shallow depth of field with focus falling off across the curve of the metal. Cool silver-grey palette with warm oxidised highlights. Forensic, intimate atmosphere. Ultra-realistic, no people, no text, no watermark, no logo. 16:9, 8K.
The Pattern Hidden in Every Famous Decipherment
Once the mechanism is visible in Linear Elamite, it becomes impossible not to see it everywhere else — including in the two decipherments that every schoolchild is taught.
The Rosetta Stone is celebrated as the object that unlocked ancient Egypt. It was recovered in 1799. Egyptian hieroglyphs were not read until the 1820s. The popular account passes lightly over that interval of more than two decades, and the interval is the interesting part: a bilingual inscription, by itself, does not decipher anything. The Greek text told scholars what the Egyptian said. It did not tell them which sign made which sound.
What Jean-François Champollion did was narrower and far more decisive. He attacked the cartouches — the oval rings enclosing royal names — on the reasoning that Ptolemy and Cleopatra were Greek names for which no Egyptian logogram could exist, and which therefore had to be spelled out phonetically. By comparing the two, he could assign sound values to individual signs and test them against further royal names. The bilingual mattered because it told him which names to expect. The names did the work.
Michael Ventris’s decipherment of Linear B in 1952 confirms the point from the opposite direction, because Ventris had no bilingual at all. He built grids of sign relationships from the internal structure of the Knossos and Pylos tablets, then tested them against a hypothesis: that certain frequently recurring sign groups were Cretan place names — Amnisos, Knossos, Tulissos — which had survived into later Greek geography and were therefore pronounceable. When the grid produced recognisable toponyms, and then recognisable Greek words clustering around them, the system unfolded within months. Roughly six thousand Linear B inscriptions survive, a corpus the field regards as modest. It was not the size that mattered. It was that Crete’s place names had outlived the Bronze Age that made them.
The pattern holds for Maya glyphs, read progressively across the twentieth century, where Diego de Landa’s sixteenth-century sign list — long dismissed as a garbled misunderstanding — together with rulers’ names and dynastic emblem glyphs supplied the phonetic anchors.
Four decipherments, four centuries apart in their methods, and in none of them was corpus size the deciding factor. In all four, a name whose pronunciation was known from outside the script provided the way in.
The bilingual told Champollion which names to expect. It was the names that did the work.
FIGURE 4 — INFOGRAPHIC. Clean editorial comparison diagram, museum publication quality. Horizontal bar chart on a warm off-white ground: six ancient scripts on the vertical axis, surviving corpus size plotted on a logarithmic horizontal scale. Deciphered scripts rendered as solid rust-orange bars, undeciphered scripts as flat charcoal grey. Thin baseline rules, generous white space, small-caps sans-serif labels, no gridlines, no drop shadows, no three-dimensional effects. The visual argument is that the orange bars are conspicuously not the longest ones. Minimal, premium, no watermark, no logo. 16:9, 300 dpi.
The Ones That Stayed Silent
Apply that lens to the scripts still waiting, and their failures stop looking mysterious. Each one is missing the same thing.
Linear A is the cruellest case, because in one sense it is not undeciphered at all. The Mycenaeans built Linear B out of it, and the two share a large number of sign shapes. Epigraphers can therefore assign provisional sound values to much of Linear A by direct transfer, which means the Minoan tablets can be read aloud this afternoon by anyone who has learned the signary. What emerges is not a language anyone recognises. It is not Greek, not Semitic, and it attaches to no Indo-European branch convincingly. The sounds are available; the meanings are not. Names certainly fill those administrative tablets — but they are the names of Minoan officials and Minoan villages, recorded by nobody else. Egypt traded with Crete and mentioned it, yet left no usable register of Minoan officeholders spelled out in a readable script. There is nothing to check the sounds against.
The Indus script poses a harder question, and one that has to be answered before decipherment can even begin sensibly. Its inscriptions are extraordinarily short: analyses of the standard databases give an average of about 4.6 signs, with the longest text on a single surface running to seventeen. In 2004 Steve Farmer, Richard Sproat and Michael Witzel published an argument that the system never encoded speech at all — that the Harappans were not literate in the usual sense. Their case rested on the brevity, on the near-absence of the unforced sign repetition that every genuinely linguistic corpus displays, and on the observation that the system barely evolved across six centuries. Living scripts change; this one, they argued, does not behave like one. The claim has never won consensus, and researchers pointing to positional regularities and language-like transition probabilities argue the opposite. Sproat’s rejoinder is the methodologically sharp one: medieval heraldic rolls also score within the range of natural languages on individual statistical tests. Passing one test is not evidence of language.
The stakes are not purely academic. In January 2025, marking the centenary of the civilisation’s announcement to the world, the chief minister of Tamil Nadu, M. K. Stalin, offered one million dollars to anyone who deciphers the script to expert satisfaction. The state’s archaeology department simultaneously published a study reporting extensive parallels between Indus signs and graffiti marks from Tamil excavations — a finding that would tie Dravidian southern India to the Harappan past, arriving inside a long-running national argument about who inherits India’s oldest civilisation. That context does not make the finding wrong. It does mean it awaits independent replication.
Rongorongo, from Rapa Nui, shows what happens when the external register was human rather than written. Twenty-seven authenticated wooden objects survive, carved in reverse boustrophedon — each line inverted relative to the last, so a reader must rotate the tablet a hundred and eighty degrees at the end of every row. Not one remains on the island; they are scattered across institutions on four continents. In February 2024 a team led by Silvia Ferrara of the University of Bologna published radiocarbon dates for four tablets held in Rome. Three came from trees felled in the eighteenth or nineteenth centuries. The fourth returned a secure date of 1493 to 1509 — two centuries before European ships arrived in 1722. Radiocarbon dates the felling of the wood rather than the carving of the glyphs, and the authors say so; but if the inference holds, Rongorongo was invented independently, one of a handful of times in human history that anyone has invented writing from nothing. What cannot be recovered is the reading. Production stopped in the 1860s amid slaving raids and epidemics, and the people who understood the tablets died before anyone thought to ask them properly.
The Phaistos disc fails for a simpler reason: there is only one of it. Recovered in Crete in 1908, it carries 241 signs stamped from a repertoire of 45 individual punches — the earliest known printing with reusable type, three millennia early. Elizabeth Barber stated the problem with a precision nobody has improved on: there is not enough information for a claimed decipherment to be proved, and by exactly the same token not enough for it to be disproved. The disc has consequently been decoded dozens of times, as Greek, as a hymn, as a calendar, as an astronomical table, and the readings cannot be adjudicated.
And the Vinča symbols of Neolithic south-eastern Europe belong on this list only by courtesy. The prevailing view is that these incised marks are proto-writing — conventional, meaningful, organised signs indicating ownership, quantity or ritual status, without encoding speech. Nothing in the corpus shows syntax or sequential structure. To call them undeciphered is to assume the very thing that needs demonstrating.
FIGURE 5 — SECTION IMAGE. Cinematic documentary photography. A wooden Rongorongo tablet lying in a conservation drawer under low tungsten-balanced museum light, carved glyph rows running diagonally across the frame, timber grain and old cracks fully resolved. Three-quarter high angle, 35 mm lens, moderate depth of field holding the near two-thirds of the tablet sharp. Warm brown and amber palette against deep neutral shadow. Melancholy, archival atmosphere. Ultra-realistic, unromanticised, no people, no text, no watermark, no logo. 3:2, 8K.
What the Machines Have Actually Done
Every year brings the claim that artificial intelligence is about to read the unreadable. The reality is both more impressive and more limited.
The strongest work comes from a collaboration between Google DeepMind and academic epigraphers. Ithaca, released in 2022, restores, dates and geographically attributes ancient Greek inscriptions. Aeneas, published in Nature in July 2025, does the same for Latin, trained on a consolidated dataset of more than 176,000 inscriptions. It retrieves parallel texts in seconds — the slow expert labour of finding comparable phrasing across scattered corpora — and restores damaged passages, including gaps of unknown length, achieving roughly 73 per cent accuracy where up to ten characters are missing. The published results are strongest when historian and model work together.
Read that description again and the limitation is inside it. Ithaca works on Greek. Aeneas works on Latin. Both are languages deciphered long ago, by people, and both systems depend on vast curated corpora that are already readable. These models restore and contextualise. Neither has deciphered anything, and neither was built to.
The obstacle is not insufficient ambition. Machine learning extracts structure by exploiting redundancy, and the undeciphered scripts are defined by having almost none. No algorithm infers grammar from inscriptions averaging under five signs, or reconstructs a language from texts that were never written down. With small samples and flexible models, spurious structure is easy to find and very hard to falsify — which is why the Phaistos disc has been decoded by algorithm as often as by hand, with the same evidential weight.
What computation has genuinely changed is the infrastructure. Palaeographic databases have made Linear A searchable outside its printed five-volume corpus. Reflectance imaging has put the Proto-Elamite tablets in front of anyone who wants them. Photogrammetric modelling has produced new transcriptions of the Rongorongo glyphs. None of this reads a script. All of it improves the odds that a human being notices the sequence that matters.
FIGURE 6 — SECTION IMAGE. Editorial science photography. A researcher’s hands, gloved in white nitrile, positioning a clay tablet beneath a multi-light reflectance imaging dome in a darkened laboratory, the dome’s LED array casting a cool ring of illumination across the tablet surface. Low three-quarter angle from tablet height, 50 mm lens, shallow depth of field with the equipment behind dissolving into bokeh. Cool blue-white instrument light against warm clay. Quiet, methodical atmosphere. Ultra-realistic, face not visible, no text, no watermark, no logo. 16:9, 8K.
The Silence of Those Who Survive Alone
There is a habit of writing about undeciphered scripts as though they were dignified riddles, and the civilisations behind them a set of puzzles arranged for our benefit. It is worth resisting.
These were working systems, used by people with deadlines. Officials counted grain with them. Merchants recorded consignments. Priests wrote down the formula that had to be spoken over the libation and copied it onto vessel after vessel until the phrasing was fixed. Someone in Susa, five thousand years ago, sat with a stylus and wrote out a list of names, and every one of those names belonged to a person who woke that morning and expected to be remembered.
What has been lost is not the writing. The writing is in remarkable condition. The Phaistos disc was deliberately fired, which is why it has outlasted most of the tablets found beside it. The Rapa Nui glyphs are crisp enough to model in three dimensions. Linear A can be pronounced aloud by any competent epigrapher today. The marks survive with a clarity that would have satisfied the people who made them.
What has been lost is the web of external reference that would let those marks be tested against anything at all — the neighbours who wrote the same names down in a different hand, the descendants who carried the language forward, the readers who could have explained what the rows of glyphs were for. Decipherment turns out to be less an act of intelligence than an act of triangulation. It requires at least two independent lines of sight onto the same word.
This reframes what it means for a civilisation to fall silent. It is not a question of how much they wrote, or how well their materials endured, or how clever their eventual investigators prove to be. A culture becomes unreadable when everyone who could corroborate it disappears — when it is left as the only surviving witness to its own vocabulary.
Linear Elamite was rescued because the Elamites had neighbours who found their kings worth mentioning. That is the whole difference. The scripts still waiting are not silent because too little of them survives. They are silent because they survive alone, and no quantity of excavation can supply a second voice where none was ever recorded.
Somewhere in a museum drawer or an unopened trench, there may be a tablet bearing a foreign name — a king, a port, a god borrowed from over the water — spelled out phonetically by a scribe who had no idea he was leaving a key. Every decipherment in history has turned on such an accident. The remaining ones are waiting for theirs.
FIGURE 7 — CLOSING IMAGE. Cinematic documentary photography. Wide interior of a museum storage facility at night, rows of shallow archival drawers receding into darkness, a single drawer pulled open and lit from within to reveal inscribed fragments. Low symmetrical eye-level angle down the central aisle, 24 mm lens, deep focus from foreground drawer to vanishing point. Desaturated grey-green palette with one warm pool of light. Still, expectant atmosphere. Ultra-realistic, no people, no text, no watermark, no logo. 16:9, 8K.
Key Takeaways
— Linear Elamite was deciphered in 2022 from roughly forty inscriptions totalling 1,863 signs — one of the smallest corpora ever to yield a decipherment.
— Corpus size does not predict success. Proto-Elamite has over 1,600 tablets, the Indus civilisation four to five thousand inscribed objects, and Linear A some 1,427 specimens. All remain unread.
— The decisive lever in every major decipherment has been proper names whose pronunciation was already known from an external, readable source.
— Names force phonetic spelling. A logogram can hide a word’s sound; a foreign king’s name cannot, because no sign exists for it.
— The Rosetta Stone did not decipher Egyptian by itself — twenty years separated its discovery from Champollion’s reading, and the breakthrough came through royal cartouches.
— Ventris deciphered Linear B in 1952 with no bilingual whatever, using Cretan place names that survived into later Greek geography.
— Whether the Indus script encodes language at all remains genuinely contested, and that question is prior to any decipherment attempt.
— AI systems such as Ithaca and Aeneas restore and contextualise Greek and Latin — languages deciphered long ago. No machine has deciphered an unknown script.
Timeline
c. 5700–4500 BCE Vinča culture produces incised pottery marks across the Danube basin — most scholars classify them as proto-writing.
c. 3100 BCE Proto-Elamite emerges at Susa. More than 1,600 tablets survive; over half of all sign tokens are numerical.
c. 2600–1900 BCE The Indus Valley civilisation flourishes, leaving four to five thousand inscribed objects averaging 4.6 signs each.
c. 2300–1880 BCE Linear Elamite in use, including inscriptions naming Sukkalmah rulers Ebarat and Šilhaha.
c. 1800–1450 BCE Linear A serves Minoan palace administration across Crete and the Aegean.
c. 1700 BCE The Phaistos disc is manufactured, stamped with 45 reusable punches.
c. 1450–1200 BCE Linear B adapts the Minoan signary to record Mycenaean Greek.
1493–1509 CE Radiocarbon range for the earliest securely dated Rongorongo tablet.
1799 The Rosetta Stone is recovered at el-Rashid, Egypt.
1822–1824 Champollion reads Egyptian hieroglyphs, working outward from royal cartouches.
1903 Linear Elamite inscriptions emerge from the French excavations at Susa.
1908 Luigi Pernier recovers the Phaistos disc in southern Crete.
1952 Michael Ventris deciphers Linear B without a bilingual text.
2004 Farmer, Sproat and Witzel argue the Indus script never encoded speech.
2018 Desset publishes nine inscribed gunagi vessels; twelve Linear Elamite signs have values.
July 2022 The decipherment of Linear Elamite is announced.
February 2024 Ferrara and colleagues publish radiocarbon dates for four Rongorongo tablets.
January 2025 Tamil Nadu announces a one-million-dollar prize for deciphering the Indus script.
July 2025 DeepMind’s Aeneas model for Latin epigraphy is published in Nature.
Did You Know?
— The Phaistos disc is the earliest known object printed with reusable type — roughly three thousand years before Gutenberg, and it was never repeated.
— Rongorongo is written in reverse boustrophedon: every line is inverted relative to the one above, so the reader must physically rotate the tablet at the end of each row.
— Not one Rongorongo tablet remains on Rapa Nui. All twenty-seven authenticated objects are held by institutions on four other continents.
— Scholars can pronounce Linear A. They simply cannot understand a word of what they are saying, because the sounds correspond to no known language.
— The Phaistos disc has never undergone thermoluminescence testing, which would settle its date of manufacture — more than a century after its discovery.
— Diego de Landa, the sixteenth-century bishop who burned Maya books, also recorded a sign list that later proved essential to deciphering the script he helped destroy.
SEO Package
SEO Title: The Ancient Scripts We Still Cannot Read
Meta Description: Linear Elamite fell in 2022 from just forty inscriptions, while vast archives stay silent. The real key to deciphering ancient scripts is not size — it is names.
URL Slug: ancient-undeciphered-scripts-decipherment
Focus Keyword: undeciphered ancient scripts
SEO Tags: undeciphered scripts, Linear Elamite decipherment, Linear A Minoan, Indus Valley script, Proto-Elamite tablets, Rongorongo Easter Island, Phaistos disc, Vinca symbols, Rosetta Stone Champollion, Michael Ventris Linear B, ancient writing systems, archaeology decipherment, epigraphy, AI ancient texts, lost languages
Fact-Check Ledger
| Statement | Confidence | Source |
| Linear Elamite corpus: c. 40 inscriptions, 1,863 signs, 1,810 legible | HIGH | Desset et al. (2022); corpus statistics reported via National Geographic |
| Decipherment anchored on Sukkalmah royal names on gunagi vessels | HIGH | Desset (2018); Desset et al. (2022) |
| The decipherment is complete and the script purely syllabic | MEDIUM | Broadly accepted, but Dahl disputes pure syllabicity and Mäder argues c. 30% logographic |
| Linear A: c. 1,427 specimens, 7,362–7,396 signs | MEDIUM-HIGH | GORILA corpus (Godart & Olivier, 1976–1985); counts vary by inclusion criteria |
| Indus inscriptions average 4.6 signs; longest single surface 17 | MEDIUM-HIGH | Farmer, Sproat & Witzel (2004); consistent across independent analyses |
| Indus corpus of 4,000–5,000 inscribed objects | MEDIUM | Widely cited range; databases differ on object definition and provenance |
| Tamil Nadu study reporting extensive Indus–Tamil graffiti parallels | LOW-MEDIUM | State archaeology department, Jan 2025; not independently replicated |
| Proto-Elamite: 1,600+ tablets, over half of sign tokens numerical | MEDIUM-HIGH | Dahl (2019); many tablets remain unpublished, so the total is a floor |
| 27 authenticated Rongorongo objects, none held on Rapa Nui | HIGH | Ferrara et al. (2024); some catalogues give 23 or 25 by stricter criteria |
| Rongorongo tablet dated 1493–1509 | HIGH | Ferrara et al. (2024), Scientific Reports |
| Rongorongo was invented before European contact | MEDIUM | Inference from the above; radiocarbon dates the timber, not the carving |
| Phaistos disc: 241 impressions from 45 distinct signs | HIGH | Standard published counts; some give 242, one sign unrecognisable |
| The Phaistos disc is authentic rather than a 1908 forgery | MEDIUM-HIGH | Associated Linear A tablet, scribal corrections, independently excavated sealing; no TL test performed |
| Vinča marks are proto-writing rather than a writing system | MEDIUM-HIGH | Prevailing scholarly view; Tărtăria stratigraphy disputed and direct dating precluded |
| Aeneas: 176,000+ Latin inscriptions, c. 73% restoration accuracy | HIGH | Assael et al. (2025), Nature |
Further Reading
Assael, Y., Sommerschield, T., Cooley, A., Shillingford, B., Prag, J., & Pavlopoulos, J. (2025). Contextualizing ancient texts with generative neural networks. Nature, 644, 1–8. https://doi.org/10.1038/s41586-025-09292-5
Barber, E. J. W. (1974). Archaeological decipherment: A handbook. Princeton University Press.
Chadwick, J. (1958). The decipherment of Linear B. Cambridge University Press.
Dahl, J. L. (2019). Tablettes et fragments proto-élamites (Textes cunéiformes du Louvre 32). Louvre Éditions.
Desset, F. (2018). Nine Linear Elamite texts inscribed on silver “gunagi” vessels: New data on Linear Elamite writing and the history of the Sukkalmah dynasty. Iran, 56(2), 105–143.
Desset, F., Tabibzadeh, K., Kervran, M., Basello, G. P., & Marchesi, G. (2022). The decipherment of Linear Elamite writing. Zeitschrift für Assyriologie und vorderasiatische Archäologie, 112(1), 11–60.
Farmer, S., Sproat, R., & Witzel, M. (2004). The collapse of the Indus-script thesis: The myth of a literate Harappan civilization. Electronic Journal of Vedic Studies, 11(2), 19–57.
Ferrara, S., Lastilla, L., Ravanelli, R., Valério, M., Talamo, S., et al. (2024). The invention of writing on Rapa Nui (Easter Island): New radiocarbon dates on the Rongorongo script. Scientific Reports, 14, 2794. https://doi.org/10.1038/s41598-024-53063-7
Godart, L., & Olivier, J.-P. (1976–1985). Recueil des inscriptions en Linéaire A (GORILA, Vols. 1–5). École française d’Athènes.
Robinson, A. (2002). Lost languages: The enigma of the world’s undeciphered scripts. McGraw-Hill.
Sproat, R. (2014). A statistical comparison of written language and nonlinguistic symbol systems. Language, 90(2), 457–481.
The British Museum. Collection online: Rosetta Stone (EA24). https://www.britishmuseum.org/collection
Musée du Louvre, Département des Antiquités orientales. Susa collections, Paris.
