SDSU researchers used AI to compare writing systems across distant regions.
Peer-Reviewed Publication
image:
From left, characters in the Ethiopic (portions only), Armenian, Georgian and Caucasian Albanian alphabets.
view more
Credit: Daniel Zemene, Esatu Zemene, Atharv Sankpal, Eskinder Sahle, Vyshak Athreya Bellur Keshavamurthy, Samuel Kinde Kassegne, Machine learning techniques for exploring influence, commonalities, and shared origin of scripts: cases of Ethiopic, Armenian, Georgian, and Caucasian Albanian scripts, Digital Scholarship in the Humanities, 2026;, fqag029, st1yle=”box-sizing:border-box;margin:5px 0px 15px”>With artificial intelligence (AI) as an essential tool, San Diego State University researchers have discovered surprising similarities among ancient writing systems from Africa and the Caucasus region of Eurasia. Their study suggests the Armenian alphabet may be more closely related in structure to the ancient Ethiopic writing system than linguists and historians previously thought.
For many years, historians noticed some Armenian, Georgian and Caucasian Albanian letters look similar to letters from Ethiopic, also known as Ge’ez, a writing system developed in the Horn of Africa more than 1,600 years ago.
Most of these early studies, however, relied on scholars’ own visual inspection of the letters to determine whether they appeared alike.
Researchers from the Department of Mechanical Engineering in the College of Engineering tested this idea using AI instead of human judgment. They trained a computer program to study more than 28,000 images of Ethiopic characters so it could learn the basic shapes and patterns in the writing system. The program learned to recognize curves, straight lines, angles and the overall structure of each letter.
Importantly, the computer had no data on history, religion, geography or culture. It only looked at shapes. After learning the Ethiopic characters, the program compared them to letters from the Armenian, Georgian and Caucasian Albanian alphabets. It then calculated how similar the shapes were.
The results, published March 25 in Digital Scholarship in the Humanities, were striking.
Among the three alphabets tested, Armenian letters showed the strongest similarity to Ethiopic letters. Caucasian Albanian letters showed a moderate level of similarity, while Georgian letters showed some similarities but were less consistent. As a comparison, the researchers also tested the Latin alphabet — the one used in English — and found it showed much lower similarity.
“Our aim was to move beyond visual impressions that are difficult to test or replicate,” said Sam Kassegne, a professor of mechanical engineering and lead investigator. “By making our criteria explicit and mathematical, we introduced an objective computational approach that is easily reproducible. We believe that this reproducibility is the key contribution of our method.”
New findings
One of the most surprising findings was that the Armenian alphabet appeared almost as similar to Ethiopic as Ethiopic is to its own earlier version. That suggests the resemblance may not be accidental.
The Armenian alphabet was created around 405 CE. Around that same time, the Ethiopic writing system was expanding and becoming more widely used. Historical records show people from Ethiopia traveled to such places as Jerusalem, Egypt and Syria during this period. The creator of the Armenian alphabet, Mesrop Mashtots, also traveled through parts of the Middle East. While the study does not prove one writing system copied the other, it suggests cultural contact and influence between these regions may have been possible.
The study also shows how modern technology can help answer ancient questions.
We are already familiar with AI being used for self-driving cars and medical imaging. In this case, it was used to study the shapes of letters from ages ago revealing some level of historical cultural interactions. By teaching a computer to carefully measure similarities, researchers were able to move beyond the limitations of visual impressions and provide numerical evidence.
Daniel Zemene, an SDSU graduate student and AI and machine learning researcher at SDSU’s NanoFAB Lab, emphasized the broader implications of the findings.
“What makes the research significant is that computational geometry and historical scholarship converge on the same scripts and time period,” said Zemene, the study’s first author.
“The model had no access to historical records, yet it learned purely from visual and structural data and identified Armenian as the closest structural match to Ethiopic within the very timeframe historians have long debated. That convergence between computation and history is powerful.”
The researchers emphasize similarity does not automatically mean direct borrowing. However, the findings make it more reasonable to consider that these cultures may have influenced one another. Throughout history, societies have shared ideas, including systems of writing. Greek, Roman, Persian and Arabic civilizations all influenced one another in different ways.
This new research suggests Ethiopia’s ancient writing culture may also have played a meaningful role in the exchange of ideas across regions. It also shows AI is not just about modern technology, but a tool that can help us understand literary heritage with a new level of precision.
Journal
Digital Scholarship in the Humanities
DOI
10.1093/llc/fqag029
Method of Research
Computational simulation/modeling
Subject of Research
Not applicable
Article Title
Machine learning techniques for exploring influence, commonalities, and shared origin of scripts: cases of Ethiopic, Armenian, Georgian, and Caucasian Albanian scripts
Article Publication Date
25-Mar-2026
—