“Symbolic precursors to writing” — a system of conventional signs — from the Paleolithic?

Below is a guest post by Richard Sproat and Rafael Núñez. It summarizes a recently published paper by d’Errico, Santos da Rosa, Courtenay, Núñez, Liu, Sproat, & Blasi, " Inferring Conventional Sign Systems from Paleolithic Engravings: Methodological and Theoretical Challenges ".

Back in February, Christian Bentz and Ewa Dutkiewicz (henceforth BD) published a paper in the Proceedings of the National Academy of Sciences, that presented their study of markings on Paleolithic artifacts, mostly in bone, antler and ivory, from the Swabian Aurignacian (43,000–34,000 BP). Their conclusion, based on various statistical analyses, including entropic measures, and a couple of sign repetition rate measures, was that these marks constituted a structured system of deliberate and conventional signs. By translating the engravings into 1D sequences of standardized characters, they reported strong statistical overlap with early Mesopotamian protocuneiform (Uruk V) and clear divergence from later systems as well as modern writing systems.

Predictably, the paper attracted a lot of attention, especially from the popular science press, ever eager for a story that “rewrites history as we know it”. Coverage included CNN, Science Magazine, Smithsonian Magazine, among many others. To their credit, and unlike some other recent work (e.g. Bacon et al. 2023), BD did not emphasize guesses as to the meanings of the marks. Also, they were relatively careful in their article to avoid making strong claims about these markings being “protowriting”, much less “writing”. The authors, however, were subtle, but explicit, in stressing pretentious and sensational claims. For instance, in the conclusions they stated that the reported “statistical fingerprint” was “very similar” to the “earliest numeric and numero-ideographic protocuneiform tablets stemming from the Uruk V period”, deliberately establishing a link to sophisticated higher-order cognition — numerical cognition and notation. (In a related article that they cite, the 2 nd author explicitly claimed that the star artifact in DB's Figure 1A (so-called "Adorant") “constitute[s] deliberate numerical representations” — Dutkiewicz, 2023.)

Such an evoked representational “numerical” interpretative framework, although flashy, ignores basic issues regarding the conventionalization of symbolic-reference subserving the origin of numerical cognition (Núñez, 2017), and obviously, the crucial fact that numerical practices in Uruk V emerged from a complex stratified society fundamentally different from that of the Swabian Aurignacian hunther-gatherers. Of course, none of these considerations stopped the press from picking up on the sensational claims, with the Science piece, for example, enthusing that “the artifacts… might contain the earliest known symbolic precursors to writing.”

Needless to say, given the roughly 40,000 year gap between these artifacts and anything we can truly identify as writing, such claims are completely unfounded. Whether one takes “ precursor” to trivially mean that early signs temporally preceded writing (and somehow later mutated to it), or that such early signs eventually inevitably led to the development of writing, the claim is unfounded: There simply isn’t the evidence to draw any connection between these marks and the much later development of pre-literate signs to which we can assign definite meanings.

In an attempt to get the discussion of these interesting artifacts back on a more tenable track, we (along with other colleagues) have published a preprint ( d’Errico, et al. 2026 ) that critiques the BD analysis from a number of angles, noting that while the formal numerical correlations are interesting, they are semantically neutral. Statistical similarity does not translate into semiotic or functional equivalence, leaving the status of conventionality for these Paleolithic marks unproven. At the same time, we propose ways to deal with such data noting that “[t]he broader challenge is not to determine whether Paleolithic engravings resemble writing systems, but to develop robust frameworks capable of linking statistical structure to the behaviours and cultural practices that produced it.”

We identify several serious problems in BD’s study:

Instability of mark classification. BD’s initial classification of marks suffers from low coding reliability. While raw inter-coder agreement was high (91–94%), the chance-corrected Cohen's kappa scores were extremely low (0.29 to 0.44), signifying only "fair-to-moderate" consensus. Treating these highly subjective categories as fixed, objective statistical units over-stabilizes the dataset artificially.

The confound of sequence length. The high overlap on metrics like unigram entropy and type-token ratio occurred strictly with the Uruk V phase. However the Aurignacian and Uruk V datasets share nearly identical sequence length distributions. Because shorter paths intrinsically suppress combinatorial entropy and inflate apparent similarity, the statistical proximity could merely be an artifact of sequence length rather than shared structural complexity.

Privileging a single metric for behavior. BD argued that figurines, in particular anthropomorphic and zoomorphic ivory figurines, showed sequences with higher entropy rates than other artifacts, such as tools. They take this as evidence that different marking patterns were applied to artifacts with different uses, which is at least consistent with the idea that the marks had meaning. But this conclusion rests selectively on a single variable (entropy rate), without considering whether other structural indices (like repetition rates or type-token ratios) converge to support the same behavioral hypothesis.

No comparison against control sets of purely decorative systems. BD, while arguing that the Aurignacian marks were an early conventionalized sign system, are of course aware of the alternative possibility that they were simply decorative. Their answer to this is that “categories such as ‘decoration’ and ‘numerical system’ (but see above), or ‘decoration’ and ‘writing system’ are not mutually exclusive”. In and of itself, this is a valid point. Yet BD provide no analysis of how the marks compare to known decorative systems, comparing only against pre-literate accounting texts from Uruk, and later written language. As we discuss in the preprint, previous work (Sproat, 2014) has shown that on many measures, even purely decorative systems such as Pennsylvania barn stars can look like conventional sign systems, and so the analysis BD provide does not in fact eliminate the possibility that the marks had no communicative function other than aesthetics.

In summary, BD’s impressive-seeming results are based on a number of questionable points, including assumptions about sign classification, confounds such as sequence length, the selective focus on particular measures in drawing some conclusions, and the lack of any comparison of the Aurignacian marks to known decorative systems. This latter oversight is particularly significant since BD’s analysis does not, it seems to us, allow one to reject the null hypothesis that the Aurignacian marks were simply intended as decoration.

References

Bacon, B.; Khatiri, A.; Palmer, J.; Freeth, T.; Pettitt, P.; & Kentridge, R. (2023). An Upper Palaeolithic proto-writing system and phenological calendar. Cambridge Archaeological Journal, doi.org/10.1017/S0959774322000415.

Bentz, C. & Dutkiewicz, E. (2026). Humans 40,000 y ago developed a system of conventional signs. Proceedings of the National Academy of Sciences, 123(9), Article e2520385123. doi.org/10.1073/pnas.2520385123.

d’Errico, F.; Santos da Rosa, N.; Courtenay, L. A.; Núñez, R.; Liu, C.; Sproat, R.; & Blasi, D.. Inferring Conventional Sign Systems from Paleolithic Engravings: Methodological and Theoretical Challenges (2026). Available at SSRN: ssrn.com/abstract=7053279 or dx.doi.org/10.2139/ssrn.7053279…