Nobody has read Linear A in 3,500 years. An agent is going through it, sign by sign.
The Minoans of Bronze Age Crete built Europe's first civilization and wrote in Linear A, and no one has been able to read a word of it since. kober works on it the way Alice Kober worked on its sister script: not by guessing what the signs mean, but by cataloguing what they do. It reads half of the documents sign by sign, keeps a card for every word, and seals each pattern it notices before it opens the other half to check. What holds goes into its notebook. What fails stays on the record.
At the desk
Live. Everyone watching sees the same document, the same sign and the same card.
What it has learned
Claims that held on documents it hadn't read.
Claims
Every claim, held or not. Sealed while it reads one half of the documents, checked on the other.
Does it add up?
The tablets' own arithmetic, checked line by line.
Many tablets end with 𐙂𐘁 KU-RO, which scholars read as "total", because the number after it is usually the sum of the entries above it. When the agent reaches one, it adds the entries up and compares. Fractions take the values in the published transcription; anything damaged means it can't tell.
| tablet | entries | written | verdict |
|---|
Make a claim
Same record, same test. Each claim is sealed once per pass; if someone got there first, you get their result.
Pick a sign and say where you think it sits in a word. Your claim is sealed into the record, measured on the half the agent reads, and checked straight away on the half it has set aside. If it fails, the entry carries your name.
Pick a sign above.
Visitor claims
Its notes
What it wrote down when it sealed a claim, checked a sum or opened the seals.
The record
Checking the chain…
Every claim is written down before the agent opens the half that tests it, and every result after. Each entry contains the hash of the one before it, so nothing can be changed or quietly removed. Your browser has just recomputed every hash. Click an entry to open it.
| no. | written | entry | about | hash |
|---|
Method
How a claim is made and how it is tested.
The corpus is every published Linear A inscription: 1,722 documents, mostly clay tablets and sealings from Haghia Triada, with Khania, Phaistos, Knossos, Zakros and a few dozen other sites. Seal impressions and roundels that carry a single sign are left out: there is nothing on them to read.
The agent works in passes. Each pass splits the documents in two by a hash of their names (both faces of a tablet stay together) and the agent reads one half, document by document, in catalogue order. After each document it measures what it has read so far. When a pattern becomes unlikely by chance (a one-sided binomial test at p < 0.001 for where signs sit, p < 0.0001 for pairs of signs), it seals the claim and its evidence into the record, before it can see the other half.
When its half is done, it reads the sealed half the same way, tallying every claim as it goes. A claim holds if the sealed half agrees at p < 0.05. Then it starts again on a fresh split, counting from zero, so every pass is an independent check. A claim stays in the notebook only while it holds in every pass that could test it.
| position | a sign starts or ends words more (or less) often than signs in general do |
| pair | one sign follows another inside words more often than their frequencies predict |
| goods | a word is listed against the same commodity (grain, wine, oil, figs…) |
| phrase | a word is followed by the same word, as in the libation formula |
| place | a word is written at one site only, outside Haghia Triada |
| sum | the entries above a KU-RO add up to the number after it; no halves, this is arithmetic |
Sign names are the conventional ones: Linear B values given to Linear A signs that look the same. They are labels, not claims about how Minoan was pronounced.
Alice Kober
Background, sources and limits.
Alice Kober (1906–1950) taught classics at Brooklyn College and spent her nights on Linear B, the script of Mycenaean Greece. She refused to guess. She made index cards, some 180,000 of them, cut from scrap paper during wartime shortages and filed in cigarette cartons, and counted what the signs did. Her "triplets", words that share a stem but change their ending, showed that the language behind Linear B was inflected. Michael Ventris built on her work and deciphered it in 1952, two years after she died.
Linear A is older, written by the Minoans on Crete from about 1800 to 1450 BC. Linear B borrowed most of its signs, so they can be given sound values, but the words they spell match no known language. There are only about 7,500 signs of it in all, and most of it is short accounting. No one has read it.
Sources
- Godart, L. & Olivier, J.-P. Recueil des inscriptions en linéaire A (GORILA), 5 vols. (1976–1985).
- Corpus text from lineara.xyz (tabulation by George Douros, commentary by John Younger); see the repository for provenance.
- Kober, A. Inflection in Linear Class B: 1 — Declension. American Journal of Archaeology 50, 268–276 (1946).
- Corazza, M. et al. The mathematical values of fraction signs in the Linear A script. Journal of Archaeological Science 125, 105214 (2021).
- Salgarella, E. & Castellan, S. SigLA: The Signs of Linear A. sigla.phis.me
- Fox, M. The Riddle of the Labyrinth. Ecco (2013).
Limits
- The corpus is small. Many claims are about the few words that recur, such as KU-RO and the libation formula, more than about the language.
- The halves are not independent people or places: most documents in both come from Haghia Triada. Passes reuse the same documents, so agreeing passes are a check on luck, not new evidence.
- Damage, word breaks and readings follow one tabulation of the corpus. Other readings exist.
- A sum that doesn't add up may be a damaged entry, a sub-total, a list that continues on the other face, or a slip by the scribe.
- Nothing here reads Linear A. It records what the script does, which is how decipherments have started before.