Thanks to artificial intelligence, the mysterious dialect of our ancient ancestors could finally be fully deciphered.
A million cuneiform tablets still exist in the world, experts estimate, but these writings left by the ancient Mesopotamians require the tedious work of archaeologists to translate and catalog their contents.
That’s estimated 90 percent of the cuneiform texts remains untranslated.
But now a team of German researchers has devised a new way to train computers to recognize cuneiform script and even make the contents of millennia-old tablets searchable like a website, making it possible to digitize larger libraries of these ancient texts and to put together.
This could unlock previously unknown details about ancient life, as the tablets contain details of achievements as important as the construction of temples, down to petty squabbles like customer service complaints.
The German academics trained an AI in two cuneiform languages, Sumerian and Akkadian.
Sumerian was spoken about 5,000 years ago and was eventually replaced by Akkadian, but both languages were used in writing until the beginning of the Christian era in Mesopotamia, which occupied modern-day Iraq and parts of what became Iran, Kuwait and Syria. and Turkey.
There are an estimated one million cuneiform tablets in the world. AI tools may allow scientists to quickly and easily search their content
The remaining cuneiform tablets are therefore not only written in multiple languages, but are also thousands of years old.
The wedge-shaped cuneiform signs that formed the basis of written languages in ancient Mesopotamia were carved into clay tablets, making them three-dimensional.
Combined with the fact that the ancient script has been weathered by time and use, their qualities can make them difficult to scan into a computer so that historians and archaeologists can use them for research.
Now, using 3D models from about 2,000 tablets, they’ve trained a computer program to scan and transcribe their text — like using your smartphone’s camera to turn a handwritten note into a text document.
The purpose of this study was not to translate the content of the tablets, but to enable other researchers to do so more easily.
And not only has the ravages of time worn away their unfired clay surfaces, making translation more difficult, but a single tablet or even a small set of text can be difficult to understand without context – like trying to understand a book by reading one sentence on one page . right in the middle.
The new AI program could help fill the gaps by allowing translators to work more efficiently.
“Until now, it has been difficult to access the contents of many cuneiform tablets at the same time – you actually have to know exactly what you are looking for and where,” says senior study author Hubert Mara, assistant professor at Martin Luther University Halle. Wittenburg in Germany, in a rack.
The tablets they used to train their computer program came from an open-access set of 3D scans, which included Sumerian cuneiform tablets – from the earliest known civilization in southern Mesopotamia, which is now south-central Iraq.
The new system not only helps researchers decipher the contents of cuneiform tablets, but also allows them to create a kind of searchable document.
The contents of these tablets will help humanities scholars give us a better understanding of what life was like in ancient Mesopotamia.
‘You can find everything on it: from shopping lists to court decisions. The tablets offer a glimpse into humanity’s past, several millennia ago. However, they are heavily weathered and therefore difficult to decipher, even for trained eyes,” Mara said.
3D scans of ancient cuneiform tablets were used to train an AI to recognize not only the script’s characteristic wedges, but also its symbols
Part of the challenge was training the AI to recognize the wedges and glyphs that make up the cuneiform script.
The researchers fed the program with 21,000 characters and 4,700 wedges, creating a new data set that can be used by other researchers wanting to study cuneiform writing.
After training the AI, they tested it on other tablets to see how reliable it was.
They found that it could accurately detect cuneiform wedges and glyphs with an accuracy of about 76 percent.
And it didn’t just work with high-quality 3D scans.
“We were surprised to find that our system works well even with photographs, which are actually a poorer source material,” says Ernst Stötzner, a student in Mara’s lab.
Ernst Stötzner performs a 3D scan of an ancient cuneiform tablet, wearing gloves to protect the millennia-old object
Stötzner and Mara’s team plans to use an even larger number of tablets to train their AI and achieve more accurate measurement.
They suspect that the relatively small number of tablets might have limited its accuracy.
By comparison, another AI trained to recognize another cuneiform-based language achieved 90 percent accuracy.
Another option is to split the tablet images into smaller segments so that the AI can process a smaller amount of information at a time.
The study appeared in the November issue of Eurographics Workshop on graphics and cultural heritage.