Tens of thousands of cuneiform tablets are sitting around, just waiting to be translated. It’s not an easy job; the ancient language is based on wedge-shaped pictograms and includes more than 1,000 unique characters that vary by era, geography, and individual writer.
But decoding the pictograms could be a culturally and historically significant task. Cuneiform arose about 5,000 years ago in Mesopotamia, in what is now Iraq. It is one of four known pristine languages — writing systems with no known influences from any other. Some translated cuneiform tablets have revealed contents as banal as a record of inventory for shipping. Others have been more profound — like the “Epic of Gilgamesh," the first known written work of literature.
Those translations, done by a relatively few individuals who know the language, required a lot of labor — and perhaps some guesswork. Decoding such complexity would be the perfect job for artificial intelligence, thought some Cornell University researchers, who, with colleagues at Tel Aviv University, created a system to do just that, they report in a paper to be presented at an April 2025 conference.