On the Palestine Archaeological Museum (PAM) Plates

It’s no exaggeration to say that scholarly analysis of the Judean Desert scrolls is indebted to technology in ways unquantifiable. One of the first methodological steps that John Trever made was to use his photography skills to photograph the scrolls. Trever managed to photograph three scrolls that were brought to him at the American Schools of Oriental Research in 1947. Trever reasoned that quality photographs would enable a thorough analysis of palaeography, without having to handle the physical artefact over and over again.1 Trever’s images are priceless for scholarly analysis of manuscripts such as 1QS (The Rule of the Community) or 1QIsaa (The Greaty Isaiah Scroll).

But the history of the images does not stop with Trever. In addition to Trever’s collection, there are several other invaluable collections of images; three collections of significant note are: the Palestine Archaeological Museum (PAM), the Shrine Collection (SHR), and the Leon Levy Dead Sea Scrolls Electronic Library (IAA).2 The PAM images are indexed in a very helpful volume: Stephen Reed, The Dead Sea Scrolls Catalogue: Documents, Photographs and Museum Inventory Numbers, ed. Marilyn J. Lundberg and Michael B. Phelps, SBL Resources for Biblical Study (Atlanta: Society of Biblical Literature, 1994).3

The PAM images, in particular, are important for several reasons. These images are often earliest and thus reveal the state of the fragments within months from their having been removed from the Qumran Caves. In addition, the images served as notes to the editors4 and thus the movement of a fragments on various PAM plates can often times reveal material connections.

Consider, for example, PAM 40.5115

In this PAM image, we can see how Jósef Milik made notes about the relationship between each fragment, as he peeled each layer (= a fragment) from a wad. He even seems to have used damage caused by a “stone” that was embedded in the scroll. This is crucial information as to how this scroll would be reconstructed (1Q22, דברי משה), and how damage patterns can aid the process of material reconstruction.

It is for reasons above, that the PAM plates themselves cannot be neglected. But how might we find other fragments that could have been part of this scroll? Or how might we find where the same fragments imaged above exist on other PAM plates? For the most part, one can begin by checking Reed’s The Dead Sea Scrolls Catalogue and Tov’s Revised List, but in my experience these lists are not comprehensive. I discovered this when I was searching through the PAM images some years ago, and stumbled on a plate of 4Q266 fragments that had not been indexed. This observation prompted to work through all the PAMs, but this is a daunting amount of work to do for one person, especially when time is precious.

Thus, I decided I would harness the speed of the computer, and put it to work for me. So, I started to tag the PAM photos several years ago.6 By tagging, I mean each PAM is tagged with respect to the fragments that were imaged on it. As the following example diagram makes clear:

Since the fragments are tagged with respect to their Region of Interest on the width and height of the fragment, it is possible to use this data in Computer Vision searches. When stored in a relational database model (e.g., postgreSQL or MariaDB), it is possible to draw on the significant power of OpenCV, Sift, and Fastai to search these images in ways unimaginable. Here is a sketch of the algorithm I have authored to conduct research for my University of Toronto doctoral dissertation and scrolls research:

Fragment Identification Neural Network
Fragment Identification Neural Network
Find Fragments via Shape
<b>Find Fragments via Shape</b>
Find Fragment via Transcription and more
Find Fragment via Transcription and more
Computer Vision
Computer Vision
Every fragment on a PAM image is tagged with respect to its region on the image. These specified Regions of Interest (ROI) are then stored in a database for additional computer vision computational methods (e.g., Scale Invariant Feature Transform = SIFT).
[Not supported by viewer]
The Database architecture is bilingual. Meaning that, the the metadata about the PAM images are stored in PostgreSQL, but this database is linked with the Natural Language Processing (NLP) Graph Database (Neo4j).
[Not supported by viewer]
Neural Network Algorithms to Compute Probabilities of Matches
[Not supported by viewer]
Possible Queries:
• Find one fragment (based on shape) in all PAM Images
• Find all fragments with ל in similar idiographic form as one in this fragment
[Not supported by viewer]
frag Unid A
frag Unid A
Possible Queries:
• Find all fragments with א followed by ת within 50 characters
• Find all fragments with אל at line initial location
• Find all fragments with stitching
• Find all fragments of 4Q265
[Not supported by viewer]
ל
ל
אל
אל
Natural Language Processing
Natural Language Processing
The Natural Language Processing side of the algorithm works with a GraphDatabase, which contains linguistically tagged data for the fragments and scrolls of the Judean Desert, as well as the Hebrew Bible, Septuagint (including variants), Peshitta, Targumim, Vulgate, Mishnah, Tosefta, and parts of Talmud Yerushalmi. These data create an enormous amount of potentials for linguistic identifications, and is designed for more advanced linguistic research beyond simplistic questions of fragment identification.
[Not supported by viewer]
© 2019 James M. Tucker
© 2019 James M. Tucker
  1. Of course, Trever desired that William F. Albright, who was in the United States of America at the time, could examine the scrolls so as to validate their authenticity and date. For more information, see his memoirs, The Untold Story of Qumran (Westwood, New Jersey: Fleming H. Revell Company, 1965).
  2. See here: https://www.deadseascrolls.org.il
  3. In addition to this volume, one also needs to consult Emaneul Tov, ed., Companion Volume to the Dead Sea Scrolls Microfiche Edition (Leiden: Brill, 1995), John M. Allego, The Allegro Qumran Collection: Supplement to The Dead Sea Scrolls on Microfiche, ed. George Brooke (Leiden: Brill, 1996), and Emanuel Tov, Revised Lists of the Texts from the Judaean Desert (Leiden: Brill, 2010).
  4. See, e.g., Strugnell’s essay, “On the History of the Photographing of the Discoveries of the Judean Desert for the International Team,” in the Companion Volume.
  5. See also https://twitter.com/James_M_Tucker/status/1093469615794122753?s=20
  6. In accordance with digital humanities best practices, this data is open source and available at my github page: https://github.com/JamesMTucker.

2 thoughts on “On the Palestine Archaeological Museum (PAM) Plates”

  1. Thanks, Loren. I think there is certainly a role for experienced instinct, but often times this instinct is not spelled out. And when it isn’t spelled out, others cannot follow and/or have access to the observations that are informing the instinct. For example, I’ve looked at the S and D scrolls so much that I can skim through PAMs and spot the hands immediately—and this has led to finding fragments of S and D in PAM plates that have not been indexed as containing these compositions. But having an algorithm that _also_ searches the corpus of PAM images can provide another mechanism to ensure we finally get a thorough index of all the fragments. The prime example of this is 4Q256 frag. 1. Alexander and Vermes were cognisant of its existence on only one PAM plate. I have found it on two other plates, and the fragment was also published in another DJD volume.

    I suppose the frustrating thing for me is that I am more interested in studying the development of law and legal traditions, and not necessarily the mechanical features as discussed above. But I suppose scholarship is both the creation of tools and the historical analysis of the transmission of ideas. If both can be done, all the better. Thanks for reading.

Leave a Reply

Your email address will not be published. Required fields are marked *