AI, Statistics, and the Hebrew Bible: Computational Directions in the Study of Ancient Hebrew Texts and Philological Limits of AI

Philology has always read ancient texts qualitatively. Much of the recent computational work asks a narrower question: what happens when we count instead? I will follow that question through three scales of evidence from the Hebrew Bible and its world — the names people were given, the spellings scribes chose, and the letter-shapes they drew. Diversity statistics on the Iron Age onomasticon distinguish Israel from Judah and invert the expected relation between capital and periphery. Column-by-column orthographic counts in the Great Isaiah Scroll point to three scribal segments rather than the classical two. Algorithmic analysis of the Tel Arad ostraca identifies at least five distinct hands, bearing on the question of literacy in late-monarchic Judah. But counting is not neutral, and the rest of the talk is about how it misleads. Our own work shows that stylometry readily mistakes theme for style, and proposes ways of separating them — in general, and in the Priestly source, where the very continuity of the sources breaks the obvious statistical test. Finally, in ongoing work on dating an ancient corpus, what looked like a model reading time in the language proved to be reading the proper names in it: so far, no method we have tested recovers chronology from linguistic structure alone. Each result required inventing the control that made it believable, and I will argue that those controls, not the models, are what computation contributes to philology.

Barak Sober, PhD, is Senior Lecturer of Statistics and Data Science and Digital Humanities at the Hebrew University of Jerusalem. His research operates at the intersection of mathematical statistics, machine learning, artificial intelligence, digital humanities, archaeology, ancient Hebrew epigraphy, and the computational analysis of Biblical Hebrew and other ancient texts.

His work applies statistical and computational methods to questions traditionally addressed through philology, archaeology, palaeography, and textual analysis. His research has included computational analysis of Biblical Hebrew, statistical investigation of literary structure in Genesis and Exodus, algorithmic handwriting analysis of ancient Hebrew inscriptions, multispectral and hyperspectral imaging of ostraca, and quantitative investigation of literacy, administration, and social patterns in the kingdoms of Israel and Judah.

Sober's interdisciplinary research has contributed to new approaches for evaluating textual classification and hypothesized literary divisions within the Hebrew Bible. His work with collaborators on the Priestly material in Genesis and Exodus combines statistical modelling and computational classification with traditional philological and biblical scholarship, providing an important methodological bridge between data science and the humanities.

He has also participated in major computational and imaging studies of Iron Age inscriptions from sites and collections associated with Arad, Samaria, Lachish, and other contexts. These projects combine mathematical modelling, computer vision, advanced imaging, archaeology, epigraphy, and historical interpretation.

Before joining the Hebrew University of Jerusalem, Sober was a Phillip Griffiths Assistant Research Professor in the Department of Mathematics at Duke University and worked with the Rhodes Information Initiative. He completed his doctoral work in applied mathematics at Tel Aviv University under Prof. David Levin, while his master's research was jointly supervised by mathematician David Levin and archaeologist Israel Finkelstein.