Mysterious second writer of Dead Sea Scroll uncovered by AI
A famous Dead Sea Scroll manuscript was written by not just one but two scribes, according to a new study that used artificial intelligence (AI) and statistics to detect subtle differences in handwriting on the ancient document.
The two scribes wrote in such a similar manner that the differences between the two aren’t visible to the naked eye, the analysis revealed — a detail that suggests the scribes might have received similar training, perhaps at a school or in a close social setting, the researchers wrote in the study.
“This is just the first step,” study principal investigator Mladen Popović, a professor of the Hebrew Bible and ancient Judaism at the University of Groningen in the Netherlands, told Live Science in an email. “We have opened the door to the microlevel of individual scribes; this will open new possibilities to study all the scribes behind the Dead Sea Scrolls and put us in a new and potentially better position to understand with what kind of collection, or collections of manuscripts we’re dealing [with] here.”
The Dead Sea Scrolls were first discovered in the late 1940s, when a young shepherd looking for a stray goat found several manuscripts in a cave in Qumran, in the West Bank. Over the next decade, researchers and local Bedouins found more than 900 manuscripts in 11 caves. These manuscripts are the oldest remaining texts of the Hebrew Bible, dating from the fourth century B.C. to the second century A.D. But it’s unclear who or even how many people wrote them, because the scribes didn’t sign their names, the researchers of the new study said.
That hasn’t stopped biblical scholars from guessing how many scribes were involved in penning the various Dead Sea Scroll manuscripts. “They would try to find a ‘smoking gun’ in the handwriting, for example, a very specific trait in a letter which would identify a scribe,’ Popović, who is also the director of the University of Groningen’s Qumran Institute, said in a statement. But these “smoking gun” analyses were often subjective and, as a result, hotly debated, he said.
So, Popović and his colleagues used another approach — AI and statistics — to investigate the Great Isaiah Scroll, one of the seven scrolls originally found by the Bedouin shepherd. This well-preserved scroll, which dates to about 125 B.C., is lengthy — it measures 24 feet (7.3 meters) long and 10 inches (26 centimeters) high — and contains 54 columns of Hebrew text. One spot, in particular, caught Popović’s eye; between columns 27 and 28, there is a small break in the text and a new “page,” where two sheets have been sewn together. Other researchers had already debated whether this scroll was written by one or two scribes, and Popović’s team wanted to see if they could solve the mystery.
In effect, the team wanted to determine “whether subtle differences in writing should be regarded as normal variations in the handwriting of one scribe or as similar scripts of two different scribes,” they wrote in the study.
The researchers’ methods detected “subtle and nuanced differences in [the] handwriting that we cannot [discern] with the human eye only,” Popović told Live Science. The discovery that two scribes collaborated on the Great Isaiah Scroll reveals that ancient scribes “worked in teams,” he said. And, unlike the “smoking gun” analyses, this research “is not just a conjecture, but based on evidence now,” Popović added.
How they did it
When designing the algorithm, the researchers had to train it to differentiate the text, or the ink, from the background — the animal skin or papyrus. This distinction, known as binarization, was designed by study co-researcher Maruf Dhali, a doctoral student in the artificial intelligence department at the University of Groningen, who created an artificial neural network that could be trained using deep learning. This neural network recorded the original ink traces on the manuscript, even when these ancient letters were transformed into digital images.
“This is important because the ancient ink traces relate directly to a person’s muscle movement and are person-specific,” study senior researcher Lambert Schomaker, a professor of computer science and artificial intelligence at the University of Groningen, said in the statement.