Tech

Could Game Designers Crack Egyptian Hieroglyphics With Machine Learning?

Developers from "Assassin's Creed" are trying to make life easier for historians.
Image via Shutterstock

Note: the author was flown to London and fed wine by Ubisoft.

For a long time Egyptian hieroglyphics were unreadable. There are records from the ninth century of scholars, both in Europe and the Middle East, baffled by the sheer number of these ancient symbols—appearing in seemingly random patterns. By the 15th century, we were no further along. A German scholar by the name of Athanasius Kircher basically claimed hieroglyphs were impossible to decipher because they "cannot be translated by words, but expressed only by marks, characters, and figures."

Advertisement

But this all changed in 1799 when Napoleon's army invaded Egypt, and an army corporal found what we now refer to as the Rosetta Stone. The stone was inscribed with a single message written in three different languages. The top and middle texts were in two forms of Ancient Egyptian, but vitally the bottom text was in Ancient Greek. Finally, the modern world had a key. By the 1820s, hieroglyphics were cracked.

But, even to this day, translation remains a messy, painstaking process. Because as Egyptian historian Perrine Poiron explains, hieroglyphs are just so uniquely complicated. "Let's say we take the word 'duck,'" she explains, sitting in the Egyptian section of the London Museum. "If I write the word 'duck,' using only a stroke, it means a duck but if I use the duck and I put a sun disk on it, it means 'sun beam.' So the duck can mean 'duck' but it can also mean 'sun.' Hieroglyphs are reliant on context."

The Rosetta Stone, housed at the London Museum

To make matters worse there are over 1,000 individual hieroglyphs, originating from several historical periods, all able to be read in both rows and columns. Translations are also open to interpretation, so academics often need to cite their work, explaining why they chose particular translations over others. All in all, translations can take weeks and produce results that are eternally open to debate.

For a team of video game designers, accustomed to technology solving everything, none of this seemed good enough.

Advertisement

Assassin's Creed: Origins is the 10th instalment in the game franchise about history. It's set in Cleopatra's Egypt, around the turn of the first century, and researchers working for the game's design team were surprised to learn there wasn't a faster way to decipher ancient texts.

"We thought that surely technology would have improved the translation of hieroglyphics," explains Pierre Miazga, who is a project coordinator at Ubisoft. "But it remains something that takes time, and we wanted to see if we could help."

Close up on the stone. The text at the bottom is in Greek, while the top is the Egyptian Demotic script

Ubisoft, creator of Assassin's Creed and one of the world's largest video game companies, had connections at Google and the two companies set about trying to use machine learning to decipher texts. That project became known as the The Hieroglyphics Initiative, and is now set to release an open source decypher tool at the end of the year.

Powered by machine learning, or "Deep Learning" as Google calls it, the technology will use a fairly new and complicated process to read hieroglyphs. The "deep" in Deep Learning refers to the layers of processing networks that relay information back and forth, self-checking and correcting until a consensus is established. If you want a more thorough technical description, there's a good explainer here.

Alex Fry, project director for the Hieroglyphics Initiative, explains that the first step is to program the technology with a database of all known hieroglyphs and their variations. Then the technology can cross-reference images of hieroglyphs with its database, in the same way that Google Images can look for cats, even though all cats look different.

The order of the hieroglyphs will then need to be established to determine whether they're to be read in rows or columns. Blocks of symbols will then be identified, so that individual symbols are read in context. "Once we've done that we're much closer to a piece of English or piece of French and then we can use the translation approach."

The developers are admittedly cagey about how the technology will appear. They say an app is unlikely, and only that they aim to make translation faster, but not actually replace the role of the traditional translator. So for historians like Perrine Poiron some manual work will still be necessary.

"I also don't know in which form this will appear or how or when but I am confident that they can do something really useful," she says. "Just anything to speed the process up would be incredible."