Did Humans Invent Writing 43,000 Years Ago? The Statistics Say \"Maybe, Sort Of\"
Writing is older than you think. But maybe not that old. In February 2026, researchers Christian Bentz and Ewa Dutkiewicz published a paper in the Proceedings of the National Academy of Sciences arguing that marks scratched into bones and antlers 43,000 years ago might be the earliest ancestors of written language. The internet did what the internet does. CNN ran it. Smithsonian Magazine ran it. Science Magazine ran it. And then the linguists started fighting.
Bones, Antlers, and a Very Big Claim
The artifacts in question come from the Swabian Aurignacian period, roughly 43,000 to 34,000 years ago. They're mostly made of bone, antler, and ivory. Someone scratched marks into them. Those marks sat in the archaeological record for tens of thousands of years, and now two researchers looked at them and said: this looks like it could be writing.
To make that argument, Bentz and Dutkiewicz compared the Paleolithic engravings to early Mesopotamian protocuneiform from the Uruk V period. Protocuneiform is some of the oldest recognized writing we have, basically ancient accountants recording grain and sheep. The comparison found statistical similarities. Hence the headlines.
There's just one problem with the statistics.
91% Agreement Sounds Great Until You Read the Fine Print
Here's where word nerds need to pay attention, because this is genuinely interesting and also a bit of a trap.
The study reported raw inter-coder agreement of 91 to 94 percent for classifying the marks. That number sounds impressive. When multiple researchers look at the same marks and agree 91 percent of the time, that feels like solid science.
But statisticians have a tool called Cohen's kappa, which corrects for how often people would agree just by chance. When you apply it to Bentz and Dutkiewicz's data, the scores drop to 0.29 to 0.44. The formal label for that range is "fair-to-moderate" consensus. Not "strong." Not "conclusive." Fair to moderate.
A critique preprint by d'Errico, Santos da Rosa, Courtenay, Núñez, Liu, Sproat, and Blasi (available at SSRN abstract 7053279) flagged this gap. The raw numbers look good. The chance-corrected numbers tell a different story. If you're going to claim something predates recognizable writing by 40,000 years, the kappa matters.
The Pennsylvania Barn Star Problem
The critique raises a second issue that might be the most entertaining problem in this whole debate: the missing control sets.
Bentz and Dutkiewicz compared the Aurignacian engravings to pre-literate Uruk accounting texts and later written language. What they did not compare them to: decorative systems that are not writing at all.
Why does that matter? Because of a 2014 paper by Sproat showing that decorative systems, including Pennsylvania barn stars, can resemble conventional sign systems on statistical measures. Pennsylvania barn stars are not writing. They are hex signs painted on barns. But run the right statistical test and they start to look like they might be.
If you want to prove that marks on a 43,000-year-old bone look like writing rather than decoration, you need to show they look more like writing than decoration does. Bentz and Dutkiewicz did not include decorative control sets in their analysis. The critique points this out. It's a meaningful gap.
This Is Not the First Time Someone Has Said This
To be fair to Bentz and Dutkiewicz, this conversation has been heating up for a while. Bacon et al. published a paper in the Cambridge Archaeological Journal in 2023 claiming an Upper Palaeolithic proto-writing system and phenological calendar. Dutkiewicz herself, also in 2023, argued that an artifact called the "Adorant" constitutes deliberate numerical representations. There's a pattern here of researchers looking at very old marks and seeing the seeds of language.
The Bentz and Dutkiewicz paper is published in PNAS volume 123, issue 9, article e2520385123. That's a serious journal. These are serious researchers. The question isn't whether the work is worth doing. It's whether the evidence yet supports the conclusion.
What It All Means for the Story of Writing
Here's the honest answer: nobody is claiming your Aurignacian ancestors were writing grocery lists. The Swabian artifacts predate anything identifiable as writing by roughly 40,000 years. Even if the marks have some structural similarities to early writing systems, "similar" is not the same as "ancestral."
What the debate is really about is where symbolic communication begins. When does a scratch become a sign? When does a pattern become language? These are questions that make linguists, archaeologists, and philosophers argue at conferences, and the Bentz and Dutkiewicz paper just threw fresh fuel on that fire.
For now, the evidence is suggestive but contested. The statistics are interesting but incomplete. And somewhere, a researcher is probably staring at a very old piece of bone, wondering if someone with thoughts and intentions made those marks on purpose.
That's not a bad thing to wonder about.
Source: Languagelog