An amateur used AI to claim a 408-word Linear A lexicon. Linguists aren't buying it.

For a century, two ancient scripts have been the locked safes of historical linguistics. Linear A, the script of Bronze Age Minoan Crete, is a true language isolate—no Rosetta Stone, no known descendant. Etruscan, pre-Roman Italy’s tongue, is only slightly less opaque, yielding a partial vocabulary from tombs but guarding its grammar jealously. These are precisely the kinds of puzzles where AI’s pattern-matching muscle could, in theory, supercharge human intuition.

A dramatic claim from June 2026 shows the seductive promise of this approach. A self-taught engineer, pairing his own linguistic hunch with AI-built scripts, proposed that a single word in a Linear A prayer was Semitic for “to dwell.” From that one seed, his system extrapolated outward, assigning sound values to 40 signs and generating a 408-word lexicon. His conclusion rewrites history: Linear A, he says, isn’t an isolate at all, but an extinct branch of the Semitic language family that includes Hebrew and Aramaic.

But here’s the catch that separates pattern generation from actual decipherment. The AI didn’t have the breakthrough idea; the human did. The software was a testing engine for a speculative guess, not an independent oracle. It found what it was programmed to look for, and if that initial “to dwell” seed is wrong, the entire structure is a house built on sand. Real decipherment requires an anchor—a bilingual text or a known relative—that Linear A simply lacks.

What this episode really exposes is the yawning gap between computational suggestion and scholarly proof. The claim hasn’t swayed the field, and it likely won’t without external verification that no algorithm can conjure. AI can tirelessly test a theory against a corpus, but it can’t tell you if the theory is any good to begin with. For now, Linear A remains silent, waiting for a key no amount of processing power can fabricate from thin air.