Arch-Tech: Using AI to Identify Scroll Scribes

024

After the Bible, the Dead Sea Scrolls are the most famous collection of writings from ancient Israel. However, the individuals who penned the scrolls remain anonymous. To identify individual scribes in the Dead Sea Scrolls, my team of scroll scholars and artificial intelligence (AI) experts has developed a tool that analyzes the scrolls on the level of individual scribal strokes.

Scholars have previously suggested that single scribes wrote some manuscripts.^a They based their claims on the detailed examination of handwriting. But traditional paleography (the study of ancient writing) is always subjective. Paleographers typically try to find a “smoking gun” in the form of a very specific trait in a letter that would identify a scribe. However, scribes may show a range in a variety of forms of individual letters within one or across more manuscripts. At the same time, different scribes may have almost the same handwriting. So the challenge for paleographers is to determine which differences in handwriting are likely to be idiographic (unique to an individual writer) and thus significant, which requires lots of experience.

To tackle this problem, our team used digital images of the scrolls and applied pattern recognition and artificial intelligence techniques. We performed a case study on the Great Isaiah Scroll from Qumran Cave 1 that offers a good example of the lack of a robust method in Dead Sea Scrolls paleography for determining and verifying writer identity. Because its handwriting is nearly uniform, some scholars say one scribe wrote this 025scroll. Others, however, suggest there were two scribes. Our team was able to present new evidence for two scribes.¹ Here’s how we did it.

The Great Isaiah Scroll, which is about 23 feet long with 54 columns of text, contains the letter aleph at least 5,000 times. It is humanly impossible to compare all these occurrences by eye. Computers, on the other hand, are well suited to analyze large datasets and can make all sorts of calculations at the level of letters and even individual strokes and curvatures. Although the human eye is amazing and presumably considers these levels, too, the process is often not transparent or quantifiable, especially when there is much data to consider.

The first hurdle was to train an algorithm to separate the inscribed text (ink) from its background (leather or papyrus). For this separation, or binarization, Maruf Dhali, a member of our research team, developed a state-of-the-art artificial neural network that can be trained using deep learning (machine learning method that imitates the way humans learn). The neural network keeps the original ink traces intact as they appear on the digital images. This is important because the ancient ink traces relate directly to a person’s muscle movement and are person-specific.

Our first analysis showed that the columns of text in the Great Isaiah Scroll fell into two distinct groups and that these were not distributed randomly through the scroll, but rather were clustered, with a transition around the halfway mark (between columns 27 and 28).

Expecting more than one writer, we then recomputed the similarities between the columns, this time using the patterns of letter fragments. This second analysis confirmed the presence of two different scribes. When we added “noise” (spurious letter forms) to the data, the results did not change. We also succeeded in demonstrating that the second scribe shows more variation within his writing than the first, although their writing is very similar.

Finally, we visualized the collected data in heatmaps for individual letters that show all attested letter shapes morphed into one character and coded in color from cold to hot, depending on the statistical frequency of individual shapes. These heatmaps show minute but significant differences between characters from the two halves of the scroll.

Certain aspects of the Great Isaiah Scroll and the positioning of the text had led some scholars to suggest that a new scribe had begun writing after column 27, but this suggestion was not generally accepted. Now, we can confirm this with a quantitative analysis of the handwriting as well as with robust statistical analyses. Instead of basing judgment on more-or-less impressionistic evidence, we can now demonstrate that the separation is statistically significant using computer-based artificial intelligence.

Our breakthrough in scroll paleography is that we can now access the level of individual scribes and carefully observe how they worked on these manuscripts. Applying our digital tool to other scrolls may reveal relations between manuscripts and texts that have previously gone unnoticed. The ability to identify scribes can contribute to a finer understanding of the different groupings and collections of manuscripts within the Dead Sea Scrolls, contributing to a better understanding of the dynamics of manuscript production and collection in biblical times.

We are now able to distinguish between individual scribes. Although we will never know their names, with artificial intelligence, it at least feels as if we can finally shake hands with them through their handwriting.

After the Bible, the Dead Sea Scrolls are the most famous collection of writings from ancient Israel. However, the individuals who penned the scrolls remain anonymous. To identify individual scribes in the Dead Sea Scrolls, my team of scroll scholars and artificial intelligence (AI) experts has developed a tool that analyzes the scrolls on the level of individual scribal strokes. Scholars have previously suggested that single scribes wrote some manuscripts.a They based their claims on the detailed examination of handwriting. But traditional paleography (the study of ancient writing) is always subjective. Paleographers typically try to find a “smoking gun” […]