Answers are grounded in this paper and cite their sources.
The Proto-Elamite script (Iran, c. 3100–2900 BCE) is undeciphered: the language of its scribes and the value of most of its signs are unknown. We present a fully reproducible analysis of the 1,504 transliterated tablets of the Cuneiform Digital Library Initiative (CDLI), starting from what can be verified by arithmetic. (1) The chain of grain capacity units can be read directly from the data, through the maximum number of repetitions of each numerical sign: among 2,306 occurrences in intact notations there is a single exception. (2) An adult ration equals 5 N24 units; a system of equations over 113 tablets and 50 personnel groups, solved by non-negative least squares with bootstrap resampling, separates a 5-unit category (M388, M054), a 4-unit category (M056~f) and counted signs that receive no grain (M263, M354). (3) In goat-herding records, cheese follows a norm of one N30C unit per goat (7 of 9 herds), alongside four fixed dues. (4) Tablets bearing the same seal share 26% of their signs, against 10% at random (p = 0.0002); the compound |M153+M342| is 2.4–2.9 times more frequent on sealed tablets and accompanies a ration-voucher form used under three different seals. (5) The phonetic values of Linear Elamite, applied to the related Proto-Elamite signs, produce no known Elamite word (p = 1.0). (6) Finally, we propose an independent statistical test of the Linear Elamite decipherment (Desset et al. 2022): judged by a language model built on 20,397 Elamite words from the Persepolis Fortification Tablets (Hallock 1969), its readings are more probable than 2,000 permutations of its own values (z = 1.9–2.2; p = 0.012–0.020), a moderate positive signal whose limits we discuss. An Elamo-Dravidian hypothesis tested with Kessler's method shows no signal (p = 0.44). Analyses carried out with the assistance of Claude (Claude Opus 5.5), an AI model developed by Anthropic, and of SophIA; see the statement on the use of AI in the paper. A French version of the paper is included.
No takes yet. Share an insight, caveat, or question.
Fabian Vanderauwera (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: