A self-taught engineer used AI-built programming scripts in June 2026 to assign phonetic values to 40 signs of Linear A, the undeciphered Bronze Age script of Minoan Crete, and compile a 408-word lexicon arguing the language is an extinct member of the Semitic family. The claim, which places Linear A alongside Hebrew, Aramaic, and Ugaritic, has not been peer-reviewed. It arrives after roughly a century of failed human-only attempts.
The mechanism is the story. The engineer began with a single hypothesis — that one word in a prayer inscription derived from a Semitic root meaning "to dwell" — and used AI to test that sound pattern against the entire Linear A corpus in minutes rather than months. Linear A's full surviving corpus runs to about 7,500 characters, small enough to fit on a single screen, which makes exhaustive cross-checking newly tractable.
“Linear A is described as a "language isolate" because it has no confirmed link to any known language, living or dead.”— Jane Adkins, PhD candidate, School of Computing, Dublin City University
Every ancient language deciphered to date has needed an anchor: a bilingual text like the Rosetta Stone or a confirmed relative language. Linear A has neither, which is why linguists classify it as a language isolate. Etruscan, spoken across Italy before the rise of the Roman Empire, has fared only slightly better — a partial vocabulary has been assembled from short funerary inscriptions, but grammar and deeper meaning remain out of reach.
Key facts
- 01A self-taught engineer used AI scripts in June 2026 to assign values to 40 Linear A signs and compile a 408-word lexicon.
- 02The claim argues Linear A belongs to the Semitic language group, alongside Hebrew, Aramaic, and Ugaritic.
- 03Linear A's entire surviving corpus is roughly 7,500 characters — short enough to fit on a single screen.
- 04Ugaritic, spoken in Syria between 1300 and 1190 BC, has been productively analyzed with AI cross-lingual transfer because its language family is known.
- 05The analysis was published July 29, 2026 by Jane Adkins, a PhD candidate at Dublin City University's School of Computing.
Writing in The Conversation on July 29, 2026, Jane Adkins, a PhD candidate at Dublin City University's School of Computing, argues the June breakthrough shows what AI is genuinely good at and where it stalls. The idea belonged to the human. The model was a research assistant that cross-checked one hypothesis against thousands of characters at a speed no scholar could match by hand.
The strengths are concrete. Large-scale pattern testing lets a researcher verify whether a hunch about one sign holds across an archive in minutes. Models can also spot repeated sequences a human eye would miss and predict likely missing characters in damaged inscriptions. A technique called cross-lingual transfer, where a model trained on a known language infers patterns in a related unknown one, has already produced results on Ugaritic, an extinct Semitic language spoken in the coastal city of Ugarit in what is now Syria between roughly 1300 and 1190 BC.
“AI is fantastic at spotting patterns, but human insight is the key.”— Jane Adkins, PhD candidate, School of Computing, Dublin City University
The ceiling is equally concrete. Statistical pattern matching cannot manufacture meaning from nothing. It needs an anchor — a known language family or a bilingual text — to distinguish meaningful patterns from coincidence. With only 7,500 characters of Linear A to work from, almost any hypothesis can find scattered matches to support it, which is why the June claim requires independent expert scrutiny rather than a statistical confidence score.
Even a hypothetical AI system trained to fluency on a corpus like Etruscan or Linear A would face a harder problem than translation. It could learn which signs follow which and which words cluster together without ever knowing what any of them refer to in the physical world. Fluency and meaning are not the same object, and there are no native speakers left to check the answer against.
That verification problem is the deepest one in the field. For a normal translation, a researcher checks against native speakers, other texts, or an expert consensus built over decades. For a genuinely undeciphered language, none of that infrastructure exists. Adkins notes the distinction between an AI finding a pattern and an AI finding the correct meaning is easy to blur — and it is exactly the blur that peer review is supposed to catch.
“AI's role in deciphering languages will stay what it is today: a very fast assistant to a very old, very human puzzle.”— Jane Adkins, PhD candidate, School of Computing, Dublin City University
None of this makes AI a dead end for lost languages. It compresses years of manual cross-referencing into minutes and lets amateurs attempt problems that were previously locked behind institutional resources. The June Linear A claim, whatever its ultimate fate, was almost certainly not going to be attempted by a single unaffiliated researcher in a pre-AI world.
For the AI industry, the case is a useful counterweight to the broader agentic-research pitch. Language models are being marketed as autonomous scientists across domains from drug discovery to code generation. The Linear A example is a clean illustration of the current division of labor: the model accelerates verification of a human hypothesis by orders of magnitude, and produces nothing without one. In the fields where an anchor exists — a known family, a parallel corpus, a benchmark — that acceleration is worth a great deal. In the fields where it doesn't, no amount of compute invents one.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
