AI Tool Interprets Ancient Cuneiform Writing System
The world has been swept up in the AI revolution, with tools like ChatGPT and Google Bard transforming how we access and process information. These tools have made even complex tasks simple, and now a team of archaeologists and computer scientists from Israel are using AI to uncover the secrets of the past. Their groundbreaking project aims to translate a 5000-year-old writing system into English in mere seconds, making it a highly advanced version of Google Translate.
We are translating an Akkadian cuneiform book
Nuolenpää is a more than 5000-year-old logosyllabic writing with typical wedge-like impressions. It was in use from the Bronze Age to the Common Era, which spans nearly 3,000 years, but has not been used for 2,000 years. Today, museums, libraries and galleries hold millions of clay tablets inscribed with arrowheads. However, due to the extremely small number of Akkadian readers, translating them into a comprehensible language has been difficult, but not anymore.
A tool similar to Google Translate was developed as a thesis at Tel Aviv University. In a study published in the peer-reviewed journal PNAS Nexus, Oxford University Press, Gai Gutherz, a computer scientist who was part of the program’s development team, said: “It’s so amazing that I don’t need to understand Akkadian at all to turn [the tablet] and find out what’s behind the arrowhead. is. I can only use an algorithm to understand and discover what the past has to say.”
Developing an AI tool
According to the study, the AI tool uses a mathematical formula known as Neural Machine Translation (NMT), an artificial neural network used to translate text from one language to another. Although it is difficult to judge what a good translation is, the researchers used a machine translation evaluation tool called Best Bilingual Evaluation Understudy 4 (BLEU4) to determine accuracy.
The exact BLEU4 score ranges from 0 to 100, with 0 being the lowest and 100 being perfect, which is unprecedented. The Akkadian translation tool returned a score of 36.52 for cuneiform to English and a score of 37.47 for transliterated cuneiform to English, which Gutherz says is “fairly good”.
Against obstacles
Since the language is so old, the biggest challenge was finding material to train the AI model. The researchers combed through images of cuneiform tablets and even sampled the University of Pennsylvania’s online database called the Open Richly Annotated Cuneiform Corpus (ORACC). The AI model was trained with 50,544 sentences, while 2,808 sentences each were reserved for validation and testing purposes.
In addition, Akkadian was used for over 3,000 years, leading to huge differences in cuneiform symbols and even dialects during this time, adding to the complexity of translating the old writing system.
Where do you get it?
If you want to try this ground-breaking AI tool, an early demo version is available on The Babylon Engine. In addition, if you want to develop your own tool, you can find its source code on GitHub in Akademia and Collaboratory.
