One Million Words the World Has Never Read
Somewhere in Vienna, there are tens of thousands of papyri that nobody has read. Not in modern times. Not ever, since they were written. Ancient Greek, pressed into plant fiber, waiting.
Worldwide, an estimated one million Greek papyri sit in the same situation. Unread. Unknown. Full of words.
If that doesn't make your word-lover heart ache a little, you might be on the wrong website.
Enter Apollo
The Austrian Academy of Sciences, working with Mistral AI and Sail Reply (a Reply Group company), has built something called Apollo. Named after the Greek god of light and patron of the arts and sciences. The name fits.
Apollo is described as the world's first advanced multimodal LLM for an ancient language. It was trained on the largest digital corpus of historical Greek assembled to date. Its job: read what humans haven't had time to read.
The project is led by Anna Dolganov, an ancient historian and papyrologist at the Austrian Archaeological Institute of the OeAW. A papyrologist is exactly what it sounds like: someone who studies papyri for a living. Imagine making Ancient Greek your whole specialty and then getting to build AI to help crack a million unread documents. That's a good career arc.
What Apollo Actually Does
Two things, mainly. Advanced searching through ancient Greek texts. And automatic text restoration.
That second one is where it gets interesting for word people. Ancient papyri are damaged. Torn. Stained. Parts of words missing, parts of sentences gone. Text restoration means filling in what's lost based on context, grammar, and patterns across the entire corpus of historical Greek.
Apollo targets hundreds of thousands of undeciphered papyri and inscriptions. Work that previously took years can now be captured in hours.
Why This Is a Harder Problem Than It Sounds
Here's something worth sitting with: LLMs have not previously been developed for a historical language evolving over many centuries. Ancient Greek isn't a snapshot. It changed dramatically across the centuries it was in use. The same word could mean different things depending on when and where it was written.
That's a harder problem than modern language. Training a model on a living language means you're hitting a moving target, but at least the target is still moving. Ancient Greek is a stationary target spread across an enormous span of time, and huge portions of it were never digitized at all.
The Papyrus Collection of the Austrian National Library alone holds tens of thousands of unread Greek papyri. Those documents could contain anything. Letters, records, notes, things people wrote when nobody important was watching. The everyday words of ordinary people who never expected anyone to read them two thousand years later.
Apollo is about to read them.
The Words We Don't Know Yet
You already use Ancient Greek constantly. Every crossword puzzle is half Greek roots whether you realize it or not. "Cryptic." "Lexicon." "Enigma." The vocabulary of word games is built on a language that stopped being anyone's native tongue over a thousand years ago.
One million unread papyri could contain vocabulary, phrases, and usages that scholars haven't seen before. Words that dropped out of the written record. Regional variations. Slang. Things that got lost.
This could mean new etymology. New answers to questions linguists have been arguing about for decades. Or it could mean more shopping lists and tax records. History is mostly accounting.
But even a tax record tells you something about language. How people wrote when nobody important was watching. What words they reached for without thinking.
A Very Old Corpus
Apollo was trained on the largest digital corpus of historical Greek to date. That's a sentence that would have sounded like fiction twenty years ago and is just a Tuesday now.
The fact that an ancient language has enough digitized material to train a large language model at all is remarkable. Decades of scholarly work, done by people who probably had no idea an AI would eventually show up to use it.
Now it has. The god of light, reading in the dark.
Source: Languagehat