PhD project: Dual-mode Time Axis Calibration for the Dead Sea Scrolls

Name: M.A. (Maruf) Dhali

Supervisors:
Prof. dr. Lambert Schomaker
Prof.dr. Mladen Popović
Period: February 2016 – February 2020

Summary of PhD project:

This PhD project is a part of the ERC project titled ‘The Hands that Wrote the Bible’, which integrates three knowledge sources: pattern recognition, 14 C-dating and palaeographic background knowledge to generate statistically supported hypotheses on fragments of text from the famous Dead Sea Scrolls (DSS). The unique strength of this project is the combined power of new radiocarbon samples and digital palaeography as quantitative methods, providing a much-needed scientific and empirical basis for the typological estimations of traditional palaeography of the DSS. The tasks of this particular PhD project mainly focus on the computational intelligence (digital palaeography) part of this project in order to align temporal developments in script style and to identify the writer (‘Hands’). Current developments in pattern recognition and machine learning have given new possibilities for this digital palaeography.

Unlike traditional palaeography of relying on the human eye only, this project applies new techniques and methods developed within computational intelligence. Powerful algorithms have already been developed at ALICE (Artificial Intelligence and Cognitive Engineering Institute) and applied in a wide range of forensic and palaeographic research contexts for the identification of the writer (the hand) and for style-based dating of documents. The combination of textural (curvature-based) and allographic letter fragments allows for a robust characterisation of writers. Though it is possible to use a manually segmented set of individual letters, the textural approach has distinct advantages in terms of manual labour, performance, and reduced human interference.

For the purpose of writer identification and handwriting recognition for the DSS, a secure software environment with High-performance-computing (HPC) ability is utilized using the cutting-edge tools within Monk system (designed by Prof. Schomaker's research group at ALICE). Using these tools of digital palaeography presents a significant step forward for the DSS-images enabling the use of quantitative methods for palaeographic assessments.

---

There are two main goals for this PhD project:

1. Identification of writers (‘hands’) and handwriting recognition

The primary goal consists of generating quantitative data for palaeographic writer identification and handwriting recognition by means of respectively GIWIS and Monk, and setting up a database of the development of handwriting styles and a database of scribal production by clustering manuscripts on the basis of writer identification and thus exploring how many manuscripts were copied by a particular scribe and on which manuscripts more than one scribe worked, either at the same time or at a later date, additions, corrections, etc.

2. Align temporal developments in script style

The other goal is to align temporal developments in script style relative to a reference scale provided by the 14C - dating and other available knowledge. With a proper set of 14C-dated Dead Sea Scrolls constituting the temporal reference, the corresponding handwritten style features may be used to attempt reverse date estimation for an undated sample from the corpus of manuscripts. An attempt can be made to connect this to writer identification and clustering of manuscripts as scribal profiles, if that works this project has the opportunity to locate scribes and their scribal production in time, based on empirical data.

To achieve both the goals, challenges exist at several levels of Computer Vision and Artificial Intelligence. Initial analysis has been done for the proper extraction of characters (foreground) from the background, which is mostly either animal-skin or papyrus in the case of DSS. Several image processing techniques have been applied for optimum results of segmentation. Starting with edge detection, morphological operations, filling gaps and then finding connected-components help in automatically segment the hand-written fragments. Then further processing can localize and extract the characters. Due to the difference in the textures of papyrus and animal-skin, individual measures are taken on their distinctive periodic structures. This has particularly enhanced the texts. The use of artificial neural network can also be used for this task. Different feature extraction techniques have already been tested on the images to further explore the possibilities.

The project aims to bridge a great divide between computational science and traditional palaeography, and will represent a big step forward in interdisciplinary method and communication between researchers in these fields, with a potential impact on digital palaeography beyond Dead Sea Scrolls studies.