Google AI Decodes Broken Greek Texts Better than Humans
Artificial Intelligence (AI) is expected to revolutionize the world in the coming decades. An example of this comes from Google’s AI research division (DeepMind) which has used the technology to decipher broken and fragmentary ancient Greek texts. The DeepMind project has proven that it is superior to humans when it comes to understanding and deciphering ancient texts that have puzzled researchers for years.
Google’s artificial intelligence (AI) research arm, DeepMind, and the University of Oxford conducted a project that involved the analysis of Greek inscriptions . Many of them date back millennia and they are often broken and missing letters. Inscriptions are typically deciphered by epigraphers, but the work is not always successful, often because of missing words.
Ancient Greek inscription containing account of the supervisors for the construction of the chryselephantine (gold and ivory) statue of Athena Parthenos in the Parthenon by the sculptor Phidias, 440/439 BC. New Acropolis Museum, Athens. (Tilemahos Efthimiadis/ CC BY SA 2.0 )
The Google AI Model Called Pythia
According to the New Scientist , ‘Yannis Assael at DeepMind and his colleagues trained a neural network, a type of AI algorithm, to guess missing words or characters from Greek inscriptions’. The neural network was named Pythia, after a Greek priestess who communicated the prophecies of Apollo at the Oracle of Delphi . Newsweek reports that Pythia is the ‘first ancient text restoration model’.
- Greek Gods: The 7 Core Males Exposed
- Garden of Eden Depicted in Ancient Greek Religious Art
- High-Technology Discovered in Classical Mythology Reveals The Ancient Origins of Artificial Intelligence
The model was designed to interpret ancient Greek words and writing . The project involved trying to decipher fragmentary inscriptions that are between 1500 and 2000 years old. Pythia learned approximately 3 million ancient Greek words and inscriptions from artifacts and relics. It was then instructed to apply what it learned to predict ‘the missing characters or words on a new set of damaged relics,’ reports the Observer.
Ancient Greek man with wax tablet. Painting by ancient Greek painter Douris (about 500 BC). ( CC BY SA 3.0 )
A study was carried out to test how effective the neural network was. The algorithm was placed in a ‘head-to-head contest against historians to guess missing words in 2,949 damaged inscriptions,’ according to the Observer. Remarkably the neural network was able to decipher the inscriptions more accurately than the human experts. Pythia had an accuracy of almost 70%, compared to the historians who were only right just over 40% of the time. Moreover, the neural network completed its work in seconds, while the experts took almost two hours.
A Better Understanding of Ancient Greek Texts
The research project shows the potential of AI in historical research. By deciphering broken inscriptions the technology could lead to new insights on ancient Greek society , culture, and politics. Thea Sommerschield, a University of O xford historian, who worked on the project, stated, “The reward is huge because it tells us about almost every aspect of the religion , social and economic life of the ancient world,” according to the Observer. The new ancient text restoration model could open ‘up the field of digital epigraphy and ancient text restoration’ reports Newsweek.
However, it is not expected that the neural network will replace academics or historians anytime soon. The algorithm is designed to provide up to 20 suggestions with regard to the meaning of a text. The human historian will then select the one that is correct, based on their experience and knowledge. Pythia is regarded as being an assistive tool for researchers as they restore inscriptions and decipher them. Humans would also be needed to piece together any broken inscriptions for the AI to decipher.
- Artificial Intelligence Inching Closer to Deciphering Long Lost Languages
- Easy as Alep, Bet, Gimel? Cambridge Research Explores Social Context of Ancient Writing
- Egyptian Child’s Greek Homework from 1,800 Years Ago Goes on Display
The Potential of AI Programs
While Pythia may be very impressive it does not get it right every time. Philippa Steele of the University of Cambridge believes that “the highest success rates would be achieved when… we are just missing small parts of a long text,” according to the New Scientist . The technology is also likely to be most effective when there are similar complete inscriptions available to researchers.
The latest project is another example of the power and potential of Google’s DeepMind project, which has already astonished the world. In 2017, its AlphaGo program was able to easily beat professional players at the notoriously complex game known as Go.
Lee Sedol (B) vs AlphaGo (W) – Game 1. (Wesalius/ CC BY SA 4.0 )
It is possible that in the future the lab will develop models that could decipher incomplete inscriptions in other languages . This could help researchers to understand other civilizations and cultures from the past.
By Ed Whelan