Linear A, and why I want to take a crack at it

Early post. The approach I'm developing isn't ready to write about yet. This is background on the problem and why I think it's worth working on.

Linear A is a writing system from Minoan Crete. About 1,400 inscriptions survive, written somewhere between 1800 and 1450 BC. Nobody can read them. The language underneath is a complete isolate with no known relatives, no bilingual text, no Rosetta Stone equivalent. It's been undeciphered for 70 years. Most people who've worked on it seriously think it's going to stay that way. I got into it a few months ago and I genuinely can't stop thinking about it.

What the Minoans actually were

The Minoans weren't a backwater. They ran trade networks across the Bronze Age Mediterranean: copper from Cyprus, tin from Afghanistan, olive oil and textiles flowing through Crete to Egypt and the Levant and back. The palace at Knossos was one of the most sophisticated administrative centers in the ancient world, with indoor plumbing, multi-story architecture, and a bureaucracy complex enough to require a dedicated writing system to run it.

That's the thing about Linear A. It's not cave markings. It's a professional administrative script used by palace scribes to track commodities, record tribute, manage storage. The people who used it were running a real economy, and we've found their records. Tablets from Hagia Triada, Zakros, Knossos, and Akrotiri survive. We just can't read a single word of what they say.

When Thera erupted around 1628 BC, the ash preserved sites like Akrotiri perfectly. We have their painted walls, their furniture, their pottery, and Linear A tablets from the administrative rooms. We know what their storage jars looked like. We don't know what the labels on them say.

Why it hasn't been solved

The standard approach is phonetic matching. Linear A and Linear B share around 60% of their signs. Linear B was deciphered by Michael Ventris in 1952 and turned out to be an early form of Greek. So the obvious move is to take the known Linear B sound values, map them onto the matching Linear A signs, and see what comes out. The problem is that Minoan isn't Greek. Every phonetic mapping produces output that looks like nothing in any known language family. At this point the consensus is that Minoan is a genuine isolate.

The corpus makes it harder. 1,400 inscriptions isn't much. Most are short, many are damaged. The longest coherent texts are the Hagia Triada tablets, storage inventories and tribute records. We can parse the structure: numbers, commodity logograms, totals. We just can't read the words between them.

What I've done so far

I've downloaded the full corpus from the SigLA database and lineara.xyz, which have every known inscription digitized with photographs, sign-by-sign annotations, and site metadata. 847 inscriptions from SigLA alone. I've been going through the archaeological literature, particularly the palace-economy work on Hagia Triada and the sign-frequency analyses done by Yves Duhoux and John Younger.

The partially-readable parts are useful. The numbers use the same decimal system as Linear B. Common commodity logograms like grain, figs, wine, wool and livestock can be identified by context and pictographic similarity. When a tablet shows a sheep-like logogram followed by a number, it's probably a sheep count. That gives you footholds even when the phonetic values are unknown. I have an approach I want to try, but I'm not writing about it until I've tested it.

Why now is actually different

Most serious Linear A work has come from linguists and archaeologists working the phonetic-matching angle. The computational direction hasn't been explored much. The tools available now are different from even five years ago: you can do pattern recognition across a small corpus in ways that weren't practical before, and you can embed everything known about Minoan material culture, trade networks, and palace economics and actually use it. We know a lot about the Bronze Age Mediterranean economy. What moved where, what was stored, who traded with whom. That's real context for reading administrative records, even ones you can't decode phonetically. It just hasn't been used that way.

Whether any of this works, I don't know yet. But it's one of the last unsolved writing systems from the ancient world and the tools are different now. That's enough for me to want to try.

If you work in this area and want to talk, my email is contact@phnix.dev.