Lost And Found: AI Helps Unlock Ancient Languages

Artificial Intelligence is at the forefront of technology, and while it’s usually thought of in futuristic terms, AI is providing incredible insights into the ancient world. Just over a year ago, Futurism reported that MIT Computer Science and Artificial Intelligence Laboratory researchers created an AI that can automatically decipher long lost languages. But things have gotten even more exciting in the last few months. AI and ancient languages are once again making headlines. In March, reports revealed that a DeepMind-developed AI tool is helping historians to accurately restore and date ancient Greek texts.

AI Goes Ancient Greek

Ancient Greek texts aren’t usually the first thing that comes to mind when people hear ‘AI.’ Advancements in medicine, self-driving cars, autonomous robots, arms, and the metaverse are the norm. AI private exchange has specifically become a major talking point in the medical and financial sectors. This technology allows for the quick and efficient analysis of information and provides an encrypted channel for the transfer of confidential data. In a world where data is the new oil and cybercrime is on the rise, AI private exchange is making data easier to analyze, utilize, and secure.

But now, AI has another function. Enter DeepMind’s Ithaca AI tool. According to New Scientist, historians have been using the tool to restore ancient Greek texts with an accuracy rate as high as 72%. They’re also using the tool to date inscriptions to within 30 years of their actual age and to predict where in the ancient Mediterranean world those texts were written with an accuracy rate higher than 70%.

Ca’ Foscari University of Venice’s Thea Sommerschield told the publication that inscriptions are evidence of past civilizations’ history, language, society, and thought. The problem is that most surviving inscriptions are fragmentary or illegible due to centuries of damage. Some inscriptions may also have been moved far from their original locations. Finding the relevant data that can help historians restore, date, and prove the origins of those inscriptions can be exceedingly difficult.

Findings from the Research

Sommerschield and her colleagues worked with DeepMind researchers to train the machine-learning Ithaca AI tool to perform all three tasks. The team used approximately 60,000 well-studied ancient Greek texts from around the Mediterranean written between 700 BC and AD 500 in the training. The team covered some characters in the text and then compared the tool’s predictions for the missing characters with the actual inscriptions. The team then used a data set of almost 8,000 well-studied inscriptions to test the tool’s performance. This was done alone or with two ancient historians. The tool’s solo performance had a 62% accuracy rate with text restoration, while the ancient historians’ attempts at text restoration had an accuracy rate of approximately 25%. However, when those historians used the tool, their accuracy rate increased to 72%.

New York University’s Tom Elliott told the publication that, while the work was groundbreaking and important, the tool needed further testing. He added that people who use the tool will require training as well as technical support. DeepMind’s Yannis Assael said he hopes ancient historians will find the tool easy to use, as they can type the text into an online interface, and the tool will produce various visualizations they can use.

Language AI Is Big News

The news about DeepMind’s Ithaca AI tool shouldn’t be much of a surprise to anyone who has kept up with AI predictions and trends. Last year, Forbes predicted that language AI will take the spotlight in 2022. According to the publication, more startups are being funded in natural language processing (NLP) than in any other AI category. If researchers can accurately automate language, it will open an array of opportunities across the economy.

Share on: