What we work on

Research

From the geometry of high-dimensional data to ink on ancient sherds.

Ancient Script & Text Evolution

How and why writing changes over time — its letterforms, its language, and the societies behind it.

Our flagship program treats writing as something that evolves — through reproduction, variation, and selection: scribes copy existing signs, copies differ through unavoidable neuro-muscular noise, and cultural context, not conscious design, decides which variants persist. At the level of the individual letterform we use the kinematic (ΣΛ) theory of rapid human movement; at the level of whole corpora we use statistics and information theory to test how scripts, spellings, and styles shift across centuries.

The work runs from Iron-Age Hebrew and Phoenician scribal schools (the “Writing Through Time” project, co-led with Christian Casey, FU Berlin) to Egyptian and Akkadian literary traditions, and into the Hebrew Bible and Dead Sea Scrolls. A recurring theme is an efficiency law — grapheme complexity tracks information content, and the coupling tightens over time. Adjacent efforts include Reed-Pen Paleography (recovering the reed nib’s geometry and grip angle as a scribal biometric) and the invited Oxford Handbook chapter on epigraphy and digital resources for Northwest-Semitic studies.

Reproduction–Variation–SelectionΣΛ kinematic modelComputational paleographyStylometry & authorshipInformation-theoretic efficiency

Key collaborators: Christian Casey (FU Berlin) · Nathan Wasserman (HUJI) · Eli Piasetzky (TAU) · Israel Finkelstein (TAU)

Funding & support: Volkswagenstiftung · Schmidt Sciences

Manifold-Based Estimation

Provable-rate estimation of manifolds — and the geometry riding on them — from finite, noisy, high-dimensional data.

A longstanding methodological collaboration with Yariv Aizenbud (Tel-Aviv University) building the statistical and approximation-theoretic core of the Manifold Moving Least-Squares (MMLS) program. Given noisy samples of a low-dimensional manifold sitting in a high-dimensional space, we estimate the manifold itself — together with its tangent spaces and curvature — directly, without first doing dimension reduction, with linear cost in the ambient dimension and provable convergence rates.

Recent work lifts a pointwise estimator to a global, smooth, Hausdorff-convergent manifold estimate, and connects the geometry back to concrete humanities questions such as the diachrony of Akkadian literature. Along the way the program has surfaced clean theoretical results — including a previously unnoticed equivalence between iterative least-squares regression and Principal Component Analysis.

Moving Least-SquaresOptimal ratesTangent & curvature estimationGeodesic distancesLS–PCA equivalence

Key collaborators: Yariv Aizenbud (TAU) · David Levin (TAU) · Ingrid Daubechies (Duke)

Hyperspectral Imaging of Inscriptions

Non-invasive spectral imaging that both reveals unreadable ancient ink and characterizes the materials that made it.

The lab’s imaging engine — the Spectra half of our name — documents inscriptions with multispectral (MSI) and hyperspectral / reflectance imaging spectroscopy (HSI/RIS). Because a hyperspectral scan captures a full spectrum at every pixel, a single acquisition can both enhance legibility of faded or invisible ink and characterize materials (ink stratification, spectral clustering). This is the first hyperspectral-imaging lab for the Humanities in Israel, and the in-house data source for our paleography and reed-pen work.

The approach has produced concrete discoveries — a previously unnoticed Iron-Age inscription on the reverse of Arad ostracon 16, revealed via MSI, and a newly imaged Hebrew ostracon from Lachish. Planned expansion adds full VNIR–SWIR reflectance capability and a spectral database of Iron-Age inscriptions.

MSI / HSI / RISVNIR–SWIR datacubesPCA + MNF enhancementInk stratificationSpectral databases

See the Imaging Lab & request documentation →

Featured directions

Foundational result

Onomastic Diversity & Social Dynamics

A single, foundational result with an outsized reach. Borrowing diversity statistics from ecology — Hill numbers, evenness curves, and small-sample tests — we read social dynamics from the distribution of personal names preserved on Iron-Age Hebrew inscriptions and seals (PNAS, 2025). The Kingdom of Israel emerges as more cosmopolitan than Judah, with a capital–periphery inversion in late Judah.

The method is deliberately general: it applies to any historical corpus of names, and the same mathematics later seeded a molecular-diversity biosignature for planetary missions in Nature Astronomy — a clean example of how a humanities question can grow a tool that travels far beyond its origin.

Also in the lab

Causal Philology. An empirical, causal-inference framework for how ancient written traditions evolve — bringing explicit causal reasoning to the study of textual change.
Dating Akkadian literature. Continuous causal-probing to date Akkadian texts, and to audit what low-resource ancient-language language-models actually encode (with N. Wasserman & G. Stanovsky; Schmidt Sciences).
Fractals via neural networks. A theory result: fractals from iterated function systems are representable by neural networks with O(k) parameters (JSAIT, 2020).