Claude Code breaks 3500 year old dead language, biggest archaeological linguistic breakthrough in 74 years

The password from 3500 years ago was pried open by someone 5 months ago.

On June 16th, an article signed by Tom Di Mino exploded in the AI community – he claimed to have systematically deciphered the Minoan civilization’s “Linear A” using a Python script written by Claude Code.

A dictionary with 408 entries. Confirmation of 40 sound values. A 9-page grammar manuscript.

Only those who understand the history of Linear Text A can truly weigh the weight behind these numbers.

Because this set of symbols engraved on clay tablets has left the world’s smartest linguists helpless for 70 years.

The glory of 1952 and the silence of 70 years
To understand the weight of this matter, we must first go back to 1952.

That year, British architect Michael Venturis announced on BBC radio that he had deciphered the linear script B.

This is the writing system used by the Mycenaean civilization on the island of Crete, and Ventis proves that it records an ancient Greek language.

The news made it to the front page of The New York Times.

Ventris was deified overnight.

But Ventis also left a regret – he didn’t even touch the predecessor of Linear B, Linear A.

Linear script A appeared around 1800 BC, about 350 years earlier than linear script B.

It is the native script of the Minoan civilization, using syllabic symbols (mainly consonant+vowel pairs) and ideographic symbols, with approximately 60 core syllables shared with linear script B.

But the key difference is that Linear Text B records Greek, which has a large number of known languages for comparison; The language behind Linear Text A, no one knows what it is.

Without a known language as a key, it’s like facing a lock without even knowing the shape of the key.

Over the past 70 years, countless scholars have attempted. No one succeeded.

A 7-year obsession of an amateur athlete
Tom Di Mino is not an academic.

He lives in the Hudson Valley of New York and is a self-taught AI engineer who is also an amateur linguist

Started studying classical history and linguistics at the age of 18, proficient in 8 languages including Attic Greek, Classical Latin, Sanskrit, Arabic, and Ugarit.

8 languages. Most of them are dead languages.

He spent a full 7 years researching Linear Text A and personally visited Crete twice for field investigations.

But the real turning point occurred in January 2026.

Di Mino started building a Python script using Claude Code that can systematically query and cross reference the digitized Linear Text A corpus (GORILA and SigLA databases).

He turned AI into a high-powered ‘cognitive microscope’ in his hands.

Breakthrough Moment: A Symbol Boosts the Entire System
The core of Di Mino’s methodology is to analyze prayer inscriptions.

The sacrificial inscriptions unearthed from five mountaintop sanctuaries on Crete follow a specific prayer formula.

Di Mino noticed that if these inscriptions are read as “prayer texts”, their structure is highly similar to the prayer tradition of the Semitic language family.

The key breakthrough occurred on May 22, 2026.

He locked onto a symbol that had no recognized pitch value before – * 301.

Through the Claude Code script’s exhaustive cross referencing of the corpus, he confirmed that this symbol is pronounced as’ na ‘.

The confirmation of this sound value triggered a chain reaction like a domino.

‘na’ unlocks a verb root: nawaya, meaning ‘to reside, to inhabit’.

This N-W-Y consonant pattern is highly consistent with cognate roots in Hebrew and Akkadian.

The Semitic language hypothesis – a theory proposed by scholar Cyrus Gordon in 1957 but never widely accepted – suddenly has substantial evidence to support it.

Di Mino pushed forward all the way: 13 previously completely unknown linear characters A exclusive symbols obtained sound values, and 5 symbols with unknown sound values in linear characters B were also solved together.

The final output includes a dictionary of 408 entries and a 9-page grammar manuscript titled ‘Ya Diktu: Grammar of the Minoan Peak Sanctuary’.

One thing must be made clear: Claude Code did not “decipher” the linear text A.

What Di Mino does with Claude Code is to compress the corpus cross referencing work that human researchers need weeks or even months to complete to a speed that can be iterated in real-time.

Claude Code’s role is not a codebreaker, but an accelerator

Cautious footnote: Experts have not yet stamped
An important limitation must be added here.

As of now, linguistic experts from Rutgers University and Cambridge University are reviewing Di Mino’s findings. The conclusion has not yet been announced.

The “decryption statement” of Linear Text A has appeared more than once in academic history, and the vast majority have ultimately been proven false. Scholars in this field remain highly vigilant about any new claims – this is normal and necessary.

Di Mino himself also used cautious wording in the original text: this is’ proposed phonetic values’, not ‘confirmed’.

But even if the final conclusion is discounted, the systematic output of 408 entries, internal consistency across five site inscriptions, and structural correspondence with the Semitic language family have far exceeded the depth of any previous attempts.

3500 years of waiting, met with 5 months of acceleration
Pull apart the timeline to see:

In 1800 BC, the Minoans carved these symbols on clay tablets.

In 1450 BC, the linear script A ceased to be used and the Minoan civilization disappeared.

In 1952, Ventris deciphered its “descendant” Linear Script B, while Linear Script A itself remained silent.

In 1957, Gordon proposed the Semitic language hypothesis, which was not widely accepted.

For the next 70 years, no one broke through.

In January 2026, an amateur linguist living in the Hudson Valley opened Claude Code.

Five months later, there were 408 entries.

If Di Mino’s results can withstand expert review, it will be the most significant breakthrough in the field of paleography research since Ventis.

And its methodological significance may be more profound than the conclusion itself——

AI doesn’t need to ‘understand’ a language to help humans decipher it, it just needs to be fast enough for every human intuition to be instantly verified.