A model brute-forced Linear A

Tom Di Mino, a self-taught AI engineer and amateur linguist based in the Hudson Valley, says he has deciphered Linear A, the Bronze Age Minoan script that has resisted translation for more than a century. His work is under review by linguists at Rutgers and Cambridge. The stakes are real: when the related script Linear B was deciphered in 1952, it made the front page of the New York Times.

The claim rests on one sign. Di Mino had read about Linear A for seven years and began working on it seriously in January. On May 22 he was looking at a set of prayer inscriptions from Minoan peak sanctuaries, all following the same fixed formula. Every word in each line was already readable from the sounds Linear A shares with Linear B, except the first, which was the same verb root in different regional spellings across five sanctuary sites. That root carried five known Linear B signs plus *301, a sign that appears only in Linear A and had no accepted sound value. Di Mino read *301 as “na,” which gave him the root nawaya, “to dwell.” It matches the Semitic triconsonantal root N-W-Y, used across Hebrew, Akkadian and other Semitic languages for words meaning to dwell or inhabit.

Some context on why the script stayed closed. Linear A appeared around 1800 BC and was used until 1450 BC, when Mycenaean Greeks took Crete and adapted its symbols into Linear B. The two scripts share about 60 core syllabic signs, both built mostly from consonant-vowel pairs, and both use logograms for whole words. Linear B was deciphered as Greek in 1952 by Michael Ventris, an architect and cryptographer building on Alice Kober’s statistical work, so scholars could guess the sounds of the shared signs. They could not say what those sounds meant, because the underlying language was unknown, and 13 Linear A signs had no agreed sound value at all. Many surviving Linear A inscriptions are commodity inventories, which give little grammar to work with.

Di Mino’s reading makes Linear A an extinct Semitic language, a precursor to Biblical Hebrew, Arabic and Aramaic in the way Latin precedes Italian. The peak-sanctuary prayer then looks like later Hebrew prayers, addressed to a goddess. He is not the first to call Linear A Semitic; Cyrus Gordon argued it in a 1957 Antiquity paper that never produced working translations or won acceptance. Di Mino says he is the first to tie the Linear A inscriptions to Hebrew prayers specifically, and that his account of Linear A logograms also clears up problems in some existing Linear B translations, which he treats as a cross-check on the result.

The model’s role is narrower than “brute-forced” suggests. Di Mino used Claude Code to build a suite of Python scripts that query, cross-reference and organize the digitized Linear A corpus, drawn from the GORILA and SigLA databases. That let him test hypotheses systematically across the full body of inscriptions at a scale that would have been impractical by hand. The scripts made the search tractable; the reading of *301 as a Semitic dwelling-root came from a person who reads eight languages, including Ugaritic, and had visited Crete twice.

So far the work has produced proposed readings for 40 signs, including 13 whose phonetic values were previously unknown, plus resolved sound values for five Linear B signs that had stayed open. There is a lexicon of 408 Linear A terms translated into English, and a nine-page draft titled “Ya Diktu: Grammar of the Minoan Peak Sanctuary Libation Formula,” which Di Mino may expand into a peer-reviewed submission. Whether it holds is now up to the reviewers at Rutgers and Cambridge.