Home / Science / Artificial intelligence helps decode the Code of the Dead Sea Scrolls.

Artificial intelligence helps decode the Code of the Dead Sea Scrolls.

Analysis of the Dead Sea Scrolls handwriting

Scroll text analysis clip art, Dead Sea Credit: Mladen Popovi.

The Dead Sea Scrolls, discovered about seventy years ago, are famous for containing the oldest manuscripts of the Hebrew Scriptures. (Old Testament) and ancient Jewish texts unknown to the present day. But each person behind the scroll evaded scientists because the scribes were anonymous. Now, by combining science and humanities, researchers at the University of Groningen have deciphered the code that helped them discover the scribe behind the scroll. They present the findings in a journal. PLOS ONE On April 21

, 2021

The scribes who made the scrolls did not sign their work. Scholars recommend that portions of the manuscript be attributed to a single scribe, based on the handwriting. “They will try to find a ‘smoking gun’ in the handwriting, for example a character in a letter that identifies the author,” says Mladen Popović, professor of Biblical languages. The Hebrews and the ancient Judaism of the Faculty of Theology and Religion are described. Education at the University of Groningen He is also the director of the university’s Qumran Institute dedicated to the study of the Dead Sea Scrolls.However, these identifications are quite personal and are often highly controversial.


That is why Popović, in The Hands that Wrote the Bible, funded by the European Research Council, in collaboration with his colleague Lambert Schomaker, professor of computer science and artificial intelligence at the School of Science and Engineering, Schomaker has worked on a variety of techniques. For a long time to allow computers to read handwriting, often from historical documents. He also conducted studies to determine whether biomechanical characteristics, such as how someone held a pen or stylus, would affect handwriting.

Analysis of the Dead Sea Scrolls mail

Two 12 × 12 Kohonen maps (blue maps) of full characters and bets from the Dead Sea Scroll collection.Each character in the Kohonen map is made up of several similar characters. (Indicated by a zoom box with a red line) These maps are useful for analyzing the chronological pattern development. In today’s author identification studies, Fraglets (fragmented character shapes) are used instead of full character shapes for more accurate (strong) results. Credit: Maruf A.Dhali, University of Groningen.

In this study, in collaboration with PhD candidate Maruf Dhali, they focused on a single roll: the Great Isaiah Scroll (1QIsa.g) From Qumran Cave 1. The handwriting in this scroll looks almost identical. But it is suggested that it was written by two writers with similar writing styles. So how do you decide? Schomaker: “This roll has at least the letter aleph or“ a ”five thousand times. It’s impossible to compare them all by eye. ”Computers are well suited to analyze large data sets, such as 5,000 handwritten characters. Digital photography makes all sorts of computer calculations possible at the micro level of letters, such as measuring curvature. (Called texture) includes all characters (called allographic).

Neural network

“The human eye is amazing and worth taking into account these levels. This allows experts to “see” the hands of many authors, but that decision is often inaccessible through a transparent process, ”Popović said. The large amount of information the scroll provides ”that is why their results are often inconclusive.

Binarization Dead Sea Scrolls

(Left to right) A grayscale image of Column 15 of the Great Isaiah Scroll, consistent binarized image using BiNet, and clean edited image. From the red box of the last two images, we can see how the rotation and the geometric transform have been corrected for better images for further processing.Credit: Reprint from Lim TH, Alexander PS Volume 1. .In: The Dead Sea Scrolls Electronic Library. 1995 under license CC BY licensed from Brill Publishers Original copyright 1995.

The first obstacle is to practice an algorithm to separate (ink) text from the background. For this separation or “binarization,” Dhali developed a state-of-the-art neural network that can be practiced using deep learning. The neural network has helped the original ink trace that the writer wrote more than 2,000 years ago to remain the same as it appears on the digital image. It’s personal and it’s personal, ”Schomaker explains.


Dhali conducted the first analytical test of this study. His analysis of the texture and Allographic features showed that 54 columns of text in the Great Isaiah Scroll were divided into two distinct groups that were not randomly distributed across the scrolls. But it is a group with a change around the halfway mark.

Aleph Dead Sea Scrolls

An illustration on how to generate a heatmap of the normal average character shape for each letter. (In this example: aleph) Credit: Maruf A.Dhali, University of Groningen.

Noting that there could be more than one writer, Dhali then sent the data to Schomaker, who recalculated the similarity between the columns using the scrap format. This second analysis step confirms that there are two different things. Several additional checks and controls were carried out, Schomaker: “When we added extra noise to the data, the results were unchanged. We have also been successful in showing that the second scribe showed more differences in his writings than the first, although their writings were very similar. ”


In the third step, Popović, Dhali, and Schomaker performed a visual analysis. They created a “heatmap” that incorporated all the patterns of the characters into the scroll. They then created this character average version for the first 27 columns and the last 27 columns. An eye comparison of these two average letters shows that they are different. This links computational and statistical analysis to the approximate interpretation of human data because heat maps are not based on or built on primary and secondary analyzes.

Certain aspects of the scroll and its placement have led some scholars to suggest that after column 27 a new scribe was started. But this is not generally accepted, Popović: “We can now confirm this with a quantitative handwriting and powerful statistical analysis. Rather than relying on more or less impressionist judgment with the smart help of computers, we can show that separation is statistically significant. ”

New window

In addition to potential fossil changes – and other ancient manuscripts – this study of the Great Isaiah Scroll opens a new way to analyze the Qumran text based on physical characteristics. Researchers can now reach the micro level of each scribe and observe carefully how they work with these manuscripts.

Popovich: “This is very exciting because it opens up a new window to the ancient world that can reveal a more complex connection between the scribes who created the scroll. In this study, we found evidence of a very similar style of writing, shared by two Great Isaiah Scroll writers, pointing out a training or general origin. Our next step will be to examine other scrolls where we may find different origins or the training of scribes. ”

In this way, we can learn more about the communities that produced the Dead Sea Scrolls. “Now we can identify the different markers,” concluded Popovich. “We won’t know their names. But after seventy years of study, it felt as though we could finally hold their hand with their handwriting. ”

Reference: “Identifying authors using artificial intelligence provides new evidence for unknown writers of the Dead Sea Scrolls, for example by the Great Isaiah Scroll (1QIsa.g) ”By Mladen Popović, Maruf A.Dhali and Lambert Schomaker April 21, 2021. PLOS ONE.
DOI: 10.1371 / journal.pone.0249769

Digital images of the Dead Sea Scrolls and Great Isaiah Scrolls kindly provided by Brill Publishers and the Israel Antiquities Authority (Leon Levy Dead Sea Scrolls Digital Library).

Source link