Computational Dating for the Nuzi Cuneiform Archive: The Least Squares Constrained by Family Trees and Synchronisms

TL;DR

This work develops a computational dating framework for Nuzi cuneiform archives by extending the Nuzi Personal Names index with post-1943 kinships (RNPN2) and solving a constrained least-squares problem to estimate birth and publication years. It leverages generation-length constraints and document dates to produce chronologies that align with Maidman’s Assyriological findings, while resolving name-identity ambiguities and reconstructing kinship networks. The approach demonstrates how to scale identity resolution, lineage reconstruction, and time-slicing to large archival datasets, yielding insights into seniority, document order, and landholding dynamics. By showing that the method generalizes to other clay-tablet archives once name indexes are in the NPN format, the work offers a practical path to systematic temporal dating in ancient corpora and invites future AI-assisted updates of name indices.

Abstract

We introduce a computational method of dating for an archive in ancient Mesopotamia. We use the name index Nuzi Personal Names (NPN) published in 1943. We made an electronic version of NPN and added the kinships of the two powerful families to NPN to reflect the Nuzi studies after 1943. Nuzi is a town from the 15th - 14th century B.C.E.for a period of some five generations in Arrapha. The cuneiform tablets listed in NPN are for contracts on land transactions, marriage, loans, slavery, etc. In NPN, the kinships and cuneiform tablets (contracts, documents, texts) involved are listed for each person. We reconstruct family trees from the added NPN to formulate the least squares problem with the constraints: a person's father is at least 22.5 years older than the person, contractors were living at the time of the contract, etc. Our results agree with the Assyriological results of M. P. Maidman on the seniority among siblings of a powerful family. Our method could be applied to the other clay tablet archives once we have the name index in the format of NPN.

Figures (7)

• Figure 1: Number of initial family trees classified by size and height and number of reconstructed family trees classified by size and height
• Figure 2: Kizzuk family tree (Dosch and Deller dd Maidman mai3).
• Figure 3: Teẖip-tilla family tree for persons of NPN in Fig \ref{['tehipfamily']}, a part of the family tree of Maidman maimai3.
• Figure 4: Birth years for men of the Teẖip-tilla family. For example, we see that Teẖip-tilla son of Puẖi-šenni is in 502 texts, and the mean of birth years by 10 computations, birth1,...,birth10, is -1419.54.
• Figure 5: Logistic growth of the number of published texts each year obtained from RNPN2. The maximum log-likelihood value for the logistic distribution is -6130.72, while that for the normal distribution is -6220.25. The maximum log-likelihood for normal distribution is attained by the parameters $\mu=-1366.94, \sigma=10.1447$, while those for the logistic distribution are $\mu=-1366.84, \beta=5.2842$. The colored line shows the data. The black line shows the estimated logistic distribution.