Computer reads seven metres of the Dead Sea Scroll and ‘sees’ that there were two writers
This entry was posted on April 28, 2021.
The Dead Sea Scrolls. How many writers worked on the Dead Sea Scrolls? Researchers from Groningen, the Netherlands, have answered this question for one of the scrolls.
By Jona Lendering
Researchers from the Qumran Institute (University of Groningen) have established that the over 2,000-year-old Dead Sea Scroll with the text of the prophet Isaiah was made by two scribes. This conclusion, published in the scientific journal PLOS ONE, marks a scientific breakthrough.
It was not human eyes staring at the letters on the ancient pieces of parchment, but algorithms that managed to distinguish two individual types of handwriting on what is known as the Great Isaiah scroll. That there were two copyists had been suggested before, but the subjective human arguments were never convincing. The more objective calculations made by the research team, led by Mladen Popović, are now changing that.
The Dead Sea Scrolls, discovered in 1947 in several caves near a ruin called Qumran, are a large collection of religious literature from around the beginning of our era. As many of these ancient texts were written on fragile parchment, and have therefore been fragmentarily preserved, their full publication took until 2009. In fact, sixty years after their discovery, research has started all over again.
Previously, this type of research was done by humans. Now the computer has learned how to identify letters
In view of the interpretation of all these texts, it is important to determine whether the Dead Sea Scrolls together form one library of, for example, a Jewish sect like the Essenes, or whether they are a more or less accidental collection of texts from several Jewish groups. Do they represent one sect or Judaism as a whole? To answer that question, it is important to identify the hands of the different scribes.
Identifying individual handwriting, however, is notoriously tricky. Everyone's handwriting has variations, for example, due to fatigue. What deviations in letter form can be characterized as normal, and which indicate a different writer? Another problem is that professional copyists could adapt their handwriting, for instance to that of a colleague whose work they were completing.
Recognizable letter shapes
Until now, this kind of research has been largely done by humans. Paleographers, specialists who study ancient manuscripts, traditionally work with tables containing typical, easily recognizable letterforms, for example, twenty representative As, twenty Bs, etc. As long as researchers use the same table, they can communicate with each other, but the underlying choice of what should be considered typical is subjective. Worse still, the size of such a palaeographic table and the human ability to remember its contents are limited. Paleography is therefore partly Fingerspitzengefühl.
Although this doesn't automatically lead to false results, subjectivity is of course a problem from a scientific point of view. The good thing about computers being used in the field of Paleography is that they lack these human limitations and can process endless amounts of data.
This however does not mean that handwriting recognition with artificial intelligence is easy. Experiments with automatic handwriting recognition have been done before, for example with Byzantine manuscripts and, just last year, with ancient inscriptions in Judaea and Samaria. However, researchers Mladen Popović, Maruf Dhali, and Lambert Schomaker still had to overcome a multitude of problems. One hurdle had already been overcome: the Isaiah scroll had already been digitally photographed, also in infrared light. This was important because bleached ink, invisible to the human eye, sometimes remains perceptible at certain light frequencies.
Differences Between Ink and Parchment
The next hurdle was to teach the computer to identify letters, which basically comes down to optically recognizing the difference between ink and parchment. For example, a dark spot on a digital photo can be a dot as well as an irregularity on the writing surface. Working with centuries-old parchment also has special requirements. Antique parchment dries out over the centuries and warps a bit. Dhali developed an artificial neural network that can also be used for the analysis of photographs of other ancient texts.
Then the computer had to learn to recognize individual handwriting. Unlike in the traditional palaeographic table, which includes only the letter shapes that one researcher has considered representative, a computer can oversee all letters. For a long text such as Isaiah, there could be hundreds, or even thousands of variations: for example 5,011 letters for aleph, the first letter of the Hebrew alphabet.
There are several ways through which software can be programmed to discover patterns. A well-known approach is to look at just how skewed the letters are and how much space there is between them. One of the innovative aspects of the approach the research team from Groningen took was the establishment of a category that they call hinge, which produces an index that defines both the slant and the roundness of the letters. Another way to help the computer get a grip on the antique letters is to record the contours of a letter, known as so-called fraglets.
Safekeeping in a jar
The Great Isaiah scroll was one of the first seven scrolls to be discovered. It was made in the second century BC and preserved in a jar, which is why it is virtually intact. It is also one of the largest scrolls, counting fifty-four columns spanning nearly seven and a half meters.
The Great Isaiah Scroll is not the only copy of this ancient book : twenty other copies were found. Apparently Isaiah was an important text in Judaism at this time, which explains not only why it was included in what was to become the canon of the Jewish Bible, but also why this scroll was so carefully stored. The researchers could expect careful copying too.
Earlier researchers dealing with this Great Isaiah Scroll believed that it had been copied by one scribe, although there were some dissident voices. In 2004, Qumranologist Emanuel Tov suggested that there had been two copyists who each copied about half the book. His hypothesis, based on non-palaeographic grounds, did not convince his colleagues. They argued that if there were any differences between the two halves, this would have to do with the fact that one copyist had worked with two originals.
Two copyists? But where did the second take over from the first? Or was there one writer working with various originals? The researchers determined that the fifty-four columns could be divided into two groups by using various statistical analyses. They also established that the second writer had more variation in his writing than the first.
Popovic, Dhali, and Schomaker then went in search of the place where the second writer had relieved his predecessor. They found the point at the end of column 27: a feat that was only possible with the new sophisticated software. In retrospect, it is exactly the spot that a human, if he had reasons to be sure that there were two writers might have pointed out by mere sight, since three lines have been left unwritten here. However, without the preceding letter analyzes, this alone would have convinced no one.
As a final test, a kind of self-control if you will, the researchers paradoxically returned to the old practice of looking. They electronically overlaid all the letters from the Great Isaiah scroll, alefs on alefs, bets on bets, gimels on gimels, thus creating a heat map of the average letter and deviations. While looking at the letters, the researchers now suddenly recognized some subtle differences between the two copyists. And so the research ended with a new palaeographic table which was no longer based on a subjective judgement of a paleographer, but on all letters, thus becoming more robust.
Looking at other texts
The fact that copyists could adapt their handwriting so closely to that of others, offers us more insight into the practice of copying than ever deemed possible, and forces ancient historians to look at other texts again. One candidate up for re-examination is the Codex Sinaiticus, an important Greek Bible text dated to the fourth-century AD.
In Groningen, research of the Dead Sea Scrolls has not yet been completed. The final task the researchers have set for themselves is to establish which copyist wrote which scrolls. To do this, all scrolls must be analyzed first and the researchers must create profiles of the various copyists. The period in which they lived will have to be determined by means of carbon dating.
In other words: there is still much work to be done.
Want to know more about the Ancient World?