Evolutionary Insights by Anthropology.net

In the Sakhnin Valley in Israel’s Lower Galilee, a survey recorded more than 200 handaxes from the Lower Paleolithic, the earliest part of the Old Stone Age. About ten of them were shaped from flint nodules with a fossil imprint or an odd geological feature inside, and in each the maker left the feature in the middle of the tool’s face, where it could be seen and touched. Such nodules are harder to work, and the valley’s other handaxe makers avoided them.
These handaxes are part of a new case for Stone Age souvenirs, one that runs against a view most readers probably hold. Keeping an object for the memory it carries is usually treated as a late habit of modern humans, from the end of the Old Stone Age onward, and archaeologists have mostly read such objects socially, as carriers of shared meaning or markers of rank.
The paper, a review in Evolutionary Anthropology: Issues, News, and Reviews, is by Ran Barkai, a professor of archaeology at Tel Aviv University; Yoav Ram of the university’s School of Zoology; and four colleagues. Its data statement says no new datasets were generated or analyzed. Its core examples come from the team’s earlier work in Israel, which they read as consistent with people keeping objects beyond practical need from about 500 kya (thousand years ago). They propose that these were personal keepsakes, or souvenirs, working as “portable external memory devices.” Once keeping them became routine, the habit may have changed the selection pressures on autobiographical memory, which they define as memory of one’s own experiences plus the family stories folded into one’s identity.
What is the evidence for Stone Age souvenirs?
The archaeological case rests on three groups of objects, all described in earlier papers by members of this team. At Qesem Cave, people deliberately carried in small, round pebbles of colorful flint from about 400 kya onward, through the roughly 200,000 years they used the cave. Ella Assaf, one of the authors, reported in 2018 that the pebbles show no sign of use; the review gives no count. The Sakhnin handaxes, published this year by Barkai and Shalata, carry fossils and formations that match outcrops in the valley, tying each tool to a particular spot. The authors grant these were probably working tools and ask why someone kept a feature that did nothing for the tool and made it harder to produce.
In my reading the third group is the strongest, because the comparison that matters is built into the sites. At Revadim and Jaljulia, two Late Acheulean sites (the Acheulean is the Lower Paleolithic handaxe tradition), and at Qesem, which comes after it, people picked up flint tools made and thrown away long before. The giveaway is patina, a weathered skin that forms on discarded flint and takes its color from the conditions where it lay. On a Qesem scraper shown in the review, the old surfaces are whitish and the resharpened edges are brown, fresh flint. Reworking on such pieces was minimal, and magnified wear traces show their second use was generally light. Because fresh tools dominate the assemblage (the full set of recovered artifacts), Bar Efrati of the University of Oxford, a co-author, argues that the old faces were kept by choice rather than for lack of good stone.
The memory half of the argument rests on what the team calls authentication. Psychologists use the term source monitoring for telling whether a memory came from something that happened or from one’s own thoughts. A keepsake adds sensory detail to that judgment; it can’t prove a memory true, the authors stress, only narrow down where it came from. Where alliances and claims to places depend on credible personal memory, people who are better at checking memories against objects should do better. That skill varies between people and is partly heritable, the paper notes, so selection would have had something to act on. Routine curation beyond practical need is what the authors call a “souvenir niche,” after niche construction theory, in which organisms alter their surroundings in ways that feed back on their own evolution.
People had lived in the Levant (the eastern Mediterranean) for at least 1.5 million years before this behavior appears, so the review looks for a trigger and finds it in big game. Prey size there fell through the Pleistocene, the epoch of the ice ages, and the elephant disappeared around 400 kya, along with Homo erectus and the Acheulean. As hunting smaller, more evasive prey pushed people into new territory, the authors propose, portable pieces of places became worth keeping. Barkai said in the press release that objects taken from the new hunting grounds “may have helped them maintain a personal connection with the places that would henceforth provide their livelihood.” The review predicts that the practice should track the decline of large herbivores rather than come before it.
Can archaeology tell a keepsake from a tool?
The patina deserves a slower look, because the time it records works against part of the argument. The whitish skin on that Qesem scraper formed while the flake lay abandoned, and the authors acknowledge that the patinas imply the pieces were found again rather than handed down. So the long interval written on these objects is time when nobody had them. The abstract speaks of the “prolonged retention” of patinated tools, but nothing the review presents measures how long the second owner kept one. Light wear on a resharpened edge fits an object handled now and then for years, and it fits a tool used briefly and dropped; the team itself calls wear its weakest criterion.
That gap matters because of the review’s own test for a souvenir: an individual keeps one object safe over a long period, neither discarding it nor trading it for food, and comes back to it from time to time. The authors use this standard to set aside animal cases, including captive chimpanzees that kept a large crystal in their sleeping quarters for two days before trading it back for food. They concede that such behavior could go unseen, because detecting it means following a known individual and a known object for months. No Paleolithic deposit offers that. A site preserves where objects ended up, and how long anyone kept them, or why, has to be inferred.
I also think Barkai and Ram chose the hardest level of meaning to detect. They focus on personal rather than social meaning because it assumes the least, needing no shared convention and no audience. That is sound as theory, and it is also why a personal keepsake is nearly invisible to archaeology, since by the paper’s own account its meaning need not be legible to anyone but its keeper. The definition of the souvenir niche sidesteps the problem. It covers routine curation “regardless of motive,” while the memory argument depends entirely on the motive.
How strong is the case that keepsakes shaped memory?
I find the evolutionary step the weakest link, because its evidence comes from studies of living people and laboratory animals, such as experiments in which learning to make and use tools changed brain networks in macaques and humans. Such studies can make a mechanism plausible, but they can’t show selection at work in the Middle Pleistocene, when these stones were collected.
The 500 kya figure is a round number for a period, not the age of any object in the review. It gives no dating method for any of the Israeli material and no stratigraphic context (the layer and setting that would date them) for the surveyed Sakhnin handaxes. Its only number for that evidence is the 400 kya start of the Qesem pebbles, and the authors allow that routine souvenir-keeping may have arrived “later than the earliest examples in the archaeological record.”
Their test of failure is also forgiving. The hypothesis fails only when curated items are optimized for use, local and uniform…


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.anthropology.net/subscribe

In a new plot of brain shape, the Taung child, a roughly 3-million-year-old skull found in 1924 at the Buxton Limeworks quarry in South Africa’s Northwest Province, sits almost on top of MLD 37/38, an adult from Makapansgat, 670 km to the northeast. The plot compares endocasts, casts of the inside of the braincase that record the brain’s outer shape, and Taung’s comes from a child whose first permanent molar had erupted while the second was still forming. In growth series of 101 chimpanzees (Pan troglodytes) and 84 gorillas (Gorilla gorilla), apes at that dental stage do not overlap their own adults in brain shape, yet Taung plots in the middle of the adult fossils from Sterkfontein. The authors call this peculiar, and I think it is the most important result in the paper, because it is where their case about species is both strongest and most exposed.
That matters because Taung is the holotype of Australopithecus africanus, the specimen that anchors the species name. Most readers picture a well-stocked species behind it, with Sts 5, the Sterkfontein skull nicknamed “Mrs. Ples,” as its typical adult. The paper argues that Taung is, strictly speaking, the only fossil that belongs to A. africanus without dispute, and that several adults long filed under the name, MLD 37/38 among them, probably belong to other species.
Gabriele Macho (Vrije Universiteit Brussel and University College London) and seven coauthors, including Simon Neubauer and Clement Zanolli, report the work in npj Palaeoecosystems. A flowstone lying above the fossil-bearing deposit gives a minimum age of 2.962 ± 0.059 million years (Ma) for the skull, and an earlier paleomagnetic estimate caps it at 3.03 Ma. Using tooth development, brain growth and molar shape, the team asks whether Taung could have grown into the adults assigned to its species, and concludes that the brain-growth results leave little doubt that the Sterkfontein fossils include more than one species. To explain that diversity, the authors propose that climate swings kept moving populations around, with interbreeding probably following where they met.
How old is the Taung child?
Southern African sites lack the volcanic ash layers that give eastern African fossils reliable dates, and Taung’s exact find spot was lost during mining. Published estimates have ranged from 942 thousand years (ka), by uranium-series dating, to 3.03 Ma, the start of the last normal-polarity interval of the Gauss magnetochron, a period identified from the magnetic direction recorded in rock.
The flowstone, a sheet of calcite laid down by flowing water, was dated by the decay of its uranium to lead. The route previously advocated for South African flowstones, uranium-238 to lead-206, needs a correction for uranium isotopes that were out of balance when the calcite formed, and beyond 1.5 Ma the uncertainty in that correction can exceed the correction itself. The uranium-235 to lead-207 route is largely immune, and it gave 2.962 ± 0.059 Ma with little scatter (an MSWD, the standard measure of that scatter, of 0.7).
I find the measurement convincing, because the method was chosen for a stated reason and the result is precise, but what it means for the fossil is less certain than the paper or the press release suggests. The two routes give ages for the same flowstone that do not overlap, a mismatch the authors tentatively blame on uranium moving after the rock formed, and they read both ages as minimums by assuming the rock only gained uranium; loss would make them too old, and the main text does not say why gain is likelier. No radiometric ceiling has been found despite further drilling, so the 3.03 Ma cap rests on paleomagnetic data from a 2013 chapter by A. I. R. Herries and colleagues.
Does the Taung child share a species with the adult fossils?
The adults in question are Sts 5, Sts 60, Sts 71 and StW 505 from Sterkfontein, in the Cradle of Humankind about 390 km northeast of Taung, plus MLD 37/38, all attributed to A. africanus. The weakest of the three comparisons is tooth development: scored for 11 Australopithecus and 16 Paranthropus fossils, it gives the A. africanus fossils a markedly different slope from every other taxon when first-molar development is set against first-incisor development, a possible sign that more than one species has been lumped together. One Makapansgat specimen, MLD 2, may be skewing that line, and the authors warn against over-interpreting such small samples, so I take the slopes as a hint.
Brain size narrows the field only a little. The adults’ endocranial volumes range from 391 to 568 ml, a spread the authors find compatible with one species by chimpanzee and gorilla standards. Apes at Taung’s stage have reached about 92 percent (chimpanzees) and 87 percent (gorillas) of average adult volume, whereas Taung, at 402 to 410 ml, would have reached only 71 percent if it were going to grow into the large-brained StW 505.
The argument gets its force from brain shape, measured with 935 landmarks and semilandmarks (matching anatomical points, plus points spread along curves and surfaces) on every endocast, fossil and ape, after the fragmentary fossils were rebuilt from CT scans. Projecting the direction and amount of change seen in growing apes forward from Taung, the team finds the child could have grown into a brain shaped like Sts 5 or Sts 60 but not like Sts 71, StW 505 or MLD 37/38, even allowing for differences in the timing and amount of growth. As I read the plot, MLD 37/38 drops out precisely because it sits where Taung sits: a child with a lot of shape change still ahead of it should end up somewhere else.
I find this the strongest evidence in the paper for more than one species, because it uses dense shape data and large ape growth samples rather than a few scored teeth. Its weak point is the assumption underneath, which the authors state openly: that these brains kept changing shape after the first molar erupted, in the direction and roughly the amount seen in living apes. Chimpanzee, gorilla and human growth models all predict large changes from Taung’s stage onward, so the assumption is reasonable, but the paper has no australopith growth series to test it, and the plot covers only two shape axes, which carry 49.5 percent of the variation.
The same kind of reasoning has already given a different answer. A 2006 simulation of Taung’s facial growth by K. P. McNulty, S. R. Frost and D. S. Strait found Taung closer to Sts 71 than to Sts 5, judging the gap from Sts 5 extreme by the standards of living African apes and humans. Two projections along growth paths borrowed from living species disagree about the same pair of adults, and until they are reconciled I treat the Sts 71 question as open.
The molars pull Taung somewhere else again: on the dentino-enamel junction, the buried boundary between enamel and dentine that forms early and barely changes with wear, Taung’s lower first molars sit at the edge of the A. africanus cluster and inside the range of Paranthropus robustus on the main axis, though the first two axes explain less than 38 percent of the shape variation.
What the Taung study can’t tell us
The authors are frank that no clear picture emerges, with no consistent clustering of specimens, which they put down to small samples and fossils scattered widely in time and space.
The climate and hybridization proposal is the part the data support least, since the paper never tests hominin hybridization directly. Marine cores off the Limpopo record more variable climate between 4 and 2.5 Ma than from 2.5 to 1.6 Ma, and the authors suggest that hominins tracked shifting habitats and could have interbred where populations met along the Zambezian and Natal/Kalahari biogeographic boundary, which today runs through the Limpopo valley at Makapansgat. They raise the obvious problem themselves, that the Cradle of Humankind lies about 400 km south of the Limpopo and Taung farther still, and answer that hybrids c…


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.anthropology.net/subscribe

Chornancap Burial 48 comes from the lower Lambayeque Valley on Peru’s north coast and dates to the Chimú period, around 1375 to 1490 CE. The left side of the forehead carries a fracture from a blunt blow, and its edges show no healing, which bioarchaeologists (specialists in human remains from archaeological sites) call perimortem: the injury happened at or near the time of death. A new study of violence in ancient Peru uses this skull to illustrate what its authors counted as interpersonal violence. Among 1,519 skeletons from the valley preserved well enough to examine, spanning more than 3,000 years, they found 25 with skull injuries of that kind.
That is a low count for a coast best known in the press for ritual killing. At Huaca de la Luna, a Moche ceremonial center, up to 107 adult men were sacrificed in six closely timed events around 550 to 600 CE, amid evidence of mud and heavy rain from an exceptionally strong El Niño, the periodic warming of Pacific waters that drenches this desert coast. At Huanchaquito-Las Llamas in the Moche Valley, more than 140 children and over 200 young camelids (llamas and their relatives) were sacrificed around 1450 CE, associated with a mud layer that may record El Niño flooding. Neither site is in the new dataset. Finds like these invite a reasonable inference: when the floods came, these societies turned violent, and mass sacrifice was the visible peak of a wider rise in killing over scarce food.
Dr. Thomas J. Snyder of New York University and Dr. Haagen D. Klaus of George Mason University test that inference in a paper published November 9, 2026, in Frontiers in Environmental Archaeology, using skeletal records from the Lambayeque Valley Biohistory Project, which Klaus founded in 2002. The Lambayeque Valley Complex (the Lambayeque Valley, for short) consists of five river drainages more than 700 kilometers north of Lima, joined into one water network by the late first millennium CE. The team compiled records for 1,659 people buried at 30 sites, from the Cupisnique period to the Spanish colonial era. Twenty-six were dropped because they couldn’t be placed securely in time, and 1,519 had bones preserved well enough to score. Of those, 25 (1.6 percent) had skull injuries consistent with interpersonal violence, and 77 (5.1 percent) had fractures below the skull of the kind usually attributed to accidents.
Neither kind of injury showed a detectable relationship with the two climate measures tested, highland rainfall and El Niño frequency. The authors also propose, without testing it, that the coast’s mass sacrifices were political and religious acts rather than a product of food shortage. For scale, they note that average rates of skull injury in some regions and periods elsewhere in the Andes reach 30 to 45 percent (Snyder and Arkush 2024).
How do archaeologists measure violence from skeletons?
The method rests on a convention in bioarchaeology: skull injuries count as probable violence, and fractures below the skull as probable accidents. Snyder and Klaus say plainly that this proxy, or stand-in, errs both ways, since a beating that bruises without breaking bone leaves nothing behind and some skull fractures come from falls. Their defense is specific to their question: an undercount doesn’t distort the comparison unless it changes with climate, as it would if fights in drought years tended to stop short of breaking bone.
Climate is harder to measure, because the north coast is a desert that gets a few millimeters of rain a year; the water in its rivers and canals falls in the highlands above about 1,500 meters. So the authors measured drought at the source, using CHELSA-TraCE21k, a global climate simulation that estimates past rainfall in 100-year steps, sampled at random points across the valley’s highland headwaters. Simulations can’t reconstruct how often El Niño struck, so the authors used a per-century frequency record from sediment cores at Laguna Pallcacocha, a lake the paper describes only as near the valley.
The hypothesis is what the authors call a Malthusian pathway, where scarcity breeds violent competition: drought or frequent floods should shrink harvests and push people toward fighting over food, and toward longer hours of risky work that breaks more bones by accident. Each person was assigned to one of 12 cultural periods, and each period got one average for rainfall and one for El Niño frequency. A Bayesian model then estimated how the odds of each kind of injury shifted with those values, alongside age, sex, colonial rule, and burial site, reporting each effect as a credible interval, a range of plausible values. For both kinds of injury, the 89 percent intervals for both climate measures straddle zero: the data can’t tell whether either one raised the odds or lowered them.
I find the skull count convincing and the climate test much weaker, and the weakness comes from that design. Everyone who died in the Late Sicán period, from 1100 to 1375 CE, carries the same rainfall value, so although the model holds 1,633 people, the climate effect rests on at most a dozen period averages. The paper’s own history section describes an event this can’t resolve: a powerful El Niño between 1050 and 1100 CE, followed by a 30-year drought. Folded into averages for the Middle Sicán period (900 to 1050/1100 CE) and the Late Sicán period, each 150 years or longer, that drought shrinks to a modest shift in one or two numbers, set against only 25 injured people. Snyder and Klaus concede that rarity limits their power, which is why the abstract reports no strong evidence for an association rather than evidence against one.
What counts as everyday violence in ancient Peru?
The choice that shapes the headline number most is one the methods section barely describes: which deaths count. The authors study what they call quotidian, or everyday, violence, so like the regional surveys they build on, they remove sacrifice and massacre victims. It matters here: the paper cites evidence that perhaps more than 1,200 people were buried as offerings atop six monumental huacas, or sacred mounds, at the Sicán capital within as little as 50 years at the start of the Middle Sicán period (Klaus and Shimada 2016). Later Muchik communities, the Indigenous people of this coast, carried out smaller sacrifices at the Chotuna-Chornancap complex and at Túcume, both on the study’s site list.
I don’t object to the split itself. Elite-staged mass sacrifice and a fight between neighbors are different social acts, and pooling them would make Lambayeque incomparable with other regions. My objection is that the paper never says how many individuals were removed on these grounds, or how a burial was judged ritual. Burial 48 is a person struck on the head around the time of death at a site with documented sacrifices, and the authors present it as interpersonal violence; I’d want to know what settled that call. And with massacre victims removed too, the measure captures violence between individuals, which may not be where competition over food shows up.
Colonial burials are where I think the measure’s limits show most clearly. After the Spanish conquest, Muchik people were forced into colonial towns and exploited for their labor, and the region lost about 40 percent of its population before a slow Indigenous recovery began around 1600 CE. Yet the authors describe skull injuries in their two large colonial samples, from Mórrope and Eten, as nearly absent. The…


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.anthropology.net/subscribe

In the central burial chamber of a princely tomb in Tyva, in southern Siberia, human hair survived for roughly 2,800 years, and several samples give clear chemical signals consistent with cannabinol, a compound that can form when the main intoxicant in cannabis oxidizes. The tomb belonged to high-ranking members of Scythian society, the horse-riding culture of the Eurasian steppe, and it also held about 1,000 cannabis seeds. A new paper in Science presents these finds as the earliest direct evidence of ancient cannabis use.
The belief this tests is one many of us hold without much scrutiny, that cannabis in an ancient grave means someone was getting high. People have often suggested as much, but the evidence is awkward. Cannabis also supplied oil and fiber, so seeds in a burial don’t prove intoxication, and residues on vessels and heated stones from later Central Asian sites point to inhalation while recording an object rather than a person. Hair could close that gap because it comes from a body.
The study’s lead author is Dr. Gino Caspari, an archaeologist at the Institute of Archaeological Sciences of the University of Bern and the Max Planck Institute of Geoanthropology. He and eight coauthors, including Timur Sadykov and Daria Fomicheva, conclude that cannabis consumption dates back at least 2,800 years. In the press release issued with the paper, Caspari takes the next step: “The molecular traces in the hair and the cannabis seeds come from the same grave and suggest that cannabis was used for its intoxicating properties.”
I should say exactly what I read. The full text is behind Science‘s paywall, and for this post I worked from the abstract, the editor’s summary, and the release, without the methods or the supplement and its eight figures and seven tables. When I call a detail missing below, I mean missing from those documents. On that evidence, I think the case that people in this tomb used cannabis in some form is good, and the case that they used it to get high is weaker.
How do you find cannabis in 2,800-year-old hair?
The date comes from the seeds. Radiocarbon dating measures the decay of carbon-14 in once-living material, and six seeds were dated directly. I consider this the firmest part of the study because the date belongs to the cannabis itself and not to something found near it. Together with other dating results, those measurements put the seeds in the grave in the early eighth century B.C., which matches the 2,800 years in the paper’s title. The error ranges, and what the other dates were taken on, are missing from what I read.
The chemistry comes from the hair, and it is easier to follow once you know the two compounds involved. Tetrahydrocannabinol, or THC, is the main intoxicating compound in cannabis. Cannabinol, or CBN, is one of the things THC becomes when it oxidizes, which makes it a sensible target in very old material. The team used high-resolution mass spectrometry, among other methods; the technique sorts molecules by mass precisely enough to match a signal to a candidate compound. Several samples gave clear signals consistent with CBN, and the team also saw weak signals whose mass is consistent with THC. I can’t tell you how many samples were tested, how many people they came from, or those people’s ages and sex.
Does cannabis in ancient hair prove people got high?
The step I’d slow down on runs from the hair to the word “psychoactive,” because “direct evidence” fits the presence of a compound better than it fits anyone’s state of mind. I also trust the CBN more than the THC. The CBN signal is described as clear, although the release says THC oxidation is only one way CBN forms. The THC signal is described as weak and rests on mass, and a mass match says a molecule of the right weight was present without settling which molecule of that weight it was. The editor’s summary, which says the team identified psychoactive compounds in the hair, is firmer than the release’s own language.
Quantity is the larger problem, because intoxication depends on how much a person took in, and the release says the hair can’t give a reliable amount. These seeds likely came from wild plants, and people had other uses for cannabis, so I can’t rule out from what I read that a person with no interest in an altered state could carry a similar trace. Caspari’s word, “suggest,” seems right to me, and it is more careful than the paper’s title.
The ritual reading has a similar gap. An earlier paper by the team described 18 horses arranged on the burial mound, and several dozen more horse sacrifices have turned up since, along with scattered human bones. Caspari says the burials were accompanied by “most likely, human sacrifices as well.” I accept that the seeds belong to that funerary setting. Hair records a person’s body, though, and nothing I read says whether the cannabis signal reflects a funeral or the years before death, so the editor’s summary’s “ritualistic psychoactive use” joins two findings that may describe different moments.
What the Scythian tomb study can’t tell us
The gap I’d most like closed is contamination, which the materials I read don’t address. The release says the team can’t determine how the compounds entered the body, which assumes they did enter it. This hair lay for about 2,800 years in a tomb containing cannabis offerings, and if anything was burned at the funeral, smoke could have settled on it too. A test separating compounds taken up by the body from compounds sitting on the hair is the first thing I’d look for in the supplement, along with control samples from the surrounding sediment.
The date and the chemistry also attach to different objects. The release calls this the oldest directly dated evidence of intoxicating use, but the direct dates it describes are on seeds, and it gives no separate date for the hair. The two are linked by coming from one grave, and in a tomb with more than one burial and scattered human bones, I’d want that association spelled out.
On potency and reach, the abstract says more than the release does. It floats the idea that specific wild populations might have had naturally higher levels of active compounds, while the release says the findings allow no conclusions about the plants’ THC content. The abstract also says the findings suggest use across Eurasia and that they anchor current debates over cannabis in law, public health, and other fields. On the evidence I could see, one tomb in Tyva can’t carry those last two claims, and I would have cut them. The release’s headline has its own problems. It says flatly that cannabis “was used as an intoxicant,” where the abstract says the findings “suggest” psychoactive use. It also says “as early as 2,800 years ago” where the abstract says “at least 2800 years”; the first reads like a starting point, and the second leaves room for an older practice.
What would settle the question of ancient cannabis use?
My verdict is that the study is partly convincing. It adds a human body to evidence that, in the documents I read, otherwise comes from plants and objects, and the direct dates tie the seeds themselves to the early eighth century B.C. If the supplement deals properly with contamination, I’d accept that people in this tomb had taken in cannabis by about the early eighth century B.C. Intoxication seems to me the likeliest explanation, given the funerary setting and the later evidence of inhalation, but it remains an inference from context, so my con…


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.anthropology.net/subscribe

On the Danish islands, the local ancestry that had been widespread between about 2200 and 1600 BP had essentially disappeared by 1250 BP, roughly a century before the Viking Age. (BP means calibrated years before present, with the present fixed at 1950 CE.) In its place, the Scandinavian-related ancestry of most individuals is best modelled by an Iron Age source from northern Germany or southern Jutland. That is the central finding on Viking ancestry in a new ancient DNA study of how the Germanic languages spread. It was published with a companion paper that applies the same data to the origin of the Celtic languages.
The usual picture has Viking Age Danes and Swedes descending from people settled in southern Scandinavia since the Bronze Age, speaking a language that developed in place. Earlier genetic studies detected southern ancestry arriving but, the authors note, generally assumed local continuity. The new work says that continuity broke in Denmark and southern Sweden during the Migration Period (1575 to 1375 BP), when Old Norse, the language of the Vikings, took shape.
Both papers appear in Nature. Hugh McColl is first author on both, sharing that position with the linguist Guus Kroonen on the Celtic paper, and Eske Willerslev is last author. The team sequenced 712 new ancient genomes and combined them with published data into a working set of 4,587 individuals spanning roughly 5,000 years. The Germanic paper proposes that a population it calls Eastern Scandinavian, linked to the northeastern Baltic, reached Sweden by about 4000 BP and carried the earliest form of Germanic. That is some 800 years after the Corded Ware groups traditionally credited with bringing it. The paper also traces most of the Scandinavian-related ancestry of the Anglo-Saxons, Frisians and Langobards to Jutland and northern Germany.
The companion paper tests four competing models for how Celtic reached an Iron Age range that ran from Iberia to the Balkans. It detects three large movements of people between 4400 and 2800 BP. It judges the last of them to be the most parsimonious vector for Celtic, meaning the one that needs the fewest extra assumptions: a westward spread from eastern central Europe between 3200 and 2800 BP, associated with the cremation-dominated Urnfield Culture. That puts the paper behind the longstanding Celtic-from-the-East model. In both papers the most important conclusion sits beside the largest gap in the sampling, and that shapes how far I would trust each one.
How does ancient DNA trace the spread of a language?
DNA cannot show what anyone spoke, as both papers acknowledge, but it does show who descended from whom. The method uses identity-by-descent (IBD) segments, which are stretches of chromosome that two people share because they inherited them from a common ancestor. Recombination shortens these stretches every generation, so long segments mark recent relatives and short ones deeper links. Many genomes here are thin, down to 0.1× coverage, meaning only about one position in ten was read even once. The team filled in the missing genotypes by imputation from a reference panel of present-day genomes.
In IBD mixture modelling, each individual gets a palette recording how much DNA they share with each of 386 genetic clusters. A target’s palette is then fitted as a blend of the palettes of chosen source groups. Because every palette draws on the whole dataset, a Viking Age Dane can be linked to an Iron Age source population even when the two share few segments directly.
Applied to Scandinavians older than 2800 BP, the method finds three clusters. The Eastern Scandinavian one carries extra hunter-gatherer ancestry best matched by hunter-gatherers from Latvia and Lithuania. The authors attribute it to an unsampled population between Finland and northeastern Sweden; no genomes yet exist from that region and period. Between about 3200 and 2450 BP, early Germanic took loanwords from Celtic to the south and lent words to Finno-Saamic languages to the east. Neither of the other clusters reached both contact zones, so the authors call Eastern Scandinavian ancestry the most parsimonious vector.
Tracking later movements is harder, because everyone who postdates the arrival of Steppe ancestry in Europe is so closely related. The Celtic paper’s solution, which I think is the smartest move in either study, is to follow which regional variety of older Neolithic farmer ancestry people carried. Bell Beaker groups absorbed different local farmers in different places, so a source of confusion becomes a regional tag. In England, ancestry tied to Neolithic farmers of Iberia and France overtakes local farmer ancestry between 3800 and 3200 BP. That matches an earlier report of large-scale migration into Britain (Patterson and colleagues, 2022). Between 3200 and 2800 BP, ancestry tied to Bronze Age Hungary and Serbia rises in England, becoming widespread by at least 2600 BP.
Iberian individuals from Urnfield contexts carry that eastern ancestry by 3150 BP. The authors use ancestry related to the Knovíz Culture of the Czech Republic as a stand-in for Urnfield populations generally. By 2800 BP it is common across former Urnfield regions, though its effect weakens westward and the older Bell Beaker-related ancestry still dominates in Iron Age England. The authors then checked each movement against published linguistic dates. These covered the formation of Proto-Celtic, the undocumented ancestor of all Celtic languages, and the later split between its Brittonic branch (including Welsh) and Goidelic branch (including Irish). Only the third movement fit most estimates on every count.
Why does cremation leave gaps in the ancient DNA record?
What I most want readers to notice is where the missing genomes fall. Both papers place their most consequential change in centuries when most people were cremated, and cremation destroys DNA. For the Celtic paper this is the Urnfield period of 3200 to 2800 BP. The authors say the timing of the transition in France cannot be measured directly for lack of Late Bronze Age samples, so they infer it from Britain and Iberia.
For the Germanic paper, the gap runs from 1575 to 1200 BP, when Old Norse formed. I counted the individuals in the paper’s ancestry plot for Denmark and Sweden (Extended Data Fig. 13). The panel for 1575 to 1230 BP holds 15 people, against roughly 200 on each side, and only four are Danish genomes older than 1250 BP. Because hundreds of genomes anchor each end, I accept that the population of the Danish islands changed, but the genomes cannot say when within the window. The paper’s date, probably 1550 to 1450 BP, comes from non-genetic evidence, including tree rings from 654 Danish oak samples.
Norway is the result I think historical linguists should weigh most. Old Norse took shape there too, yet the paper finds genetic continuity in Norway from the Iron Age to the Viking Age. The authors conclude that demographic and cultural mechanisms combined to drive the language. They also observe that the later boundary between East Norse and West Norse dialects roughly follows the boundary between the newcomers’ ancestry and Norwegian Iron Age ancestry, even in Viking areas of Britain. That is a firmer link between genes and language, because it concerns a split inside a language rather than its birth. The authors also suggest the expansion through Denmark may partly explain why Old Norse speakers called their language dǫnsk tunga, the Danish tongue.


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.anthropology.net/subscribe