Researchers test how AI can help decode lost languages like Linear A

In short: More researchers are using AI to speed up work on undeciphered ancient languages, but AI cannot confirm meanings without strong outside clues.

What's going on

Some ancient writing systems are still not understood today. Two famous examples are Linear A from the Minoan civilisation on Crete, and Etruscan from pre Roman Italy. These are hard cases because there is no clear “anchor” to compare against, like a bilingual text (think of the Rosetta Stone as a side by side translation).

A recent example shows both the promise and the limits. In June 2026, a self taught engineer and amateur linguist claimed progress on Linear A. He started with a human guess that one word in a prayer might relate to a Semitic root meaning “to live” or “to dwell”, and then used AI driven scripts to test that guess across the known Linear A texts. He reported assigning sounds to about 40 signs and building a 408 word list, arguing Linear A was part of the Semitic language family, which includes Hebrew and Aramaic. The claim is still under review.

Researchers say this is the pattern that matters. AI acts like a very fast research assistant. It can check a hunch across thousands of characters in minutes, spot repeating patterns, and even predict missing symbols in damaged inscriptions. It can sometimes transfer knowledge from a related known language to an unknown one, like how Spanish can help you guess Portuguese.

What to watch

AI still hits a wall when evidence is scarce. Linear A has only about 7,500 surviving characters, and without an anchor, patterns can be coincidence. Watch for independent expert checks and peer review, because “AI found a pattern” is not the same as “we know what it means.”

Source: Arstechnica