The romance of the ancients is in the Delphic scraps of the past

Academics at Oxford have devised a way of using Artificial Intelligence to read ancient Greek inscriptions. Rather than slave over deciphering the illegible or missing bits of text, they can now ask a computer to do the work for them.

Their ingenious machine, named Pythia after the priestess of Apollo at Delphi, has been taught to predict which letters are most likely to have filled gaps left in stone or ceramic inscriptions after decay or damage. What’s more, Pythia has been shown to make fewer mistakes than do humans, who have been honing the art of epigraphy for centuries.

This is obviously a brilliant feat. Deciphering inscriptions can be painstaking and timeconsuming work. But is there not also something slightly deflating about being able to remove the mystery from something so old at the click of a button?

I have always found romance in the fragments of the past. It is in the little gaps and infelicities of ancient texts that the imagination is stirred. Not knowing exactly what an ancient writer meant prompts you to consider the possibilities. The more deeply you ponder, the more engaged you become in the pattern of words before you.

As it is, languages are full of ambiguities, as poor Nigel Molesworth memorably discovered at school. Cursed with the unedifying challenge of Latin before “brekfast hav even settled”, he found himself having to translate from English back into Latin sentences such as, “The ramparts of the enemy are long.” “How long, that is the point?” he wondered. Leagues, miles, steps, yards?

The Pythia herself spoke with such ambiguity that people would often misinterpret the oracles she gave them. The wealthy King Croesus of Lydia famously made war on the Persians after she said he would destroy a great empire. Little did he suspect she meant his own.

Unlike her namesake, the Pythia machine delivers a series of hypotheses and “confidence level” percentages for each. When she is as meticulous as this, one cannot help but wonder whether she might inadvertently correct an ancient writer’s mistakes. Names and abbreviations are frequently misspelt on vase paintings and tombstones. Per humanos errores. Can an algorithm know to recover the shoddy handiwork of a poorly paid engraver?

Pythia isn’t deciphering the Enigma code. The stakes are not that high. But you wouldn’t want tell-tale details to be lost through overcorrection. A missing consonant or incorrect vowel can offer tantalising clues as to the dialect of the person behind the inscription. Obliterating mistakes like these would be the equivalent of replacing “Down with Skool” with “Down with School”.

It will, of course, have its uses. Greek is an inflected language – words change depending on their function in the sentence. A letter or two can make all the difference to the sense of a line. If Pythia can resolve whether, say, a man was killed or was the killer, she’ll have done a good day’s work.

Pythia may be a godsend to epigraphers – and we should look forward to seeing what she recovers – but even machines need a break. With thousands of inscriptions to work on, the least she could do is leave the odd mystery for us humans to ponder.