Artificial intelligence is changing the pace of work on lost languages, but it has not changed the basic problem. A system can compare signs, detect repetition, and test a proposed link across a corpus far faster than a person working by hand. What it cannot do is create a reliable meaning where the evidence does not provide one.
That distinction matters for Linear A, the writing system of the Bronze Age Minoan civilisation on the Greek island of Crete, and for Etruscan, a language of Italy used before the rise of the Roman Empire. Both remain difficult because the usual supports for decipherment are missing, weak, or incomplete.
Why lost languages need an anchor
Every ancient language that has been deciphered has depended on some kind of anchor. The source article points to a bilingual text such as the Rosetta Stone, or a known related language that can be used for comparison.
Linear A has neither. It is described as a “language isolate” because no confirmed link has been established with any known language, living or dead. That absence has kept linguists from reaching a settled decipherment over the past century.
Etruscan is not in exactly the same position, but it remains deeply limited as evidence. Linguists have gathered a partial vocabulary from short funerary inscriptions, yet its grammar and fuller meaning remain unresolved.
This is the central challenge for AI. Pattern recognition can be powerful, but decipherment is not only about finding patterns. It is about deciding which patterns correspond to real words, grammar, and meaning.
What AI can do well
AI is valuable because it can make slow comparative work much faster. If a researcher has a hypothesis about a sign, sound, or word, a model or AI-built script can test that idea across thousands of characters in a short time.
The source gives a recent example from June 2026. A self-taught AI engineer and amateur linguist claimed a breakthrough on Linear A after starting with a single proposed connection: one unknown word in a prayer inscription, he suggested, came from a Semitic root meaning “to dwell” or “to inhabit”.
He then used AI-built programming scripts to test that sound pattern against a collected Linear A corpus. According to the source, he was reportedly able to assign values to 40 signs and compile a 408-word lexicon. In his view, that pointed to Linear A as an extinct member of the Semitic language group, which includes languages such as Hebrew and Aramaic.
The important point is not that this claim settles the issue. The source says the claim is still under review. The important point is the division of labor: the human supplied the idea, and AI helped test it quickly.
In practical terms, AI can help researchers by:
- checking whether a proposed sign or word pattern appears across an archive;
- finding repeated sequences that could be easy to miss by eye;
- suggesting likely missing characters in damaged or fragmentary inscriptions;
- testing whether a known language may help explain an unknown one.
That last ability is known as “cross-lingual transfer”. A model trained on a known language can sometimes infer patterns in a closely related unknown language. The source compares this to how knowing Spanish can help someone make guesses about Portuguese.
Where pattern matching reaches its limit
The strongest AI results come when the language family is already known. Researchers have used this kind of approach successfully on scripts such as Ugaritic, where the family relationship is established. Ugaritic is an extinct Semitic language spoken during the late Bronze Age (about 1300-1190BC) in the ancient coastal city of Ugarit, in what is now Syria.
Linear A and Etruscan are harder because the needed anchor is absent or incomplete. A model can show that symbols repeat, that certain sequences cluster, or that one proposed comparison produces matches. But those matches may still be coincidence.
This creates a sharp difference between fluency-like behavior and translation. With a large enough corpus, the source says an AI model could hypothetically become fluent enough to converse with a Minoan or Etruscan speaker on its own statistical terms. It might recognize which words and structures tend to appear together and reuse them in another context.
Even then, it would not necessarily be able to tell a modern reader what the language means. Knowing which signs follow each other is not the same as knowing what those signs refer to in the world.
Why verification is so difficult
AI-assisted decipherment claims face a verification problem. Normally, researchers can compare a proposed translation with native speakers, other texts, or a long-standing expert consensus. For a genuinely undeciphered language, those checks are not available.
The data problem is especially severe for Linear A. Its entire surviving corpus is about 7,500 characters, which the source describes as short enough to fit on a single screen. With so little material, many hypotheses can produce scattered matches that look suggestive without proving much.
That is why independent expert scrutiny and peer-review matter more than confidence scores alone. A statistical result may show that a pattern exists. It does not automatically show that the pattern carries the meaning a researcher proposes.
This is also why the phrase “AI found a pattern” should not be treated as equivalent to “AI found the correct meaning”. Those are different claims, and the gap between them is where much of the real scholarly work remains.
The likely role for AI
AI is not a dead end for lost languages. It can compress years of manual cross-referencing into minutes and let more people test ideas that would once have required far more institutional support.
But the source article’s larger lesson is that AI does not remove the two requirements that decipherment has always depended on. Researchers still need a real comparative anchor. They also need human review capable of separating a breakthrough from an appealing coincidence.
For Linear A and Etruscan, that means AI is best understood as an unusually fast assistant. It can widen the search, sharpen the tests, and expose possible relationships. Until stronger anchors or stronger confirmation appear, the oldest part of the puzzle remains human: deciding whether a pattern is evidence of meaning.