How WEIRD is Cognitive Archaeology? Engaging with the Challenge of Cultural Variation and Sample Diversity

In their landmark 2010 paper, “The weirdest people in the world?”, Henrich, Heine, and Norenzayan outlined a serious methodological problem for the psychological and behavioural sciences. Most of the studies produced in the field use people from Western, Educated, Industrialised, Rich and Democratic (WEIRD) societies, yet inferences are often drawn to the species as a whole. In drawing such inferences, researchers implicitly assume that either there is little variation across human populations, or that WEIRD populations are generally representative of the species. Yet neither of these assumptions is justified. In many psychological and behavioural domains, cultural variation begets cognitive variation, and WEIRD samples are recurrently shown to be outliers. In the years since the article was published, attention has focused on the implications this has for research on extant human populations. Here we extend those implications to the study of ancient H. sapiens, their hominin forebears, and cousin lineages. We assess a range of characteristic arguments and key studies in the cognitive archaeology literature, identifying issues stemming from the problem of sample diversity. We then look at how worrying the problem is, and consider some conditions under which inferences to ancient populations via cognitive models might be provisionally justified.


Prelude
In 2010 Henrich and colleagues argued that the behavioural sciences face a serious methodological issue: most of the results in the field are produced using participants from WEIRD populations, yet these results often fail to replicate in cross-cultural studies (Henrich et al. 2010).Moreover, cross-cultural research suggests that WEIRD people are often outliers with respect to many cognitive and behavioural traits.So it seems that inferences from culturally localised samples to species-wide psychological claims are unjustified.The take home lesson is that, when it comes to behavioural psychology, culture matters; and often matters profoundly.There is no prior guarantee that an effect will be reproduced beyond the group it is found in, nor that it is an inlier with respect to the broader population.1 Therefore, generalise with care.
So far, so worrying.But we think there is more to be worried about.The bulk of scholarly attention in the wake of Henrich and colleagues' article has focused on the issue of synchronous generalisation; that is, inferring from WEIRD populations to the (extant) human species at large ('generalisations').However, as we will show, in the field of human cognitive evolution we often make diachronic inferences; that is, inferences from modern WEIRD populations to populations of ancient humans, their hominin forebears, or cousin lineages ('past projections'). 2 And this too should be worrying-after all, ancient humans were certainly non-WEIRD.
Our focus is the discipline of cognitive archaeology.Cognitive archaeologists attempt to reconstruct the cognitive and cultural lives of ancient humans, Neanderthals, and their hominin ancestors by studying their material traces.One aim of the discipline is to shed some light on the evolution of human cognition.However, artefacts alone are insufficient to link information about behaviour to underlying cognitive capacity: a mid-range theory is required (Currie 2018;Binford 1972).So cognitive archaeologists typically appeal-either implicitly or explicitly-to a theory or model from the cognitive/psychological/behavioural (henceforth 'cognitive') sciences; one which is thought to be independently plausible.Using this model, inferences are made from artefacts to cognitive capacity.
We begin by introducing the challenge to cognitive archaeology in a little more detail, and provide some motivating examples (Section 2).We then outline four case studies which exemplify the issue of cross-cultural sample diversity in cognitive archaeology (Section 3).Next we look at how worrying the issue is, and briefly consider some conditions under which inferences from contemporary samplesmost often WEIRD humans-to ancient populations might be provisionally justified (Section 4).Throughout, our mantra will be that further cross-cultural testing of cognitive models stands to strengthen or undermine the cognitive archaeological by evidence from more diverse samples than the standard WEIRD range, and where possible to provide details of the sample so that inferences from WEIRD or otherwise narrow samples can be seen for what they are.
We do not have to look far for motivating cases.In their contribution to a special issue on the archaeology of children and childcare, Lew-Levy et al. (2020) argue that the results of developmental psychology experiments researching the innovation of children in WEIRD societies-results which suggest that children are poor innovators (e.g.Lister et al. 2020)-are not generalisable to children of small-scale, forager societies.And seemingly with good reason.Sterelny (2021) sums it thus: "...forager children are likely to innovate much more than WEIRD children, given that they grow up in an environment in which they have a lot of autonomy to play, to experiment and to explore the affordances of the material substrates to which they have access.In many forager cultures, experimental learning for oneself is positively encouraged.Furthermore, they routinely engage in peer-peer learning.While adults support social learning, they do relatively little directive teaching.Instead, they scaffold learning with equipment and raw materials, they provide occasional advice and they allow children to involve themselves with adult activities.Finally, children, even quite young children, do not just imitate in play adult economic and social activities, they practice those activities.They engage in subsistence activities, often in distinctive ways."(Sterelny 2021, p. 5) The general concern should be clear: if the results of innovation studies do not generalise beyond WEIRD populations in extant humans, then we have reason to question their applicability to ancient humans.Furthermore, if we think that the learning environments of ancient populations were more akin to those of small-scale forager societies, then we should be prepared to treat samples from WEIRD populations as outliers.
Of course, innovation is typically assumed to be heavily influenced by sociocultural processes.So the preceding example is perhaps not particularly striking.It might be thought, however, that more 'base-level' processes-such as those involved in visual perception-would not vary much across contemporary human populations.Henrich et al. (2010) provide a motivating case that undermines this line of thinking-the Müller-Lyer illusion-which is now a classic example for establishing cross-cultural variability (Segal and Campbell 1966). 5The two lines in the illusion are of equal length, though (to us and our colleagues, at least) they do not appear to be equal (see Fig. 1).Line A appears to us to be shorter.Researchers have tested the strength of the illusion by asking subjects how much longer Line A needs to be 5 For review, see Phillips (2011).There is some debate about the replicability of the original results and the extent to which the cultural variation exists.For example, Deręgowski (2013) argues that only Müller-Lyer figures with interfin angles between 90° and 180° range are appropriate for testing cultural variation.Modified versions of the experiment continue to demonstrate cultural variation (e.g.Nestor 2017), however.While we acknowledge that the strength of the variation may be less than initially thought, we follow Henrich et al. (2010) in utilising this phenomenon as a motivating example, one which still demonstrates cultural variation in what we think is a surprising domain.
than Line B in order for the two lines to appear equal (see Fig. 2).This quantity varies widely across societies, along a continuum.At one end, WEIRD undergraduates require Line A to be about 20% longer than Line B; at the other end, the difference in line length San foragers required was indistinguishable from zero.For the San, there is no illusion.Much has been suggested about the causal role one's environment plays in bringing the illusion into effect (carpentered corners being ubiquitous in WEIRD societies and absent in the Kalahari; see McCauley and Henrich 2006;Henrich 2008).6However, a causal explanation of such variation is not what is at stake here.Rather, the case illustrates that even base-level visual processing can be subject to cultural variation.And, if even visual perception can vary across populations, the range of cognitive/psychological processes that are sensitive to culture is plausibly very broad.Moreover, the fact that the magnitude of the effect is strongest in WEIRD subjects is of concern, as this demonstrates that WEIRD people are outliers.We cannot thus argue that despite cultural variation WEIRD subjects are generally representative of the population.And if the inference from WEIRD people to San peoples is not justified, then we should be likewise concerned about any inferences to ancient populations, for, plausibly, the socioecology of the San more closely resembles that of ancient humans than that of WEIRD populations.
There are other examples, not involving illusions.For instance, researchers once thought, based on their studies of university students at their home institutions, that humans exhibit a right-hemisphere bias in face-recognition processing (see Henrich 2020, pp. 3-7).However, we now know that the process of becoming literate rewires our neural circuitry.One effect of this is that face-recognition processing is shifted to the right hemisphere (Dehaene et al. 2010(Dehaene et al. , 2015)).So the generalisation from the all-literate sample to H. sapiens is undermined, as it appears that people from nonliterate societies do not exhibit such a bias (and, as it happens, are better facial-recognisers than literate people).Given that literate societies are relatively new, from an evolutionary perspective it would be an obvious mistake to infer on the grounds of the initial studies that ancient humans exhibited a right-hemisphere bias in facerecognition processing.
The purpose of this article is to apply this general line of critique to cognitive archaeology.We aim to draw out some of the issues raised by the challenge of cultural variation and sample diversity within the context of the historical sciences.

Case Studies
In this section we outline four examples from work in cognitive archaeology where WEIRD sampling issues are at play.In the first two, we target specific inferences concerning the evolution of spatial cognition and long-term working memory.We then expand our scope to look at research programs as a whole; namely the affordances framework and neuroarchaeology.

Case Study 1: Wynn & Piagetian Theory
Thomas Wynn is a cognitive archaeology trailblazer,7 and one of the most cited representatives of the discipline (as of the time of writing: over 6,200 citations according to GoogleScholar).Whether one agrees with his arguments or not, Wynn has been incredibly influential in the development of the field.For this reason, we pay particular attention to his work.Of course, we must be highly selective in an article of this length; but it is not from some fringe corner of the field that we select our examples.Indeed, Wynn's application of Piagetian psychological theory to the stone tool record is one of the earliest examples of a cognitive archaeological argument.This demonstrates that the problem of sample diversity was inherited (and indeed acknowledged) by cognitive archaeology from the outset; Piagetian theory's failure to generalise is one of the reasons Wynn discarded it in later work (see Wynn 2016).
Our other examples demonstrate that the problem persists in various guises in the contemporary literature.
Before beginning, we note that our comments are not intended to be an indictment of Wynn's work in general; there are many areas of his work that are not subject to WEIRD concerns.Wynn (1985), for instance, utilises comparative data on non-human primates, and in other work, Wynn considers the implications of gross brain morphology as gained from palaeoneurological data.Our analysis merely draws attention to specific instances from Wynn's oeuvre where sample diversity issues are at play. 8he work of twentieth century Swiss psychologist Jean Piaget has been incredibly influential in developmental psychology and beyond.Piaget proposed a sequential stage model for individual intellectual development known as "genetic epistemology" (Piaget 1972).The stages are invariant, each necessary for the next, and based on a structuralism that conceives of intelligence as governed by a set of operational principles.9In 1979's seminal "The intelligence of later Acheulean hominids", Wynn applied Piaget's genetic epistemology model to the evolution of hominin intelligence.Specifically, Wynn's goal was to identify the spatial cognition capacities of ancient hominins by interpreting the 300,000 year old stone tools from Isimila Prehistoric Site, Tanzania, in terms of the final stage of Piaget's model, 'operational thought'.
Each stage of Piaget's model presents a set of operational principles-characterising development from birth until adolescence-which regulate thought.The final stage of the model sees individuals between 11 and 16 years old develop abstract reasoning, logical thought, and metacognitive reasoning resources.Moreover, according to the theory, reversibility and conservation are the two fundamental regulatory operators of these cognitive principles.Reasoning with the principle of transitivity exemplifies conservation.Schematically, if x=z because x=y and y=z then it is due to the conservation of some property; an inference from x=y and y=z to the conclusion x=z is based on the same principle.Reversibility is the operation that takes one, through inversion, back to the starting point (e.g., 0+n-n=0) or, through reciprocity, to an equivalence (e.g., p≥q and p≤q lead to p=q).Although the lines of reasoning here can be formalised with symbols, they are intended to capture operations which, at least most of the time, we employ informally in casual settings.(For further examples of these operations-including their relevance to classifications of kinship systems-see Wynn's paper.) Wynn's interpretation of the stone tools from Isimila identified the targets of these psychological concepts in the tools' production.In doing so, he ran a minimum-capacity inference: "In order to manufacture all but the most rudimentary stone tools, however, flake removals must be related to one another in a fashion yielding the appropriate configuration or pattern.If a stone artefact presents a pattern of flake removals that could only have been organised by means of reversibility and/or conservation, then it must be concluded that the maker possessed  2010), reproduced here with the permission of Cambridge University Press operational intelligence.I will show that the later Acheulean artefacts from the Isimila Prehistoric Site present such patterns."(Wynn 1979, p. 374) His argument ran as follows.The reduction sequence from raw material to finished tool required the toolmaker to apply four kinds of operational spatial constructs: whole-part relations, qualitative displacement, spatio-temporal substitution and symmetry.Because each of these operations requires conservation and reversibility, the final stage of Piaget's model can be located in the makers of these tools.The bulk of Wynn's paper, then, is dedicated to explaining how one can infer the four spatial constructs from the knapping method identified in the Isimila bifaces, and how these relate to Piaget's two core ingredients of operational thought.(Although the details are interesting and informative, we must skip over them here; see Wynn 1979; also Wynn 1985 for discussion.) Wynn's specific conclusions aside, our concern is with the use of Piaget's model as mid-range theory.Did it provide epistemic licence to Wynn's projection to the past?There are many reasons that Piagetian theory has gone out of fashion (Bjorklund 1997(Bjorklund , 2018)).According to Parker and Gibson (1979, p. 400) its evolutionary application is "Lamarckian and vitalistic"; and for other researchers, its reliance on the principle of recapitulation is "dangerous" (Renfrew 1982, p. 14).Importantly for our purposes though, Piaget's developmental model has long been criticised for its limited sample size: it was based on the observations of middle-class European children, and yet proposed as a general model of human cognitive development.Furthermore, as argued by Lancy (2010) and Shweder (2010), the general applicability of Piaget's model has been thoroughly undermined by cross-cultural research.Mead (1932) demonstrated that Piaget's developmental model (in its 1929 form) fails to generalise given her studies of an Admiralty Islands small-scale society, and Luria (1976) described alternative patterns of reasoning in Uzbek peasants in Central Asia.Wynn himself notes that a desiderata of a mid-range theory is that it is cross-culturally confirmed (Wynn 1985, p. 33), yet Piaget's theory fails this test.And if the theory cannot generalise from middle-class European children to children from the Admiralty Islands, then we should be skeptical of inferences made via the theory regarding the makers of the tools found at Ismilia.Moreover, we should be even more skeptical of any generalisations from those individual tool producers to other contemporaneous hominins, which Piaget's theory-and the scope of his conception of ontogeny as a singular, invariant process-appears to allow (cf.Wynn 1979, p. 383ff).It's no surprise that in later work Wynn discontinued this line of reasoning, eschewing the commitment to Piaget.It is important to note that this does not mean that the individuals who produced the Ismilia tools did not have the spatial constructs Wynn identifies; just that Piaget's cognitive model isn't the way to licence that inference.Wynn himself recognised this, and in later work he turned to comparative and palaeoneurological data, as well as a broad range of cognitive theories and frameworks via his subsequent teaming with psychologist Frederick Coolidge.
Wynn's early work is lauded by many cognitive archaeologists (e.g., Mithen 1996;Stade and Gamble 2019;McGrew et al. 2019;Davidson 2019), even though its hardline Piagetian justification would have few-if any-advocates today, as it helped to pave the way for a more mature discipline (and see Wynn 2016, pp. 8-10, for retroactive reflection on his Piagetian days).The case demonstrates that the problem of sample diversity in psychology was present at the beginnings of the cognitive archaeological project, and that its negative effects were acknowledged.The next subsection details a much more promising inference to the long-term working memory capacities of Neanderthals (Wynn and Coolidge 2004) by way of contemporary psychological and cognitive anthropological models.

Case study 2: Neanderthals and Long-term Working Memory
Recall that a minimum-capacity inference takes an artefact or tool industry-say, the Levallois technocomplex, associated with Neanderthals-and, via a cognitive theory, reaches a claim about the production of the trace-say, that the producers of those prepared-core technologies had the capacity for advanced (essentially, modern or very near-modern) long-term working memory.This is the claim defended by Wynn and Coolidge (2004). 10evallois reduction is a task comprising multiple steps: "The first prepares a core with two distinct but related surfaces, one, a more convex platform surface that will include the striking platform, and a second flatter production surface from which the blank or blanks will be removed.The second step prepares the striking platform itself in relation to the axis of the intended blank.The third step is the removal, by hard hammer, of the blank or blanks."(Wynn and Coolidge 2004, pp. 473-474; see Figures 3 and 4 for a schematic representation and a photograph).When one blank is prepared, the method is called preferential and when two or more are prepared it is called recurrent.The recurrent method contains unidirectional, bidirectional and centripedal variants, each requiring different knapping techniques and platform preparation.Consequently, the 'end products' (e.g.Levallois points) are the result of not a single action-sequence but a plurality of operational schemas/methods (Boëda 1995).
Now, what of the mid-range theory through which the cognitive claim above is epistemically licenced?In this case, Wynn and Coolidge appeal to two cognitive models, one from cognitive neuropsychology (Ericsson and Kintsch 1995;Ericsson and Delaney 1999;Ericsson et al. 2000) and the other from cognitive anthropology (Keller and Keller 1996).Here we will focus on the latter.
Keller and Keller use the practice of smithing as their primary case study.According to their model, there are three crucial aspects to skilled expertise.First, 'the stock of knowledge', which is the information pool a smith acquires, builds, and maintains over many years of experience.This would include semantic/symbolic information, but also sensory information: visual, sonic, and tactile 'images' of procedures, materials, and so on.Second, an 'umbrella plan'; a mental model or representation of the end product intended as well as the tasks and subtasks required for its production.Again, this would include both semantic/symbolic information and visual, sonic, and tactile information.Umbrella plans are similar to the concept of retrieval structures employed in cognitive neuropsychology: a given task recalls the relevant retrieval structure from long term memory, containing cues that facilitate encoding and retrieving associatively linked knowledge.Third, 'constellations', which are the requisite ideas and mental images, as well as the materials and tools, that enable each step of the process to be successfully accomplished or begun anew.Depending on the tasks involved, constellations can be deployed more or less automatically, or with full conscious attention.Active feedback between the smith and the constellation shapes the smith's actions and decisions in deploying the constellation.
Wynn and Coolidge claim that, in light of this model, Levallois reduction (and perhaps other Neanderthal technologies) demonstrates that the producers had longterm working memory capacities.They argue that, like smithing, prepared-core technology involves successfully completing a complex sequence of tasks requiring a stock of knowledge, an umbrella plan, and constellation, and that these concepts can be "directly applied" (p.474).They say: "The sequence of actions that can be reconstructed for Levallois reduction resembles the sequence of action documented by the Kellers for blacksmithing: a sequential task with definable steps during which the artisan makes choices among a variety of specific techniques and procedures in order to complete each step and, ultimately, produce a finished product" (Wynn and Coolidge 2004, p. 474).
As is further outlined by Pain (2019), Wynn and Coolidge provide good reason to think the models' concepts map nicely onto aspects of the target phenomenon (the Levallois), so the conceptual framework is at least plausible. 11Nonetheless, smithing may well be more culturally variable than the Kellers and Wynn and Coolidge implicitly assume, potentially restricting the scope of the model.Smithing, of course, is not an exclusively WEIRD activity.So cross-cultural research stands to strengthen or undermine the plausibility of the model's general application.As such, the cognitive capacities involved in the production of Levallois technologies may be distinct from that of modern smithing, undermining the analogy.This raises an important question: are all cases of artisan expertise organised in broadly similar ways, cognitively speaking?The concepts might appear to 'fit', but that fitting might yet overstate the long-term working memory of those ancient knappers (some interpretations from ecological psychology, and more broadly anti-representational approaches, would be consistent with this line of reasoning).So, even if the Keller's 11 What's less plausible is Wynn and Coolidge's subsequent species-wide generalisation from the individual producers of the Levallois to Neanderthals broadly, e.g., "We believe that Neandertals regularly employed L-TWM [long-term working memory] that was as effective as any documented for modern humans" (Wynn and Coolidge 2004, p. 473).It is important to distinguish between a past projection and a species-wide generalisation from that projection.For even if some Neanderthals demonstrated a capacity (a producer of some tool) that does not imply that the capacity is distributed universally through the population/species.model is applicable to skilled expertise in particular domains, it not only might fail to generalise, but also might not be suitable for projection to the deep past (after all, the Levallois dates as far back as roughly 300,000 years ago, and the last common ancestor of Neanderthals and H. sapiens lived roughly half a million years ago).Wynn and Coolidge admittedly are aware of this problem: "But are the cognitive underpinnings the same, that is, did Neandertal stone knappers have stocks of knowledge, make umbrella plans, and use constellations?"(p.474).In their efforts to epistemically licence their inference, Wynn and Coolidge turn to modern skilled knappers, and they cite the work of Boëda (1995) and Baumler (1995).But these papers only describe the techniques involved in Levallois reduction, not cognitive underpinnings.Of course, it is relevant to take into account tool morphology, assemblage analysis and the constraints of knapping processes. 12But testing the cognitive model, ideally utilising a diverse sample of participants, is also required.
Of course, all too often cognitive models based on culturally diverse samples are simply not available-precisely because of the bias towards WEIRD samples in the cognitive and behavioural sciences.Furthermore, for the most part, it is not the role of cognitive archaeologists to develop such models.Cognitive archaeologists are typically consumers, not producers of such models.But clearly, all things being equal, models licenced by data from culturally diverse samples should be preferred over those that are not.And where such models are not yet available, this could be acknowledged.The same point can be made about anthropological data: it is likely to be anthropologists, not cognitive archaeologists, who would provide rich descriptions, for example, of African indigenous smithing.Nonetheless, where available, such data stands to strengthen or undermine the epistemic licence for inferences such as Wynn and Coolidge's above. 13his example, like the Piagetian one preceding it, demonstrates that, in relying on models from the cognitive sciences, minimum-capacity inferences are at risk of inheriting the problem of cultural variation and sample diversity.However, perhaps unlike the Piagetian one, here we have an example of an argument that looks prima facie plausible (Pain 2019) before considering that challenge.In sum: further testing of the cognitive model using culturally diverse samples stands to strengthen or undermine the projection to the past.

Case Study 3: Affordances
The previous two examples targeted specific inferences in cognitive archaeology, focusing on the work of Wynn.In the following two case studies, we broaden our scope to look at the challenge of cross-cultural variation and sample diversity to research programs more generally.
The dominant theoretical paradigm in contemporary cognitive science is representationalism/computationalism.Over the last 30 or so years, this dominance has come under increasing pressure from advocates of so-called 4E (embodied, enactive, extended and embedded) approaches.However, the situation in cognitive archaeology is somewhat different.This is partly due to a "founder effect" (Mayr 1942).Two archaeologist/psychologist duos were particularly influential in the development of cognitive archaeology: Coolidge and Wynn (as previously discussed), and Noble and Davidson (see e.g.Noble and Davidson 1996).And while the former duo work primarily in a representational framework, Noble and Davidson appeal to Gibsonian ecological psychology (see Davidson 2019 for a retrospective on the debate between these two).So frameworks that reject internalist, representation-focused accounts of cognition have been, and continue to be, a mainstay of cognitive archaeology (e.g., Malafouris 2020; Overmann 2017; see papers in DeMarrais et al. 2004;indeed, even Wynn (2020) has begun to engage with affordance theory).
So far our discussion has focused on Wynn's early-and mid-career work, which means it has been confined to representational/computational approaches to cognitive archaeology.In this section we want to broaden it to 4E approaches.We will focus on affordance theory, as this was Noble and Davidson's chosen framework.We will not target specific cognitive archaeological inferences here.14Rather, we want to track some work in the development of affordance theory, and show that the concept in general is likely to target phenomena which are culturally influenced.Consequently, care must be taken when inferring from WEIRD samples to ancient populations when appealing to the framework.
Affordances are understood as products of interactions between an organism and its environment.As different organisms have different body types, the same environment will produce different affordances for different organisms.For instance, to use an oft-quoted example: the wall in my office does not offer me the affordance of walking (I cannot walk on walls). 15This is not the case for the spider that lives in the corner of my office (spiders can walk on walls).Walls (and ceilings) offer spiders the affordance of walking.Gibson (1979) argued that organisms perceive affordances directly, and the theory has subsequently become popular with those attempting to challenge the representational paradigm (e.g.Chemero 2009).Initial experimental work with human subjects focused on biomechanical models, where affordances were treated as a ratio between some aspect of a subject's body scale and some feature of the environment; for instance, between leg height and the height of a stair, or between eye height and the width of a gap (see e.g.Mark 1987;Burton 1992Burton , 1994;;Mark et al. 1999).A range of experiments appeared to generate reliable results.Most famously, Warren found a consistent ratio between leg height and stair height with regards to the boundary between stairs deemed "climbable" and stairs deemed "unclimbable" (Warren 1984).Similar effects have been found in the case of eye height and the width of a gap deemed "crossable" (Jiang and Mark 1994).This work appears to lend weight to the body-scale account of affordances, and a body-scale account is one that would plausibly generalise across human populations.
However, as Chemero (2009, p. 143) notes, the property of organisms we are really interested in when it comes to affordances is their ability to perform some action.Aspects of body scale, such as leg height, are just a proxy for our ability to climb stairs that is easily quantifiable.This idea was tested by Cesari et al. (2003).Warren's original stair climbing experiment used US college students as subjects, so the participants were mainly younger adults.By contrast, Cesari et al. used subjects with a range of different ages, including elderly people.They found that the ratio between leg height and stair height when it comes to the transition between climbable and unclimbable stairs varied considerably according to age.For elderly people, this ratio was considerably lower than the rest of the participants.Rather, flexibility seemed to also be playing an important role.Similar results have been produced in the case of gap crossing tasks (Chemero 2009, p. 145).Other work has contrasted subjects' judgments of their ability to cross a gap and gap length with leg length and gap length (Chemero 2009).The former ratio was found to be much more highly correlated with the maximum gap judged crossable than the latter.Taken together, these results suggest that it is ability, not body scale, which is relevant in the perception of affordances.
All this is important for our purposes when we take into consideration two points.First, body-scale is a biological category, whereas ability is influenced by a range of biological and cultural factors.Stair climbing ability is a product of properties like leg height and flexibility, but it is also a learnt skill, and hence, to some extent, a product of culture.Furthermore, stair climbing ability will be to some extent determined by the prevalence of stairs in the environment-again, a cultural property.So populations that do not inhabit built environments will have different stair climbing abilities to those that do.The bottom line is that, if affordances are abilities, then they are influenced by culture.The situation here is roughly analogous to the case of the Müller-Lyer illusion (see section 2).There a 'base-level' perceptual effect was shown to be a product of architectural environments.Something similar is true of the trajectory of experimental work in affordances: an effect once thought to be more strictly confined to the biological domain is now understood as, to some extent, culturally determined.Second, the cited research above was carried out, not merely on WEIRD subjects, but, for the most part, on US college students. 16And US colleges tend to be stair-rich environments.17Together, these two points suggest that any kind of generalised claim about affordances (beyond, perhaps, their existence) will need to be based on cross-cultural samples.And this is no different in the case of cognitive archaeology's use of affordance theory.

Case Study 4: Experimental Neuroarchaeology
Recently, cognitive archaeologists have developed a new strategy for producing inferences to the past.As we have seen, traditional strategies-such as minimumcapacity inferences-rely on interpreting the archaeological record using a model from the cognitive sciences.In contrast, experimental neuroarchaeology takes modern human subjects and investigates their brain activity during knapping tasks using neuroimaging techniques.This technique inherits a different set of theoretical and methodological problems.First and foremost among these is that modern human subjects are not, for instance, H. erectus; so any argument produced by this strategy is one by homology.This means that the strength of the inference involved depends on the neural similarity between modern H. sapiens and (again, for example) H. erectus, and the value of that ratio is very difficult to ascertain.
We will put these concerns aside here.Instead, we want to focus on issues of sample diversity in this research.In many ways, neuroarchaeology suffers similar problems to those we have identified in more traditional cognitive archaeology.But there is an upside here: the experimental aspect of neuroarchaeology allows us to more directly test cultural impacts on cognitive and behavioural capacities.In turn, this further illustrates the importance of sample diversity.
The majority of work in neuroarchaeology has so far focused on tool-language co-evolutionary hypotheses.In particular, research has tried to identify if there is any neural overlap between toolmaking and language production (e.g.Stout et al. 2008;Putt 2019).For instance, in their 2008 study Stout and colleagues took three expert tool knappers and used fluorodeoxyglucose positron emission tomography to assess the areas of the brain co-opted by both Oldowan and late Acheulean tool production tasks (Stout et al. 2008).They found that, compared with Oldowan toolmaking, late Acheulean toolmaking produced increased brain activity in areas of the brain associated with language production (in particular, Broca's area).More recently, Shelby Putt and colleagues have expanded the scope of this work using functional near-infrared spectroscopy (Putt et al. 2017; see Putt 2019 for an overview).Putt and colleagues were particularly interested in investigating whether the mode of learning-either verbal or non-verbal-had any effect on the brain regions co-opted by Oldowan and late Acheulean tasks.Participants in the experiment were taught to knap using either spoken language or via visual aids alone.Their results indicate that the mode of learning has a significant impact on parts of the brain coopted during Acheulean-style toolmaking.This suggests that modern day knappers may be rehearsing the verbal instructions they were exposed to during learning.
Importantly though, the participants in these experiments were all WEIRD, and indeed either university professors or college students from the United States. 18nd this is worrying, as the need for sample diversity does not stop at Henrich et al.'s level of explanation (behavioural sciences) but is also a priority for neuroscience (Chiao and Cheon 2010).According to Chiao (2009), 90% of peer-reviewed neuroimaging research comes from Western countries.Yet even between Westerners and East Asians there appear to be neuroscientific-level instances of cultural variation: "Westerners engage brain regions associated with object processing to a greater extent relative to East Asians, who are less likely to focus on objects within a complex visual scene (Gutchess et al. 2006).Westerners show differences in medial prefrontal activity when thinking about themselves relative to close others, but East Asians do not (Zhu et al. 2007).Activations in frontal and parietal regions associated with attentional control show greater response when Westerners and East Asians are engaged in culturally preferred judgments (Hedden et al. 2008).Even evolutionarily ancient limbic regions, such as the human amygdala, respond preferentially to fearful faces of one's own cultural group (Chiao et al. 2008, [...]).Taken together, these findings show cultural differences in brain functioning across a wide variety of psychological domains and demonstrate the importance of comparing, rather than generalizing, between Westerners and East Asians at a neural level."(Chiao and Cheon 2010, p. 89).
Moreover, there is a dearth of neuroimaging work utilising individuals from small-scale populations.While fMRI machines are not easily transportable, EEG methods (for instance) are non-invasive and the technology far more mobile.Chiao and Cheon's call for more effort to investigate neuroscientific research questions via sampling more diverse populations thus looks well justified. 19Meanwhile, in the case of neuroarchaeology, it is important that inferences licenced via culturally localised samples are treated with caution.
Indeed, even the trajectory of research from Stout to Putt illustrates this.Stout purported to show neural overlap between toolmaking and language.Putt observed that Stout's experiments did not control for the method of learning of the participants.In testing the difference between visually taught participants and verbally taught participants Putt demonstrated that a cultural force-learning-could influence the results of neuroscientific testing.This trajectory serves as a cautionary tale, yet also illustrates an important opportunity.As an experimental research program, neuroarchaeology can actually test cultural variables, even when operating with limited sample diversity-Putt's work shows that the way in which a skill is learnt affects the neural substrates it co-opts.This is not an option for traditional cognitive archaeology, which operates by interpreting the archaeological record through a cognitive model.

How Worried Should We Be?
In the previous section, we outlined examples in which cognitive archaeologists use models, frameworks, or experimental research from the cognitive sciences which have been built using limited samples or are otherwise susceptible (or potentially susceptible) to the challenge from cultural variation.In this section, we look in more detail at how worrying this situation is.We begin with some conceptual work clarifying the issue of generalisation and sample diversity in the context of inference to the deep past.Then, drawing on Henrich et al. (2010), we look at some conditions under which generalisations from limited sample sizes might be justified.

Sample Diversity, Generalisation, and Inference to the Past
Recall the issue of sample diversity as it faces the cognitive sciences.The problem is that, by generalising from WEIRD samples, researchers assume that either there is no cross-cultural variation, or that WEIRD people are generally representative of the species.However, cross-cultural research has shown that in many cases these assumptions do not hold.So sample diversity is required to make reliable claims at the larger population-or species-level.Now, take the case in which we use a cognitive model derived from WEIRD samples alone, and then apply that model to some artefact from the stone tool record using a minimum-capacity inference.What precisely is the worry here?In the first instance, notice that this does not necessarily undermine the inference to the cognitive capacity of the maker of the artefact.If the minimum-capacity inference holdsand that is a big 'if'-then we should be confident that the maker of the artefact at least had the relevant capacity.Remember, the lesson from Henrich et al. ( 2010) is generalise with care; at this stage in the process, no generalisation has been made.Though perhaps unlikely given the differences in lifeways between WEIRD people and past peoples, it is at least logically possible that we are successfully inferring from one culturally localised sample to another.A distinct problem arises, however, when we generalise from that minimum-capacity inference to a population-wide or species-wide claim in the deep past.That generalisation would be justified if we had reason to believe that there was no cultural variation in the population (which is perhaps only plausible for pre-Homo populations) or that WEIRD people were generally representative of the species and diachronically so, but of course this is not the case.If, however, the cognitive model at hand has been corroborated by cross-cultural research, then, all other things being equal, researchers have greater epistemic licence to extend the inference to that of the population-or species-level.
To illustrate this point, consider Wynn and Coolidge's minimum-capacity inference concerning the long-term working memory abilities of Neanderthals (section 3.2).The inference here used Keller and Keller's cognitive anthropological model of skilled expertise, based on their observations of smithing.In the absence of cross-cultural testing of that model, the minimum-capacity inference may well licence claims regarding the long-term working memory capacities of those individuals, and perhaps their wider cultural group, found in situ with Levallois technology (that is, if we are satisfied that the past projection via that cognitive model is provisionally licenced-we have discussed above some challenges for this).
In other words, we may be satisfied that the inference takes us from the technology, via the cognitive model, to the cognitive capacity of the individuals who produced the technology.But even then we would not be justified in making species-wide claims regarding Neanderthals-that Neanderthals in general had modern or nearmodern long-term working memory-as Wynn and Coolidge do (and then go on to draw additional inferences based on that generalisation, see Wynn and Coolidge 2004, pp. 478ff).On the other hand, if Keller and Keller's model was corroborated by cross-cultural research, all other things being equal, the inference would more reliably allow species-wide generalisation.At the very least, such inferences would gain credibility.The lesson is this: sample diversity allows us to make more plausible generalisations in the present, and this is important insofar as it provides greater epistemic licence for generalisation in the past.

Are
There Cases Where Generalisations From Weird Samples are 'Safe'?
We have thus far argued that lack of sample diversity is inferentially problematic for cognitive archaeology, insofar as it utilises cognitive models licenced by data on WEIRD subjects, and outlined a range of cases-both specific and more general.In this section, we want to examine some cases where traditional minimum-capacity inferences might be more plausible.Henrich et al. (2010) suggest a range of conditions under which generalising from WEIRD samples to contemporary humans might be provisionally justified: 1. Effects found in "[...] cognitive domains related to attention, memory and percep-tion…".2. Effects "... measured at a physiological or genetic level".3. "... generalisations from one well-studied universal phenomenon to another similar phenomenon".4. Effects demonstrated "...in other species, such as rats or pigeons…".5. Effects "...which are evident among infants".6.Effects in brain regions "...less responsive to experience".(Henrich et al. 2010, p. 79) We think this list can be reduced to two conditions: [a] where an effect is reliably thought to be a product of biology; and [b] where an effect is reliably thought to be a product of a cultural universal.2 is a way of reaching the conclusion in [a], while 4, 5 and 6 are methods of eliminating culture as a potential cause of an effect, thereby increasing the likelihood that it is biological.3 is one way of reaching the conclusion in [b].Finally, the intuition driving 1 looks to be related to [a], but Henrich et al. note that the work of their paper "...does not bolster this intuition" (Henrich et al. 2010, p. 79).Simply put, there are two reasons why we might think a cognitive effect might be universal-it is either part of human biology or it is, for whatever reason, common to all human cultures.Are there areas of research to which cognitive archaeology contributes that satisfy these conditions?We consider cases from research on language and theory of mind, but begin with two general observations.First, a distinct possibility is that satisfying [a] in the present may be insufficient for the production of reliable inferences to our deep past.This will occur as the morphology of hominin body shapes changes through phylogenetic space.Thus effects that generalise across modern day human populations due to their being measured at a physiological level will not apply as we move further back in time.
Second, it might be thought that long periods of cultural uniformity, such as the Acheulean, signal that either [a] or [b] has been satisfied. 20This would only apply to those who commit to cognitive explanations of uniformity-those who posit demographic or environmental causes might not consider either to be satisfied (see Pain 2019 for an overview).
As a result of Chomsky's influence, the view that our ability to produce language is innate and domain specific (e.g.Berwick and Chomsky 2015;Chomsky 2007) is widespread.If we thought this view was true, then we might think that [a] would be satisfied.The evolution of language was a key focus of early cognitive archaeological studies (e.g., Gibson and Ingold 1993;Mithen 1996;Noble and Davidson 1996), so perhaps much of this work is immune from sample diversity concerns.However, there is increasing acknowledgement amongst researchers regarding the causal importance of culture in the evolution of language.This includes gene-culture coevolutionary accounts (e.g., Laland 2017;Tomasello 2005Tomasello , 2010)), and more radical accounts that deny biologically produced language-specific capacities in human ontogeny (e.g., Christiansen and Chater 2016;Heyes 2018).If these theories are on the right track for language, [a] would either not be satisfied at all, or look much more difficult to satisfy.In addition, recent tool-language co-evolutionary hypotheses look to develop accounts where the evolution of syntactical features of language involves co-opting existing capacities evolved to support toolmaking (e.g., Stout et al. 2008;Stout and Chaminade 2012;Planer and Sterelny 2021).Syntactical capacities are sometimes thought to be one of-if not the-biologically-endowed, domain-specific language capacity.These accounts, however, suggest that the phylogenetic ancestry of that capacity was heavily driven by culture.This again raises questions about the ability to satisfy [a].
'Theory of mind' or 'mindreading' refers to our ability to infer, understand, or simulate the mental states of another individual.For instance, one of us might infer from the yawns of an audience that participants in the lecture are bored.Our theory of mind capacities are embedded in a broader framework of orders of intentionality.These orders begin with the awareness of our own mental states, and progress from there.For instance: a lecturer intends (1); that their audience understand (2); that Stout believes (3); that Chomsky disagrees (4); with Tomasello's commitment to domain-general processes in theorising about language (5); and so on.Theory of mind is thus located in the second order of intentionality and beyond.Recently, Cole (e.g., Cole 2016; Cole 2019) has produced a range of studies attempting to correlate theory of mind capacities and orders of intentionality with the lithic record (see also Planer 2017).Now, one might think that the ability to interpret other people's mental states is a culturally produced capacity (Heyes 2018) and is perhaps also distinctive of human beings.Theorists of this persuasion may well think that theory of mind is something approaching a cultural universal, and hence that it satisfies [b].However, Cole (2019) has argued that the record suggests more variation in orders of intentionality capacities than universal models indicate, which would undermine this conclusion.
We have run, very briefly, through two cases where one might have reason to think that [a] or [b] is satisfied-however, recent work tells against this conclusion in both cases.Of necessity we have given these research programs short shrift; a full analysis is beyond the scope of this article and is an avenue for future research.Henrich et al. (2010) identified an important problem for the behavioural and cognitive sciences: the tension between the empirical reality of cultural variation and the narrow cultural representativeness of most study samples.Cultural variation in many domains engenders cognitive variation, and WEIRD participants are often not representative of the human species.We have argued that cognitive archaeology inherits this problem insofar as it uses models (and frameworks, experimental research, etc.) from the cognitive sciences to provide epistemic licence for its inferences to the past.Our examples demonstrate the breadth of the challenge to cognitive archaeology: it is not restricted to any specialisation or theoretical paradigm, and has historical roots.

Conclusion and Future Directions
We have claimed that rather than being methodologically flawed, inferences to cognitive capacity from a physical trace via a cognitive model stand to be strengthened or called into question by further testing of the cognitive model against more culturally diverse samples.Corroboration is possible, so the situation is not totally dire; there is cause for optimism (Currie 2018).We have also suggested that there may be some cases where the force of the problem is mitigated-hypothetically, some minimum-capacity inferences from particular cognitive models might be more or less plausible in light of independent reason to think an inference can be applied outside the sample, e.g., when an effect is reliably thought to be the product of biology, or when an effect is reliably thought to be the product of a cultural universal.That said, our brief analysis suggests that these conditions look difficult to satisfy.A full analysis is an important avenue of future research. 21inally, as we have stressed throughout, theoretical resources from the cognitive sciences are an important part of the cognitive archaeologist's inferential toolkit.Yet cognitive science is in a state of live debate.This poses a problem for cognitive archaeologists: which principles should guide cognitive archaeologists when selecting a cognitive model for use as a mid-range theory?We (Killin and Pain 2021) have recently proposed two complementary, though non-exhaustive solutions: theory choice should be guided by consilience (convergence from multiple independent lines of evidence) and methodological pluralism.This article proposes a further, compatible principle.That is, all other things being equal, cognitive archaeologists should prefer models tested and corroborated by diverse samples; or, at the very least, where effort has been undertaken to assess how well the results (very often from WEIRD samples) might project to the deep past.Since cultural effects produce cognitive and behavioural variations even in contemporary populations, due caution must be taken when projecting inferences to ancient hominins, and it would be no bad thing to heed this caution in model selection.After all, the above considerations suggest that past projections require a developmental cognitive mid-range theory, one that takes into account the effects of culture on cognition.In turn, this implies that reconstructing ancient minds requires, to some degree at least, reconstructing their cultures. 22This challenge too is suggestive of future research.

Fig. 3
Fig. 3 Stages in preparedcore tool production.Source: Ambrose (2001), reproduced here with the permission of the American Association for the Advancement of Science