10

Culture as Cognitive Technology:

An Evolutionary Perspective
Stephen C. Levinson

1 Some Reservations About Current

Cultural Models
Cultural model theory is perhaps the most prominent trend in contem-
porary cognitive anthropology. It comes in two main flavors: a qualita-
tive, discourse-based version (CM) and a quantitative version often
denoted ‘cultural consensus analysis’ (CCA). Both versions attempt to get
at propositions that underlie cultural opinions and motivate culture-­
typical actions, and they tend to emphasize the largely out-of-awareness
moral injunctions that guide cultural behavior. The CCA variety comes
with sophisticated methods for assessing the degree of consensus across a
social group and the individual’s degree of conformity to it.
There are a number of reasons why many anthropologists may have
reservations about both varieties of the theory. The theory is an outgrowth

S. C. Levinson (*)
Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 241
G. Bennardo et al. (eds.), Cognition In and Out of the Mind, Culture, Mind, and
Society, https://doi.org/10.1007/978-3-031-48181-9_10
242 S. C. Levinson

of earlier steps toward a cognitive anthropology, notably Goodenough’s

definition of culture as an entirely mental phenomenon:

A society’s culture consists of whatever it is one has to know or believe in

order to operate in a manner acceptable to its members.… Culture is not a
material phenomenon; it does not consist of things, people, behavior, or
emotions. It is rather an organization of these things. It is the forms of
things that people have in mind, their models for perceiving, relating, and
otherwise interpreting them. (Goodenough 1957, p. 167)

Goodenough’s definition is highly reminiscent of Chomsky’s defini-

tion of “competence” as the grammatical knowledge of an idealized
speaker-hearer and has the same potential flaws: a characterization of, say,
“American culture” is bound to represent the cultural knowledge as
belonging to some kind of averaged or idealized individual, or at best of
some prototypical group. A related problem is that a cultural model is
often conceived of as some coherent schema or set of consistent proposi-
tions, whereas in many cases we hold inconsistent or inchoate beliefs. So,
we are presented with a cultural model of American marriage, or a foun-
dational cultural model of Tongan society or the like, where these varia-
tional issues are often backgrounded.
A ‘cultural model’ also often seems to presuppose a homogeneity that
is certainly alien to complex societies. Take a society like Southern India.
Even in a small village with say 20 castes, each caste may have a different
kinship system, different rites and different professions and may well
speak different languages and profess different religions. In such a society
it makes no sense to argue, for example, a cultural model of marriage: one
caste will have a preference for marrying father’s sister’s daughter with
dowry and asymmetry between wife-givers and takers, while another will
have a preference for mother’s brother’s daughter without asymmetry or
dowry (and yet others a preference for sister’s daughter marriage; Beck
1972; Levinson 1982). In addition, each caste will have a default profes-
sion (weaving, metallurgy, priesthood, temple cooking, etc.) and thus its
own expertise. There is, in effect, an elaborate, indeed exquisite, division
of epistemic labor and little cultural consensus (although what there is, is
highly significant). Quantitative versions of CCA escape this idealized
 10 Culture as Cognitive Technology: An Evolutionary Perspective 243

homogeneity, but the summary statistics are not in the head of anybody
except the analyst of course, so they are not in that sense mental models.
Caste societies may be one end of the spectrum, but even the simplest
societies with minimal division of labor do not show high degrees of cul-
tural consensus. For example, Rossel Island is a tiny island with perhaps
6000 people, but it supports two major dialects—in a survey across the
dialects, I found corresponding words shared only 20% identical forms,
and another 20% belonged to different cognate classes (i.e., were unre-
lated; Levinson 2022a). Material culture like sago-processing gear also
shows ‘dialects’ across the island. Testing the application of terms to pri-
mary percepts (color, shape, taste, smell, hearing, etc.) again yields sig-
nificant variation both within the culture and in contrast to other cultures
(Majid et al. 2018). Indeed, greater homogeneity is likely to be found in
literate, state-based societies, considering these results on judgments of
perception across cultures.
Another issue is that most cultural models tend to focus on declarative
knowledge, rather than the procedural knowledge of the kind that
Goodenough seemed to have in mind when he defined culture as what
one has to know in order to behave in culturally appropriate ways. But
procedural knowledge may not be organized in terms of propositions as
opposed to recipes for action, often encoded as much in habitual motor
programs as in explicit propositions. Moreover, procedural knowledge
has the backgrounded, out-of-awareness character that cultural model
theory claims to be of special interest. It is worth noting that early exer-
cises in cognitive anthropology stressed the procedural in a way that cur-
rent efforts mostly do not (but see Frake 1964; Gatewood, Chap. 6, this
volume). Consequently, most current efforts in cultural model theory
propose a schema that seems too static and too propositional to capture
the fact that it is cognition that drives cultural action that we are actually
interested in.
A final issue is the point of departure for this chapter. Following
Goodenough, ‘cultural model’ theory posits that culture is an entirely
mental thing—it lies wholly between the two ears (Bennardo, Chap. 1,
this volume). This is a peculiar position. The criterial property of culture
is that it is shared to some large degree, on the basis of social transmission
via some kind of externalization. That entails that culture does not lie
244 S. C. Levinson

entirely between my or your ears. Anything that is only in the head can-
not therefore be cultural. The point of course was made by Durkheim
(1895): sociology or anthropology cannot be reduced to psychology on
the grounds that social facts have an existence independent of any indi-
vidual. He was pointing, correctly in my opinion, to the peculiar onto-
logical status of social facts. Here I am going to pursue the Durkheimian
insight and argue for a different direction for cognitive anthropology.
But first, let me concede two points. First, there is a wide variety of
opinion among cultural model enthusiasts, and the strictures above may
miss the mark here and there. Second, and more importantly, there are
genuine insights and very useful methodologies that have been developed
within cultural model theory. We do utilize models, schemas, frames and
the like to navigate an uncertain world. And I too have enjoyed the exe-
getical triumphs of finding “semplates” (semantic templates) hidden in
vocabularies (Levinson and Burenhult 2009) or finding ergative pattern-
ing at the heart of an exotic language (Levinson 2022a). But these are
only a part—and possibly a minor part—of the cognitive anthropology
we should be promoting. In addition, an idealized mental competence is
an emphasis that threatens to cut us off from the rest of anthropology—it
doesn’t connect easily to variationist studies, to studies of the organiza-
tion of discourse and behavior, to social interaction or to the material side
of human life that is the preoccupation of archeology. So, let me sketch
an alternative framework, which I would argue should be at the very least
a central part of the picture.
I start from the position that culture is such an improbable biological
outcome that it needs an evolutionary perspective to explain its emer-
gence.1 We could begin by asking: What is the function of culture? Given
the many overheads (including the cognitive effort, the costs of behaving
altruistically and giving away good tricks), what are the payoffs? The cen-
tral function, I would suggest, is for sharing things that are useful (i.e.,
tricks for doing things), and the payoffs are that individuals get more out
than they put in. What they get out I suggest is a whole new form of

1
In Levinson (2022b) I suggest that the source of the human propensity to share lies in an early
evolved use of alloparenting, requiring some kind of matching of empathy and communication.
 10 Culture as Cognitive Technology: An Evolutionary Perspective 245

cognition, which has a curious ontological status and is not wholly in

the head.
A way of thinking about the shared tricks that constitute culture is to
think of culture as composed of multitudes of tools, some concrete and
some abstract. The concrete tools are mostly devices that amplify the
body: hammers, knives, levers, pulleys—even horses, jet planes and jack-
hammers are essentially bodily prostheses. The kind of exoskeleton that
deep sea divers use is a kind of prototype to bear in mind. In contrast to
these bodily prostheses, the more abstract tools are devices that amplify
the mind: they are what I will call cognitive artifacts. They may be con-
crete things like a slide rule, an abacus or a computer, but their job is not
to amplify the body, but to extend our mental powers. The rest of this
chapter outlines these mind-expanding tools and tries to show how our
cultural worlds are pervaded and sustained by them and indeed how they
constitute the backbone of culture.

2 Culture as Cognitive Technology2

2.1 The Idea of a Cognitive Artifact

Nearly half a century ago, Jack Goody (1977) introduced the idea of a
“technology of the intellect.” His key example was the invention of writ-
ing. When writing is first introduced in a society, he noted, it is used for
very simple mundane uses, of which making lists is the most prominent.
Such lists have two rather different uses: they can be forward-looking
to-do lists (like shopping lists or recipes), and they can be backward-
looking lists (as in records of tribute received). The first has a directive or
computational character, while the second performs a mnemonic func-
tion. In both cases, he noted, in externalizing thought, and freezing the

2
An extended version of this argument can be found in Levinson (2020). In turn, this draws
on the work of Ed Hutchins (1995), Jean Lave (1988), Suchman (2007), Andy Clark (Clark
and Chalmers 1998), ethnoarcheologists (e.g. Malafouris 2013; Overmann and Malafouris 2018),
human-machine interaction (Norman 1993) and ultimately Vygotsky (1986)—movements that
have gone under the rubrics of activity theory, distributed cognition, extended mind, Situated
cognition, material engagement and so forth.
246 S. C. Levinson

evanescent linguistic signal, they make new modes of cognition available:

you can inspect the signal, reorder it, count the elements and edit it. It
enables meta-cognition. He held that these cognitive effects were revolu-
tionary and that writing therefore “changes the type of data an individual
is dealing with, and it changes the repertoire of programs he has available
for treating this data. Whether or not it changes the hardware…is another
matter, but on the analogy of language the possibility is there” (1977,
pp. 109-10).
That speculation turns out to be correct: literacy radically rewires the
brain, colonizing the fusiform gyrus at the expense of our face recogni-
tion and strengthening the white-matter connections between the hemi-
spheres and elsewhere (Dehaene 2009; Carreiras et al. 2009). It is a
revolutionary piece of cultural technology. Goody also pointed to the
extraordinary sociocultural effects of literacy, which allows communica-
tions over space and time, with consequences for the very distinct histo-
ries of the literate civilizations of Eurasia and the non-literate civilizations
of sub-Saharan Africa. Literacy after all allows the cumulative use of data
and records which lie behind the growth of science and technology, the
bureaucracies of vast empires, the rise of mass communications and,
indeed, many aspects of the modern condition.3
Goody concentrated solely on literacy, but I believe the idea of tech-
nologies of the intellect has in fact far wider applicability (see Levinson
2020). A ‘cognitive artifact’ is, in my favored sense to be developed, an
external aid to internal computation (see also Norman 1991). There’s no
shortage of obvious examples, from the abacus to the slide rule, from the
theodolite to the compass, from the map to the diagram or from the
electronic calculator to the computer. But there are much simpler exam-
ples, and they are all around us.4

3
Giovanni Bennardo (this volume) holds that “the locus of knowledge is the human mind”
(Bennardo 2022, p. 1), but for modern literate societies this is not where the bulk of information
resides.
4
Although they are sometimes so simple that they are not always easy to recognize. Consider the
transparent window on a bottle of engine oil that allows you to see how much is left, or the depth
gauge on a river crossing, or the LED that lets you know your device is plugged in, or any symbol
like a road sign warning of rock falls.
 10 Culture as Cognitive Technology: An Evolutionary Perspective 247

Consider this simple example: a Tzeltal man, who is typically a subsis-

tence farmer, carries with him a wooden staff a meter long. As well as a
walking stick, a weapon of self-defense against dogs on a trail and some-
times a rod of office, it is basically a dibber: when planting corn he pokes
it three times into a little heap of soil and plants three corn kernels. Then
he uses the full length of the stick to measure the place for the next little
heap and so along the row; then he goes up one stick’s length and plants
the parallel row. If he plants 20 heaps per row, and there are 20 rows, he
knows he can expect 1200 (400 times 3) corn cobs (unless the variety
yields two cobs per plant, in which case he can double 400). So, he can
calculate the likely harvest and whether to plant more. If he hires some-
one to do it, he will pay by the number of rows. If he is thinking of selling
the field, or buying it, he knows how much of the family’s needs he will
lose or gain. That simple stick gives him a system of precise estimation.
Every system of weights and measures is a simple piece of cognitive
technology, and most of them had their origin in just such calculations:
an acre, for example, was the area that medieval peasants could plow in
one day with a team of oxen. The enormous value of a simple measure of
length was brought home to me watching Rossel Islanders making houses
and canoes from bush materials. Rossel Islanders, traditionally at least,
use no measures. For example, looking for a ridge pole they trot off into
the jungle and find the longest straight timber of the requisite diameter
they can find. They carry these very heavy 300+ kg timbers back down
the mountain, haul them up on top of the house structure and chop
them off to fit the emerging shape. If they had a tape measure, they could
save themselves a lot of sweat and a lot of wasted materials. The measure
would be a way to carry with them a precise estimation of what is needed.
Rossel Island has no market economy and no pressing use for weights and
measures, unlike traditional Mayan communities like those that speak
Tzeltal.
A cognitive artifact then is an aid to solving a mental problem by
means of an external instrument, which returns some value which can be
re-internalized. Figure 10.1 represents this export of a mental problem
out of the head via a query to an external device, which then returns a
value which can be re-imported into mental computations. In the
248 S. C. Levinson

Fig. 10.1 Cognitive artifacts as coupled systems bridging mind and matter

parlance of the philosopher Andy Clark (2011),5 this is a ‘coupled sys-

tem,’ a circuit that depends on both inner and outer components. It is
because the outer instrument is externalized that it can be shared and
indeed is subject to the processes of cultural evolution, which typically
hones the instrument into an ever more efficient form. Thinking of cul-
ture as composed largely of such coupled systems (and I will enlarge the
scope below) gives us a satisfying way of cashing out the Durkheimian
ontological insight that ‘social facts’ are ‘things’ that cannot be reduced to
psychology.
Let us turn to obvious examples of cognitive artifacts to get further
clarity. There is no dearth of examples in the history of technology. Aids
to spatial navigation—where the problem is to solve location and direc-
tion—have been much studied and include Micronesian star maps,
medieval portolan charts, Roman strip maps, compass roses, astrolabes,
sextants and the modern GPS and radar systems. Traditional Polynesian
navigation relied on master navigators memorizing star rising and setting
points and internalizing rules of thumb about wave interference, oro-
graphic cloud and the predominant settings of winds and currents
(Gladwin 1970; Feinberg 1988; Lewis 1972). The Western tradition
externalized these processes, measuring speed by a chip log, star azimuths
with the astrolabe, direction by compass and so on, using these measure-
ments to plot location on a chart (Hutchins 1995), so answering
“Where am I?”

5
I should however distance myself from the metaphysical claims of Clark and others (Clark and
Chalmers 1998; Clark 2011) that mind can then literally spill into the environment and is no
longer confined to the head; such ontological claims are not germane to the present argument.
 10 Culture as Cognitive Technology: An Evolutionary Perspective 249

Another much-studied domain is the measurement of time, both diur-

nal and calendrical. Simple systems include noting the location of celes-
tial bodies and noting the fruiting or migration times of natural species.
Thus, Rossel Islanders arrange to meet on the morrow by pointing to
where the sun will be, while Tenejapan Mayans may arrange to meet
before dawn by pointing to where the moon will be, which requires more
careful observation—here, gestures are acting as externalized aids to com-
putation (see also Floyd 2016). On Rossel Island, the seasons are esti-
mated by changes in prevailing wind directions and the coastal running
of fish, like sardines and mullets, triggering events in the agricultural cal-
endar. So, the noting of external events can answer the problem of when
to plant seed taros.
From such humble beginnings, artifacts and machines for temporal
reckoning slowly developed: for diurnal reckoning, water clocks, sundials
and clockwork clocks which counted divisions of the day for purposes as
diverse as shared irrigation (Iran), allocating time for plaintiff and defense
in a court of law (Rome), the opening of markets (Greece and Rome) or
times of worship (Aveni 1989). For calendrical reckoning, complex celes-
tial measurements were undertaken using instruments as massive as
Stonehenge or the Caracol at Chichen Itzá. Huge investments in astro-
nomical observation were made to bring the lunar and solar cycles into
alignment with the solar year.
A final obvious example is geometry and mathematics. Innate number
concepts are restricted to subitizing (recognizing between one and five
objects at a glance) and estimating masses: it is counting that seems to
bridge across the systems, giving us precise quantities (Dehaene 1996).
Peoples who lack number words like the Pirahã are famously unable to
match precise quantities (Gordon 2004; Frank et al. 2008). First steps
involve matching external counters with the things to be enumerated,
and the body provides a convenient set of digits for this purpose (Bender
and Beller 2012). Other early methods of reckoning involve tallies—the
20,000-year-old Ishango bone may even have acted as an early slide rule
(Ifrah 1998; Marshak 1972). Rossel Islanders keep track of thousands of
shell ‘coins’ by collecting the strings on which they were strung in tens,
counting the strings and then multiplying in their decimal number sys-
tem. Devices like the abacus, going back to Sumerian times, speed up
250 S. C. Levinson

mental computations for seasoned number users (Wang 2020), while

today most of us use calculator apps. The actual form of the externaliza-
tion of number concepts crucially matters, as history shows: the Roman
non-positional system hindered calculations, while the Arabic system
favored them. More exactly, as Chrisomalis (2020) explains, although
Roman number representations could easily be supplemented with calcu-
lations on the abacus, there was no easy way to show the workings of the
calculations in the numerical notation—hence double-entry bookkeep-
ing was instrumental in ushering in the Arabic notation. Geometry is
another domain where externalization makes available mental calcula-
tions that are otherwise not achievable—the Babylonians already had
approximations to pi, ways of calculating the areas of various figures, and
even calculus (Ossendrijver 2016)
These examples are sufficient I think to illustrate the idea of a cognitive
artifact. Let us now firm up the concept. Figure 10.2 provides a charac-
terization, using the simple example of a measurement system (like a
modern tape measure) to estimate how big a door I need to fill the door
frame. As spelled out there, there is a mental problem to be solved, repre-
sented here as a function with an unknown value, and the use of an
external device—the cognitive artifact—yields the value in a format eas-
ily returned to the mind. The simple example makes clear that we are not

A computational cognitive artefact has the following properties:

1.) There is a recurrent cognitive problem, finding the value of f(x) = ?

e.g., How big is the gap? (measuring for a door)
2.) The artefact is externalized, in a publicly accessible medium
e.g., tape measure
3.) The instantiated artefact is shared in type, and honed by cultural evolution
e.g., standardized units on a compact retractable metal coil
4.) There’s a procedure for operating on the instantiation
e.g., holding the tape against the gap
5.) The process is economical: the cognitive advantage outweighs the costs of
externalization
6.) The output of the process must be re-internalizable (e.g., memorable)
if f(x) = y, y must be easily assimilated
e.g., the number of centimeters

Fig. 10.2 Characterization of a computational cognitive artifact

 10 Culture as Cognitive Technology: An Evolutionary Perspective 251

talking about artificial intelligence here—the artifact itself can be as sim-

ple as the Tenejapan measuring stick.
This is, however, only one species of cognitive artifact, namely one that
aids active computation. There are others. Jack Goody (1977) drew atten-
tion in the case of literacy to its two functions: a list could be an inven-
tory of what I need to do, or it could be a record of what I have already
done. In the latter case, the writing serves the purpose of recall—perhaps
by an actor other than the writer. All of our historical, bureaucratic and
scientific records are of this type of course, and they enable that amazing
accumulation of knowledge that characterizes advanced literate societies.
This is another type of cognitive artifact, which, rather than enabling
computation, enables the retrieval of an earlier solution.
Curiously, it seems that the very origin of Sumerian writing (and thus
ours) lies in mnemonic clues to the contents carved in the top of clay
envelopes which contained tokens (representing, e.g., a debt of five sheep;
Schmandt-Besserat 1996). After a while, the tokens became superfluous,
and the mnemonics took over. Once again, though, we can generalize
away from literacy and look at other ways in which simpler retrieval sys-
tems work. Consider the knotted handkerchief, to remind me to contact
Jill in the morning. It works because encountering the knotted handker-
chief in my pocket in the morning, I’m reminded of the intention to call
Jill, which caused me to knot it. Or consider the use of notches in a bone,
used to keep a tally of the days passed in this location, or the successful
kills in the hunt—such tallies seem to go back to the Upper Paleolithic
(Bahn 2016, p. 324ff).
Simple mnemonics may be private, as with the handkerchief, or they
may be public like road signs reminding one not to speed. On Rossel
Island, people erect taboo signs to signal and remind people that a coco-
nut plantation is no longer free for foraging, while many kinds of bound-
ary markers in other societies serve a similar purpose (tim trees in the case
of Tenejapa, planted at the boundaries of fields). Non-literate retrieval
systems can be vastly more complex than this. Consider the ancient
Peruvian quipu—knotted cords that served (among other things) to keep
track of tribute paid in the Inca empire, using a positional system that
kept track of vast numbers (Urton 2003)—a system still used a century
ago by herders keeping tabs on their flocks.
252 S. C. Levinson

Let us call this a mnemonic cognitive artifact and give the following
characterization, as in Fig. 10.3.

2.2 How Cognitive Artifacts Work—

Thought Transduction

It is not entirely self-evident why cognitive artifacts have the efficacy they
have. Why does translating a mental problem into an external medium
seem to automatically aid computation? One principal reason seems to
be that the problem now has a double representation—one in an internal
mental medium and the other in an external medium. Each kind of rep-
resentation may offer different kinds of cognitive affordances. It has been
noted, for example, that children faced with Piagetian problems gesture
the correct answer before they learn to expound it verbally (Goldin-­
Meadow 2015), and if one gives route directions, one will find that one’s
gestures precede the verbal instructions as one literally ‘feels out’ the
solution.
Another feature is that the transfer from inside to outside involves
some kind of transduction, something other than a direct 1:1 mapping (in
engineering, a transducer is a device that converts energy from one form

A mnemonic cognitive artefact has the following properties:

1. A cognitive artefact for recall has the function of encoding the thought A
at time t, in such a way that it can be retrieved at a later time t+n
e.g., How much did I pay for the ticket?
2. There must be some external marker of A, call it ά, such that encountering
ά brings A to mind
e.g., a printed receipt
3. To be recoverable by random others, there must be a shared convention
that ά stands for A
e.g., the receipt has a standard form
4. The procedure for encoding ά is effective and simple enough to be
economical
5. The encoded thought A must be recoverable, useful and easily assimilated
at time t+n

Fig. 10.3 Characterization of a mnemonic cognitive artifact

 10 Culture as Cognitive Technology: An Evolutionary Perspective 253

into another—e.g., sound waves into electrical current and back into
sound waves). An abacus used by a speaker of a decimal system re-­
represents numbers in terms of units of fives and ones. Linear speech is
converted into two-dimensional orthography stripped of many expres-
sive features. A guess at the size of a timber is converted into a precise set
of conventional units. Transduction typically involves the following
transformations:

(a) Transduction often involves a mapping into a higher or lower dimen-

sional space—1D to 2D (musical notation, writing, written addition
of a string of orthographic numbers), 1D to 3D (sundials, astrolabes),
3D to 2D (maps, blueprints, geometry of solids), 3D to 1D (oral
recitations of, e.g., Aboriginal songlines, or a reduction of an itiner-
ary to a list);
(b) The transduction mostly involves a mapping into a more concrete
medium—e.g., abstract number into the tactile medium of the aba-
cus, or abstract direction into gesture—where it can literally be
manipulated;
(c) At the same time, the external representation simplifies and strips
away incidental distracting properties: consider a map, which only
represents a few chosen features of the landscape or a musical score
that abstracts away from variable phrasing and performance;
(d) The transduction may involve a mapping from a weaker sense into a
more dominant one, e.g., internal representations of sound into
vision (writing, musical notation), abstract reasoning into visual-­
spatial representations (graphs, Venn diagrams, numbers) or abstract
order or number into tactile manipulations (prayer beads, abacus);
(e) It may involve the transduction of a fleeting or changing signal into
a static one where time is frozen, as in writing or musical notation, or
a seismograph;
(f ) The external device may physically constrain the solution space, so
reducing errors, as with an abacus, or a pilot’s checklist with tabs to
be flipped one by one;
(g) Crucially, by externalizing a computation, parts of it can be put in an
external memory buffer, so overcoming the highly limited buffer
capacities of our working memory, as explained further below. The
254 S. C. Levinson

importance of this is evident when doing long division, remembering

a string of digits or going through a checklist in preparation for
a journey.

These transductions of thought into different types of representation

seem to capture Goody’s (1977) insight that the early uses of literacy
involve the exploitation of metacognitive rumination, which is less avail-
able for flashes of thought in the mind or the ephemeral signals of speech.
In addition, they offer some account of what Goody, following Bruner
(1966), called the amplifying function of externalization. From the point
of view of the user, the cognitive artifact may substitute an entirely differ-
ent and easier task for the target one, as when logarithms are added to
multiply large numbers, a procedure partially automated in a slide rule
(Norman 1991).
Another prominent property of cognitive artifacts is that they involve
both externalization and re-internalization, often repeated recursively, as
when writing and redrafting or doing complex calculations. Such repeated
re-ingestion will inevitably remake the internal representation into a for-
mat closer to the external representation. The oral performance of a liter-
ate person is not like the oral performance of an illiterate praise-singer (as
Goody 1987 notes), nor is the mental arithmetic of a mathematician like
the mental arithmetic of a peasant. Skilled and practiced users of these
repetitive transductions may end up not actually needing the externaliza-
tion, as with skilled abacus users who can use a mental representation of
the 3D thing without resorting to the external aid (Frank and Barner
2012; Barner et al. 2016). In the same way, while I may need notes to
keep my lecture on track, skilled orators have since Cicero (Yates 1966)
used internal systems of mnemonics based on imagined externalization
(e.g., by traversing an image of a familiar room). What the example of
literacy shows is that this recursive re-ingestion induces the mutual adap-
tation of brain to external device and external device to brain, in what
Dehaene and Cohen ( 2007) have called a “cultural recycling of cortical
maps.” It is this mutual adaptation that characterizes a ‘coupled system’
between mind and device.
It is possible to invent private, secret cognitive artifacts, as when one
uses external clues to remember software passwords—this is prominent
 10 Culture as Cognitive Technology: An Evolutionary Perspective 255

especially in mnemonic cognitive artifacts. But a huge part of the efficacy

of cognitive artifacts is that they are mostly shared, public representa-
tions. As a consequence, they have been honed, often by eons of cultural
evolution (as with our alphabet or number systems) to maximize the
metacognitive affordances of the external representation (e.g., making
addition or multiplication easy) and to minimize the effort of transduc-
tion when externalizing and re-internalizing the representations.
Sometimes there will be an arbitrary quality to the external representa-
tion—e.g., it could be in inches or centimeters—drawing attention to
the fact that part of what makes it valuable is the very fact that it is a
shared, standardized system even when the format is sub-optimal (like
the QWERTY keyboard). Figure 10.4 offers a summary of the argument.

2.3 Language as a Cognitive Artifact

So far, the proponents of cultural models may retort, “Sure, those are
interesting examples, but they hardly add up to that massive knowledge
base that we call a culture.” Yes, but we have not finished yet. Any theory

Summary: How Cognitive Artefacts work

• Double representations in different formats give extra ‘handles’ for

computations or retrieval
• Transductions into a different dimensional media afford different operations,
allowing metacognitive rumination, inspection and the ‘freezing’ of temporal
succession
• Stronger senses can be used to reinforce weaker ones, as with the visual
representation of abstract properties
• Externalization offers mid-computation ‘memory buffers’, thus overcoming
the severe limitations of working memory
• Externalized representations can be honed by cultural evolution to maximize
the metacognitive handles and minimize the efforts of transduction
• External representations tend to simplify, selecting a few dominant features
of the underlying thought
• Recursive externalization and re-ingestion bring inner and outer formats
closer in alignment, easing the difficulties of transduction

Fig. 10.4 The efficacy of cognitive artifacts

256 S. C. Levinson

of culture has to come to grips with language. One of the most striking
things about the human species is that it is the only one on the planet
which has a communication system that varies so fundamentally across
social groups at every level of structure, from the sounds to the syntax,
from the syntax to the semantics (Evans and Levinson 2009). Language
is the foundation for cultural diversity, one of the prime modes of its
transmission across generations.
Language of course is also a cognitive artifact, if an enormously com-
plex one. It transduces a thought into sound waves. It solves the problem
of making overt an intended meaning. It has an external form or repre-
sentation which is shared—we learned our languages in an open forum.
A language offers some of the same computational advantages we have
already noted of dual representations—when I express a thought, I
streamline it, simplify and clarify it to myself. Thus, I can re-internalize
my now clarified thought. Writing a paper like this, the thought goes in
and out a few times before stabilizing on the page. Now, just like the tape
measure or the calculator, I have inherited the technology which has been
honed by cultural evolution to serve a community of speakers. What I
have inherited, among other glorious things, is a vast repository of ready-­
made concepts packed into words.
Most of these words encode notions that would take a lot of round-
about description to convey by other means (ogee, logarithm, architrave,
sonata, algorithm, pi, shaman, etc.). At the outset of cognitive science,
George Miller (1956) pointed out that these perform a crucial computa-
tional function. We can only hold, he suggested, 7 ± 2 items at once in
working memory (we now know the mode is much less than this, more
like four; Cowan 2001). So, the only way we can compute complex
things is by packing complex concepts into memorable chunks, as we do
in maths when we let x stand for the output of another operation. Think
about words as pointers to complex concepts in virtually unlimited long-­
term memory—but unless they themselves form a chunk (as in a sen-
tence), you can only hold four or five such random pointers in short-term
memory at a time. A culture, then, provides you with a huge stock of
words and expressions, a bonanza of ‘zipped’ thoughts, little nuggets of
intensive meaning. These allow us to get complex multi-stranded thoughts
through the bottleneck of working memory. And although linguists tend
 10 Culture as Cognitive Technology: An Evolutionary Perspective 257

to think of the lexicon as an unstructured repository, it comes with imma-

nent structure, as in the taxonomies, partonomies, semplates and seman-
tic fields which are familiar objects of anthropological enquiry.
In addition, every language provides us with its own recursive struc-
ture for building complex propositions. Studies have shown that those
who have been deprived of full-blown language in early childhood have
difficulty conceptualizing complex embedded propositions (Pyers and
Senghas 2009). Besides the major constructions of the central syntax of a
language, languages have additional mini-grammars like those found in
their numeral systems or their kinship systems, which again can recur-
sively construct specialized thoughts of arbitrary complexity.
The transduction of thought into speech is an amazing process involv-
ing upward of a second and a half of intensive mental processing and the
deployment of over a hundred muscles, and in conversation it often has
to work in parallel to processing the incoming turn at talk (e.g., Levinson
2016). We ease the process by learning to regiment our thoughts into a
form that fits the categories of the particular language (Slobin 1996). It is
because of this that our mental categories come to match language-­
specific grammatical and lexical categories, yielding many features of ‘lin-
guistic relativity’ (see Levinson 2012). The socially shaped patterns of
usage come to dominate the way we think: in one culture we think in
terms of left versus right, while in another in terms of north versus south;
in one culture we think of relatives in terms of a Crow system, while in
another according to an Omaha system of reckoning; in one culture we
think in terms of blue versus green, while in another we think in terms of
grue; and so on.
In sum, languages have all the hallmarks of cognitive artifacts. They
transduce thoughts into a socially shared medium in order to solve myr-
iad problems; they are tools for persuading, cajoling, encouraging, order-
ing, comforting and exchanging information with our social others. We
re-imbibe thoughts clothed in language, benefiting from the transduc-
tion into a streamlined and culturally shaped medium.
258 S. C. Levinson

2.4 Generalizing the Idea: Other Kinds

of Cognitive Artifact

I have outlined two main types of cognitive artifact, the computational

and the mnemonic. These two types are central exemplars, but there are
almost certainly other kinds of cognitive artifact. At the risk of over-­
widening the category, consider the following. Cultures use various
means of inducing mood changes, that is, altering the emotional stance
of participants. These may include mood-changing substances like alco-
hol or kava or emotion-inducing cultural performances like music and
dance. One might object that these changes are mechanical, natural cau-
sations, but it is noticeable that they do not necessarily work across cul-
tures: kava and Chinese opera leave me cold, while alcohol often turns
my Pacific friends into zombies rather than socialites. Ritual flagellation
may induce trance in a Hindu devotee, but I doubt it would work on me.
Nor will magic mushrooms automatically make a shaman out of you.
Within a culture, these mood-changing artifacts work in the requisite
direction because there are already expectations of the state that should
be reached. Presuming that emotional states are part of our cognition
(and not in a separate compartment of mind), then devices for changing
those states might also be candidate cognitive artifacts (see Fig. 10.5 for a
rough characterization).

There is a goal state, an emotional tone T, to be obtained

There is an artefact or performance that is publicly recognized to induce this tone T
The artefact or performance is honed by cultural evolution to induce this tone T
There’s a specific procedure for using this artefact or performance
There’s a good probability that undertaking the procedure will induce T as desired
Acquiring and internalizing T may be a precondition to playing an active part in
the performance that will induce T in others

Fig. 10.5 Characterization of an emotion-inducing cognitive artifact or

performance
 10 Culture as Cognitive Technology: An Evolutionary Perspective 259

Music and dance might be the most powerful and most common
devices of this sort, but more extreme events like human sacrifice or
sacred mortal combat (as in the Mayan ball game) might be others,
inducing mixtures of fear and wonder. These kinds of human activities
are hard to explain in terms other than their intended emotion-inducing
functions.
Representational art may also have this function—after all, it is histori-
cally deeply associated with ritual, religion and multi-modal perfor-
mances. Consider too the architecture intended to make those who enter
feel small and insignificant or wonder at the crystallization of massive
labor and expense—the emotions one feels on the thresholds of castles,
cathedrals, palaces or parliaments. Clearly, huge amounts of effort and
resources are expended in mind-bending external devices, performances
and installations.
A more straightforward extension of the notion of cognitive artifact is
to perception. The history of science is populated by devices that extend
the range or acuity of human perception. If telescopes and microscopes
allow us to see things somewhat beyond our normal vision, radar, infra-
red sensors and X-rays make visible the invisible. The current direction
with wearable technology is likely to make cyborgs of us all (Clark 2003).
Perceptual prostheses, however, may not be technologically complex at
all—as Bateson (1972, p. 359) famously pointed out, the blind man’s
stick makes an extended cognitive system. Barking guard dogs or the
Capitoline geese also served as extra eyes and ears.
A completely different line of reasoning may lead in the direction of
seeing social teams as computational devices. The argument here has
been well rehearsed by Ed Hutchins (1995) who showed how, for exam-
ple, a team of sailors on the bridge of a ship may navigate the behemoth
via a division of labor in which each sailor reports readings from different
devices (sonar, radar, line of sight) to a central navigator plotting the
course. The ‘distributed cognition’ movement may itself have lost steam,
but the insights are perfectly valid: we solve a lot of problems by out-
sourcing them to a team. In the simplest case, I can ask you when we
went to Rome together, using your recall device to solve my failed one.
In more complex cases, a team of scientists or developers can jointly
solve long-standing problems. Whole bureaucracies can operate merely
260 S. C. Levinson

to answer the questions posed by the state. In complex societies there are
elaborate divisions of epistemic expertise and labor, and we employ
accountants, lawyers, surveyors, web designers and translators. The mod-
ern trend of course is to try and shrink these teams by substituting artifi-
cial intelligence, merely the latest step in a long line of experiments with
cognitive artifacts. The insights of the ‘distributed cognition’ movement
very nicely tie cognitive anthropology into the study of social interaction
and social organization (see Enfield and Kockelman 2017).

2.5 The Payoffs of Viewing Cultures as Built out

of Cognitive Artifacts

I started this chapter by complaining that cultural models are often envis-
aged in a way that presumes homogeneity, requires idealization, overem-
phasizes the internal at the expense of the external shared nature of
cognition and focuses too much on static templates rather than recipes
for computation and performance. In contrast, the framework of cogni-
tive artifacts can easily encompass variation (not everyone needs to use
the same tools) and therefore has no need to idealize away from variation,
imperfect performance and real time process. Above all, cognitive arti-
facts escape the head: they are in the environment and can be empirically
studied with comparative ease. Cognitive artifacts are tools in active use,
not data structures, and the theory can much more easily encompass
rapid technological and mental change.
It follows that human cognition has a curious ontology (Clark and
Chalmers 1998; Overmann 2017). Is it my recollection or my digital
diary’s that I have an appointment on April 1, 2023? Is it my belief or my
calculator’s that the square root of 3 is 1.73205080757? Is it my estima-
tion or that of the bubble in my spirit level that this door jamb is level? Is
it my estimation or the map’s that it’s 64 miles from Chicago to DeKalb?
Clearly, human minds form coupled systems with their cognitive tools—
we have partially outsourced our cognition and in doing so hugely ampli-
fied it, as Goody (1977, p. 109ff) was fond of pointing out. Now, that’s a
fit and proper subject for cognitive anthropology.
 10 Culture as Cognitive Technology: An Evolutionary Perspective 261

The signal advantage, it seems to me, of the cognitive artifacts perspec-

tive is, firstly, it connects directly to external culture—to artifacts, to their
use, to social interaction, to social transmission and by way of the cogni-
tive division of labor to the organization of society. In this way it does not
leave a cognitive anthropology marooned in the past, but one that can
interrelate with mainstream trends not only in sociocultural anthropol-
ogy, but in archeology, history of science, even art history and musicol-
ogy, while maintaining strong links with the cognitive sciences. In
addition, it offers some kind of account of how culture enhances cogni-
tion and thus an account of how culture endowed humans with an evo-
lutionary advantage. It suggests that culture is not some outcome of
oversized brains, but rather of the way in which cognition interacted with
the social environment in a feedback relation over hundreds of thousands
of years resulting in the gradual development of this expensive organ.
The complaint will still be, I imagine, that this kind of a focus does not
encompass all the things that a cognitive anthropology should be con-
cerned with. But the limits of this kind of framework have hardly been
explored—I have just sketched some of the further developments that
may be expected. Another complaint is likely to be that the whole point
of cultural models is that they offer an integration of many different ideas
and practices into some kind of coherent whole, while the present account
is atomistic, yielding a collection of small cognitive tricks—but that I
think is to underplay the potential. Consider how a solution to one prob-
lem may give the key to another: for example, how describing an array on
a table (variously in terms of left/right or north/south) ties into another
(describing how to get from X to Y) and thus into way-finding in a land-
scape (Levinson 2003). Distinct styles of thought come to characterize a
series of cognitive artifacts. Tools after all come in sets for solving a col-
lection of related problems, and there may be just as much of a complex
network covering a large domain as there appears to be in the alternative
framework of cultural models.
To conclude, I have offered here an alternative vision for the future of
cognitive anthropology. It is cobbled together from the previous efforts of
many scholars, not only in cultural anthropology (in the situated and
distributed cognition traditions especially), but in archeology, history of
science, philosophy, psychology, linguistics and other sciences. I view the
262 S. C. Levinson

mongrel nature of the ideas as a positive virtue, because if cognitive

anthropology is to have a future, it must connect more directly to a wide
spectrum of the surrounding sciences. How we evolved this deep cou-
pling of our cognition with our shared culturally shaped environment is
a topic of the first importance for a range of disciplines.6

References
Aveni, Anthony. 1989. Empires of Time. New York: Basic Books.
Bahn, Paul G. 2016. Images of the Ice Age. Oxford: Oxford University Press.
Barner, David, George Alvarez, Jessica Sullivan, Neon Brooks, Mahesh
Srinivasan, and Michael Frank. 2016. Learning Mathematics in a Visuospatial
Format: a Randomized, Controlled Trial of Mental Abacus Instruction. Child
Development 87 (4): 985–1311. https://doi.org/10.1111/cdev.12515.
Bateson, Gregory. 1972. Steps to an Ecology of Mind. New York: Ballantine.
Beck, Brenda. 1972. Peasant Society in Konku. Vancouver: University of British
Columbia Press.
Bender, Andrea, and Sieghard Beller. 2012. Nature and Culture of Finger
Counting: Diversity and Representational Effects of an Embodied Cognitive
Tool. Cognition 124: 156–182.
Bennardo, Giovanni. (2022). Cognitive Foundations and Cultural Models: The
Mental Life of Culture. Wenner-Gren Conference [Position Paper].
Bruner, Jerome S. 1966. Studies in Cognitive Growth. New York: John Wiley.
Carreiras, Manuel, Mohamed L. Seghier, Silvia Baquero, Adelina Estévez,
Alfonso Lozano, Joseph T. Devlin, and Cathy J. Price. 2009. An Anatomical
Signature for Literacy. Nature 461 (7266): 983–986. https://doi.org/10.1038/
nature08461.
Chrisomalis, Stephen. 2020. Reckonings: Numerals, Cognition, and History.
Cambridge, MA: MIT Press.
Clark, Andy. 2003. Natural-Born Cyborgs. Oxford: Oxford University Press.
———. 2011. Supersizing the Mind. Oxford: Oxford University Press.
Clark, Andy, and David Chalmers. 1998. The Extended Mind. Analysis
58: 10–23.

6
Niche construction theory offers a pan-species approach to feedback relations between con-
structed environment and organism (Odling-Smee et al. 2013); the full human exploitation is
going to require special explanation (one factor sketched in Levinson 2022b).
 10 Culture as Cognitive Technology: An Evolutionary Perspective 263

Cowan, Nelson. 2001. The Magical Number 4 in Short-term Memory: A

Reconsideration of Mental Storage Capacity. Behavioral and Brain
Sciences 24: 87.
Dehaene, Stanislas. 1996. The Number Sense. Oxford: Oxford University Press.
———. 2009. Reading in the Brain. New York: Viking.
Dehaene, Stanislas, and Lauren Cohen. 2007. Cultural Recycling of Cortical
Maps. Neuron 56 (2): 384–398.
Durkheim, Emile. 1895. Les Règles de la Méthode Sociologique. Paris: Felix Alcan.
Enfield, Nick J., and Paul Kockelman. 2017. Distributed Agency. Oxford: Oxford
University Press.
Evans, Nicholas, and Stephen C. Levinson. 2009. The Myth of Language
Universals. Behavioral and Brain Sciences 32 (5): 429–492.
Feinberg, Richard. 1988. Polynesian Seafaring and Navigation. Kent, OH: Kent
State University Press.
Floyd, Simeon. 2016. Celestial Pointing for Time-of-Day Reference in
Nheengatú. Language 92 (1): 31–64.
Frake, Charles O. 1964. How to Ask for a Drink in Subanun. American
Anthropologist 66: 127–132. https://doi.org/10.1525/aa.1964.66.
suppl_3.02a00080.
Frank, Michael C., and David Barner. 2012. Representing Exact Number
Visually Using Mental Abacus. Journal of Experimental Psychology: General
141 (1): 134–149. https://doi.org/10.1037/a0024427.
Frank, Michael, Daniel L. Everett, Evelina Fedorenko, and Edward Gibson.
2008. Number as a Cognitive Technology: Evidence from Pirahã Language
and Cognition. Cognition 108 (3): 819–824.
Gladwin, Thomas. 1970. East is a Big Bird. Cambridge, MA: Harvard
University Press.
Goldin-Meadow, Susan. 2015. From Action to Abstraction: Gesture as a
Mechanism of Change. Developmental Review 38: 167–184.
Goodenough, Ward H. 1957. Cultural Anthropology and Linguistics. In
Georgetown University Monograph Series on Languages and Linguistics, vol. 9,
167–173. Washington, D.C.: Georgetown University Press.
Goody, Jack. 1977. The Domestication of the Savage Mind. Cambridge:
Cambridge University Press.
———. 1987. The Interface Between the Written and the Oral. Cambridge:
Cambridge University Press.
Gordon, Peter. 2004. Numerical Cognition Without Words: Evidence from
Amazonia. Science 306: 496–499.
264 S. C. Levinson

Hutchins, Edwin. 1995. Cognition in the Wild. Cambridge, MA: MIT Press.
Ifrah, Georges. 1998. The Universal History of Numbers. London: Harvill.
Lave, Jean. 1988. Cognition in Practice. Cambridge: Cambridge University Press.
Levinson, Stephen C. 1982. Caste Rank and Verbal Interaction in Western
Tamilnadu. In Caste Ideology and Interaction, ed. Dennis B. McGilvray,
98–203. Cambridge University Press.
———. 2003. Space in Language and Cognition. Cambridge: Cambridge
University Press.
———. 2012. Foreword. In Language, Thought, and Reality: Selected Writings of
Benjamin Lee Whorf, ed. J.B. Carroll, S.C. Levinson, and P. Lee, 2nd ed.,
vii–xxiii. Cambridge, MA: MIT Press.
———. 2016. Turn-taking in Human Communication—Origins and
Implications for Language Processing. Trends in Cognitive Sciences 20
(1): 6–14.
———. 2020. On Technologies of the Intellect: Goody Lecture 2020. Halle: Max
Planck Institute for Social Anthropology.
———. 2022a. A Grammar of Yélî Dnye. Berlin: de Gruyter. https://www.
degruyter.com/document/doi/10.1515/9783110733853/html.
———. 2022b. The Interaction Engine: Cuteness Selection and the Evolution
of the Interactional Base for Language. Philosophical Transactions of the Royal
Society of London, Series B: Biological Sciences. https://doi.org/10.1098/
rstb.2021.0108.
Levinson, Stephen C., and Niclas Burenhult. 2009. Semplates: A New Concept
in Lexical Semantics? Language 85: 153–174.
Lewis, David. 1972. We the Navigators. Canberra: ANU Press.
Majid, Asifa, Seán G. Roberts, Ludy Cilissen, Karen Emmorey, Brenda
Nicodemus, Lucinda O’Grady, Bencie Woll, Barbara LeLan, Hilário de
Sousa, Brian L. Cansler, Shakila Shayan, Connie de Vos, N.J. Gunter Senft,
Rogayah A. Enfield, Razak Sebastian Fedden, Sylvia Tufvesson, Mark
Dingemanse, Ozge Ozturk, Penelope Brown, Clair Hill, Olivier Le Guen,
Vincent Hirtzel, Rik van Gijn, Mark A. Sicoli, and Stephen C. Levinson.
2018. Differential Coding of Perception in the World’s Languages. PNAS
115 (45): 11369–11376.
Malafouris, Lambros. 2013. How Things Shape the Mind: a Theory of Material
Engagement. Cambridge, MA: MIT Press.
Marshak, Alexander. 1972. The Roots of Civilization. New York: McGraw Hill.
 10 Culture as Cognitive Technology: An Evolutionary Perspective 265

Miller, George A. 1956. The Magical Number Seven, Plus or Minus Two: Some
Limits on Our Capacity for Processing Information. Psychological Review 63
(2): 81–97.
Norman, Don A. 1991. Cognitive artifacts. In Designing Interaction: Psychology
at the human-Computer Interface, ed. J.M. Carroll, 17–38. Cambridge:
Cambridge University Press.
———. 1993. Things That Make Us Smart. New York: Basic Books.
Odling-Smee, John, Douglas H. Erwin, Eric P. Palkovacs, Marcus W. Feldman,
and Kevin N. Laland. 2013. Niche Construction Theory: A Practical Guide
for Ecologists. Quarterly Review of Biology 88 (1): 4–28.
Ossendrijver, Mathieu. 2016. Ancient Babylonian Astronomers Calculated
Jupiter’s Position from the Area Under a Time-Velocity Graph. Science 351
(6272): 482–484.
Overmann, Karenleigh A. 2017. Thinking Materially: Cognition as Extended
and Enacted. Journal of Cognition and Culture 17 (3–4): 354–373.
Overmann, Karenleigh, and Lambros Malafouris. 2018. Situated Cognition. In
International Encyclopedia of Anthropology: Anthropology Beyond Text, ed.
H. Callan, 1–8. New York: Wiley-Blackwell.
Pyers, Jennie E., and Ann Senghas. 2009. Language Promotes False-Belief
Understanding: Evidence from Learners of a New Sign Language. Psychological
Science 20 (7): 805–812. https://doi.org/10.1111/j.1467-­9280.2009.
02377.x.
Schmandt-Besserat, Denise. 1996. How Writing Came About. Austin: University
of Texas Press.
Slobin, Dan I. 1996. From “Thought and Language” to “Thinking for Speaking”.
In Rethinking Linguistic Relativity, ed. J.J. Gumperz and S.C. Levinson,
70–96. Cambridge: Cambridge University Press.
Suchman, Lucy A. 2007. Human-Machine Reconfigurations: Plans and Situated
Actions. Cambridge: Cambridge University Press.
Urton, Gary. 2003. Signs of the Inka Khipu: Binary Coding in the Andean Knotted-­
String Records. Austin: University of Texas Press.
Vygotsky, Lev S. 1986. Thought and Language. Cambridge, MA: MIT Press.
Wang, Chunjie. 2020. A Review of the Effects of Abacus Training on Cognitive
Functions and Neural Systems in Humans. Frontiers in Neuroscience 14
(913): 1–12. https://doi.org/10.3389/fnins.2020.00913.
Yates, Frances. 1966. The Art of Memory. London: Routledge.