If your feed this week told you that scientists discovered 40,000-year-old writing in a German cave, it didn’t. That claim has spread across social media in multiple forms – “earliest known writing,” “Stone Age people recorded their thoughts,” “oldest form of writing found” – and every version overstates what the research actually found.

What researchers did find is genuinely significant. A peer-reviewed paper published February 23 in Proceedings of the National Academy of Sciences shows that geometric marks engraved on Ice Age ivory artifacts in southwestern Germany were not random decoration. They were systematic, conventional, and statistically structured in ways that resemble the earliest known proto-writing – from roughly 37,000 years later.

The distinction matters. This is not writing. It may be one of writing’s oldest ancestors.

What the Paper Found

Linguist Christian Bentz of Saarland University and archaeologist Ewa Dutkiewicz of the Museum für Vor- und Frühgeschichte in Berlin analyzed more than 3,000 engraved geometric signs on approximately 260 portable artifacts from four cave sites in the Swabian Jura region of Baden-Württemberg. The artifacts – figurines, tools, ornaments, and musical instruments carved from mammoth ivory and bone – date to the Aurignacian period, roughly 43,000 to 34,000 years ago.

The sign types include notches, dots, lines, crosses, and star shapes. Using computational methods, Bentz and Dutkiewicz measured the statistical properties of these sign sequences – how varied they were, how often signs repeated, how much information each sequence carried – and compared them against two baselines: modern written language samples and proto-cuneiform tablets from Uruk, Mesopotamia (dating to before 3200 BC).

The result: the Aurignacian sign sequences are clearly distinguishable from modern writing, but their statistical fingerprint – measured by metrics like unigram entropy, entropy rate, and type-token ratio – is comparable to the earliest proto-cuneiform.

“We would argue that these sign sequences go beyond decoration … these signs were applied selectively and conventionally.”

– Christian Bentz, Reuters

Critically, the patterns were not random. When the researchers compared the original sign sequences against randomized versions of the same data, the originals showed significantly more structure – ruling out the possibility that the marks were arbitrary scratches.

“The convention to carve certain sign types only into surfaces of certain artifacts must have been handed down over many generations…”

– Christian Bentz, Reuters

Specific sign types were associated with specific object categories. Crosses appeared on tools and animal figurines but not on other artifact classes. Dots were absent from tools. These associations held stable across roughly 10,000 years of the Aurignacian record – evidence of shared rules, not individual preference.

Archaeologist Dr. Hugh Thomas, who reviewed the paper independently, highlighted this as the study’s most compelling dimension:

“What’s really exciting about this paper is the fact that they found that certain types of objects generally had more of one specific type of symbol on it. There was a pattern to it. So crosses appear on animal figurines like horses and our chunky little mammoth, but generally not on anthropomorph figurines. Dots are on figurines, but aren’t on tools.”

– Dr. Hugh Thomas, archaeologist

The Caves

The artifacts come from one of the richest concentrations of early human symbolic culture anywhere on Earth. The Swabian Jura cave cluster – inscribed as a UNESCO World Heritage site in 2017 – spans the Ach and Lone valleys in southwestern Germany and includes four major sites used in this study:

  • Vogelherd Cave (Lone Valley) – famous for a large assemblage of small ivory animal figurines, including a mammoth figurine bearing engraved crosses and dots
  • Geißenklösterle Cave (Ach Valley) – source of the Adorant, a small ivory plaquette (~38 mm × 14 mm) with an anthropomorphic figure on one side and deliberate sequences of notches and dots on the other
  • Hohlenstein-Stadel (Lone Valley) – findspot of the Lion-Man (Löwenmensch), the iconic 40,000-year-old human-lion hybrid figurine, which itself carries regularly spaced notches along one arm
  • Hohle Fels (Ach Valley) – home to the Venus of Hohle Fels and some of the world’s oldest known musical instruments (bone and ivory flutes)

These caves have already rewritten the timeline of human creativity. The new paper adds another dimension: not just art and music, but structured information systems.

How They Measured It

The methodology is worth understanding because it’s what separates this paper from earlier claims about Paleolithic “proto-writing.”

Bentz and Dutkiewicz digitized the signs into a database and treated them as sequences rather than isolated marks. They then quantified four properties per sequence:

  • Unigram entropy – how varied the sign types are within a sequence (higher = more different signs)
  • Entropy rate – how predictable the next sign is given the preceding signs (lower = more structured)
  • Type-token ratio – the proportion of unique sign types to total signs (a basic measure of vocabulary richness)
  • Repetition rate – how often signs repeat

They then ran classification algorithms – K-Nearest Neighbors (KNN) and a Multi-Layer Perceptron (MLP) neural network – to test whether the Aurignacian sequences clustered with proto-cuneiform or with modern writing. They clustered with proto-cuneiform.

A principal component analysis (PCA) visualization confirmed the separation: modern writing occupied one region of the feature space, proto-cuneiform another, and the Aurignacian signs fell squarely with the proto-cuneiform group.

The full dataset and analysis code are publicly available on GitHub.

What It Doesn’t Mean

The paper is precise about its limits, even if social media wasn’t.

“Our analyses demonstrate that these sign sequences have nothing to do with the writing systems of today…”

– Christian Bentz, Saarland University press release

Modern writing encodes spoken language. It has high information density. The Aurignacian signs show frequent repetition and low variety – the opposite of what you’d expect from a system recording speech. The authors do not claim to have found writing. They do not claim to have deciphered any meanings. They cannot determine what, specifically, the signs recorded.

“We can only speculate about the status of spoken languages at the time.”

– Ewa Dutkiewicz, Reuters

As Thomas put it:

“This isn’t writing, but it’s not random. These symbols probably had some form of meaning behind them.”

– Dr. Hugh Thomas, archaeologist

What the paper does argue is that these marks were conventional (rule-governed), transmitted across generations, and carried structured information – properties that place them on a continuum with later administrative sign systems, not with decoration or idle scratching.

Paleoanthropologist Genevieve von Petzinger, who has cataloged recurring geometric signs across European cave art, told Scientific American that interpreting specific meanings from such marks is “extremely difficult or practically impossible,” but that the pattern-based approach – testing for intentionality, repetition, organization, and surface selection – is a strong method for assessing whether marks were meaningful beyond decoration.

A Long Road to Writing

The Aurignacian signs are not the only ancient mark-making system, nor the oldest evidence of abstract symbolic behavior. They occupy one position on a long timeline that stretches from Middle Stone Age Africa to the administrative centers of ancient Mesopotamia.

PeriodSystemLocationAge
Middle Stone AgeEngraved ochre (cross-hatch patterns)Blombos Cave, South Africa~77,000 years
Middle Stone AgePerforated shell beadsBlombos Cave, South Africa~75,000 years
Early Upper PaleolithicRed disks and hand stencilsEl Castillo Cave, Spain~40,800 years
AurignacianGeometric sign sequences on ivorySwabian Jura, Germany~43,000–34,000 years
Early Upper PaleolithicFigurative paintings with abstract signsChauvet Cave, France~36,000 years
Upper PaleolithicNotched bone with grouped marksIshango, DR Congo~20,000 years
MagdalenianAnimal paintings with dot clusters and gridsLascaux, France~17,000 years
Late 4th millennium BCProto-cuneiform administrative tabletsUruk, Mesopotamia~5,200 years
Before 3200 BCCuneiform scriptMesopotamia~5,200 years

What makes the Swabian Jura signs distinctive is not their age alone – the Blombos engravings are 30,000 years older – but the combination of features: repeated sign types, structured sequences on portable objects, consistent associations between sign types and object categories, and stability over millennia.

“Our research is helping us uncover the unique statistical properties – or statistical fingerprint – of these sign systems, which are an early predecessor to writing.”

– Christian Bentz, Saarland University press release

As the British Museum notes, cuneiform – originating in what is now Iraq before 3200 BC – remains “the oldest form of writing in the world” by the standard scholarly definition. The Aurignacian signs don’t change that. What they suggest is that the impulse to encode information in conventional marks is far older than anyone previously demonstrated with this level of statistical rigor.

The Misframing

Headlines claiming “oldest writing discovered” or “Stone Age humans recorded their thoughts” misrepresent the paper in ways the authors themselves have pushed back on. The New York Post ran “Stone Age symbols could rewrite the history of writing.” New Scientist’s Facebook post described “a simple form of writing.” An Instagram post claimed the signs were used “to record their thoughts.”

This is a familiar problem in science journalism. As Thomas noted from experience:

“When a media release goes out about an archaeological paper, often it’s going to have to have some kind of hook to make people want to click on it and read it. And as an archaeologist, you very rarely get to decide what that hook is.”

– Dr. Hugh Thomas, archaeologist

None of this reflects what the paper says. The study tests statistical structure and conventionality. It does not identify content, decode meaning, or establish a connection to spoken language. Calling these marks “writing” is like calling a tally stick a novel – it mistakes the precursor for the thing itself.

The actual finding – that Ice Age humans in what is now Germany maintained a shared, rule-governed system of engraved signs for thousands of years, one whose information structure resembles the earliest administrative records from Mesopotamia – doesn’t need exaggeration. It stands on its own.


Sources: Bentz & Dutkiewicz, PNAS (2026) · PubMed Record · Saarland University Press Release · Reuters · Scientific American · ScienceDaily · PaleoSigns GitHub Repository · UNESCO: Caves and Ice Age Art in the Swabian Jura · British Museum: Cuneiform · SignBase Dataset (Nature Scientific Data) · Dr. Hugh Thomas, “40,000-Year-Old Writing” (YouTube)