Historians are increasingly turning to artificial intelligence technology to restore our forgotten past.
- Researchers use AI to restore ancient Greek texts
- Can machine learning help us translate unknown languages?
- Scientists develop an algorithm that can reassemble damaged archaeological artefacts
- AI uncovers a hidden geoglyph in the Nazca Lines
- How AI is helping us bring history back to life
History is of crucial importance to our understanding of the world. By providing valuable insights into our past, it allows us to make sense of our current situation and helps us better prepare for what the future brings. Sadly, much of our history still remains a mystery to us. In an attempt to solve this puzzle, researchers are increasingly turning to emerging technologies. Artificial intelligence (AI) has proven particularly useful in this quest, helping historians restore ancient texts that have been damaged over the centuries, translate lost languages, reassemble broken archaeological artefacts, and uncover hidden designs left by past societies.
Researchers use AI to restore ancient Greek texts
Ancient texts written on stone tablets, such as those found in Greece, provide invaluable insight into the history of past civilisations. However, being as much as 2,600 years old, these priceless inscriptions haven’t been immune to the ravages of time. They’re often damaged and littered with chips and cracks, sometimes even missing entire fragments of text. To fill these gaps and reconstruct the missing sections, historians typically rely on a discipline called epigraphy, which uses “grammatical and linguistic considerations, layout and shape, textual parallels, and historical context” to estimate what the text originally contained. As this is a very complex and time-consuming process, the team behind Google’s DeepMind AI joined forces with the University of Oxford to develop Pythia, the first ancient text restoration model that uses deep neural networks to recover missing characters from damaged text.
To train the system, the researchers first converted PHI, the largest digital corpus of ancient Greek inscriptions, into text that would be understandable to machine learning algorithms. Then, they created an algorithm that can guess the missing character sequences in damaged text, providing a ranked list of possible solutions and assigning a confidence score to each suggestion. From there, human experts can step in and use their own judgment and expertise to choose the most appropriate solution. To test the system, the researchers tasked it with filling the gaps in 2,949 damaged inscriptions. The system managed to complete the task with a 30.1 per cent character error rate, as opposed to a 57.3 per cent error rate displayed by human experts. Furthermore, the system was able to go through all of the inscriptions in a matter of seconds, while human experts needed 2 hours just to finish 50. “The reward is huge because it tells us about almost every aspect of the religion, social and economic life of the ancient world,” says Thea Sommerschield from the University of Oxford, who was part of the team.
Can machine learning help us translate unknown languages?
What if the ancient text isn’t damaged, but rather written in an unknown language? Over the years, archaeologists have discovered numerous inscriptions written in long-lost languages that are still waiting to be deciphered, keeping their true meaning hidden from us. One of the most famous examples is ancient stones and tablets found on the Mediterranean island of Crete. Dating as far back as 1800 BCE, they’re written in two different scripts, known as Linear A and Linear B. While Linear B was finally deciphered in 1953, Linear A still remains a mystery. To solve this problem, a team of researchers from MIT and Google’s AI Lab developed a machine learning system capable of translating lost languages.
To achieve this, the researchers used an innovative approach to machine translation that takes into account the way languages are known to evolve over time, eliminating the need for large datasets. Based on the idea that languages can only change in certain ways, the new approach focuses on “four key properties related to the context and alignment of the characters to be deciphered – distributional similarity, monotonic character mapping, structural sparsity, and significant cognate overlap.” The only drawback to this approach is that it requires you to know the progenitor language, or the language your lost language is related to. To demonstrate its capabilities, the researchers first had it translate Linear B and Ugaritic, two lost languages with known progenitor languages. The system was able to translate both languages with remarkable accuracy, correctly translating 67.3 per cent of Linear B cognates into their Greek equivalents. Sadly, as we don’t know what the progenitor language of Linear A is, this technique won’t be applicable to it, but it does represent a significant milestone in machine translation that brings us one step closer to solving the mystery of lost languages.
Scientists develop an algorithm that can reassemble damaged archaeological artefacts
It’s not just texts inscribed on stone tablets that have suffered damage over the centuries. The same can be said for countless other archaeological objects, most of which are in a poor or fragmentary state. As a result, archaeologists are often forced to reassemble these fragments manually, which can be a gruelling and tedious process. To address this issue and help historians solve the archaeological puzzle, a team of researchers from Technion – Israel Institute of Technology and the University of Haifa have developed an algorithm that can automatically reassemble fragments of archaeological artefacts.
When trying to restore damaged artefacts, archaeologists face three major issues: abrasion, colour fading, and continuity. Abrasion creates gaps between pieces and makes it harder to match adjacent fragments, while colour fading creates false edges, which can be difficult to distinguish the real edges and gradients. Finally, the irregular nature of the fragments produces continuity issues by offering an almost infinite number of possible configurations. To resolve these difficulties, the algorithm first scans and extrapolates each fragment, predicting what should surround it and identifying its neighbouring piece. Then, taking into account unique characteristics such as the gaps between pieces, colour fading, spurious edges, varying lengths of matching boundaries, and imprecise transformations, the algorithm assigns each fragment pair dissimilarity and confidence scores to find the best possible match. The algorithm was tested on various real archaeological objects from the British Museum and frescoes from churches around the world, successfully reassembling the majority of broken artefacts.
AI uncovers a hidden geoglyph in the Nazca Lines
Located in the desert area of Southern Peru, the Nazca Lines are drawings on the ground made by removing rocks and earth to reveal the white sand beneath. The lines join and overlap one another to form a wide variety of designs, ranging from simple geometric shapes to humanoid and animal figures. Despite numerous theories proposed over the years, their true purpose remains a mystery to this very day. Known as geoglyphs, they’re believed to have been made by the Nazca people that once occupied Peru, with some of the designs dating as far back as 100 BC. Thanks to the dry climate and winds that sweep away the sand, the lines remain well preserved despite their age. However, human activity and floods have obscured some of them, making it increasingly difficult to identify new designs. To solve this problem, a team of researchers from Yamagata University joined forces with IBM Japan to develop a machine learning algorithm that could automatically identify hidden designs in the Nazca Lines.
The algorithm was able to find a small, five-metre-tall humanoid figure holding a cane or a club. Believed to date to between 100 BC and 500 AD, the figure was discovered by analysing on-site surveying and aerial imagery data gathered by earlier studies. As there are less than 100 geoglyphs already identified, featuring wild variations in shape and size, teaching the algorithm what to look for proved to be quite challenging. “We specifically built techniques in the deep learning framework to learn and distinguish between these different patterns and sizes of the geoglyphs,” explains Akihisa Sakurai, a researcher from IBM Japan. In the end, the algorithm managed to identify several hundred potential candidates for new geoglyphs, which were then manually examined by the researchers, revealing the little guy with the cane.
How AI is helping us bring history back to life
Artificial intelligence technology has had a profound impact on a wide variety of industries in recent years, transforming almost every aspect of our lives in the process. While most of the talk surrounding AI has focused on how it will affect the human workforce, it can’t be denied that it’s found some very promising applications in certain areas. One of these areas is archaeology, where AI’s ability to analyse large amounts of data in a short amount of time and uncover hidden patterns is very useful. Whether it’s used to restore ancient Greek texts, translate long-lost languages, reassemble damaged archaeological artefacts, or uncover hidden designs in the Nazca Lines, AI is becoming an increasingly important tool in archaeologists’ arsenal, helping them bring history back to life.