RSTB_363_1509.

qxp 9/30/08 5:06 PM Page 1

Phil. Trans. R. Soc. B | vol. 363 no. 1509 pp. 3467–3603 | 12 Nov 2008
ISSN 0962-8436

volume 363

12 November 2008 number 1509

volume 363 . number 1509 . pages 3467–3603
pages 3467–3603
Cultural transmission and the evolution
of human behaviour In this issue

Papers of a Theme Issue compiled and edited by Kenny Smith, Michael L. Kalish, Thomas L. Griffiths Cultural transmission and the evolution
and Stephan Lewandowsky
of human behaviour
Introduction. Cultural transmission and the evolution of human behaviour 3469
K. Smith, M. L. Kalish, T. L. Griffiths & S. Lewandowsky
Papers of a Theme Issue compiled and edited by Kenny Smith, Michael L. Kalish, Thomas L. Griffiths
Review. Establishing an experimental science of culture: animal social diffusion experiments 3477 and Stephan Lewandowsky
A. Whiten & A. Mesoudi
Review. The multiple roles of cultural transmission experiments in understanding human
cultural evolution 3489
A. Mesoudi & A. Whiten
Review. Theoretical and empirical evidence for the impact of inductive biases on cultural evolution 3503
T. L. Griffiths, M. L. Kalish & S. Lewandowsky
Beyond existence and aiming outside the laboratory: estimating frequency-dependent and
pay-off-biased social learning strategies 3515
R. McElreath, A. V. Bell, C. Efferson, M. Lubell, P. J. Richerson & T. Waring
Review. Studying cumulative cultural evolution in the laboratory 3529
C. A. Caldwell & A. E. Millen

Cultural transmission and the evolution of human behaviour

Investigating children as cultural magnets: do young children transmit redundant information
along diffusion chains? 3541
E. Flynn
The fitness and functionality of culturally evolved communication systems 3553
N. Fay, S. Garrod & L. Roberts
Culture, embodiment and genes: unravelling the triple helix 3563
M. Wheeler & A. Clark
Exploring gene-culture interactions: insights from handedness, sexual selection and niche-construction
case studies 3577
K. N. Laland
Cultural evolution: implications for understanding the human language faculty and its evolution 3591
K. Smith & S. Kirby

The world’s longest running international science journal

Founded in 1660, the Royal Society
is the independent scientific academy
of the UK, dedicated to promoting
excellence in science
journals.royalsociety.org
Registered Charity No 207043 Published in Great Britain by the Royal Society, 12 November 2008
6–9 Carlton House Terrace, London SW1Y 5AG
RSTB_363_1509.qxp 9/30/08 5:06 PM Page 2

GUIDANCE FOR AUTHORS

Editor
Professor Georgina Mace Selection criteria to satisfy most non-specialist readers. Supplementary
The criteria for selection are scientific excellence, data up to 10Mb is placed on the Society's website free
Publishing Editor originality and interest across disciplines within biology. of charge. Larger datasets must be deposited in
Claire Rawlinson The Editors are responsible for all editorial decisions and recognised public domain databases by the author.
they make these decisions based on the reports received
Editorial Board from the referees and/or Editorial Board members. Many
more good proposals and articles are submitted to us Conditions of publication
Neuroscience and Cognition Organismal, environmental and evolutionary
than we have space to print, we give preference to Articles must not have been published previously, nor be
Dr Brian Billups biology
Professor Georgina Mace those that are of broad interest and of high scientific under consideration for publication elsewhere. The main
Dr Andrew Glennerster
Professor Bill Harris Professor Yadvinder Malhi quality. findings of the article should not have been reported in
Professor Atsushi Iriki Professor Manfred Milinski the mass media. Like many journals, Phil. Trans. R. Soc. B
Professor Trevor Lamb Professor Peter Mumby Publishing format employs a strict embargo policy where the reporting of
Professor Obaid Siddiqi Professor Peter Raven
Phil. Trans. R. Soc. B articles are published regularly online a scientific article by the media is embargoed until a
Professor Marc Tessier-Lavigne Professor Karl Sigmund
Professor Monkombu Swaminathan and in print issues twice a month. Along with all Royal specific time. The Executive Editor has final authority in
Professor Andrew Whiten
Society journals, we are committed to archiving and all matters relating to publication.
Cell and developmental biology Health and Disease providing perpetual access. The journal also offers the
Professor Julie Ahringer Professor Zhu Chen facility for including Electronic Supplementary Material
Dr Buzz Baum Professor Mark Enright (ESM) to papers. Contents of the ESM might include Electronic Submission details
Dr Anne Donaldson Professor Peter Goodfellow details of methods, derivations of equations, large tables For full submission guidelines and access to all journal
Professor Richard Gardner Professor Michael Malim
of data, DNA sequences and computer programs. content please visit the Phil. Trans. R. Soc. B website at
Professor John Gray Professor Lap-Chee Tsui
Professor Keith Gull Professor Nicholas Wald However, the printed version must include enough detail publishing.royalsociety.org/philtransb.
Professor Fotis Kafatos Professor Bob Williamson
Professor Elliot Meyerowitz
Professor Anthony Pawson

Publishing Editor: Claire Rawlinson AIMS AND SCOPE

Each issue of Phil. Trans. R. Soc. B is devoted to a specific area of the biological sciences.
(tel: +44 (0)20 7451 2602; fax: +44 (0)20 7976 1837; This area will define a research frontier that is advancing rapidly, often bridging The Royal Society is an independent scientific academy For further information on the Society’s activities, please
claire.rawlinson@royalsociety.org) traditional disciplines. Phil. Trans. R. Soc. B is essential reading for scientists working
across the biological sciences. In particular, the journal is focused on the following four
founded in 1660 and self-governing under Royal contact the following departments on the extensions
cluster areas: neuroscience and cognition; organismal and evolutionary biology; Charter. The Society has three roles, as the scientific listed by dialling +44 (0)20 7839 5561, or visit the
cell and developmental biology; and health and disease. As well as theme issues, academy of the United Kingdom, as a learned society, Society’s Web site (www.royalsociety.org).
the journal publishes papers from the Royal Society’s biological discussion meetings.
Production Editor: Victoria Brown For information on submitting a proposal for a theme issue, consult the journal‘s
as a funding body.
website at publishing.royalsociety.org/philtransb. Research Support (UK grants and fellowships)
The objectives of the Royal Society are to Research appointments: 2547
Reviews. The journal also publishes reviews in the broad areas of research listed
above. Review articles will often be commissioned, but the Editor is happy to consider
Research grants: 2539
6–9 Carlton House Terrace, London SW1Y 5AG, UK suggestions / proposals for unsolicited review articles. Please submit an abstract and a • recognise excellence in science Conference grants: 2540
covering letter to the Editorial Office for approval for consideration. 6,000 words is
publishing.royalsociety.org usually a maximum length for reviews. • support leading-edge scientific research and its Science Advice
applications General enquiries: 2585
ISBN: 978-0-85403-708-7
• stimulate international interaction
Copyright © 2008 The Royal Society Science Communication
• further the role of science, engineering General enquiries: 2572
Except as otherwise permitted under the Copyright, Designs and Patents Act, 1988, this publication may only be reproduced, stored or and technology in society
transmitted, in any form or by any other means, with the prior permission in writing of the publisher, or in the case of reprographic International Exchanges (for grants enabling
reproduction, in accordance with the terms of a licence issued by the Copyright Licensing Agency. In particular, the Society permits the
• promote the public’s understanding of science research visits between the UK and most other
making of a single photocopy of an article from this issue (under Sections 29 and 38 of this Act) for an individual for the purposes of • provide independent authoritative advice on countries (except the USA))
research or private study. matters relating to science, engineering General enquiries: 2550
and technology
Library and Information Services
SUBSCRIPTIONS
• encourage research into the history of science Library/archive enquiries: 2606
In 2009 Phil. Trans. R. Soc. B (ISSN 0962-8436) will be
published twice a month. Full details of subscriptions and
single issue sales may be obtained either by contacting Subscription prices All other
our journal fulfilment agent, Portland Customer 2009 calendar year Europe USA & Canada countries
Services, Commerce Way, Colchester CO2 8HP; tel: printed version plus £2024/US$3845 £2136/US$4058 £2186/US$4153
+44 (0)1206 796351; fax: +44 (0)1206 799331; email: Cover image: Chinese Whispers I–IX (Series I ) 2006. By Helen Flanagan. Nine carborundum prints on Somerset paper
electronic access /€2631
sales@portland-services.com or by visiting our website (Edition of five) Each approx 56.5
44cm.
at publishing.royalsociety.org/subscribers. The Royal
Society is a Registered Charity No. 207043.
Typeset in India by the Alden Group, Oxfordshire. Printed by Latimer Trend, Plymouth.
This paper meets the requirements of ISO 9706:1994(E) and ANSI/NISO Z39.48-1992 (Permanence of Paper) effective with volume 335, issue 1273, 1992.
Philosophical Transactions of the Royal Society B (ISSN: 0962-8436) is published twice a morth by the Royal Society and distributed in the USA by DSW, 75
Aberdeen Road, Emigsville PA 17318-0437. Periodicals postage paid at Emigsville PA. POSTMASTER: send address changes to Philosophical Transactions
of the Royal Society B, c/o PO Box 437 Emigsville PA 17318-0437.
Cultural transmission and the evolution
of human behaviour
Papers of a Theme Issue compiled and edited by Kenny Smith,
Michael L. Kalish, Thomas L. Griffiths and Stephan Lewandowsky

Contents

Introduction. Cultural transmission and the evolution of human behaviour 3469

K. Smith, M. L. Kalish, T. L. Griffiths and S. Lewandowsky

Establishing an experimental science of culture: animal social diffusion experiments 3477

A. Whiten and A. Mesoudi

The multiple roles of cultural transmission experiments in understanding human 3489

cultural evolution
A. Mesoudi and A. Whiten

Theoretical and empirical evidence for the impact of inductive biases on cultural 3503
evolution
T. L. Griffiths, M. L. Kalish and S. Lewandowsky

Beyond existence and aiming outside the laboratory: estimating frequency-dependent 3515
and pay-off-biased social learning strategies
R. McElreath, A. V. Bell, C. Efferson, M. Lubell, P. J. Richerson and T. Waring

Studying cumulative cultural evolution in the laboratory 3529

C. A. Caldwell and A. E. Millen

Investigating children as cultural magnets: do young children transmit redundant 3541

information along diffusion chains?
E. Flynn

The fitness and functionality of culturally evolved communication systems 3553

N. Fay, S. Garrod and L. Roberts

Culture, embodiment and genes: unravelling the triple helix 3563

M. Wheeler and A. Clark

Exploring gene–culture interactions: insights from handedness, sexual selection and 3577
niche-construction case studies
K. N. Laland

Cultural evolution: implications for understanding the human language faculty and 3591
its evolution
K. Smith and S. Kirby

3467
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3469–3476

doi:10.1098/rstb.2008.0147
Published online 17 September 2008

Introduction. Cultural transmission and the

evolution of human behaviour
Kenny Smith1,*, Michael L. Kalish2, Thomas L. Griffiths3
and Stephan Lewandowsky4
1
Division of Psychology, Cognition and Communication Research Centre, Northumbria University,
Northumberland Building, Northumberland Road, Newcastle upon Tyne NE1 8ST, UK
2
Institute of Cognitive Science, University of Louisiana at Lafayette, Lafayette, LA 70504-3772, USA
3
Department of Psychology, University of California, Berkeley, CA 94720-1500, USA
4
Department of Psychology, University of Western Australia, Crawley, WA 6009, Australia
The articles in this theme issue seek to understand the evolutionary bases of social learning and the
consequences of cultural transmission for the evolution of human behaviour. In this introductory
article, we provide a summary of these articles (seven articles on the experimental exploration of
cultural transmission and three articles on the role of gene–culture coevolution in shaping human
behaviour) and a personal view of some promising lines of development suggested by the work
summarized here.
Keywords: social learning; cultural transmission; cultural evolution; human evolution;
evolutionary psychology; diffusion chain

1. INTRODUCTION 2. AN EXPERIMENTAL APPROACH TO CULTURE

Humans learn from other humans in a wide variety of
domains. Consequently, systems of knowledge and
behaviour are culturally transmitted in human popu-
lations. The articles in this theme issue seek to
understand the evolutionary bases and consequences
of cultural transmission: how widespread is cultural
transmission in the animal kingdom; how does cultural
transmission work in human populations; what
products does cultural evolution deliver; and how has
culture interacted with biological evolution to shape
our species?
Rather than outline our own research on cultural
transmission and human behaviour (which is presented
at length in the papers by Griffiths et al. (2008) and
Smith & Kirby (2008)), our aim in this paper is to
summarize the content of the articles in the of this
issue, identify common themes and offer a personal
view on the directions in which this research pro-
gramme should be developed.
The articles in this issue fall into two groups. The
first seven papers deal with experimental approaches to
studying cultural transmission and cultural evolution—
these contributions are sketched out in §2. The second
group of articles (the final three articles in this issue,
described in §3) deal with the interactions between
biological and cultural evolution and, in particular, the
relationship between coevolutionary theories and
theories that seek to explain human behaviour purely
or primarily in terms of biological evolution (the
Evolutionary Psychology approach).
* Author for correspondence (kenny.smith@northumbria.ac.uk).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
Inquiry into the evolutionary bases and consequences
of cultural transmission is of course not a new
endeavour: evolutionary approaches to culture have
a distinguished history (e.g. Darwin 1879/2004,
pp. 112–114, draws direct parallels between biologi-
cal evolution and the cultural evolution of words
and languages), and the study of cultural trans-
mission and cultural evolution is a vibrant and
growing research area (see Mesoudi et al. (2006b)
for a programmatic review). Much of this research
has been theoretical or observational in nature, based
on formal models of evolutionary processes (e.g.
Cavalli-Sforza & Feldman 1981; Boyd & Richerson
1985; Richerson & Boyd 2005) or observational
study of real-world cultural phenomena (e.g. Durham
1991; Rogers 1995).
While these remain important tools for studying
cultural evolution, they are not the only ones available.
A further possibility is to adopt an experimental
approach to explore the mechanisms and dynamics of
cultural transmission—experimental study offers a
potential bridge between the generality and control of
the formal model and the naturalism of observation
of real behaviour in real cultural environments.
A powerful experimental approach, with a long history
but undergoing a renaissance in recent years, is to
investigate cultural evolution directly in simple labora-
tory populations under controlled conditions, in order
to establish what actually happens when people learn
from other people (or, indeed, when non-humans
learn from non-humans). This theme issue brings
together, for the first time, a recent and growing body
of work that applies these experimental techniques
(sometimes called diffusion chains or transmission chains)
to investigate cultural evolution.

3469 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3470 K. Smith et al. Introduction
The experimental body of this issue can be further
subdivided into three sections, discussed in turn below.
The first section consists of two papers by Mesoudi &
Whiten (2008) and Whiten & Mesoudi (2008), which
review the methodologies and early findings from
experimental studies of cultural transmission and
evolution (see §2a). The second section (§2b), which
consists of papers by Griffiths et al. (2008) and
McElreath et al. (2008), looks in more detail at two
key issues in the experimental study of cultural
transmission: what do culturally transmitted systems
adapt to and who do we learn from when learning
socially? The final set of experimental papers from
Caldwell & Millen (2008), Fay et al. (2008) and Flynn
(2008), summarized in §2c, consider the functionality
of the products of cultural evolution: to what extent do
culturally transmitted behaviours accumulate modifi-
cations over time to produce complex and well-adapted
behaviours?
(a) Review of experimental methods
The primary methodology for the experimental study
of cultural transmission (the diffusion chain experi-
ment) dates back at least to Bartlett’s (1932) serial
reproduction experiments. In common with dyadic
studies of social learning (e.g. Bandura 1977), diffusion
chain experiments are based around a pairwise learning
interaction, whereby one individual produces a
behaviour for observation by another individual, who
attempts to learn or reproduce that behaviour. The
novelty of the diffusion chain method is that the second
individual is then used as the model, producing
behaviour for a third individual and so on—a mini-
culture is created in the laboratory. Despite its
venerability, the diffusion chain has been something
of a fringe paradigm, used by a small number of
researchers in a wide range of disciplines (most notably,
comparative biology and psychology) to address a fairly
eclectic set of research questions. This situation has
recently undergone a dramatic shift, as methodological
advances have seen an increase in the use of the
diffusion chain method and an increasing awareness
across disciplinary boundaries of the techniques in use
and the questions addressed. The contributions from
Mesoudi & Whiten (2008) and Whiten & Mesoudi
(2008) review these developments, outlining the
diversity of diffusion chain methodologies available
(the ‘linear chain’ method outlined above is but one of
several) and their application to the questions of animal
culture and the determinants of cultural evolution in
human populations.
One of the fundamental questions in understanding
the human capacity for culture is to identify its
evolutionary origins: is this a recent ability, or an ancient
one which simply appears in an unusual form in our
species? Whiten & Mesoudi review the literature on
diffusion studies in non-human animals, focusing on the
range of experimental methodologies employed and
their ability to distinguish social learning and cultural
transmission from other mechanisms capable of produ-
cing similar group-level behaviours (e.g. individual
learning). The achievements in this area have been
impressive: there is clear evidence for cultural trans-
mission in a number of non-human species (primates,
Phil. Trans. R. Soc. B (2008)
but also rodents, birds and fishes). Furthermore,
transmission is seen under a range of experimental
regimes, ranging from the highly controlled linear
chain design (as described above) through to the less
controlled but more naturalistic open-diffusion design,
where a behaviour is seeded in a population and allowed
to spread through that population in a spontaneous and
uncontrolled fashion.
In the process of this review, Whiten & Mesoudi also
identify the limits of this literature: the range of species
which have to date been studied in this fashion is fairly
limited, and the range of social learning tasks is also
somewhat restricted. In addition to broadening taxo-
nomic and task coverage, Whiten & Mesoudi identify
one of the major challenges facing the burgeoning
animal diffusion literature as the move from studies
involving captive animals to controlled studies in the
wild. Such studies would serve to narrow the current
divide between naturalistic but uncontrolled (and
therefore often uninterpretable) studies of putative
cultural behaviour in the wild and controlled but fairly
artificial studies in captivity.
The material reviewed by Whiten & Mesoudi speaks
to establishing the existence (or at least the capacity for
supporting) culturally transmitted traditions in various
species. In our species, the question is not one of the
existence of culture, but the details of the cultural
transmission process and the cultural evolutionary
dynamic it engenders. Mesoudi & Whiten review the
historical and contemporary literature on human
cultural transmission experiments, with a focus on
how this literature addresses four issues: (i) what kinds
of information are stable over repeated episodes of
cultural transmission, (ii) who do social learners chose
to learn from when learning socially, (iii) when is social
learning favoured over alternatives, and (iv) how, on a
mechanistic level, does social learning work? To give
brief examples: addressing the ‘what’ question, linear
diffusion chain studies show that human learners bring
a number of biases for particular sorts of content to
social learning tasks (e.g. biases in favour of social over
non-social information; Mesoudi et al. 2006a) and
these biases result in more faithful transmission of
information that meets the content biases of individ-
uals; addressing the ‘who’ question, closed-group
studies (where a group of individuals repeatedly
interact; e.g. Efferson et al. 2008) show that at least
some humans exploit frequency information when
confronted with a social learning problem, preferen-
tially copying the behaviour of the population majority.
Again, in common with the review of the non-
human diffusion literature provided by Whiten &
Mesoudi, this review reveals a picture of a healthy but
relatively youthful discipline: an exciting proliferation
of methods and promising early results, but relatively
little systematic evaluation of experimental tools or
integration of studies addressing each of the four
issues above.
(b) What and who
In their paper, Griffiths et al. provide several case
studies that seek to address Mesoudi & Whiten’s ‘what’
question: what kinds of culturally transmitted
behaviours are stable over time, and when a culturally
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
transmitted behaviour changes over time, what is it
changing towards? Griffiths et al. adopt a mix of
mathematical and experimental diffusion chain tech-
niques to demonstrate that culturally transmitted
behaviours adapt to fit the inductive biases of learners.
Any learning process has some bias—some behaviours
are easier to learn (require less data to learn) than
others, due to the architecture of the learning system
and the constraints inherent in it. Culturally trans-
mitted systems repeatedly undergo filtering through
these inductive biases of learners as they are passed
from individual to individual.
Griffiths et al. summarize their own mathematical
work (Griffiths & Kalish 2005, 2007; Kirby et al. 2007),
which shows that, under a fairly broad set of assump-
tions, cultural evolution will lead to systems that mirror
the inductive biases of individuals—seen in this light, the
various examples provided by Mesoudi & Whiten would
then be specific instances of a more general phenom-
enon. Furthermore, these inductive biases can over-
whelm contrary pressures from natural selection—in
conditions where the learning biases of individuals
favour one behaviour and natural selection favours
another, inductive biases win out under a broad range of
conditions. Griffiths et al. support this formal modelling
work with a summary of their laboratory experiments
(Kalish et al. 2007; Griffiths et al. 2008) in two domains
where the inductive biases of individuals are already well
established—function learning and categorization—and
show that cultural versions of these tasks result in
convergence to behaviours (functions or categories) that
match the inductive biases of individuals.
We would highlight one final contribution by
Griffiths et al., derived from formal modelling. They
show an equivalence between the equilibria of cultural
evolution in linear transmission chains and populations
where there are multiple individuals per generation.
Specifically, the stable outcome of cultural evolution
(the stationary distribution) arrived at by each process
should be the same—after cultural evolution has run its
course, the probability that a particular individual in a
linear chain will exhibit a particular behaviour is equal
to the proportion of individuals exhibiting that
behaviour in a population. In other words, studying
simple linear chains of transmission potentially offers a
short cut to determining the outcomes of cultural
evolution in populations. This constitutes an additional
justification for studying cultural evolution in simpli-
fied, manageable laboratory populations, and establish-
ing the range of conditions under which behaviour in
laboratory populations approximates behaviour of
larger and more complex populations would be a
worthwhile next step.
Rather than asking what inductive biases learners
bring to social learning tasks, McElreath et al. seek to
understand the extent to which humans use social
information and, importantly, how multiple types of
social information are integrated. Social learning is not
the only way in which individuals can adapt to
challenges posed by their environment (an alternative
is to learn individually), and social learners face choices
about who they learn from (e.g. conforming to the
majority behaviour or preferentially copying more
successful individuals). Furthermore, such behaviours
Phil. Trans. R. Soc. B (2008)
Introduction K. Smith et al. 3471
need not be applied exclusively—learners can learn
through a combination of individual and social learning,
and apply a combination of social learning strategies
(e.g. by weighted or hierarchical combinations of pay-
off-based and conformity-based strategies).
McElreath et al. use an abstract task (‘crop
selection’, where different crops have different yields
and the pay-off changes periodically) that can be solved
by individual or social means—participants have access
to the pay-offs associated with their own past choices
but also the choices and pay-offs of several other
individuals. McElreath et al. then use model-fitting
techniques to identify which combinations of individ-
ual and social learning strategies best describe the
actual choices that their experimental participants
made (similarly to the approach used in, for example,
Efferson et al. (2008)). They find that their participants
combine individual and social learning, attending to a
hierarchically organized combination of pay-off and
frequency information when learning socially (prefer-
entially copying high pay-off behaviours, but selecting
the most frequently chosen response when the
difference in yields is less marked). This use of pay-
off-based social learning is predicted by McElreath
et al.’s mathematical analysis to be the most successful
strategy under a wide range of assumptions about pay-
offs and environmental variability.
While this is in itself an interesting result, McElreath
et al. see the main contribution of this approach as a
means of studying social learning in the wild: while they
apply their fitting technique to laboratory results, the
same approach could be applied to real-world data (e.g.
the diffusion of competing innovations in an open-
diffusion study of the type outlined by Whiten &
Mesoudi (2008)). This approach therefore offers an
alternative to existing experimental approaches (dyadic
or diffusion chain) to teasing apart social learning
strategies—it may be that in some cases the behavioural
signatures of different social learning strategies are
sufficient to identify those strategies.
(c) Cultural evolution and functionality
One of the main motivations for understanding the
human capacity for culture is that it appears to form the
basis of some of humanity’s most surprising achieve-
ments. Sophisticated technologies, highly developed
sciences and elaborate social or religious rituals are
products of a cumulative process of cultural evolution,
whereby each generation builds on the achievements of
their predecessors in a gradual, approximately mono-
tonic ratcheting up of complexity and functionality
( Tomasello 1999). The final three experimental
articles in this issue apply the methods reviewed by
Mesoudi & Whiten to an exploration of this class of
phenomenon: to what extent does cultural trans-
mission yield products that are well designed, and can
we use experimental techniques to delve into the
processes that produce these functional outcomes?
Caldwell & Millen provide an introduction to the
area of cumulative cultural evolution: its taxonomic
spread (its presence in non-humans is contentious); the
mechanisms underpinning it (it remains unclear
whether sophisticated imitation is required for cumu-
lative cultural evolution, or whether more basic social
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3472 K. Smith et al. Introduction
learning mechanisms will suffice); and the types of
outcomes it yields (e.g. whether cumulative cultural
evolution can deliver behaviours that are universal
cross-culturally). The latter is a crucial issue: cross-
cultural universality is often taken as a hallmark of
non-cultural transmission—for example, fundamental
structural similarities across diverse languages are often
taken as evidence for a universal genetically specified
language blueprint (Chomsky 1965). Caldwell & Millen
summarize their own experimental work (Caldwell &
Millen 2008) which uses a diffusion chain approach
to explore the cumulative cultural evolution of tech-
nological artefacts. As well as demonstrating the
phenomenon under laboratory conditions, they show
convergent evolution across separate populations
towards similar artefact designs, indicating that, under
certain circumstances, cumulative cultural evolution can
potentially offer a non-genetic explanation for cross-
cultural universals.
Flynn provides a second illustration of cultural
transmission delivering improved traditions, building on
previous dyadic work (e.g. McGuigan et al. 2007) which
suggests that children are prone to over-imitation—they
copy both task-relevant and task-irrelevant (and
therefore non- or a-functional) behaviours. Flynn
presents infants (aged 2–3 years) with a box-opening
(‘artificial fruit’) task. Diffusion chains are initialized with
a mix of relevant (directed to retrieving a sticker from
the box) and irrelevant (not contributing to releasing the
sticker) behaviours. While task-relevant actions are
faithfully transmitted down multiple generations of these
chains, irrelevant actions are rapidly filtered from the
populations’ behavioural repertoire. The culturally trans-
mitted patterns of behaviour in these populations there-
fore become more efficient over transmission events, in
line with the notion of cumulative cultural evolution.
Finally, Fay et al. offer a detailed experimental
evaluation of the optimality of the products of cultural
evolution. They focus on graphical communication
systems that are produced in an experimental paradigm
(described in detail in Garrod et al. (2007)) where adult
human participants negotiate communication systems
through repeatedly playing a graphical communication
game similar to the parlour game Pictionary. Fay et al.
contrast the graphical communication systems that
emerge through two different routes: repeated
interaction between a single pair of participants
(isolated pair systems) and repeated interaction within
a community of multiple individuals (community
systems). Both isolated pairs and communities start
off with iconic systems of representation (based around
relatively complex drawings that resemble the concepts
they refer to) and develop more streamlined symbolic
communication systems (drawings become consider-
ably simplified and abstract). This symbolization is
attributable to pressure for the participants to minimize
their effort in producing graphics, while still maintain-
ing distinct symbols for distinct concepts—in this sense
both isolated pair and community systems are highly
functional. Fay et al. show, however, that the commu-
nity systems also simultaneously optimize their trans-
missibility (see Kirby et al. (2008) for a related result).
In communities, the ideal communicative symbol
will not only be (i) economical to produce and
Phil. Trans. R. Soc. B (2008)
(ii) distinctive, but (iii) will have some residual iconicity
that allows an individual who has not seen this
particular symbol before to infer its meaning—this
pressure does not exist in purely pair-based systems,
where both participants are privy to every symbol’s
iconic roots. Consequently, community-evolved sym-
bols are optimized along this third dimension and
therefore (as Fay et al. (2008) show) easier for naive
individuals to learn.
(d) Experimental models of cultural
transmission: a summary
The experimental study of cultural transmission is a
rapidly developing and coalescing field: as the articles
in the body of this issue show, the processes of
developing a consensus on the appropriate experi-
mental methodologies, the overarching theoretical
predictions and the key sub-topics have begun.
However, this consensus building is at an early
stage and much remains to be done. Some of this
outstanding work is methodological in nature. For
example, while Whiten & Mesoudi are able to compare
results obtained across experimental designs, little
work directed explicitly at evaluating the impact of
different experimental designs has been done to date
(but see Whiten et al. (2005), Horner et al. (2006) and
Griffiths et al. (2008), which show that some results can
be replicated with different diffusion chain designs).
Furthermore, there remains little agreement on the
validity of the different available methods for addres-
sing particular questions. To take an example touched
upon in this issue: while there is a general agreement
that cumulative cultural evolution is an important sub-
topic to address, there is less agreement on the best
method to explore it. While Flynn uses a linear
diffusion chain to study cumulative cultural evolution,
Caldwell & Millen are somewhat critical of the
suitability of this experimental design for investigating
this phenomenon. One of the challenges for the future
is to explore more fully the methodological space and
address these issues head-on—as experimentalists, we
should be prepared to subject our methodologies to
experimental test, in particular testing for consistency
across different diffusion chain designs.
Of course, the points of dispute are not merely
methodological. Again, to take an example from this
issue: while some theoretical accounts of cumulative
cultural evolution (e.g. Tomasello 1999) emphasize the
importance of a cultural ratchet, such that functional
modifications are preserved and not lost (the ratchet
prevents the evolving behaviour slipping backwards
towards non-functionality), the chain-by-chain results
of cultural evolution presented by Caldwell & Millen
(2008, fig. 2 in their paper) look anything but
ratcheted—performance of the evolving artefacts
frequently decreases from generation to generation,
although the overall trend is upwards. While this could
be explained as a consequence of a slightly noisy
mapping from quality of design to measured function-
ality in this particular experiment (even the best
designed spaghetti tower will collapse if constructed
from substandard ingredients), this explanation
works less well in the case of highly non-functional
innovations in Flynn’s diffusion chains (while the
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
general trend is to eliminate irrelevant actions, one
child introduced multiple unnecessary movements
of the box door). Ideally, these kinds of experimental
phenomenon should be fed back into a refined theory
(in this case, can our theory tolerate a slippy ratchet?),
generating new predictions to be tested experimentally
(e.g. how slippy can the ratchet be if we are still to see
cumulative cultural evolution in the laboratory?), and
perhaps touching upon the sort of methodological
questions outlined above (e.g. do certain transmission
dynamics, such as linearity, lead to less of a ratchet
effect and reduced cumulativity?).
3. THE RELATIONSHIP BETWEEN GENES
AND CULTURE
The final three articles in this issue move beyond the
experimental study of cultural transmission to consider
the wider issue of how cultural evolution interacts with
that other source of adaptive behaviour in the natural
world, biological evolution. The deeply cultural nature
of human cognition must ultimately be rooted in our
biology: it has, for example, been attributed to uniquely
human social learning mechanisms (e.g. Tomasello
(1999) and discussion in the paper by Whiten &
Mesoudi (2008) and references therein). Culture also
influences biology: for example, it is often argued that
our cognitive capacities have been massively adapted to
work in conjunction with the human cultural inheri-
tance (Sperber 1996). Furthermore, culture provides
a second inheritance system for human behaviour
(Boyd & Richerson 1985; Whiten 2005; Mesoudi et al.
2006b). The appearance of design in human behaviour
therefore has at least two possible causes, biological or
cultural evolution, and explaining the origins of
complex and adaptive human behaviours requires us
to understand which inheritance systems carry and
shape which behaviours, as well as understanding how
these two inheritance systems interact.
The articles in this section address these issues of
interactions between biology and culture. Further-
more, all three are explicitly concerned with addressing
the relationship between explanations of human
behaviour involving cultural evolution and the popular
Evolutionary Psychology approach (henceforth EP;
e.g. Cosmides & Tooby 1987; Pinker 1997). Unlike
cultural accounts, the EP school of thought is wide-
spread in the psychological community and, indeed, in
the popular consciousness. As such, pinning down the
relationship between cultural and EP accounts is an
important issue for proponents of cultural or coevolu-
tionary explanations of human behaviour, both on a
practical level (to assist in the promulgation of these
theories) and from a scientific standpoint (to determine
which theory has greater explanatory power).
The classic EP account sees human behaviour as
governed by a set of hard-wired, task-specific mental
modules evolved to deal with specific ecological
challenges posed by the ancestral human environment.
As pointed out by Wheeler & Clark in their contribution
here, the EP explanatory approach seems fundamentally
at odds with two alternative and powerful explanations
of human behaviour: cultural evolution and embodied
cognition. Cultural evolutionary accounts allow for a
Phil. Trans. R. Soc. B (2008)
Introduction K. Smith et al. 3473
role for extra-genetic transmission and adaptation.
Embodied accounts of cognition emphasize the recipro-
cal relationship between an organism and its environ-
ment, such that the environment is exploited to reduce
the cognitive burden on the brain and structure in the
environment in turn impacts on the way in which
the brain seeks to solve problems. Both emphasize the
capacity for non-genetic factors to influence behaviour
and highlight the self-constructing and bootstrapping
nature of an organism’s or population’s interaction with
its environment.
Wheeler & Clark argue that this apparent incompat-
ibility between EP, embodied cognition and culture can
be resolved by a more nuanced view of how an evolved
mental module might interact with its (self- and
culturally constructed) environment. For example,
the initial disposition of the cognitive system (poten-
tially a component of our evolved mind) interacts with
the environment (which may be constructed and
exploited by the individual and/or their cultural
predecessors) via an incremental bootstrapping pro-
cess, such that the brain develops along the route
primed by the genes but shaped through interaction
with the environment. Under the most extreme
interactionist version of this argument ‘what is special
about human brains. may be precisely their ability
(.) to enter into deep, complex and ultimately
architecture-determining relationships with an open-
ended variety of culturally transmitted practices,
endowments, and non-biological constructs, props
and aids’ (Wheeler & Clark 2008). At the other end
of the spectrum lies something resembling the classic
EP position, where interaction with the environment is
downplayed. Wheeler & Clark see the challenge facing
an integrated, embodied cultural EP as identifying
where on this spectrum from heavy genetic influence to
emergent mind each aspect of human cognition resides.
Laland offers three case studies on the intimate
coevolutionary relationship between culture and genes
in shaping human behaviour. While adopting a far less
conciliatory tone towards EP, Laland’s (2008) con-
clusion is broadly similar to that of Wheeler & Clark:
‘human minds and human environments have engaged
in a long-standing, intimate exchange of information.
leaving each beautifully fashioned in the other’s image’.
Laland’s first case study (Laland et al. 1995) deals
with explaining variation in handedness in human
populations. While purely genetic accounts of handed-
ness are highly influential, Laland shows that the best
fit to the observed data on human handedness is
obtained by a model where genes and culture (in the
form of parental shaping of offspring handedness)
interact: no purely genetic account fits the data on
heritability and cross-cultural variation. In other words,
EP-style accounts that ignore cultural influences on
behaviour (such as handedness) risk falling at the first
hurdle of explaining observed human behaviour.
The second and third case studies deal with
situations where culturally transmitted traits (mate
preferences in the second case study and niche-
constructing capacity in the third) impact on or change
the course of biological evolution. Preferences for
sexual partners (one of the core areas of EP; e.g. Buss
1994; Miller 2001) can be influenced by the observed
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3474 K. Smith et al. Introduction
preferences of others (Jones et al. 2007), and Laland’s
(1994) formal modelling work shows that such socially
learned mate preferences can generate selection acting
on biological evolution that takes the preferred trait to
fixation in the opposite sex. Culturally transmitted
niche-constructing behaviours can generate selection
pressures that drive evolution in directions differing
from those that would be favoured by the unmodified
environment (Laland et al. 2001), suggesting that
heavy niche constructors (such as humans) should be
less responsive to selection pressures arising from
changes in the environment, because they can modify
that environment to attenuate those pressures. As
Laland points out, this is at odds with the tenet of EP
that humans are operating with a set of mental modules
adapted for our ancestral environment and possibly
maladapted to our current environment—to a large
extent, we construct our environment to suit ourselves.
The final article by Smith & Kirby (2008) similarly
tackles gene–culture interactions and the EP approach
to explaining human behaviour, with a specific focus on
language. Language underpins many culturally trans-
mitted human behaviours, but is itself a culturally
transmitted system: we learn the language we hear
around us as we grow up. Despite this fairly obvious
contribution from culture, explanations of language
design (why does human language have the particular
characteristics it does?) have typically been biological
rather than cultural: following the classic EP model,
the argument is that language looks the way it does
because the mental module dedicated to language, the
language faculty, evolved to build in those features,
primarily because they are useful for communication
(e.g. Pinker & Bloom 1990).
Smith & Kirby argue against the necessity of this
strong EP position on language in two ways. First, they
review a body of computational work, developed over
the past 10 years, which shows that cultural evolution
can explain certain aspects of language design.
Specifically, a language, like any other cultural system,
can only survive repeated transmission if it can be
reliably learned, and certain design features of language
can be seen as cultural adaptations to these learnability
constraints. Second, they show that cultural trans-
mission has the potential to fundamentally alter the
sorts of language faculty that natural selection
favours—under certain scenarios, selection acting on
the language faculty pushes evolution into regions of
design space where the language faculty only weakly
constrains the structure of language. In other words,
not only does cultural evolution potentially offer an
alternative explanation for some aspects of language
design, but it also potentially changes the extent to
which biological evolutionary accounts work at all.
(a) Genes and culture: a summary
The significance of gene–culture coevolutionary theory
has not to date been widely grasped in the section
of the research community for whom it is most rele-
vant: psychologists concerned with evolutionary expla-
nations of human behaviour. The more reductionist
explanations of EP, focusing on biology to the exclusion
of culture, hold sway in the broader consciousness, and
the final three articles in this section all represent
Phil. Trans. R. Soc. B (2008)
attempts to remedy this lack of penetration, by either
suggesting a synthesis (Wheeler & Clark 2008) or
attacking the foundations of EP (Laland 2008; Smith &
Kirby 2008). Of course, taking cultural transmission
seriously does not offer instant insights into the causes of
human behaviour—as highlighted by Laland (2008)
and Wheeler & Clark (2008), the relationship between
environment, genes and culture is rather intricate and
requires us to probe deeper into how we think the
various component parts of the theory work and how the
component parts interact. These are tough questions
and as such lack some of the appeal of clean EP
explanations for human behaviour. The challenge, as
met in Laland’s handedness case study, is to show that
coevolutionary theories provide a better fit to observed
human behaviour.
4. LOOKING AHEAD
As the articles gathered in this issue show, under-
standing cultural transmission is key to understanding
human behaviour. Many aspects of human behaviour
are influenced by social learning, including some of the
features that are often taken to differentiate humans
from other animals (e.g. complex technologies or
language), and purely biological explanations for the
evolution of such behaviours, as offered by EP, are
therefore deeply problematic. The explosion of interest
in the experimental study of social learning and cultural
transmission provides a promising and powerful tool
for understanding the relationship between cognition
and cultural evolution, bridging the gap between
theoretical and observational approaches.
One of our goals in editing this theme issue was to
provide a snapshot of the state of the field as it currently
stands, as a useful reference for researchers already
working in this area and a starting point for newcomers.
Another was to help drive the field forward, not least by
providing such a starting point. Given this second goal,
it seems only fair that we should provide some personal
thoughts on what we see as the potential near future of
the field.
We have already offered some suggestions in the
summary sections above on possible lines of develop-
ment, including: more systematic exploration of the
diffusion chain methodology (e.g. wider use of the
methods summarized by Whiten & Mesoudi and
explicit testing for convergence of results across
experimental designs, greater coverage of species and
social learning tasks); improving interaction between
theoretical and experimental results; following Laland’s
lead in challenging EP on conceptual and explanatory
grounds. We would highlight one further overarching
objective here, which recurs throughout the articles in
this issue: the desirability of a tighter coupling between
the three tools of theoretical model, experimental
model and real-world data.
Several of the articles here explicitly address this
triumvirate of approaches. McElreath et al. provide a
method for linking mathematical models (both
evolutionary and behavioural) to real behaviour, albeit
in the laboratory, and argue that this same technique
can be extended further, to explore and explain cultural
behaviour in the real world. Griffiths et al. similarly
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
provide integrated formal and experimental models
(based at present around linear diffusion chains,
although other forms of transmission could be
explored), and some suggestions regarding the real-
world cultural phenomena these models relate to.
Working from the other end, Whiten & Mesoudi suggest
that the divide between real-world and laboratory
studies of cultural evolution could be bridged by running
experimental techniques in the field. In the ideal world,
all these techniques might be brought to bear on a given
cultural behaviour of interest. For example, we might
first identify the phenomenology of a real-world human
behaviour which we expect involves social learning.
Formal models of possible social learning strategies or
inductive biases that aim to explain that behaviour
would be fitted to the real-world behaviour, providing an
indication of which sort of social learning strategy or bias
best describes that phenomenon. Moving to experi-
mental approaches, laboratory analogues and field
experiments could provide an opportunity to examine
model predictions, and therefore both potentially falsify
models and also unveil added levels of detail on the
social learning processes underpinning the behaviour of
interest. Finally, further cultural or acultural experi-
ments could be used to tease apart the fine details of the
social learning mechanisms, before we return to the real
world to establish whether the predictions of the newly
developed model are consistent with the details of the
real-world behaviour.
Taking this kind of multi-pronged approach to
explaining human behaviour is of course extremely
challenging, not least because it requires a research
team with a detailed grasp of the real-world behaviour
of interest, familiarity with a range of experimental
methodologies and access to sophisticated mathemat-
ical modelling techniques. While we are much more
interested in getting on with doing the work than
agonizing over frameworks and terminology, it is a
truism that one of the barriers to this kind of
interdisciplinary research is the absence of common
expectations and terminology. At the very least, we
hope that this issue will provide the foundations for the
shared vocabulary and body of knowledge required by
this approach to explaining the role of cultural
transmission in shaping human behaviour.
This volume arose from a seminar on ‘Formal and experi-
mental models of cultural evolution’, held in April 2007, hosted
by the University of Edinburgh and funded by an Economic
and Social Research Council Research Seminar Series Award
held by Kenny Smith and Andrew Whiten. Preparation of this
article was supported by grants 0704034 and 0544705 from
the US National Science Foundation (to T.L.G. and M.L.K.,
respectively) and by a Discovery Project grant from the
Australian Research council to S.L. and Nic Fay.
REFERENCES
Bandura, A. 1977 Social learning theory. Oxford, UK:
Prentice-Hall.
Bartlett, F. C. 1932 Remembering. Oxford, UK: Macmillan.
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Buss, D. M. 1994 The evolution of desire: strategies of human
mating. New York, NY: Basic Books.
Phil. Trans. R. Soc. B (2008)
Introduction K. Smith et al. 3475
Caldwell, C. A. & Millen, A. E. 2008 Experimental models
for testing hypotheses about cumulative cultural
evolution. Evol. Hum. Behav. 29, 165–171. (doi:10.
1016/j.evolhumbehav.2007.12.001)
Caldwell, C. A. & Millen, A. E. 2008 Studying cumulative
cultural evolution in the laboratory. Phil. Trans. R. Soc. B
363, 3529–3539. (doi:10.1098/rstb.2008.0133)
Cavalli-Sforza, L. L. & Feldman, M. W. 1981 Cultural
transmission and evolution. Princeton, NJ: Princeton
University Press.
Chomsky, N. 1965 Aspects of the theory of syntax. Cambridge,
MA: MIT Press.
Cosmides, L. & Tooby, J. 1987 From evolution to behavior:
evolutionary psychology as the missing link. In The latest
on the best: essays on evolution and optimality (ed. J. Dupre),
pp. 227–306. Cambridge, MA: MIT Press.
Darwin, C. 1879/2004 The descent of man. London, UK:
Penguin.
Durham, W. H. 1991 Coevolution: genes, culture and human
diversity. Stanford, CA: Stanford University Press.
Efferson, C., Lalive, R., Richerson, P. J., McElreath, R. &
Lubell, M. 2008 Conformists and mavericks: the
empirics of frequency-dependent cultural transmission.
Evol. Hum. Behav. 29, 56–64. (doi:10.1016/j.evolhum-
behav.2007.08.003)
Fay, N., Garrod, S. & Roberts, L. 2008 The fitness and
functionality of culturally evolved communication
systems. Phil. Trans. R. Soc. B 363, 3553–3561. (doi:10.
1098/rstb.2008.0130)
Flynn, E. 2008 Investigating children as cultural magnets: do
young children transmit redundant information along
diffusion chains? Phil. Trans. R. Soc. B 363, 3541–3551.
(doi:10.1098/rstb.2008.0136)
Garrod, S., Fay, N., Lee, J., Oberlander, J. & MacLeod, T.
2007 Foundations of representation: where might graphi-
cal symbol systems come from? Cogn. Sci. 31, 961–987.
(doi:10.1080/03640210701703659)
Griffiths, T. L. & Kalish, M. L. 2005 A Bayesian view of
language evolution by iterated learning. In Proc. 27th Annual
Conf. of the Cognitive Science Society (eds B. G. Bara,
L. Barsalou & M. Bucciarelli), pp. 827–832. Mahwah,
NJ: Erlbaum.
Griffiths, T. L. & Kalish, M. L. 2007 A Bayesian view of
language evolution by iterated learning. Cogn. Sci. 31,
441–480.
Griffiths, T. L., Christian, B. R. & Kalish, M. L. 2008 Using
category structures to test iterated learning as a method for
identifying inductive biases. Cogn. Sci. 32, 68–107.
(doi:10.1080/03640210701801974)
Griffiths, T. L., Kalish, M. L. & Lewandowsky, S. 2008
Theoretical and empirical evidence for the impact of
inductive biases on cultural evolution. Phil. Trans. R. Soc.
B 363, 3503–3514. (doi:10.1098/rstb.2008.0146)
Horner, V., Whiten, A., Flynn, E. & de Waal, F. B. M. 2006
Faithful replication of foraging techniques along cultural
transmission chains by chimpanzees and children. Proc.
Natl Acad. Sci. USA 103, 13 878–13 883. (doi:10.1073/
pnas.0606015103)
Jones, B. C., DeBruine, L. M., Little, A. C., Burriss, R. P. &
Feinberg, D. R. 2007 Social transmission of face
preferences among humans. Proc. R. Soc. B 274,
899–903. (doi:10.1098/rspb.2006.0205)
Kalish, M. L., Griffiths, T. L. & Lewandowsky, S. 2007
Iterated learning: intergenerational knowledge transmission
reveals inductive biases. Psychon. Bull. Rev. 14, 288–294.
Kirby, S., Dowman, M. & Griffiths, T. L. 2007 Innateness
and culture in the evolution of language. Proc. Natl Acad.
Sci. USA 104, 5241–5245. (doi:10.1073/pnas.06082
22104)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3476 K. Smith et al. Introduction
Kirby, S., Cornish, H. & Smith, K. 2008 Cumulative cultural
evolution in the laboratory: an experimental approach to
the origins of structure in human language. Proc. Natl
Acad. Sci. USA 105, 10 681–10 686. (doi:10.1073/pnas.
0707835105)
Laland, K. N. 1994 On the evolutionary consequences of
sexual imprinting. Evolution 48, 477–489. (doi:10.2307/
2410106)
Laland, K. N. 2008 Exploring gene–culture interactions:
insights from handedness, sexual selection and niche-
construction case studies. Phil. Trans. R. Soc. B 363,
3577–3589. (doi:10.1098/rstb.2008.0132)
Laland, K. N., Kumm, J., Van Horn, J. D. & Feldman, M. W.
1995 A gene-culture model of handedness. Behav. Genet.
25, 433–445. (doi:10.1007/BF02253372)
Laland, K. N., Odling-Smee, F. J. & Feldman, M. W. 2001
Cultural niche construction and human evolution. J. Evol.
Biol. 14, 22–33. (doi:10.1046/j.1420-9101.2001.00262.x)
McElreath, R., Bell, A. V., Efferson, C., Lubell, M.,
Richerson, P. J. & Waring, T. 2008 Beyond existence and
aiming outside the laboratory: estimating frequency-
dependent and pay-off-biased social learning strategies.
Phil. Trans. R. Soc. B 363, 3515–3528. (doi:10.1098/rstb.
2008.0131)
McGuigan, N., Whiten, A., Flynn, E. & Horner, V. 2007
Imitation of causally-opaque versus causally-transparent
tool use by 3- and 5-year-old children. Cogn. Dev. 22,
353–364. (doi:10.1016/j.cogdev.2007.01.001)
Mesoudi, A. & Whiten, A. 2008 The multiple roles of cultural
transmission experiments in understanding human cul-
tural evolution. Phil. Trans. R. Soc. B 363, 3489–3501.
(doi:10.1098/rstb.2008.0129)
Mesoudi, A., Whiten, A. & Dunbar, R. I. M. 2006a A bias
for social information in human cultural transmission.
Phil. Trans. R. Soc. B (2008)
Br. J. Psychol. 97, 405–423. (doi:10.1348/000712605
X85871)
Mesoudi, A., Whiten, A. & Laland, K. N. 2006b Towards a
unified science of cultural evolution. Behav. Brain Sci. 29,
329–383. (doi:10.1017/S0140525x06009083)
Miller, G. 2001 The mating mind: how sexual choice shaped
human nature. London, UK: Vintage.
Pinker, S. 1997 How the mind works. New York, NY: Norton.
Pinker, S. & Bloom, P. 1990 Natural language and natural
selection. Behav. Brain Sci. 13, 707–784.
Richerson, P. J. & Boyd, R. 2005 Not by genes alone. Chicago,
IL: University of Chicago Press.
Rogers, E. 1995 The diffusion of innovations. New York, NY:
Free Press.
Smith, K. & Kirby, S. 2008 Cultural evolution: implications
for understanding the human language faculty and its
evolution. Phil. Trans. R. Soc. B 363, 3591–3603. (doi:10.
1098/rstb.2008.0145)
Sperber, D. 1996 Explaining culture: a naturalistic approach.
Oxford, UK: Oxford University Press.
Tomasello, M. 1999 The cultural origins of human cognition.
Boston, MA: Harvard University Press.
Wheeler, M. & Clark, A. 2008 Culture, embodiment and
genes: unravelling the triple helix. Phil. Trans. R. Soc. B
363, 3563–3575. (doi:10.1098/rstb.2008.0135)
Whiten, A. 2005 The second inheritance system of
chimpanzees and humans. Nature 437, 52–55. (doi:10.
1038/nature04023)
Whiten, A. & Mesoudi, A. 2008 Establishing an experimental
science of culture: animal social diffusion experiments.
Phil. Trans. R. Soc. B 363, 3477–3488. (doi:10.1098/rstb.
2008.0134)
Whiten, A., Horner, V. & de Waal, F. B. M. 2005 Conformity
to cultural norms of tool use in chimpanzees. Nature 437,
737–740. (doi:10.1038/nature04047)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3477–3488

doi:10.1098/rstb.2008.0134
Published online 17 September 2008

Review

Establishing an experimental science of culture:

animal social diffusion experiments
Andrew Whiten1,* and Alex Mesoudi2
1
Centre for Social Learning and Cognitive Evolution, School of Psychology,
University of St Andrews, St Andrews KY16 9JP, UK
2
Department of Social and Developmental Psychology, University of Cambridge,
Free School Lane, Cambridge CB2 3RQ, UK
A growing set of observational studies documenting putative cultural variations in wild animal
populations has been complemented by experimental studies that can more rigorously distinguish
between social and individual learning. However, these experiments typically examine only what one
animal learns from another. Since the spread of culture is inherently a group-level phenomenon, greater
validity can be achieved through ‘diffusion experiments’, in which founder behaviours are
experimentally manipulated and their spread across multiple individuals tested. Here we review the
existing corpus of 33 such studies in fishes, birds, rodents and primates and offer the first systematic
analysis of the diversity of experimental designs that have arisen. We distinguish three main
transmission designs and seven different experimental/control approaches, generating an array with 21
possible cells, 15 of which are currently represented by published studies. Most but not all of the
adequately controlled diffusion experiments have provided robust evidence for cultural transmission in
at least some taxa, with transmission spreading across populations of up to 24 individuals and along
chains of up to 14 transmission events. We survey the achievements of this work, its prospects for the
future and its relationship to diffusion studies with humans discussed in this theme issue and elsewhere.
Keywords: culture; cultural transmission; social learning; diffusion experiments; diffusion chains;
transmission chains

1. INTRODUCTION: MIND THE GAP
The study of cultural processes in animals can now
boast approximately half a century of achievement,
generally considered to have been launched by the
famous efforts of Japanese researchers to document the
spread of novel behaviour patterns among groups of
macaque monkeys (Itani & Nishimura 1973; McGrew
1998). Reports of cultural phenomena in other
mammals, birds and fishes have since accumulated,
their frequency rising in recent years as decades of field
research on some species have facilitated the identifi-
cation of regional variations in behaviour, attributable
to social learning (learning from others; Whiten &
van Schaik 2007; Laland & Galef 2008). Where there is
evidence that such variations are sustained (e.g. across
generations) they are typically referred to as traditions
or cultural variations.
Such phenomena are of considerable theoretical
significance for evolutionary biology, because they offer
(i) a means of inheritance and adaptation much more
rapid than the genetic transmission processes on whose
shoulders they have evolved and (ii) the prospect of a
* Author for correspondence (a.whiten@st-andrews.ac.uk).
Electronic supplementary material is available at http://dx.doi.org/10.
1098/rstb.2008.0134 or via http://journals.royalsociety.org.
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
secondary form of behavioural evolution at the cultural
level ( Whiten 2005; Mesoudi et al. 2006). The animal
studies are additionally of interest in identifying the
roots of the cultural processes that are so distinctive in
our own species (Whiten in press).
However, purely observational studies of wild
populations are constrained in the inferences they can
draw about the social learning mechanisms involved.
Owing to this, a complementary corpus of experi-
mental studies has arisen, in which the role of social
learning can be robustly tested by comparing a
condition permitting observational learning with one
that offers no such opportunities. The literature based
on this kind of approach now spans over a century and
accommodates scores of studies identifying and
differentiating various forms of social learning in
different animal taxa (Galef & Heyes 2004). We think
this literature suffers a major limitation, however, in
relation to the topic of culture. Typically, researchers
examine only what a single animal learns from another:
in other words, they study only a single transmission
event. Given that by its nature, culture requires
multiple transmission episodes, examining social
learning only at the dyadic level falls far short of the
methodology that is needed. Hence our exhortation to
‘mind the gap’. There is a yawning gap between the
dyadic norm of the experimental literature and the
typical, and proper, focus of observational field studies

3477 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3478 A. Whiten & A. Mesoudi Review. Experimental science of culture
on group-level phenomena. The latter, which include
regional differences among groups and diffusion of
novel behaviour patterns through groups, are what
cultural analyses should properly be concerned with.
We propose that the best opportunity currently
available to us to bridge this gap is the cultural diffusion
(or ‘transmission’) experiment. Here, rather than
focusing on only a model–observer dyad, experi-
mentally controlled innovations in behaviour are
seeded into groups of individuals and the spread (or
otherwise) of the innovation is tracked and documen-
ted. Such approaches have been represented in the
research literature for some time, but only sparsely and
spasmodically in comparison to the dyadic design. We
advocate here that the current interest in animal culture
means that their time has come. They combine the
power of experimental control with group-level
analysis. Accordingly in §2 we offer a brief resumé of
the history of these experiments, which in turn leads to
an effort to systematize the variations in design that
have proliferated.
2. DEVELOPMENT AND EVOLUTION OF
DIFFUSION EXPERIMENTS
The first clear example of a diffusion experiment
appeared in the celebrated work of Bartlett (1932),
who studied how story narratives were either preserved
or modified as they were transmitted along a chain of
human subjects. Bartlett recognized and discussed the
relevance of this approach for investigating cultural
transmission, but his primary interest was in what the
successive transmissions told us about the nature of
memory. Over the next decade or so, Bartlett’s
pioneering methods were adopted and developed by
numerous disciples, but then the transmission experi-
ment temporarily faded from the literature. It was
rejuvenated as a tool to study cultural transmission by
Jacobs & Campbell’s (1961) explicitly titled ‘perpetu-
ation of an arbitrary tradition through several gener-
ations of a laboratory microculture’. Through the
remainder of the century other human studies steadily
built on this, but only recently has the power of such
‘laboratory microculture’ diffusion experiments with
human subjects been fully appreciated and a rapid
expansion of this literature occurred. In a companion
paper to the present one (Mesoudi & Whiten 2008) we
review this corpus of human diffusion studies from
Bartlett to the present day.
In the non-human animal (henceforth, ‘animal’)
diffusion literature, the first study most commonly
cited in the cultural transmission literature that
followed it is the work of Curio et al. (1978a,b).
These authors conditioned blackbirds to make alarm
calls in relation to novel stimuli and showed that such
responses would pass along a transmission chain of six
successive pairs of birds (A–B, B–C, C–D and so on)
without decrement, contrasting with baseline rates of
alarm calls. The authors interpreted these results as
support for a ‘cultural transmission hypothesis’.
The next controlled diffusion experiments con-
cerned foraging behaviour in pigeons and rats.
Lefebvre (1986) released pigeons trained to peck
through food covers into whole flocks of naive birds.
Phil. Trans. R. Soc. B (2008)
In this way, he showed that the piercing technique was
learned rapidly by some observing birds, contrasting
with control birds that saw no model, and in the seeded
population it continued to spread in the ensuing weeks.
Laland & Plotkin (1990, 1992, 1993) returned to the
principle of the linear diffusion chain, applying it to the
transmission of digging up pieces of hidden food
through consecutive expert–novice pairings of rats.
These studies cited the earlier experiments of Curio
et al. (1978a,b) and over the last 15 years a reasonably
thorough ‘citation genealogy’ has been built on these
foundations. However, what appears to be the first true
diffusion experiment in animals remained uncited until
recently. In this study, Menzel et al. (1972) investigated
habituation to two anxiety-inducing objects by juvenile
chimpanzees, applying a ‘replacement method’ that
started with a founder group of three chimpanzees that
avoided the novel objects. One chimpanzee was then
replaced by a naive chimpanzee and this process was
repeated through 17 consecutive trios. Between the
fourth and eighth ‘generation’ in this process, habitu-
ation occurred in some chimpanzees and gradually
became pervasive, such that later trios routinely
engaged with the objects. Menzel et al. accordingly
concluded that ‘a culture-like process was at work’.
This first study effectively underlines why the diffusion
method is indispensable for studying cultural trans-
mission; the changes identified by Menzel et al. would
never have been documented in a merely dyadic study
because they were inherently cumulative.
Building on these pioneering foundations, diffu-
sion studies have appeared with accelerating fre-
quency (nearly half in the present century). They
have now extended to a variety of species of fishes,
birds and mammals and a diverse assortment of
behavioural categories including predator avoidance,
foraging, tool use, route choice and communication,
which are of considerable potential adaptive signi-
ficance. Below we survey this growing corpus of
studies and, most importantly, offer the first system-
atization of the diversity it encompasses. In the
electronic supplementary material (table S1), we
offer detailed information about the scope of each
of these studies. Table 1 below is a succinct overview
derived directly from table S1 in the electronic
supplementary material.
3. CLASSIFYING THE EMERGING PARADIGMS
Our approach to systematizing these studies involves
two broad sets of distinctions that are constituted,
respectively, by the columns and rows of table 2. The
value of this operation is that, after a period in which this
small field has grown by the gradual accretion of a
number of individual studies, we can now start to survey
all the methodological options ‘in the round’, together
with their various limitations, pay-offs and prospects for
more informed and strategic work in future.
(a) Transmission designs
We distinguish three broad types of experimental
design that form the columns in table 2.
Phil. Trans. R. Soc. B (2008)

Table 1. Chronological table of diffusion experiments. (This table is directly derived from table S1 in the electronic supplementary material, which offers a comprehensive survey of methods
and conclusions, together with Latin names of species studied, and further evaluative comments. Des. str.Zdesign strength, following the scheme described in table 2 and explained fully in the
text, where higher numbers represent designs judged more powerful in identifying diffusion based upon social learning: those of levels 3 and above incorporate control conditions that
discriminate social from non-social learning, and are represented in italics. The column ‘cultural diffusion’ summarizes evidence for diffusion of the behaviour pattern of interest, where des.
str.Z3 or more. Numbers of transmissions within chains are shown in parentheses; ?Znumber of transmissions unknown. For detailed information see table S1 in the electronic
supplementary material.)

cultural diffusion? (no. of

study content species studied publication des. str. transmissions)

Successive replacements in trios exposed to alarming objects chimpanzees Menzel et al. (1972) 3B habituation effect stable (17)
Transmission chains, seeded with alarm calls to arbitrary object blackbirds Curio et al. (1978a,b) 3C alarm calling stable (6)

Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Whole groups exposed to novel cues to buried food; spread of baboons, vervets Cambefort (1981) 1A ?
discovery in groups documented
Three nut-cracking chimpanzees mixed with nine naive ones chimpanzees Sumita et al. (1985) 2A ?
Models pecked through paper covers, in wild and captive flocks pigeons Lefebvre (1986) 4A piercing stable over 55 days
Spread of spontaneously initiated nut-cracking recorded in group chimpanzees Hannah & McGrew (1987) 2A ?

Review. Experimental science of culture

Transmission chains seeded with models digging up carrot pieces rats Laland & Plotkin (1990) 4C digging stable (8)
Replication of 1990 study but incorporating a 24 hour delay rats Laland & Plotkin (1992) 4C stability less, with delay
Opportunity provided for group to use tools to probe for honey chimpanzees Paquette (1992) 1A ?
Rearing conciliatory stump-tailed macaques with rhesus, among rhesus de Waal & Johanowicz (1993) 5A effect stable over six-week post-
whom reconciliation is relatively rare macaques model phase
Transmission chains seeded with rats preferring different flavours rats Laland & Plotkin (1993) 7C transmission shown (8) but
fidelity variable
Nuts cracked elsewhere introduced; spread in group recorded over chimpanzees Matsuzawa (1994) and 1A ?
several years Biro et al. (2003)
Successive replacements in groups with initial flavour preferences rats Galef & Allen (1995) 5B transmission shown, with slight
waning (over 14)
Wild groups seeded with individuals trained to open specific small magpie jays (wild) Langan (1996) 7A door opening stable over 3 days,
doors to feed but not door chosen
Spread of dipping for honey with specific natural tool documented chimpanzees Tonooka et al. (1997) 1A ?
Founder shoals were seeded with preference for one of two routes guppies Laland & Williams (1997) 5B differences transmitted (7) but
waned by half
Similar to 1997 paper but more efficient alternative available guppies Laland & Williams (1998) 5B differences transmitted (7) but

A. Whiten & A. Mesoudi

waned
Groups with mixed naive and experienced fishes created guppies Reader & Laland (2000) 2A ?
Young cowbirds housed with adults singing either of two different cowbirds Freeberg (1998) and Freeberg 5A ?
songs; repeated once first cohort became adults (models) et al. (2001)
Spread of route preference: models in familiarity and experience guppies Swaney et al. (2001) 2A ?
Spread logged off using either of two routes to escape a threat guppies Brown & Laland (2002) 7A social transmission not found
Flocks exposed to models feeding on blood from mock hen chickens Cloutier et al. (2002) 4A rapid waning over three
transitions
Wild, banded birds exposed to model using novel foraging method keas (wild) Gajdon et al. (2004) 3A no significant evidence of social
transmission
Captive chimpanzees given rough leaves used medicinally in wild chimpanzees Huffman & Hirata (2004) 1A ?

3479
(Continued.)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3480 A. Whiten & A. Mesoudi Review. Experimental science of culture

(i) Open group diffusion (column A )

stable after 5 days with no model,

stable, spread across three groups
minimal evidence of social trans-
moderate fidelity but stable over

spread to half group: stable, one

Here, a behaviour of interest is introduced into a whole

techniques spread but social

group, as in the study of Lefebvre (1986) described
cultural diffusion? (no. of

above. Rather than exert experimental control over the

potential channels of social learning such as was done
learning unclear
in the work of Curio et al. (1978a), the open group

one corruption
two months
approach leaves open for investigation which other
transmissions)

individuals might attend to, learn from and possibly

corruption

stable (13)
adopt the behaviour they see (figure 1). The open

mission
stable (8)

stable (4)
group approach thus scores high in ecological validity,
reflecting a situation common in nature, where an
individual skilled in some technique is observed by
naive individuals, as in intergroup migration.
Weighing against this approach is that the results are
des. str.

likely to be more ‘messy’ than those of more

6A

7A

7A

6A
7A

6A
7C

7C
4B
constraining methods described below. At the stage
where a second and third individual adopts the seeded
behaviour, it may already have been difficult to
distinguish whether the third learned it from the first
or the second (or both), and with each new learner, the
question of who learned by observing whom may
Price & Caldwell (2007)

become difficult to disentangle.

Fragaszy et al. (2004)

Hopper et al. (2007)

Whiten et al. (2005)

Horner et al. (2006)

Whiten et al. (2007)

Stanley et al. (2008)

Bonnie et al. (2006)

Dindo et al. (2008)

(ii) Linear chain (column C )

We address this next because it represents an opposite
extreme to the open group approach. As in the study of
publication

Curio et al., each step in the diffusion is constrained to

involve just one model and one naive observer, with the
latter then becoming the model for the next in the chain,
and so on, resembling the children’s game ‘Chinese
Whispers’ or ‘Telephone’ (figure 1). This is sometimes
referred to as a ‘diffusion chain’ or ‘transmission chain’
method. It allows the experimenter to track precisely
Two groups each seeded with a different foraging technique (via video) colobus monkeys
brown capuchins

brown capuchins

what happens at each step in the diffusion process, and

guppies, playfish
species studied

identify, for example, at what point a particular level of

chimpanzees

chimpanzees

chimpanzees

chimpanzees
chimpanzees

corruption occurs, contrasting with the complex

children

interactions that may occur in an open diffusion context.

The cost of this might be thought to be a loss of the kind
of ecological validity inherent in the open group
approach; however, there are many cases in the wild
where transmission may routinely be one to one, as in
Replication of Horner et al. (2006) with appropriately modified task
Juveniles exposed to adults using either of two methods to get juice

some parent–offspring relationships. For these, the linear

Two transmission chains; opening artificial ‘fruit’ using alternative

Two groups, each seeded with arbitrary convention to obtain food

As for 2005, but transfer between groups in foraging techniques

chain design can be seen as simulating repeated

Two groups each seeded with model using tool in different way

Replication of Whiten et al. (2005), with one untrained model

intergenerational transmission, collapsing what in the

Successive replacements of fishes focused on novel food task

wild may take decades into a diffusion chain experiment

that may occupy only weeks (Horner et al. 2006).
Excluding the studies in parentheses (which ident-
methods, plus control condition lacking model

ified chains of social learning only via examining

transmission across three groups) in column C,
table 2 records the completion of only four true
linear-chain experiments in the animal literature. This
contrasts with numerous transmission chain studies in
the human literature (Mesoudi & Whiten 2008). One
reason for the paucity of such experiments in animals
may be that it is necessary to ensure that each pair in
of related species model

the chain is both comfortable with being isolated from

the remainder of their group and compatible with each
Table 1. (Continued.)

other. Experience indicates that in primates at least,

these can be very exacting requirements (Horner et al.
study content

2006; Dindo et al. 2008).

(iii) Replacement (column B )

The replacement method, such as the linear chain,
involves a systematic series of steps or ‘cultural

Phil. Trans. R. Soc. B (2008)

Phil. Trans. R. Soc. B (2008)

Table 2. Experimental designs used to study the spread of traditions. (Criteria for inclusion are that a behaviour pattern is either facilitated (row 1) or explicitly seeded, usually through training

Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

of an initial model (rows 2–7), and the spread of such patterns is subsequently documented. Experimental designs are classified according to (i) experimental versus control conditions
(condition designs: rows 1–7) and (ii) methods used to examine the spread of any traditions emerging (transmission designs: columns A–C), distinctions fully explained in the text. Studies
marked with a single asterisk showed spread within a first group in an open diffusion design (column A) then on to a second or third group, thus demonstrating a chain of transmission (column
C). For more information on each study see table 1, and table S1 in the electronic supplementary material.)

transmission designs (A–C)

Review. Experimental science of culture

condition designs (1–7) A. open group B. replacement C. linear chain

1. one group, presented with novel learning opportunities Cambefort (1981), Paquette (1992), Tonooka et al.
(1997), Biro et al. (2003) and Huffman & Hirata (2004)
2. action explicitly seeded in one group but no baseline Sumita et al. (1985), Hannah & McGrew (1987), Reader
& Laland (2000) and Swaney et al. (2001)
3. one experimental group with one trained, seeded action, Gajdon et al. (2004) Menzel et al. (1972) Curio et al. (1978a)
following no-model baseline
4. one experimental condition with one trained, seeded Lefebvre (1986) and Cloutier et al. (2002) Stanley et al. (2008) Laland & Plotkin (1990, 1992)
action, versus no-model control condition
5. two experimental conditions, with alternative actions de Waal & Johanowicz (1993), Freeberg (1998) and Galef & Allen (1995) and Laland (Freeberg (1998), Freeberg
seeded in each Freeberg et al. (2001) & Williams (1997, 1998) et al. 2001 see also Col. A)
6. two experimental conditions, with alternative actions Fragaszy et al. (2004), Price & Caldwell (2007) and (Whiten et al. (2007),
seeded in each, after baseline, no-model control period Whiten et al. (2007) see also Col. A)
7. two experimental conditions, with alternative actions Langan (1996), Brown & Laland (2002), Whiten et al. Laland & Plotkin (1993),

A. Whiten & A. Mesoudi

seeded in each, plus third, no-model control condition (2005), Bonnie et al. (2006) and Hopper et al. (2007) Horner et al. (2006) and
Dindo et al. (2008)

3481
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3482 A. Whiten & A. Mesoudi Review. Experimental science of culture

(a) (b) (c) (b) Experimental conditions designs

In the seven rows of table 2, we distinguish what we call
(i) A A (i) A (i)
‘conditions designs’ on the basis of the experimental
B B B B and control conditions applied. Each of rows 2–7
C
C C C C includes the introduction of a model, usually a trained
H D D D D D one but sometimes capitalizing on the natural emer-
E
F E E E E E gence of an innovator. Whether we have successfully
D F F F F captured all of the relevant published studies in the
G G G present paper or not, the criteria for inclusion of a study
G
H H in each of these rows appear quite clear.
(ii) A (ii) (ii) The same cannot be said of row 1, where the
A A
B B experimental approach is the most minimal, simply
B
G B C C C offering novel learning experiences in such a way that if
C
D D
an innovation occurs, its potential subsequent spread
C D D D
D can be systematically tracked. In this row we have
E E E E E
included studies that express this intent. For example,
H F F F F
E Paquette (1992) gave four chimpanzees the opportu-
F G G G
nity to use tools to dip for honey in an artificial ‘termite
H H
mound’ and recorded the emergence and spread of this
(iii) (iii) (iii) behaviour. However, it could be argued that many
G other studies of the spread of behaviour patterns, not
G B E B
B included here, share essential features with those listed
D C D C D C in row 1. For example, the original Japanese macaque
G ‘preculture’ studies documented the spread of
H H F
E behaviour patterns elicited by novel learning experi-
F F H E ences, such as washing human-supplied foodstuffs
(Kawai 1965). Studies in table 2 are differentiated
Figure 1. Three principal diffusion experiment designs. from these by the authors’ intent to conduct an
(a) Open diffusion with one model seeded in each group experiment tracing diffusion, but the evidence for
and all members free to observe, learn or not; (b) replacement social learning remains only of the weakest, circum-
method with an experienced individual replaced by a naive stantial kind, owing to a lack of any control condition
one at each step; (c) linear chain with order of any where social learning is not possible. In the Paquette
transmission determined by experimenter. (i,ii) (corre- study referred to above, for example, we cannot be sure
sponding to the designs described in row 7 of table 2) that the spread of the dipping behaviour was not simply
Seeding with a model acting in different ways (shaded); (iii) a the result of each chimpanzee developing this on its
no-model control group limited to individual learning. own account, rather than through observing those
Arrows illustrate hypothetical diffusion of information,
already dipping for honey. Our principal interest in the
beginning with seeded model ‘A’.
present paper is thus in the lower rows of the table.
Broadly, as we further descend the rows of table 2,
generations’ that are experimentally imposed, unlike in
the power of the experimental designs to identify social
the open diffusion approach. However, at each step,
learning, and in turn cultural transmission in the
one naive individual replaces one of a group of spread of the behaviour of interest, is enhanced.
experienced individuals, so that in this respect there Row 2 differs from row 1 in that a known model is
are resemblances to the open group context; the seeded, providing added focus about what behaviour
experimenter will not necessarily know from which of pattern is to be subsequently tracked. However, the
the available models the novice learns, or if it learns absence of a comparison with a control condition
from several; and again, the more experienced individ- where there is no model means that evidence for social
uals may be influenced by how the later recruits behave. learning here still remains weak (see table S1 in the
The replacement method can thus be regarded as electronic supplementary material for details).
something of an intermediate design, lying between the Row 3 is the first where we see the incorporation of a
open group and linear chain approaches in our table control condition for individual learning, in this case
(table 2 and figure 1). through an initial baseline phase of exposure to the
One important aspect of this method is that if the problem of interest, before subjects witness a model in
animals being studied are influenced by the number of the experimental phase. As in the case of the original
models they witness, being predisposed to ‘copy the ape and avian experiments of Menzel et al. and Curio
majority’ (a form of conformity, discussed further et al., respectively, where responses changed dramati-
below), then positive social learning effects might be cally between baseline and social learning conditions,
documented by a replacement approach, yet missed in compelling evidence of cultural transmission can here
a linear design that fails to sufficiently stimulate the be obtained. However, the use of a baseline as the
social learning mechanisms available. In general, the reference condition may remain weaker in some
replacement approach provides a good model of other contexts, wherein the behaviour is more likely
natural situations in which there is gradual turnover to appear through exploration the longer the period
in a group. A human example is in the present issue of exposure; this may be the case for solving novel
(Caldwell & Millen 2008). foraging problems, for example.

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Experimental science of culture
Row 4 overcomes this limitation through a between-
subjects design: an experimental condition where
seeding via a model takes place and with a separate
control condition where no such model is available.
Studies in this row thus have the power to provide
clear evidence of social learning. Of course, whether
cultural diffusion was found in each of these studies is
a different matter, and indeed the extent of cultural
spread documented varies across the studies. The
important point for now is that the method has the
capability to determine the extent of any cultural
diffusion that occurs.
Rows 5–7 list studies that expose each of two
populations to different models. This is an approach
borrowed from dyadic studies of social learning
particularly concerned to identify imitation, as opposed
to simpler forms of social learning (Heyes 1996). The
idea here was that whether subjects that observe a
model using either a behaviour pattern A or pattern B
subsequently show a significant tendency to preferen-
tially match the pattern they saw, imitation is
implicated. In the context of diffusion experiments,
this kind of discriminatory power can be contrasted
with the approach seen in row 4, where there is just one
experimental and one control condition. Where a
behaviour pattern spreads only in the experimental
condition, we have good evidence the cause is social
learning. However, the social learning could be of the
simplest kind. For example, Laland & Plotkin’s (1990)
study demonstrated in this way that carrot digging by
rats diffused socially along a chain of eight steps.
However, the rats did not necessarily learn about
digging: perhaps they learned only that there was
buried food available. By contrast, in a design of the
type shown in rows 5–7, the diffusion may start with
two different techniques to recover the food: and if
these each diffuse with significant fidelity, we know that
the social learning is sophisticated enough to involve
some degree of replication or copying of these
alternative forms of behaviour. For example, Whiten
et al. (2005) exposed groups of chimpanzees to either of
two types of tool-based foraging techniques and found
that each spread with significant fidelity in the group
they were seeded in, creating different traditions.
However, each of any such pairs of techniques may be
latent in the animals under study, and be merely elicited
through witnessing a model (contagion). Conditions 6
and 7 therefore add the refinement of controls for
individual learning—a baseline procedure in row 6 and a
between-group control in row 7, paralleling those
involved in the one-model designs of rows 3 and 4. For
example, in the study of Whiten et al. (2005) noted
above, chimpanzees exposed to the problem without
benefit of a model failed to solve it, indicating that in the
experimental conditions the different techniques spread
to become local traditions because individuals acquired
the techniques by observational learning.
In these respects, the approaches identified in rows 6
and 7 represent the most powerful designs so far
developed for the experimental investigation of cultural
transmission, and we advocate that in future they should
be adopted wherever possible. The two could even be
juxtaposed, using both baseline and between-group
Phil. Trans. R. Soc. B (2008)
A. Whiten & A. Mesoudi 3483
controls to identify the role of social learning in
particularly rigorous fashion.
Having said that, a final word in this section should
be said in support of one aspect of the approach
illustrated in row 1. It is important to remember that
cultural processes must rely on both social learning
and on innovation (Reader & Laland 2000). Only
once innovations emerge can social learning drive the
spread of new cultural variations. In table 2, the lower
rows represent the more powerful means of identifying
social learning and row 1 lists the weakest; however,
the approaches in the lower rows all depend on the
experimenter creating, through training or other
means, the initial innovation. Thus, these procedures
really focus on identifying just one ‘side’ of the culture
process, social learning. Perhaps, with ingenuity, it may
be possible in the future to combine the key elements of
row 6 or 7 with the element of spontaneous innovation
that is in play in the works listed in row 1. In the present
issue, McElreath et al. (2008), in human studies, has
offered a different approach to dealing with important,
spontaneously generated social information.
4. THE SCOPE OF DIFFUSION EXPERIMENTS
TO DATE: METHODS, TAXONOMIC COVERAGE
AND TYPES OF BEHAVIOUR STUDIED
Inspection of table 2 shows that some of the methods
distinguished remain to be exploited by more than a
handful of studies. Among the main transmission
designs distinguished in columns A–C, open diffusion
is the most common with 23 studies, whereas replace-
ment and linear chain designs account for only five and
six studies, respectively. Moreover, several cells in the
table, denoting the intersection of specific transmission
designs with specific condition designs, remain empty.
We count only 11 studies that have employed the most
powerful condition designs (rows 6 and 7).
Taxonomic coverage shows a primate focus typical
of the field of social learning: there are 17 primate
studies (12 of them on chimpanzees) but just 4 on other
mammals (all rodents) and only 7 and 6 on birds and
fishes, respectively. However, the extent of the primate
bias is a very recent phenomenon: in fact, until 2005
none of the approaches with control conditions (rows
3–7) had been extended to primates.
There is a marked homogeneity in the types of
behaviour that existing studies focus on. The early
study of Curio et al. remains the only one concerning
responses to predators, that of Menzel et al. the only
one on habituation to alarming objects. The two avian
studies by Freeberg and colleagues concern courtship,
focusing particularly on vocalizations. The bulk of the
studies—the other two dozen—all concern foraging
behaviour (including drinking behaviour and ingestion
of putative medicinal items), in nine cases through the
use of tools.
In sum, the present corpus of studies is patchy and
uneven in its coverage of methods, taxa and types of
behaviour. Nevertheless, the field has generated a
sufficient diversity of methods and findings to populate
a table already as elaborate as table 2, providing a
working map of the variety of methodological routes
that further investigations may consider following or
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3484 A. Whiten & A. Mesoudi Review. Experimental science of culture
surpassing, as well as the areas where there is still a low
density of coverage taxonomically and behaviourally.
The limited size and uneven distribution of the
existing corpus of diffusion studies make it premature
to attempt any very systematic comparative analysis of
the findings they have generated. However, some initial
generalizations and repeating themes are worth high-
lighting at this stage.
5. PRINCIPAL QUESTIONS ADDRESSED
(a) How well do traditions spread in the species
and context studied?
The primary question driving all diffusion studies is
essentially about how well the seeded behaviour
patterns do or do not spread. If they spread, are they
maintained at the levels seeded, or are they instead
degraded or corrupted in some fashion, or even lost
altogether at some point? That these basic questions
are those of the studies to date is no doubt due to
the youth of the field, and contrasts with the wider
range of questions tackled by studies with human
subjects (Mesoudi & Whiten 2008). In human studies
the existence of cultural transmission is of course
already assured, whereas for the animal studies this
remains the core issue.
Questions about the success with which seeded
behaviours are transmitted will be addressed in
different ways according to the transmission design.
In the case of linear chains, one can count the number
of transmission episodes through which the behaviour
of interest passes, at levels significantly above those of
baseline or other control conditions, and/or one can
examine the fidelity to the founder patterns as the
chain proceeds, as Flynn (2008) has done for children
elsewhere in this issue. In the case of replacement
studies, one can proceed in a similar fashion as
successive replacements are examined. In open diffu-
sion studies, however, it may be difficult or impossible
to enumerate the number of transmission events;
instead, one can consider the extent to which seeded
behaviour spreads across the groups studied and
whether behavioural mutations emerge.
In surveying the results of the studies from this
perspective, we focus on rows 3–7, where the
incorporation of controls allows relatively clear answers
to be given to questions concerning social transmission.
(i) Linear transmission chains
We address studies using this design first since it gives
the most direct answer to the question of fidelity of
transmission across cultural generations. Interestingly,
a majority of the studies of this kind have demonstrated
statistically significant diffusion relative to control
conditions along all or most of the chains. This fidelity
has also been maintained along all or most of the chain
steps tested in the studies (chains have included up to
four (Dindo et al. 2008), six (Curio et al. 1978a;
Horner et al. 2006) or eight steps (Laland & Plotkin
1990, 1992, 1993)). These studies involve an eclectic
mix of species (birds, rats and primates) and behaviour
patterns (mobbing and varied aspects of foraging). The
latter, coupled with the small numbers of studies
completed, means that no substantial comparative
Phil. Trans. R. Soc. B (2008)
conclusions are yet warranted. The focus has been to
establish whether a functional degree of fidelity of
transmission is maintained along significant chains for
the species and task examined.
(ii) Replacement ‘chains’
There are just five replacement studies completed with
fishes, rats and primates but they include some of the
higher numbers of transmission steps, in this case
counted as the number of replacements made.
However, in such cases, it is not so straightforward to
denote how many cultural generations or transmission
episodes are involved. How long it takes for a group to
completely replace its cultural ancestors depends on
the size of groups and how many are replaced at each
step. In the study of Menzel et al. (1972) for example,
17 replacements were completed, creating a series in
which six successive sets of trios, each involving
different chimpanzees, existed over time (as in the
first and the last of the initial series of individuals 123,
234, 345, 456) and the habituation to novel objects that
were built up by the replacements 4–8 (depending on
the stimulus) were maintained throughout the remain-
der of the transitions. In the rat diet study of Galef &
Allen (1995), 14 consecutive replacements likewise
generated four entire group replacements over the
course of the study and the differential dietary (flavour)
preferences of the rats were sustained, although it
waned throughout this period.
Laland & Williams (1997) likewise showed that over
seven replacement episodes, preferences of guppies to
adopt one route over another were sustained, although
they waned in their magnitude. In this case, approxi-
mately half the difference between the two experimen-
tally initiated preferences for one route over the other
eroded over this period, suggesting that such alterna-
tive traditions would no longer exist after roughly twice
this many transitions.
The extent to which animal traditions are transitory
or sustained in the long term are of paramount
theoretical significance. The answer in any one case is
likely to depend on a multitude of factors, including the
behavioural and psychological constitution of the
species, the nature of the behavioural features (as simple
as diet choice, for example, or as complex as use of a
tool set), spatio-temporal variance in the environment
and the costs and benefits of the behaviour relative to
alternative options. This has been little addressed so far,
but an illustrative, systematic attempt was made by
Laland & Williams (1998), working with guppies. These
authors compared the sustainability of traditions across
seven replacements in which fishes had a choice of two
doorways to travel through, each coupled with either a
short route to food or a route that was three times longer
and thus more costly. When the routes were short, the
founder fishes’ trained preferences for one door over the
other were strongly maintained over the seven replace-
ments, but when the routes taken were maladaptively
long, the alternative traditions steadily eroded and were
non-significant after five replacements.
(iii) Open diffusions
Open diffusion experiments provide important infor-
mation about the extent to which traditions spread across
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Experimental science of culture
potential recruits, and how much loss or corruption
occurs in relation to the founding behaviour patterns.
The studies available in this category (column A,
rows 3–7 in table 2) reveal extensive variation in these
respects. At one extreme, Whiten et al. (2007)
documented the spread of two alternative foraging
techniques first to each of two groups of chimpanzees,
and then to two further groups in each case without
corruption to the alternative technique. One technique
spread across a total of 24 individuals, with only 4
others never succeeding in mastering the task. At the
other extreme, Gajdon et al. (2004) found no evidence
of uptake and transmission of a novel foraging
technique by wild keas exposed to a founder model,
despite the trainability of this (wild) model and the
readiness of captive birds to acquire the technique.
Other studies found intermediate degrees of spread
(table 1, and table S1 in the electronic supplementary
material). The factors that determine the extent of
spread are important theoretically but remain little
studied as yet. They should be priorities for future
studies building on the foundation of the first tranches
of studies reviewed here.
(b) What are the underlying social learning
mechanisms?
Diffusion experiments are designed primarily to answer
questions about whether behavioural variations spread
and how faithfully they do so, rather than what social
learning mechanisms are responsible. The latter may be
regarded as an orthogonal, but important supplementary
question to the more basic one concerning the extent to
which cultural transmission is experimentally demon-
strated in the first place. If transmission is demonstrated,
it may in principle be due to a range of alternative
learning mechanisms that then become of interest.
At first sight, this principle may appear to conflict
with our earlier statement that the two-action
procedures described in rows 5–7 of table 2 can
discriminate some of the most basic processes that
might underlie diffusion. For example, Whiten et al.
(2005) found that in one chimpanzee group seeded
with a model that used a ‘poke’ tool-use technique to
release food trapped in a ‘pan pipes’ foraging device,
this technique spread, but it was not discovered by a
control group that saw no model. That result by itself
equates to the designs listed in row 4 of table 2. It
demonstrates diffusion due to social learning.
However, the social learning could be of the most
basic kind, in which the observer had merely learned
that a tool could be used to extract food, and they then
applied a method already within their repertoire to
achieve this. By contrast, the introduction of a model
employing a different (‘lift’) technique into a second
group, where this alternative technique spread prefer-
entially ( Whiten et al. 2005), implicates a more
structured social learning process capable of producing
copies of the poke and lift techniques. An additional
no-model control condition in which chimpanzees
performed neither technique showed further that
what the naive chimpanzees learned involved more
than merely eliciting an existing functional response.
The two-action methods of rows 5–7 thus tell us
something about the nature of the social learning in
Phil. Trans. R. Soc. B (2008)
A. Whiten & A. Mesoudi 3485
operation. Nevertheless, this is limited to a fairly crude
specification. Numerous alternatives known within the
social learning literature (Whiten et al. 2004) might be
operating and the diffusion experiment itself is largely
mute on which are involved. One approach taken by a
few investigators has been to complete a group-level
diffusion experiment and then, if cultural transmission
has been demonstrated, to use separate, dyadic
experiments to tease out the mechanisms involved. In
the case of the pan pipes study described above, ‘ghost
experiments’, in which the relevant tools and other
objects were experimentally operated without the
agency of a chimpanzee model, showed that observers
were not able to learn to emulate these physical effects,
suggesting that imitation of the actions of a model is
more likely involved in the diffusions documented
earlier (Hopper et al. 2007). In similar fashion,
Lefebvre (1986) demonstrated diffusion of pecking
through covers to access grain among pigeon flocks,
and in separate dyadic experiments Palameta &
Lefebvre (1985) showed that watching another bird
execute the piercing and feeding was significantly more
effective than observing piercing behaviour alone.
Incorporating such investigations directly into the
conduct of a diffusion experiment is more challenging
and has been attempted little to date. Perhaps the only
good example so far concerns the diffusion of food
flavour preferences demonstrated in rats by Laland &
Plotkin (1993). These authors went on to show that
diffusion was facilitated both by gustatory cues on the
rats’ breath and by excretory cues, and that these
factors can interact to produce more robust trans-
mission. We encourage further studies that experimen-
tally dissect learning mechanisms within an ongoing
diffusion in this fashion, rather than separately.
(c) Comparative and evolutionary analyses of the
content of cultural behaviour
Elsewhere we have recently offered broad-ranging
analyses of the relationships between biological and
cultural evolution (Mesoudi et al. 2006) and the
comparative scope of cultural phenomena in humans
and non-human animals ( Whiten et al. 2003; Whiten
2005). Here our comparative focus is the tighter one of
diffusion experiments.
If we focus only on transmissibility, there is
considerable comparability of findings across the fish,
bird and mammal studies reviewed here. Some of the
longest chains demonstrating fidelity of transmission
are in the fish and bird studies (see table S1 in the
electronic supplementary material for details). This
suggests that human culture, although of course vastly
more complex than anything seen in non-humans, may
have evolved from a biological base that supports the
social transmission of information in widespread ways
among vertebrates.
However, when we look more closely at the content
of what is transmitted, we note significant differences
between major taxa. The fish studies all concern the
following of a particular route, and whether this requires
social learning in the full sense can be questioned.
Although the naive fishes in these experiments are often
described as ‘observers’ and the experienced ones as
‘demonstrators’, there is no evidence that the former
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3486 A. Whiten & A. Mesoudi Review. Experimental science of culture
learn from the latter by observing what they do: rather,
the fishes have a preference to shoal together and so the
naive fishes come to learn the route they follow (by
individual learning) as they swim along with the
experienced fishes. This appears to fit what Whiten &
Ham (1992) denote as ‘social influence’, rather than
‘social learning’ in which naive individuals learn directly
from models (e.g. by observing what they do). However,
this means that the studies of Laland & Williams (1997,
1998) are interesting in showing that even such social
influence can be sufficient for the diffusion of traditions
of route choice.
Among the bird and mammal studies there is
evidence of observational learning, as when particular
foraging techniques are acquired. There appears to be
broad comparability between the experimental diffu-
sions involving birds learning to open flaps (Langan
1996), rats learning to dig up hidden food (Laland &
Plotkin 1990) and primates learning to open ‘artificial
fruits’ (Horner et al. 2006; Dindo et al. 2008).
Diffusion through such forms of observational learning
thus appears to reflect a functionally important and
fundamental capacity shared with much of human
cultural transmission (Hurley & Chater 2005).
Two ways in which the primate studies go beyond
those in other taxa appear worth remarking on so far. In
chimpanzees, these have involved extensive diffusion of
(i) different kinds of tool use (Whiten et al. 2005, 2007)
and (ii) techniques that involve hierarchically organized
sequences of different subcomponent actions (Whiten
et al. 2007). Each of these may reflect more specific
shared cognitive ancestry with humans, although the
corpus of studies available remains so small that until
more comparative studies are completed these must
remain as only tentative hypotheses.
6. CONCLUSIONS
Given the state of the field reviewed, we see a principal
contribution of the present paper as methodological,
systematizing the current corpus of diffusion studies in
the manner summarized in table 2. A majority (15) of
the 3!7 array of options we distinguish there
correspond to one or more of the small set of published
studies, although these are inevitably spread thinly
across the table, as yet.
We think the distinctions among the columns and
the rows in the array we have arrived at are of
differential significance. On the one hand, each of the
three transmission designs corresponding to the three
columns has made an important contribution to our
understanding; indeed, an ideal study can now be
seen to profitably apply all three in turn, for each
offers different and complementary information, as
has been discussed. By contrast, when we turn to the
rows in the table, we conclude that the lower ones,
particularly 6 and 7, offer greater analytical power
than those above and in general should be preferred
for future studies.
However, we note that by far the majority of the
studies so far have been completed in captivity that
limits the validity of the field as a whole. This may be
another correlate of the youth of this field, but ethology
has an illustrious history of field experiments and it is to
Phil. Trans. R. Soc. B (2008)
be hoped that as their methodological and theoretical
significance becomes better appreciated, the future will
see more diffusion experiments completed in the wild.
This will require better solutions to the practical and
logistic difficulties entailed, such as training alternative
models without observation by other group members.
Tables 1 and 2 list only three such field studies and we
note that in contrast to the overwhelming proportion of
positive diffusion results for the captive studies, two of
the three field studies found no (Gajdon et al. 2004)
or restricted (Langan 1996) diffusion from the models
so industriously introduced into the wild populations.
The existence of only three field experiments is too few
to elicit real concern over a laboratory/field mismatch,
and in any case Lefebvre (1986) found more extensive
diffusion in feral than captive pigeons, so the negative
outcomes may be the result of contextual factors that
are not yet well understood. More field experiments are
clearly needed.
Beyond the current set of demonstrations that
cultural transmission can be experimentally established
in a wide variety of species and types of behaviour, our
conclusions about similarities and differences in the
forms this takes across animal taxa must be viewed as
very tentative. Above (and in detail in table S1 in the
electronic supplementary material) we summarized the
current picture for fishes, birds, rodents and primates
as it currently appears, suggesting both that elementary
forms of cultural transmission are widespread across
this taxonomic range, and that more complex contents
and mechanisms are identifiable in the avian and
mammal studies, particularly in the primate ones where
they extend to tool use and more elaborate manip-
ulative and foraging techniques. This pattern suggests a
series of phases in the evolutionary elaboration of
cultural transmission that paved the way for human
culture. However, the principal function of this paper is
to provide a first overview of the contribution of
diffusion experiments that can guide future research in
this area in a more informed fashion. We expect
diffusion experiments to provide an increasingly
productive and robust bridge between observational
studies of animal cultures in the wild, the extensive
and well-established field of dyadic social learning
experiments and the literature on human diffusion
experiments illustrated elsewhere in this issue.
The principal research findings of A.W. reported here result
from support by BBSRC, ESRC and the Leverhulme Trust.
A.W. was supported by a Royal Society Leverhulme Trust
Senior Research Fellowship. A.M. is supported by a Mellon
Foundation postdoctoral fellowship. We are grateful to
Victoria Horner, Ludwig Huber, Stephan Lewandowsky
and Sadie Ryan for their comments on an earlier version of
the manuscript.
REFERENCES
Bartlett, F. C. 1932 Remembering. Oxford, UK: Macmillan.
Biro, D., Inoue-Nakamura, N., Tonooka, R., Yamakoshi, G.,
Sousa, C. & Matsuzawa, T. 2003 Cultural innovation and
transmission of tool use in wild chimpanzees: evidence
from field experiments. Anim. Cogn. 6, 213–223. (doi:10.
1007/s10071-003-0183-x)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Experimental science of culture
Bonnie, K. E., Horner, V., Whiten, A. & de Waal, F. B. M.
2006 Spread of arbitrary customs among chimpanzees: a
controlled experiment. Proc. R. Soc. B 274, 367–372.
(doi:10.1098/rspb.2006.3733)
Brown, C. & Laland, K. N. 2002 Social learning of a novel
avoidance task in the guppy: conformity and social release.
Anim. Behav. 64, 41–47. (doi:10.1006/anbe.2002.3021)
Caldwell, C. A. & Millen, A. E. 2008 Studying cumulative
cultural evolution in the laboratory. Phil. Trans. R. Soc. B
363, 3529–3539. (doi:10.1098/rstb.2008.0133)
Cambefort, J. P. 1981 A comparative study of culturally
transmitted patterns of feeding habits in the chacma
baboon Papio ursinus and the vervet monkey Cercopithecus
aethiops. Folia Primatol. 36, 243–263.
Cloutier, S., Newberry, R. C., Honda, K. & Alldredge, J. R.
2002 Cannibalistic behaviour spread by social learning.
Anim. Behav. 63, 1153–1162. (doi:10.1006/anbe.2002.
3017)
Curio, E., Ulrich, E. & Vieth, W. 1978a Cultural transmission
of enemy recognition: one function of avian mobbing.
Science 202, 899–901. (doi:10.1126/science.202.4370.899)
Curio, E., Ulrich, E. & Vieth, W. 1978b The adaptive
significance of avian mobbing: II. Cultural transmission of
enemy recognition in blackbirds: effectiveness and some
constraints. Z. Tierpsychol. 48, 184–202.
de Waal, F. B. M. & Johanowicz, D. L. 1993 Modification of
reconciliation behaviour through social experience: an
experiment with two macaque species. Child Dev. 64,
897–908.
Dindo, M., Thierry, B. & Whiten, A. 2008 Social diffusion of
novel foraging methods in brown capuchin monkeys
(Cebus apella). Proc. R. Soc. B 275, 187–193. (doi:10.
1098/rspb.2007.1318)
Investigating children as cultural magnets: do young children
transmit redundant information along diffusion chains?
2008. Phil. Trans. R. Soc. B 363, 3541–3551. (doi:10.
1098/rstb.2008.0136)
Fragaszy, D. M., Visalberghi, E. & Fedigan, L. M. 2004 The
complete capuchin: the biology of the genus Cebus. Cambridge,
UK: Cambridge University Press.
Freeberg, T. M. 1998 The cultural transmission of courtship
patterns in cowbirds, Molothrus ater. Anim. Behav. 56,
1063–1073. (doi:10.1006/anbe.1998.0870)
Freeberg, T. M., King, A. P. & West, M. J. 2001 Cultural
transmission of vocal traditions in cowbirds (Molothrus
ater) influences courtship patterns and mate preferences.
J. Comp. Psychol. 115, 201–211. (doi:10.1037/0735-7036.
115.2.201)
Gajdon, G., Fijn, N. & Huber, L. 2004 Testing social learing
in a wild mountain parrot, the Kea (Nestor notabilis).
Learn. Behav. 32, 62–71.
Galef Jr, B. G. & Allen, C. 1995 A new model system for
studying animal traditions. Anim. Behav. 50, 705–717.
(doi:10.1016/0003-3472(95)80131-6)
Galef Jr, B. G. & Heyes, C. M. (eds) 2004 Learn. Behav.
Special Issue on Social Learning and Imitation, whole
issue 32, 1–144.
Hannah, A. C. & McGrew, W. C. 1987 Chimpanzees using
stones to crack open oil palm nuts in Liberia. Primates 28,
31–46. (doi:10.1007/BF02382181)
Heyes, C. M. 1996 Genuine imitation? In Social learning in
animals: the roots of culture (eds C. M. Heyes & B. G. Galef),
pp. 371–389. London, UK: Academic Press.
Hopper, L. M., Spiteri, A., Lambeth, S. P., Schapiro, S. J.,
Horner, V. & Whiten, A. 2007 Experimental studies of
traditions and underlying transmission processes in
chimpanzees. Anim. Behav. 73, 1021–1032. (doi:10.
1016/j.anbehav.2006.07.016)
Horner, V., Whiten, A., Flynn, E. & de Waal, F. B. M. 2006
Faithful replication of foraging techniques along cultural
Phil. Trans. R. Soc. B (2008)
A. Whiten & A. Mesoudi 3487
transmission chains by chimpanzees and children. Proc.
Natl Acad. Sci. USA 103, 13 878–13 883. (doi:10.1073/
pnas.0606015103)
Huffman, M. A. & Hirata, S. 2004 An experimental study of
leaf swallowing in captive chimpanzees: insights into the
origin of a self-medicative behavior and the role of social
learning. Primates 45, 113–118. (doi:10.1007/s10329-
003-0065-5)
Hurley, S. & Chater, N. (eds) 2005 Perspectives on imitation:
from mirror neurons to memes. Boston, MA: MIT Press.
Itani, J. & Nishimura, A. 1973 The study of infrahuman
culture in Japan: a review. In Precultural behaviour (ed.
E. W. Menzel Jr), pp. 26–50. Basel, Switzerland: Karger.
Jacobs, R. C. & Campbell, D. T. 1961 The perpetuation of an
arbitrary tradition through several generations of a
laboratory microculture. J. Abnorm. Soc. Psychol. 62,
649–658. (doi:10.1037/h0044182)
Kawai, M. 1965 Newly-acquired pre-cultral behavior of the
natural troop of Japanese monkeys on Koshima islet.
Primates 6, 1–30. (doi:10.1007/BF01794457)
Laland, K. N. & Galef Jr, B. G. (eds) 2008 The question of
animal culture. Cambridge, MA: Harvard University Press.
Laland, K. N. & Plotkin, H. C. 1990 Social learning and
social transmission of foraging information in Norway rats
(Rattus norvegicus). Anim. Learn. Behav. 18, 246–251.
Laland, K. N. & Plotkin, H. C. 1992 Further experimental
analysis of the social learning and transmission of foraging
information among Norway rats. Behav. Proc. 27, 53–64.
(doi:10.1016/0376-6357(92)90040-K)
Laland, K. N. & Plotkin, H. C. 1993 Social transmission of
food preferences among Norway rats by marking of food
sites and by gustatory contact. Anim. Learn. Behav. 21,
35–41.
Laland, K. N. & Williams, K. 1997 Shoaling generates social
learning of foraging information in guppies. Anim. Behav.
53, 1161–1169. (doi:10.1006/anbe.1996.0318)
Laland, K. N. & Williams, K. 1998 Social transmission of
maladaptive information in the guppy. Behav. Ecol. 9,
493–499. (doi:10.1093/beheco/9.5.493)
Langan, T. A. 1996 Social learning of a novel foraging skill by
white-throated magpie-jays (Calocitta formosa, Corvidae):
a field experiment. Ethology 102, 157–166.
Lefebvre, L. 1986 Cultural diffusion of a novel food-finding
behaviour in urban pigeons: an experimental field test.
Ethology 71, 295–304.
Matsuzawa, T. 1994 Field experiments on use of stone tools
by chimpanzees in the wild. In Chimpanzee cultures (eds
R. W. Wrangham, W. C. McGrew, F. B. M. de Waal &
P. Heltne), pp. 351–370. Cambridge, MA: Harvard
University Press.
McElreath, R., Bell, A. V., Efferson, C., Lubell, M.,
Richerson, P. J. & Waring, T. 2008 Beyond existence and
aiming outside the laboratory: estimating frequency-
dependent and pay-off-biased social learning strategies.
Phil. Trans. R. Soc. B 363, 3515–3528. (doi:10.1098/rstb.
2008.0131)
McGrew, W. C. 1998 Culture in non-human primates? Ann.
Rev. Anthropol. 27, 301–328. (doi:10.1146/annurev.
anthro.27.1.301)
Menzel, E. W., Devenport, R. K. & Rogers, C. M. 1972
Proto-cultural aspects of chimpanzees’ responsiveness to
novel objects. Folia Primatol. 17, 161–170.
Mesoudi, A. & Whiten, A. 2008 The multiple roles of cultural
transmission experiments in understanding human cul-
tural evolution. Phil. Trans. R. Soc. B 363, 3489–3501.
(doi:10.1098/rstb.2008.0129)
Mesoudi, A., Whiten, A. & Laland, K. N. 2006 Towards a
unified science of cultural evolution. Behav. Brain Sci. 29,
329–383. (doi:10.1017/S0140525x06009083)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3488 A. Whiten & A. Mesoudi Review. Experimental science of culture
Palameta, B. & Lefebvre, L. 1985 The social transmission of
a food-finding technique in pigeons: what is learned?
Anim. Behav. 33, 892–896. (doi:10.1016/S0003-3472(85)
80023-3)
Paquette, D. 1992 Discovering and learning tool-use for
fishing honey by captive chimpanzees. Hum. Evol. 7,
17–30. (doi:10.1007/BF02436257)
Price, E. & Caldwell, C. A. 2007 Artificially-generated
cultural variation between two groups of captive monkeys,
Colobus guereza kikuyensis. Behav. Proc. 74, 13–20. (doi:10.
1016/j.beproc.2006.09.003)
Reader, S. M. & Laland, K. N. 2000 Diffusion of foraging
innovations in the guppy. Anim. Behav. 60, 175–180.
(doi:10.1006/anbe.2000.1450)
Stanley, E. L., Kendal, R. L., Kendal, J. R., Grounds, S. &
Laland, K. N. 2008 The effects of group size, rate of
turnover and disruption to demonstration on the stability
of foraging traditions in fishes. Anim. Behav. 75, 565–572.
(doi:10.1016/j.anbehav.2007.06.014)
Sumita, K., Kitahara-Frisch, J. & Norikoshi, K. 1985 The
acquisition of stone-tool use in captive chimpanzees.
Primates 26, 168–181. (doi:10.1007/BF02382016)
Swaney, W., Kendal, J., Capon, H., Brown, C. & Laland,
K. N. 2001 Familiarity facilitates social learning of
foraging information in the guppy. Anim. Behav. 62,
591–598. (doi:10.1006/anbe.2001.1788)
Tonooka, R., Tomonaga, M. & Matsuzawa, T. 1997
Acquisition and transmission of tool making and use
for drinking juice in a group of captive chimpanzees
Phil. Trans. R. Soc. B (2008)
(Pan troglodytes). Jpn Psychol. Res. 39, 253–265. (doi:10.
1111/1468-5884.00058)
Whiten, A. 2005 The second inheritance system of
chimpanzees and humans. Nature 437, 52–55. (doi:10.
1038/nature04023)
Whiten, A. In press. The identification of culture in
chimpanzees and other animals: from natural history to
diffusion experiments. In The question of animal culture,
(eds K. N. Laland & B. G. Galef ), Cambridge, MA:
Harvard University Press.
Whiten, A. & Ham, R. 1992 On the nature and evolution of
imitation in the animal kingdom: reappraisal of a century
of research. Adv. Stud. Behav. 21, 239–283. (doi:10.
1016/S0065-3454(08)60146-1)
Whiten, A. & van Schaik, C. P. 2007 The evolution of animal
‘cultures’ and social intelligence. Phil. Trans. R. Soc. B
362, 603–620. (doi:10.1098/rstb.2006.1998)
Whiten, A., Horner, V. & Marshall-Pescini, S. 2003 Cultural
panthropology. Evol. Anthropol. 12, 92–105. (doi:10.1002/
evan.10107)
Whiten, A., Horner, V., Litchfield, C. A. & Marshall-Pescini, S.
2004 How do apes ape? Learn. Behav. 32, 36–52.
Whiten, A., Horner, V. & de Waal, F. B. M. 2005 Conformity
to cultural norms of tool use in chimpanzees. Nature 437,
737–740. (doi:10.1038/nature04047)
Whiten, A., Spiteri, A., Horner, V., Bonnie, K. E., Lambeth,
S. P., Schapiro, S. J. & de Waal, F. B. M. 2007
Transmission of multiple traditions within and between
chimpanzee groups. Curr. Biol. 17, 1038–1043. (doi:10.
1016/j.cub.2007.05.031)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3489–3501

doi:10.1098/rstb.2008.0129
Published online 19 September 2008

Review

The multiple roles of cultural transmission

experiments in understanding human
cultural evolution
Alex Mesoudi1,* and Andrew Whiten2
1
Department of Social and Developmental Psychology, University of Cambridge,
Free School Lane, Cambridge CB2 3RQ, UK
2
Centre for Social Learning and Cognitive Evolution, School of Psychology, University of St Andrews,
St Andrews KY16 9JP, UK
In this paper, we explore how experimental studies of cultural transmission in adult humans can address
general questions regarding the ‘who, what, when and how’ of human cultural transmission, and
consequently inform a theory of human cultural evolution. Three methods are discussed. The
transmission chain method, in which information is passed along linear chains of participants, has been
used to identify content biases in cultural transmission. These concern the kind of information that is
transmitted. Several such candidate content biases have now emerged from the experimental literature.
The replacement method, in which participants in groups are gradually replaced or moved across groups,
has been used to study phenomena such as cumulative cultural evolution, cultural group selection and
cultural innovation. The closed-group method, in which participants learn in groups with no replacement,
has been used to explore issues such as who people choose to learn from and when they learn culturally as
opposed to individually. A number of the studies reviewed here have received relatively little attention
within their own disciplines, but we suggest that these, and future experimental studies of cultural
transmission that build on them, can play an important role in a broader science of cultural evolution.
Keywords: cultural evolution; cultural transmission; laboratory experiments; diffusion experiments;
social learning

1. CULTURAL TRANSMISSION: MORE persistence of socially learned information in chains or

QUESTIONS THAN ANSWERS
Cultural transmission is the process by which infor-
mation is passed from individual to individual via social
learning mechanisms such as imitation, teaching or
language. This can be contrasted with the acquisition
of information via genetic inheritance from biological
parents, and with individual learning, where there is no
influence from conspecifics. A great deal is known
about both genetic inheritance and individual learning,
in no small part through extensive laboratory experi-
ments conducted respectively by population geneticists
(Hartl & Clark 1997) and experimental psychologists
(Mackintosh 1983). Far less experimental research
has examined cultural transmission. While there has
been some experimental research into social learning
within social psychology (e.g. Bandura 1977), these
studies have usually been restricted to a single model
and a single learner, with few studies examining the
* Author and address for correspondence: School of Biological and
Chemical Sciences, Queen Mary, University of London, Mile End
Road, London E1 4NS, UK (am786@cam.ac.uk).
Electronic supplementary material is available at http://dx.doi.org/10.
1098/rstb.2008.0129 or via http://journals.royalsociety.org.
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
groups that involve larger numbers of individuals. Yet
such multiple-individual or multigenerational experi-
mental designs would appear to be essential to test
hypotheses concerning broader cultural patterns and
trends that are inherently group-level phenomena.
Encouragingly, this situation is changing, and, in the
last few years, there has been a surge of interest
in the experimental study of cultural transmission in
adults, children and non-human species. In this paper,
we review recent and past experimental studies of
cultural transmission in adult humans, complementing
related reviews concerning non-humans ( Whiten &
Mesoudi 2008) and children (Flynn 2008).
Questions regarding cultural transmission can be
broadly summarized in terms of ‘what, who, when and
how’ (following Laland 2004): what is copied? (i.e.
what kind of information is most easily remembered
and most often transmitted?); who is copied? (i.e. the
identity of the model(s) from whom information is
acquired); when do individuals copy? (e.g. is copying
more likely when the task at hand is easy or difficult, or
when the environment is constant or changing?); and
how do individuals copy? (e.g. using imitation,
emulation, or spoken or written language?). Various
experimental studies in the past several decades have
addressed all four of these types of question, and used

3489 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3490 A. Mesoudi & A. Whiten Review. Roles of cultural transmission experiments
various methods in doing so. However, perhaps due to
the sparseness of past experimental studies and the lack
of any guiding theoretical framework, these questions
and methods have not been addressed in a systematic
fashion, and answers to each must be said to be sketchy
at best. Our aim here is to make links between these
disparate studies, which have often emerged in isolated
fringes of different disciplines, such as psychology,
sociology, anthropology and economics, and draw
them to the attention of a wider audience. We also
think that a cultural evolutionary framework offers the
best prospect for such a cross-disciplinary synthesis, an
argument which is elaborated in §2.
2. CULTURAL TRANSMISSION AND
CULTURAL EVOLUTION
We believe that experimental studies of cultural trans-
mission will be most valuable if they are pursued within a
framework of cultural evolution. This body of theory
contends that human culture evolves according to basic
Darwinian principles, in important respects similar to
those by which biological species evolve (Campbell
1974; Cavalli-Sforza & Feldman 1981; Boyd &
Richerson 1985; Plotkin 1994; Mesoudi et al. 2004;
Richerson & Boyd 2005; Mesoudi et al. 2006b). These
Darwinian principles are variation, differential fitness and
inheritance, and just as Darwin (1859/1968) showed
these basic principles to characterize the evolution of
biological organisms, they can also be observed in
human culture (Mesoudi et al. 2004): (i) cultural traits
(beliefs, attitudes, skills, knowledge, etc.) vary across and
within individuals and groups; (ii) not all cultural traits
are equally likely to be preserved and copied due to
competition for expression, attention or memory space,
some ideas are more memorable or attractive than
others, and some models are more likely to be copied;
and (iii) cultural traits are inherited or transmitted from
model(s) to learner(s) via social learning.
As indicated in point (iii), cultural transmission is a
fundamental component of cultural evolution. Without
transmission there can be no evolution, and the form
that this transmission takes can significantly influence
the evolutionary dynamics of culture. As such, the
cultural evolution literature already contains definitions,
classifications and rigorous mathematical analyses of
many aspects of cultural transmission. For example,
Cavalli-Sforza & Feldman (1981) modelled vertical
(from biological parent to offspring), oblique (from
parental generation to offspring generation excluding
kin) and horizontal (within-generational) cultural trans-
mission, while Boyd & Richerson (1985) modelled
conformist transmission (preferentially copying the
most popular variant) and prestige/indirect bias (pre-
ferentially copying the cultural trait of the most
prestigious or successful member of the group). All
these analyses address ‘who’ should be copied and the
consequences of doing so. Other models have addressed
‘when’ cultural transmission should be favoured over
individual learning and/or genetic evolution (Rogers
1988; Boyd & Richerson 1995; Aoki et al. 2005),
generally concluding that cultural transmission should
be favoured when (i) environments change too rapidly
for genes to track them effectively, but not so rapidly that
Phil. Trans. R. Soc. B (2008)
the behaviour of a potential model becomes outdated,
and/or (ii) individual learning is particularly costly or
difficult. In the literature review below, we highlight
experimental studies that have addressed these distinc-
tions and findings.
A second advantage of adopting a cultural evolution-
ary approach to cultural transmission is that it
encourages links to be made between small-scale
transmission processes that can be observed in a
restricted number of individuals, as typically studied
in experiments, and population-level patterns gener-
ated by people in real-life situations over longer time
periods. This population-level thinking is inherent in
Darwinian evolutionary theory, and ever since the
evolutionary synthesis of the 1930s and 1940s (Mayr &
Provine 1980), evolutionary biologists have made links
between small-scale microevolutionary processes, such
as natural selection, sexual selection, mutation and
drift (often studied experimentally), and population-
level macroevolutionary patterns in time or space (as
studied by palaeobiologists and biogeographers), with
the latter patterns understood to be generated in part
by the former. The same population thinking can be
applied to cultural evolution (Richerson & Boyd 2005),
and placing cultural transmission within an evolutionary
framework potentially allows a similar interdisciplinary
evolutionary synthesis for the cultural sciences
(Mesoudi 2008a). Thus, the forces and biases of
cultural transmission studied experimentally in the
laboratory can be seen as at least partly generating the
population-level patterns of cultural change documen-
ted by socio-cultural anthropologists, archaeologists,
sociologists and other social scientists. This gives
cultural transmission experiments added significance:
cultural transmission should not only be studied for its
own sake (i.e. in order to better understand cultural
transmission itself ), but also in order to explain broader
cultural patterns and trends, all as part of a unified
science of cultural evolution (Mesoudi et al. 2006b).
Conversely, cultural evolution theory can benefit
greatly from more detailed empirical studies of
cultural transmission. Past cultural evolution research
has predominantly involved the analysis of formal
mathematical models (Cavalli-Sforza & Feldman
1981; Boyd & Richerson 1985), and sorely lacks
empirical studies that test the assumptions and
findings of those models. Experiments offer a means
of performing this using actual people but retaining
much of the rigour and control of mathematical
models. The value of experimental tests of theoretical
models can be seen in the field of experimental
economics, where recent experimental findings that
conflict with prior theoretical predictions (e.g. the
ultimatum game, for which people universally exhibit
‘non-rational’, i.e. non-self-interested, behaviour;
Henrich et al. 2005) have forced a productive
reconsideration of theoretical assumptions. Below,
we note several similar cases in which participants
deviate significantly from theoretically derived pre-
dictions, which may force a similarly productive
re-examination of the theoretical assumptions of
some cultural evolutionary models.
The following sections briefly outline experimental
studies concerning cultural transmission in adult
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Roles of cultural transmission experiments
humans, focusing on their implications for the field of
cultural evolution. To count as a study of cultural
transmission, there must be some kind of transmission
of information (knowledge or behaviour) along a chain
or within a group of more than two participants. The
studies are categorized according to their methodology;
we discuss in turn the linear transmission chain
method, the replacement method and the closed-
group method. A more detailed literature review
using the same classification can be found in Mesoudi
(2007), and we direct readers interested in fuller
descriptions of the studies mentioned here to consult
that publication. For further reference, table S1 of the
electronic supplementary material provides a summary
of all adult human cultural transmission studies that we
are aware of, listing for each one the methodology used,
the participant sample, the material/behaviour that was
transmitted and the study authors’ main conclusions.
3. THE LINEAR TRANSMISSION CHAIN METHOD
The linear transmission (or diffusion) chain method
represents perhaps the simplest experimental
procedure for studying cultural transmission. Devised
by Bartlett (1932), this method resembles the children’s
game ‘Chinese whispers’ or ‘Telephone’, wherein some
material relevant to a particular hypothesis is passed
along linear chains of participants (figure 1). The first
participant in the chain reads or hears some material
(typically text or pictures), and then attempts to recall
it. This recalled information is given to the second
participant, who reads it and later recalls it in a similar
way; this recall is passed on to the third participant, and
so on along the chain. By measuring the changes that
occur within the material as it is passed along the
chain, or by comparing the rates at which different
kinds of material degrades, the researcher can infer the
operation of systematic biases in cultural transmission.
Bartlett (1932) conducted a series of transmission
chain studies using various types of material, from
Native American folk tales to descriptions of sporting
events. As transmission proceeded along the chains,
Bartlett (1932) noted that the material became much
shorter in length and lost many of the details, with
only the overall gist remaining. Participants also tended
to distort the material, making it more coherent
and consistent with their own pre-existing knowledge.
The folk tales from non-industrial societies, for
example, contained many supernatural elements that
were nonsensical to the English participants and were
subsequently removed or replaced with more familiar
events. These two processes, loss of detail and
assimilation to prior knowledge, led Bartlett (1932) to
propose that remembering is primarily a reconstructive
process, and seldom a process of exact replication.
Only the gist or overall impression of the material is
preserved and rebuilt around pre-existing knowledge
structures or schemas. Accordingly, Bartlett (1932)
found that folk stories were transmitted with greater
accuracy than any of the other material, which he
argued was because people already possess story
schemas that contain the structure of a typical folk
tale, thus aiding recall.
Phil. Trans. R. Soc. B (2008)
A. Mesoudi & A. Whiten 3491
generation
1 2 3 4
A
original B
chain
material
C
D
Figure 1. Design of a typical transmission chain study. The
original material is passed along parallel chains of participants
(represented by circles). Here, there are four chains (A–D),
each comprising four generations (1–4). Adapted from
Mesoudi (2007).
The two decades following Bartlett’s (1932) original
study saw the publication of several transmission chain
studies that shared Bartlett’s general methodology but
varied in the material used and participants tested
(Maxwell 1936; Northway 1936; Allport & Postman
1947; Ward 1949; Hall 1951). The results of these
studies largely supported Bartlett’s original findings of
increasing generalization and assimilation to pre-existing
knowledge. Although the later twentieth century saw a
decline in the popularity of the transmission chain
method, several recent studies have sought to reintro-
duce the method as a means of studying cultural change,
and have updated the transmission chain method to
conform to modern standards of experimental psychol-
ogy (Bangerter 2000; Kashima 2000; Barrett & Nyhof
2001; Mesoudi & Whiten 2004; Mesoudi et al. 2006a;
Kalish et al. 2007; Griffiths et al. 2008; see Mesoudi
2007). These recent studies, too, support Bartlett’s
(1932) conclusions. For example, Mesoudi & Whiten
(2004) confirmed and updated Bartlett’s (1932) notion
of ‘generalization’ by drawing on script theories from
cognitive psychology, finding that descriptions of every-
day events were described at increasingly abstract levels
of a hierarchically organized knowledge structure as they
were passed along transmission chains. Other studies
have supported Bartlett’s claim of assimilation to
previous knowledge, finding that transmitted infor-
mation gradually converges upon pre-existing gender
stereotypes (Bangerter 2000; Kashima 2000) and prior
cognitive biases (Kalish et al. 2007; Griffiths et al. 2008;
see Griffiths et al. 2008).
How can the transmission chain method, and the
findings of transmission chain studies, inform research
into cultural evolution? The transmission chain
method, as it has been used predominantly to date,
seems most suited to identifying what Richerson &
Boyd (2005) have called ‘content-based’ or ‘direct’
biases, in which transmission is determined by the
content of the information being transmitted (i.e.
‘what’ is transmitted). However, content-based biases
have received relatively little attention from mathe-
matical modellers such as Cavalli-Sforza & Feldman
(1981) and Boyd & Richerson (1985), who focus more
on model-based biases (‘who’ is copied; see §§5 and
6c). Content-based biases have received much more
attention from cognitively minded anthropologists
such as Boyer (1994), Sperber (1996, 2000)
and Atran (1998, 2001). Content-biased cultural
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3492 A. Mesoudi & A. Whiten Review. Roles of cultural transmission experiments
transmission resembles what Sperber (1996) has called
‘cultural attraction’, where culturally acquired rep-
resentations are transformed or distorted to become
more similar to a particular form, or ‘attractor’, that is
favoured by pre-existing cognitive biases (which are
often argued to be genetically specified products of
natural selection). The findings from the transmission
chain experiments that cultural transmission is recon-
structive strongly support Sperber’s (1996) argument
that content biases will readily operate to distort
cultural information in particular directions. To give
a specific example, Barrett & Nyhof (2001) found
that descriptions of living things, physical objects
and intentional agents that are ‘minimally counter-
intuitive’, i.e. contain a small number of features that
violate some common intuitions of folk biology, folk
physics and folk psychology, were passed along
transmission chains with significantly higher fidelity
than items that were either intuitive (did not violate folk
knowledge) or bizarre (were highly unusual but did not
violate folk knowledge). In another study, Mesoudi
et al. (2006a) found that information concerning third-
party social interactions was transmitted with higher
fidelity than equivalent non-social information, in line
with the hypothesis that primate intelligence evolved
particularly to solve complex social problems (Byrne &
Whiten 1988; Dunbar 2003), suggesting the operation
of a ‘social bias’ in cultural transmission. When we add
the counter-intuitive bias (Barrett & Nyhof 2001) and
the social bias (Mesoudi et al. 2006a) to the hierarchical
bias (Mesoudi & Whiten 2004) and gender-stereotype
bias (Bangerter 2000; Kashima 2000) noted earlier, we
can begin to see a provisional list of content biases
emerging from the experimental literature.
In §2 we noted that cultural evolutionary population
thinking encourages the extrapolation of individual-
level biases to explain population-level patterns in
actual cultural datasets. Sperber & Hirschfeld (2004)
have attempted just this for some of the cognitive biases
noted above. They argue that certain patterns of
human cultural diversity and stability can be explained
by cultural attraction towards the domains of pre-
existing cognitive modules. For example, the rich and
similarly structured ecological knowledge shown by a
large number of otherwise dissimilar hunter-gatherer
societies worldwide can be explained by the operation
of a universal folk-biology module, which favours the
acquisition of similarly structured biological knowledge
(Atran 1998). Cultural diversity, meanwhile, can be
explained in part because the proper domain of a
cognitive module (the domain it evolved to deal with,
e.g. for a face recognition module, human faces) may
not always correspond to its actual domain (the set of
environmental stimuli that activate the module, e.g.
diverse masks, caricatures and portraits) due to errors
in perception or exploitation by others. Finally, super-
natural concepts may spread because they activate
more than one domain. For example, ghosts have
human-like intentions (a folk psychology module) but
in being able to pass through solid objects violate
another (a folk-physics module). The aforementioned
study by Barrett & Nyhof (2001) supports this claim,
with population-level consequences of this bias seen in
the widespread popularity of supernatural or religious
Phil. Trans. R. Soc. B (2008)
beliefs across the world (Boyer 1994) and the
persistence of minimally counter-intuitive folk tales
through history (Norenzayan et al. 2006).
What of the other experimental findings noted
above? A cognitive hierarchy bias (Bartlett 1932;
Mesoudi & Whiten 2004) might lead to the prediction
that information that has persisted for many gener-
ations should have a gist-like form that can easily be
reconstructed. Accordingly, Rubin (1995) showed that
many orally transmitted folk tales have been preserved
over many generations precisely owing to their abstract,
schema-like content. However, this should be qualified
with Barrett & Nyhof ’s (2001) finding that minimally
counter-intuitive items, which by definition do not
conform to a generalized schema, are favoured during
transmission. Perhaps these two findings are not so
contradictory, however: a counter-intuitive belief
cannot spread unless people already possess the folk
schemas that it violates, making these two biases
mutually reinforcing. Moreover, Norenzayan et al.
(2006) found that too many counter-intuitive elements
decrease the memorability of narratives, suggesting a
trade-off between counter-intuitive and schematic
properties. Operation of the gender-stereotype bias
(Kashima 2000; Bangerter 2000) might be observed in
everyday language, which tends to contain more male-
favourable terms than female-favourable terms (e.g.
‘chairman’), possible evidence of gender stereotypes
influencing cultural transmission of grammar and
vocabulary ( Lakoff 1975). Finally, a social bias
(Mesoudi et al. 2006a) might be partially responsible
for the fact that socially oriented magazines and
newspapers tend to have circulations orders of magni-
tude higher than non-social or factual publications
(A. Mesoudi 2005, unpublished PhD thesis). Some of
these claims remain quite tentative, however, and there is
much opportunity here to more formally link small-scale
cultural transmission experiments with actual cultural
datasets from sociology and anthropology.
The experimental finding that cultural transmission
resembles reconstruction rather than replication has
been used by some (e.g. Sperber 2000; Atran 2001) to
argue against memetic models of cultural change, in
which cultural evolution proceeds through the differ-
ential selection of high-fidelity cultural replicators or
memes (Blackmore 1999). However, while this criticism
may be valid when directed towards certain versions of
memetics, the broader cultural evolution literature has
long recognized that cultural transmission can be
imperfect, vulnerable to distortion by content biases,
and based on continuous rather than discrete (meme-
like) traits (Cavalli-Sforza & Feldman 1981; Boyd &
Richerson 1985). Models that make these assumptions
are just as useful as models that assume high-fidelity
particulate inheritance (Henrich & Boyd 2002). Simi-
larly, while certain patterns of cultural variation might
be explained by the operation of cognitive attractors, as
argued by Sperber & Hirschfeld (2004), this should not
preclude the possibility that cultural variation can be
influenced by other cultural transmission biases too (e.g.
conformity, see §5), as acknowledged by Claidiere &
Sperber (2007). Or perhaps both model-based and
content-based biases operate simultaneously but at
different levels: for example, content biases might favour
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Roles of cultural transmission experiments
the transmission of minimally counter-intuitive con-
cepts in general, but which specific minimally counter-
intuitive concept a person adopts is determined by
model-based biases such as conformity.
The final study we discuss in this section used the
transmission chain method to address not what people
copy but how they copy, and comes not from psychology
or anthropology but from experimental economics.
Schotter & Sopher (2003) had successive pairs of
participants play the ‘Battle of the Sexes’ game, in
which two players must choose one of two options with
no communication. If the players choose different
options, then neither player gets any pay-off; if both
players choose the same option, then they both get a
pay-off. This rule encourages cooperation. However, the
two options differ in their pay-offs to the two players: if
both players choose the first option, then player 1 gets a
larger pay-off than player 2; and if both players choose
the second option, then player 2 gets a larger pay-off.
This rule encourages competition. Two modes of
transmission between successive generations were
allowed: either (i) a behavioural history of the choices
(option 1 or 2) made by pairs of players in every previous
generation and their associated pay-offs, or (ii) explicit
verbal advice given by the previous generation as to
which option the present generation should choose and
why. Verbal advice was found to generate stable
conventions, i.e. long periods during which both players
agreed on which option to choose, punctuated with brief
periods of rapid change. Viewing behavioural history
without verbal advice, on the other hand, did not
generate stable conventions, resulting instead in
continuous fluctuation. This study nicely demonstrates
how the transmission chain method can be used to test
the effect of different transmission mechanisms and that
these mechanisms can have striking effects on the rate
and form of cultural change.
4. THE REPLACEMENT METHOD
The replacement method, originally proposed by Gerard
et al. (1956), involves groups of participants repeatedly
engaging in a task or game that is designed to capture
some aspect of actual cultural change. One by one, the
participants in the groups are replaced with new
participants, with each replacement representing a single
‘cultural generation’ (figure 2). Researchers can then
examine how group performance changes over succes-
sive generations, and how the socialization of each new
participant into the group affects this change. In some
replacement studies, a norm or bias is artificially
introduced into the first generation of participants,
either by explicitly training the participants to follow
this norm or by using confederates to introduce the norm
surreptitiously. The extent to which this artificially
introduced norm remains in the group during successive
generations then represents a measure of its transmission
to the new members. Generally, the replacement method
is useful for simulating cultural change that occurs with
changing group membership, as is found, for example, in
business organizations with frequent staff turnover or
traditional hunter-gatherer societies in which small
groups maintain stable traditions despite continual
population replacement via births, deaths and migration.
Phil. Trans. R. Soc. B (2008)
A. Mesoudi & A. Whiten 3493
generation
1 2 3 4
A B C D
B C D E
C D E F
D E F G
Figure 2. Design of a typical replacement study. Four
participants (A–D) engage in a learning task, and in each
generation one member of the group is replaced with a new
participant. Adapted from Mesoudi (2007).
As an illustrative example, Jacobs & Campbell
(1961) used the replacement method to study the
conformist transmission of artificially exaggerated
judgements of an ambiguous perceptual illusion. In
an earlier study by Sherif (1936), participants res-
ponded to a perceptual illusion in which a stationary
point of light in an otherwise pitch-black room is
perceived as constantly moving by a few centimetres.
The participants were asked to publicly estimate the
distance which the light moved after several other
participants, actually confederates of the experimenter,
had given unrealistically exaggerated judgements.
Sherif ’s (1936) now-classic finding was that the
majority of participants gave similar estimates to the
confederates despite that estimate being patently false,
illustrating the powerful effect of conformity in group
settings. Jacobs & Campbell (1961) repeated Sherif’s
(1936) experiment with the additional step that, after
the group had made their estimates, one group member
was replaced with a new naive participant and the
new group estimated again. Significant evidence of
the artificially introduced norm remained for about
four or five generations following the replacement of all
of the confederates, after which the perceptual
judgement tended to return to that exhibited by naive
control groups. This finding indicates some degree of
conformist transmission but no long-term persistence.
Several other studies have used the replacement
method with various tasks and tested various
hypotheses (Rose & Felton 1955; Zucker 1977; Insko
et al. 1980, 1983; Baum et al. 2004; Caldwell & Millen
2008a; see Mesoudi 2007). Here, we highlight the
implications that these studies have had or potentially
could have on three areas of cultural evolution research
in particular: cultural group selection; cumulative
cultural evolution; and cultural innovation.
Cultural group selection has been proposed by
Richerson & Boyd (2005) to explain the widespread
non-kin and non-reciprocal altruism that is observed in
human societies. This theory holds that, during human
evolutionary history, more-cooperative and more-
cohesive groups tied together by conformity and
policed by the punishment of non-cooperators would
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3494 A. Mesoudi & A. Whiten Review. Roles of cultural transmission experiments
have out-competed less-cooperative and less-cohesive
groups, resulting in the evolution of ‘tribal social
instincts’, which motivate cooperation with ingroup
members and hostility towards outgroup members.
Although they did not directly address this theory, two
replacement studies lend support to this cultural group
selection hypothesis. First, Zucker (1977) repeated
Jacobs & Campbell’s (1961) study but with the
addition that participants were given instructions
emphasizing membership of an institution or organi-
zation, and found that transmission of the arbitrary
norm significantly increased in fidelity. This suggests
that conformist transmission is particularly effective
when it operates explicitly within groups, possibly
indicating evidence of the aforementioned tribal social
instincts. Second, Insko et al. (1983) used the replace-
ment method to simulate between- and within-group
cooperation in the trading of goods. Groups of partici-
pants were taught to produce different types of paper
models, with pay-offs increased when paper models
from different groups were combined. In a ‘voluntar-
istic’ condition, groups could voluntarily trade their
goods. In a ‘coercive’ condition, one group could
forcibly confiscate the goods produced by other groups.
Periodically, one member of each group was replaced,
in order to simulate the continual group turnover of
actual societies. It was found that voluntaristic societies
were significantly more productive and earned signi-
ficantly more money than coercive societies, due to
sabotages, strikes and slowdowns in the latter.
Although Insko et al. (1983) did not explicitly frame
their study as a simulation of cultural group selection,
we might infer from their results that societies
composed of mutually cooperative subgroups would
have out-competed more competitive, less-cohesive
societies, potentially favouring the spread of coopera-
tive norms via cultural group selection.
Future studies might explicitly test cultural group
selection theories, perhaps by allowing groups to
compete more directly and allowing unsuccessful
groups to go extinct either by removal from the
experiment or by switching to a different group norm.
This may require the modification of the replacement
method along the lines of a recent study conducted by
Gurerk et al. (2006), in which participants playing a
public goods game could choose whether to participate
in a sanctioning society, in which free-riders could be
punished, or a non-sanctioning society, in which
punishment was not possible. By the end of the
experiment, virtually every participant had migrated
to the sanctioning society, providing experimental
support for the theoretical finding that moralistic
punishment is one way of facilitating the cultural
group selection of cooperative norms (Boyd et al.
2003). An important point to note from Gurerk et al.’s
(2006) study is that initially only approximately one-
third of the participants chose the sanctioning societies,
indicating an a priori aversion (or at best indifference)
to the use of punishment. Despite this initial pre-
ference, eventually, all participants migrated to the
sanctioning societies. This initial variability and
subsequent flexibility in participant behaviour suggests
that cooperative norms for strong reciprocity may not
be genetically hard-wired ‘instincts’ as sometimes
Phil. Trans. R. Soc. B (2008)
suggested; rather, people are diverse and flexible in
their behaviour, and cooperative group norms may be
an entirely cultural invention (given broad, genetically
specified capacities for social learning, individual
recognition, etc.).
Cumulative culture (Boyd & Richerson 1996;
Tomasello 1999; Caldwell & Millen 2008b) describes
the capacity to accumulate cultural innovations in
successive generations, with each new generation
learning from and adding to the previous generations’
cultural knowledge. While many species exhibit regional
differences in behaviour that appear to be attributable
to cultural transmission (Whiten et al. 1999), these
behaviours, such as nut-cracking or termite-fishing in
chimpanzees, do not appear to be the product of
cumulative culture (Tomasello 1999). This contrasts
with the products of much human culture, such as
computers or quantum physics, that have accumulated
over multiple generations and could not plausibly have
been invented by a single individual in a single lifetime.
Several replacement studies have found that the
performance on the prescribed task improved over
generations, plausibly indicative of cumulative cultural
evolution, where each new participant acquires the
existing group customs and successively improves these
customs. For example, Insko et al. (1980, 1983) found
that the voluntaristic groups of traders increased their
productivity and earnings during successive replace-
ments due to the emergence and intergenerational
transmission of increasingly efficient trading tactics
(e.g. soft bargaining: giving more than is received) and
division of labour (e.g. seniority rules for leadership,
where the longest serving member took charge). Baum
et al. (2004) found that replacement groups faced with a
choice of solving anagrams that gave either small,
immediate pay-offs or larger, delayed pay-offs gradually
converged on the optimal choice. This was due to the
emergence of intergenerational traditions in which
existing group members encouraged new members to
choose the optimal choice by transmitting accurate or
inaccurate information about pay-offs. Interestingly, this
echoes Schotter & Sopher’s (2003) finding that explicit
advice is particularly effective at maintaining optimal
behaviour. Finally, Caldwell & Millen 2008a found that
replacement groups constructed increasingly effective
artefacts across successive generations: paper aeroplanes
and spaghetti towers that were constructed by later
generations flew significantly further or were signi-
ficantly taller, respectively, than aeroplanes and towers
constructed by earlier generations, suggesting the
preservation and accumulation of increasingly effective
manufacturing techniques.
A note of caution, however, is that none of these
studies included an individual learning control con-
dition in which a single individual engaged in the same
task for the same amount of time or trials as the
replacement chains (see Whiten & Mesoudi (2008) for
further discussion on the use of, and need for, control
conditions in diffusion experiments). Without this
control condition it is difficult to conclude with
certainty that these experiments have demonstrated
true cumulative culture, in which a society accumulates
a cultural trait that could not have been invented by
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Roles of cultural transmission experiments
a single individual alone; this remains a challenge for
future studies.
An issue that has been seldom addressed by the
cultural evolution literature is that of innovation, or the
emergence and spread of novel cultural traits. In an
early study, sociologists Rose & Felton (1955) used a
modified form of the replacement method to ask under
what conditions cultural innovation is likely to occur.
Groups of participants discussed their interpretations
of Rorschach ink blots, and over successive generations
participants were systematically swapped across groups
in order to see how rates of cultural innovation and
transmission (in this case, of ink-blot interpretations)
were affected by different forms of migration/replace-
ment. The somewhat surprising result was that closed
societies with no participant migration were signi-
ficantly more innovative in generating novel interpre-
tations than open societies in which members
frequently switched groups. With hindsight, this result
is somewhat intuitive: migrants into a new group could
simply repeat the interpretations that they generated in
previous groups, whereas the participants in closed
groups were forced to come up with novel interpre-
tations. However, as Rose & Felton (1955) noted, this
finding contradicts the commonly held notion that
cosmopolitan societies with many immigrants (e.g.
large cities such as New York or London) are more
creative/innovative than closed societies that prohibit
migration (e.g. the Amish). Although different experi-
mental findings might be obtained with functional
rather than subjective/arbitrary cultural traits, Rose &
Felton’s (1955) study shows how experiments can be
useful in challenging intuitive beliefs concerning
cultural processes, and points to how the replacement
method might be used to explore the effect of migration
on cultural phenomena such as innovation.
5. THE CLOSED-GROUP METHOD
The closed-group (or constant-group) method involves
simulating cultural transmission within small groups
of participants with no replacement of members.
Individuals within a group repeatedly engage in a task
or game over the course of the experiment, and the
experimenter can manipulate the opportunities for
cultural transmission (i.e. who can view and copy other
participants’ behaviour and when) within the group
(figure 3). This method is useful for simulating under
controlled conditions the various cultural transmission
biases modelled in the cultural evolution literature
concerning ‘who’ people copy, such as conformity or
prestige bias, as well as testing cultural evolutionary
hypotheses regarding the conditions under which
cultural transmission is predicted to be employed
relative to individual learning (‘when’ questions).
Consequently, closed-group experiments typically
employ an individual learning control condition in
which participants engage in the same task as the
participants in groups, but with no social interaction.
In practical terms, the closed-group method requires
fewer participants and is less time consuming than the
replacement method, which requires a steady stream of
new participants to introduce into the groups. Conse-
quently, several closed-group studies have appeared
Phil. Trans. R. Soc. B (2008)
A. Mesoudi & A. Whiten 3495
generation
1 2 3
A B A B A B
social
learning
C D C D C D
A B A B A B
individual
learning
C D C D C D
Figure 3. Design of a typical closed-group study. In the social
learning condition, four participants (A–D) repeatedly engage
in a learning task. Arrows indicate the flow of information via
social learning, e.g. in generation 1, A learns from C, B learns
from A and C, and C and D learn from each other. In
generations 2 and 3, A, C and D all learn from B, who might
have been recognized (or manipulated) to be particularly
successful or prestigious. In the individual learning control
condition, four participants engage in the same task but
with no social interaction. Adapted from Mesoudi (2007).
in the last few years (Kameda & Nakanishi 2002,
2003; McElreath et al. 2005; Efferson et al. 2007,
2008; Mesoudi & O’Brien 2008; Mesoudi 2008b;
see Mesoudi 2007).
Unlike many of the transmission chain and replace-
ment method studies, these closed-group studies have
often been explicitly designed to test the assumptions
and findings of existing theoretical models of cultural
evolution. Accordingly, it is easier to draw direct links
between experiments and models (indeed, many of these
studies present both theoretical models and experiments
in the same paper). For example, Kameda & Nakanishi
(2002, 2003) explored experimentally the conditions
under which cultural learning is adaptive relative to
individual learning. A previous theoretical model
(Rogers 1988) suggested that the reason that culture is
adaptive is not, contrary to popular belief, that cultural
learning helps to avoid the costs of individual learning.
This is because in a population of cultural and individual
learners, the cultural learners become free-riding ‘infor-
mation scroungers’ who copy adaptive behaviour from
individual learners (‘information producers’) without
paying the associated costs of individual learning. If
the frequency of cultural learners becomes too high,
however, then there are not enough individual learners
to effectively track environmental change. Thus, cultural
learners copy outdated, maladaptive behaviour from
each other, such that cultural learners decrease in
frequency and individual learners increase in frequency.
Kameda & Nakanishi (2002) tested these predictions
experimentally. Participants in groups had to choose one
of two locations to search for a rabbit, one of which was
correct, using either individual or cultural learning. The
results confirmed that groups of learners do indeed
divide themselves into cultural learners (information
scroungers) and individual learners (information pro-
ducers) and that both types coexist at equilibrium. The
theoretical prediction that cultural learning should be
more common when individual learning is costly
(Boyd & Richerson 1995) was also supported: increas-
ing the cost of individual learning increased the
proportion of cultural learners. Finally, the experiment
revealed that this equilibrium was polymorphic, i.e.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3496 A. Mesoudi & A. Whiten Review. Roles of cultural transmission experiments
a proportion p of participants always learned individu-
ally and a proportion 1Kp always learned culturally,
rather than monomorphic, i.e. all participants learn
individually with a fixed probability p and culturally with
a fixed probability 1Kp, a distinction that could not be
made using theoretical models alone.
A follow-up study by Kameda & Nakanishi (2003),
by contrast, found a mismatch between the predictions
of theoretical models and experimental results. Using
the same task as before, Kameda & Nakanishi (2003)
found that, against the prediction of Rogers’ (1988)
model, groups in which cultural learning was permitted
significantly outperformed groups of pure individual
learners, despite the presumed detrimental effect of
information scroungers in the former. Further analyses
suggested that the cultural group did not divide into
fixed individual learners (who always engaged in
individual learning) and fixed cultural learners (who
always engaged in cultural learning) as assumed in
Rogers’ (1988) model. Rather, the participants flexibly
switched between individual learning (when individual
learning was accurate) and cultural learning (when
individual learning was inaccurate). (A similar flexible
learning strategy was observed by Mesoudi (2008b)
using a different task, and similarly enhanced fitness
relative to individual learning controls.) Kameda &
Nakanishi (2003) presented theoretical models which
confirmed that flexible cultural learners do indeed
outperform the fixed cultural learners of Rogers’
(1988) model (see also Boyd & Richerson 1995).
Kameda & Nakanishi’s (2002, 2003) work is a good
example of how experiments and models can be most
effective when combined: experiments can be used to
test the predictions of models, and where a mismatch is
found, further models can then be used to explore the
reasons for this mismatch, which are then subject to
further experimental tests, and so on.
Other studies have addressed ‘who’ is copied, or
what Richerson & Boyd (2005) have called model-
based and frequency-based biases. Several studies have
focused on conformity (disproportionately adopting
the most common trait in a population), following
cultural evolutionary models which suggest that
conformity is adaptive under a wide range of conditions
(Boyd & Richerson 1985; Henrich & Boyd 1998).
McElreath et al. (2005) asked participants to select one
of two crops to plant, one of which gave a higher pay-off
than the other. The participants could view the choices
of either one randomly selected group member
(allowing simple cultural learning) or all other group
members (potentially allowing conformity). Substan-
tial individual variation in learning strategies was
found, with a sizeable proportion of participants not
engaging in cultural learning, even where models
suggested cultural learning would have given higher
pay-offs. Of those who did copy, conformity was only
used when the environment (i.e. which crop was
optimal) changed, despite models suggesting that
conformity is the most adaptive strategy under all
conditions. A further study (Efferson et al. 2008) using
a similar task also found individual variation in the use
of social information: while the majority (70%) of
participants who could potentially use social infor-
mation engaged in conformity, resulting in significantly
Phil. Trans. R. Soc. B (2008)
higher earnings in line with theoretical expectations, a
substantial minority (30%) did not, instead ignoring
information about the behaviours’ frequency in the
group. Finally, Efferson et al. (2007) conducted a
similar experiment with Bolivian subsistence pastoral-
ists. One group of participants was shown the choice
of the player who received the highest pay-off in the
previous round (allowing a ‘copy-best’ strategy), while
another group was shown the choices of all players
from the previous round (potentially allowing con-
formity). Although the latter group outperformed the
copy-best group and individual controls, analyses of
the participants’ learning strategies indicated that
neither group of cultural learners actually used the
social information that was presented to them, given
that their learning strategies were indistinguishable
from those of the individual controls. Efferson et al.
(2007) suggested that social facilitation (improved
performance due to the mere presence of conspecifics)
may have contributed to the better performance of the
total distribution group. In general, experimental
studies of conformity (McElreath et al. 2005; Efferson
et al. 2007, 2008) have found that while many
participants do conform (and receive higher earnings
for doing so), there is often considerable individual
variation in participants’ use of social information,
such that sizeable numbers of participants fail to
engage in conformity despite theoretical models
showing it to be the optimal learning strategy.
A set of studies conducted by the first author have
examined the cultural learning strategy of copying the
most successful individual in a group (Mesoudi &
O’Brien 2008, 2008; Mesoudi 2008b). Mesoudi &
O’Brien (2008) used the closed-group method to
experimentally simulate the cultural transmission of
arrowhead designs, in order to test a specific hypothesis
concerning actual arrowhead variation in the archae-
ological record (Bettinger & Eerkens 1999). Participants
designed their own ‘virtual arrowheads’ and received
pay-offs partly determined by their designs. Arrowhead
designs could be improved either by trial-and-error
individual learning or by copying the most successful
fellow group member. As predicted, periods of individ-
ual learning resulted in increasingly diverse sets of
arrowhead designs as participants followed their own
idiosyncratic learning strategies, while periods during
which participants could engage in copy-successful-
individuals cultural learning resulted in more uniform
arrowhead designs, as participants converge on the
design of the most successful player. These patterns of
variation match corresponding patterns of arrowhead
diversity observed in the prehistoric Great Basin
(Bettinger & Eerkens 1999): high diversity in prehistoric
California, indicative of individual learning, and low
diversity in prehistoric Nevada, indicative of copy-
successful-individuals learning.
As well as simply recreating past patterns of cultural
transmission, however, experiments can also be used to
determine the adaptiveness of different learning
strategies and systematically manipulate variables of
interest, both of which are extremely difficult with
historical data alone. Mesoudi & O’Brien (2008) found
the copy-successful-individuals bias to be significantly
more adaptive than individual learning, especially
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Roles of cultural transmission experiments
when individual learning was costly, consistent with
previous theoretical models (Boyd & Richerson 1995).
Given that the environment in prehistoric Nevada is
thought to have been harsher than the prehistoric
Californian environment, this provides a potential
explanation for differences between the two regions in
learning strategies. Importantly, however, all of these
conclusions are crucially dependent on the shape of the
fitness functions underlying pay-offs of different arrow-
head designs. Mesoudi & O’Brien (2008) assumed a
multimodal adaptive landscape underlying arrowhead
fitness, with multiple locally optimal designs (‘peaks’
in the landscape). Consequently, during periods of
individual learning, different participants converged on
different peaks in this adaptive landscape, thus main-
taining within-group diversity in arrowhead designs.
During periods of cultural learning, different partici-
pants converged on the high-fitness peak found by the
most successful group member, thus reducing diversity
and increasing overall group fitness. However, Mesoudi
(2008b) showed that when the adaptive landscape is
unimodal—with a single peak and a single optimal
arrowhead design—then individual learners easily
converge on this single peak and perform just as well as
the cultural learners, thus eliminating the adaptive
advantage of cultural learning. An important message
here, then, is that the adaptiveness and use of a particular
cultural learning strategy, e.g. copying successful
individuals, critically depends on the shape of the
underlying adaptive landscape (just as the shape of
genetic adaptive landscapes can dramatically affect
genetic evolution; Arnold et al. 2001).
6. DISCUSSION
At the beginning of this paper, we identified the goals
of cultural transmission experiments as answering the
what, who, when and how questions posed by Laland
(2004): what is copied; who is copied; when do
individuals copy; and how do individuals copy? These
questions were motivated by the insight from theoretical
models, some of which were discussed above, that
indiscriminate social learning is not universally adaptive
(e.g. Rogers 1988; Boyd & Richerson 1995; Henrich &
Boyd 1998). These models suggest instead that
individuals should use social information selectively;
that is, they should be selective in who they learn
from, what they copy, how they copy and when
they copy. The various experiments discussed above
support this prediction, and begin to provide specific
answers to these ‘who, what, when and how’ questions
regarding cultural transmission. The following sections
summarize these findings and point to directions for
future research.
(a) What is copied?
Transmission chain studies, in which information is
passed along linear chains of participants, show that
human cultural transmission can often be vulnerable to
distortions and biases rather than constituting a
process of high-fidelity replication. These studies have
identified several candidate content biases in cultural
transmission: a counter-intuitive bias (Barrett & Nyhof
2001); a hierarchical bias (Mesoudi & Whiten 2004);
Phil. Trans. R. Soc. B (2008)
A. Mesoudi & A. Whiten 3497
a gender-stereotype bias (Bangerter 2000; Kashima
2000); and a social bias (Mesoudi et al. 2006a).
(We qualify these as ‘candidate’ biases given that each
is supported by one or at most two experiments; as the
field expands, we anticipate future experiments to
provide further support or qualifications to these initial
findings.) Cognitive, evolutionary and social psychol-
ogy offer a wealth of hypotheses regarding other
potential content biases that might be tested formally
using the transmission chain method. For example,
Heath et al. (2001) have proposed that cultural
transmission is affected by emotional reactions of
disgust, and showed using historical data that rumours
that elicit disgust are more likely to survive than
rumours that do not. Nairne et al. (2008) found that
words that are processed within a survival context (e.g.
relating to food or predators) are recalled better than
those same words presented in non-survival contexts,
suggesting an ‘ecological’ bias in cultural transmission
that might under certain conditions rival the social bias
found by Mesoudi et al. (2006a). The transmission
chain method might be used to experimentally test
whether disgust, ecological or other content biases,
which are currently supported only by observational
evidence or single-generation memory experiments,
operate in cultural transmission, i.e. to what extent they
extend beyond single individuals. It would also be
useful, following the example of the closed-group
method, to implement individual control conditions
in which a single individual repeatedly recalls their own
recalled material (much similar to Bartlett’s (1932)
method of repeated reproduction) in order to quantify
exactly how (or whether) the cumulative, cultural recall
of multiple participants differs from the recall of a
single participant.
(b) How is it copied?
Few diffusion studies with humans have explicitly
addressed the mechanism through which cultural
transmission operates. Schotter & Sopher (2003)
found that explicit verbal advice or rules are more
effective in generating stable behavioural conventions
in an economic game than simply observing past
behaviour. This is echoed by Insko et al.’s (1980,
1983) and Baum et al.’s (2004) findings that optimal
traditions were maintained by explicit verbal rules. The
latter studies found cumulative improvement in
performance, suggesting that explicit verbal rules
might maintain cumulative culture. However, apart
from Schotter & Sopher’s (2003) study (which did not
allow cumulative improvement), there has been no
formal experimental comparison of different cultural
transmission mechanisms, such as imitation, emulation
and stimulus enhancement ( Whiten et al. 2004).
Although such work has only just begun in the non-
human animal literature (Hopper et al. 2007; Whiten
in press), future studies using human adults could
similarly profit from the detailed taxonomies of social
learning and methods of the non-human social learning
research (Want & Harris 2002; Whiten et al. 2004) in
order to identify the cognitive mechanisms underlying
particular forms of cultural transmission.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3498 A. Mesoudi & A. Whiten Review. Roles of cultural transmission experiments
(c) Who is copied?
Recent studies have used the closed-group method to
test the adaptiveness and consequences of two ‘who’
biases: copying the majority (conformity) and copying
the most successful individual. Efferson et al. (2008)
found that some, but not all, participants engage in
conformity, and, as a result, they do better than non-
conformists, consistent with theoretical expectations
(Henrich & Boyd 1998). It is puzzling, then, why not
all of Efferson et al.’s (2008) or McElreath et al.’s (2005)
participants and none of Efferson et al.’s (2007)
participants engaged in conformity, even though
conformity would have yielded higher earnings.
Establishing the reasons for this discrepancy is an
important task for future experiments. Mesoudi &
O’Brien (2008) and Mesoudi (2008b) simulated a
copy-successful-individuals bias, and found, by con-
trast, that almost all participants readily discarded the
artefact that they had spent several trials designing to
adopt the artefact design of the most successful
member of the group. (N.B. this appears to contrast
with recent studies of non-human primates that
showed a marked tendency to stick to a known,
satisfactory technique rather than upgrade to a more
productive one being used by another individual, a
difference that might help explain why complex
cumulative culture is such a distinctive human
attribute; Marshall-Pescini & Whiten 2008). Efferson
et al. (2007), however, found that the behaviour of the
most successful group member was not adopted. The
discrepancy between these findings and those of
Mesoudi & O’Brien (2008) might be explained by
differences in task and participant sample. For
example, Efferson et al. (2007) used a simple task of
choosing one of two discrete options (one of two
technologies), whereas Mesoudi & O’Brien (2008)
used a more complex task with multiple variables, some
continuous and some discrete, some functional and
some neutral, and with multimodal adaptive land-
scapes underlying artefact success. As shown by
Mesoudi (2008b), the shape of this adaptive landscape
can dramatically affect the adaptiveness of cultural
learning. Further studies might look more system-
atically at the nature of the task set for participants and
the underlying fitness functions determining task
success.
(d) When do people copy?
Experimental studies have broadly confirmed the
theoretical predictions that cultural learning should
be more frequent relative to individual learning when
environments do not change (McElreath et al. 2005)
and when individual learning is inaccurate (McElreath
et al. 2005) and/or costly (Kameda & Nakanishi 2002;
Mesoudi & O’Brien 2008). However, several studies
(McElreath et al. 2005; Efferson et al. 2007, 2008) have
found that people engage in cultural transmission far
less than would be optimal. Moreover, much individual
variation has been found in these and other studies,
with participants often differing widely in their
tendency to learn from others. At present, the cause
of this individual variation is a mystery, and points to a
more general question about this and the model-based
biases (§6c): what is the origin of these cultural
Phil. Trans. R. Soc. B (2008)
transmission biases and strategies? Are they genetically
specified, as sometimes assumed in cultural evolution
models (e.g. Boyd & Richerson 1985), or are they
learned during ontogeny (a kind of ‘learning of learning
strategies’)? Perhaps cultural learning strategies are
themselves learned from others, such that conformists
conform because they have copied from others a
tendency to conform. Developmental studies would
be valuable here in determining how individual
variation in experimental behaviour might be explained
by different learning opportunities during ontogeny.
Recent twin studies (McEwen et al. 2007; Fenstermacher &
Saudino 2007) have suggested that individual differ-
ences in the capacity of 2-year-olds to imitate can be
partly attributed to genetic variance and partly to
environmental factors, although the studies disagreed
as to the relative influence of each and whether the
environment is shared (e.g. interaction with parents) or
non-shared (e.g. individual reinforcement histories). It
should also be noted that a capacity for imitation, one
particular social learning mechanism, may be unrelated
to the voluntary use of cultural learning strategies (e.g.
conformity or copy-successful-individuals) tested in
the experiments reviewed here; future developmental
and twin studies might examine these more specific
cultural learning strategies as well as broader capacities
such as imitation, and in children of varying ages.
Cross-cultural studies might also be used to explore the
cultural learning of cultural learning strategies. So far,
the vast majority of cultural transmission experiments
have been conducted using western (USA or UK)
participants (see table S1 in the electronic supple-
mentary material), with the exception of Kameda &
Nakanishi’s (2002, 2003) studies in Japan and Efferson
et al.’s (2007) study in Bolivia. Non-transmission
experiments have found that more ‘collectivist’ East
Asian participants show higher levels of conformity
than more ‘individualist’ western participants (Bond &
Smith 1996); perhaps the low levels of conformity seen
in some of the experiments reviewed above are due to
the use of western participant samples. Kameda &
Nakanishi (2002) found that a majority of Japanese
participants engaged in conformist transmission, but
whether this was systematically different from observed
rates of conformity in studies that used western
participant samples is unclear given differences in
experimental tasks and procedures.
(e) Integrating questions
So far, we have assumed that questions regarding
cultural transmission—who, what, how and when—can
be considered separately. In reality, it is unlikely that
cultural behaviour is neatly divided in this way. A
promising avenue for future research would be to pit
different biases against one another. For example, what
happens when the group majority exhibits a different
behaviour or belief to the most successful individual?
What happens when low-prestige models pass on
minimally counter-intuitive information? Perhaps
certain biases will be found to dominate cultural
transmission, or perhaps equilibria will be observed
where different biases operate simultaneously. Inspi-
ration might be sought from evolutionary biology,
where experimental studies are used to explore the
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Roles of cultural transmission experiments
simultaneous operation of multiple evolutionary pro-
cesses in the same population (e.g. natural and sexual
selection; Skroblin & Blows 2006).
7. CONCLUSIONS
Our understanding of human cultural change can greatly
benefit from laboratory experiments. While mathe-
matical models in the gene–culture coevolution/cultural
evolution tradition have produced invaluable insights into
the processes of cultural change, laboratory experiments
are needed to test the assumptions and findings of these
models with actual people. Similarly, while, historical,
ethnographic and archaeological studies of cultural
evolution (Basalla 1988; Hewlett et al. 2002; O’Brien &
Lyman 2002) are invaluable in providing real-world data
regarding cultural change, laboratory experiments offer a
degree of control and manipulation that is impossible to
achieve with naturalistic studies. Of course, laboratory
experiments also have their shortcomings, most
obviously deficits in external validity resulting from the
simple tasks and artificial laboratory settings involved.
However, when experiments are used in conjunction with
other methods, as part of a unified science of cultural
evolution (Mesoudi et al. 2006b), then a better under-
standing of cultural phenomena can be attained than
when a single method is used alone. The studies reviewed
here suggest that this interdisciplinarity is beginning
to pay dividends, with studies such as Kameda &
Nakanishi (2003), McElreath et al. (2005) and Efferson
et al. (2008) explicitly tying experiments to mathematical
models, and others (e.g. Insko et al. 1983; Mesoudi &
O’Brien 2008) using experiments to explicitly test
hypotheses from cultural anthropology and archaeology.
Much of the older work reviewed here has so far had
relatively little direct impact in the fields in which they
originated, typically social psychology but also
sociology, economics and anthropology. For example,
Jacobs & Campbell’s (1961) pioneering experimental
simulation of conformist cultural transmission has,
in the last 47 years, only been cited between eight (Web
of Knowledge, 1970–present) and ten (PsycInfo,
1961–present) times. Similarly, in the quarter-century
since Insko et al.’s (1983) hugely innovative study was
published in one of the leading social psychology
journals (the Journal of Personality and Social Psychol-
ogy), that paper has been cited only eight times
according to Web of Knowledge, while PsycInfo
records no citations for it at all. For whatever reasons,
social psychologists have not considered cultural
transmission to be worthy of study in this way, and
cultural anthropologists have not considered experi-
ments to be particularly relevant to their work. By
drawing such studies to the attention of a wider body of
researchers in the cultural evolution tradition, and
linking them to each other and to formal cultural
evolution theory, we hope to offer added value and renew
interest in experimental studies of cultural transmission.
A.M. was supported by a Mellon Foundation Postdoctoral
Fellowship. A.W. was supported by a Royal Society
Leverhulme Trust Senior Research Fellowship. We are
grateful for valuable comments and advice from Tom
Griffiths and two anonymous reviewers.
Phil. Trans. R. Soc. B (2008)
A. Mesoudi & A. Whiten 3499
REFERENCES
Allport, G. W. & Postman, L. 1947 The psychology of rumor.
Oxford, UK: Henry Holt.
Aoki, K., Wakano, J. Y. & Feldman, M. W. 2005 The
emergence of social learning in a temporally changing
environment. Curr. Anthropol. 46, 334–340. (doi:10.1086/
428791)
Arnold, S. J., Pfrender, M. E. & Jones, A. G. 2001 The
adaptive landscape as a conceptual bridge between micro-
and macroevolution. Genetica 112–113, 9–32. (doi:10.
1023/A:1013373907708)
Atran, S. 1998 Folk biology and the anthropology of science.
Behav. Brain Sci. 21, 547–609.
Atran, S. 2001 The trouble with memes. Hum. Nat. 12,
351–381. (doi:10.1007/s12110-001-1003-0)
Bandura, A. 1977 Social learning theory. Oxford, UK:
Prentice-Hall.
Bangerter, A. 2000 Transformation between scientific and
social representations of conception. Br. J. Soc. Psychol.
39, 521–535. (doi:10.1348/014466600164615)
Barrett, J. L. & Nyhof, M. A. 2001 Spreading non-natural
concepts: the role of intuitive conceptual structures in
memory and transmission of cultural materials. J. Cogn.
Cult. 1, 69–100. (doi:10.1163/156853701300063589)
Bartlett, F. C. 1932 Remembering. Oxford, UK: Macmillan.
Basalla, G. 1988 The evolution of technology. Cambridge, UK:
Cambridge University Press.
Baum, W. M., Richerson, P. J., Efferson, C. M. & Paciotti,
B. M. 2004 Cultural evolution in laboratory microsoci-
eties including traditions of rule giving and rule following.
Evol. Hum. Behav. 25, 305–326. (doi:10.1016/j.evolhum-
behav.2004.05.003)
Bettinger, R. L. & Eerkens, J. 1999 Point typologies, cultural
transmission, and the spread of bow-and-arrow tech-
nology in the prehistoric Great Basin. Am. Antiq. 64,
231–242. (doi:10.2307/2694276)
Blackmore, S. 1999 The meme machine. Oxford, UK: Oxford
University Press.
Bond, R. & Smith, P. B. 1996 Culture and conformity: a
meta-analysis of studies using Asch’s (1952b, 1956) line
judgment task. Psychol. Bull. 119, 111–137. (doi:10.1037/
0033-2909.119.1.111)
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Boyd, R. & Richerson, P. J. 1995 Why does culture increase
human adaptability? Ethol. Sociobiol. 16, 125–143.
(doi:10.1016/0162-3095(94)00073-G)
Boyd, R. & Richerson, P. J. 1996 Why culture is common, but
cultural evolution is rare. Proc. Br. Acad. 88, 77–93.
Boyd, R., Gintis, H., Bowles, S. & Richerson, P. J. 2003 The
evolution of altruistic punishment. Proc. Natl Acad. Sci.
USA 100, 3531–3535. (doi:10.1073/pnas.0630443100)
Boyer, P. 1994 The naturalness of religious ideas. Berkeley, CA:
University of California Press.
Byrne, R. W. & Whiten, A. 1988 Machiavellian intelligence.
Oxford, UK: Clarendon Press.
Caldwell, C. A. & Millen, A. E. 2008a Experimental models
for testing hypotheses about cumulative cultural
evolution. Evol. Hum. Behav. 29, 165–171. (doi:10.
1016/j.evolhumbehav.2007.12.001)
Caldwell, C. A. & Millen, A. E. 2008b Studying cumulative
cultural evolution in the laboratory. Phil. Trans. R. Soc. B
363, 3529–3539. (doi:10.1098/rstb.2008.0133)
Campbell, D. T. 1974 Evolutionary epistemology. In The
philosophy of Karl Popper (ed. P. A. Schilpp), pp. 413–463.
La Salle, IL: Open Court.
Cavalli-Sforza, L. L. & Feldman, M. W. 1981 Cultural
transmission and evolution. Princeton, NJ: Princeton
University Press.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3500 A. Mesoudi & A. Whiten Review. Roles of cultural transmission experiments
Claidiere, N. & Sperber, D. 2007 The role of attraction in
cultural evolution. J. Cogn. Cult. 7, 89–111. (doi:10.1163/
156853707X171829)
Darwin, C. 1859/1968 The origin of species. London, UK:
Penguin.
Dunbar, R. 2003 The social brain. Ann. Rev. Anthropol. 32,
163–181. (doi:10.1146/annurev.anthro.32.061002.093158)
Efferson, C., Richerson, P., McElreath, R., Lubell, M.,
Edsten, E., Waring, T., Paciotti, B. & Baum, W. 2007
Learning, productivity, and noise: an experimental study
of cultural transmission on the Bolivian Altiplano. Evol.
Hum. Behav. 28, 11–17. (doi:10.1016/j.evolhumbehav.
2006.05.005)
Efferson, C., Lalive, R., Richerson, P. J., McElreath, R. &
Lubell, M. 2008 Conformists and mavericks: the empirics
of frequency-dependent cultural transmission. Evol.
Hum. Behav. 29, 56–64. (doi:10.1016/j.evolhumbehav.
2007.08.003)
Fenstermacher, S. K. & Saudino, K. J. 2007 Toddler see,
toddler do? Genetic and environmental influences on
laboratory-assessed elicited imitation. Behav. Genet. 37,
639–647. (doi:10.1007/s10519-007-9160-5)
Flynn, E. G. 2008 Investigating children as cultural magnets:
do young children transmit redundant information along
diffusion chains? Phil. Trans. R. Soc. B. 363, 3541–3551.
(doi:10.1098/rstb.2008.0136)
Gerard, R. W., Kluckhohn, C. & Rapoport, A. 1956
Biological and cultural evolution. Behav. Sci. 1, 6–34.
Griffiths, T. L., Christian, B. R. & Kalish, M. L. 2008 Using
category structures to test iterated learning as a method for
identifying inductive biases. Cogn. Sci. 32, 68–107.
(doi:10.1080/03640210701801974)
Griffiths, T. L., Kalish, M. L. & Lewandowsky, S. 2008
Theoretical and empirical evidence for the impact of
inductive biases on cultural evolution. Phil. Trans. R. Soc. B
363, 3503–3514. (doi:10.1098/rstb.2008.0146)
Gurerk, O., Irlenbusch, B. & Rockenbach, B. 2006 The
competitive advantage of sanctioning institutions. Science
312, 108–111. (doi:10.1126/science.1123633)
Hall, K. R. L. 1951 The effect of names and titles upon the
serial reproduction of pictorial and verbal material. Br.
J. Psychol. 41, 109–121.
Hartl, D. L. & Clark, A. G. 1997 Principles of population
genetics. Sunderland, MA: Sinauer.
Heath, C., Bell, C. & Sternberg, E. 2001 Emotional selection
in memes. J. Pers. Soc. Psychol. 81, 1028–1041. (doi:10.
1037/0022-3514.81.6.1028)
Henrich, J. & Boyd, R. 1998 The evolution of conformist
transmission and the emergence of between-group
differences. Evol. Hum. Behav. 19, 215–241. (doi:10.
1016/S1090-5138(98)00018-X)
Henrich, J. & Boyd, R. 2002 On modeling cognition and
culture: why cultural evolution does not require replica-
tion of representations. J. Cogn. Cult. 2, 87–112. (doi:10.
1163/156853702320281836)
Henrich, J. et al. 2005 ‘Economic Man’ in cross-cultural
perspective. Behav. Brain Sci. 28, 795–855.
Hewlett, B., De Silvestri, A. & Guglielmino, C. R. 2002
Semes and genes in Africa. Curr. Anthropol. 43, 313–321.
(doi:10.1086/339379)
Hopper, L. M., Spiteri, A., Lambeth, S. P., Schapiro, S.,
Horner, V. & Whiten, A. 2007 Experimental studies of
traditions and underlying transmission processes in
chimpanzees. Anim. Behav. 73, 1021–1032. (doi:10.
1016/j.anbehav.2006.07.016)
Insko, C. A., Thibaut, J. W., Moehle, D., Wilson, M.,
Diamond, W. D., Gilmore, R., Solomon, M. R. & Lipsitz,
A. 1980 Social evolution and the emergence of leadership.
J. Pers. Soc. Psychol. 39, 431–448. (doi:10.1037/0022-
3514.39.3.431)
Phil. Trans. R. Soc. B (2008)
Insko, C. A., Gilmore, R., Drenan, S., Lipsitz, A., Moehle, D.
& Thibaut, J. W. 1983 Trade versus expropriation in open
groups. J. Pers. Soc. Psychol. 44, 977–999. (doi:10.1037/
0022-3514.44.5.977)
Jacobs, R. C. & Campbell, D. T. 1961 The perpetuation of an
arbitrary tradition through several generations of a
laboratory microculture. J. Abnorm. Soc. Psychol. 62,
649–658. (doi:10.1037/h0044182)
Kalish, M. L., Griffiths, T. L. & Lewandowsky, S. 2007
Iterated learning: intergenerational knowledge trans-
mission reveals inductive biases. Psychon. Bull. Rev. 14,
288–294.
Kameda, T. & Nakanishi, D. 2002 Cost-benefit analysis of
social/cultural learning in a nonstationary uncertain
environment. Evol. Hum. Behav. 23, 373–393. (doi:10.
1016/S1090-5138(02)00101-0)
Kameda, T. & Nakanishi, D. 2003 Does social/cultural
learning increase human adaptability? Evol. Hum. Behav.
24, 242–260. (doi:10.1016/S1090-5138(03)00015-1)
Kashima, Y. 2000 Maintaining cultural stereotypes in the
serial reproduction of narratives. Pers. Soc. Psychol. B. 26,
594–604. (doi:10.1177/0146167200267007)
Lakoff, R. 1975 Language and woman’s place. New York, NY:
Harper & Row.
Laland, K. N. 2004 Social learning strategies. Learn. Behav.
32, 4–14.
Mackintosh, N. 1983 Conditioning and associative learning.
Oxford, UK: Oxford University Press.
Marshall-Pescini, S. & Whiten, A. 2008 Chimpanzees (Pan
troglodytes) and the question of cumulative culture. Anim.
Cogn. 11, 449–456. (doi:10.1007/s10071-007-0135-y)
Maxwell, R. S. 1936 Remembering in different social groups.
Br. J. Psychol. 27, 30–40.
Mayr, E. & Provine, W. 1980 The evolutionary synthesis.
Cambridge, MA: Harvard University Press.
McElreath, R., Lubell, M., Richerson, P. J., Waring, T. M.,
Baum, W., Edsten, E., Efferson, C. & Paciotti, B. 2005
Applying evolutionary models to the laboratory study of
social learning. Evol. Hum. Behav. 26, 483–508. (doi:10.
1016/j.evolhumbehav.2005.04.003)
McEwen, F., Happé, F., Bolton, P., Rijsdijk, F., Ronald, A.,
Dworzynski, K. & Plomin, R. 2007 Origins of individual
differences in imitation. Child Dev. 78, 474–492. (doi:10.
1111/j.1467-8624.2007.01010.x)
Mesoudi, A. 2007 Using the methods of experimental social
psychology to study cultural evolution. J. Soc. Evol. Cult.
Psychol. 1, 35–58.
Mesoudi, A. 2008a A Darwinian theory of cultural evolution
can promote an evolutionary synthesis for the social
sciences. Biol. Theor. 2, 263–275. (doi:10.1162/biot.2007.
2.3.263)
Mesoudi, A. 2008b An experimental simulation of the “copy-
successful-individuals” cultural learning strategy: adaptive
landscapes, producer-scrounger dynamics, and informa-
tional access costs. Evol. Hum. Behav. 29, 350–363.
(doi:10.1016/j.evolhumbehav.2008.04.005)
Mesoudi, A. & O’Brien, M. J. 2008 The cultural transmission
of Great Basin projectile point technology I: an experi-
mental simulation. Am. Antiq. 73, 3–28.
Mesoudi, A. & O’Brien, M. J. 2008 The cultural transmission
of Great Basin projectile point technology II: an agent-
based computer simulation. Am. Antiq 21, 350–363.
(doi:10.1016/j.evolhumbehav.2008.04.005)
Mesoudi, A. & Whiten, A. 2004 The hierarchical transfor-
mation of event knowledge in human cultural trans-
mission. J. Cogn. Cult. 4, 1–24. (doi:10.1163/15685370
4323074732)
Mesoudi, A., Whiten, A. & Laland, K. N. 2004 Is human
cultural evolution Darwinian? Evolution 58, 1–11. (doi:10.
1554/03-212)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Roles of cultural transmission experiments
Mesoudi, A., Whiten, A. & Dunbar, R. I. M. 2006a A bias
for social information in human cultural transmission.
Br. J. Psychol. 97, 405–423. (doi:10.1348/000712605X
85871)
Mesoudi, A., Whiten, A. & Laland, K. N. 2006b Towards a
unified science of cultural evolution. Behav. Brain Sci. 29,
329–383. (doi:10.1017/S0140525X06009083)
Nairne, J. S., Pandeirada, J. N. S. & Thompson, S. R. 2008
Adaptive memory: the comparative value of survival
processing. Psychol. Sci. 19, 176–180. (doi:10.1111/
j.1467-9280.2008.02064.x)
Norenzayan, A., Atran, S., Faulkner, J. & Schaller, M. 2006
Memory and mystery: the cultural selection of minimally
counterintuitive narratives. Cogn. Sci. 30, 531–553.
(doi:10.1207/s15516709cog0000_68)
Northway, M. L. 1936 The influence of age and social group
on children’s remembering. Br. J. Psychol. 27, 11–29.
O’Brien, M. J. & Lyman, R. L. 2002 Evolutionary archeology.
Evol. Anthropol. 11, 26–36. (doi:10.1002/evan.10007)
Plotkin, H. 1994 Darwin machines and the nature of knowledge.
London, UK: Penguin.
Richerson, P. J. & Boyd, R. 2005 Not by genes alone. Chicago,
IL: University of Chicago Press.
Rogers, A. 1988 Does biology constrain culture? Am.
Anthropol. 90, 819–831. (doi:10.1525/aa.1988.90.4.02a0
0030)
Rose, E. & Felton, W. 1955 Experimental histories of culture.
Am. Sociol. Rev. 20, 383–392. (doi:10.2307/2092735)
Rubin, D. C. 1995 Memory in oral traditions. Oxford, UK:
Oxford University Press.
Schotter, A. & Sopher, B. 2003 Social learning and
coordination conventions in intergenerational games.
J. Pol. Econ. 111, 498–529. (doi:10.1086/374187)
Sherif, M. 1936 The psychology of social norms. Oxford, UK:
Harper.
Phil. Trans. R. Soc. B (2008)
A. Mesoudi & A. Whiten 3501
Skroblin, A. & Blows, M. W. 2006 Measuring natural and
sexual selection on breeding values of male display traits in
Drosophila serrata. J. Evol. Biol. 19, 35–41. (doi:10.1111/
j.1420-9101.2005.00986.x)
Sperber, D. 1996 Explaining culture. Oxford, UK: Oxford
University Press.
Sperber, D. 2000 Why memes won’t do. In Darwinizing
culture (ed. R. Aunger), pp. 163–174. Oxford, UK: Oxford
University Press.
Sperber, D. & Hirschfeld, L. 2004 The cognitive foundations
of cultural stability and diversity. Trends Cogn. Sci. 8,
42–46. (doi:10.1016/j.tics.2003.11.002)
Tomasello, M. 1999 The cultural origins of human cognition.
Cambridge, MA: Harvard University Press.
Want, S. C. & Harris, P. L. 2002 How do children ape? Dev.
Sci. 5, 1–13. (doi:10.1111/1467-7687.00194)
Ward, T. H. G. 1949 An experiment on serial reproduction
with special reference to the changes in the design of early
coin types. Br. J. Psychol. 39, 142–147.
Whiten, A. In press. Coming of age for cultural panthropology.
In The mind of the chimpanzee (eds E. Lonsdorf, S. Ross & T.
Matsuzawa). Chicago, IL: University of Chicago Press.
Whiten, A. & Mesoudi, A. 2008 Establishing an experimental
science of culture: animal social diffusion experiments.
Phil. Trans. R. Soc. B 363, 3477–3488. (doi:10.1098/rstb.
2008.0134)
Whiten, A., Goodall, J., McGrew, W. C., Nishida, T.,
Reynolds, V., Sugiyama, Y., Tutin, C. E. G., Wrangham,
R. W. & Boesch, C. 1999 Cultures in chimpanzees. Nature
399, 682–685. (doi:10.1038/21415)
Whiten, A., Horner, V., Litchfield, C. & Marshall-Pescini, S.
2004 How do apes ape? Learn. Behav. 32, 36–52.
Zucker, L. G. 1977 The role of institutionalization in cultural
persistence. Am. Sociol. Rev. 42, 726–743. (doi:10.2307/
2094862)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3503–3514

doi:10.1098/rstb.2008.0146
Published online 19 September 2008

Review

Theoretical and empirical evidence for the impact

of inductive biases on cultural evolution
Thomas L. Griffiths1,*, Michael L. Kalish2 and Stephan Lewandowsky3
1
Department of Psychology, University of California, 3210 Tolman Hall No. 1650,
Berkeley, CA 94720-1650, USA
2
Institute of Cognitive Science, University of Louisiana at Lafayette, Lafayette, LA 70501, USA
3
Department of Psychology, University of Western Australia, Perth, WA 6009, Australia
The question of how much the outcomes of cultural evolution are shaped by the cognitive capacities
of human learners has been explored in several disciplines, including psychology, anthropology and
linguistics. We address this question through a detailed investigation of transmission chains, in which
each person passes information to another along a chain. We review mathematical and empirical
evidence that shows that under general conditions, and across experimental paradigms, the
information passed along transmission chains will be affected by the inductive biases of the people
involved—the constraints on learning and memory, which influence conclusions from limited data.
The mathematical analysis considers the case where each person is a rational Bayesian agent. The
empirical work consists of behavioural experiments in which human participants are shown to
operate in the manner predicted by the Bayesian framework. Specifically, in situations in which each
person’s response is used to determine the data seen by the next person, people converge on concepts
consistent with their inductive biases irrespective of the information seen by the first member of the
chain. We then relate the Bayesian analysis of transmission chains to models of biological evolution,
clarifying how chains of individuals correspond to population-level models and how selective forces
can be incorporated into our models. Taken together, these results indicate how laboratory studies of
transmission chains can provide information about the dynamics of cultural evolution and illustrate
that inductive biases can have a significant impact on these dynamics.
Keywords: cultural evolution; Bayesian models; learning

1. INTRODUCTION information is passed from one person to another

Much of human knowledge is acquired not by
interacting directly with the physical world, but by
interacting with other people. The concepts we use, the
social conventions we obey and the languages we speak
are often learned by observing examples, behaviour
or speech produced by other people. These processes
of knowledge transmission constitute a basic element of
cultural evolution and have been the object of extensive
research in psychology (e.g. Bartlett 1932; Mesoudi
2007), anthropology (e.g. Cavalli-Sforza & Feldman
1981; Boyd & Richerson 1985; Sperber 1996) and
linguistics (e.g. Kirby 2001; Briscoe 2002; Nowak et al.
2002). A key question in all cases is how the minds of
human learners shape the outcomes of cultural
evolution: how inductive biases—the constraints on
learning and memory, which influence our conclusions
from limited data—relate to the concepts, conventions
and languages which appear in human societies.1
In this paper, we explore one part of this question by
analysing the effects of inductive biases on one simple
form of knowledge transmission: the case where
* Author for correspondence (tom_griffiths@berkeley.edu).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
(figure 1). In this case, each person observes data
generated by the previous person, forms a hypothesis
about the process that produced those data and then
uses that hypothesis to generate data for the next
person. For example, a language learner might infer the
grammar of a language by hearing the utterances of
another person, and then use that grammar to generate
utterances that are heard by someone else. The
languages spoken by the people in this chain will
gradually change over time as a consequence of this
process. Transmission chains of this kind represent each
generation of learners with just one person, and thus do
not allow us to explore the influences of individuals
within a generation on one another; nonetheless, they
provide a powerful tool for exploring how knowledge
changes when transmitted across generations.
Our analysis of transmission chains (also known as
diffusion chains) uses a mixture of mathematical
modelling and laboratory experiments with human
participants. Mathematical models are widely used in
the study of cultural evolution, often drawing on the rich
body of mathematical models of biological evolution
(Cavalli-Sforza & Feldman 1981; Boyd & Richerson
1985; Nowak et al. 2002). Laboratory experiments are
used more rarely, although there exist both classic and

3503 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3504 T. L. Griffiths et al. Review. Inductive biases and cultural evolution
hypothesis hypothesis
data data data
Figure 1. Transmission chains provide a simple setting for
studying cultural transmission that has been used in
psychology, anthropology and linguistics. In a transmission
chain, each agent observes the data generated by the previous
agent, forms a hypothesis about the source of these data and
then uses that hypothesis to generate data for the next agent.
more recent studies of this kind (see Mesoudi 2007;
Caldwell & Millen 2008; Mesoudi & Whiten 2008).
Combining mathematical modelling with laboratory
experiments gives us the opportunity to test the
predictions of our models. Because the mechanisms of
cultural evolution are fundamentally psychological,
involving processes such as learning, memory and
decision-making, using the methods of cognitive psy-
chology allows us to determine whether we have
accurately characterized these mechanisms.
We seek to describe how human inductive biases
change the information being transmitted. Both
learning and remembering involve inductive problems,
requiring people to form hypotheses that go beyond the
limited data that are available to them (e.g. Anderson
1990). Learning language is a classic example of an
inductive problem, with the grammar of the language
being underdetermined by the utterances a learner
observes. Similar problems arise in other settings, such
as determining whether a social convention such as
tipping applies based on a few examples or reconstruct-
ing a briefly glimpsed experimental stimulus. Inductive
biases are the factors that lead a learner to choose one
hypothesis over another when both are equally
consistent with the observed data. In language
learning, such biases might favour languages of certain
forms over others, whereas in the case of tipping they
might reflect beliefs about social structures. While
previous work has explored how relatively simple
‘direct biases’ that influence whether an agent adopts
a hypothesis affect knowledge transmission (Boyd &
Richerson 1985), we aim to obtain general results
characterizing the consequences of arbitrarily complex
inductive biases.
Exploring the effects of inductive biases on knowledge
transmission requires having a means of expressing these
biases. We do this by analysing transmission chains
formed of agents who use Bayesian inference, a
mathematical theory that provides a rational solution
to inductive problems. Bayesian models make inductive
biases explicit and have accounted for human
learning (Anderson 1991; Tenenbaum & Griffiths
2001; Griffiths & Tenenbaum 2005) and memory
(Anderson & Milson 1989; Shiffrin & Steyvers 1997;
Griffiths et al. 2007) with considerable success. Examin-
ing how knowledge transmission by Bayesian agents is
affected by the inductive biases of those agents gives us a
very general framework, whose assumptions overlap
with accounts of rational behaviour in economics and
statistics. This framework makes predictions about the
outcomes of cultural evolution, which we can test in the
laboratory with human participants.
Phil. Trans. R. Soc. B (2008)
Our central thesis is that the inductive biases of
individuals have a significant effect on the information
conveyed along a transmission chain, and that this
…
suggests that inductive biases may play a significant role
in cultural evolution more broadly. In support of this
thesis, we present a basic mathematical result—that
information passed along a transmission chain formed
of the Bayesian agents ultimately comes to reflect the
inductive biases of those agents (Griffiths & Kalish
2005, 2007; Kirby et al. 2007)—and summarize a series
of experiments with human participants, which bear
out this prediction (Kalish et al. 2007; Griffiths et al.
2008). We also show that this analysis can be
generalized to populations as well as chains of
individuals, producing parallels with formal models of
biological evolution, and that in such a context the
inductive biases of individual learners can have a
greater effect on the outcome of cultural evolution
than selective forces.
We proceed as follows: §2 reviews the significance of
questions about inductive biases and cultural evolution
in anthropology, psychology and linguistics; §3 discusses
how these different disciplines have converged on the
use of transmission chains and summarizes our math-
ematical analyses; §4 presents empirical results bearing
out the predictions of this account; §5 outlines how our
approach relates to the models of biological evolution
and the relative importance of inductive biases and
selective forces in cultural evolution; and §6 presents
our conclusions.
2. RELATING INDUCTIVE BIASES AND
CULTURAL EVOLUTION
Inductive problems feature prominently in cognition.
Questions about how people learn categories,
functional relationships or languages ultimately reduce
to questions about human inductive biases. Typically,
research with adult participants explores the form of
these biases, such as what kinds of categories are easy to
learn (Shepard et al. 1961), whereas researchers in
cognitive development seek to understand the origins
of those biases (e.g. Spelke et al. 1992; Gopnik &
Meltzoff 1997). Recently, evolutionary psychologists
have suggested that we can obtain answers to these
questions by looking at ‘human universals’ (Brown
1991)—the beliefs and practices which seem to be
common to all human societies (e.g. Pinker 2002).
Anthropologists have explicitly explored the
relationship between inductive biases and cultural
evolution. Sperber (1985, 1996), Boyer (1994, 1998)
and Atran (2001, 2002) have argued that processes of
cultural transmission provide the opportunity for
inductive biases, such as ontological commitments
about the kinds of entities that exist, to manifest
themselves in culture. This argument is based on the
significant role that learning and memory play in
cultural transmission. Sperber (1996, p. 84) states
that ‘the ease with which a particular representation
can be memorized’ will affect its transmission, and
Boyer (1994, 1998) and Atran (2001) emphasize the
effects of inductive biases on memory. This idea has
some empirical support. For example, Nichols (2004)
showed that social conventions based on disgust were
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Inductive biases and cultural evolution
more likely to survive several decades of cultural
transmission than those without this emotional com-
ponent. This advantage is consonant with the large
body of research showing that emotional events are
often remembered better than comparable events that
are lacking an emotional component (for a review, see
Buchanan 2007).
The role of memory and learning in cultural
transmission has also led to arguments against applying
mathematical models of biological evolution to cultural
evolution (e.g. Cavalli-Sforza & Feldman 1981; Boyd &
Richerson 1985), on the grounds that imperfect
inferential transmission is very different from the more
reliable copying of genes, which underlies biological
evolution (Boyer 1998; Atran 2001; Sperber & Claidiére
2006). In particular, cognitive factors that transform
knowledge in a way that is analogous to the mutation of
genes may play a more significant role in cultural
evolution than external selective forces that favour one
piece of knowledge over another. Henrich & Boyd
(2002) presented several simple models intended to
defuse these arguments. For example, one model
showed that in the presence of strong ‘cognitive
attractors’ that make agents more likely to adopt
particular pieces of knowledge, weak selective forces
that increased the value of different knowledge were
sufficient to favour one attractor over another as the
outcome of cultural evolution. We return to the question
of how inductive biases and selection interact in §5.
Research on language evolution also explores the
relationship between inductive biases and cultural
transmission, examining how constraints on language
learning influence the languages that a population of
learners comes to speak. Human languages form a
subset of all logically possible communication schemes,
with some properties being shared by all languages
(Greenberg 1963; Comrie 1981; Hawkins 1988).
Traditionally, these ‘linguistic universals’ are explained
by appealing to the constraints of an innate system
specific to the acquisition of language (e.g. Chomsky
1965). A popular alternative explanation is that the
universal properties of human languages have arisen as
a consequence of languages being learned anew by each
generation, with each learner having only weak,
domain-general inductive biases (e.g. Kirby 2001).
This alternative explanation relies upon the possibility
that cultural transmission can emphasize the inductive
biases of language learners, allowing such weak biases
to be translated into strong and systematic universals of
the kind seen in human languages.
The effects of cultural transmission on languages
have also been the subject of extensive observational
and experimental analysis. Creolization, the formation
of a more regular system of communication from a
piecemeal pidgin, has traditionally been one of the
strongest arguments for constraints on language
acquisition influencing the structure of languages
(Bickerton 1981), and typically occurs when a language
is passed from one generation to the next. Experiments
investigating how adults and children learn artificial
but realistic languages have provided support for the
idea that language learning by children plays an
important role in this process, showing that children
tend to regularize probabilistic elements of languages
Phil. Trans. R. Soc. B (2008)
T. L. Griffiths et al. 3505
(making them more deterministic) to a greater extent
than adults (Hudson-Kam & Newport 2005). Recent
work has also explored how languages are formed and
change across generations through the observation of
the development and transmission of sign languages
(Senghas et al. 2004), complementing an extensive
theoretical and empirical literature on language
creation and change (DeGraff 1999).
The preceding examples illustrate that all the three
disciplines discussed—psychology, anthropology and
linguistics—could be informed by a deeper under-
standing of how inductive biases affect knowledge
transmission.
3. USING TRANSMISSION CHAINS TO MODEL
CULTURAL EVOLUTION
In addition to sharing common questions about the
influence of inductive biases on cultural transmission,
psychologists, anthropologists and linguists have all used
a common paradigm to explore these questions,
examining what happens when information is trans-
mitted along a single chain of individuals, as illustrated
in figure 1. Such transmission chains provide a way to
study one of the basic elements of cultural evolution—
how information changes when passed from one person
to another—in isolation, making it possible to study it in
detail. While this analysis ignores many of the other
factors that are important to the creation and change of
concepts and languages, such as interactions between
individuals within a generation (Steels 2003; Galantucci
2005; Garrod et al. 2007), understanding how each of
these factors operates in isolation will ultimately help
understand their combination.
The use of transmission chains in psychology was
pioneered by Bartlett’s (1932) ‘serial reproduction’
experiments, in which participants were shown a
stimulus and then asked to reproduce it from memory,
with their recalled version being presented to the next
participant and so on. Bartlett argued that reproductions
seem to become more consistent with the cultural biases
of the participants as the number of successive
reproductions increases. However, these arguments
were largely anecdotal and lacked quantitative rigor.
Nonetheless, serial reproduction has become one of the
primary methods that psychologists have used to explore
the effects of cultural transmission, and similar experi-
ments are used by anthropologists and biologists to
examine what kinds of cultural concepts persist over
time and whether non-human animals can transmit
information across generations (for a review, see
Mesoudi 2007; Mesoudi & Whiten 2008; Whiten &
Mesoudi 2008).
In linguistics, the study of transmission chains has
largely been restricted to simulations of the process of
language change. In these ‘iterated learning’
simulations, a sequence of agents each learns a
language by observing the utterances of the previous
agent, and then in turn produces utterances that are
observed by the next agent (Kirby 2001; see Smith &
Kirby 2008). Simulations have shown that languages
with interesting structure emerge from iterated learning
with a variety of learning algorithms (Kirby 2001;
Brighton 2002; Smith et al. 2003). In particular, basic
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3506 T. L. Griffiths et al. Review. Inductive biases and cultural evolution
properties of human languages such as composition-
ality—the use of different parts of an utterance to
describe different aspects of an event—can be produced
by very simple learning algorithms, without requiring
innate language-specific constraints on learning (e.g.
Smith et al. 2003).
The prevalence of transmission chains in research on
cultural evolution is due in part to their simplicity as a
model of knowledge transmission. This simplicity also
makes transmission chains amenable to mathematical
analysis. In the remainder of this section, we sum-
marize the behaviour of transmission chains consisting
of a sequence of the Bayesian agents (Griffiths & Kalish
2005, 2007; Kirby et al. 2007).
(a) Chains of Bayesian agents
Following the schema shown in figure 1, we have a
sequence of agents, each of whom observes data d and
forms a hypothesis h about the knowledge of the previous
agent responsible for generating those data. What form
the data and hypotheses take will depend on the kind of
knowledge being transmitted: for concepts, data could
be instances of that concept and hypotheses rules that
characterize it; for social conventions, data could be
observations of the behaviour of others and hypotheses
the circumstances under which a convention applies; and
for languages, data could be a set of utterances and
hypotheses grammars. We assume that each learner
selects a hypothesis by sampling from a distribution
PLA(hjd ), where LA refers to some learning algorithm,
and generates data by sampling from a distribution
PPA(hjd ), where PA refers to some production algorithm.
Using hn and dn to represent the hypothesis formed and
the data generated by the nth learner, respectively, this
defines a stochastic process on (hn, dn) pairs.
A first observation is that this process is a Markov
chain: a sequence of random variables in which each
variable depends only on that which precedes it. In our
case, the hypothesis–data pair (hn, dn) is independent of
all preceding pairs given (hnK1, dnK1). Marginalizing
out (i.e. summing over) either hypotheses or data
makes it possible to define Markov chains on just dn or
hn, respectively. It is often particularly convenient to
study the Markov chain on hypotheses. If the number
of hypotheses is finite, the probability of the nth learner
adopting hypothesis i given that the nK1th learner held
hypothesis j is given by the transition matrix Q,
with entries
qij Z Pðhn Z ijhnK1 Z j Þ
X The learning algorithms we will consider are based
Z PLA ðhn Z ijdÞPPA ðdjhnK1 Z j Þ; ð3:1Þ
d Bayes’ rule. ‘Learning’ in the present context refers
which will depend on the learning and production
algorithms adopted by the learners.
Reducing the process of cultural transmission to a
Markov chain makes it easy to ask questions about the
outcome of such a process. Provided the Markov chain
satisfies a set of easily checked conditions, it will
converge asymptotically to a stationary distribution
(Norris 1997). In the case of the Markov chain on
hypotheses identified above, this means that the
probability that the nth learner entertains a particular
hypothesis will converge to a fixed value as n becomes
Phil. Trans. R. Soc. B (2008)
large, regardless of the hypothesis entertained by the
first learner. Determining the consequences of using a
particular learning algorithm is thus a matter of
determining how that learning algorithm influences
the stationary distribution. This distribution can be
found numerically by computing the first eigenvector of
the transition matrix (such as the matrix Q defined in
equation (3.1), but in some cases it is also possible to
give an analytic characterization.
Transmission chains formed of the Bayesian agents
provide one case in which an analytic stationary
distribution can be obtained. If we use a probability
distribution over hypotheses P(h) to encode an agent’s
degrees of belief in each hypothesis before seeing the
data (known as the prior distribution), the corresponding
distribution PðhjdÞ after seeing the data d (known as the
posterior distribution) is obtained by applying Bayes’ rule
PðdjhÞPðhÞ
PðhjdÞ Z P ; ð3:2Þ
Pðdjh 0 ÞPðh 0 Þ
h 0 2H
where P(djh) (known as the likelihood) is the probability
of seeing the particular data d if the particular hypothesis
h is true, and the sum in the denominator ranges over
the set of all possible hypotheses, H. The Bayesian
inference provides a useful framework for exploring
questions about inductive biases, since the prior P(h)
effectively encodes the inductive biases of the agent,
being a source of additional information or constraints
that discriminate between hypotheses with equal like-
lihoods. Thus, hypotheses with lower prior probability
are harder to learn or remember, requiring stronger
evidence to achieve high posterior probability.
The assumption that the agents use Bayesian
inference reduces the psychological complexities of
learning to a single equation. At first glance, this might
appear to ignore a long tradition of work on under-
standing human learning by cognitive psychologists;
however, rather than ignoring that knowledge, our
approach merely characterizes human learning at a
higher level of abstraction, often referred to as the
‘computational level’ (Marr 1982). That is, we are
exclusively concerned with the outcome of learning but
have no commitment to a specific process by which it
occurs. Many available models of learning and skill
acquisition may provide helpful process instantiations
of the Bayesian agents in our computational level of
description, and formal equivalences exist between
some of these process models and Bayesian inference
(e.g. Ashby & Alfonso-Reese 1995).
on the posterior distribution produced by applying
to the choice of a hypothesis about the data, so perhaps
the simplest algorithm is to sample a hypothesis from
the posterior. Using this algorithm, the distribution
PLA(hjd ) becomes
PPA ðdjhÞPðhÞ
Psamp ðhjdÞ Z P ; ð3:3Þ
h 0 2HPPA ðdjh 0 ÞPðh 0 Þ
where we place no constraints on the production
algorithm PA, but assume that the learning algorithm
employed by the agents draws on accurate knowledge
of this distribution.2
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Inductive biases and cultural evolution
With these specific assumptions about the form of
the learning algorithm in hand, we are able to analyse
the stationary distribution of the resulting Markov
chain. Griffiths & Kalish (2005) showed that the
stationary distribution of the Markov chain on
hypotheses is the prior distribution, P(h). A more
extensive analysis performed by Griffiths & Kalish
(2007) also provided stationary distributions for
Markov chains on data and hypothesis–data pairs,
and pointed out a correspondence between the
latter and a Markov chain Monte Carlo algorithm
called Gibbs sampling (Geman & Geman 1984),
commonly used in Bayesian statistics. In a nutshell,
these mathematical results imply that irrespective of
the stimuli presented at the outset, the final result of
iterated learning across generations is the expression
of the learners’ inductive biases.
Convergence to the prior provides a simple answer
to the question of how the inductive biases of
individuals affect the outcome of cultural evolution. It
indicates that the probability that a particular
hypothesis—a language, religious concept or social
norm—will emerge as the result of being transmitted
from one person to another is simply the prior probabi-
lity of that hypothesis. This means that inductive
biases—the constraints on learning that characterize
the minds of individuals—will lie in a direct one-to-one
correspondence with the outcomes of knowledge
transmission. Returning to the various claims about
cultural evolution made above, this analysis is consist-
ent with Bartlett’s conclusions about serial reproduc-
tion revealing cultural biases, with the arguments of
Boyer (1994, 1998), Sperber (1996) and Atran (2001)
concerning the role of human cognition in shaping the
information being transmitted, and with the analysis of
linguistic universals as the direct outcome of con-
straints on language acquisition.3
Making what might seem like a small change to the
assumptions about the learning algorithm used by our
Bayesian agents has significant consequences. An
alternative to sampling from the posterior distribution
is to choose the hypothesis that has the highest
posterior probability (known as maximum a posteriori
or MAP estimation). In this case, the probability of
selecting a particular hypothesis becomes
( Pðh Z 1jd Z 1Þ Z
1; h maximizes Pðhjd Þ;
PMAP ðhjd Þf ð3:4Þ
0; otherwise;
where Pðhjd Þ is computed as in equation (3.3), and the
constant of proportionality is determined by the
number of maxima of Pðhjd Þ. Griffiths & Kalish
(2007) showed that in this case a small difference in
the prior P(h) can result in a big difference in the
stationary probability of a hypothesis. Kirby et al.
(2007) showed that moving from sampling to MAP
estimation increases the magnitude of the effect of the
prior on the outcome of knowledge transmission, with
hypotheses that are slightly favoured by the prior being
over-represented in the stationary distribution. These
results paint a slightly different picture of the relation-
ship between inductive biases and cultural universals,
showing that weak inductive biases can be magnified by
the process of cultural transmission to produce strong
Phil. Trans. R. Soc. B (2008)
T. L. Griffiths et al. 3507
universals. This is still consistent with the claims of
psychologists and anthropologists about the import-
ance of cognitive factors in cultural evolution.
However, it undermines the inference from cultural
universals to equivalently strong constraints on
learning, which is part of the traditional interpretation
of linguistic universals: if weak biases can be magnified
by cultural evolution, then we no longer need to
postulate strong constraints to account for the consist-
ency observed in human languages.
(b) A simple example: two hypotheses
We illustrate the dynamics of the Bayesian transmission
chains with a simple example. In this example, we
assume that agents choose between two hypotheses by
sampling from their posterior distributions. A similar
example covering both sampling and MAP estimation
is analysed in detail by Griffiths & Kalish (2007).
The case of two hypotheses naturally maps onto a
variety of simple pieces of knowledge that might be
transmitted across generations, such as whether the verb
in a sentence precedes the object, a certain class of foods
is considered sacred or to tip taxi drivers. Inductive
biases from a variety of sources, from innate constraints
on language learning to the social perception of tipping,
could influence the transmission of this knowledge.
Using numbers to denote hypotheses, we can summarize
the prior distribution over these hypotheses by using p to
designate P(hZ1). Each agent in a chain has the
opportunity to observe a piece of data generated by the
previous agent, such as a set of utterances, a labelling of
sacred objects or some tipping behaviour. To simplify, we
will assume that this piece of data can also take on two
values and that these values are indicative of the
hypothesis entertained by the agent. This can be done
by taking Pðd Z kjhZ kÞZ 1Ke for k 2 f1; 2g, where e
is a parameter indicating the amount of noise in
transmission.
These assumptions provide us with all the infor-
mation we need to compute the transition matrix of the
Markov chain on hypotheses. The prior and likelihood
specified by p and e can be substituted into equation
(3.2) to give the posterior distributions,
ð1KeÞp
;
ð1KeÞp C eð1KpÞ
ep
Pðh Z 1jd Z 2Þ Z ;
ep C ð1KeÞð1KpÞ
where the probabilities for hZ2 are obtained from the
fact that the posterior sums to 1. Substitution into
equation (3.1) can be used to compute the transition
matrix, summing over the values d 2 f1; 2g. Since
probabilities sum to 1, we need to specify only two of
the entries of Q, such as q12 and q21, to give the full
transition matrix. An elementary calculation yields
q12 Z cp q21 Z cð1KpÞ; ð3:5Þ
where cZeð1KeÞð1=ð1KeKpC2epÞC1=ðeCpK2epÞÞ.
This indicates that the probability of moving from
hypothesis 2 to 1 is proportional to the prior probability
p, but the constant of proportionality is strongly
influenced by the noise rate e, increasing as e increases.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3508 T. L. Griffiths et al. Review. Inductive biases and cultural evolution
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 1 2 3 4 5 6 7 8 9
Figure 2. Dynamics of the probability of an agent adopting
hypothesis 1 as a function of the number of generations of
transmission. As the number of generations increases, the
probability of choosing h1 converges to the prior probability,
pZ0.2. The noise parameter e determines the rate of
convergence, with eZ0.01 (solid lines) converging more
slowly than eZ0.05 (dotted lines).
The transition matrix can be used to characterize the
dynamics and asymptotic consequences of cultural
transmission. The probability that an agent chooses a
particular hypothesis after n iterations is given by Qnp,
where p is a vector specifying the distribution over
hypotheses used to generate the first piece of data.
Figure 2 shows how this quantity evolves over time for
pZ0.2 and e2{0.01,0.05}. Regardless of whether the
first piece of data was generated from hypothesis 1 or 2,
just 10 iterations are sufficient to bring the probability
that an agent selects a hypothesis close to the prior
probability p. Increasing the value of e (and hence the
noise in the transmission) increases the rate of
convergence, making it easier for an agent to entertain
a hypothesis different from that of the previous agent.
The first eigenvector of Q is a vector q such that
QqZq. It makes intuitive sense that this should be the
stationary distribution of the Markov chain, since this
defines a distribution that does not change through
further application of the stochastic process defined by
Q (i.e. by definition of eigenvectors, QnqZq for all n).
Since q2 Z 1K q1 , we can reduce this definition to an
equation in a single variable,
ð1K q21 Þq1 C q12 ð1K q1 Þ Z q1 ;
which has the solution q1 Z q12 =ðq12 C q21 Þ. Substituting
the values for q12 and q21 from equation (3.5) into this
solution, we obtain q1 Z p. This indicates that the
stationary probability of hypothesis 1 is p, being equal
to its prior probability and consistent with the
convergence shown in figure 2.
(c) Summary
Transmission chains provide a simple way to study one
of the basic forces in cultural evolution—the way that
knowledge changes when transmitted from person to
person. This simplicity is paralleled in the mathemat-
ical analysis of such systems that reduce to Markov
chains. When the chain is composed of Bayesian
agents, we can make precise predictions about the
Phil. Trans. R. Soc. B (2008)
effects of inductive biases (expressed in the priors of
those agents) on knowledge transmission: the prob-
ability that an agent considers a hypothesis will
converge to the prior probability of that hypothesis.
We next examine whether these predictions are borne
out in the laboratory.
4. SIMULATING CULTURAL EVOLUTION IN
THE LABORATORY
Empirical tests of the idea that transmission chains
converge to the agents’ prior distributions face two
obstacles. First, we must know what the priors are, so
that we can recognize how closely they are approxi-
10 mated by the stationary distribution. Second, we must
be able to determine when (and if ) a chain has
converged. The first constraint led us to consider two
simple tasks for which previous research provided
strong evidence as to the general structure of the prior.
The second constraint led us to a design that employed
multiple chains starting from different initial states.
Convergence has occurred when all chains produce
similar results despite their diverse initial conditions.
(a) Learning categories
The simplest example of this method, and perhaps the
best instance of a known prior in an appropriate domain,
is a study in which people learned to extend a partially
specified category to a set of novel items (Griffiths et al.
2008, Experiment 1B). The items all varied on three
binary dimensions and the categories divided the eight
items into two classes of four. If we do not distinguish
structures that differ only in the assignment of physical
features to the binary dimensions, there are only six
types of such categories (figure 3a). To illustrate, if the
three binary dimensions defined geometric objects by
shape (e.g. circle or square), size (e.g. small or large) and
colour (e.g. black or white), then a type I category might
differentiate all squares (regardless of size or colour)
from the circles, whereas a type II category might pick
out white squares and differentiate them from black
circles (regardless of size).
Psychological research has told us a good deal about
how people learn these categories. In particular,
Shepard et al. (1961) showed that the six types of
categories have a canonical ordering of difficulty, with
types I and II being significantly easier to learn than the
ð3:6Þ others. The robustness of this finding (e.g. Nosofsky
et al. 1994) suggests that it is an effective index of the
prior over the six category types: the more difficult a
category is, the more data it requires to learn and hence
the lower its prior probability.
We used these category types to explore whether
people’s inductive biases—reflected in the difficulty-
of-learning results—would influence the outcome of
cultural transmission. Our stimuli were ‘amoebae’
whose nuclei varied along the three dimensions of
shape, size and colour mentioned above (after Feldman
2000). People were asked to make inferences about
‘species’ of amoebae based on examples. On each trial
of the experiment, a participant was shown three
amoebae that were stated to belong to a species, and
asked to identify the fourth amoeba belonging to that
species. To do so, all possible four-item categories that
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Review. Inductive biases and cultural evolution T. L. Griffiths et al. 3509

(a) (b) 0.12 (i) (ii)

(i) (ii) (iii)
0.10

selecting concept
probability of
0.08

0.06
(iv) (v) (vi)
0.04

0.02

0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
iteration iteration
Figure 3. Transmission chains for categories. (a) If we consider categories that are sets of four objects defined on three binary
dimensions and ignore the assignment of the dimensions to the physical properties of those objects, there are just six possible
category types (i–vi), types I–VI (Shepard et al. 1961). Each of the six types is illustrated on a cube, where each dimension of the
cube corresponds to one of the binary dimensions and the vertices are the eight objects. Filled circles represent members of an
example category of that type. Type I categories are defined on one dimension; type II uses two dimensions; types III, IV and V are
one dimension plus an exception and type VI uses all three dimensions. (b) Transmission chains were constructed by showing
people three objects drawn from a category and asking them to indicate, from a set of possible alternatives, which object completed
the set. The objects seen by the next person were selected at random from the set selected by the previous person. The probability
with which people selected categories of the six types changes as a function of the number of generations of a transmission chain, as
predicted by a Bayesian model using a prior estimated from human learning data. In particular, the probabilities of types I and VI
increase and decrease, respectively. (i) Human participants and (ii) Bayesian model (circles, type I; crosses, type II; triangles, type
III; squares, type IV; five-point stars, type V; six-point stars, type VI ). Further details are provided in Griffiths et al. (2008).

contained the three original amoebae and one other
amoeba were presented to the participant who selected
the category deemed most likely. Formally, the three
original amoebae are the data d and the response
alternatives are the hypotheses h. Participants were
implicitly being asked to compute pðhjd Þ and use it to
select one of the alternatives.
Each of the participants in the experiment
completed a series of trials, of which a subset were
linked to the responses of other people via transmission
chains. Specifically, the participants were randomly
grouped into seven ‘families’ of 10 generations each,
with responses transmitted between members of each
family. For the first participant in each family, the
amoebae seen on each trial were sampled uniformly at
random from the set of four matching a category
structure of one of the six types, with the six types
appearing with equal probability. The amoebae seen by
the next participant were then sampled from the set of
four selected by the first participant and so forth.
Under the mathematical analysis presented above,
the frequency of each category type in each generation
should come to approximate the prior as the number of
generations increases. This is precisely what was
observed empirically: the frequency of type I concepts
increased and type VI decreased over the course of
the experiment, and types I and II dominated responses
by the end of the experiment (figure 3b). The use of a
finite hypothesis space made it possible to compute
a full transition matrix for this Markov chain, and
the numerical predictions of the resulting Bayesian
model were strongly consistent with the observed
data (figure 3b).
(b) Learning functions
In contrast to the limited set of hypotheses available
to learners with the concepts described above,
most inductive problems allow for a vast number of
hypotheses. One such task is function learning, where a
metric stimulus value (such as the dosage of a drug or
driving speed) is related to a metric criterion (such as the
response to the drug or stopping distance). Such
relationships can have arbitrary complexity, but people
nonetheless appear to have strong priors over the space of
possible relationships. Kalish et al. (2004), in reviewing
the literature on function learning, observed that people
generally assume (and are the quickest to learn)
increasing linear functions where the criterion increases
in direct proportion to the stimulus. This is consistent
with an inductive bias that favours such functions.
Exploiting knowledge about human inductive biases
for this task, Kalish et al. (2007) conducted an
experiment in which people formed a transmission
chain for function concepts. In this experiment, each
generation of participants received 50 trials of training
on a single function. On each trial, the value of the
stimulus was presented as a visual magnitude, being the
width of a horizontal bar on a computer screen.
Participants responded by adjusting the height of a
vertical bar and then received corrective feedback (by
displaying the correct magnitude next to the response
bar). After training, participants responded to 100
stimuli that covered the entire possible range of
magnitudes without receiving feedback.
As in the experiment described above, the data seen
by the participants were influenced by the responses of
other participants. Participants were arranged into
eight families of nine generations, for each of four
conditions. The conditions differed with respect to the
function used to generate the training data seen by the
first generation of participants: those initial values were
drawn either from a positive linear, negative linear or
quadratic function, or entirely at random. For example,
a participant trained on the negative linear function
would see a series of training pairings where large
stimulus values (i.e. long bars) were paired with small

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3510 T. L. Griffiths et al. Review. Inductive biases and cultural evolution

(a) (i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x)

(b)

(c)

(d )

Figure 4. Representative results for transmission chains with human participants in which people learn functions. (a–d ) Each
row shows a single chain. (i) The (x, y) pairs were presented to the first participant in the chain, being represented as the width
and height of horizontal and vertical rectangles, respectively. Participants then made predictions of the value of y for new x values
((ii) nZ1, (iii) nZ2, (iv) nZ3, (v) nZ4, (vi) nZ5, (vii) nZ6, (viii) nZ7, (ix) nZ8, (x) nZ9). These predictions formed the
(x, y) pairs given to the next person in the chain, whose data appear in (ii)–(x) and so forth. Consistent with the previous research
exploring human inductive biases for function learning, chains produced linear functions with mostly positive slopes, regardless
of whether they were initialized with (a) a positive linear function, (b) a negative linear function, (c) a nonlinear function or (d ) a
random collection of points.

criterion values (i.e. short bars) and vice versa. The
responses of each participant on 50 of the test trials
were taken as the data used to train the participant in
the next generation of that family.
Representative families from the four conditions are
shown in figure 4. Two features of the data from these
chains are immediately apparent. First, striking
changes in the stimulus–criterion functions across
generations were observed, but only sporadically.
This indicates that people’s acquired functions were
generally very easy for the next generation to learn.
Second, notwithstanding the dramatic differences
between functions at the outset, across generations all
of the initial functions gradually disappeared and
transited into only one of two stable functions: positive
linear (28 out of 32 families) and negative linear (4 out
of 32), both with approximately unit slope. These
results are consistent with the previous work suggesting
that people’s priors are centred on positive and negative
linear functions and they support the predictions of our
formal analysis.
(c) Summary
Laboratory experiments involving transmission chains
for concepts that have been extensively studied by
psychologists provide a direct test of the predictions of
our formal framework. By using categories and
functions—concepts for which human inductive biases
are well understood—we were able to investigate
whether these biases influence the outcome of knowl-
edge transmission. The results support the conclusion
that knowledge transmission converges to an equili-
brium determined by the inductive biases of learners,
with categories and functions that people find easier to
learn becoming more prevalent across generations.
Flynn (2008) reports an analogous result with small
children, who very quickly discard irrelevant infor-
mation when transmitting a sequence of problem-
solving moves to an observer in the next generation.
Our laboratory results have implication for views of
human cultural evolution. In particular, the data are
consonant with the view that cultural representations
tend to be ‘recurrent’—that is, many aspects of culture
transcend beyond isolated times and places (e.g. Boyer
1998). Our repeated demonstrations that inductive
biases determine the final outcome of knowledge
transmission provide an empirical foundation for
claims by anthropologists and psychologists that
human cognitive capacities will influence the ideas
that appear in human societies, such as Boyer’s (1998)
claim that religious concepts are influenced by people’s
‘intuitive ontologies’—i.e. the distinctions they draw
between classes of objects from a very early age.
5. RELATING CULTURAL AND BIOLOGICAL
EVOLUTION
We next consider some connections between the
theoretical and empirical analyses presented thus far
and mathematical models of biological evolution.
These connections generalize our results beyond the
simple case of transmission chains. Mathematical
models of biological evolution are often applied to
cultural evolution (Cavalli-Sforza & Feldman 1981;
Boyd & Richerson 1985), and it is common to see both
informal (Deacon 1997; Kirby 1999) and formal
(Nowak et al. 2002) analogies between languages and
genotypes as objects of evolution. We first discuss how
our results relate to standard analyses of evolutionary
dynamics, by showing that the evolution of population
proportions in the absence of selection is intimately
related to the behaviour of transmission chains. We
then discuss what this connection tells us about the role
of selection in cultural evolution.
(a) Transmission chains and the replicator
dynamics
The basic model of deterministic evolution is based on
the replicator dynamics (e.g. Hofbauer & Sigmund

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Review. Inductive biases and cultural evolution T. L. Griffiths et al. 3511

1998). Let xi denote the proportion of a population of additional justification for the use of transmission
agents entertaining hypothesis i at a given moment t, chains in studying cultural evolution: the parallel
and qij denote the probability that a learner chooses between the stationary distributions of such chains
hypothesis i after seeing the data generated from and the equilibria of the replicator dynamics in
hypothesis j, as defined in equation (3.1). If we assume populations provides a way to gather clues about the
that each learner learns from a random member of the behaviour of populations using a paradigm that is easily
population, then the population proportions evolve as simulated in the laboratory.
dxi X
Z fj qij xj Kfxi ; ð5:1Þ (b) Inductive biases can overwhelm selective
dt j pressures
In addition to indicating how transmission chains can
where fj is the fitness
P of people who subscribe to
inform the study of cultural evolution more broadly,
hypothesis j; fZ k fk xk is the mean fitness; and the
this connection provides us with a way to generalize our
second
P term on the right-hand side ensures that
mathematical results to cases where selective forces also
i xi Z 1. In biological evolution, fitness reflects the
influence the adoption of hypotheses. This can allow us
number of offspring produced by an individual of a
to evaluate whether inductive biases can play a more
particular type. In cultural evolution, it is more natural
significant role in cultural evolution than selection, as
to interpret fitness as influencing the probability with
suggested by Sperber (1996), Boyer (1998) and Atran
which an individual chooses an agent from the previous
(2001), or whether selection is the more powerful force,
generation as a source of data. If agents are selected with
as argued by Henrich & Boyd (2002). While obtaining
probability determined by fj, the same dynamics hold.4
general analytical results is difficult, we can at least gain
Equation (5.1) has been extensively applied to
an idea of how these forces interact by returning to our
cultural evolution for the case of languages, in the
example with just two hypotheses.
form of the ‘language dynamical equation’ explored by
With the two hypotheses, the fact that x 1Cx 2Z1
Nowak et al. (2001, 2002). In this work, fitness is
means that we can work with just one variable. We will
typically assumed to be a function of how well speakers
use x 1, the proportion of agents choosing hypothesis 1,
of a particular language can communicate with the
and denote this x for simplicity. In §3b, we defined the
population at large, implementing a selection pressure
matrix Q as a function of the prior probability of
for communication. If we instead assume that all
hypothesis 1, p, and the noise rate, e. In the neutral
speakers have equal fitness, fjZ1, equation (5.1)
model from §5a, where the fitness of both hypotheses is
simplifies to
equal (i.e. each generation chooses an agent to learn
dxi X from at random from the previous generation with
Z qij xj K xi ; ð5:2Þ uniform probabilities), the equilibrium of the system is
dt j given by finding a value of x such that equation (5.2) is
equal to zero. It is straightforward to show that this
which is a linear dynamical system. This is a ‘neutral’
is equivalent to solving equation (3.6), and thus the
model, in which there are no selective forces favouring
equilibrium is given by xZp. The critical question is
one language or hypothesis over another. A special
how this equilibrium is affected by selection, as
case of this model was analysed by Komarova &
represented by unequal fitness for the two hypotheses.
Nowak (2003).
We will assume that the fitness of hypothesis 1 is
The neutral model characterizes the evolution of a
f1Zs and hypothesis 2 has constant fitness f2Z1. We
population in the absence of selection, and thus
are interested in the case where sO1. This higher fitness
provides a valuable null hypothesis against which to
might reflect higher social status accorded to those
evaluate claims about selective forces, as well as a way
who adopt the hypothesis, greater success in solving
to study the effects of mutation. It also gives us a way to
problems posed by the environment as a consequence
connect the replicator dynamics to transmission chains.
of having this belief or some other indicator of success
The asymptotic behaviour of this linear dynamical
that might make others more likely to try to learn from
system is straightforward to analyse: it converges
these ‘fit’ individuals. The equilibrium of the resulting
towards an equilibrium at the first eigenvector of the
system is given by finding x such that equation (5.1) is
transition matrix Q (for details, see Griffiths & Kalish
equal to zero. Simplification for the case of the two
2007). This means that the neutral form of the
hypotheses reduces this to the quadratic equation
replicator dynamics displays asymptotic behaviour
that is very similar to that of transmission chains dx
involving discrete generations of single learners. The Z ð1KsÞx2 C ðð1K q21 Þs K q12 K 1Þx C q12 ; ð5:3Þ
dt
key difference is in the nature of the quantities that
converge: with discrete generations of single learners, it which can be solved by standard methods to find an
is the probability with which a particular learner equilibrium for a particular choice of s, q12 and q21.
entertains hypothesis i that converges to the stationary Figure 5a shows how the equilibrium changes as a
probability; under the replicator dynamics, it is the function of s for pZ0.2 and e2 0:01; 0:05. As might be
proportion of the population that entertains hypothesis i expected, increasing s increases the representation of
that converges to this probability. hypothesis 1 in the equilibrium solution.
The results from the previous sections characterize We can use equation (5.3) to explore the relative
the consequences of cultural evolution not only for contributions of the prior probability of a hypothesis p
individuals but also for populations. This provides an and the strength of selection s on the equilibrium of this

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3512 T. L. Griffiths et al. Review. Inductive biases and cultural evolution
(a)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
1 10 102
Figure 5. The interaction of selection with inductive biases. (a) Increasing the selective pressure in favour of hypothesis 1
increases the representation of that hypothesis in the population. The equilibrium probability of hypothesis 1 for pZ0.2,
e2{0.01,0.05} (solid line, dotted line, respectively), and a range of values of the selective pressure s are shown. (b) Threshold on
s for hypothesis 1 to obtain an equilibrium probability greater than 0.5 as a function of p and e. For values of p and e such that
q21O0.5, no value of s produces an equilibrium favouring hypothesis 1.
system. When sZ1, we know that the equilibrium value
of x will be p. If p is less than 0.5, the equilibrium will
be biased against h1. We might thus ask how large s will
have to be in order to overcome this bias, making the
equilibrium value of x greater than 0.5. The functions
shown in figure 5a indicate that this happens relatively
quickly for the values of p and e considered above, with
the equilibrium passing 0.5 for values of s not much
greater than 1. In appendix A, we show that the
threshold value of s is
1K2q12
sZ ; ð5:4Þ
1K2q21
provided q12 ! q21 ! 0:5. The first part of this condition
follows automatically from the fact that p! ð1KpÞ, but
the second part is more interesting. If q21 O 0:5, then
there is no value of s such that the equilibrium favours
hypothesis 1. Intuitively, if more than half the agents
learning from endorsers of hypothesis 1 adopt
hypothesis 2, there is no way that increasing the fitness
of hypothesis 1 can push the equilibrium past 0.5.
The requirement that q21 be less than 0.5 places
strong constraints on the values of p and e, which can
support equilibria favouring hypothesis 1. Figure 5b
shows how the threshold on s behaves as a function of p
and e. The threshold rapidly increases as p and e
approach values that make q21 close to 0.5, and any
value of p less than 0.5 has some value of e for which no
amount of selection will yield an equilibrium favouring
hypothesis 1. For example, pZ0.2 results in reasonable
thresholds on s for small values of e of the kind used in
the examples above, but taking eZ0.16 allows the prior
to have a sufficiently strong influence on the inferences
of the agents that no amount of selection can overcome
it. These results thus illustrate how inductive biases can
lead a population to an equilibrium that reflects those
biases, even if there are other social or environmental
factors that strongly favour a different outcome.
6. CONCLUSION
At the start of this paper, we asked a very general
question concerning cultural evolution, namely how
people’s inductive biases (their knowledge and
Phil. Trans. R. Soc. B (2008)
(b)
100
0.5
10
0.4
0.3
el,
lev
1 0.2
ise
0.5 0.4 0.1
no
0.3 0.2
0.1 0
expectations) affect the transmission of languages and
concepts. We analysed this general question in the more
circumscribed context of transmission chains, in which
knowledge is passed from one person to the next.
Within this paradigm, the general question about
inductive biases becomes the question of how these
biases change the information being transmitted. We
provided two converging answers: one based on an
abstract mathematical analysis and the other based on
evidence from behavioural experiments. Both answers
suggest that in many circumstances, transmission
chains converge to an equilibrium that reflects people’s
inductive biases.
The mathematical results we summarized apply to
learning algorithms based on the Bayesian inference in
which observed data are combined with inductive
biases expressed as a prior distribution over hypotheses.
In this case, the probability with which a person at the
end of a transmission chain selects a particular
hypothesis converges to a distribution determined by
the prior. The data from several experiments were
found to be in accord with this prediction: after
transmission across a fairly small number of gener-
ations, people’s responses approximated their known
inductive biases in terms of the proportions with which
they chose competing hypotheses, for both categorical
concepts and continuous functions. In both cases,
people’s biases were established independently through
previous experiments, and, with categorical concepts,
direct measurement within the same experiment. The
fact that the products of our transmission chains were
consistent with these inductive biases suggests that the
way people behave in these tasks is sufficiently similar
to the Bayesian inference to permit the conclusion that
our mathematical results accurately characterize the
dynamics of cultural transmission.
These mathematical analyses and experimental
results imply two strong statements about cultural
evolution in general. First, they indicate that the power
of inductive biases can trump the potential stabilization
provided by faithful learning. Recall that in the
function learning experiment of Kalish et al. (2007),
the first generation of learners was presented with
widely different functions, ranging from positive linear
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Inductive biases and cultural evolution
to quadratic and entirely random—nonetheless, after
only four or five generations, those different starting
points had been absorbed and responses converged to a
function that remained stable across further gener-
ations. Learning from the data produced by the
previous participant was thus insufficient to guarantee
faithful cultural transmission, with the influence of
inductive biases accumulating with each generation.
Second, the analyses reported in §5 suggest that prior
biases may even trump selection pressures in determin-
ing the dynamics of cultural evolution: a highly
counter-intuitive hypothesis will fail to dominate a
population, even if there are strong advantages to
adopting it. These results suggest that one of the
consequences of cultural transmission will be the
adaptation of concepts and languages to match
human inductive biases.
This work was supported by grants 0704034 and 0544705
from the National Science Foundation (to T.L.G. and
M.L.K., respectively) and by a Discovery Project grant
from the Australian Research council to S.L. We thank four
anonymous reviewers for their comments.
ENDNOTES
1
We refer to these constraints as inductive biases by analogy to the
machine learning literature, in which the inductive bias of a learning
algorithm is the set of assumptions that lead the algorithm to select
one hypothesis over another ( Mitchell 1997). By considering human
learning as one such algorithm, we use inductive biases to refer to all
factors, such as prior knowledge or expectations, that make ideas
easier to learn or remember, whether they are derived from innate
constraints or from experience.
2
Note that PA refers to the agent from the previous generation in this
equation, as the data are the utterances produced by the previous
learner. We assume that PA and LA are the same across all learners,
which amount to the assumption that the prior distribution P(h) is
also shared.
3
It is worth emphasizing that this analysis only justifies a connection
between the prior and the consequences of knowledge transmission: it
does not indicate where the inductive biases expressed in the prior
distribution of hypotheses come from, and thus does not in itself
provide justification for the claims about modular cognitive
architectures or innate domain-specific constraints on linguistic or
ontological knowledge, which are associated with these positions (for
further discussion of this point, see Griffiths & Kalish 2007).
4
While much recent work applying these models (e.g. Nowak et al.
2002) has focused on the effects of frequency-dependent selection, we
restrict ourselves here to the case where fitness does not depend on
the composition of the population. Exploring the consequences of
Bayesian learning in the context of frequency-dependent selection is
an exciting direction for future work.
APPENDIX A
To derive the threshold on s, we observe that dx/dt is a
negative quadratic function in x, and takes positive
value when xZ0 (dx/dtZq12) and negative values when
xZ1 (dx/dtZKq21s). It follows that dx/dtZ0 at exactly
one point in [0,1]. When sZ1, this point is p. If p!0.5,
then we can ask what value of s is required such that the
crossing point is greater than 0.5. The derivative of
dx/dt with respect to s is Kx2 C ð1K q21 Þx, which is
positive at 0.5 provided q21!0.5. Solving for s such that
dx/dtZ0 when xZ0.5 thus gives us a threshold above
which the equilibrium value of x will be greater than
0.5. Substituting 0.5 for x into 9 and solving for s gives
Phil. Trans. R. Soc. B (2008)
T. L. Griffiths et al. 3513
equation (5.4). When q21O0.5, the derivative of dx/dt
with respect to s at 0.5 is negative. Consequently,
increasing s can only decrease dx/dt at this point. We
know that dx/dt at 0.5 is negative when sZ1, so no sO1
can result in an equilibrium in which the probability of
hypothesis 1 is 0.5 or greater.
REFERENCES
Anderson, J. R. 1990 The adaptive character of thought.
Hillsdale, NJ: Erlbaum.
Anderson, J. R. 1991 The adaptive nature of human
categorization. Psychol. Rev. 98, 409–429. (doi:10.1037/
0033-295X.98.3.409)
Anderson, J. R. & Milson, R. 1989 Human memory: an
adaptive perspective. Psychol. Rev. 96, 703–719. (doi:10.
1037/0033-295X.96.4.703)
Ashby, F. G. & Alfonso-Reese, L. A. 1995 Categorization as
probability density estimation. J. Math. Psychol. 39,
216–233. (doi:10.1006/jmps.1995.1021)
Atran, S. 2001 The trouble with memes: inferences versus
imitation in cultural creation. Hum. Nat. 12, 351–381.
(doi:10.1007/s12110-001-1003-0)
Atran, S. 2002 In gods we trust: the evolutionary landscape of
religion. Oxford, UK: Oxford University Press.
Bartlett, F. C. 1932 Remembering: a study in experimental and
social psychology. Cambridge, UK: Cambridge University
Press.
Bickerton, D. 1981 Roots of language. Ann Arbor, MI:
Karoma.
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Boyer, P. 1994 The naturalness of religious ideas: a cognitive
theory of religion. Berkeley, CA: University of California
Press.
Boyer, P. 1998 Cognitive tracks of cultural inheritance: how
evolved intuitive ontology governs cultural transmission.
Am. Anthropol. 100, 876–889. (doi:10.1525/aa.1998.100.
4.876)
Brighton, H. 2002 Compositional syntax from cultural
transmission. Artif. Life 8, 25–54. (doi:10.1162/10645
4602753694756)
Briscoe, E. (ed.) 2002 Linguistic evolution through language
acquisition: formal and computational models, Cambridge,
UK: Cambridge University Press.
Brown, D. E. 1991 Human universals. New York, NY:
McGraw-Hill.
Buchanan, T. W. 2007 Retrieval of emotional memories.
Psychol. Bull. 133, 761–779. (doi:10.1037/0033-2909.
133.5.761)
Caldwell, C. & Millen, A. E. 2008 Studying cumulative
cultural evolution in the laboratory. Phil. Trans. R. Soc. B
363, 3529–3539. (doi:10.1098/rstb.2008.0133)
Cavalli-Sforza, L. L. & Feldman, M. W. 1981 Cultural
transmission and evolution. Princeton, NJ: Princeton
University Press.
Chomsky, N. 1965 Aspects of the theory of syntax. Cambridge,
MA: MIT Press.
Comrie, B. 1981 Language universals and linguistic typology.
Chicago, IL: University of Chicago Press.
Deacon, T. W. 1997 The symbolic species: the co-evolution of
language and the brain. New York, NY: Norton.
DeGraff, M. (ed.) 1999 Language creation and language
change: creolization, diachrony, and development,
Cambridge, MA: MIT Press.
Feldman, J. 2000 Minimization of Boolean complexity in
human concept learning. Nature 407, 630–633. (doi:10.
1038/35036586)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3514 T. L. Griffiths et al. Review. Inductive biases and cultural evolution
Flynn, E. 2008 Investigating children as cultural magnets: do
young children transmit redundant information along
diffusion chains? Phil. Trans. R. Soc. B 363, 3541–3551.
(doi:10.1098/rstb.2008.0136)
Galantucci, B. 2005 An experimental study of the emergence
of human communication systems. Cogn. Sci. 29,
737–767. (doi:10.1207/s15516709cog0000_34)
Garrod, S., Fay, N., Lee, J., Oberlander, J. & Macleod, T.
2007 Foundations of representation: where might graphi-
cal symbol systems come from? Cogn. Sci. 31, 961–988.
(doi:10.1080/03640210701703659)
Geman, S. & Geman, D. 1984 Stochastic relaxation, Gibbs
distributions, and the Bayesian restoration of images.
IEEE Trans. Pattern Anal. Mach. Intell. 6, 721–741.
Gopnik, A. & Meltzoff, A. N. 1997 Words, thoughts, and
theories. Cambridge, MA: MIT Press.
Greenberg, J. (ed.) 1963 Universals of language, Cambridge,
MA: MIT Press.
Griffiths, T. L. & Kalish, M. L. 2005 A Bayesian view of
language evolution by iterated learning. In Proc. Twenty-
Seventh Annual Conf. of the Cognitive Science Society (eds
B. G. Bara, L. Barsalou & M. Bucciarelli), pp. 827–832.
Mahwah, NJ: Erlbaum.
Griffiths, T. L. & Kalish, M. L. 2007 A Bayesian view of
language evolution by iterated learning. Cogn. Sci. 31,
441–480.
Griffiths, T. L. & Tenenbaum, J. B. 2005 Structure and
strength in causal induction. Cogn. Psychol. 51, 354–384.
(doi:10.1016/j.cogpsych.2005.05.004)
Griffiths, T. L., Steyvers, M. & Tenenbaum, J. B. 2007 Topics
in semantic association. Psychol. Rev. 114, 211–244.
(doi:10.1037/0033-295X.114.2.211)
Griffiths, T. L., Christian, B. R. & Kalish, M. L. 2008 Using
category structures to test iterated learning as a method
for identifying inductive biases. Cogn. Sci. 32, 68–107.
(doi:10.1080/03640210701801974)
Hawkins, J. (ed.) 1988 Explaining language universals. Oxford,
UK: Blackwell.
Henrich, J. & Boyd, R. 2002 Culture and cognition: why
cultural evolution does not require replication of rep-
resentations. Cult. Cogn. 2, 87–112. (doi:10.1163/
156853702320281836)
Hofbauer, J. & Sigmund, K. 1998 Evolutionary games and
population dynamics. Cambridge, UK: Cambridge
University Press.
Hudson-Kam, C. L. & Newport, E. L. 2005 Regularizing
unpredictable variation: the roles of adult and child
learners in language formation and change. Lang. Learn.
Dev. 1, 151–195. (doi:10.1207/s15473341lld0102_3)
Kalish, M., Lewandowsky, S. & Kruschke, J. 2004 Popu-
lation of linear experts: knowledge partitioning and
function learning. Psychol. Rev. 111, 1072–1099. (doi:10.
1037/0033-295X.111.4.1072)
Kalish, M. L., Griffiths, T. L. & Lewandowsky, S. 2007
Iterated learning: intergenerational knowledge trans-
mission reveals inductive biases. Psychon. Bull. Rev. 14,
288–294.
Kirby, S. 1999 Function, selection and innateness: the emergence
of language universals. Oxford, UK: Oxford University
Press.
Kirby, S. 2001 Spontaneous evolution of linguistic structure:
an iterated learning model of the emergence of regularity
and irregularity. IEEE J. Evol. Comput. 5, 102–110.
(doi:10.1109/4235.918430)
Kirby, S., Dowman, M. & Griffiths, T. L. 2007 Innateness
and culture in the evolution of language. Proc. Natl Acad.
Sci. USA 104, 5241–5245. (doi:10.1073/pnas.06082
22104)
Phil. Trans. R. Soc. B (2008)
Komarova, N. L. & Nowak, M. A. 2003 Language dynamics
in finite populations. J. Theor. Biol. 221, 445–457. (doi:10.
1006/jtbi.2003.3199)
Marr, D. 1982 Vision. San Francisco, CA: W. H. Freeman.
Mesoudi, A. 2007 Using the methods of experimental social
psychology to study cultural evolution. J. Soc. Evol. Cult.
Psychol. 1, 35–58.
Mesoudi, A. & Whiten, A. 2008 The multiple roles of cultural
transmission experiments in understanding human cul-
tural evolution. Phil. Trans. R. Soc. B 363, 3489–3501.
(doi:10.1098/rstb.2008.0129)
Mitchell, T. M. 1997 Machine learning. New York, NY:
McGraw Hill.
Nichols, S. 2004 A fragment of the genealogy of norms.
Sentimental Rules 1, 118–141. (doi:10.1093/0195169344.
003.0006)
Norris, J. R. 1997 Markov chains. Cambridge, UK:
Cambridge University Press.
Nosofsky, R. M., Gluck, M., Palmeri, T. J., McKinley, S. C.
& Glauthier, P. 1994 Comparing models of rule-based
classification learning: a replication and extension of
Shepard, Hovland, and Jenkins (1961). Mem. Cognit. 22,
352–369.
Nowak, M. A., Komarova, N. L. & Niyogi, P. 2001 Evolution
of universal grammar. Science 291, 114–118. (doi:10.
1126/science.291.5501.114)
Nowak, M. A., Komarova, N. L. & Niyogi, P. 2002
Computational and evolutionary aspects of language.
Nature 417, 611–617. (doi:10.1038/nature00771)
Pinker, S. 2002 The blank slate: the modern denial of human
nature. New York, NY: Viking.
Senghas, A., Kita, S. & Özyürek, A. 2004 Children creating
core properties of language: evidence from an emerging
sign language in Nicaragua. Science 305, 1779–1782.
(doi:10.1126/science.1100199)
Shepard, R. N., Hovland, C. I. & Jenkins, H. M. 1961
Learning and memorization of classifications. Psychol.
Monogr. 75, 1–42.
Shiffrin, R. M. & Steyvers, M. 1997 A model for recognition
memory: REM: retrieving effectively from memory.
Psychon. Bull. Rev. 4, 145–166.
Smith, K. & Kirby, S. 2008 Cultural evolution: implications
for understanding the human language faculty and its
evolution. Phil. Trans. R. Soc. B 363, 3591–3603. (doi:10.
1098/rstb.2008.0145)
Smith, K., Kirby, S. & Brighton, H. 2003 Iterated learning: a
framework for the emergence of language. Artif. Life 9,
371–386. (doi:10.1162/106454603322694825)
Spelke, E. S., Breinlinger, K., Macomber, J. & Jacobson, K.
1992 Origins of knowledge. Psychol. Rev. 99, 605–632.
(doi:10.1037/0033-295X.99.4.605)
Sperber, D. 1985 Anthropology and psychology: towards an
epidemiology of representations. Man 20, 73–89. (doi:10.
2307/2802222)
Sperber, D. 1996 Explaining culture: a naturalistic approach.
Oxford, UK: Blackwell.
Sperber, D. & Claidiére, N. 2006 Why modeling cultural
evolution is still such a challenge. Biol. Theory 1, 20–22.
(doi:10.1162/biot.2006.1.1.20)
Steels, L. 2003 Evolving grounded communication for
robots. Trends Cogn. Sci. 7, 308–312. (doi:10.1016/
S1364-6613(03)00129-3)
Tenenbaum, J. B. & Griffiths, T. L. 2001 Generalization,
similarity, and Bayesian inference. Behav. Brain Sci. 24,
629–641. (doi:10.1017/S0140525X01000061)
Whiten, A. & Mesoudi, A. 2008 Establishing an experimental
science of culture: animal social diffusion experiments.
Phil. Trans. R. Soc. B 363, 3477–3488. (doi:10.1098/rstb.
2008.0134)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3515–3528

doi:10.1098/rstb.2008.0131
Published online 17 September 2008

Beyond existence and aiming outside the

laboratory: estimating frequency-dependent
and pay-off-biased social learning strategies
Richard McElreath1,*, Adrian V. Bell2, Charles Efferson3, Mark Lubell2,
Peter J. Richerson2 and Timothy Waring2
1
Department of Anthropology, and 2Department of Environmental Science and Policy,
University of California Davis, Davis, CA 95616, USA
3
Institute for Empirical Research in Economics, Blümlisalpstrasse 10, 8006 Zürich, Switzerland
The existence of social learning has been confirmed in diverse taxa, from apes to guppies. In order to
advance our understanding of the consequences of social transmission and evolution of behaviour,
however, we require statistical tools that can distinguish among diverse social learning strategies. In
this paper, we advance two main ideas. First, social learning is diverse, in the sense that individuals
can take advantage of different kinds of information and combine them in different ways. Examining
learning strategies for different information conditions illuminates the more detailed design of social
learning. We construct and analyse an evolutionary model of diverse social learning heuristics, in
order to generate predictions and illustrate the impact of design differences on an organism’s fitness.
Second, in order to eventually escape the laboratory and apply social learning models to natural
behaviour, we require statistical methods that do not depend upon tight experimental control.
Therefore, we examine strategic social learning in an experimental setting in which the social
information itself is endogenous to the experimental group, as it is in natural settings. We develop
statistical models for distinguishing among different strategic uses of social information. The
experimental data strongly suggest that most participants employ a hierarchical strategy that uses
both average observed pay-offs of options as well as frequency information, the same model predicted
by our evolutionary analysis to dominate a wide range of conditions.
Keywords: cultural evolution; social learning; quantitative methods

1. INTRODUCTION serious exchange between the evolutionary anthro-

Under a broad definition, social learning is common in
nature. The behaviour of conspecifics influences
individual behaviour through modification of the
environment, emulation of goals and imitation of
patterns (cf. Whiten & Ham 1992). This psychological
set of distinctions has directed years of research in
animal behaviour, especially the study of social learning
in non-human apes. Distinguishing between emulation
and imitation, and the interaction of the two (Horner &
Whiten 2005), has generated a literature testifying to
the breadth and diversity of social learning in nature
(Fragaszy & Perry 2003).
More recent high-profile experiments with chim-
panzees (Whiten et al. 2005, 2007) have demonstrated
that short-lived traditions can evolve in chimpanzee
social groups, backing up studies that claim that
behavioural variation among wild populations of
chimpanzees are ‘cultural’ (Boesch & Tomasello
1998; Whiten et al. 1999; Boesch 2003). While the
finding of short-lived socially transmitted traditions
may not be surprising to students of Galef ’s rat
experiments (Galef & Whiskin 1997), the findings
suggest that the time may be right to attempt a more
* Author for correspondence (mcelreath@ucdavis.edu).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
pology literature on social learning—which emphasizes
a toolbox of social learning strategies, such as majority
rule conformity and pay-off-biased learning (Boyd &
Richerson 1985; Henrich & McElreath 2003)—and the
animal literature—which tends to emphasize the
existence or not of culture.
There are at least two good reasons to try. First, non-
human animals may also have special-purpose social
learning strategies that combine and recombine
different kinds of social information, yet usually no
effort is made to look for these (Laland 2004). Finding
such cases of analogy (or possibly homology, in the case
of other apes) would potentiate advances in the general
understanding of the evolution of adaptations for
processing social information. Second, many biologists
and anthropologists remain sceptical of the evidence
of animal, and especially great ape, culture (Laland &
Janik 2006). This is partly a result of the difficulty
of inferring patterns of learning from cross sections of
behavioural variation. However, statistical tools
developed to study dynamic learning in human groups
can be leveraged to study diverse social learning
strategies in other animals, as well.
In this paper, we illustrate an approach for analysing
different strategies for combining social cues from
multiple conspecifics, in less poorly controlled settings.

3515 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3516 R. McElreath et al. Analysis on social learning strategies
We use a stylized evolutionary model to generate
broad predictions for which of several candidate
strategies we expect to find in nature and under what
conditions. We then apply these stylized predictions to
a laboratory experiment that allows participants great
flexibility in from whom and how they learn. Instead of
asking if social learning occurs, we develop likelihood
models that allow us to ask how participants socially
learn. While the precise example we present uses
very detailed information, the same approach can be
applied to more naturalistic contexts, in which
incomplete time series or purely cross-sectional data
are all that are available.
While neither the appreciation of strategic diversity
nor our model-based approach is particularly new in
itself, we think the combination is of value. The key
insight is that each social learning strategy implies
different outcomes, under at least some sets of available
information, both for each individual and entire groups
of individuals. This is not a new point (Cavalli-Sforza &
Feldman 1981; Boyd & Richerson 1985; Galef &
Whiskin 1997), but statistical approaches are usually
not up to the task of exploring it adequately. Those who
do study distinctions among strategies may be inclined
to rely upon highly controlled and artificial experi-
ments. Even when an experimenter is clever enough to
design a series of treatments that can carefully
distinguish among diverse strategies in the laboratory,
scientists will still debate the lessons of behaviour in the
wild. In order to resolve animal culture debates and
gain a more detailed behavioural understanding of
social learning, whether in humans or other animals,
we will need analytical approaches that do not require
precise experimental control of social information.
Another reason to develop statistical methods for less
controlled contexts is that part of the action in social
learning is evolution of behaviour, and experiments
that control social information do not allow us to study
these population-level effects nor how strategies are
adapted to them.
The general approach we suggest is to (i) nominate
a series of candidate social learning strategies, (ii)
translate each of these into an expression for the
conditional probability of behaviour, given an infor-
mational context for an individual animal, (iii) use these
expressions to generate likelihoods of observing field or
laboratory data, and (iv) compare the fits of these
strategies to the data with information theoretic criteria,
such as Akaike information criterion (AIC) or Bayesian
information criterion (BIC). Approaching the problem
as a task of discriminating among a toolbox of potential
strategies, rather than a task of demonstrating the
existence of social learning, may allow all of us to
squeeze more from both our experiments and field
studies than we previously imagined.
2. MANY WAYS TO LEARN SOCIALLY
There was a time when biology wondered if natural
selection occurred. Now no one—within evolutionary
biology—seriously questions the existence of natural
selection as an evolutionary force. Instead, we debate
its relative strength and character in different
environmental and biological contexts. Both sexual
Phil. Trans. R. Soc. B (2008)
(Kokko et al. 2006) and social selection ( Frank 2006)
have generated special literatures of theory and
evidence that testify to the subtle diversity of the
action of natural selection. One could seriously say
that there are many natural selections.
In a similar sense, there are many social learnings.
Psychologists and animal behaviourists have long
recognized taxonomic distinctions between, for
example, social facilitation and imitation (Zajonc
1965). But many of the highest profile publications
still address basic existence questions, asking if other
animals have human-like social learning and human-
like traditions or culture ( Whiten et al. 2005). These
publications are probably taking the right rhetorical
approach. Many anthropologists remain unconvinced
that chimpanzee or crow culture is much like human
culture (Boesch 2003).
However, many scientists have enough interest in
the details of social learning in humans, as well as other
animals, to step aside the ‘is it human enough?’ debate.
As social learning is diverse, it has diverse effects. Some
mechanisms generate rather short-lived traditions, if
any at all (Galef & Whiskin 1997). Human cultural
traditions can be both ephemeral and demonstrate
tremendous inertia (Richerson & Boyd 2005), depend-
ing in part upon the strategic diversity of social learning
and the details of the social context (Cavalli-Sforza &
Feldman 1981; Boyd & Richerson 1992). Studying the
mechanistic and algorithmic diversity of social learning
will be just as important as arguing that it exists, and
our hunch is that most researchers in both anthro-
pology and animal behaviour are prepared to move in
this direction.
In this section, we briefly review evolutionary work
on structurally different social learning strategies.
Most of this literature has been concerned with
human social and cultural learning (Boyd & Richerson
1985; Henrich & McElreath 2003), but there is no
reason these models cannot apply to other organisms
(Laland 2004). Before moving on to apply these
different strategies to experimental data, we hope
to convince the reader that it is worth asking, for
example, if chimpanzees also use majority rule social
learning or are guided by observed cues of others’
success. While no single strategy is imagined to
dominate at all times nor to exist in the absence of
individual learning, the dynamic consequences of each
strategy can be appreciated most easily by first
examining them in isolation.
(a) Unbiased social learning
One of the simplest social learning strategies is to select
a random target individual and copy his or her
behaviour. We call this kind of social learning
‘unbiased’, as it tends to maintain the frequencies of
different behaviour (Cavalli-Sforza & Feldman 1981;
Boyd & Richerson 1985). One adaptive advantage of
unbiased social learning is economizing on individual
learning costs (Boyd & Richerson 1985).
(b) Frequency-dependent social learning
When individuals can sample more than one con-
specific, a large family of frequency-dependent
strategies become possible. The most commonly
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Analysis on social learning strategies R. McElreath et al. 3517

(a) positive frequency dependence (b) positive frequency dependence

1.0 1.0

expected frequency of trait

0.8 0.8

frequency of trait
after imitation 0.6 0.6

0.4 0.4

0.2 0.2

0 0

(c) pay-off biased (d) pay-off biased + frequency dependence

1.0 1.0
expected frequency of trait

0.8 0.8

frequency of trait
after imitation

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1.0 5 10 15 20 25
frequency of trait before imitation time
Figure 1. (a,c) Instantaneous and (b,d ) evolutionary dynamics of frequency-dependent and pay-off-biased social learning.

studied of these is positive frequency dependence, which
preferentially copies the most common behaviour
variants in the sample. Such a strategy has very deep
intellectual roots, being studied formally at least as far
back as 1785, in Condorcet’s jury theorem (see Estlund
1994). Evolutionary treatments of positive frequency
dependence, ‘conformity’, emphasize its adaptive value
for individuals (Boyd & Richerson 1985; Henrich &
Boyd 1998).
Figure 1a,b plots the instantaneous and evolutionary
dynamics of positive frequency dependence. In
figure 1a, for any frequency of one of two alternative
learned behaviours on the horizontal axis, the solid
curve gives the expected frequency (or probability of
adoption) after social learning. If p is the value on the
horizontal axis, then pC pð1K pÞð2p K 1Þ is the value
on the vertical (Boyd & Richerson 1985—we re-derive
this function in §2e). The dashed line illustrates the
expected frequency under unbiased social learning. In
figure 1b, the evolution of behaviour within a
population of learners who practice positive frequency
dependence depends on whether the initial frequency
of behaviour is below or above one-half. Positive
frequency dependence tends to increase the more
common variants and decrease the others.
(c) Pay-off-biased social learning
When individuals have information about the pay-offs
of others, it is possible to use these cues of success to
adaptively bias social learning. Such pay-off-, success-,
or prestige-biased social learning can be very individually
adaptive, provided cues are reliable, leading to
evolutionary dynamics that can be very similar to
natural selection (Boyd & Richerson 1985; Schlag
1998, 1999; Henrich & Gil-White 2001). A key
property of these strategies may be their tendency to
lead to the copying of neutral or mildly maladaptive
behaviour that was initially associated with successful
individuals (Boyd & Richerson 1985), but recombina-
tion is also a possibility (Boyd & Richerson 2002).
Figure 1c,d plots the instantaneous and evolutionary
dynamics of simple pay-off-biased learning. In figure 1c,
frequency of trait after social learning as a function of
the frequency before social learning is shown. If p is
the value on the horizontal axis, then pC pð1K pÞb
is the value on the vertical axis (derived in McElreath &
Boyd 2007, ch. 1). The parameter b determines the
strength of pay-off bias and is analogous to a selection
coefficient, in genetic evolutionary theory. We plot
bZ1/2 here. The dashed line is again the expectation
under unbiased social learning. In figure 1d, the
evolutionary dynamics produce a classic logistic growth
curve (solid curve). Pay-off-biased social learning tends
to increase the frequency of adaptive behaviour, but at
the cost of greater information demands.
(d) Integrated social learning
Many mixes of the above kinds of social learning are
possible (Laland 2004; Whiten et al. 2004). Aside from
the likely possibility that individual asymmetries—age,
sex, skill, position in social network—will make some
strategies more common among some individuals,
strategies can be hierarchically ranked within each
individual. Mixes of strategies produce their own

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3518 R. McElreath et al. Analysis on social learning strategies

evolutionary trajectories, as well (Henrich 2001). The Table 1. Probabilities of acquiring optimal behaviour via
dashed curve in figure 1d is the dynamics of a mix social learning, for the three nonlinear strategies positive
of pay-off bias and positive frequency dependence. frequency dependence (C), pay-off bias (S) and pay-off
For different mixes of these and other strategies, conformity (SC). The sample column gives all possible
different evolutionary dynamics are expected. samples of three adults. Uppercase letters indicate that the
individual sampled received a large (B) pay-off from their
behaviour whereas lowercase indicates the opposite. A or a
(e) Modelling integrated pay-off-biased and indicates optimal behaviour and B or b indicates non-optimal
frequency-dependent social learning behaviour. By multiplying each probability of a specific
While there has been modelling effort devoted to sample occurring by chance in a particular strategy column,
studying linear, unbiased social learning, frequency- one can sum these products to compute an expected
probability of acquiring optimal behaviour via a given
dependent social learning and pay-off-biased social
strategy. In the Pr(sample) column, q is the frequency of
learning, to our knowledge no theoretical study has optimal behaviour in the entire adult population and a and b
simultaneously examined these options in the same are the probabilities of optimal and non-optimal behaviours,
context. Therefore, we finish this section by presenting respectively, returning large pay-offs.
an extension of existing evolutionary theory that
includes frequency-dependent bias, pay-off bias and Pr Pr Pr
a hierarchical integration of the two. We construct sample Pr(sample) (1rC) (1rS) (1rSC)
recursions for the dynamics of genes controlling these
different learning strategies, as well as for the frequency AAA q3$a3 1 1 1
of adaptive learned behaviour. We then analyse this AAa q3$3a2(1Ka) 1 1 1
gene-culture system in order to understand what Aaa q3$3a(1Ka)2 1 1 1
environments favour different strategies. aaa q3$(1Ka)3 1 1 1
AAB 3q2(1Kq)$a2b 1 2/3 1
Consider a large population living in a uniform but
AAb 3q2(1Kq)$a2(1Kb) 1 1 1
temporally varying environment. Each individual faces AaB 3q2(1Kq)$2a(1Ka)b 1 0 0
a choice of two discrete behaviours. One of these Aab 3q2(1Kq)$2a(1Ka) 1 1 1
choices yields a fitness benefit B, a proportion a of the (1Kb)
time, yielding an average of aB. The other yields an aaB 3q2(1Kq)$(1Ka)2b 1 0 0
average bB!aB. For each generation, there is a chance aab 3q2(1Kq)$(1Ka)2 1 2/3 1
u that the better behaviour switches to the other option. (1Kb)
These changes cannot be observed by individuals. ABB 3q2(1Kq)2$ab2 0 1/3 0
Behaviour is acquired via learning, either individu- ABb 3qð1KqÞ2 $a2bð1KbÞ 0 1 1
ally or socially. Individual learning (I) pays an average Abb 3qð1KqÞ2 $að1KbÞ2 0 1 1
learning cost in order to determine which option is abb 3q(1Kq)2$(1Ka)(1Kb)2 0 1/3 0
better. This makes the fitness of an individual learner: aBb 3qð1KqÞ2 $ð1KaÞ2bð1KbÞ 0 0 0
aBB 3qð1KqÞ2 $ð1KaÞb2 0 0 0
W ðIÞ Z w0 C aBKc; BBB ð1KqÞ3 $b3 0 0 0
BBb ð1KqÞ3 $3b2 ð1KbÞ 0 0 0
where w0 is baseline fitness from other behaviour and c Bbb ð1KqÞ3 $3bð1KbÞ2 0 0 0
is the average cost of learning. bbb ð1KqÞ3 $ð1KbÞ3 0 0 0
Social learning can be unbiased (linear, L),
frequency dependent (conformist, C), pay-off biased
(S) or pay-off conformity (SC). Linear social learning conformist learner acquiring optimal behaviour is
copies a random adult from the previous generation,
resulting in average fitness: qC Z q C qð1KqÞð2q K1Þ:

W ðLÞ Z w0 C aBq C bBð1KqÞ Using this expression gives us a mean fitness for C,

Z w0 C Bðaq C bð1KqÞÞ: W ðCÞ Z w0 C BðaqC C bð1K qC ÞÞ:

Pay-off-biased social learning (S) samples three
The frequency of currently optimal behaviour, q, has
individuals and adopts the behaviour with the highest
its own dynamics, which we define below. The
average observed pay-off. We compute the expected
important point here is that linear social learning
probability of acquiring optimal behaviour through this
does not transform this proportion in any direct way.
heuristic in the same fashion as for conformity: each of
On average, it replicates the frequency of optimal
the three models sampled has a chance q of having
behaviour across generations.
optimal behaviour and each model then has a chance
Positive frequency dependence, conformity (C),
either a or b of displaying a pay-off of B. Thus, the
does however transform q. We assume perhaps the
probability of any combination of underlying behaviour
simplest conformity heuristic. The learner samples
and displayed pay-offs can be computed from the
three random adults from the previous generation and
binomial distribution (table 1). This results in a chance
then adopts the most common behaviour among these
of acquiring optimal behaviour:
three models. Since the chance that any one model has
optimal behaviour is q, the binomial distribution qS Z q C qð1KqÞðað2 C bð2K3bð1KqÞK4qÞÞ
(table 1) allows us to compute the probability of any
combination and therefore the probability of the C a2 ð3b K1Þq C bðbð1KqÞK2ÞÞ:

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Analysis on social learning strategies R. McElreath et al. 3519

(i) (ii) (iii) (iv) (v)

(a) 0
I L

C S SC
0.6
0 b 0.5
(i) (ii) (iii) (iv) (v)
(b) 0
I
a–b

L C S SC
0.5
0 b 0.5
(i) (ii) (iii) (iv) (v)
(c) 0
I L
u
C S SC
0.6
2 10
B
Figure 2. Sensitivity analysis for the evolutionary model of social learning strategies in the main text. Each row plots the frequencies
of five different strategies (individual learning, unbiased social learning, positive frequency dependence, pay-off bias and pay-off
conformity) for two-dimensional combinations of parameters. Each individual plot is the frequency of a single strategy after 5000
generations of simulations at all combinations of the two parameters labelled on each axis. (a(i)–(v)) u varied from 0 to 0.5, b varied
from 0 to 0.5, while aZ0.5Cb. (b(i)–(v)) aKb varied from 0 to 0.5, b again from 0 to 0.5. (c(i)–(v)) u again varied from 0 to 0.5,
B from 2 to 10. All other parameters not on axes were fixed at B/cZ6, uZ0.1, aZ3/4, bZ1/4, w0Z2. The most powerful inference
from these simulations is that either pay-off bias (S) or pay-off conformity (SC) dominates the population, unless the environment
is very unstable and individual learning is too costly, relative to fitness benefits of optimal behaviour.

The fitness of S is therefore, The complete evolutionary system is very difficult to

W ðSÞ Z w0 C BðaqS C bð1K qS ÞÞ:
Finally, we consider the integrated strategy pay-off
conformity (SC). This strategy attempts pay-off-biased
social learning just as S, but falls back on positive
frequency dependence whenever observed pay-offs are
tied. Just as before, it is possible to compute the
expected chance of acquiring optimal behaviour
through this heuristic, by using the binomial distri-
bution (table 1). This gives us
qSC Z q C qð1KqÞð3að1Kb2 Þð1KqÞ
C 3a2 bqKqð3bK2Þ K 1Þ:
Again, this implies mean fitness:
W ðSCÞ Z w0 C BðaqSC C bð1K qSC ÞÞ:
The dynamics of q are governed by the proportions
of each strategy in the population. For proportions
fI ; fL ; fC ; fS ; fSC , the frequency of optimal behaviour in
the next generation in the absence of environmental
change is given by
q 0 Z fI C fL q C fC qC C fS qS C fSC qSC :
Now accounting for environmental change, we arrive at
the recursion for the frequency of optimal behaviour in
the next generation:
q 00 Z ð1K ut Þq 0 C ut ð1Kq 0 Þ;
where ut 2 ½0; 1
is a random variable indicating
whether the environment changed in generation t.
This random variable has chance u of being 1, as u is
the long-run rate of environmental change.
analyse, because the recursion for q is highly nonlinear.
This means there is no guarantee that q even reaches a
stationary distribution, and so the fast–slow dynamics
approach often employed in these situations (see
McElreath & Boyd 2007, ch. 6) is risky. Even if we
adopt the fast–slow approach, the implied equilibrium
of q is itself the solution to a cubic in q and very difficult
to analyse.
Therefore, we adopt a simple simulation approach
to analysing this system. We conduct simulations for
a large number of parameter combinations in order to
map out the conditions that favour different strategies.
The fitness expressions and the recursion for q allow
us to define a set of difference equations that define
the evolutionary dynamics of the system. For any
initial frequencies of the strategies and values for
w0 ; B; c; a; b; u, simulating this system amounts to
generating a random variable u t and recursively
computing the frequencies of each strategy after
selection. After 5000 simulated generations at each
parameter combination, we record the frequency of
each strategy. While frequencies could in principle be
highly stochastic, fluctuating as selection fluctuates, the
results show that taking the final frequency delivers
the correct inferences. It also turns out that initial
frequencies have no effect on the long-run evolution of
the system, allowing us to present simulation results for
uniform initial conditions in which all strategies had
initially equal frequency.
Figure 2 plots the frequencies of each strategy
at simulation end, for two-dimensional sensitivity
analyses. Black indicates a frequency of 1, white
indicates a frequency of 0 and grey indicates

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3520 R. McElreath et al. Analysis on social learning strategies
intermediate frequencies, on a linear gradient. Baseline
parameter values in these simulations were B/cZ6,
uZ0.1, aZ3/4, bZ1/4, w0Z2. In figure 2a, the
horizontal axis takes b, the rate of good pay-offs from
the non-optimal choice, from 0 to 0.5, holding the
value of aZ0.5Cb. Thus, the degree to which
the optimal choice is better remains constant, but the
absolute level of profitability of both options increase,
as one moves left to right on the horizontal axis. The
vertical axis takes u, the rate of environmental change,
from 0 to 0.6, moving top to bottom.
When u is large, the environment changes rapidly,
and individual learning excludes the other strategies
(figure 2a(i)). When the environment is sufficiently
stable, however, either pay-off-biased social learning
(S) or pay-off-conformity social learning (SC) excludes
the other strategies. When b is small, S excludes SC.
The second row varies the difference between the
optimal and non-optimal option, aKb, from 0 to 0.5,
on the vertical axis. The difference in profitability
between the two options interacts only very weakly
with the absolute level of profitability, shown again on
the horizontal axis. At the extreme limit of aKbZ0,
learning does not pay at all, and so all strategies remain
at their initial frequencies (the grey line at the top of
the plots in figure 2b(ii)–(v)), except for individual
learning (I), which is eliminated for trying to learn
and paying a direct cost to do so.
The third row of simulations interacts environmental
uncertainty, u, with the magnitude of pay-offs, B. The
vertical axis is identical to that of the first row, but the
horizontal varies B from 2 to 10 (centred on the value
BZ6 that generated the other rows). We can see now
that, when B is sufficiently small, individual learning is
always excluded, even when the environment is highly
unstable. Pay-off-biased social learning, however,
excludes the other strategies for these parameter
combinations. Pay-off conformity only dominates, as
the environment becomes more stable. This stands to
reason, as conformity—combined with pay-off bias or
not—suffers more from changes in the environment than
does pure pay-off bias. To understand this, consider
what happens to a conformist just after a change in
the environment. Chances are, majority behaviour is
suboptimal, and therefore conformity tends to reduce
the frequency of optimal behaviour even more. Pay-off
bias, however, can still use pay-offs as a cue to optimality.
(f ) Analysis summary
The most obvious result of this analysis is to emphasize
the adaptive significance of pay-off-biased social
learning, whether combined with frequency depen-
dence or not. Provided pay-offs can be observed
with sufficient accuracy, adopting behavioural options
with higher observed average pay-offs excludes other
strategies under a wide range of conditions. Unless the
environment is extremely stochastic (in which case
individual learning dominates) or almost perfectly stable
(in which case pure conformity dominates), some kind
of pay-off-biased learning is an evolutionarily stable
strategy, in our simulations.
The integrated social learning strategy, pay-off
conformity, excludes pure pay-off bias when
the environment is not too unstable. Being partly
Phil. Trans. R. Soc. B (2008)
frequency dependent, it needs the optimal behaviour to
be the more common behaviour, at least long enough
to realize fitness gains. Otherwise, ignoring frequency
information is more adaptive. The other factor
affecting whether pay-off conformity dominates pure
pay-off bias appears to be the magnitudes of a and b,
the chances optimal and non-options behaviour yield
large pay-offs. In the simulations, when aO1/2, the
integrated pay-off-conformity strategy outperforms
pay-off bias alone, holding the difference aKb constant.
We are unsure what is causing this advantage. The
expression qSCOqS can be reduced, but it yields a
complicated expression that is difficult to interpret. It
is also not the whole story, because the average value
of q is not described by this condition, and a and b
will have large effects on this value.
An interesting feature of pay-off-biased strategies is
that they can eliminate individual learning, because any
variation among individuals in choice can be used to
discriminate good and bad options by pay-offs. All
of the nonlinear social learning strategies—positive
frequency-dependence, pay-off bias and pay-off con-
formity—can in fact do this, because their nonlinear
effects can, under the right conditions, accomplish the
same thing as individual learning.
In §3, we present an experimental design that allows
for a large number of different and integrated social
learning strategies. In light of these simulations, we
expect a heavy reliance on pay-off bias. Also, because
the environment is quite stable in the experiment
(changing every 15 periods, or a rate of 0.07), the
integrated pay-off-conformity strategy should exclude
pure pay-off bias. We do not think these exact
predictions will describe the results—even simple
experiments are much more complex than the
theory that motivates them. However, if the theory
we have presented here gets at the right economic
considerations, then the qualitative results should
show a much stronger reliance on pay-off bias than
frequency bias.
3. EXPERIMENTAL DESIGN
In order to study the diversity of social learning strategies, we
require a decision context complex enough to make both
frequency dependence and pay-off bias simultaneously
possible. Our social learning experiments create social
contexts in which groups of individuals can evolve beha-
vioural traditions, through a combination of their own
experience and the available social information. These
‘microsociety’ (Schotter & Sopher 2003; Baum et al. 2004)
experiments are highly controlled, relative to field studies of
social learning, and as a result, we know which social and
individual information each participant examines at each
time step. Unlike most experiments, however, our experi-
mental groups generate all social information endogenously,
without any experimenter deception. This both allows us to
examine the emergent properties of social learning and
develop statistical methods that can address less controlled
natural sources of data.
The experiment allows participants to access both the
frequencies of different choices and associated pay-offs,
within their own social groups. Over a series of rounds, they
may or may not use this information to learn, and we use the
complete time series of decisions and records of which
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Analysis on social learning strategies
participants access which information in order to test the
different models of social learning, pay-off biased or
frequency biased.
We have used a similar social decision environment in
previous work (McElreath et al. 2005), and the environment
itself is a social-learning extension of familiar multi-arm
bandits used in diverse fields to study individual learning. By
using a well-studied decision environment, we can begin with
good candidate individual learning models and study the
effects of adding different kinds of social information. Our
previous experimental studies have omitted pay-off infor-
mation, and so we could not consider pay-off-biased
strategies. And while we have used the statistical approach
in our previous papers, we have not previously emphasized
the methodological value of the statistics themselves, for
analysing data collected in ‘wild’ contexts.
(a) Participants
One hundred and sixty-three participants, students at the
University of California at Davis, interacted with one another
via a computer network. We recruited participants through an
advertisement in the campus newspaper. Participants
received between $5 and $20 for their participation, based
upon their performance. We used no deception in this
experiment. Participants read a complete set of instructions
and successfully completed a set of test questions about their
knowledge of the experiment, before beginning.
(b) Group structure
Participants were sorted into random, anonymous groups
of four to seven individuals, in sessions of between 8 and
20 participants. Each session was a single experiment on a
single date. While participants in the same session made
choices in the same room, these participants did not know
which of the other participants they were sorted into a group
with. Groups were constrained to be always greater than
three individuals, in order for frequency bias to be effective,
as three neighbours are the required minimum for positive
frequency dependence. Depending upon the total number of
participants showing up for a given session, group sizes were
arranged to create as many groups of four as possible. All
remaining participants in that session were placed in a single
larger group.
(c) Decision
Over a series of 60 periods, ‘seasons’, each participant made a
series of 60 crop choice decisions. These 60 periods were
divided into four ‘farms’ of 15 periods each. These farms
served to signal to participants that conditions might have
changed. On any given farm, one of two crops, ‘wheat’ or
‘potatoes’, had a higher average yield than the other. Across
farms, which crop was optimal was determined at random.
Thus in each period, each participant chose a single crop
to plant and receive a yield from. Yields were summed across
all periods, and participants received cash payment so that
they earned between $5 and $20, depending upon per-
formance. The vast majority of participants earned between
$15 and $18.
The number of farms and periods in each finesses the
trade-offs of (i) having only limited time to keep participants
before they grow bored and unmotivated and (ii) desiring the
most varied data on learning. Thus the total number of
periods, 4!15, is set by the time constraint. The number of
periods per farm is set to maximize information about
learning dynamics. If we had a single farm of 60 periods,
most of the later periods would add little to nothing to the
Phil. Trans. R. Soc. B (2008)
R. McElreath et al. 3521
analysis, because all participants would be sure of the best
option by then, as we have learned from previous experiments
(McElreath et al. 2005). If a farm is too short, however, we
never witness the full dynamics of any learning process.
Therefore, guided by pilot experiments and our simulation
studies, we decided on 15 periods per farm, as this is the
approximate value that maximized our ability to correctly
distinguish simulated strategies.
(d) Social information
On the first period of each farm, no social information was
available. However, on each period after the first, participants
could access social information from the most recent period.
Participants could examine their own most recent crop
choices and resulting yields. Each participant could also
examine the most recent crop choices and yields of each
member of their own group. This information was displayed
on screen in boxes labelled by the type of information. When
a participant moused over a box, the information in it was
displayed. The experiment software tracked millisecond
access to this information, resulting in a time series of
information access. This kind of ‘mouse-tracking’ experiment
has been used to great effect in judgement and decision-
making research (Payne et al. 1993). The order of the rows,
yield and crop was randomized for each participant, each
period, and the order of neighbours was also randomized.
The order of the crop choices at the bottom was also
randomized within each participant and period.
(e) Pay-offs
Both crops generated pay-offs from normal distributions with
the same variance, while the better crop had a mean pay-off of 13
units and the worse 10 units (set from previous experience and
simulation study). Participants knew that one crop had a
constant higher mean than the other, but had no prior
information that would allow them to determine which of the
two was better.
The variance of yields was constant within farms but could
be either 1/2 or 4, determined randomly but in a way to
ensure two farms with a variance of 1/2 and two farms with a
variance of 4. The different variances comprise a learning
difficulty treatment that we have used in previous experiments
(McElreath et al. 2005).
(f ) Simulating the experiment
While there is not space here to describe our simulation
in detail, we used the statistical models we will present
later to produce simulated experimental play, under a
variety of group sizes and other experiment parameters.
These simulations simply use the probability models to
produce stochastic learning and choice. We then run
the data produced through the exact statistical analysis
we use on the real data. These simulations allowed us to
(i) choose good experimental design parameters and
(ii) verify that our statistical analysis works (i.e. recovers
true simulated strategies).
4. RESULTS
Like our previous experiments (McElreath et al. 2005;
Efferson et al. 2007), participants learn the optimal
crop for each farm, over time. Figure 3 shows
the proportion of participants making optimal choices,
as a function of period within each farm. The rate
of improvement is much faster than in previous
experiments, which omitted pay-off information for
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3522 R. McElreath et al. Analysis on social learning strategies
(a) 1.0
0.9
proportion choosing
optimal crop 0.8
0.7
0.6
0.5
2 4 6 8 10
round in farm
Figure 3. (a) Proportion of optimal crop decisions, by round within farm. Vertical lines show 95% profile-likelihood confidence
intervals. (b) Proportion of neighbours’ crop decisions (circles) and yields (curve) inspected, by round within farm.
neighbours (McElreath et al. 2005). The increase
between periods 2 and 15 is much smaller than the
increase between periods 1 and 2.
Perhaps as a result of the marginal gains in
optimality declining after the second period, rates of
inspecting the choices (which crop was planted) and
yields (how much profit was made last period) of
neighbours decline from the second period onward
(figure 3). The average rate never falls below a
majority of neighbours, however. Note that rates of
inspecting yields slightly exceed those for inspecting
crop choices. This implies that some participants were
using something similar to an elimination by aspects
strategy, in which one important cue is used to first
narrow down the number of cases one will consider
(see Payne et al. 1993). In this case, some participants
may have first eliminated neighbours to examine crop
choices from, by first scanning the yields from the
previous period. This would result in the kind of
pattern seen in figure 3b. Our statistical analyses in §5
use only the yields and crops actually inspected by
each participant, and so take the search strategy as a
given. We think the design of the search strategy is a
worthwhile question, however. But we doubt such
details—truly observing information search—will
often be possible in natural settings.
5. ANALYSIS
We adopt a statistical approach that allows us to (i)
directly use mathematical models of social learning
strategies as statistical models and (ii) evaluate several
plausible, non-null statistical models simultaneously.
The question is not whether social information is
used—few would expect a complete absence of social
learning in such a context—but rather how social
information is used.
(a) Strategies
We translate each hypothetical learning strategy into an
expression that yields the conditional probability of an
individual choosing any behavioural option i in any
period t, given private information and the social
information the individual accessed. Each strategy
consists of two parts. The first part is the definition of
a recursion for updating the attraction scores of all
behavioural options. The second part is a convex
Phil. Trans. R. Soc. B (2008)
(b) 0.65
proportion neighbours
0.60
examined
0.55
0.50
12 14 2 4 6 8 10 12 14
round in farm
combination of individual choice and the influence of
social information.
A large number of meaningfully different strategies
can be constructed by varying these two components
(Camerer & Ho 1999; Stahl 2000; Camerer 2003). As
our purpose in this paper is to illustrate the approach
in the simplest manner, we do not explore a large
strategy space, but instead restrict ourselves to those
nominated by the basic research question and existing
evolutionary literature: how do people use frequency-
dependent and pay-off-biased social learning, when
both are possible?
We examine five different models that combine
elements of frequency dependence and/or pay-off bias.
First, we define (i) individual learning, (ii) frequency-
dependent social learning, (iii) pay-off-biased social
learning. We then define hierarchical strategies that
combine pay-off-biased learning with the frequency
dependence or individual learning: (iv) hierarchical
compare means and individual learning and (v)
hierarchical compare means and frequency depen-
dence. We do not present analyses of strategies that
reverse the hierarchical order of information use,
frequency dependence and compare means, for
example. These strategies fit very poorly to our data,
as will become clear when we examine the fits of each
basic model, and so we omit them for simplicity
of presentation.
(i) Individual learning
We use a standard, successful reinforcement learning
model as the basis of individual updating (Camerer
2003, ch. 6). The attraction score of option i in period
t C1 is given by
Ai;tC1 Z ð1KfÞAi;t C fpi;t ;
where f is a parameter determining the weight given to
new experience and pi,t is the pay-off observed for
option i in period t. When option i was not sampled in
period t, pi,tZ0. Since there is no reason to expect
participants to have strong priors favouring either
behavioural option, we set A1,0ZA2,0Z0.
The attraction scores are transformed into probabil-
istic choice with a ‘softmax’ choice rule, again typical
of the learning in games literature. The probability of
choosing option i in period tC1 is given by
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Analysis on social learning strategies R. McElreath et al. 3523

expðl Ai;t Þ 1.0

PrðijAt ; QÞtC1 Z ;
expðl A1;t Þ C expðl A2;t Þ

proportion of pay-off-biased
where Q indicates a vector of all parameters and l is a 0.8
parameter that measures the influence of differences

social learning
between attraction scores on choice. When lZ0, 0.6
choice is random with respect to attraction scores. As
l/N, choice becomes deterministic, in favour of the 0.4
option with the higher attraction score.

(ii) Frequency-dependent social learning 0.2

To model the family of strategies that use the frequency
of behaviour among group members, we modify the 0
learning model above to cue choice by the frequency of
–4 –2 0 2 4
options seen. Attractions are updated as before, but
difference in observed mean pay-offs
choice is given by the rule
expðl Ai;t Þ Figure 4. The function that determines the reliance on pay-
PrðijAt ; QÞtC1 Z ð1KgÞ off-biased learning, as a function of the observed difference in
expðl A1;t Þ C expðl A2;t Þ means, p 1;t Kp 2;t . See the description of the hierarchical
nf means/conformity strategy in the text. Solid curve, dZ1/10;
C g f i;t f ; dashed curve, dZ2.
n 1;t C n 2;t
where ni;t is the count of neighbours observed to have symmetrical logistic function to model the change in
chosen option i in period t; g measures the weight of reliance on pay-offs, as the distance between the
social information on choice; and f determines how observed means increases. Let Y ðd; p 1;t ; p 2;t Þ be the
nonlinear frequency dependence is. When fZ1, proportion of choice that is driven by individual
imitation is unbiased. When f O1, however, more updating, where d is a new parameter that determines
common options have exaggerated chances of being how quickly reliance on pay-offs decreases, as the
copied, resulting in positive frequency dependence, difference in observed means increases,
such as majority rule conformity. When f!1, frequency 2
dependence is negative, and more commonly observed Y ðd; p 1;t ; p 2;t Þ Z hY:
1 C expðdðp 1;t Kp 2;t Þ2 Þ
options are less likely to be copied.
Since changes in choice feedback to changes in Figure 4 plots this function for two values of d. The
attraction scores, even though this strategy has the same probability of choosing i under the hierarchical
attraction updating recursion as individual learning, compare means/individual strategy is
reinforcement patterns may be quite different.
expðl Ai;t Þ
PrðijAt ; QÞtC1 Z ð1KgÞ
(iii) Compare means expðl A1;t Þ C expðl A2;t Þ
This pay-off-biased strategy attends to neighbours’ p 100
i;t
yields and chooses the option with the highest observed C g ð1KY Þ
p 100
1;t C p 100
2;t
mean. It uses the choice rule !
expðl Ai;t Þ expðl Ai;t Þ
PrðijAt ; QÞtC1 Z ð1KgÞ CY :
expðl A1;t Þ C expðl A2;t Þ expðl A1;t Þ C expðl A2;t Þ

p 100
i;t For similar observed means, the individual learning
Cg ; component will dominate the social learning term.
p 100
1;t C p 100
2;t
P Otherwise, the individual will mainly attend to
where p i;t Z j pi; j;t =ni;t is the mean pay-off observed for differences in observed means. However, if d is a very
option i in period t over all group members j, including large number, then only a very narrow range of very
oneself. Raising these average pay-offs to a large power similar observed means will lead to falling back on
creates an approximate step function, so that one or the individual updating.
other option is favoured by the social component of
choice. When one or both options are unobserved in
period t, this strategy behaves as individual learning. (v) Hierarchical compare means/frequency-dependent
We fix f Z100 in order to force the model to match our social learning
theory, i.e. a threshold behaviour. This model is like the previous, but falls back on
frequency-dependent social learning, as the difference
(iv) Hierarchical compare means/individual learning in observed means increases.
This strategy uses the comparison of choice means and expðl Ai;t Þ
individual updating, but in a manner different from the PrðijAt ;QÞtC1 Zð1KgÞ
expðl A1;t ÞCexpðl A2;t Þ
pure compare means model. Using the distance
between estimated means as a cue of uncertainty, the !
p 100
i;t ni;tf
strategy falls back on individual learning (attraction Cg ð1KY Þ 100 CY f :
updating) when the means are similar. We use a p 1;t C p 100
2;t n1;t Cn2;tf

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3524 R. McElreath et al. Analysis on social learning strategies
(b) Fitting strategies to data
The 19 experiment sessions involving 163 participants
provided 7900 decisions, under full information
conditions that might allow us to distinguish between
frequency-dependent and pay-off-biased social
learning. We fit the above models to these decisions,
producing for each model a negative log likelihood of
observing the true data, given the assumption that the
model is true: Klog LðDjx; QÞ for a model x with set of
parameters Q, where D is the data, a vector of ‘crop’
choices. The likelihood is defined as
Y
LðDjx; QÞ Z PrðDj jAtK1 ; QÞt ;
t
Q where k is the number of free parameters in model x.
where t indicates the product over all rows t. The
usual practice in likelihood estimation, and the practice
we follow here, is to take natural logs of each
conditional probability and then sum these to find
log LðDjx; PÞ:
X nKk K 1
Klog LðDjx; QÞ ZK log PrðDj jAtK1 ; QÞt :
t
Taking logarthirms first results in greater precision,
owing to the way most computers handle floating point
values. The parameters Q are fit via maximum
likelihood, and therefore the fitting exercise also yields
information on the best estimates of flexible com-
ponents of the learning rules.
We conducted this fitting exercise, as well as the
validating simulations, in R and using the helpful
package bbmle (Bolker and based on stats4 by the
R Development Core Team 2008; R Development
Core Team 2008). All analysis code is available from
the corresponding author. We confirmed via simulation
that our analysis could recover true parameter values
and strategies, when the true strategy was among the
set of strategies considered. The validation exercise is
helpful, because not all distinct models can be
distinguished by all kinds of data (this problem has
plagued the individual learning literature, see Camerer
2003, ch. 6).
(c) Comparing models
We compare the fit of the social learning models using
Akaike information criteria (Akaike 1974; Burnham &
Anderson 2002). Unlike null hypothesis testing,
comparing models with Akaike information criteria
(AIC—called by Akaike himself simply ‘An infor-
mation criterion,’ but subsequently renamed by the
scientific community), or another information
criterion, allows a researcher to assess the relative
explanatory power of any number of different compet-
ing and plausible models, without favouring any ‘null’
model. AIC is an estimate of the information lost by
using any particular model to estimate reality.
The advantages of the information theoretic
approach over customary null hypothesis testing has
been discussed for several decades (see citations in
Cohen 1994; Anderson et al. 2000), so we will not
repeat them here. Readers should note, however, that
there will be no p values in our presentation. Like many
statisticians, we do not find much inferential value in
p-values, especially when multiple plausible models
Phil. Trans. R. Soc. B (2008)
are under consideration. AIC and related approaches
are becoming increasingly popular in the evolutionary
sciences, because they permit more nuanced questions
and are not plagued by the same sample size biases
of null hypothesis testing ( Johnson & Omland 2003).
They also allow for more powerful analysis of
observational data, collected without precise experi-
mental control.
In order to compare the models, each negative log
likelihood from the fitting exercise is transformed into
an AIC:
AICx ZK2 log LðDjx; PÞ C 2k;
We use the common sample-size-adjusted version of
the above, AICc (Burnham & Anderson 2002), and this
is what we display in our results:
2kðk C 1Þ
AICc;x ZK2 log LðDjx; PÞ C 2k C ;
where n is the number of observations to be
predicted by the model. The penalty for number of
parameters is not arbitrary—it adjusts precisely for
the expected overfitting that arises whenever free
parameters are added.
AIC can be used to select a single ‘best’ model, if an
analyst desires. However, since the ‘true’ model, in all
its detail, is certainly not contained in the set of models
fit to data, it is perhaps a more productive approach to
treat it as a continuous measure of the degree to which
each model estimates ‘truth’ (Forster & Sober 1994).
AIC estimates the out-of-sample predictive accuracy of
each model, and one easy way of ranking these
estimates relative to the models in the analysis is by
using Akaike weights (Burnham & Anderson 2002).
The weight of any model x is given by


exp K 12 Dx
wx Z P
1 ;
j exp K 2 Dj
where Dx Z AICx KAICmin , the difference between the
AIC of model x and the smallest AIC in the set of
compared models. For the best-fitting model with the
smallest AIC, DZ0. These weights are numbers
between 0 and 1 that estimate the relative likelihoods
of each model being the best model in the set.
A useful way we have found to explain this approach
is to consider a horse race. There are many horses in
each race, and while the fastest horse will not always
win, it usually will. If the best horse loses, it should not
usually lose by much. Thus both the rank of finishes—
which horse was first, second, etc.—and the time
differences in finishes are informative. In the same way,
the true model may not always fit the data best (just as
‘significant’ p values do not always identify important
effects). But it will usually have a high Akaike weight,
even if not the highest. So, just as a photo finish tells
you that it is difficult to say, without another race,
which of two horses is faster, when two models have
very similar Akaike weights, there is uncertainty as to
which would make the best out-of-sample predictions.
When one model has an Akaike weight much larger
than the others, however, we can be confident that it is
the best of the models considered.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Analysis on social learning strategies R. McElreath et al. 3525

Table 2. Comparison of social learning models fit to experimental data. AICc indicates the adjusted fit of each model to the entire
sequence of choices, for each subject, after accounting for model complexity (number of parameters). Models are ordered from
best fit to worst. The Akaike weights estimate the proportion of evidence in favour of each model. (Meanings of parameters:
f 2 ½0; 1
, strength of attraction updating; lO0, influence of attraction differences of choice; g 2 ½0; 1
, weight given to social
information; dO0, decline in probability of pay-off-biased imitation, as difference in observed means increases. HCMFD,
Hierarchical compare means/frequency dependence; HCMINDIV, Hierarchical compare means/individual learning.)

parameter estimates

Akaike
model AICc weight f l g d f

HCMFD 6519.59 z1 0.6605 0.1645 0.3365 3.210 1.953

HCMINDIV 6918.89 !0.001 0.4620 0.1917 0.1400 4.982 n.a.
compare means 6924.23 !0.001 0.4611 0.1921 0.1239 n.a. n.a.
frequency dependent 6929.34 !0.001 0.4998 0.1814 0.1349 n.a. 3.396
individual 7004.25 !0.001 0.4382 0.1866 n.a. n.a. n.a.

Table 2 presents the AICc, Akaike weight and 1.0

parameter estimates for each model, sorted from best
to worst fitting model. The bulk of evidence favours

probability of copying choice

model 5, hierarchical compare means/frequency
dependence. While there is no doubt about hetero-
geneity among participants, the strength of this result
leaves little doubt that any of the simpler strategies
accounts for any sizeable fraction of participants. The
maximum-likelihood estimate for f, the degree of
positive or negative frequency dependence, is just
under 2, indicating mild positive frequency depen-
dence or conformity (figure 5). The maximum-
likelihood estimate of d (not shown in figure) produces
a steep fall-off in reliance on pay-off bias for a distance
above approximately 1 unit. We caution that there is
uncertainty in these estimates, but emphasize that a
model with d fixed to a large value, say 100, does not
produce a better fit, even accounting for the reduction
of one parameter.
Many readers may wonder what proportion of
variance in choice is explained by the best-fitting
model. As is usual with binomial models, there is no
true equivalent of R 2, the proportion of variation
explained by the fit model. However, it is possible to
construct an analogue that compares the raw like-
lihoods of each model to a random choice model. A
random choice model just chooses randomly at
each time t. Over 7900 choices, this model will always
have a negative log likelihood of K7900 !logð1=2ÞZ
5475:863. This is a reasonable benchmark for the worst
any model can do, predicting the data. The negative log
likelihood of the best-fitting model is 3259.792. There-
fore, an analogous calculation of the variance explained
by any model x is 1KlogðLðDjxÞÞ=logðLðDjrandomÞÞ. In
our case, 1K3259.792/5475.863Z0.4047. For the
second-best model, 1K3459.442/5475.863Z0.3682.
These measures do not account for model complexity,
but they do provide a rough guide to additional raw
variance explained by the best model. We caution,
however, that substantial components of choice may
be truly random, and therefore any behavioural model
will fail to achieve a negative log likelihood of zero.
In cases in which measurement error is possible, as in
field studies or data coded from video, measurement
error will also make it impossible for even the true
model to achieve a negative log likelihood near zero.
0.8
0.6
0.4
0.2
0
0 0.5 1.0 1.5 2.0 2.5 3.0
frequency of choice
Figure 5. Maximum-likelihood estimate of strength of
positive frequency dependence for the best-fitting model,
hierarchical compare means/frequency dependence. Solid
curve indicates estimated probability of copying a choice,
given its frequency in the group. Dashed line indicates same
probability under fZ1, unbiased social learning.
6. DISCUSSION
We have analysed an interdependent time series of
profit-oriented choice behaviour in humans. Our
experiment did not precisely control the social
information available to each participant. Instead, we
allowed all social information to arise endogenously,
through the actual behaviour and information seeking
of participants. While one major tradition in laboratory
experiments frowns upon such a design, we consider it
an asset, for two reasons.
First, if the study of social learning is ever to link
the psychological to the population level, statistical
techniques that can accommodate observational and
noisy data are needed. The model comparison
approach we adopt in this paper is general to any set
of strategies a researcher might imagine. Caution is
needed to ensure that the kind of data available can
discriminate among the possible models. But provided
the different models are identifiable in this way, the
likelihood-based information criteria can quantify the
relative explanatory power of different hypotheses.
These dynamic models can then be reasonably asked
to produce out-of-system predictions that provide

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3526 R. McElreath et al. Analysis on social learning strategies
another avenue of disconfirmation. By contrast, effect
sizes from ANOVA cannot reasonably be expected to
predict out-of-experiment effects, because no genuine
model of learning is present.
Second, the emergent population-level conse-
quences of social learning can only be studied where
the experimenter allows them to occur—in settings in
which social information itself is not controlled
experimentally. This advantage is twofold. Being
able to study population-level effects, such as the
emergence of traditions or rates of diffusion, is
important. But in a cultural species, such as humans
and possibly other species, social learning strategies
themselves are probably adapted to a cultural
environment (Henrich & McElreath 2007). Thus, it
will eventually be difficult to study the functional
design of strategic social learning without appreciating
the cultural environment it is adapted to. This will
be true even (especially) if learning strategies them-
selves are culturally transmitted, because the popu-
lation will exert downward causation on individuals’
learning strategies.
The major scientific finding of our analysis of the
experiment is that our human participants relied
heavily on pay-off-biased social learning, as predicted
by the evolutionary model. We think predictions
generated by an economic, rather than evolutionary
model, would make similar predictions, provided social
information was endogenous to the model. When there
is no additional cost to access pay-offs and the
information is subject to no error, as in this experiment,
then it is no surprise perhaps that a successful strategy
will attend to pay-offs. What might be more counter-
intuitive is the hierarchical combination of pay-off
and frequency biases. The evidence strongly suggests
that our participants used a strategy akin to: (i) Are the
two choices’ pay-offs similar on average? (ii) If yes,
which is more common? (iii) If no, which has the higher
average pay-off?
It also worth noting that participants did not require
any training time to learn to attend strongly to pay-
offs—they did so from the first period when social
information was present. We make no strong claims
about the source of these strategies. Social learning
strategies may of course themselves be learned socially,
and we have wondered about the effects of this in
previous experiments (McElreath et al. 2005). Indeed,
there is likely hidden strategic variation among
participants. Our analysis approach, fitting a single
model at a time to the entire set of data, is a common
approach, because rarely do we have enough data on
each participant to reliably distinguish differences in
strategy. However, in principle, the statistical methods
here do not require one to conduct the analysis this
way. Each participant can be analysed separately, or a
series of fixed effects parameters can be used to
statistically model individual differences. In the analysis
here, the overwhelming support of the best model
implies little strategic variation that could be detected
by the considered models. However, we do not think
this means all participants used the same strategy,
merely that we have not modelled the kind of
differences that exist.
Phil. Trans. R. Soc. B (2008)
A common reaction, both by ourselves and our
colleagues, to experiments with students is to be
sceptical of the generality of the results. True,
university students are a special population that is
likely not typical of the human species. However, no
single population will likely be representative of the
human species. That is, every culture and subculture
may be a special case. We think there are serious limits
to how much we can generalize from experiments with
students. But we also think that being able to explain
learning in any case is an advance. Just as studying the
evolution of beetle larva in the laboratory does not tell
us exactly how evolution works in any other species (or
even in wild beetles), the clarity of the results does
generate insights than can transfer across cases. Our
feeling is that no one should conclude the human
species is just like university students, anymore than
one should conclude all insects have the evolutionary
dynamics of flour beetles. But nor should one ignore
flour beetles, as if their evolution is not worth
explaining. University students are real people with
real learning strategies, and being able to model this
learning is worthwhile.
It is always possible that another, unconsidered,
strategy is a better description of the social learning
process. The same weakness is common to all analytical
approaches, however, and we caution readers not to
consider this a flaw special to the information criterion
and model comparison approach. But despite the
strong weight of evidence for this strategy here, we
think there is no substitute for replication and the
variation of experimental design in order to test the
robustness of a result. Both our experiments and
theoretical analysis are special, like all experiments
and models. Whatever the source of social learning
strategies—cultural or genetic or (likely) both—the
strategies we find in our experiments certainly did not
evolve in the laboratory. And however useful simple
evolutionary models are for exploring the logic of
population dynamics, they cannot and do not attempt
to replicate reality. We have emphasized the generality
of the statistical approach, as it is not tied to any
particular experiment or set of predictions, but it is
worth noting key assumptions of both the experiments
and models.
First, the experimental environment we have used
provides highly accurate (noise free) pay-off infor-
mation, whereas real social environments certainly do
not. In addition, real social environments may provide
cues of success, but these cues will often be integrations
of the contributions of many separate behaviours. For
example, if someone in your town is healthy, is it a
result of her diet, her religion or her close bonds with
kin? This integrated nature of cues of success means
that people may copy many traits from successful or
prestigious individuals, with potentially important
effects on cultural dynamics (Boyd & Richerson
1985; Henrich & Gil-White 2001). Relevant to our
experimental results and the prediction of the model
that pay-off bias would dominate, it may be that the
clear advantage of pay-off bias depends upon the ability
to know that any cue of success arises from a particular
behaviour. If not, other forms of social learning may be
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Analysis on social learning strategies
more competitive. Some of our ongoing experiments
explore this consideration.
Second, our evolutionary analysis is built upon a
number of existing models of the evolution of social
learning (Boyd & Richerson 1988, 1995, 1996; Rogers
1988; McElreath & Strimling 2008). By doing so, it is
comparable to these models, but also considers a fairly
special life history. In all of these models, generations
are barely overlapping: adults survive only long enough
to be imitated. Grandparents never survive to be
imitated. There is no population structure, including
within the biological family, and therefore any effects of
gene-culture covariation are ignored (see however
McElreath & Strimling 2008). While this kind of
model provides perhaps the purest evaluation of the
logic and economics of social learning strategies, actual
strategies may have evolved (culturally or genetically)
under rather special conditions or in order to exploit
overlapping generations. If so, the inferences derived
from these models will be misleading. How they will be
misleading is hard to say, until more social learning
theory exploring population structure and overlapping
generations appears.
This lacuna of theory aside, the existing evolutionary
literature is sufficient to motivate the search for positive
frequency dependence and kinds of pay-off bias in
other apes, if not crows, whales and rats. In the search
for the psychological differences that make human
cultural evolution qualitatively different from that of
other animals, the existence of frequency-dependent
and refined pay-off bias is often ignored. For example,
experiments in which apes see three ape demonstrators
access food through a two-action problem, with two
demonstrators performing one action and the third
another, will produce data that can estimate the
magnitude of positive frequency dependence.
This research was funded by the National Science
Foundation.
REFERENCES
Akaike, H. 1974 A new look at the statistical model
identification. IEEE Trans. Autom. Control 19, 716–723.
(doi:10.1109/TAC.1974.1100705)
Anderson, D. R., Burnham, K. P. & Thompson, W. L. 2000
Null hypothesis testing: problems, prevalence, and an
alternative. J. Wildl. Manage. 64, 912–923. (doi:10.2307/
3803199)
Baum, W. M., Richerson, P. J., Efferson, C. M. & Paciotti,
B. M. 2004 Cultural evolution in laboratory microsoci-
eties including traditions of rule giving and rule following.
Evol. Hum. Behav. 25, 305–326. (doi:10.1016/j.evolhum-
behav.2004.05.003)
Boesch, C. 2003 Is culture a golden barrier between human
and chimpanzee? Evol. Anthropol. 12, 82–91. (doi:10.
1002/evan.10106)
Boesch, C. & Tomasello, M. 1998 Chimpanzee and human
culture. Curr. Anthropol. 39, 591–604. (doi:10.1086/
204785)
Bolker, B. and based on stats4 by the R Development Core
Team 2008 bbmle: tools for general maximum likelihood
estimation. R package v. 0.8.5.
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Boyd, R. & Richerson, P. 1988 An evolutionary model of
social learning: The effects of spatial and temporal
Phil. Trans. R. Soc. B (2008)
R. McElreath et al. 3527
variation. In Social learning: a psychological and biological
approach (eds T. Zentall & B. G. Galef ), pp. 29–48.
Hillsdale, NJ: Lawrence Erlbaum Associates.
Boyd, R. & Richerson, P. J. 1992 How microevolutionary
processes give rise to history. In Evolution and history (ed.
M. Niteki), pp. 149–178. Chicago, IL: University of
Chicago Press.
Boyd, R. & Richerson, P. 1995 Why does culture increase
human adaptability? Ethol. Sociobiol. 16, 125–143. (doi:10.
1016/0162-3095(94)00073-G)
Boyd, R. & Richerson, P. J. 1996 Why culture is common, but
cultural evolution is rare. In Evolution of social behaviour
patterns in primates and man (eds W. G. Runciman &
R. I. M. D. John Maynard Smith) Proc. British Academy,
pp. 77–93. Oxford, UK: Oxford University Press.
Boyd, R. & Richerson, P. J. 2002 Group beneficial norms
spread rapidly in a structured population. J. Theor. Biol.
215, 287–296. (doi:10.1006/jtbi.2001.2515)
Burnham, K. & Anderson, D. 2002 Model selection and multi-
model inference: a practical information–theoretic approach.
Berlin, Germany: Springer.
Camerer, C. 2003 Behavioral game theory: experiments in
strategic interaction. The Roundtable Series in Behavioral
Economics. Princeton, NJ: Princeton University Press.
Camerer, C. F. & Ho, T. 1999 Experience-weighted
attraction learning in normal form games. Econometrica
67, 827–874. (doi:10.1111/1468-0262.00054)
Cavalli-Sforza, L. L. & Feldman, M. W. 1981 Cultural
transmission and evolution: a quantitative approach. Mono-
graphs in Population Biology, vol. 16. Princeton, NJ:
Princeton University Press. [L. L. Cavalli-Sforza and
M. W. Feldman. ill.; 23 cm. Includes index.]
Cohen, J. 1994 The earth is round ( p!0.05). Am. Psychol.
49, 997–1000. (doi:10.1037/0003-066X.49.12.997)
Efferson, C., Richerson, P. J., McElreath, R., Lubell, M.,
Edsten, E., Waring, T. M., Paciotti, B. & Baum, W. 2007
Learning, productivity, and noise: an experimental study
of cultural transmission on the bolivian altiplano. Evol.
Hum. Behav. 28, 11–17. (doi:10.1016/j.evolhumbehav.
2006.05.005)
Estlund, D. M. 1994 Opinion leaders, independence, and
condorcet’s jury theorem. Theory Decision 36, 131–162.
[In English.] (doi:10.1007/BF01079210)
Forster, M. R. & Sober, E. 1994 How to tell when simpler,
more unified, or less ad hoc theories will provide more
accurate predictions. Br. J. Phil. Sci. 45, 35.
Fragaszy, D. & Perry, S. (eds) 2003 The biology of traditions:
models and evidence. Cambridge, UK: Cambridge
University Press.
Frank, S. A. 2006 Social selection. In Evolutionary genetics:
concepts and case studies (eds C. W. Fox & J. B. Wolf ),
pp. 350–363. Oxford, UK: Oxford University Press.
Galef, B. G. & Whiskin, E. 1997 Effects of social and asocial
learning on longevity of food-preference traditions. Anim.
Behav. 53, 1313–1322. (doi:10.1006/anbe.1996.0366)
Henrich, J. 2001 Cultural transmission and the diffusion of
innovations: adoption dynamics indicate that biased
cultural transmission is the predominate force in beha-
vioral change. Am. Anthropol. 103, 992–1013. (doi:10.
1525/aa.2001.103.4.992)
Henrich, J. & Boyd, R. 1998 The evolution of conformist
transmission and between-group differences. Evol. Hum.
Behav. 19, 215–242. (doi:10.1016/S1090-5138(98)
00018-X)
Henrich, J. & Gil-White, F. 2001 The evolution of prestige:
freely conferred status as a mechanism for enhancing the
benefits of cultural transmission. Evol. Hum. Behav. 22,
1–32. (doi:10.1016/S1090-5138(00)00071-4)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3528 R. McElreath et al. Analysis on social learning strategies
Henrich, J. & McElreath, R. 2003 The evolution of cultural
evolution. Evol. Anthropol. 12, 123–135. (doi:10.1002/
evan.10110)
Henrich, J. & McElreath, R. 2007 Dual inheritance theory:
the evolution of human cultural capacities and cultural
evolution. In Oxford handbook of evolutionary psychology
(eds R. Dunbar & L. Barrett), pp. 555–570. Oxford, UK:
Oxford University Press.
Horner, V. & Whiten, A. 2005 Causal knowledge
and imitation/emulation switching in chimpanzees (Pan
troglodytes) and children (Homo sapiens). Anim. Cognit. 8,
164–181. (doi:10.1007/s10071-004-0239-6)
Johnson, J. B. & Omland, K. S. 2003 Model selection in
ecology and evolution. Trends Ecol. Evol. 19, 101–108.
(doi:10.1016/j.tree.2003.10.013)
Kokko, H., Jennions, M. & Brooks, R. 2006 Unifying and testing
models of sexual selection. Annu. Rev. Ecol. Evol. Syst. 37,
43–66. (doi:10.1146/annurev.ecolsys.37.091305.110259)
Laland, K. N. 2004 Social learning strategies. Learn. Behav.
32, 4–14.
Laland, K. N. & Janik, V. M. 2006 The animal cultures
debate. Trends Ecol. Evol. 21, 544–547. (doi:10.1016/
j.tree.2006.06.005)
McElreath, R. & Boyd, R. 2007 Mathematical models of social
evolution: a guide for the perplexed. Chicago, IL: University
of Chicago Press.
McElreath, R. & Strimling, P. 2008 When natural selection
favors imitation of parents. Curr. Anthropol. 49, 307–316.
(doi:10.1086/524364)
McElreath, R., Lubell, M., Richerson, P. J., Waring, T. M.,
Baum, W., Edsten, E., Efferson, C. & Paciotti, B. 2005
Applying evolutionary models to the laboratory study of
social learning. Evol. Hum. Behav. 26, 483–508. (doi:10.
1016/j.evolhumbehav.2005.04.003)
Payne, J. W., Bettman, J. R. & Johnson, E. J. 1993 The
adaptive decision maker. Cambridge, UK: Cambridge
University Press.
R Development Core Team 2008 R: a language and
environment for statistical computing. Vienna, Austria: R
Foundation for Statistical Computing.
Phil. Trans. R. Soc. B (2008)
Richerson, P. J. & Boyd, R. 2005 Not by genes alone: how
culture transformed human biology. Chicago, IL: University
of Chicago Press.
Rogers, A. R. 1988 Does biology constrain culture? Am.
Anthropol. 90, 819–831. (doi:10.1525/aa.1988.90.4.02a
00030)
Schlag, K. H. 1998 Why imitate, and if so, how? J. Econ.
Theory 78, 130–156. (doi:10.1006/jeth.1997.2347)
Schlag, K. H. 1999 Which one should I imitate?
J. Math. Econ. 31, 493–522. (doi:10.1016/S0304-4068
(97)00068-2)
Schotter, A. & Sopher, B. 2003 Social learning and
coordination conventions in intergenerational games: an
experimental study. J. Polit. Econ. 111, 498–529. (doi:10.
1086/374187)
Stahl, D. O. 2000 Rule learning in symmetric normal-form
games: theory and evidence. Games Econ. Behav. 32,
105–138. (doi:10.1006/game.1999.0754)
Whiten, A. & Ham, R. 1992 On the nature and evolution
of imitation in the animal kingdom: reappraisal of a
century of research. In On the nature and evolution of
imitation in the animal kingdom: reappraisal of a century of
research (eds P. Slater, J. Rosenblatt, C. Beer &
M. Milinski), pp. 239–283. New York, NY: Academic.
Whiten, A., Goodall, J., McGrew, W., Nishida, T., Reynolds,
V., Sugiyama, Y., Tutin, C., Wrangham, R. & Boesch, C.
1999 Cultures in chimpanzees. Nature 399, 682–685.
(doi:10.1038/21415)
Whiten, A., Horner, V., Litchfield, C. & Marshall-Pescini, S.
2004 How do apes ape? Learn. Behav. 32, 36–52.
Whiten, A., Horner, V. & de Waal, F. 2005 Conformity to
cultural norms of tool use in chimpanzees. Nature 437,
737–740. (doi:10.1038/nature04047)
Whiten, A., Spiteri, A., Horner, V., Bonnie, K. E., Lambeth,
S. P., Schapiro, S. J. & de Waal, F. B. M. 2007
Transmission of multiple traditions within and between
chimpanzee groups. Curr. Biol. 17, 1038–1043. (doi:10.
1016/j.cub.2007.05.031)
Zajonc, R. B. 1965 Social facilitation. Science 149, 269–274.
(doi:10.1126/science.149.3681.269)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3529–3539

doi:10.1098/rstb.2008.0133
Published online 17 September 2008

Review

Studying cumulative cultural evolution

in the laboratory
Christine A. Caldwell* and Ailsa E. Millen
Department of Psychology, University of Stirling, Stirling FK9 4LA, UK
Cumulative cultural evolution is the term given to a particular kind of social learning, which allows
for the accumulation of modifications over time, involving a ratchet-like effect where successful
modifications are maintained until they can be improved upon. There has been great interest in the
topic of cumulative cultural evolution from researchers from a wide variety of disciplines, but until
recently there were no experimental studies of this phenomenon. Here, we describe our motivations
for developing experimental methods for studying cumulative cultural evolution and review the results
we have obtained using these techniques. The results that we describe have provided insights into
understanding the outcomes of cultural processes at the population level. Our experiments show that
cumulative cultural evolution can result in adaptive complexity in behaviour and can also produce
convergence in behaviour. These findings lend support to ideas that some behaviours commonly
attributed to natural selection and innate tendencies could in fact be shaped by cultural processes.
Keywords: culture; cumulative cultural evolution; ratchet effect; social learning

1. BACKGROUND only creative invention but also, and just as importantly,

In this review, we aim to explain why there is currently a
need for an experimental science of cumulative cultural
evolution. We will discuss methods that we have
developed, which we believe can be employed in
order to test experimental hypotheses about cumulative
cultural evolution. We will also discuss the results that
we have obtained using these methods and the
implications that these findings have for our under-
standing of the effects of cumulative cultural evolution
on human behaviour.
(a) What is cumulative cultural evolution?
In order to explain the motivation behind our research
on cumulative cultural evolution, we begin by explain-
ing why it is an interesting behavioural phenomenon
and an important topic of study. Cumulative cultural
evolution is distinct from culture in the general sense in
a number of ways. While culture is accepted by most to
refer to a socially transmitted heritage peculiar to a
particular society (Boyd & Richerson 1985),1 the
definition of cumulative cultural evolution is consider-
ably narrower. Boyd & Richerson (1994) showed that
social learning could increase the average fitness of a
population if it permitted ‘learned improvements to
accumulate from one generation to the next’ (p. 134),
essentially describing what they later termed cumulative
cultural evolution (Boyd & Richerson 1996). Tomasello
(1990, 1999; Tomasello et al. 1993) has coined the term
‘the ratchet effect’ to capture a similar notion: ‘The
process of cumulative cultural evolution requires not
* Author for correspondence (c.a.caldwell@stir.ac.uk).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
faithful social transmission that can work as a ratchet to
prevent slippage backward—so that the newly invented
artifact or practice preserves its new and improved form
at least somewhat faithfully until a further modification
or improvement comes along.’ (Tomasello 1999, p. 5).
Hence, cumulative cultural evolution refers to situations
in which social transmission allows for successive
improvements to performance over generations of
learners, generated by the accumulation of modifi-
cations to the transmitted behaviours.
Cumulative cultural evolution in this sense should
be distinguished from cultural evolution that does not
lead to appreciable improvement in the efficiency of the
behaviours in question. Mesoudi et al. (2004) have
argued that all human culture constitutes an evolution-
ary process, since it involves variation (multiple traits
that may be copied), heritability (similarity between
traits as a result of copying) and competition (some
traits are copied more than others), leading to the
accumulation of modifications over time. However, not
all such examples involve increasing efficiency or
complexity. They therefore do not constitute the kind
of learning that Boyd & Richerson (1994) or Tomasello
(1999) were referring to, whereby each generation is
provided with short cuts to the end results of extensive
trial and error learning amassed by their cultural
ancestors. So, although methods developed in
evolutionary biology are currently proving extremely
useful in reconstructing the history of cultural
products, such as textile designs (Tehrani & Collard
2002), linguistic forms (Gray & Jordan 2000) and
stories (Barbrook et al. 1998), these examples do not
represent the kind of process that we are referring to as
cumulative cultural evolution.

3529 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3530 C. A. Caldwell & A. E. Millen Review. Cumulative culture in the laboratory
All the same, even in these narrow terms, cumulative
cultural evolution seems to pervade human society.
Each generation builds on the knowledge, inventions
and achievements of the previous one. Our present-day
technologies exist only as a result of our ability to
understand and make use of the imparted knowledge
and artefacts of others. By contrast, the phenomenon of
cumulative cultural evolution seems to be intriguingly
rare in non-humans. This is despite the fact that there
are plenty of cases of animal culture in the more general
sense. To take just one well-known example, the sweet
potato washing behaviour of the Japanese macaques of
Koshima (e.g. Kawamura 1959; Kawai 1965) appears
to have been acquired through social learning, since the
behaviour spread initially to the close associates of the
first monkey to use this technique, and then on to those
individuals’ associates. However, as a number of
researchers have pointed out (most notably Boyd &
Richerson 1996), examples of animal social learning
such as this typically involve relatively simple
behaviours that could also be readily learned through
individual trial and error processes. There is little
evidence of successive improvement over generations,
or of the accumulation of modifications, and therefore
no suggestion that the behaviours concerned could not
have been invented by a single individual.
(b) Debates in cumulative cultural evolution
A strong motivation for us in developing experimental
methods for studying cumulative cultural evolution was
the fact that there are a number of important
unresolved issues surrounding this topic, which have
been the focus of much debate. The issue of whether
cumulative cultural evolution is unique to humans, or
merely relatively rare in non-humans, is just one of
these. In addition, there is still disagreement over the
learning mechanisms upon which cumulative cultural
evolution may depend, and also the extent to which it is
responsible for complex adaptive human behaviours.
The details of these debates are summarized below.
(i) Human uniqueness
The question that has probably drawn the most attention
surrounding cumulative cultural evolution is that of
whether the phenomenon is unique to humans. As noted
above, Boyd & Richerson (1996), among others, have
drawn attention to the fact that although social learning
is relatively common in the animal kingdom, cumulative
cultural evolution appears to be extremely rare: ‘cumu-
lative cultural evolution resulting in behaviors that no
individual could invent on their own is limited to
humans, song birds, and perhaps chimpanzees’
(Boyd & Richerson 1996, p. 77). Likewise, Heyes
(1993) has drawn a distinction between the sorts of
behaviours that non-humans appear to learn socially, in
comparison with those of humans: ‘the human attributes
that are described as ‘cultural’ in ordinary discourse,
seem to be a good deal more complex than, for example,
potato washing and termite-fishing.and it is plausible
that their greater complexity derives from the accumu-
lation of modifications’ (Heyes 1993, p. 1004).
Even stronger statements have been made by others.
For example, Galef (1992) has stated that ‘human
culture accumulates over generations and can lead to
Phil. Trans. R. Soc. B (2008)
invention and transmission of increasingly complex
behaviours. No one has claimed that any animal learns
any behaviour from conspecifics that it could not learn
independently through interaction with its physical
environment’ (Galef 1992, p. 161). Along similar lines,
Tomasello (1999) has asserted that ‘the cultural
traditions and artifacts of human beings accumulate
modifications over time in a way that those of other
animal species do not’ (Tomasello 1999, p. 5).
Of course, these authors are not claiming that social
learning does not help animals to acquire useful
behaviours. The example of the Koshima macaques
illustrates how an advantageous invention can readily
spread through a group. The issue is again the notion of
the cultural ratchet, and of later generations exploiting
the labours of previous ones in a way that allows them
to make use of behaviours that they could not have
learned by themselves.
All the same, some researchers have argued that there
is in fact compelling evidence for cumulative cultural
evolution in non-humans. For example, Boesch (2003)
cited three examples of chimpanzee behaviours that he
suggests may have arisen through accumulated modifi-
cations of socially learned behaviour. One of these is nut-
cracking behaviour, which is observed in West African
chimpanzee populations (Whiten et al. 1999). While
some chimpanzees use fixed anvils, such as tree roots,
others use loose stones. Furthermore, Sugiyama (1997)
reported that some chimpanzees at the Bossou field site
sometimes used an extra stone to stabilize the stone
anvils upon which the nut is placed for cracking, perhaps
indicating a refinement of the simpler technique.
Whiten et al. (2003) have put forward similar
examples of possible cumulative culture in chimpan-
zees, including the above-mentioned nut-cracking
behaviour. They also suggested that the alternative
tool-use techniques used to forage on ants could show
evidence of ratcheting. Of the two methods used, one
is more elaborate, involving bimanual coordination,
and also results in a greater yield ( Humle &
Matsuzawa 2002) and therefore Whiten et al. (2003)
suggested that this could represent an elaboration on
the simpler version.
Cumulative culture has also been proposed in
species other than chimpanzees. Hunt & Gray (2003)
proposed that the tool manufacture skills observed in
New Caledonian crows had been acquired through
cumulative cultural evolution. They were able to
document population-specific methods of tool manu-
facture, which were apparently unrelated to local
ecological conditions, and these different methods
showed varying degrees of complexity. They therefore
argued that these different techniques had been
developed through cumulative cultural evolution.
However, it is still unclear whether these techniques
are socially learned at all, as it seems that tool
manufacture abilities may be largely under genetic
control in this species (Kenward et al. 2005).
There is clearly good reason to remain open-minded
with regard to the question of whether cumulative
cultural evolution is unique to humans. Boesch &
Tomasello (1998), for example, have made the point
that we have very little data on the behaviour of
previous generations of chimpanzees, since long-term
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Cumulative culture in the laboratory
studies of their behaviour began only ca 40 years ago.
This inevitably makes it difficult to judge whether their
behaviour has shown any ratcheting over time.
(ii) The cognitive underpinnings of cumulative
cultural evolution
Related to the above debate, researchers have also
deliberated over the issue of the cognitive mechanisms
that may or may not be necessary for cumulative
culture. This is necessarily tied in with the issue of
which species exhibit this phenomenon, since beliefs
about which cognitive processes may be involved have
typically lain behind opinions about which species
possess the capacity.
For example, Boyd & Richerson (1996) have made a
compelling argument that a capacity for ‘true imitation’
is necessary for cumulative cultural evolution. By true
imitation they refer to any form of learning in which
individuals can learn a new behaviour by perceiving
another individual’s performance of that behaviour.
Boyd & Richerson (1996) also used the term ‘obser-
vational learning’, which may be slightly misleading in
this context as visual perception is not necessarily
involved, since their definition encompasses learning
about vocal communication, such as the learning of
grammatical rules (Richerson & Boyd 2005) and bird
song learning. In any case, other types of social
learning, which they specify cannot support cumulative
cultural evolution, are those in which the behaviour
itself is not what is learned from another individual.
Similar behaviours may arise between two individuals
because the actions of one individual function to draw
the attention of another to a particular location (‘local
enhancement’) or class of objects (‘stimulus enhance-
ment’; Whiten & Ham 1992). However, in such cases,
the behaviour is learned by trial and error, and the
presence of the other individual has simply made that
learning more likely.
Boyd & Richerson (1996) argued that, since
cumulative cultural evolution by definition must allow
learners to proceed from a more advanced starting
point than was possible for previous generations, true
imitation is a necessary condition for its occurrence.
Learning that relies upon trial and error processes will
inevitably mean that each new learner has to start from
scratch, thereby wiping out any useful innovations that
may have been chanced upon by others.
Tomasello (e.g. Tomasello et al. 1993; Tomasello
1999) has made a similar argument, proposing that
imitation and teaching, each dependent on a capacity
for taking the perspective of another, are the foun-
dations of cumulative cultural evolution. Like Boyd &
Richerson, Tomasello has not only stressed the
importance of faithful transmission (and therefore
imitation) but has also emphasized the need for an
understanding of the goals of other individuals. Under-
standing what cultural practices are about, such as what
a tool is used for, or what a particular communicative
signal means, is crucial to human cultural learning
(Tomasello 1999), so he has suggested this to be
another necessary feature of cumulative cultural
evolution (see also Hermann et al. (2007) for a recent
conceptualization of this view).
Phil. Trans. R. Soc. B (2008)
C. A. Caldwell & A. E. Millen 3531
However, not all theorists see things in this way.
Heyes (1993), for example, has stated that there is no
reason why imitation should be particularly crucial to
the generation of cumulative behavioural change. Heyes
(1993) proposed that the particular learning mechanism
involved is in fact irrelevant to the issue of faithful
transmission, and that what really matters is whether or
not a behaviour will extinguish once learned. Heyes
(1993) cited Galef et al.’s (1986) data from a two-action
study on budgerigars: subjects that observed cons-
pecifics accessing hidden food using either their beak or
their foot tended to match the demonstrated technique
for only the first two trials post demonstration. In later
trials, the difference between the groups disappeared.
Therefore, she has argued that behaviours learned via
imitation, like any other, will be subject to modification
from trial and error learning. Heyes (1993) proposed
that in fact cumulative culture would be more likely to
be supported by a separate insulating mechanism that
protects socially learned information against the influ-
ence of trial and error learning.
Laland & Hoppitt (2003), similarly, have argued
that there is currently no reason to believe that either
imitation or teaching is particularly significant to
cumulative cultural evolution. Laland (2004) has
instead suggested that an ability to evaluate the relative
effectiveness of behavioural alternatives would be a
more plausible cognitive precursor to cumulative
culture. Along similar lines, Enquist & Ghirlanda
(2007) have argued for the importance of an ‘adaptive
filtering’ mechanism. Modelling populations of social
learners, Enquist & Ghirlanda (2007) concluded that,
without such a filter, cumulative culture would result in
the acquisition of many maladaptive traits and there-
fore could not evolve. However, a capacity to filter out
maladaptive traits causes adaptive traits to accumulate
preferentially, generating the ratchet effect that is
characteristic of human culture.
In conclusion, therefore, there are a range of
different views regarding the cognitive abilities upon
which cumulative cultural evolution may depend.
However, experimental work on this topic could
contribute greatly to addressing this question, as we
hope to elucidate later, and such work will also have
implications for questions about human uniqueness.
(iii) The origins of complex human behaviour
Just as there is disagreement over the precursors to
cumulative cultural evolution, as detailed above, there
are also conflicting views on the outcomes of this
process, which we will outline here. There is broad
agreement that cumulative cultural evolution is
responsible for a number of particularly interesting
human traits. Indeed, this goes some way to explaining
why so many researchers have been fascinated by this
phenomenon. Tomasello (e.g. 1999) has argued that,
given that our species shared a common ancestor with
chimpanzees a mere 6 million years ago, the cognitive
achievements of modern humans (such as written
language, mathematics and complex technologies)
have developed implausibly rapidly to be attributed to
natural selection on behaviour. Instead he has
proposed that cumulative cultural evolution may have
played a significant role, and that this would have
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3532 C. A. Caldwell & A. E. Millen Review. Cumulative culture in the laboratory
allowed for much more rapid behavioural change than
would genetic evolution. Boyd & Richerson (1996)
have emphasized the influence of cumulative cultural
evolution in terms of the success of humans as a
species. Our ability to exploit a range of habitats has
allowed us to become the most widespread animal on
the planet.
However, recently, arguments have been put for-
ward which propose a role for cumulative cultural
evolution in the origins of behaviours that have been
popularly believed to be largely dependent on naturally
selected innate tendencies. For example, the structural
properties shared by many different languages (usually
referred to as linguistic universals) have often been
argued to be good evidence of innate language-specific
capabilities, common to all humans (e.g. Pinker &
Bloom 1990). However, this view has recently been
challenged by Kirby et al. (2007) and Christiansen &
Chater (in press), among others (see also Smith &
Kirby 2008).
Kirby et al. (2007) have argued that even extremely
weak biases in learning (such as an expectation of
regularity) can, over generations of learners, result in
languages that are strongly adapted to those biases. The
result of this is that no language-specific capacity is
necessary in order to explain the existence of linguistic
universals, as universals could arise from slight biases in
general-purpose learning mechanisms. Within this view,
languages are therefore seen as shaped by the brain
(Christiansen & Chater in press) through the repeated
cycle of learning and use over many generations, rather
than the brain being adapted to language.
The crucial question here really is whether cumu-
lative cultural evolution, driven by general learning
mechanisms, can account for cross-cultural universals
in behaviour. Universality in complex adaptive human
behaviour is often taken as a hallmark of highly
specialized innate predispositions (e.g. Buss 1989;
Pinker & Bloom 1990). However, since cumulative
cultural evolution can similarly result in complex
adaptive behaviour, the issue is really whether it also
results in convergence in behaviour, such that separate
populations independently invent and retain similar
behaviours. While culture is generally viewed as a
source of behavioural variation between cultures, under
certain circumstances it may result in cross-cultural
convergence. We explain in §5 how our experimental
work has so far contributed to this question.
Clearly, therefore, there are currently some extremely
interesting intellectual disputes within the field of
cumulative cultural evolution, each of which could
benefit from empirical studies of this phenomenon. In
§2 we will examine existing approaches to studying
cumulative cultural evolution, and what these studies
have so far been able to contribute.
2. APPROACHES TO STUDYING CUMULATIVE
CULTURAL EVOLUTION
(a) Studying cumulative cultural evolution
in the field
In order to study the phenomenon of cumulative
cultural evolution in natural populations, it is necessary
to have access to fairly accurate information about the
Phil. Trans. R. Soc. B (2008)
past forms of behaviours. This can be difficult because
many behaviours leave no discernable trace. However,
it is possible to study cumulative cultural evolution
from cultural artefacts, including archaeological find-
ings. For example, we can infer from the archaeological
record (or rather, the lack of it) that any tools used by
hominids up until ca 2 million years ago were probably
comparable with those used by other great apes, such as
chimpanzees (Ambrose 2001). We also know that,
approximately a quarter of a million years ago, tools
manufactured by humans began to show rapid change
and development. Around this time, a wide variety of
different stone tools were being used by humans, each
tailored to a specific function.
Some authors have documented progress in science
and technology in explicitly evolutionary terms (e.g.
Wilder’s 1968 Evolution of Mathematical Concepts;
Basalla’s 1988 The Evolution of Technology). In other
texts, the notion of accumulation is more implicit.
Inventions and discoveries are dated and advancement
is assumed (e.g. Bunch & Hellemans 2004). However,
irrespective of the source consulted, evidence can
readily be found for ratcheting, with new developments
building on previous ones. The loss of information or
skills from a population is very much the exception,
rather than the rule (although it is worth noting that
such loss certainly has been documented; the decline in
complexity of the Tasmanian toolkit is probably the
best known example, e.g. Henrich 2004).
Consider two examples, the wheel and the math-
ematical notation. The invention of the wheel as an aid
to transportation is generally dated between 5000 and
6000 years ago, with early wheels constituting simple
wooden discs with a hole for the axle. The spoked wheel
was invented only more recently (ca 4000 years ago).
The invention of the wheel then gave rise to a number of
other technological innovations including cogs and
pulleys (Basalla 1988; Bunch & Hellemans 2004).
Mathematical systems of notation have also developed
considerably over time, with simple tally systems
recorded from ca 30 000 years ago. However, more
abstract symbolic representations are much more recent.
Place-value notation was being used by the Mesopota-
mians by ca 4000 years ago, but also seems to have been
independently invented by the Indians within the last
2000 years, leading to our current ‘Arabic’ notation.
The invention of a symbol to represent zero (which
made the place notation system considerably more
informative) was more recent still, in both cultures
(Wilder 1968; Bunch & Hellemans 2004).
Clearly, valuable information can be gleaned from
studying cumulative cultural evolution in natural
populations. Important insights can be gained into
the kinds of behaviours that show cumulative cultural
evolution over time and the nature of the changes that
occur. These can, up to a point, address certain issues
raised in §1. For example, adequate information on the
past forms of behaviour in populations that have had
little contact with one another could enhance our
understanding of convergence in cumulative cultural
evolution, and therefore the extent to which it may
explain cross-cultural universals. However, there are
limitations to the kinds of questions one can ask when
using these kinds of data. Mesoudi (Mesoudi 2007;
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Cumulative culture in the laboratory
Mesoudi & O’Brien 2008) has argued that historical
methods have several drawbacks when it comes to
studying cultural variation, and the same apply for
investigations of cumulative cultural evolution. The
main weakness of this approach, certainly from the
perspective of addressing some of the debates listed
above, is that it does not allow for the manipulation of
variables of interest. We can ask questions about how
cumulative cultural evolution has occurred, but we
cannot ask what might have happened, had circum-
stances been slightly different. By contrast, using an
experimental approach, we can explicitly manipulate
factors believed to be crucial in generating cumulative
cultural evolution, to test hypotheses about necessary
precursors. Likewise we can set out to study the
outcomes of cumulative cultural evolution in multiple
replicated populations under controlled conditions.
(b) Theoretical studies of cumulative
cultural evolution
By contrast, theoretical studies of cumulative cultural
evolution permit researchers to manipulate as many
variables as they desire, and controlling extraneous
factors is of course not an issue. There are many
theoretical models involving social learning, but only a
few have tested hypotheses relevant to the debates
detailed previously. For instance, Boyd & Richerson’s
(1996) paper included a theoretical analysis of why
cumulative cultural evolution seems to be a rare
phenomenon within the animal kingdom. Their
models showed that an ability to copy the behaviour
of others becomes more useful than a capacity for
individual learning only when other members of the
population are themselves engaging in behaviours with
higher pay-offs than would be achieved by individual
learning alone. Hence there is a significant obstacle to
the evolution of such capabilities since they are valuable
only once they are widespread in the population.
Enquist & Ghirlanda (2007) have addressed the issue
of the cognitive abilities necessary for cumulative
cultural evolution. As mentioned above, their models
indicate that a mechanism that can selectively filter out
maladaptive behaviours would be crucial for cumulative
cultural evolution to evolve. Work by Kirby and
colleagues (e.g. Kirby & Christiansen 2003; Smith
et al. 2003; Kirby et al. 2007; Smith & Kirby 2008) has
investigated the question of whether complex innate
competencies are a necessary feature for the evolution
of structured language. Modelling iterated learning of
communication systems, they have shown that struc-
tural features, such as compositionality, can arise
through cultural transmission over multiple generations.
However, the great strength of theoretical models, in
terms of the flexibility afforded, is also to an extent their
weakness. The constraints imposed on the models are
those selected by their creator, and the conclusions
drawn necessarily depend on the underlying assump-
tions, which may or may not be accurate. By contrast,
in experimental studies with human subjects, one can
gain important insights into the likely results of real
learning processes repeated over multiple generations.
In §3–6 we detail experimental approaches that can be
used to study culture in the laboratory (§3), and explain
Phil. Trans. R. Soc. B (2008)
C. A. Caldwell & A. E. Millen 3533
how we have applied these methods to test hypotheses
about cumulative cultural evolution (§4).
3. STUDYING CULTURE IN THE LABORATORY
There are a variety of methods that have been used for
studying culture under laboratory conditions (for
reviews see: Mesoudi 2007; Mesoudi & Whiten 2008;
Whiten & Mesoudi 2008). The aim of such approaches
is essentially to simulate cultural phenomena on a small
scale, allowing researchers to study how behaviours
change over time as a result of repeated learning and
transmission between individuals. While experimental
approaches inevitably have imperfections of their own,
we consider that the power to manipulate variables and
collect precisely the data that are required, constrained
and informed by the behaviour of real participants
strikes a very constructive balance.
The use of such methods is best illustrated with an
example of this kind of study. Jacobs & Campbell
(1961) used laboratory ‘microcultures’ (Gerard et al.
1956) aiming to ‘demonstrate a perpetuation of
‘cultural’ characteristics that transcends the replace-
ment of individual persons’ (p. 649). Their study
therefore involved simulating generational succession
through the repeated removal and replacement of
participants within small groups. Jacobs & Campbell
(1961) wanted to determine whether participants’
tendencies to conform to majority opinion could result
in long-lasting traditions of counter-intuitive beliefs.
Groups were founded by experimental confederates,
instructed to respond with a significant overestimation
of their true perception of the strength of a visual
movement illusion, but these were gradually replaced
by naive participants. In an example of one of Jacobs &
Campbell’s (1961) conditions, there were three indi-
viduals present in the test group at any one time, and at
the start of the experiment two of these individuals were
confederates and one was a naive participant. Each
individual from the group was asked to estimate the
degree of the illusory movement perceived, starting
with the confederates, and their responses were
recorded. This was repeated 30 times, after which one
of the confederates was removed and replaced by
another naive participant. Then 30 more trials were
carried out with this new group, and the remaining
confederate was then removed and replaced with a
further naive participant. This procedure continued for a
total of 10 ‘generations’. Jacobs & Campbell (1961) also
ran further conditions, manipulating the size of the test
group (varying between one individual and four) and the
number of confederates in the first generation (varying
between zero and three). Each experimental condition
was replicated three times each. Jacobs & Campbell
(1961) found that the overestimation bias induced by the
confederates persisted for several generations after the
final confederate had been removed.
Similar methods, involving the removal and replace-
ment of participants within groups, have since been
adopted by Insko et al. (1980, 1983) and also more
recently by Baum et al. (2004). The benefit of this
method lies not only in the power to manipulate certain
key variables (e.g. the size of group and the number of
confederates for Jacobs & Campbell 1961), but also in
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3534 C. A. Caldwell & A. E. Millen Review. Cumulative culture in the laboratory
the time scale required for the study. While cultural
evolution is generally assumed to occur on a time scale
of multiple human lifespans, these experiments seek to
study cultural phenomena over learner generations,
rather than reproductive generations. For this reason,
such methods have also been applied within the
literature on animal social learning (e.g. Galef &
Allen 1995; Laland & Williams 1997, 1998).
As well as methods involving removal and replace-
ment of participants in groups, there are also other
methods that can be used to study culture under
experimental conditions (for a comprehensive reviews
see Mesoudi 2007; Mesoudi & Whiten 2008; Whiten &
Mesoudi 2008), but these are less relevant in terms of
finding an experimental model for studying cumulative
cultural evolution. For example Bartlett (1932) made
use of the ‘method of serial reproduction’ in his studies
of human memory, a method that has been recently
revived by Mesoudi and colleagues (Mesoudi &
Whiten 2004; Mesoudi et al. 2006). Although chains
of multiple generations are involved, with new partici-
pants learning from previous learners, it is quite
different from the method detailed above in a number
of important respects. In this type of research,
participants are explicitly instructed to copy, in that
their aim is to reproduce information as accurately as
they possibly can. The focus is therefore on the
degradation of the information originally provided to
the first participant. While this is an excellent method
for revealing people’s unconscious cognitive biases (as
it allows researchers to investigate what sort of
information is omitted, or introduced, when partici-
pants are actively trying to reproduce material as
accurately as possible), it is clearly considerably less
appropriate for studying cumulative cultural evolution.
Any laboratory model of cumulative cultural evolution
must involve behaviours that can show measurable
improvement over generations, and it is also important
that participants understand that the choice of whether
or not to copy is their own.
Until recently, even the experimental work that has
been carried out using the replacement method has
fallen short of providing an adequate model of
cumulative cultural evolution. As noted above, the
behaviours to be transmitted must be capable of
showing measurable improvement over generations,
as a consequence of accumulated modifications. The
perceptual judgement task used by Jacobs & Campbell
(1961) is therefore far too simple a behaviour. Work
carried out by Baum et al. (2004; as well as other
work from the same group, e.g. McElreath et al. 2005,
2008; Efferson et al. 2007, 2008) does show increas-
ingly adaptive choices made by participants over
generations, but these experiments involve participants
choosing between two options each of which has an
unpredictable pay-off. This method has been used to
elucidate participants’ strategies in gauging their use of
socially and individually acquired information in order
to make best guess choices. In terms of studying
cumulative cultural evolution, however, the method is
less appropriate. While the average pay-offs seem to
increase over time, as a result of participants gravitating
towards the choice that is the best on average, the
behaviours themselves (of one choice over another) are
Phil. Trans. R. Soc. B (2008)
not ideal candidates in terms of demonstrating the
accumulation of modifications.
Interestingly, in Insko’s studies (Insko et al. 1980,
1983), which involved between-group trading of paper
origami products manufactured by group members,
groups made increasingly greater profits over gener-
ations, suggesting more efficient methods of production
were being passed on. However, it is unclear what the
nature of the improved efficiency was and whether this
was definitely attributable to transmission between
generations. In §4 we detail the method that we have
developed for studying cumulative cultural evolution
experimentally. Within this issue, studies reported by
Fay et al. (2008) and Flynn (2008) show similar effects to
ours, which may also provide promising approaches for
studying cumulative cultural evolution.
4. A METHOD FOR STUDYING CUMULATIVE
CULTURAL EVOLUTION IN THE LABORATORY
(a) The tasks
As explained above, the task presented to participants
must be chosen very carefully in order to create an
effective laboratory model of cumulative cultural
evolution. It was important that we chose a task that
could show measurable improvements in performance
over generations, based on the accumulation of
modifications. A task with a clear aim and an objective
measure of success was therefore crucial, as we needed
to be able to show that participants’ scores were, on
average, better the further they were down the chain
(therefore showing that the skills and knowledge had
accumulated over the learner generations). It was also
important for our design to choose tasks that were
simple and easy enough for participants to complete in
a short space of time, in order to make these feasible
laboratory methods. However, the tasks also needed to
be sufficiently difficult and complex that definite
benefits could be obtained from opportunities for
social learning, and that the accumulation of modifi-
cations could be documented.
We have so far used two different tasks in our
laboratory studies of cumulative cultural evolution
(Caldwell & Millen 2008). In one of our tasks,
participants are asked to build a paper aeroplane, the
goal being to build one that will fly as far as possible. In
the other task, participants are asked to build a tower
out of spaghetti and modelling clay. The goal for this
task is to build a tower that is as high as possible. Hence
we have our objective measures of the success of each
participant in relation to the goals they have been given.
Furthermore, the tasks that we have selected show
similarities with certain examples of early human
material culture, such as projectile point shaping and
shelter construction. From our point of view, this is also
helpful given the types of questions we would like to
address with these methods (see §1 above). The tasks
therefore have much in common, but there are also
important differences between them. First, while many
people have some prior experience of having built a
paper aeroplane, the spaghetti tower task is far more
novel and participants have few preconceived ideas
about how to approach the task. Second, while feedback
on performance on the spaghetti tower task is continual
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Cumulative culture in the laboratory
time participants present
minutes in test group
00.00 1 2 3
02.30 1 2 3 4
05.00 2 3 4 5
07.30 3 4 5 6
10.00 4 5 6 7
12.30 5 6 7 8
15.00 6 7 8 9
17.30 7 8 9 10
8 9 10
20.00
9 10
22.30
25.00 10
Figure 1. Group composition over time in the microsociety
design. Generational succession is simulated through the
repeated removal of experienced participants and introduc-
tion of naive participants. Each row of the table shows the
group composition at any given time, made up of observing
participants (grey) and participants actually engaged in
the task (black). Participants were randomly assigned the
positions 1–10.
during construction (participants can of course see
exactly how high their tower currently is), feedback on
performance on the paper aeroplane task is delayed until
construction is complete.
(b) The design
We used a replacement method, with each chain
totalling 10 individuals. For each task, we ran 10
replicates of these chains of 10 participants. Figure 1
shows a schematic of the replacement design, indicat-
ing which participants were present in the test group at
what point during each trial. Participants were
randomly assigned to the positions 1–10 in each
chain. In order to simulate generational succession,
the participants’ start times were staggered, such that
every 2.5 min a new person entered the group. While
they were in the test group, each participant had 5 min
of observation time, during which they could watch the
previous participants building their artefact, followed
by 5 min of building time, during which they had to
construct their own artefact. Once their time was up,
they left the test group. The staggered start and finish
times had the effect that, at any given time (except at
the very start and very end of any given chain), there
were four individuals together in the group, two of
whom were observing and two of whom were actually
engaged in the task (figure 1). So, for example, a chain
would begin with participant 1 building their artefact,
with participants 2 and 3 observing. Then, after
2.5 min, participant 2 would also start building and
participant 4 would join the group as an observer. The
aim was to simulate a miniaturized society, in which
one generation would have the opportunity to interact
with and observe individuals from the previous two
generations, but not those further back. However, we
did retain all artefacts for inspection by later partici-
pants, to reflect the more permanent record generated
by material culture. The experimenter wrote down the
relevant measurements next to each, so that this
information was also available. Participants left the
testing area once their artefact had been evaluated.
Phil. Trans. R. Soc. B (2008)
C. A. Caldwell & A. E. Millen 3535
(c) The ratchet effect
Our results showed clear evidence for improvement in
performance over the course of the chains (Caldwell &
Millen 2008). Figure 2 shows these results for
both the paper aeroplanes and the spaghetti towers.
Figure 2a(i),b(i) shows the results for each of the 10
chains separately, so each differently coloured line
represents a different chain. As is clear from these
figures, performance overall can be extremely variable.
However, when we (Caldwell & Millen 2008) analysed
the trends over generations, a strong effect of improve-
ment in terms of the goal measures (of plane flight
distance and tower height) was found. Figure 2a(ii),
b(ii) shows the average score for the participants in each
position in the chain and illustrates the steady
improvement much more clearly. Thus, skills and
knowledge do indeed appear to accumulate in the
chains, independent of individual membership,
consistent with the predictions assuming cumulative
cultural evolution. Furthermore, similar patterns were
found for both the tasks, in spite of the differences
between them, suggesting that we are tapping into a
fairly general phenomenon.
(d) Accumulation of modifications
As well as looking for improvement in performance
over generations, we were also interested in investi-
gating the inheritance of modifications. We predicted
that designs would be more similar within chains than
they were across chains (indicating cultural variation),
and also that designs that were close together in the
chain would be more similar to one another than those
that were far apart (indicating descent with modifi-
cation). For this reason, we took photographs of all the
artefacts that had been created by participants, and we
were able to use these to test predictions regarding the
similarity of designs. The photographs were rated by
naive coders, each of whom was given one of the
photographs from the set and asked to rate it in
comparison with all the others. The coders were
provided with a seven-point scale, which they were to
refer to in making their ratings. For full details of the
methods used to obtain and analyse these ratings, see
Caldwell & Millen (2008). As predicted, artefacts from
the same chain were rated as more similar than those
from different chains, for both the paper aeroplanes and
the spaghetti towers. Designs from positions close
together in the chains were also more similar than those
from positions far apart in chains, demonstrating that
the improvement in performance was associated with
the accumulation of modifications.
5. APPLICATIONS OF THESE METHODS
We see great potential for using these methods in order
to test hypotheses about cumulative cultural evolution.
In this section we discuss how some of our work to date
has helped to address some of the issues raised in the
introduction. We start by discussing our findings
regarding cultural convergence, and explain the
implications for the potential role of cumulative
cultural evolution in human behavioural universals.
We also discuss experiments explicitly designed to test
hypotheses about learning mechanisms involved in
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3536 C. A. Caldwell & A. E. Millen
(a) 1250 (i)
dist. flown by
1000
plane (cm) 750
500
250
0
(b) 100 (i)
80
tower (cm)
height of
60
40
20
0
1 2 3 4 5 6 7
position in chain
Figure 2. Measures of success over generations. (a(i)) The data for the 10 chains of paper aeroplanes, with (a(ii)) showing the
mean distance flown for each position in the chains (error bars indicate Gs.e.m.). (b(i)) The data for the 10 chains of spaghetti
towers, with (b(ii)) showing the mean height for each position in the chains (error bars indicate Gs.e.m.). Redrawn from
Caldwell & Millen (2008).
cumulative cultural evolution. This work has obvious
implications for the debate surrounding the cognitive
precursors of cumulative cultural evolution, but is also
relevant to the debate regarding the uniqueness (or
otherwise) of this phenomenon to humans.
(a) Convergence in cumulative cultural evolution
As mentioned in §1, questions about the possible role
of cumulative cultural evolution in complex human
behavioural universals essentially centre on the issue of
whether separate cultures are likely to independently
invent and retain similar behaviours. We have already
been able to investigate this issue, based on the same
dataset already discussed (Caldwell & Millen 2008).
In biological evolution, convergent evolution refers
to a process by which species that are only distantly
related independently evolve analogous adaptations in
response to similar environmental pressures. Conver-
gent cultural evolution therefore refers to situations in
which different populations independently develop
similar socially transmitted behaviours despite different
ancestral histories (Caldwell in press). In evolutionary
biology, distinctions are therefore also drawn between
traits that are homologous and analogous. Homology
refers to the ‘relationship of two characters that have
descended, usually with divergence, from a common
ancestral character’ (Fitch 2000, p. 227). By contrast,
analogy ‘is distinguished from homology in that its
characters, although similar, have descended conver-
gently from unrelated ancestral characters.’ ( Fitch
2000, p. 227).
Likewise, it is an important question when studying
cross-cultural similarities, whether those similarities
are the result of common cultural descent or conver-
gence from contrasting ancestral forms. In our
experiments it is possible to study the extent to which
convergence occurs, since we can control the contact
that different microcultures have with one another.
Phil. Trans. R. Soc. B (2008)
Review. Cumulative culture in the laboratory
(ii)
(ii)
8 9 10 1 2 3 4 5 6 7 8 9 10
position in chain
As noted in §4d, we had all our photographs of
participants’ planes and towers rated for their similarity
to one another. As well as using these ratings to look
into the accumulation of modifications, we were also
able to use the ratings to test for convergent cultural
evolution (i.e. increasing similarity between the
different chains, over generations). This would be
predicted because successful designs are liable to have
features in common, so the later designs ought to be
rated as more similar to one another, compared with
earlier designs, due to the fact that they have in effect
been shaped by similar selection pressures. In order to
analyse this, we took the similarity ratings for all pairs of
photographs that were in the same position across
chains. Positive correlations were found between the
position in the chain and the mean similarity ratings,
for both tasks. These results are displayed in figure 3.
We are by no means the first to show that ‘iterated
learning’, i.e. each generation learning from data
generated by the previous one, can result in conver-
gence across different microsocieties. Kalish et al.
(2007), for example, have well illustrated that learners’
biases interfere with transmitted information in
consistent ways, resulting in different chains of
learners—who have often been provided with quite
different information to start with—passing on very
similar data (see also Griffiths et al. 2008). However,
such findings, albeit fascinating, may not contribute
greatly to the question of whether cumulative cultural
evolution could present an alternative account of the
existence of complex human behaviours. As noted in
§1, Tomasello (1999) has argued that cumulative
cultural evolution could explain why human behaviour
is so different from that of the other great apes. In effect,
the suggestion is that these behaviours are the result
of the accrued learning of many generations, rather
than the result of natural selection. The finding that
cultural transmission degrades information in the
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Review. Cumulative culture in the laboratory C. A. Caldwell & A. E. Millen 3537

(a) 5 Using our paper aeroplane task detailed previously, we

mean similarity
have run a variety of different experimental conditions, in
4 which certain sources of social information are either
available or unavailable. While in our previous experi-
rating

3 ment (Caldwell & Millen 2008) participants could

observe the two previous participants in the chain, and
2 discuss the task with them, as well as see the results of
their efforts in the form of the completed plane and
1 information about its flight distance, in this study we
separated these sources of information. In a series of
(b) 5 different experimental conditions, participants had
access to information in the form of either: actions only;
mean similarity

4 results only; teaching only; actions plus results; actions

plus teaching; results plus teaching; or actions, results
rating

3 and teaching. If imitation (in terms of ‘learning to do an

2 crucial to cumulative cultural evolution, then conditions
1 from actions) ought to show much weaker trends towards
1 2 3 4 5 6 7 8
position in chain
Figure 3. Ratings of similarity for pairs of artefacts in the same
position across chains. The data for (a) paper aeroplanes and
(b) spaghetti towers. Error bars indicate Gs.e.m. Redrawn
from Caldwell & Millen (2008).
general direction of learners’ innate biases therefore
does not add much to this particular issue.
By contrast, our results show that increasing cultural
similarity can go hand in hand with increasing adaptive
complexity. Although universality in complex human
behaviour is often attributed to specialized innate
predispositions, our results imply that similar
behaviours may well be independently discovered and
passed on within different populations. It is in fact
examples such as ours, which show cumulative cultural
evolution being shaped by feedback from multiple
attempts, which serve to emphasize just what is so
useful about cumulative cultural evolution (e.g.
Henrich & McElreath 2003). It is a means by which
information is gained, and then retained, within
populations. The extended learning period afforded
allows us to make discoveries that would not have been
possible within a single lifetime. While our experiments
are so simple, small scale and short term, as to
somewhat trivialize this point, our findings are none-
theless illustrative of phenomena that we believe to be
operating over much longer time scales, in many realms
of behaviour.
(b) Learning mechanisms
In our ongoing work, we are using the methods that we
have developed to test hypotheses about the learning
mechanisms that may be involved in cumulative
cultural evolution. It has been proposed that cumu-
lative cultural evolution may depend on a capacity for
imitation, and that this may be the reason for its
apparent rarity in non-humans. Consequently, we have
run experiments (C. A. Caldwell & A. E. Millen 2008,
unpublished data) designed to test whether restricting
opportunities for imitation, and indeed other forms of
social learning, influences the trends that we find
towards improvement over generations of learners.
act by seeing it done’, e.g. Whiten & Ham 1992) is indeed
in which this not possible (i.e. those with no information
9 10
improvement over generations.
6. CONCLUSION
In summing up, we believe that the experimental
methods that we have developed will prove to be
extremely useful tools in helping to understand the
phenomenon of cumulative cultural evolution, and that
it will be possible to make important contributions to
some of the debates that surround this field. We hope
that our current findings help to illustrate how scaled-
down laboratory tests can be used to investigate
fundamental cultural processes.
Ethical approval for this research was granted by the University
of Stirling Department of Psychology Ethics Committee.
Many thanks to Kenny Smith for editing this issue and inviting
our contribution. This work was funded by a grant from the
Economic and Social Research Council (RES-061-23-0072).
ENDNOTE
1
It should be noted that there are many different definitions of culture
in the literature. Kroeber & Kluckhohn (1952) identified multiple
different definitions from authors from a variety of disciplines. Boyd &
Richerson (1985) pointed out that, within Kroeber & Kluckhohn’s
definitions, there is broad agreement on the notion of culture as
socially transmitted heritage peculiar to a particular society, which is
why we selected this description. Our criteria for culture in the
general sense are therefore intentionally broad and inclusive.
REFERENCES
Ambrose, S. 2001 Paleolithic technology and human
evolution. Science 291, 1748–1753. (doi:10.1126/science.
1059487)
Barbrook, A. C., Howe, C. J., Blake, N. & Robinson, P. 1998
The phylogeny of The Canterbury Tales. Nature 394, 839.
(doi:10.1038/29667)
Bartlett, F. C. 1932 Remembering. Cambridge, UK:
Cambridge University Press.
Basalla, G. 1988 The evolution of technology. Cambridge, UK:
Cambridge University Press.
Baum, W. M., Richerson, P. J., Efferson, C. M. & Paciotti,
B. M. 2004 Cultural evolution in laboratory micro-
societies including traditions of rule giving and rule
following. Evol. Hum. Behav. 25, 305–326. (doi:10.1016/
j.evolhumbehav.2004.05.003)

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3538 C. A. Caldwell & A. E. Millen Review. Cumulative culture in the laboratory
Boesch, C. 2003 Is culture a golden barrier between human
and chimpanzee? Evol. Anthropol. 12, 82–91. (doi:10.
1002/evan.10106)
Boesch, C. & Tomasello, M. 1998 Chimpanzee and human
cultures. Curr. Anthropol. 39, 591–614. (doi:10.1086/
204785)
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Boyd, R. & Richerson, P. J. 1994 Why does culture increase
human adaptability? Ethol. Sociobiol. 16, 125–143. (doi:10.
1016/0162-3095(94)00073-G)
Boyd, R. & Richerson, P. J. 1996 Why culture is common, but
cultural evolution is rare. Proc. Br. Acad. 88, 77–93.
Bunch, B. & Hellemans, A. 2004 A history of science and
technology. London, UK: Houghton Mifflin.
Buss, D. M. 1989 Sex differences in human mate preferences:
evolutionary hypotheses tested in 37 cultures. Behav.
Brain Sci. 12, 1–49.
Caldwell, C. A. In press. Convergent cultural evolution may
explain linguistic universals (commentary on Christian-
sen & Chater, Language as shaped by the brain). Behav.
Brain Sci.
Caldwell, C. A. & Millen, A. E. 2008 Experimental models
for testing hypotheses about cumulative cultural
evolution. Evol. Hum. Behav. 29, 165–171. (doi:10.
1016/j.evolhumbehav.2007.12.001)
Christiansen, M. H. & Chater, N. In press. Language as
shaped by the brain. Behav. Brain Sci.
Efferson, C., Richerson, P. J., McElreath, R., Lubell, M.,
Edsten, E., Waring, T. M., Paciotti, B. & Baum, W. 2007
Learning, productivity, and noise: an experimental study
of cultural transmission on the Bolivian Altiplano. Evol.
Hum. Behav. 28, 11–17. (doi:10.1016/j.evolhumbehav.
2006.05.005)
Efferson, C., Lalive, R., Richerson, P. J., McElreath, R. &
Lubell, M. 2008 Conformists and mavericks: the
empirics of frequency-dependent cultural transmission.
Evol. Hum. Behav. 29, 56–64. (doi:10.1016/j.evolhum-
behav.2007.08.003)
Enquist, M. & Ghirlanda, S. 2007 Evolution of social
learning does not explain the origin of human cumulative
culture. J. Theor. Biol. 246, 129–135. (doi:10.1016/j.jtbi.
2006.12.022)
Fay, N., Garrod, S. & Roberts, L. 2008 The fitness and
functionality of culturally evolved communication
systems. Phil. Trans. R. Soc. B 363, 3553–3561. (doi:10.
1098/rstb.2008.0130)
Fitch, W. M. 2000 Homology: a personal review of some of
the problems. Trends Genet. 16, 227–231. (doi:10.1016/
S0168-9525(00)02005-9)
Flynn, E. 2008 Investigating children as cultural magnets: do
young children transmit redundant information along
diffusion chains? Phil. Trans. R. Soc. B 363, 3541–3551.
(doi:10.1098/rstb.2008.0136)
Galef Jr, B. G. 1992 The question of animal culture. Hum.
Nat. 3, 157–178. (doi:10.1007/BF02692251)
Galef Jr, B. G. & Allen, C. 1995 A new model system for
studying behavioural traditions in animals. Anim. Behav.
50, 705–717. (doi:10.1016/0003-3472(95)80131-6)
Galef Jr, B. G., Manzig, L. A. & Field, R. M. 1986 Imitation
learning in budgerigars: Dawson and Foss 1965 revisited.
Behav. Process. 13, 191–202. (doi:10.1016/0376-6357(86)
90025-2)
Gerard, R. W., Kluckhohn, C. & Rapoport, A. 1956
Biological and cultural evolution: some analogies and
explorations. Behav. Sci. 1, 6–34.
Gray, R. D. & Jordan, F. M. 2000 Language trees support the
express-train sequence of Austronesian expansion. Nature
405, 1052–1055. (doi:10.1038/35016575)
Phil. Trans. R. Soc. B (2008)
Griffiths, T. L., Kalish, M. L. & Lewandowsky, S. 2008
Theoretical and empirical evidence for the impact of
inductive biases on cultural evolution. Phil. Trans. R. Soc.
B 363, 3503–3514. (doi:10.1098/rstb.2008.0146)
Henrich, J. 2004 Demography and cultural evolution: how
adaptive cultural processes can produce maladaptive
losses: the Tasmanian case. Am. Antiquity 69, 197–214.
(doi:10.2307/4128416)
Henrich, J. & McElreath, R. 2003 The evolution of cultural
evolution. Evol. Anthropol. 12, 123–135. (doi:10.1002/
evan.10110)
Hermann, E., Call, J., Hernandez-Lloreda, M. V., Hare, B. &
Tomasello, M. 2007 Humans have evolved specialized skills
of social cognition: the cultural intelligence hypothesis.
Science 317, 1360–1366. (doi:10.1126/science.1146282)
Heyes, C. M. 1993 Imitation, culture and cognition. Anim.
Behav. 46, 999–1010. (doi:10.1006/anbe.1993.1281)
Humle, T. & Matsuzawa, T. 2002 Ant-dipping among the
chimpanzees of Bossou, Guinea, and some comparisons
with other sites. Am. J. Primatol. 58, 133–148. (doi:10.
1002/ajp.10055)
Hunt, G. R. & Gray, R. D. 2003 Diversification and
cumulative evolution in New Caledonian crow tool
manufacture. Proc. R. Soc. B 270, 867–874. (doi:10.1098/
rspb.2002.2302)
Insko, C. A. et al. 1980 Social evolution and the emergence of
leadership. J. Pers. Soc. Psychol. 39, 431–448. (doi:10.
1037/0022-3514.39.3.431)
Insko, C. A., Gilmore, R., Drenan, S., Lipsitz, A., Moehle, D.
& Thibaut, J. 1983 Trade versus exploration in
open groups: a comparison of two types of social power.
J. Pers. Soc. Psychol. 44, 977–999. (doi:10.1037/0022-
3514.44.5.977)
Jacobs, R. C. & Campbell, D. T. 1961 The perpetuation of an
arbitrary tradition through several generations of labora-
tory microculture. J. Abnorm. Soc. Psych. 62, 649–658.
(doi:10.1037/h0044182)
Kalish, M. L., Griffiths, T. L. & Lewandowsky, S. 2007
Iterated learning: intergenerational knowledge trans-
mission reveals inductive biases. Psych. Bull. Rev. 14,
288–294.
Kawai, M. 1965 Newly-acquired pre-cultural behavior of the
natural troop of Japanese monkeys on Koshima Islet.
Primates 6, 1–30. (doi:10.1007/BF01794457)
Kawamura, S. 1959 The process of sub-culture propagation
among Japanese macaques. Primates 2, 43–60. (doi:10.
1007/BF01666110)
Kenward, B., Weir, A. A. S., Rutz, C. & Kacelnick, A. 2005
Tool manufacture by naive juvenile crows. Nature 433,
121. (doi:10.1038/433121a)
Kirby, S. & Christiansen, M. H. 2003 From language
learning to language evolution. In Language evolution
(eds M. Christiansen & S. Kirby), pp. 272–294. Oxford,
UK: Oxford University Press.
Kirby, S., Dowman, M. & Griffiths, T. L. 2007 Innateness and
culture in the evolution of language. Proc. Natl Acad. Sci.
USA 104, 5241–5245. (doi:10.1073/pnas.0608222104)
Kroeber, A. L. & Kluckhohn, C. 1952 Culture, a critical
review of the concepts and definitions. Pap. Peabody
Museum Am. Archeol. Ethnol. 47, 1–233.
Laland, K. N. 2004 Social learning strategies. Learn. Behav.
32, 4–14.
Laland, K. N. & Hoppitt, W. 2003 Do animals have culture?
Evol. Anthropol. 12, 150–159. (doi:10.1002/evan.10111)
Laland, K. N. & Williams, K. 1997 Shoaling generates social
learning of foraging information in guppies. Anim. Behav.
53, 1161–1169. (doi:10.1006/anbe.1996.0318)
Laland, K. N. & Williams, K. 1998 Social transmission of
maladaptive information in the guppy. Behav. Ecol. 9,
493–499. (doi:10.1093/beheco/9.5.493)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Review. Cumulative culture in the laboratory
McElreath, R., Lubell, M., Richerson, P. J., Waring, T. M.,
Baum, W., Edsten, E., Efferson, C. & Paciotti, B. 2005
Applying evolutionary models to the laboratory study of
social learning. Evol. Hum. Behav. 26, 483–508. (doi:10.
1016/j.evolhumbehav.2005.04.003)
McElreath, R., Bell, A. V., Efferson, C., Lubell, M., Richerson,
P. J. & Waring, T. 2008 Beyond existence and aiming
outside the laboratory: estimating frequency-dependent
and pay-off-biased social learning strategies. Phil. Trans. R.
Soc. B 363, 3515–3528. (doi:10.1098/rstb.2008.0131)
Mesoudi, A. 2007 Using the methods of experimental social
psychology to study cultural evolution. J. Soc. Evol. Cultur.
Psychol. 1, 35–58.
Mesoudi, A. & O’Brien, M. J. 2008 The cultural transmission
of Great Basin projectile-point technology I: an experi-
mental simulation. Am. Antiquity 73, 3–28.
Mesoudi, A. & Whiten, A. 2004 The hierarchical transformation
of event knowledge in human cultural transmission. J. Cogn.
Cult. 4, 1–24. (doi:10.1163/156853704323074732)
Mesoudi, A. & Whiten, A. 2008 The multiple roles of cultural
transmission experiments in understanding human cul-
tural evolution. Phil. Trans. R. Soc. B 363, 3489–3501.
(doi:10.1098/rstb.2008.0129)
Mesoudi, A., Whiten, A. & Laland, K. N. 2004 Is human
cultural evolution Darwinian? Evidence reviewed from the
perspective of The Origin of Species. Evolution 58, 1–11.
(doi:10.1554/03-212)
Mesoudi, A., Whiten, A. & Dunbar, R. 2006 A bias for social
information in human cultural transmission. Br. J. Psychol.
97, 405–423. (doi:10.1348/000712605X85871)
Pinker, S. & Bloom, P. 1990 Natural language and natural
selection. Behav. Brain Sci. 13, 707–727.
Richerson, P. J. & Boyd, R. 2005 Not by genes alone: how
culture transformed human evolution. Chicago, IL:
University of Chicago Press.
Smith, K. & Kirby, S. 2008 Cultural evolution: implications
for understanding the human language faculty and its
evolution. Phil. Trans. R. Soc. B 363, 3591–3603. (doi:10.
1098/rstb.2008.0145)
Phil. Trans. R. Soc. B (2008)
C. A. Caldwell & A. E. Millen 3539
Smith, K., Brighton, H. & Kirby, S. 2003 Complex systems
in language evolution: the cultural emergence of compo-
sitional structure. Adv. Complex Syst. 6, 537–558. (doi:10.
1142/S0219525903001055)
Sugiyama, Y. 1997 Social tradition and the use of tool-
composites by wild chimpanzees. Evol. Anthropol. 6, 23–27.
(doi:10.1002/(SICI)1520-6505(1997)6:1!23::AID-EVAN
7O3.0.CO;2-X)
Tehrani, J. & Collard, M. 2002 Investigating cultural
evolution through biological phylogenetic analyses of
Turkmen textiles. J. Anthropol. Archaeol. 21, 443–463.
(doi:10.1016/S0278-4165(02)00002-8)
Tomasello, M. 1990 Cultural transmission in tool use and
communicatory signalling of chimpanzees. In “Language”
and intelligence in monkeys and apes: comparative developmental
perspectives (eds S. Parker & K. Gibson), pp. 274–311.
Cambridge, UK: Cambridge University Press.
Tomasello, M. 1999 The cultural origins of human cognition.
Cambridge, MA: Harvard University Press.
Tomasello, M., Kruger, A. C. & Ratner, H. H. 1993 Cultural
learning. Behav. Brain Sci. 16, 495–552.
Whiten, A. & Ham, R. 1992 On the nature and evolution of
imitation in the animal kingdom: reappraisal of a century
of research. Adv. Stud. Behav. 21, 239–283. (doi:10.1016/
S0065-3454(08)60146-1)
Whiten, A. & Mesoudi, A. 2008 Establishing an experimental
science of culture: animal social diffusion experiments.
Phil. Trans. R. Soc. B 363, 3477–3488. (doi:10.1098/rstb.
2008.0134)
Whiten, A., Goodall, J., Mcgrew, W. C., Nishida, T.,
Reynolds, V., Sugiyama, Y., Tutin, C., Wrangham, R. &
Boesch, C. 1999 Cultures in chimpanzees. Nature 399,
682–685. (doi:10.1038/21415)
Whiten, A., Horner, V. & Marshall-Pescini, S. 2003 Cultural
panthropology. Evol. Anthropol. 12, 92–105. (doi:10.1002/
evan.10107)
Wilder, R. L. 1968 Evolution of mathematical concepts: an
elementary study. London, UK: Wiley.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3541–3551

doi:10.1098/rstb.2008.0136
Published online 17 September 2008

Investigating children as cultural magnets:

do young children transmit redundant
information along diffusion chains?
Emma Flynn*
Department of Psychology, Durham University, Science Laboratories, Durham DH1 3LE, UK
The primary goal of this study was to investigate cultural transmission in young children, with specific
reference to the phenomenon of overimitation. Diffusion chains were used to compare the imitation
of 2- and 3-year-olds on a task in which the initial child in each chain performed a series of relevant
and irrelevant actions on a puzzle box in order to retrieve a reward. Children in the chains witnessed
the actions performed on one of two boxes, one which was transparent and so the lack of causality of
the irrelevant actions was obvious, while the other was opaque and so the lack of causal relevance was
not obvious. Unlike previous dyadic research in which children overimitate a model, the irrelevant
actions were parsed out early in the diffusion chains. Even though children parsed out irrelevant
actions, they showed fidelity to the method used to perform a relevant action both within dyads and
across groups. This was true of 3-year-olds, and also 2-year-olds, therefore extending findings from
previous research.
Keywords: social learning; culture; children; overimitation; imitation; emulation

1. INTRODUCTION In the present study, definitions for two critical social

Understanding how we learn from others has been
of interest to psychologists for over a century (Baldwin
1902; Rogers & Williams 2006), and during this time it
has been investigated by a range of disciplines including
anthropology, cognitive neuroscience, robotics and
philosophy, as well as developmental, social and
comparative psychology ( Dautenhahn & Nehaniv
2002; Meltzoff & Prinz 2002; Want & Harris 2002;
Frith & Wolpert 2003; Bekkering et al. 2005; Breazeal
et al. 2005; Hayashi et al. 2005; Hurley & Chater 2005;
Kubota 2005; Lukowski et al. 2005). Research into
children’s social learning is undergoing a major
expansion, stimulated in part by the integration of
developmental and comparative perspectives ( Want &
Harris 2002; Call et al. 2005; Carpenter et al. 2005;
Tomasello et al. 2005; Horner et al. 2006; Tennie et al.
2006; McGuigan et al. 2007), which has allowed the
distinction of social learning processes from the simple,
such as local or stimulus enhancement, to the complex,
including imitation and goal emulation. Such an
examination has important implications for our under-
standing of cultural acquisition as children are cultural
magnets, with some researchers arguing that processes
such as imitation are the bedrock of the acquisition
of culture (Boyd & Richerson 1985; Tomasello 1999;
Plotkin 2003; Richerson & Boyd 2005), although others
highlight the role of trial-and-error learning in
the transmission and development of cultural forms
(Sterelny 2006).
With this expansion, there has been debate regard-
ing the definition of key terms within social learning.
*e.g.flynn@durham.ac.uk
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
learning mechanisms, emulation and imitation, are
taken from McGuigan et al. (2007), which stated ‘One
such process is emulation, where the observer attempts
to reproduce the results of a model’s actions, rather
than the more complete copy of the model’s behaviour
that distinguishes imitative learning’ (p. 353). The
present study investigated whether groups of 2- and/or
3-year-olds transmitted newly acquired behaviour
through emulation or imitation, or a combination of
the two. In order to do this, the phenomenon of
overimitation, the adoption of inefficient strategies
that an individual has seen a model use, was used,
therefore establishing whether a series of irrelevant
actions would be transmitted from child to child along
a chain of children. The phenomena of overimitation
and diffusion chain designs are explained in full in the
following sections.
(a) Overimitation
Young children have been shown to have a number of
social learning mechanisms at their disposal. For
example, 14- to 18-month-olds are likely to imitate
the method used to achieve a goal when they appear to
be intentional, but not when they are accidental
(Carpenter et al. 1998) and 12- to 18-month-olds use
context to decide whether to copy means (imitation) or
outcome (goal emulation; Gergely et al. 2002;
Carpenter et al. 2005). Twenty-month-old children
will emulate by reordering the sequence of a series of
actions that they witness so that enabling actions are
put together, even when the demonstration presented a
sequence in which enabling actions were interspersed
with non-enabling actions; thus, they reach the same
end state but do so through a different sequence
of actions (Bauer 1992). Nielsen (2006) found that by

3541 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3542 E. Flynn Investigating children as cultural magnets
(a)
Figure 1. The two glass ceiling boxes: (a) the transparent box with the tool being tapped in the upper compartment and (b) the
opaque box with the door in the lift position and the tool inserted into the opaque tube that contains the reward.
2 years of age children imitated by using a tool that
had been used by a model to open a box, even though
they often found this ineffective and would have been
more successful at the task had they emulated and
used their hand as younger children had done. Even
though very young children have a diverse set of social
learning mechanisms available to them, by 3 years
children begin to persistently imitate adults’ actions,
even when these actions are not the most task efficient
method, leading to a phenomenon that has been
labelled ‘overimitation’ (McGuigan et al. 2007). It is
unclear at which point overimitation ceases, if indeed it
does. Research with adults has shown that they are
‘optimum imitators’, using even the same digit to poke
at a bolt defence, but they are also emulators as they
discover and adopt more efficient variants of these
behaviours over trials (Custance et al. 2006).
Making a distinction between the two social learning
mechanisms, imitation and emulation, is critical to the
investigation of the phenomenon of overimitation.
In the present study, two different perspectives were
used to draw such a distinction. First, a participant
witnessed a model retrieve a reward from a box after
completing a series of seven actions, critically only two
of these actions were causally necessary to retrieve the
reward, while the other five actions in the sequence
were not relevant. Imitation could be said to occur if a
child copied the relevant and irrelevant actions, thus
producing overimitation as inefficient causally irrele-
vant actions were reproduced. Emulation of a goal
occurred if a child only reproduced the two relevant
actions. Second, within the series of actions, certain
behaviours were able to be undertaken using one of
two methods, e.g. lifting or sliding a door (Dawson &
Foss 1965; Whiten et al. 2006; Flynn & Whiten 2008,
in press). The two-method design allows a distinction
to be drawn between imitation and emulation. If the
individuals in an experimental group witness a model
use method A to complete an action (e.g. lifting a door)
and subsequently adopt method A in their own
attempts at the task, while the individuals in a second
experimental group witness a model use method B
(sliding the door) and subsequently adopt method B,
imitation can be said to have occurred. However, if
there is no distinction between the two experimental
groups, those who witnessed method A and those who
witnessed method B, in terms of the method they
adopted during their attempt then children are not
Phil. Trans. R. Soc. B (2008)
(b)
systematically imitating the method they have wit-
nessed used to reach the same end state, instead they
are emulating the goal.
The task used in the present study was the glass-
ceiling box (henceforth referred to as the GCB), which
consists of two boxes that are identical except that one
is opaque and the other transparent (figure 1). Both
boxes contained an opaque tube, and for the experi-
ment each tube was baited with a reward, a Velcro-
backed sticker. To retrieve this reward a door situated
at the front of the box had to be opened, either by lifting
or sliding left or right, and a rod tool with a Velcro end
inserted into the opaque tube. During the demon-
stration, children witnessed a series of actions that
included five irrelevant actions directed towards an
opening at the top of the box to a chamber that is
hollow and does not make any connection with the
reward or the opaque tube that contains the reward. It
should be clear to children who witness the series of
causally relevant and causally irrelevant actions on the
transparent box, as it is to chimpanzees (Horner &
Whiten 2005), that only those actions directed at the
front opening are necessary, while children who witness
the series of actions performed on the opaque box are
not privileged to such a distinction.
Horner & Whiten (2005) and McGuigan et al.
(2007) found that 3- and 5-year-olds copied all actions
a model performed on the GCB, irrespective of their
opportunity to see the causal relevance of the actions,
i.e. whether they were presented with the opaque or
transparent box. Such a finding is surprising given that
younger children are able to parse out or reorder
actions that are irrelevant to a goal (Bauer 1992).
Children were also found to be faithful to the method
used to undertake the actions, showing that not only do
3- and 5-year-olds imitate by copying causally irrele-
vant as well as causally relevant actions, but they also
imitate by copying the specific method used by a model.
McGuigan et al. (2007) found that under certain
conditions, such as watching a preconstructed video
demonstration of a model’s hands completing the
task in contrast to a live demonstration, the tendency
to overimitate decreases for 3-year-olds as they
were more likely to parse out irrelevant actions, but
5-year-olds were just as likely to overimitate after
having watched the video. Lyons et al. (2007) found
that 3- to 5-year-olds imitate irrelevant actions under
conditions that should reduce such a tendency,
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Investigating children as cultural magnets E. Flynn
including when trained to identify irrelevant actions
performed by an experimenter, or when children
believe the experiment is over and they are under a
time constraint to prepare for the next participant, and
also when given direct instructions to ignore any
unnecessary actions. It was only under conditions in
which the demonstrator’s irrelevant actions actually
broke the contact principle, i.e. the rule that mechan-
ical interactions cannot occur at a distance, that
children overcame their tendency to overimitate. It
has been suggested that overimitation occurs because
children attempt to share experience with a demon-
strator (Užgiris 1981; Carpenter 2006; Nielsen 2006,
in press) or learn about initially opaque aspects of
causality (Lyons et al. 2007), as well as when they
assume a demonstrator is trying to teach them some-
thing (Gergely & Csibra 2005, 2006).
To summarize, children younger than 3 years appear
to be able to implement different observational learning
mechanisms depending on the scenario, yet by their third
birthday children tend to overimitate. At 3 years children
imitate all aspects of a model’s demonstration, even
actions that are causally irrelevant and therefore less task
efficient. Research examining overimitation has concen-
trated on investigating overimitation during dyadic
interactions where an adult model has demonstrated to
a participating child. The present study aimed to extend
our understanding of the phenomenon of overimitation
by examining whether overimitation is transmitted
across groups of children; such an investigation is
essential to our understanding of the transmission of
traditions. The primary question addressed in the
present study was whether 2- and/or 3-year-old children
transmit traditions that contain redundant elements,
therefore transmitting a tradition that is not task efficient.
(b) Transmission of information across groups
In order to discover more about social learning in the
real world it is important to examine the transmission
of information or behaviour from a variety of models,
and across a variety of settings. For example, children
often learn by observing other children, who may be
viewed as less rational, knowledgeable and have less
authority than adult-experimenter models. Although
the majority of social learning studies have used adult-
experimenter models, some studies have used children
as models and have shown a high level of fidelity to the
demonstration witnessed (Horner et al. 2006; Flynn &
Whiten 2008). Another critical aspect of observational
learning studies is that the majority use dyadic
transmission in which an experimenter performs a
demonstration for a child who subsequently attempts
the task, but the transmission ends there. Social
learning in the real world and culture, which is so
closely related to imitation and observational learning
of others (Boyd & Richerson 1985; Tomasello 1999;
Plotkin 2003; Richerson & Boyd 2005), is bigger than
simple dyadic relations and involves the transmission of
information across generations, from one individual to
another. Therefore, it is essential that social learning
experiments mimic the transmission of information
and behaviour across groups. The present study
adopted a diffusion chain design, which offers a
controlled, micro-level representation of culture going
Phil. Trans. R. Soc. B (2008)
3543
beyond the usual dyadic designs of observational
learning studies.
A number of studies have examined such trans-
mission in non-human animals and adults using
diffusion experiments, which were introduced by
Bartlett (1932; human adults: Bangerter 2000, Kirby
et al. 2008, see Mesoudi & Whiten (2008) for
further examples; chimpanzees: Menzel et al. 1972,
Whiten et al. 2005; rats: Laland & Plotkin 1990;
guppies: Reader & Laland 2000; blackbirds: Curio
et al. 1978; pigeons: Lefebvre 1986; see Whiten &
Mesoudi (2008) for further examples). Initially
in diffusion studies a model is trained to perform
a behaviour, such as how to open a puzzle box in a
particular way. In diffusion chains, the transmission
of this behaviour is then investigated along ‘chains’ of
individuals through repeated dyadic interactions, such
that the model (individual A) is observed by individual
B while completing the action, individual B is
subsequently observed completing the task by individ-
ual C, who in turn is later observed completing the task
by individual D, and so on. Chains continue until
the number of participants is exhausted or until the
information fails to be transmitted. This allows the
transmission of information across ‘cultural gener-
ations’ to be examined, thus referring to consecutive
social transmissions from individual to individual,
which in real life may coincide with genetic generations
(parent to child) or not (e.g. peer to peer).
It is only recently that such methods have been used
to examine the transmission of information across
groups of children (Horner et al. 2006; Flynn & Whiten
2008). Horner et al. (2006) showed that 3-year-olds
showed high fidelity in the extraction of a reward from a
puzzle box by using one of two methods, either lifting
or sliding a door that concealed the reward. The
method seeded in the original child in a chain was
transmitted successfully along the chain, so that all the
children including the eighth and final child used
the same method to open the door of the puzzle box.
Flynn & Whiten (2008) used a more complex task in
which children had to use one of two tools to extract an
object from a puzzle box, either using a pronged tool to
stab the object and extract it through a hole in the roof
of the box, or a sliding tool that slid along the floor of
the box and allowed the objects to be guided to a
protruding chute that was bottomless and from which
the objects could fall. They found that children in the
diffusion chains conformed to the technique they
witnessed, with 5-year-olds displaying more robust
transmission than 3-year-olds.
The present study used diffusion chains to investi-
gate the cumulative effect of transmission of behaviour
across chains of 2- and 3-year-olds. A central question
in the study was whether irrelevant information is
transmitted, i.e. whether the overimitation seen in
dyadic studies would be transmitted across groups, and
if not, how and at which point it is parsed out. If the
irrelevant actions were faithfully transmitted along
the length of the chain this would provide evidence
that 2- and 3-year-olds transmit traditions that contain
irrelevant actions. Alternatively, the irrelevant actions
could be parsed out immediately with children in the
chain only transmitting actions that were relevant to
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3544 E. Flynn Investigating children as cultural magnets
the goal, thus showing that young children will remove
redundant items so that traditions are task efficient. If
the irrelevant actions were parsed out of the trans-
mitted information, a further point of the study was to
discover more about the process of transmission of
behaviour. Did parsing occurred suddenly, with all
elements removed at the same time or, was it gradual,
with individual elements being discarded at each
generation until they were all removed?
In order to examine the transmission of traditions
from a second perspective, i.e. the imitation and
transmission of specific actions across groups, this
study used the powerful two method, three group
design. Thus, the first child in each chain was trained to
use one of two methods to perform an action, e.g.
lifting rather than sliding a door. This produced two
experimental groups, chains seeded with method A and
chains seeded with method B. In the present study, two
sets of actions could be undertaken using different
methods, one of these actions was irrelevant (the bolts
at the top of the GCB could be either dragged from the
left using the tool or poked with the tool from the right)
and one was a relevant action (the door could be
opened by sliding left or right and by lifting). If the
specific methods seeded at the beginning of each chain
was transmitted faithfully along each chain, then this
would provide evidence of the transmission of
traditions across groups. The design also included an
important third group, referred to as a no-model
control condition, in which children were presented
with the task but receive no demonstration. Such a
control condition permits the level of success through
individual learning to be established, allowing an
analysis to be undertaken to establish whether obser-
vational learning has occurred in the diffusion chains.
Furthermore, the no-model control condition allows
an investigation of the predisposition to produce the
actions of interest. For example, if all children in
the no-model control condition remove the bolts in the
GCB and tap the tool into the upper compartment,
then it is not possible to establish whether performance
of such actions by children in the diffusion chains is due
to social or asocial learning.
(c) Predictions
The critical questions in the present study were
whether children transmitted traditions that contained
irrelevant actions and whether children faithfully
transmitted the method that a model used to complete
an action. The present study further examined whether
each form of transmission (irrelevant actions and
specific methods) was affected by age (2- versus
3-year-olds) and access to causal information (opaque
versus transparent). In line with dyadic studies of
overimitation, it was predicted that children in the
diffusion chains would reproduce and therefore
transmit irrelevant actions, thus transmitting traditions
that were not task efficient.
Overimitation increases from 3 years, and so it was
predicted that 3-year-old children would show a
significantly higher level of imitation of causally
irrelevant actions than 2-year-old children in chains
where the box was transparent. Yet, in chains where the
opaque box is presented it was predicted that 2- and
Phil. Trans. R. Soc. B (2008)
3-year-olds would show a similar level of overimitation,
reproducing the irrelevant actions, as young children
will overimitate when the goal is not clear ( Williamson &
Markman 2006).
A comparison of chains of 2- versus 3-year-olds
offered an interesting counterpoint to previous research
because until now diffusion chain studies have recruited
chains of 3- and 5-year-olds, which have shown good
fidelity in their transmission of both the methods used
and the specific action used to achieve these methods.
Including chains of 2-year-olds in the present study
addressed whether younger children are capable of
faithfully transmitting behaviour over generations in
relation to the transmission of irrelevant actions and the
specific actions used to achieve a transmitted method.
2. MATERIAL AND METHODS
(a) Participants
Eighty children participated, divided equally between 2- and
3-year-olds. Thirty-two children, half of which were 2-year-
olds, were allocated to a no-model control condition and
48 children, half of which were 2-year-olds, were allocated
to diffusion chains. Each chain contained six children of the
same age group. The mean age of the 2-year-old children in
each chain ranged from 2 years 6 months to 2 years 8 months
(standard deviation (s.d.) ranged from 3 to 4 months), and
the mean age of the 3-year-old children in each chain ranged
from 3 years 6 months to 3 years 9 months (s.d. ranged from
3 to 6 months). The mean age of the 2-year-old children in
the no-model control condition was 2 years 6 months (s.d.Z2
months) and the mean age of the 3-year-old children in
the no-model control condition was 3 years 6 months
(s.d.Z4 months).
(b) Design
The study used a between-group, diffusion chain design to
compare observational and individual learning in relation
to age (2- versus 3-year-olds) and access to causal information
(opaque versus transparent box). Children were allocated to a
no-model control condition or a diffusion chain, each chain
containing six children in total. Two chains were run for
each of the four conditions defined by these two factors (as in
Flynn & Whiten 2008), yielding eight chains in all.
(c) Materials
The GCB consists of two boxes that are identical except for
the fact that one box is transparent and the other is opaque
(figure 1). Each box has a hole on the roof, covered by a bolt
defence, and a second hole on the front face of the box
covered by a door defence. Behind the front hole is a sloping
tube, opaque in both boxes, which contains a reward
(a Velcro-backed sticker). In order to retrieve the reward
the door must be opened (either by sliding or lifting), a tool
(a 22 cm long rod with Velcro on the end) inserted and then
the reward can be pulled out. Actions directed to the front of
the box are causally necessary to retrieve the reward, whereas
actions directed to the top of the box are not, because
inserting the tool in the top hole results in hitting a barrier
(the ‘glass ceiling’) that prevents physical access between the
tool and the tube containing the reward.
(d) Two-action design
The extent of the participants’ imitation of the demonstrated
actions on the bolt and door defences was examined using the
‘two-action’ design (Horner et al. 2006; Whiten et al. 2006;
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Investigating children as cultural magnets E. Flynn
Flynn & Whiten 2008, in press). The door, which was hinged
at the top, could either be lifted or slid to the side to reveal the
opening to the tube. Similarly, the bolts could be dragged
from the left with the tool, or pushed from the right with the
tool in order to reveal the top hole. A model was trained to use
one of these two or three actions, (i) lifting the front door or
sliding it to the left or right and (ii) dragging or pushing the
bolts on the top of the box.
(e) Procedure
Each of the eight chains contained six children of the same
age group (2- versus 3-year-olds), who saw a model retrieve a
reward from either the opaque or transparent box. Testing
took place in a quiet room away from the other children in the
nursery. For children in the diffusion chains, initially the
experimenter said to the first child, ‘Okay watch me and then
you can have a go’. Then the child watched the experimenter
perform a series of actions either on the opaque or
transparent box. The child witnessed the experimenter either
push or drag the bolts from the top hole, the tool was then
inserted into the top hole and tapped on to the glass ceiling
below three times, after which the experimenter either lifted
or slid the door away from the hole at the front of the box,
inserted the tool and then removed the reward. Having
witnessed two demonstrations, the child was allowed to have
a turn, ‘Now it is your turn’; the goal of retrieving the Velcro-
backed sticker was never explicitly stated. The first child in
every chain was trained to retrieve the sticker using feedback
until she/he had incorporated all of the elements demon-
strated by the experimenter, so that the first child’s attempt
was a replication of the experimenter’s demonstration. Once
the model was proficient, the second child in the chain was
brought into the room, and told to wait while the first child
had two attempts, then it would be his/her turn. No explicit
instructions were given about watching, teaching or copying
and the tool was never handed to a child but placed on the
table in front of the GCB. The experimenter made sure that
each child had a clear view of the GCB and the actions upon
it. Children were retained as a model for the following child in
the chain as long as they attempted to remove the reward,
irrespective of the method used during their attempt.
Children were only discounted if they performed no mean-
ingful actions on the box. After the first child’s two
demonstrations, the second child, who had been present
during the demonstrations, had two solo attempts before
becoming a demonstrator for the next child in the chain. This
procedure continued to the final child, who had only two
attempts, as there was no need for him/her to demonstrate.
In the no-model control condition, children were brought
into the room and presented with the GCB and tool, being
told, ‘Lots of boys and girls have had a go, and now it is your
turn’. Testing ended if a child successfully retrieved the
sticker, refused to continue after general encouragement, or
after 4 min of interaction with the GCB. Children who
struggled in the no-model control condition were given
general encouragement, including, ‘What do you think you
do now?’, ‘You can touch it as much as you like, you can’t
break it’ and ‘You’re doing really well, what do you think you
do next?’.
Three children in the diffusion chains refused to
participate, two of these were at the end of the chains, and
one was in the second position along the chain. For the child
in the second position, this child was not included further in
the chain as he undertook no meaningful behaviour on the
box. Instead, the original model was asked to return and acted
as a model for the following child in the chain. All children,
Phil. Trans. R. Soc. B (2008)
3545
irrespective of success, received a sticker as a reward at the
end of the testing session.
(f ) Coding and inter-rater reliability
Each child’s performance was scored on four separate
variables: (i) whether she/he removed the bolts, and if so, the
method used, (ii) whether she/he tapped in the top of the box,
and if so, how many times, (iii) whether she/he opened the
door, and if so, which method she/he used, and (iv) whether
she/he inserted the rod to remove the sticker and was therefore
successful. From this coding, a score could be given for the
number of irrelevant actions undertaken from (i) the number
of bolts removed and (ii) the number of taps in the top of the
box (the original model in each chain performed five irrelevant
actions: removing both bolts and tapping three times into the
upper compartment). An independent observer, who was
blind to the rationale of the study, coded 18 per cent of the
sample (14 children made up of two chains and two control
children resulting in 50 incidents of behaviour). All Cohen’s
kappa scores (remove bolts by pulling, dragging or poking;
number of taps in upper compartment; open door by sliding
left or right or lifting; removing reward) were 0.91 or above,
showing a good level of reliability.
3. RESULTS
The analyses followed a series of questions, which were
considered in turn. First, did social learning occur and
how did children in the no-model control condition
behave? Second, did children copy the irrelevant
actions that were originally seeded in chains? Third,
did children copy the specific method they witnessed
used to perform actions? Finally, were behaviours
transmitted along chains from the original model,
therefore producing traditions?
(a) Did social learning occur and how did
children in the no-model control condition
behave?
When the level of success of the diffusion chain
children’s first attempt1 was compared to the level
of success of children in the no-model control condi-
tion it was clear that social learning had occurred.
Children in the diffusion chains were significantly
more successful at retrieving the reward (success
rateZ94%) than children in the no-model control
condition (success rateZ9%; c 21 (nZ80)Z56.93,
p!0.001). Of the 32 children in the no-model control
condition, 27 touched the GCB and/or tool suggesting
that the lack of success within this condition was not
due to a lack of interaction with the task.
It was also important to note when children in the
no-model control condition produced behaviours that
were of interest in the diffusion chains, as this provides
a baseline for their occurrence during individual
learning. For example, no child poked the bolts with
the tool, but nine children dragged the bolts, two using
the tool and the remaining seven using their hands.
Of the 32 no-model control children, 23 opened the
door of the GCB at least once. Six children lifted
the door open, 22 children opened the door by sliding
it to the right, and 20 opened the door by sliding it to
the left. Finally, none of the no-model control
children tapped the tool into the upper compartment
of the GCB.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3546 E. Flynn Investigating children as cultural magnets
(b) Did children copy the irrelevant actions that
were originally seeded in chains?
This analysis was concerned with whether children in
the diffusion chains produced any of the five irrelevant
actions (removing two bolts and tapping three times
into the upper compartment) performed by the first
child in each chain. This coding was not concerned
with the manner in which these behaviours were
performed, e.g. dragging or pushing the bolts, but in
whether the behaviour was actually undertaken. An
‘irrelevant action’ score was given for each child’s
attempt by adding the number of actions performed
out of the original five irrelevant actions. From this a
mean irrelevant score, ranging from 0 to 5, was
awarded across each child’s attempts.
A repeated measures analysis of variance was
undertaken on the children’s mean irrelevant action
scores according to the child’s position in the chain, age
and box type. As assumptions of sphericity were not
met, Huynh–Feldt corrections were used. It was found
that for the between-participant factors there was no
main effect for age (F1,2Z9.47, n.s.) or box type (F1,2Z
13.07, n.s.). For the repeated measures analysis there
was a significant effect for the number of irrelevant
actions produced depending on a child’s position in the
chain (F4.91,9.81Z37.87, p!0.001). Post hoc Bonferroni
tests showed that children in the first position
(meanZ5.00) made significantly more irrelevant actions
than children in the third, fourth, fifth and sixth
positions (meanZ0.60 (third position), 0.70 (fourth
position), 0 (fifth and sixth positions)). Children in the
first position did not differ significantly in the number of
irrelevant actions made from children in the second
position (meanZ2.90), and children in the second
position did not differ to children at any other position.
There was also an interaction between position and box
type (F4.91,9.81Z5.12, p!0.05). Children who were first
in the chains with either the opaque or transparent GCB
(mean for both was 5) made significantly more irrelevant
actions than children in all the other positions.
Figure 2 presents an illustration of the children’s
behaviour in each of the chains. It shows that the
causally irrelevant action of tapping the tool into the
top hole was never transmitted beyond the second
generation in a chain, and even then in only three of
the eight chains did the second person in each chain
tap the tool into the upper compartment. The action
of removing the bolts was also irrelevant. However,
removing the bolts was more resistant to being
discarded than the tool tapping, with bolt removal
being transmitted in four chains; two chains until
the second position and two chains until the fourth
position. A repeated measures analysis of variance
using Huynh–Feldt corrections found that there was
an effect for the number of bolt removals according
to the position that the child was in the chain
(F4.91,9.81Z5.66, p!0.05), but there was no effect for
age (F1,2Z3.79, n.s.) or box type (F1,2Z3.13, n.s.).
(c) Did children copy the specific method they
witnessed used to perform actions?
The transmission of the method used to perform
certain actions across each dyadic interaction was
investigated. Children’s actions on the bolts were not
Phil. Trans. R. Soc. B (2008)
included in this analysis because too few participants
actually performed actions on the bolts. Of the eight
children who did remove the bolts, six were faithful
on at least one of their attempts to the method they
had witnessed.
The method used to open the door could be
examined in detail, as all the children who remained
in the chains undertook this causally relevant action.
Children were coded in terms of whether the method
they used to open the door was the same as the method
they witnessed the previous child in the chain use. In
order to include the majority of children, each attempt
was analysed separately. At their first attempt signi-
ficantly more children (87%) imitated the method that
they had witnessed used to open the door than children
who used an alternative method (13%; c21(nZ37)Z
19.70, p!0.001). The five instances of lack of
imitation occurred when children slid the door in the
opposite direction to that which they had witnessed.
Similar results were produced for the second, third and
fourth attempt, with no less than 77 per cent of children
imitating the same method used to open the door to
that which they had witnessed rather than using a
different method (c2 ranged from 9.32 to 11.92 with
p!0.01). This high level of fidelity to the door method
witnessed did not differ according to age group, across
all of the attempts all c2 scores (ranging from 0.06 to
0.99) which contrasted 2- with 3-year-olds were not
significant, or box type, across all of the attempts all c2
scores (ranging from 0.01 to 0.50) which contrasted
the opaque and transparent GCB were not significant.
(d) Were behaviours transmitted along chains,
therefore producing traditions?
The previous analysis has shown that children were not
transmitting behaviour traditions that contained irre-
levant behaviour, as these were parsed out early in the
chains. However, it was possible to investigate whether
traditions were produced in relation to a relevant
behaviour, the manner in which the door was opened.
In order to establish whether traditions were trans-
mitted along the chains, in relation to the method used
to open the door, the behaviour that children produced
during their demonstrations was coded creating four
possible combinations: left–left, right–right, left–right
and right–left. These behaviours were compared to the
behaviours that the children had witnessed their model
produce, and a rating of ‘same as model’ or ‘different to
model’ was created. The number of changes (i.e. the
number of different to model scores) in the door-
opening behaviour produced along each of the chains
could then be recorded. In order to make comparisons
across all of the chains, only the first to fifth children
were included in the analysis, as not all chains
contained six children. If the method used to open
the door was faithfully reproduced along all generations
in a chain (e.g. left–left to left–left to left–left to left–left
to left–left or right–left to right–left to right–left to
right–left to right–left), then the chain would achieve a
‘change score’ of 0; however, if every child in a chain
produced a different door-opening behaviour to that
which they had witnessed (e.g. left–left to right–right to
left–left to right–right to left–left) then the chain would
be awarded a change score of 4. The fidelity of the
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Investigating children as cultural magnets E. Flynn 3547

diffusion chains

2-year-olds 3-year-olds

opaque (db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR)

(db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR)
(db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR)
(db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR)
(pb)(pb)(t)(t)(t)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (pb)(SDR)(RR)
(pb)(pb)(t)(t)(t)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (pb)(pb)(SDL)(RR)
(pb)(pb)(t)(t)(t)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (pb)(SDL)(RR)
(pb)(pb)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (pb)(SDL)(RR)
(pb)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (SDL)(RR)
(pb)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (SDL)(RR)
(pb)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (SDL)(RR)
(pb)(SDL)(RR) (SDL)(RR) (SDL)(tu)(RR) (SDL)(RR)
(pb)(pb)(SDR)(RR) (SDL)(RR) (SDR)(SDL)(tu)(RR) (SDL)(RR)
(tb)(SDR)(RR) (SDR)(RR) (SDR)(SDL)(tu)(RR) (SDR)(RR)
(hp)(hp)(SDR)(RR) (SDL)(RR) (SDR)(SDL)(tu)(RR) (SDR)(RR)
(hp)(hp)(SDR)(RR) (SDL)(RR) (SDL)(tu)(RR) (SDL)(RR)
(tb)(SDR)(RR) (SDL)(RR) (SDL)(SDR)(x20tu)(RR) (SDL)(RR)
(tb)(SDR)(RR) (SDL)(RR) (SDR)(SDL)(x11tu(RR) (SDL)(RR)
(SDR)(RR) (SDL)(SDR)(x4tu)(RR) (SDL)(RR)
(tb)(SDR)(RR) (SDL)(SDR)(SDL)(tu)((RR)) (SDL)(RR)
(SDR)(RR) (SDL)(SDR)(SDL)(tu)(RR) (SDL)(RR)
(SDR)(RR) (SDL)(SDR)(SDL)(tu)(RR) (SDL)(RR)

transparent (db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(LD)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR)

(db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(LD)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR)
(db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR)
(db)(db)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR)
(SDR)(RR) (SDL)(RR) (pb)(pb)(hp)(hp)(t)(t)(t)(SDR)(RR) (pb)(t)(t)(t)(SDL)(RR)
(SDL)(RR) (tb)(SDL)(RR) (hp)(hp)(t)(t)(t)(SDR)(RR) (pb)(t)(t)(t)(SDL)(RR)
(SDR)(RR) (SDL)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(t)(t)(t)(SDL)(RR)
(SDR)(RR) (SDL)(RR) (pb)(pb)&&(t)(t)(t)(SDR)(RR) (pb)(t)(t)(t)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDR)(RR) (SDR)(RR) (SDR)(RR) (pb)(pb)(SDL)(RR)
(SDL)(RR) (SDR)(RR) (SDR)(RR) (tb)(SDL)(tu)(RR)
(SDL)(RR) (SDR)(RR) (SDR)(RR) (SDL)(tu)(RR)
(SDL)(RR) (SDR)(RR) (SDR)(RR) (SDL)(RR)
(SDL)(RR) (SDR)(RR) (SDR)(RR) (SDL)(RR)
(SDR)(RR) (SDL)(RR)
(SDR)(RR) (SDL)(RR)

Figure 2. Transmission along diffusion chains. Each row of seven bracketed symbols represents a child’s attempt on a GCB. The
upper two rows of each set of four are attempts and the lower two rows are demonstrations: (db) represents dragging the bolts,
(pb) represents pushing the bolts, (hp) represents poking and pulling the bolts with one’s hand, (tb) refers to children who
touched the bolts but did not move them, (t) represents tapping into the upper compartment (the initial demonstrator did this
three times, only the first three taps are illustrated although some children made more), (LD) represents lifting the door open,
(SDL) or (SDR) represent sliding the door open either towards the left or right, (tu) refers to children who used the tool rather
than their hand to move the door, sometimes children moved the door more than once (e.g. !20 means moving the door 20
times) and (RR) represents retrieving the reward. A symbol that does not appear means that the behaviour it represents was not
produced during that attempt. A black rectangle to the side of a chain represents a child who was allocated to a chain but did not
participate in the chain. Actions represented with uppercase letters are causally necessary actions, while actions represented with
lowercase letters were not causally necessary.

transmission of children in the diffusion chains was compared to the expected distribution of scores
followed a binomial distribution from which the using a chi-squared statistic. The goodness-of-fit test
expected distribution of scores could be calculated. proved to be significant (c232 Z 53:63, p!0.001; Spiegel
The distribution of scores in the actual diffusion chains 1961). It was found that there were more cases with

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3548 E. Flynn Investigating children as cultural magnets
change scores of 0, 1 or 2 in the diffusion chains than
would be expected by chance, and less chains
producing change scores of 3 or 4 than expected by
chance. Observation of the data showed that 2- and
3-year-olds did not differ from one another in the
number of changes produced in the chains (mean for
2-year-oldsZ1.50; mean for 3-year-oldsZ1.50).
4. DISCUSSION
The primary goal of this study was to investigate the
cumulative effect of transmission of behaviour across
groups of 2- and 3-year-old children, addressing
questions such as whether irrelevant information is
transmitted, and if not, how and at which point it is
parsed out. Furthermore, the present study extended
previous diffusion chain studies by investigating
whether younger children to those previously tested,
specifically 2-year-olds, were capable of transmitting
the details of the method used to perform an action,
therefore producing a tradition. Before the main
findings are discussed, it must be acknowledged
that although diffusion chains offer an exciting
opportunity to investigate the transmission of traditions
and provide a micro-representation of culture, the size
of the samples used in diffusion chain studies is small.
Therefore, the strength of the conclusions made in
diffusion chain studies, including the present study,
needs to be considered in the light of this small sample
size. Specific caution must be used when interpreting
the findings in this study relating to age and box type,
as the small sample size may lead to type II errors.
(a) Overcoming overimitation
A core question of the present study addressed whether
there would be fidelity of transmission across cultural
generations when the demonstrated behaviour
contained irrelevant actions. Unlike previous studies
that have found strong and persistent fidelity to
traditions across generations of up to eight children
for all demonstrated behaviour, children in the present
study were very quick in parsing out the irrelevant
actions and transmitting only those actions that were
relevant to the goal. Indeed, the third child in the
chains, as well as all the subsequent children, made
significantly fewer irrelevant actions than the initial
model in each chain. Examination of the chains found
that for five of the eight chains irrelevant actions were
removed together, rather than seeing a gradual
removal. For the other three chains, removal of the
irrelevant actions was gradual, with the tapping being
removed from the sequence initially and later the
actions on the bolts, with the bolt actions sometimes
making it until the fourth child in the chain. Copying
only the actions on the bolts is interesting, as one would
assume that if participants did not tap into the top, the
action of removing the bolts would seem particularly
redundant. However, it is clear that access to causal
information and causal relations does not always assist
in overcoming overimitation, as children overimitate on
both opaque and transparent boxes (Horner & Whiten
2005). Lyons et al. (2007) found that certain types of
causal information, such as behaviour that breaks the
contact principle, facilitates children’s parsing of
Phil. Trans. R. Soc. B (2008)
irrelevant actions. Yet, in the present study the causal
irrelevance of removing the bolts when one does not
follow this by tapping into the upper compartment did
not seem to assist in children’s parsing of this irrelevant
act. The tendency to overimitate the removal of the
bolts may be because this was the first action within the
sequence, and was therefore subject to a primacy effect.
Future work could examine children’s imitation of
same-aged peers’ irrelevant actions on a task when such
actions are presented at different positions within a
sequence. Similarly, children’s lack of fidelity to the full
sequence of actions was not due to the memory
demands of the task, as some children were able to
replicate the full sequence of actions, and previous
work, which has used a similar sequence of actions that
were all relevant to the task, has shown that children are
capable of remembering and reproducing such
sequences (Flynn & Whiten 2008, in press).
Age and access to causal information did not affect
children’s ability to parse out the irrelevant actions. It
was predicted that as overimitation increases from 3
years, the diffusion chains containing 3-year-olds
would show a significantly higher level of imitation
of causally irrelevant actions than the chains of
2-year-olds when the GCB was transparent. Yet, in
chains where the opaque box is presented it was
predicted that 2- and 3-year-olds would show a similar
level of overimitation, reproducing the irrelevant
actions, as young children will overimitate when the
goal is not clear ( Williamson & Markman 2006). This
was not found to be the case; 3-year-olds were just as
likely to parse out the irrelevant information as 2-year-
olds, therefore showing a lack of overimitation. Also the
level of causal information available, whether the GCB
was opaque or transparent, did not affect children’s
reliance on the behaviour of the original model to
achieve the same goal. However, as stated at the
beginning of the discussion, caution must be used when
interpreting the findings in this study relating to age
and box type, as the small sample size may lead to a
type II error.
(b) Transmission across generations
Previously diffusion chain studies that have examined
the transmission of behaviour across groups of young
children have shown a strong and persistent replication
of behaviour from the first child in the chain to the last.
This is a process of canalization, where an individual’s
exploration of a task is reduced from potentially
limitless options to only a subset of behaviours that
she/he has seen performed by others (Horner et al.
2006; Flynn & Whiten 2008). The present study is the
first to include relevant and irrelevant actions within
the original model’s demonstration. It is clear from the
results that there was strong tendency to imitate the
relevant actions within the demonstration, producing
traditions similar to those produced in Horner et al.
(2006), as both studies examined the method of
opening a door. In the present study, the fidelity to
the relevant actions was strong across age groups and
box types within the dyadic interactions, i.e. children
copied the method they had witnessed. This was also
true across the whole chains, as both 2- and 3-year-olds
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Investigating children as cultural magnets E. Flynn
showed fewer changes in transmitted behaviour along
the chains than would be expected by chance.
Although based on a small sample, this study
provides an example of children’s cumulative cultural
evolution (CCE; see Caldwell & Millen (2008) for
further exploration of this concept). The term CCE is
used to describe the way that,
some individual or group of individuals first invented a
primitive version of the artefact or practice, and then
some later user or users made a modification, an
‘improvement,’ that others then adopted perhaps
without change for many generations, at which point
some other individual or group of individuals made
another modification, which was then learned and used
by others, and so on over historical time in what has
sometimes been dubbed ‘the ratchet effect’ ( Tomasello
et al. 1993)
( Tomasello 1999, p. 5).
Often CCE refers to the elaboration of cultural
techniques or artefacts, e.g. the development of
sophisticated technologies. However, in the present
study, traditions are improved by the removal of
irrelevant actions rather than the addition of behaviours,
and this creates a more efficient and streamlined
tradition. Thus, it appears that children as young as 2
years are capable of participating in CCE.
(c) Future directions
There was a significant reduction in the present study in
children’s overimitation in comparison to previous
studies. One possible cause for this difference was that
unlike previous studies that have shown overimitation in
young children, in the present study peers rather than
adult experimenters were used as models. Peers may be
viewed as less rational and knowledgeable and as having
less authority, which would explain why observing
children should be less likely to imitate the irrelevant
actions within the demonstration. It is clear that the lack
of transmission of irrelevant actions was not due to the
diffusion chain design, as the parsing of irrelevant
actions occurred early in the chains, at a point where
cumulative effects could not have built up. That is
during the point at which most parsing occurred (the
child 1 to child 2 transmission), the experiment is similar
to the usual dyadic design. Thus, a more likely
explanation is that the tendency to overcome over-
imitation is facilitated when the model is a same-age peer
rather than an adult experimenter. This finding needs
further support from a larger sample with a dyadic
design, where direct comparisons are made of children’s
imitation of adult- and peer models. Further work also
needs to examine why children are less likely to
overimitate from a peer. For example, is it due to a
difference in the perceived knowledge, authority or
rationality of the model? Alternative explanations may
be that there is less of a desire to be like one’s peers
compared to an adult, or that there is less motivation to
perpetuate the social interaction with one’s peer.
Finally, an interesting and unexpected effect within
the diffusion chains was the production and trans-
mission of idiosyncratic behaviour. For example, in one
of the chains children began to touch but not move the
bolts, a behaviour that was transmitted along
Phil. Trans. R. Soc. B (2008)
3549
generations. In another chain, children began to use
the tool to move the door rather than their hand, which
was consistently transmitted across generations, and
also the door was moved in both directions (in one case
20 times) before a child inserted the tool into the
opaque tube. Such transmissions of naturally produced
behaviour provide an interesting avenue for future
work regarding the production and transmission of
participant-produced behaviour, and most impor-
tantly, provide an opportunity to investigate the
identity of these innovators of traditions.
5. CONCLUDING REMARKS
The ability of human children to copy others outstrips
that of other animals (Tomasello 1990), and is so
significant that Meltzoff (1988) dubbed us, Homo
imitans. The present study extends our understanding
of children’s imitative abilities, by investigating
the transmission of traditions in the context of the
phenomenon of children’s overimitation, and has
resulted in a number of critical findings. The present
study is one of the first diffusion chain studies to
show that children’s transmission of behaviour across
groups does not necessarily involve a strong level of
fidelity. Previous diffusion chain studies with young
children have shown a strong replication of behaviour
from the beginning of the chain to the end. This study
shows that this is not an inbuilt phenomenon of diffusion
chains. Instead, children are able to parse out irrelevant
behaviours from a sequence of demonstrated actions.
This study, along with other diffusion studies, has
shown that the transmission of traditions by young
children can be examined within the laboratory
to explore interesting phenomena, such as the effects
of age and gender (Flynn & Whiten 2008), compari-
sons of different species (Horner et al. 2006) and the
effect of the relevance of behaviour to the goal (the
present study). Examining the transmission of
behaviour and information across groups appears to
be ripe for further exploration, as has been highlighted
by Flynn & Siegler (2007). The next step appears to be
an examination of young children’s cultural trans-
mission using open diffusion designs, in which a trained
model and a task are introduced to a group of
participants at the same time. Open diffusion studies
offer a more realistic micro-representation of culture, as
children choose when and who they observe, allowing
issues such as the role of children’s social status,
popularity and friendship patterns to be investigated.
I would like to thank all the children, parents and staff at the
nurseries who participated in this research. I would also like
to thank Kenny Smith and Stephan Lewandowsky for editing
this issue and inviting my contribution.
ENDNOTES
1
Comparing the level of success of the diffusion chain, children’s first
attempt meant that this result was not confounded with the possibility
of individual learning that may have occurred at later attempts.
2
The degree of freedom is 3 because in the analysis the expected
values were estimated from the binomial distribution and this has
been taken into account.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3550 E. Flynn Investigating children as cultural magnets
REFERENCES
Baldwin, J. M. 1902 Development and evolution. New York,
NY: MacMillan.
Bangerter, A. 2000 Transformation between scientific and
social representations of conception: the method of serial
production. Br. J. Social Psychol. 39, 521–535. (doi:10.
1348/014466600164615)
Bartlett, F. C. 1932 Remembering. Oxford, UK: Macmillan.
Bauer, P. 1992 Holding it all together: how enabling relations
facilitate young children’s event recall. Cognit. Dev. 7,
1–28. (doi:10.1016/0885-2014(92)90002-9)
Bekkering, H., Brass, M., Woschina, S. & Jacobs, A. M. 2005
Goal-directed imitation in patients with ideomotor
apraxia. Cognit. Neuropsychol. 22, 419–432. (doi:10.
1080/02643290442000275)
Boyd, R. & Richerson, P. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Breazeal, C., Buchsbaum, D., Gray, J., Gatenby, D. &
Blumberg, B. 2005 Learning from and about others:
towards using imitation to bootstrap the social under-
standing of others by robots. Artif. Life 11, 31–62. (doi:10.
1162/1064546053278955)
Caldwell, C. A. & Millen, A. E. 2008 Studying cumulative
cultural evolution in the laboratory. Phil. Trans. R. Soc. B
363, 3529–3539. (doi:10.1098/rstb.2008.0133)
Call, J., Carpenter, M. & Tomasello, M. 2005 Copying
results and copying actions in the process of social
learning: chimpanzees (Pan troglodytes) and human
children (Homo sapiens). Anim. Cogn. 8, 151–163.
(doi:10.1007/s10071-004-0237-8)
Carpenter, M. 2006 Instrumental, social, and shared
goals and intentions in imitation. In Imitation and the
social mind: autism and typical development (eds S. J. Rogers
& J. Williams), pp. 48–70. New York, NY: Guilford Press.
Carpenter, M., Akhtar, N. & Tomasello, M. 1998 Fourteen-
to 18-month-old infants differentially imitate intentional
and accidental actions. Infant Behav. Dev. 21, 315–330.
(doi:10.1016/S0163-6383(98)90009-1)
Carpenter, M., Call, J. & Tomasello, M. 2005 Twelve- and
18-month-olds copy actions in terms of goals. Dev. Sci. 8,
F13–F20. (doi:10.1111/j.1467-7687.2004.00385.x)
Curio, E., Ulrich, E. & Vieth, W. 1978 Cultural transmission
of enemy recognition: one function of avian mobbing.
Science 202, 899–901. (doi:10.1126/science.202.4370.
899)
Custance, D. M., Prato Previde, E., Spiezio, C., Rigamonti,
M. & Poli, M. 2006 Social learning in pig-tailed macaques
and adult humans on a two-action Perspex fruit. J. Comp.
Psychol. 120, 303–313. (doi:10.1037/0735-7036.120.3.
303)
Dautenhahn, K. & Nehaniv, C. L. (eds) 2002 Imitation in
animals and artifacts. Cambridge, MA: MIT Press.
Dawson, B. V. & Foss, B. M. 1965 Observational learning in
budgerigars. Anim. Behav. 13, 470–474. (doi:10.1016/
0003-3472(65)90108-9)
Flynn, E. & Siegler, R. 2007 Measuring change: current
trends and future directions in microgenetic research.
Infant Child Dev. 16, 135–149. (doi:10.1002/icd.502)
Flynn, E. & Whiten, A. 2008 Cultural transmission of tool-
use in young children: a diffusion chain study. Social Dev.
17, 699–718. (doi:10.1111/j.1467-9507.2007.00453.x)
Flynn, E. & Whiten, A. In press. Imitation of hierarchical
structure versus component details of complex actions by
3- and 5-year-olds, J. Exp. Child Psychol. (doi:10.1016/
j.jecp.2008.05.009)
Frith, C. D. & Wolpert, D. M. (eds) 2003 Neuroscience of
social interactions: decoding, influencing & imitating others.
Oxford, UK: Oxford University Press.
Phil. Trans. R. Soc. B (2008)
Gergely, G. & Csibra, G. 2005 The social construction of the
cultural mind: imitative learning as a mechanism of
human pedagogy. Interact. Stud. 6, 463–481. (doi:10.
1075/is.6.3.10ger)
Gergely, G. & Csibra, G. 2006 Sylvia’s recipe: the role of
imitation and pedagogy in the transmission of human
culture. In Roots of human sociality: culture, cognition,
and human interaction (eds N. J. Enfield & S. C. Levinson),
pp. 229–255. Oxford, UK: Berg Publishers.
Gergely, G., Bekkering, H. & Király, I. 2002 Developmental
psychology: rational imitation in pre-verbal infants. Nature
415, 755. (doi:10.1038/415755a)
Hayashi, M., Mizuno, Y. & Matsuzawa, T. 2005 How does
stone-tool use emerge? Introduction of stones and nuts to
naive chimpanzees in captivity. Primates 46, 91–102.
(doi:10.1007/s10329-004-0110-z)
Horner, V. & Whiten, A. 2005 Causal knowledge and
imitation/emulation switching in chimpanzees (Pan troglo-
dytes) and children (Homo sapiens). Anim. Cogn. 8,
164–181. (doi:10.1007/s10071-004-0239-6)
Horner, V., Whiten, A., Flynn, E. & de Waal, F. B. M. 2006
Faithful replication of foraging techniques along cultural
transmission chains by chimpanzees and children. Proc.
Natl Acad. Sci. USA 103, 13 878–13 883. (doi:10.1073/
pnas.0606015103)
Hurley, S. & Chater, N. 2005 Perspectives on imitation: from
mirror neurons to memes. Cambridge, MA: MIT Press.
Kirby, S., Cornish, H. & Smith, K. 2008 Cumulative cultural
evolution in the laboratory: an experimental approach to
the origins of structure in human language. Proc. Natl
Acad. Sci. USA 105, 10681–10686. (doi:10.1073/pnas.
0707835105)
Kubota, N. 2005 Computational intelligence for structured
learning of a partner robot based on imitation. Inform. Sci.
171, 403–429. (doi:10.1016/j.ins.2004.09.012)
Laland, K. N. & Plotkin, H. C. 1990 Social learning and
social transmission of foraging information in Norway rats
(Rattus noregicus). Anim. Learn. Behav. 18, 246–251.
Lefebvre, L. 1986 Cultural diffusion of a novel food-finding
behaviour in urban pigeons: and experimental field test.
Ethology 71, 295–304.
Lukowski, A. F., Wiebe, S. A., Haight, J. C., DeBoer, T.,
Nelson, C. A. & Bauer, P. J. 2005 Forming a stable
memory representation in the first year of life: why
imitation is more than child’s play. Dev. Sci. 8, 279–298.
(doi:10.1111/j.1467-7687.2005.00415.x)
Lyons, D., Young, A. & Keil, F. 2007 The hidden structure of
overimitation. Proc. Natl Acad. Sci. USA 104,
19 751–19 756. (doi:10.1073/pnas.0704452104)
McGuigan, N., Whiten, A., Flynn, E. & Horner, V. 2007
Imitation of causally-opaque versus causally-transparent
tool use by 3- and 5-year-old children. Cognit. Dev. 22,
353–364. (doi:10.1016/j.cogdev.2007.01.001)
Meltzoff, A. N. 1988 The human infant as Homo imitans. In
Social learning: psychological and biological perspectives (eds
T. Zentall & B. G. Galef ), pp. 319–341. Hillsdale, NJ:
Erlbaum.
Meltzoff, A. N. & Prinz, W. 2002 The imitative mind:
development, evolution, and brain bases. Cambridge, UK:
Cambridge University Press.
Menzel, E. W., Devenport, R. K. & Rogers, C. M. 1972
Proto-cultural aspects of chimpanzees’ responsiveness to
novel objects. Folia Primatol. 17, 161–170.
Mesoudi, A. & Whiten, A. 2008 The multiple roles of cultural
transmission experiments in understanding human cul-
tural evolution. Phil. Trans. R. Soc. B 363, 3489–3501.
(doi:10.1098/rstb.2008.0129)
Nielsen, M. 2006 Copying actions and copying outcomes:
social learning through the second year. Dev. Psychol. 42,
555–565. (doi:10.1037/0012-1649.42.3.555)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Investigating children as cultural magnets E. Flynn
Nielsen, M. In press. The imitative behaviour of children and
chimpanzees: a window on the transmission of cultural
traditions. Primatologie.
Plotkin, H. 2003 We-intentionality: an essential element in
understanding human culture. Perspect. Biol. Med. 46,
283–296. (doi:10.1353/pbm.2003.0028)
Reader, S. M. & Laland, K. N. 2000 Diffusion of foraging
innovations in the guppy. Anim. Behav. 60, 175–180.
(doi:10.1006/anbe.2000.1450)
Richerson, P. & Boyd, R. 2005 Not by genes alone: how culture
transformed human evolution. Chicago, IL: University of
Chicago Press.
Rogers, S. & Williams, J. (eds) 2006 Imitation and the social
mind. New York, NY: Guilford Publications.
Spiegel, M. R. 1961 Theory and problems of statistics. New
York, NY: Schaum Publishing Company.
Sterelny, K. 2006 The evolution and evolvability of culture.
Mind Lang. 21, 137–165.
Tennie, C., Call, J. & Tomasello, M. 2006 Push or pull:
imitation versus emulation in human children and great
apes. Ethology 112, 1159–1169. (doi:10.1111/j.1439-
0310.2006.01269.x)
Tomasello, M. 1990 Cultural transmission in the tool use and
communicatory signaling of chimpanzees? In Language and
intelligence in monkeys and apes: comparative developmental
perspectives (eds S. Parker & K. Gibson), pp. 274–311.
Cambridge, UK: Cambridge University Press.
Tomasello, M. 1999 The cultural origins of human cognition.
Cambridge, MA: Harvard University Press.
Phil. Trans. R. Soc. B (2008)
3551
Tomasello, M., Kruger, A. C. & Ratner, H. H. 1993 Cultural
learning. Behav. Brain Sci. 16, 495–552.
Tomasello, M., Carpenter, M., Call, J., Behne, T. & Moll, H.
2005 Understanding and sharing intentions: the origins of
cultural cognition. Behav. Brain Sci. 28, 675–691. (doi:10.
1017/S0140525X05000129)
Užgiris, I. 1981 Two functions of imitation during infancy.
Int. J. Behav. Dev. 4, 1–12. (doi:10.1016/S0163-6383
(81)80003-3)
Want, S. C. & Harris, P. L. 2002 How do children ape?
Applying concepts from the study of non-human primates
to the developmental study of ‘imitation’ in children. Dev.
Sci. 5, 1–13. (doi:10.1111/1467-7687.00194)
Whiten, A. & Mesoudi, A. 2008 Establishing an experimental
science of culture: animal social diffusion experiments.
Phil. Trans. R. Soc. B 363, 3477–3488. (doi:10.1098/rstb.
2008.0134)
Whiten, A., Horner, V. & de Waal, F. B. M. 2005 Conformity
to cultural norms of tool-use in chimpanzees. Nature 437,
737–740. (doi:10.1038/nature04047)
Whiten, A., Flynn, E., Brown, K. & Lee, K. 2006 Imitation
of hierarchical structure in actions by young children.
Dev. Sci. 9, 574–582. (doi:10.1111/j.1467-7687.2006.
00535.x)
Williamson, R. & Markman, E. 2006 Precision of imitation as
a function of preschoolers’ understanding of the goal of
the demonstration. Dev. Psychol. 42, 723–731. (doi:10.
1037/0012-1649.42.4.723)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3553–3561

doi:10.1098/rstb.2008.0130
Published online 17 September 2008

The fitness and functionality of culturally evolved

communication systems
Nicolas Fay1,*, Simon Garrod2 and Leo Roberts1
1
The University of Western Australia, Crawley, WA 6009, Australia
2
The University of Glasgow, Glasgow G12 8QQ, UK
This paper assesses whether human communication systems undergo the same progressive
adaptation seen in animal communication systems and concrete artefacts. Four experiments
compared the fitness of ad hoc sign systems created under different conditions when participants play
a graphical communication task. Experiment 1 demonstrated that when participants are organized
into interacting communities, a series of signs evolve that enhance individual learning and promote
efficient decoding. No such benefits are found for signs that result from the local interactions of
isolated pairs of interlocutors. Experiments 2 and 3 showed that the decoding benefits associated with
community evolved signs cannot be attributed to superior sign encoding or detection. Experiment 4
revealed that naive overseers were better able to identify the meaning of community evolved signs
when compared with isolated pair developed signs. Hence, the decoding benefits for community
evolved signs arise from their greater residual iconicity. We argue that community evolved sign
systems undergo a process of communicative selection and adaptation that promotes optimized sign
systems. This results from the interplay between sign diversity and a global alignment constraint;
pairwise interaction introduces a range of competing signs and the need to globally align on a single
sign-meaning mapping for each referent applies selection pressure.
Keywords: graphics; communication; signs; cultural evolution; fitness

1. INTRODUCTION Dediu & Ladd 2007), cultural transmission, the

Like everything else in the natural world, communi-
cation systems evolve: their signals adapt to best fit the
circumstances of the communication. For example,
the ‘whine-plus-chuck’ mating call of the male
Panamanian frog Physalaemus pustulosus is perfectly
adapted to the communicative situations in which it is
used. On one hand, its pitch lies within the range of the
best-hearing frequencies of the female whom it attracts;
on the other hand, the female attracting ‘chuck’ part of
the call is just short enough to make it difficult for
predatory bats to localize (Seyfarth & Cheney 2003).
This kind of biological (i.e. genetic) adaptation may
also be seen in the evolution of human languages. For
example, Dediu & Ladd (2007) showed that the
marked geographical distribution of tone languages is
a result of the distribution of recently evolved alleles of
the brain growth and development genes ASPM and
Microcephalin. They suggested that this reflects the
influences of the genes on the ability of their owners to
easily acquire the tone languages.
However, human communication systems might
also evolve through processes of cultural evolution
(Christiansen & Kirby 2003; Kirby et al. 2007). Whereas
biological adaptation optimizes the language learning
machinery via innate learning biases (Pinker 1994;
* Author and address for correspondence: School of Psychology,
University of Western Australia, 35 Stirling Highway, Crawley, WA
6009, Australia (nfay@psy.uwa.edu.au).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
historical transmission of languages across genera-
tions of learners, could optimize language via lingui-
stic selection (see Kirby & Hurford (1997) for a
computer simulation).
In this paper, we explore the analogy between
biological and cultural evolution by testing whether
cultural transmission tends to produce an optimized, or
fit, system of signs in the way that biological evolution
shapes the mating call of the Panamanian frog. In so
doing, we assess the veracity of the functionalist view of
language, which argues that language has evolved to
support precise and efficient communication (Pinker &
Jackendoff 2005).
Historical and laboratory studies of cumulative
cultural evolution, the process by which knowledge
accumulates across generations, support the function-
alist perspective. Later generations improve upon the
solutions provided by earlier generations, eventually
arriving at solutions that no single individual could
produce on their own. According to Tomasello (1999;
Tomasello et al. 2005), this incremental improvement
in the quality of solutions (ratcheting up) is based on
social learning mechanisms unique to humans. Tech-
nological evolution offers a compelling historical
example of cumulative cultural evolution. Employing
an organic–mechanical analogy, Basalla (1988) pro-
posed that human artefacts undergo a Darwinian
process of survival of the fittest. Specifically, artefacts
that are best suited for certain tasks survive, and are
subject to gradual modifications that improve their
functionality. This is seen in the progressive improve-
ment of the hammer, evolving from a crudely shaped

3553 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3554 N. Fay et al. Culturally evolved communication systems
pounding stone to today’s claw and ball-pein hammers.
A similar outcome is evident under controlled labora-
tory conditions, where performance in producing
artefacts improves, or ratchets up, over successive gene-
rations (e.g. distance travelled by paper aeroplanes and
height of spaghetti tower constructions; Caldwell &
Millen 2008; see also Caldwell & Millen 2008).
It is an open question whether these evolutionary
principles apply to the development of symbolic
artefacts (e.g. linguistic and other sign systems). In
particular, it is unclear whether linguistic systems
undergo the progressive improvement evident in bio-
logical systems and concrete artefacts. As Kirby (2002,
p. 194) observed, ‘We cannot take it for granted that
either learning or cultural evolution are adaptive
mechanisms that seek optimal solutions with regard to
communication, however intuitively appealing that may
appear’. This view was echoed by Plotkin (2002), who
noted that ‘social constructions’ are a product of shared
agreement, and as such may require a fundamentally
different explanation from concrete traits such as
technological artefacts (cited in Mesoudi et al. 2006).
2. ASSESSING THE EVOLUTION OF HUMAN
COMMUNICATION SYSTEMS
Despite the recent resurgence of interest in how present-
day languages have evolved (e.g. Lieberman et al. 2007;
Pagel et al. 2007), little is known about what drives this
evolution. In particular, it is still unclear whether
languages increasingly adapt to fit the needs of speakers
and listeners. It is difficult to assess the fitness of evolved
linguistic systems because there is little available
empirical data; almost all of the systems used today
originated in the prehistoric past. However, there has
been some interesting research on recently evolved sign
languages (e.g. Kegl et al. 1999; Goldin-Meadow 2003;
Sandler et al. 2005). One way of overcoming the lack of
linguistic fossils for spoken languages is to use computer
simulations of communicating agents to test different
hypotheses about how language might evolve (Kirby &
Hurford 2002; Steels et al. 2002; Barr 2004). While
computational models have identified several par-
ameters important for language evolution, they do not
speak to the fitness of the evolved linguistic symbols.
Steels et al. (2002) showed that a stable lexicon is
established via the interactions of a community of
computer agents, but their results do not explain, for
example, why ‘wogglesplat’ was selected among other
signs to convey a particular meaning (e.g. red triangle).
An alternative approach has been to study how
present-day humans perform novel communication
tasks without access to a previously established sign
system. Most of the tasks that have been used to do this
involve graphical communication with participants
having access to various drawing media (Galantucci
2005; Garrod et al. 2007; Healy et al. 2007). The
rationale for using graphical communication tasks to
study the emergence and evolution of sign systems is
that, with the exception of writing and reading,
graphical communication is extremely rare. So graphi-
cal communication tasks allow us to study how people
adapt to new communication media and how graphical
sign systems emerge and evolve over time.
Phil. Trans. R. Soc. B (2008)
(a) (b) (c)
1 2 1
(d ) (e) (f)
2 1 2
Figure 1. Drawing refinement and alignment for the
concept ‘Parliament’ across six games between a pair of
interlocutors playing the Pictionary task (adapted from
Fay et al. in preparation): (a) game 1, (b) game 2, (c) game 3,
(d ) game 4, (e) game 5, ( f ) game 6. Participant numbers are
given in bold on the top right of the drawing.
A consistent finding across studies is the crucial role
of feedback and interaction to the development of
graphical sign systems. In Garrod et al. (2007),
participants communicated a series of predetermined
concepts by drawing on a standard whiteboard. Like
the game Pictionary, participants were not allowed
to speak or use text in their drawings, forcing them to
create a novel sign system. In one condition, pairs of
participants graphically communicated a set of recur-
ring concepts (e.g. Art gallery, Drama, Arnold
Schwarzenegger and Television), alternating between
drawing and identifying roles from game to game.
Figure 1 illustrates the changing form of the sign
representing ‘Parliament’ across six games.
What begins as an iconic depiction of Parliament,
a figurative illustration of the debating chambers,
develops, through a process of local adaptation and
entrainment, into a simplified symbolic form (two lines
plus a circle). Not only are participants’ drawings
refined across games, but they also become increasingly
similar, or aligned. Crucially, simple repetition of
drawings was insufficient to produce simplification
and abstraction. This occurred only when there was a
feedback from the addressee. Garrod et al. (2007)
argued that interactive graphical communication
allows participants to develop shared symbolic rep-
resentations from what started out as primarily iconic
representations through a ‘grounding’ process similar
to that found in interactive spoken communication
(Clark 1996). Further research using the Pictionary
task contrasts this local process with the global
evolution of a ‘visual lexicon’ within a community of
interlocutors ( Fay et al. in preparation).
Four 8-person laboratory communities, or micro-
societies, were created via the one-to-one interactions of
partners drawn from the same pool. Participants played
six consecutive games with a partner, where each game
contained the same to-be-communicated items (16
targets plus four distracters, presented in a different
random order on each game) that were known to both
partners. As in the previous example, drawing and
identifying roles alternated from game to game. Partici-
pants then switched partners and played a further six
games with a new partner, and continued to do so until
they had interacted with each of the other community
members (table 1 displays the sequence of partner
interactions in each community). Communities were
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Table 1. The sequence of partner interactions in each
community. (Within each ‘round’, participants played six
consecutive games of the Pictionary task with a different
partner.)
round pair composition
1 1 and 2 3 and 4 5 and 6
2 1 and 4 3 and 2 5 and 8
3 1 and 6 3 and 8 5 and 2
4 1 and 8 3 and 6 5 and 4
5 1 and 3 2 and 4 5 and 7
6 1 and 5 2 and 6 3 and 7
7 1 and 7 2 and 8 3 and 5
designed such that a conventional communication
system could be established by the time participants
encountered their fourth partner. For instance, assume
person 2 adopts person 1’s sign system (round 1, table 1),
and that person 2 then influences person 3 (round 2). If
person 8 aligns with person 3 (round 3), persons 1 and 8
will share a similar communication system (round 4)
despite having never directly interacted.
The community condition was contrasted with an
isolated pair condition, in which participants interacted
with the same partner over the same number of games
(i.e. 42 games; see Garrod & Doherty (1994) for a
natural language analogue). The task was administered
using a virtual whiteboard tool (Healy et al. 2002), with
each participant seated at a computer terminal and
drawing input and item selection made via a standard
mouse. Crucially, participants were unaware of the
identity of their partner in any round. Figure 2
illustrates the global and local evolution of the sign
representing ‘Brad Pitt’ within a single community and
a corresponding number of isolated pairs.
The first drawings of Brad Pitt (figure 2, round 1)
illustrate the diversity of graphical signs; some indicate
his American origins, others his frequent casting as a
ladies man, while others use the rebus principle to
represent part of the test item (community members 5
and 6 draw a large hole in the ground to convey a ‘pit’,
whereas isolated pair member 4 draws an arrow pointing
at an arm pit). Drawing diversity at round 1 in the
community condition contrasts sharply with drawing
uniformity at round 7, where all community members
have globally converged on a refined version of person
5’s initial pit drawing. Unlike community members,
isolated pairs locally converged on a shared sign system,
but globally diverged across games. Note that in both
conditions groups arrived at a series of signs of equal
visual complexity (see Fay et al. in preparation).
In this paper, we investigate whether the signs (i.e.
drawings) produced by communities are better adapted
for use in the larger population than those produced by
isolated pairs. If these signs undergo systematic
adaptation, then we would expect signs evolved by
isolated pairs to be fit only for that pair, whereas
community evolved signs should be fit for the larger
population from whom the community had been
created. We test this hypothesis by contrasting the
accuracy and ease with which community and isolated
pair evolved signs can be learned (experiments 1 and 2)
and detected (experiment 3) by new subjects drawn
Phil. Trans. R. Soc. B (2008)
Culturally evolved communication systems N. Fay et al. 3555
from the same population. Experiments 1–3 can be
thought of as fractionating ‘levels of meaningfulness’ of
the signs: decoding of learned signs assesses the strength
of the stored sign-meaning mapping (high meaning-
fulness); encoding of learned signs addresses the effort
required to discriminate between the signs themselves
(low meaningfulness); and detection determines the
7 and 8 ease with which the signs are perceived (no meaning).
7 and 6 Clearly an effective sign should be easily detected,
7 and 4
efficiently encoded into memory and its meaning should
7 and 2
6 and 8
be accurately and efficiently derived from the sign.
4 and 8 However, it may be that community evolved signs are
4 and 6 superior to those that develop among isolated pairs
because the signs preserve more salient, concrete
information, despite their comparable visual complexity.
In other words, they have more residual iconicity. If
greater iconicity distinguishes community from isolated
pair evolved signs (i.e. the strength of the sign-meaning
mapping), then the benefits of community signs should
be most clearly seen in experiment 1 (decoding).
Experiment 4 provides a direct test of sign iconicity by
testing naive overseers’ ability to guess the meaning
associated with community and isolated pair evolved
signs. Thus, the current study tests the intuitive
hypothesis that the product of cultural evolution, or
glossogeny, is an optimized sign system, a position
consistent with the functionalist perspective of language.
3. LEARNABILITY STUDIES
The first two experiments investigate the learnability of
signs evolved by communities as compared to those
developed by isolated pairs. Experiment 1 assesses the
relative decoding of the two kinds of meaningful sign.
Experiment 2 assesses their learned discriminability.
(a) Experiment 1. Decoding community
and isolated pair evolved signs
(i) Methods
Participants and apparatus
Thirty-two undergraduate psychology students partici-
pated in exchange for payment. Participants were
tested individually in sessions lasting 45 min. Stimuli
were presented and controlled by a personal computer
with a monitor refresh rate of 100 Hz (the same
apparatus was used in experiments 2–4).
Materials and design
Stimulus materials were drawn from Fay et al.
(in preparation). The stimuli consisted of 512 commu-
nity drawings (32 participants randomly allocated to
4!8-person communities) and 512 isolated pair
drawings (32 participants randomly allocated to 16
isolated pairs). Half the drawings were sampled at
game 1 of round 1 (i.e. pre-interaction; 16 concepts!32
pairs) and the other half at game 1 of round 7 (i.e. post-
interaction; 16 concepts!32 pairs). Images were
sampled such that each participant was presented with
images produced by a community pair and an isolated
pair at round 1 or round 7. This equated to 32 images
per participant (16 community produced images at
round 1/7 and 16 isolated pair generated images at
round 1/7). Thus, the corpus was sampled once across
participants. A mixed design was used.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3556 N. Fay et al. Culturally evolved communication systems

(a) (b)
1 2 1 7

3 4 2 8

5 6 3 5

7 8 4 6

(c) (d)
1 2 1 2

3 4 3 4

5 6 5 6

7 8 7 8

Figure 2. Drawing refinement and alignment for the concept ‘Brad Pitt’ among (a,b) a community of interlocutors and (c,d )
between isolated pairs at round 1 (a,c) and round 7 (b,d ) (game 1) of the Pictionary task (adapted from Fay et al. in preparation).
Participant numbers are given in bold on the top right of the drawing.

Procedure The timed decoding task required participants to

Participants were tested individually in a quiet testing
booth. The experiment began with a training phase,
where the participants learned the identity of each
image (32) before progressing to a two alternative
forced-choice reaction time task. During training,
participants viewed each image and its associated
label (e.g. ‘Microwave’ and ‘Soap Opera’; presented
in a random order), pressing the space bar to progress
to the next image–label pair. At test, each image was
presented and participants were cued to select the
associated label from an adjacent list. Identification
accuracy of 80 per cent or better was required for
participants to proceed to the reaction time task.
If required, the training phase was repeated until
the participant achieved 80 per cent accuracy or better
on test.
decide whether a presented label matched a previously
presented image. Each trial began with the presentation
of a fixation cross in the centre of the screen (500 ms).
Next the target image was presented (40 ms), followed
immediately (0 ms onset) by a mask (50 ms). Each
mask was a scrambled version of the most complex
target image (measured in pixels). A matching or
mismatching label was then presented in the centre of
the screen, to which participants responded either
match (by pressing the ‘f ’ key) or mismatch (by
pressing the ‘j’ key). A feedback beep indicated whether
their response was correct or incorrect (see figure 3 for
a sample trial sequence).
Half the trials were ‘match’ trials, where the target
image agreed with the associated label. The remaining
trials were ‘mismatch’ trials (e.g. a drawing of

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culturally evolved communication systems
500 ms 40 ms
Figure 3. A sample trial sequence from experiment 1.
Microwave followed by the label Parliament). Each
image was displayed eight times: four times in a match
trial and four times in a mismatch trial. Images and
their associated labels were presented in a random
order throughout. Participants completed 256 trials
(16 concepts!2 pairs!2 trial types!4 repetitions).
The computer recorded their accuracy and response
latency for each trial.
Results
All participants successfully completed the training
phase before progressing to the reaction time task. A
majority learned the image–label pairing to the pre-
determined criterion (80% accuracy) at the first attempt
(21, or 66%), with a few requiring a second attempt
(8, or 25%) and fewer still requiring a third attempt (3, or
9%). Round 1 and round 7 community and isolated
pair images were equally well learned at training
(88% accuracy).
Figure 4 displays the mean hit rate (% correct
responses) and response latencies (in milliseconds) for
community and isolated pair evolved signs at round 1
and round 7 (image–label matching trials). Per-
formance at round 1 is equivalent. At round 7,
community evolved signs are more accurately recog-
nized and more efficiently processed than isolated pair
developed signs. ANOVA confirms these observations.
Participants’ mean hit rates were entered into a
mixed-design ANOVA, treating group (community
and isolated pair) as a within-subjects factor and
round (1 and 7) as a between-subjects factor (for all
F and t values reported p!0.05 unless otherwise
stated). This returned a main effect of group (F1,30Z
7.28, h2p Z0.20), but no effect of round (F1,30Z2.76,
pO0.05, h2p Z0.08). However, the main effect of group
was mediated by a reliable group by round interaction
(F1,30Z4.12, h2p Z0.12). This was due to the simple
effect of group at round 7 (F1,30Z11.17, dZ1.08), with
no such effect at round 1 (F!1). Identical findings
were returned when the ANOVA was repeated using
participants’ d scores.
The same ANOVA was carried out on participants’
response latencies. Mean response times were calcu-
lated after the removal of times 2.5 standard deviations
from the condition median. These extreme scores were
replaced by values corresponding to the median plus or
minus 2.5 standard deviations. This accounted for 2.7
per cent of the data. ANOVA returned a main effect of
group (F1,30Z12.46, h2p Z0.29), but no effect of round
(F!1). Again, the main effect of group was mediated
by a reliable group by round interaction (F1,30Z9.85,
h2p Z0.25). The interaction reflects the simple effect of
group at round 7 (F1,30Z22.23, dZ0.47), with no
difference at round 1 (F!1).
Phil. Trans. R. Soc. B (2008)
N. Fay et al. 3557
Microwave
50 ms match or mismatch
1100
83% 82% 83%
1050
1000
response time (ms)
91%
950
900
850
800
750
700
1 7
round
Figure 4. Mean hit rate (%) and response latency (ms) for
community (grey bars) and isolated pair (white bars) evolved
signs in image–label matching trials at round 1 and round 7.
Error bars indicate the standard error of the mean.
In summary, community evolved signs (round 7)
offer a substantial learning advantage over signs that
locally develop among isolated pairs. Furthermore,
these signs are more rapidly accessed when compared
with isolated pair developed signs of equal visual
complexity. To establish the validity of the iconicity-
based account (i.e. that the accessibility benefits seen for
community evolved signs are a function of their greater
residual iconicity), two competing explanations must be
ruled out: encoding and detection. Ease of sign encoding
and detection are tested in experiments 2 and 3.
(b) Experiment 2. Encoding community
and isolated pair evolved signs
The decoding benefits associated with community
evolved signs (experiment 1) may be attributable to
more efficient sign encoding (i.e. community evolved
signs are more efficiently encoded in working memory)
and greater sign discriminability (i.e. community
evolved signs are more distinct and are therefore less
confusable than isolated pair developed signs). Experi-
ment 2 tested this possibility by comparing the ease
with which community and isolated pair evolved signs
are encoded and distinguished from one another.
(i) Method
Participants
Thirty-two undergraduate psychology students partici-
pated in exchange for payment. Participants were
tested individually in sessions lasting 45 min.
Materials and design
Stimulus materials were again drawn from Fay et al.
(in preparation). Eight images were sampled from each
community pair and isolated pair at round 1 and round
7. Images were sampled such that each participant
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3558 N. Fay et al. Culturally evolved communication systems
was exposed to all 16 concepts (four targets and
four distracters sampled from a community pair and
an isolated pair at round 1 or round 7), with no
duplication of item types. This meant that half the
corpus was used. A mixed design was employed.
Procedure
The experiment consisted of a training phase in which
participants learned eight target images followed by an
inspection-time task. At training, participants viewed
eight target images presented in a random sequence,
pressing the space bar to proceed from one target to the
next. A recognition memory test followed where
participants were cued to identify each image as a
target or distracter (eight targets plus eight distracters).
Memory performance exceeding 80 per cent allowed
progression to the inspection-time task.
The inspection-time task required participants to
recognize images (as targets or distracters) presented at
varying exposure durations. A parameter estimation
by sequential testing algorithm ( Taylor & Creelman
1967) determined the minimum exposure duration
required by each participant to respond at 70 per cent
accuracy (see Treutwein (1995) for a review). Individ-
ual staircases lasting 50 trials were performed on each
image. Each trial consisted of the presentation of an
image for the determined inspection time, followed by
a mask (0 ms onset) that remained on the screen
until participants identified the image as a target (by
pressing the ‘f’ key) or a distracter (by pressing the
‘j’ key). The mask was identical to that used in
experiment 1. A feedback beep informed participants
if their response was correct or incorrect. Participants
completed 800 trials in total (16 images!50 trials).
Image presentation was randomized throughout.
Results
All participants successfully completed the training
phase at the first attempt. Community and isolated pair
target images were equally well learned at round 1 and
round 7 (97% accuracy). Participants’ mean inspection
times were calculated across the last five trials for each
target image (community or isolated pair at round 1 or
round 7). Inspection times were computed after
the removal of times 2.5 standard deviations from the
condition median. These extreme scores were replaced
by values corresponding to the median plus or minus
2.5 standard deviations. This accounted for 3.4 per
cent of the data.
Figure 5 displays the mean inspection time (in
milliseconds) required to identify each target image to
the predetermined criterion (70% accuracy). There
was no difference between conditions. This was
confirmed by ANOVA (same design as experiment 1).
There was no effect of group (F1,30Z2.78, pO0.05,
h2p Z0.08), round (F!1) or group by round interaction
(F!1). The similar inspection times for round 1 and
round 7 signs indicate that the greater visual complex-
ity of round 1 signs did not slow sign encoding,
suggesting that participants needed only to encode part
of the round 1 signs for successful recognition. More
importantly, the near-identical encoding efficiency of
round 7 community and isolated pair images indicates
that the decoding benefits of community evolved signs
Phil. Trans. R. Soc. B (2008)
40
35
30
inspection time (ms)
25
20
15
10
5
0
1 7
round
Figure 5. Mean inspection time (ms) required to correctly
recognize round 1 and round 7 community (grey bars) and
isolated pair (white bars) target images on 70 per cent of the
trials. Error bars indicate the standard error of the mean.
(experiment 1) cannot be attributed to more efficient
sign encoding or discrimination.
(c) Experiment 3. Detecting community
and isolated pair evolved signs
If community evolved signs are easier to detect than
isolated pair developed signs, then this would provide
an alternative explanation of the decoding benefits seen
for community evolved signs (experiment 1). This
possibility was tested in experiment 3.
(i) Method
Participants
Sixteen undergraduate psychology students partici-
pated in exchange for payment. Participants were
tested individually in sessions lasting 1 hour.
Materials and design
Stimulus materials were identical to those used in
experiment 1. Images were sampled such that each
participant was presented with images generated by
two community pairs (from different communities) and
two isolated pairs, randomly sampled at round 1 and
round 7. This equated to 128 images per participant
(16 concepts!2 community pairs!2 isolated pairs!2
rounds). Thus, the corpus was sampled twice across
participants. A within-subjects design was used.
Procedure
Participants completed a two alternative forced-choice
task in a quiet testing booth. Each image was presented
once at each of three exposure durations (10, 20 or
30 ms), with a corresponding number of target-absent
trials (i.e. 128 target-present trials plus 128 target-absent
trials!3 exposure durations). Images were presented in
a random order. Each trial began with the presentation
of a fixation cross in the centre of the screen (500 ms).
Next a target image or blank screen was presented
(10, 20 or 30 ms), followed immediately (0 ms onset) by
a mask (50 ms). The mask was identical to that used in
experiments 1 and 2. Participants were cued to respond
whether the target was present (by pressing the ‘f’ key) or
absent (by pressing the ‘j’ key).
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
100
hit rate (%) 90
80
70
60
50
40
30
20
R1 R7 R1 R7 R1
10 20
round (R) and exposure duration (ms)
Figure 6. Mean hit rate (%) for community (grey bars) and
isolated pair (white bars) signs at round 1 (R1) and round 7
(R7) at 10, 20 and 30 ms exposure durations. Error bars
indicate the standard error of the mean.
Results
Figure 6 displays the mean hit rate (% correct
responses in target-present trials) for each condition
at 10, 20 and 30 ms exposure durations. Community
and isolated pair evolved signs were detected equally
well at each exposure duration, although in both
conditions round 1 images were more accurately
detected than round 7 images.
Participants’ mean performance scores (% correct)
were entered into an ANOVA that treated group
(community and isolated pair), round (1 and 7) and
exposure duration (10, 20 and 30 ms) as within-subject
factors. This returned a main effect of round (F1,15Z
10.99, h2p Z0.42) and exposure duration (F1,15Z129.31,
h2p Z0.90), but no effect of group (F!1). There were no
interaction effects (Fs!2.82). Further analysis of the
main effect of exposure duration confirmed a large
improvement in detection rates between 10 and 20 ms
(t15Z12.54, dZ3.23) and a smaller improvement
between 20 and 30 ms (t15Z2.45, dZ0.71).
Thus, participants were better able to detect the
presence of an image at longer exposure durations. Not
surprisingly, participants were also more successful at
detecting the more visually complex round 1 images.
Identical findings were returned when the ANOVA was
repeated using participants’ d scores. The equivalent
detection rates for community and isolated pair images
confirm that the decoding benefits seen for community
evolved signs (experiment 1) cannot be attributed to
ease of detection.
(d) Experiment 4. The transparency of commu-
nity and isolated pair evolved signs
Experiment 4 provides a direct test of the hypothesis that
community evolved signs have greater residual iconicity
than those developed by isolated pairs. To test this, we
assessed the degree to which naive observers could guess
the original meaning of the two kinds of signs.
(i) Method
Participants and apparatus
Thirty-two undergraduate psychology students partici-
pated in exchange for payment. Participants were
tested individually in sessions lasting 30 min.
Materials and design
Stimulus materials were again drawn from Fay et al.
(in preparation). Unlike experiments 1–3, where
participants were shown static images, in experiment 4
Phil. Trans. R. Soc. B (2008)
Culturally evolved communication systems N. Fay et al. 3559
60
50
per cent correct
40
R7 30
30
20
10
0
1 7
round
Figure 7. Overseers’ mean identification accuracy (%) for
community (grey bars) and isolated pair (white bars) signs at
round 1 and round 7. Error bars indicate the standard error of
the mean.
images were animated, replicating the dynamic drawing
construction experienced by the actual matcher. The
virtual whiteboard tool (Healy et al. 2002) used by Fay
et al. (in preparation) enables pixel-by-pixel playback of
the drawing activity within each experimental trial.
Trial playback was converted to QuickTime animations
and used as stimuli in experiment 4. Each participant
attempted to identify the meaning of the animated
drawings produced by a community pair and an isolated
pair at round 1 or round 7 (32 animations in total). The
corpus was sampled once across participants. A mixed
design was used.
Procedure
Participants completed the task individually in a quiet
testing booth. Trials were initiated with the presen-
tation of a fixation cross (500 ms) followed by
animation playback. Participants then tried to identify
the referent of the animated drawing (by key press)
from an adjacent list of 20 concepts (e.g. ‘Homesick’,
‘Cartoon’ and ‘Computer Monitor’). The animated
drawings were presented in a random order.
Results
Figure 7 displays overseers’ mean identification rate
(%) for community and isolated pair evolved signs at
round 1 and round 7. Identification accuracy at round
1 is equivalent, whereas at round 7 community evolved
signs are more accurately identified. This was
confirmed by ANOVA.
Participants’ mean identification accuracy scores
were entered into a mixed-design ANOVA as per
experiments 1 and 2. This returned a main effect of
group (F1,30Z5.06, h2p Z0.14), round (F1,30Z21.35,
h2p Z0.42) and a reliable group by round interaction
(F1,30Z10.05, h2p Z0.25). The interaction is explained
by the simple effect of group at round 7 (F1,30Z14.68,
dZ1.27), with no such effect at round 1 (F!1). Thus,
community evolved signs retain more residual iconicity
when compared with isolated pair developed signs, and
this makes the translation from sign to meaning more
transparent to naive overseers.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3560 N. Fay et al. Culturally evolved communication systems
4. DISCUSSION
In this paper, we investigated the fitness of a small
visual lexicon that evolved among members of an
interacting community engaged in a graphical com-
munication task similar to the game Pictionary. For
reasons outlined earlier, the Pictionary task offers a
useful vehicle to study the emergence and evolution of
communication systems under controlled laboratory
conditions. In particular, the corpus of signs generated
by participants offers a rare opportunity to determine
whether the product of cultural evolution, or glos-
sogeny, is an optimized communication system.
The results show that graphical signs that evolve
within a community offer distinct advantages when
compared with those that locally develop among isolated
pairs. In particular, the meaning associated with a
particular sign is more accessible for a subsequent
generation of sign learners (experiment 1). This
decoding benefit cannot be explained by differences in
speed of discriminating or ease of detecting community
evolved signs (experiments 2 and 3, respectively).
Instead, the benefit arises from the greater residual
iconicity of community evolved signs (experiment 4).
From a functional point of view, this can be explained in
relation to Garrod et al.’s (2007) information theoretic
account of different kinds of signs. They argue that icons
differ from symbols in terms of where the information
they convey lies. Icons work through resemblance to the
objects they signify. Hence, they are effective to the
extent that their graphical structure (i.e. information)
maps onto the physical structure of the object. In this
way, the graphical complexity of icons is related to the
physical complexity of the signified object. By contrast,
symbols are effective to the extent that their structure
maps onto the structure of other instances of the symbol
used previously to signify that object. Other things being
equal, graphical symbols can become structurally
simpler (i.e. bear less information) than graphical
icons because they require only sufficient structure to
differentiate them from other symbols in the domain. In
turn, this simplification facilitates sign production
making it increasingly fluent. Hence, from the point of
view of communicative fitness, it makes sense for icons
(or indices) to evolve into symbols. And this is what
happens with both isolated pairs and in communities.
But what about the global evolution unique to the
community signs? In a group context, signs need to be
effective both in terms of communicative fitness within
each pair of the group and in terms of transmission
fitness for other group members. Our results indicate
that communities achieve this by developing increas-
ingly simple signs, but nevertheless signs that retain
sufficient residual iconicity to be easily recognized
(experiment 4) and learned (experiment 1) by new
members of the population from which the community
was drawn. So, in both community and isolated pair
conditions, graphical signs evolve functionally, becom-
ing progressively refined and therefore more efficiently
produced and decoded by interlocutors. However, only
community evolved signs exhibit learning and decoding
benefits for persons not actively engaged in sign
construction. As these benefits are unanticipated (i.e.
the signs are not ‘designed’ with an external audience
in mind), sign fitness is a ‘functional by-product’ of
Phil. Trans. R. Soc. B (2008)
adaptation in the community condition. Thus, like the
Panamanian frog’s mating call, community evolved
signs are optimized in two ways at once: ease of
production and ease of learning by subsequent
generations. Herein lies the fitness and functionality
of community evolved sign systems.
We turn now to the proximal mechanisms promot-
ing the fitness of community evolved sign systems.
Intuitively, and consistent with a biological account,
the benefits of community evolved sign systems may
derive from the greater pool of exemplars that
communities can draw on. Community members
have eight exemplars of each sign-referent pairing to
select from, whereas isolated pairs have only two (i.e.
one per member). While diversity is crucial to
biological and cultural evolution, diversity alone cannot
account for the observed benefits of community
evolved sign systems. As Plotkin (2002) observed,
social constructions are a product of shared agreement.
Clearly, for a sign to ‘work’ there must be substantial, if
implicit, agreement between interlocutors with regard
to what the sign signifies. In other words, communi-
cation systems rely on conceptual alignment (see
Pickering & Garrod 2004). For isolated pairs local
alignment is sufficient, whereas for communities global
alignment is necessary. We propose that the fitness of
community evolved sign systems derives from the
diversity of potential signs, and the need to globally
align on a single sign-meaning mapping. A similar
mechanism seems to operate with the development of
community-wide linguistic conventions.
Garrod & Doherty (1994) examined the linguistic
description schemes developed by pairs working together
to navigate around a computerized maze. Three
conditions were compared: isolated pairs, communities
and non-communities (participants paired with a series
of different partners not drawn from the same commu-
nity). Like Fay et al. (in preparation), isolated pairs locally
developed a range of different maze description schemes,
whereas community members globally aligned on a
single community-wide description scheme. Interest-
ingly, the descriptions of non-community members, who
were privy to a diverse range of maze descriptions,
became increasingly misaligned as they encountered new
partners, with participants tending to use individually
salient description schemes irrespective of the scheme
used by their current partner. Importantly,
the description schemes adopted by community players
were more efficient than those used by members of
isolated pairs and non-communities, requiring fewer
communicative moves to navigate successfully through
the maze. The comparison between community and non-
community participants indicates that diversity alone
cannot account for the benefits of community evolved
linguistic description schemes.
In conclusion, like the progressive adaptation charac-
teristic of biological systems and concrete artefacts,
communicative artefacts undergo a Darwinian process
of survival of the fittest that promotes optimized sign
systems. Via pairwise interaction, community members
produce a range of competing signs, and the need to
globally align on a series of sign-meaning mappings
applies selection pressure. The interplay between sign
diversity and this global alignment constraint results in a
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culturally evolved communication systems
series of schematized signs that retain substantial
residual iconicity. This aids sign production, individual
learning and the efficient translation from sign to
meaning. Thus, the present study illustrates the parallels
between phylogeny and glossogeny, a position consistent
with a functionalist view of language. Our findings also
have implications for icon design (e.g. icons for maps,
computer displays, road signs and logos). The commu-
nity evolved signs capture two core aspects of good icon
design, concreteness and simplicity, factors that enhance
individual learning and speed of processing (Gittins
1986; McDougall 2000). Harnessing the ‘apparent’
design prevalent in interacting communities offers an
exciting alternative to traditional design practices.
This research was supported by an ARC Discovery Grant
(grant DP0556991) awarded to N.F. We thank two
anonymous reviewers for their helpful comments and Mike
Anderson, Nicholas Badcock and Daniel Little for their
programming assistance.
REFERENCES
Barr, D. J. 2004 Establishing conventional communication
systems: is common knowledge necessary? Cognit. Sci. 28,
937–962. (doi:10.1016/j.cogsci.2004.07.002)
Basalla, G. 1988 The evolution of technology. Cambridge, UK:
Cambridge University Press.
Caldwell, C. A. & Millen, A. E. 2008 Experimental models
for testing hypotheses about cumulative cultural
evolution. Evol. Hum. Behav., 165–171. (doi:10.1016/
j.evolhumbehav.2007.12.001)
Caldwell, C. A. & Millen, A. E. 2008 Studying cumulative
cultural evolution in the laboratory. Phil. Trans. R. Soc. B
363, 3529–3539. (doi:10.1098/rstb.2008.0133)
Christiansen, M. H. & Kirby, S. 2003 Language evolution:
consensus and controversies. Trends Cognit. Sci. 7,
300–307. (doi:10.1016/S1364-6613(03)00136-0)
Clark, H. H. 1996 Using language. Cambridge, UK:
Cambridge University Press.
Dediu, D. D. & Ladd, D. R. D. R. 2007 Linguistic tone is
related to the population frequency of the adaptive
haplogroups of two brain size genes, ASPM and
Microcephalin. Proc. Natl Acad. Sci. USA 104,
10 944–10 949. (doi:10.1073/pnas.0610848104)
Fay, N., Garrod, S., Roberts, L. & Swoboda, N. In preparation.
The interactive evolution of communication systems.
Galantucci, B. 2005 An experimental study of the emergence
of human communication systems. Cognit. Sci. 29,
737–767. (doi:10.1207/s15516709cog0000_34)
Garrod, S. & Doherty, G. 1994 Conversation, coordination
and convention—an empirical investigation of how groups
establish linguistic conventions. Cognition 53, 181–215.
(doi:10.1016/0010-0277(94)90048-5)
Garrod, S., Fay, N., Lee, J., Oberlander, J. & MacLeod, T.
2007 Foundations of representation: where might graphi-
cal symbol systems come from? Cognit. Sci. 31, 961–987.
(doi:10.1080/03640210701703659)
Gittins, D. 1986 Icon-based human computer interaction.
Int. J. Man-Machine Stud. 24, 519–543.
Goldin-Meadow, S. 2003 The resilience of language: what
gesture creation in deaf children can tell us about how children
learn language. New York, NY: Psychology Press.
Healy, G. T., Swoboda, N. & King, J. 2002 A tool for
performing and analysing experiments on graphical
communication. In People and computers XVI: Proc.
HCI2002: The 16th British HCI Group Annual Conference
(eds X. Faulkner, J. Finlay & F. Detienne), pp. 55–68.
London, UK: Springer.
Phil. Trans. R. Soc. B (2008)
N. Fay et al. 3561
Healy, P. G. T., Swoboda, N., Umata, I. & King, J. 2007
Graphical language games: interactional constraints on
representational form. Cognit. Sci. 31, 285–309. (doi:10.
1080/15326900701221363)
Kegl, J., Senghas, A. & Coppola, M. 1999 Creation through
contact: sign language emergence and sign language change
in Nicaragua. In Language creation and language change:
creolization, diachrony and development (ed. M. DeGraff ),
pp. 179–237. Cambridge, MA: MIT Press.
Kirby, S. 2002 Natural language from artificial life. Artif. Life
8, 185–215. (doi:10.1162/106454602320184248)
Kirby, S. & Hurford, J. 1997 Learning, culture and evolution
in the origin of linguistic constraints. In Proc. Fourth
European Conference on Artificial Life (eds S. Husbands &
I. Harvey), pp. 493–502. Cambridge, MA: MIT Press.
Kirby, S. & Hurford, J. 2002 The emergence of linguistic
structure: an overview of the iterated learning model. In
Simulating the evolution of language (eds A. Cangelosi &
D. Parisi), pp. 121–148. London, UK: Springer.
Kirby, S., Dowman, M. & Griffiths, T. L. 2007 Innateness and
culture in the evolution of language. Proc. Natl Acad. Sci.
USA 104, 5241–5245. (doi:10.1073/pnas.0608222104)
Lieberman, E., Michel, J. B., Jackson, J., Tang, T. & Nowak,
M. A. 2007 Quantifying the evolutionary dynamics of
language. Nature 449, 713–716. (doi:10.1038/nature06137)
McDougall, S. 2000 Exploring the effects of icon charac-
teristics on user performance: the role of icon concrete-
ness, complexity, and distinctiveness. J. Exp. Psychol. Appl.
6, 291–306. (doi:10.1037/1076-898X.6.4.291)
Mesoudi, A., Whiten, A. & Laland, K. N. 2006 Towards a
unified science of cultural evolution. Behav. Brain Sci. 29,
329–347. (doi:10.1017/S0140525X06009083)
Pagel, M., Atkinson, Q. D. & Meade, A. 2007 Frequency of
word-use predicts rates of lexical evolution throughout
Indo-European history. Nature 449, 717–U717. (doi:10.
1038/nature06176)
Pickering, M. J. & Garrod, S. 2004 Toward a mechanistic
psychology of dialogue. Behav. Brain Sci. 27, 169–190.
(doi:10.1017/S0140525X04000056)
Pinker, S. 1994 The language instinct. New York, NY:
HarperCollins.
Pinker, S. & Jackendoff, R. 2005 The faculty of language:
what’s special about it? Cognition 95, 201–236. (doi:10.
1016/j.cognition.2004.08.004)
Plotkin, H. C. 2002 The imagined world made real. Harmonds-
worth, UK: Penguin.
Sandler, W., Meir, I., Padden, C. & Aronoff, M. 2005 The
emergence of grammar: systematic structure in a new
language. Proc. Natl Acad. Sci. USA 102, 2661–2665.
(doi:10.1073/pnas.0405448102)
Seyfarth, R. M. & Cheney, D. L. 2003 Signalers and receivers
in animal communication. Annu. Rev. Psychol. 54,
145–173. (doi:10.1146/annurev.psych.54.101601.145121)
Steels, L., Kaplan, F., McIntyre, A. & Van Looveren, J. 2002
Crucial factors in the origins of word-meaning. In The
transition to language (ed. A. Wray), pp. 252–271. Oxford,
UK: Oxford University Press.
Taylor, M. M. & Creelman, C. D. 1967 PEST: efficient
estimates in probability functions. J. Acoust. Soc. Am. 4,
782–787. (doi:10.1121/1.1910407)
Tomasello, M. 1999 The cultural origins of human cognition.
Cambridge, MA: Harvard University Press.
Tomasello, M., Carpenter, M., Call, J., Behne, T. & Moll, H.
2005 Understanding and sharing intentions: the origins of
cultural cognition. Behav. Brain Sci. 28, 675–691. (doi:10.
1017/S0140525X05000129)
Treutwein, B. 1995 Adaptive psychophysical procedures.
Vis. Res. 35, 2503–2522. (doi:10.1016/0042-6989(95)
00016-X)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3563–3575

doi:10.1098/rstb.2008.0135
Published online 19 September 2008

Culture, embodiment and genes: unravelling

the triple helix
Michael Wheeler1,* and Andy Clark2
1
Department of Philosophy, University of Stirling, Stirling FK9 4LA, UK
2
School of Philosophy, Psychology, and Language Sciences, David Hume Tower, George Square,
Edinburgh EH8 9J7, UK
Much recent work stresses the role of embodiment and action in thought and reason, and celebrates
the power of transmitted cultural and environmental structures to transform the problem-solving
activity required of individual brains. By apparent contrast, much work in evolutionary psychology has
stressed the selective fit of the biological brain to an ancestral environment of evolutionary
adaptedness, with an attendant stress upon the limitations and cognitive biases that result. On the face
of it, this suggests either a tension or, at least, a mismatch, with the symbiotic dyad of cultural evolution
and embodied cognition. In what follows, we explore this mismatch by focusing on three key ideas:
cognitive niche construction; cognitive modularity; and the existence (or otherwise) of an evolved
universal human nature. An appreciation of the power and scope of the first, combined with
consequently more nuanced visions of the latter two, allow us to begin to glimpse a much richer vision
of the combined interactive potency of biological and cultural evolution for active, embodied agents.
Keywords: cultural transmission; embodied cognition; niche construction; evolutionary psychology;
modularity; neuroconstructivism

1. INTRODUCTION: A TENSION REVEALED
There is a natural affinity between work that stresses the
role of embodiment and action in thought and reason
(examples include Varela et al. 1991; Clark 1997; Noë
2004; Wheeler 2005) and work that explores the
cognitive role of cultural evolution ( Tomasello 1999;
Kirby 2002; Sterelny 2003). Both approaches share an
emphasis on the power of non-neural structures to
transform the shape of the problem-solving activity
required of individual brains. Such potent non-neural
structures take a wide variety of forms, from the
biomechanics of the gross physical body (Collins et al.
2005), to the structural features of a linguistic code
(Kirby 2002), and on to aspects of the local, physical
and social environment (for some reviews, see Clark
1997; Wilson & Clark in press). Many of these enabling
non-neural structures are self- or species created, and
are thus both products and determinants of human
thought and activity. Such products and determinants
are also subject to cycles of transmission, alteration and
inheritance, in at least a rough analogy with genetic
inheritance systems (e.g. Jablonka & Lamb 2005). The
result (as we shall see) is a vision of the evolution, the
development and the real-time unfolding of human
cognition, in which a kaleidoscope of complex ratchet
effects fuel the flexible and, to a significant degree,
open-ended character of thought and action.
By apparent contrast, much work in evolutionary
psychology1 has stressed the selective fit of the
biological brain to some ancestral environment of
evolutionary adaptedness, with an attendant focus
* Author for correspondence (m.w.wheeler@stir.ac.uk).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
upon the limitations and cognitive biases that result
(see, canonically, Barkow et al. 1992. For more recent
coverage, see Buss 2005). On the face of it, this
suggests either a tension or, at least, a mismatch with
our symbiotic dyad of cultural evolution and embodied
cognition. In place of a dynamic and transformative
interplay of neural, bodily and (sometimes self-created)
environmental resources over different time scales, we
confront a restricted set of pre-specified adapted
functions, performed in the triggering context of
variable non-neural structures and cultural forces, by
relatively static, genetically based forms of neural
encoding and processing.
In what follows, we explore this mismatch by focusing
on three key ideas: cognitive niche construction;
cognitive modularity; and the existence of an evolved
human nature. An appreciation of the power and scope
of the first, combined with consequently more nuanced
visions of the latter two, allow us (we shall argue) to
begin to glimpse a much richer vision of the combined
potency of biological and cultural evolution for active,
embodied agents. In §2, we explain the basic idea of
cognitive niche construction. In §§3–5, we explore that
idea in a variety of settings. The outcome is a clearer
understanding of how cultural transmission and embo-
died cognition generate the first image of human cogni-
tive systems identified above. That done, §§6 and 7
unpack the alternative (evolutionary-psychological)
picture by focusing on the interlocking notions of
cognitive modularity and an evolved human nature. In
§§8–11, we endeavour to resolve some of the tension
between our two visions, by examining how, and to what
extent, the notions of cognitive modularity and an
evolved human nature may be reconstructed within a
cognitive niche-construction framework. This brings

3563 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3564 M. Wheeler & A. Clark Culture, embodiment and genes
into focus what we, adapting the original usage by
Lewontin (2000), are dubbing triple helix models of
mind and cognition. These are models in which the goal
is to take seriously, and ultimately to understand, the
multiple ways in which three tangled sets of factors—
culture, embodiment and genes—combine to make us the
beings that we are.2
2. COGNITIVE NICHE CONSTRUCTION
Niche construction, as defined by Laland et al. (2000,
p. 131), refers to:
the activities, choices and metabolic processes of
organisms, through which they define, choose, modify
and partly create their own niches. For instance, to
varying degrees, organisms choose their own habitats,
mates, and resources and construct important com-
ponents of their local environments such as nests, holes,
burrows, paths, webs, dams, and chemical environments.
Niche construction is a pervasive, though still widely
underestimated, force in nature. All animals act on their
environments and, in so doing, alter those environments
in ways that may sometimes change the fitness land-
scape of the animal itself. A classic example3 is the
spider’s web. The existence of the web modifies the
sources of natural selection within the spider’s selective
niche, allowing (for example) subsequent selection for
web-based forms of camouflage and communication.
Still further complexity is introduced when organ-
isms collectively build structures that persist beyond
their own lifetime. A familiar example is the com-
munally constructed beaver’s dam, whose physical
presence subsequently alters selection pressures on
both the beaver and its progeny, who inherit the dam
and the altered river flows it has produced. Similar
effects can be seen in the nest-building activities of
many wasps and termites, where the presence of the
nest introduces selection pressures for behaviours that
regulate nest temperature by (for example) sealing
entrances at night (von Frisch 1975).
The cultural transmission of knowledge and prac-
tices resulting from individual lifetime learning, when
combined with the physical persistence of artefacts,
yields yet another source of potentially selection-
impacting feedback. The classic example here (from
Feldman & Cavalli Sforza 1989) is the practice of
domesticating cattle and dairying, which paved the way
for selection for adult lactose tolerance in (and only in)
those human populations engaging in such activities.
In all these cases, what ultimately matters, as Laland
et al. (2000) stress, is the way niche-construction activity
leads to new feedback cycles. In the standard cases, these
feedback cycles run across evolutionary time. Animals
change the world in ways that change the selective
landscapes for biological evolution. But it is worth
pointing out that this whole process has a direct analogue
within lifetime learning. Here, the feedback cycles alter
and transform processes of individual and cultural
reasoning and learning. For example, both educational
practices and human-built structures (artefacts) are
passed on from generation to generation in ways that
dramatically alter the fitness landscape for individual
lifetime learning. To adapt an example one of us has
Phil. Trans. R. Soc. B (2008)
used elsewhere (Clark 2001), the novice bartender
inherits an array of differently shaped glassware and
cocktail furniture, and a culturally transmitted
practice of serving different drinks in different kinds of
glass. As a result, expert bartenders learn to line up
differently shaped glasses in spatial sequence corre-
sponding to the temporal sequence of drinks orders
(Beach 1988). The problem of remembering what
drink to prepare next is thus transformed, as a result
of learning within this pre-structured niche, into the
problem of perceiving the different shapes and associ-
ating each shape with a kind of drink. The bartender,
by creating persisting spatially arrayed stand-ins for the
drinks orders, actively structures the local environ-
ment so as to press more usefulness from the basic
modes of visually cued action and recall. In this way, the
exploitation of the physical situation allows relatively
lightweight cognitive strategies to reap large rewards.
This is a simple illustration of the power of cognitive
niche construction, defined as the process by which
animals build physical structures that transform
problem spaces in ways that aid (or sometimes impede)
thinking and reasoning about some target domain or
domains.4 These physical structures combine with
appropriate culturally transmitted practices to trans-
form problem solving, and (in the most dramatic cases)
to make possible whole new forms of thought and
reason.5 Sections 3–5 of this paper explore the idea of
cognitive niche construction in a variety of settings.
3. THINKING SPACE
A vast amount of contemporary human cognitive niche
construction involves the active exploitation of space,
often by way of culturally inherited artefacts and
culturally transmitted strategies. Kirsh (1995) in his
classic treatment ‘The Intelligent Use of Space’ divides
these uses into three broad (and overlapping)
categories. The first is ‘spatial arrangements that
simplify choice’, such as laying out cooking ingredients
in the order you will need them, or putting your
shopping in one bag and mine in another. The second
is ‘spatial arrangements that simplify perception’, such
as putting the washed mushrooms on the right of the
chopping board and the unwashed ones on the left, or
the colour green dominated jigsaw puzzle pieces in one
pile and the red dominated ones in another. The third
is ‘spatial dynamics that simplify internal compu-
tation’, such as repeatedly reordering the scrabble
pieces so as to prompt better recall of candidate words,
or the use of instruments such as slide rules, which
transform arithmetical operations into perceptual
alignment activities.
It is noteworthy that the majority of these spatial
arrangement ploys work, as Kirsh himself notes at the
end of his treatment, by reducing the descriptive complexity
of the environment. Space is often used as a resource for
grouping items into equivalence classes for some
purpose (e.g. washed mushrooms, red jigsaw pieces,
my shopping and so on). Human language, perhaps the
ultimate cognitive tool (Clark 1997), is itself notable for
both its open-ended expressive power and its ability to
reduce the descriptive complexity of the environment.
Reduction of descriptive complexity, however achieved,
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culture, embodiment and genes M. Wheeler & A. Clark 3565
makes new groupings available for thought and action.
In this way, the intelligent use of space and the intelligent
use of language may form a mutually reinforcing pair,
pursuing a common cognitive agenda.
Developmental investigations lend some substance
to such a hypothesis. To take just one example, Namy
et al. (1997) conducted a series of experiments
involving children’s use of space to represent similarity.
Very briefly, what the experiments suggest is that
spatial groupings of play objects (such as putting all
the balls here, and all the boxes there) are not mere
spatially expressed reflections of fully achieved grasp of
category membership, but rather part and parcel of the
process of coming to learn about categories and to
discover the use of space as a means of representing
category membership. The process the investigators
document, in rich microgenetic detail, is one of
bootstrapping that starts with early play experiences
in which the child is interested in one kind of play
object and hence ends up (as a side effect) with those
objects grouped together in space. Such self-created
groupings help the child to discover the possibility and
the value of spatial classification itself. Crucial to this
discovery is the child’s engagement in preferential play
in which one type of object is preferred over another.
This kind of play was shown to lead, over relatively
short periods of developmental time, to the emergence
of true exhaustive classification behaviour, in which
spatial organization functions as a symbolic indicator
of category membership.
This whole process is one of incremental cognitive
self-stimulation within a partially self-constructed
cognitive niche. The perceptually available (grouped)
products of the child’s own activity form the new inputs
that favour learning about exhaustive classification and
(simultaneously) about the use of space as a means of
representing category membership. The capacities of
spontaneous spatial classification that this develop-
mental bootstrapping helps create may then further
scaffold the process of learning names and labels, while
the acquisition of new names and labels in turn
promotes the exploration of new and more sophis-
ticated spatial groupings.
4. EPISTEMIC ENGINEERS
Our second example of cognitive niche construction
emphasizes the transformative power of incrementally
organized and actively engineered epistemic resources in
the evolution and development of human cognition. To
bring this phenomenon into focus, it helps to introduce
the notion, due to Sterelny (2003), of cumulative
downstream epistemic engineering. Sterelny offers an
account of human uniqueness that gives pride of place
to our extraordinary capacities as ‘ecological engineers’,
that is to say, as the active constructors of our own
cognitive niches. Having earlier argued for group
selection as a key force in human evolution, Sterelny
notes that groups of humans engineer their own
habitats, and that these are transmitted to the next
generation, who further modify the habitat. Importantly,
some of these modifications are to the epistemic
environment, and affect the informational structures
and opportunities presented to each subsequent
Phil. Trans. R. Soc. B (2008)
generation. Although other animals clearly engage in
niche construction, it is only in the human species
(Sterelny argues) that we see this potent, cumulative,
runaway (self-fuelling) process of epistemic engineering.
Niche construction is depicted by Sterelny as a kind
of additional inheritance mechanism, working along-
side (and interacting with) genetic inheritance. One
of the points of interaction concerns phenotypic
plasticity. For rampant niche construction yields a
rapid succession of selective environments, and hence
favours the (biological) evolution of phenotypic
plasticity. Hominid minds, Sterelny suggests, are
adapted to the spread of variation itself. To cope with
such variability, we are said to have evolved powerful
forms of developmental plasticity. These allow early
learning to induce persisting and stable forms of neural
reorganization, impacting our range of automatic skills,
affective responses and generally reorganizing human
cognition in deep and profound ways. The upshot is
that ‘the same initial set of developmental resources
can differentiate into quite different final cognitive
products’ (Sterelny 2003, p. 166). In this way:
transforming hominid developmental environments
transformed hominid brains themselves. As hominids
remade their own worlds, they indirectly remade
themselves.
(Sterelny 2003, p. 173)
We see this explanatory template in action in, for
example, Sterelny’s account of our capacity to interpret
others as intentional agents. Thus:
Selection for interpretative skills could lead to a different
evolutionary trajectory: selection on parents (and via
group selection on the band as a whole) for actions
which scaffold the development of the interpretative
capacities. Selection rebuilds the epistemic environment
to scaffold the development of those capacities.
(Sterelny 2003, p. 221)
Basic perceptual adaptations, for example, gaze
monitoring, etc., are thus supposed to be bootstrapped
up to a full-blown ‘mind-reading’ ability via the
predictable effects of intense social scaffolding: the
child is surrounded by exemplars of mind-reading in
action; she is nudged by cultural inventions such as the
use of simplified narratives6 (and, ultimately, books
and pictures); prompted by parental rehearsal of her
own intentions; and provided with a rich palate of
linguistic tools such as words for mental states. Such
‘incremental environmental engineering’ provides, we
are told, a ‘wealth of the stimulus’ argument against the
innateness hypothesis (Sterelny 2003, p. 223). Our
theory of mind, according to this argument, is not
wired in at birth, but acquired by rich developmental
immersion. Such immersion may itself have ‘architec-
tural consequences’ (Sterelny 2003, p. 225), but these
are the upshot, not the precondition, of learning. This
explanatory strategy thus depicts much of what is most
distinctive in human cognition as rooted in the reliable
effects, on developmentally plastic brains, of immersion
in a well-engineered, cumulatively constructed cogni-
tive niche.
Sterelny’s emphasis is thus very much upon the
direct neural consequences of the culturally and
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3566 M. Wheeler & A. Clark Culture, embodiment and genes
artefactually scaffolded training regimes applied to
young human minds. But while such consequences are
surely of the utmost importance, they do not yet
exhaust the cognition-transforming effects of material
artefacts and culture. For many of the new cognitive
regimes supported by our best bouts of incremental
epistemic engineering seem to resist full internal-
ization. It is no use, as Ed Hutchins (personal
communication) points out, trying to imagine a slide
rule when you need to work out a log or cosine! Plastic
human brains may nonetheless learn to factor the
operation and information-bearing role of such
external props and artefacts deep into their own
problem-solving routines, creating hybrid cognitive
circuits that are themselves the physical mechanisms
underlying specific problem-solving performances. We
thus come to our final and arguably most radical take
on cognitive niche construction.
5. EXTENDED COGNITIVE SYSTEMS
Under certain conditions, non-organic props and aids,
many of which are either culturally inherited tools or
structures manipulated by culturally transmitted prac-
tices, might themselves count as proper parts of
extended cognitive processes (e.g. Clark & Chalmers
1998; Hurley 1998; Rowlands 1999; Wilson 2004;
Clark in press). Consider an accountant, Ada, who is
extremely good at dealing with long tables of figures.
Over the years, Ada has learnt how to solve specific
classes of accounting problems by rapidly scanning the
columns, copying some numbers onto a paper notepad,
then looking to and from those numbers (carefully
arrayed on the page) back to the columns of figures. This
is all now second nature to Ada, who scribbles at
lightning speed deploying a variety of ‘minimal memory
strategies’ (Ballard et al. 1997). Instead of attempting to
commit multiple complex numerical quantities and
dependencies to biological short-term memory, Ada
creates and follows trails through the scribbled numbers,
relying on self-created external traces every time an
intermediate result is obtained. These traces are visited
and revisited on a ‘just in time, need to know’ basis,
briefly shunting specific items of information into and
out of short-term organic memory, in much the same
way as a serial computer shifts information to and from
the central registers in the course of carrying out some
computation. This extended process may be best
analysed as a set of problem-solving state transitions
whose implementation happens to involve a distributed
combination of organic memory, motor actions, external
symbolic storage and just-in-time perceptual access.
Wilson’s (1994, 2004) notion of ‘wide computation’
captures the key features of such an extended approach.
According to wide computationalism, ‘at least some of
the computational systems that drive cognition reach
beyond the limits of the organismic boundary’ (Wilson
2004, p. 165). The larger systems thus constituted are,
Wilson insists, unified wholes such that ‘the resulting
mind–world computational system itself, and not just
the part of it inside the head, is genuinely cognitive’
(Wilson 2004, p. 167). Extended cognitive systems
theorists thus reject the image of mind as a kind
of input–output sandwich with cognition as the filling
Phil. Trans. R. Soc. B (2008)
(for this picture, and many more arguments for its
rejection, see Hurley 1998; see also Clark & Chalmers
1998; Wheeler 2005). Instead, we confront an image of
the local mechanisms of human cognition quite literally
bleeding out into body and world.
6. DARWINIAN MODULES
And now for something completely different—or so it
would seem. We have been mapping out an account of
ourselves in which the human brain is depicted as a
vortex of large-scale developmental and adaptive
plasticity, positioned in an ongoing and co-determining
interactive relationship with a dynamic flow of
culturally evolving non-neural elements. However,
what looks, on the face of things, to be a very different
vision of our evolved neural engine and of how it relates
to its cultural environment finds expression in the pages
of the evolutionary psychology literature. It is time to
scout that alternative vision.
Evolutionary psychology starts from the assumption
that just as there are anatomical adaptations (bodily
structures shaped by natural selection to solve certain
adaptive problems), so there are psychological adap-
tations (internal information processing mechanisms
shaped by natural selection to solve certain other
adaptive problems). As Cosmides & Tooby (1987,
p. 282) put it, ‘[the] evolutionary function of the
human brain is to process information in ways that lead
to adaptive behavior’. Evolutionary psychologists argue
that it follows from this ‘Darwinized’ conception of
information processing psychology that our innate
cognitive endowment, as shared by all developmentally
normal human beings, is not a domain-general learning
and reasoning engine (as many social scientists and
others have claimed), but rather (to use a now famous
image) a psychological Swiss army knife, in that it
comprises a large collection of specialized cognitive
tools. This collection of tools is depicted as a suite of
genetically specified, domain-specific computational
mechanisms, often called modules, each of which (i) is
triggered by informational inputs specific to a parti-
cular evolutionarily salient domain (e.g. choosing a mate
and social exchange) and (ii) has access to internally
stored information about that domain alone. Thus, the
Swiss army knife account of mind is sometimes glossed
as the massive modularity hypothesis (Sperber 1996;
Samuels 1998).7
Two immediate clarifications of this picture are in
order. First, it is important to note a distinguishing
feature of the tabled approach to modularity. Accor-
ding to the evolutionary-psychological picture, the
modules that comprise our innate cognitive endow-
ment are to be demarcated at a functional level of
analysis, an implication of which is that they need not
be realized in localized regions of neural hardware
(Gaulin & McBurney 2001). Secondly, evolutionary
psychologists argue that in order to give an account of
our adapted cognitive modules, one needs to identify
the appropriate selective environment. This is a local
application of a general principle. When one attempts
to explain adaptation, one needs to have in view
the ‘composite of environmental properties of the
most recent segment of a species’ evolution that
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culture, embodiment and genes M. Wheeler & A. Clark 3567
encompasses the period during which its modern
collection of adaptations assumed their present form’
(Tooby & Cosmides 1990, p. 388). This crucial slice of
selective history is what evolutionary psychologists call
a trait’s environment of evolutionary adaptedness (EEA).
Of course, the relevant EEA may well not be the
current environment in which a trait operates.
Environments sometimes change, and evolution by
cumulative Darwinian selection is typically thought of
as a rather slow process that may lag well behind such
change. This is especially probable in the case of a trait
as complex as the human brain, embedded in an
environment rich in historically unfolding cultural
dynamics. Applying this logic, evolutionary psycho-
logists typically argue that the last time any significant
modifications were made by selection to the human
brain’s functional architecture was during the Pleisto-
cene epoch (ca 2 Myr to 10 kyr ago), when humans
were hunter-gatherers. So the composite of selection
pressures at work in the Pleistocene constitutes our
brain’s EEA (see Crawford 1998 for discussion). This
is where one finds the adaptive problems to which the
modules housed by the modern brain—modules which
have been inherited essentially unchanged from our
Pleistocene hunter-gatherer ancestors—constitute
evolved solutions.
Although the identification of the human EEA with
the hunter-gatherer Pleistocene environment is an idea
that has attracted a good deal of critical fire (e.g. Gould
2000; Smith et al. 2001), it does help the evolutionary
psychologist to account for the fact that some of our
behaviour fails to maximize fitness in modern cultural
environments. For example, modern human males do
not adopt a fitness-enhancing strategy of widespread
sperm donation because our reproductive strategies are
designed for Pleistocene conditions. And the fitness-
decreasing obesity brought about by an overindulgence
in sugar-rich foods in technologically advanced
countries may be explained by the fact that our sweet
tooth, which was adaptive in the nutritional challenges
posed by the Pleistocene, has since been rendered
maladaptive in such countries by the mass availability
of refined sugar.
This image of a species-wide assemblage of evolved
domain-specific information processing mechanisms,
meshed with ancestral environmental factors, provides
the background to a further aspect of the overall
evolutionary-psychological picture that will be import-
ant in what follows. Evolutionary psychologists claim
that behind all the manifest diversity in human cultural
behaviour, there sits an evolved universal human nature.
In what, then, does this evolved universal human
nature consist, and how, given its alleged species-wide
homogeneity, does it generate that remarkable diversity
in cultural behaviour?
7. HUMAN NATURE
From what we have seen so far, it might seem that the
evolutionary-psychological notion of an evolved
universal human nature will be cashed out in terms
of a suite of Darwinian modules possessed by all
developmentally normal adult human beings. However,
we need to be careful in how we handle this idea because
Phil. Trans. R. Soc. B (2008)
the fact is that that suite of modules, even as portrayed in
evolutionary psychology, is not strictly universal. For
example, whether or not a particular psychological
adaptation is ultimately ‘wired up’ in a certain way in a
specific individual will typically depend on the presence
of certain environmental triggers that, under normal
circumstances, occur reliably at critical stages during
development. (For a dramatic example, consider the
need for a rich linguistic environment to be present
during language development.) Moreover, there may be
alternative psychological adaptations available to
development that are under the control of genetic
switches (roughly, mechanisms by which genes are
turned on or off through the absence or presence of
DNA-binding proteins). Indeed, evolutionary psychol-
ogists argue that men and women confront divergent,
sex-relative adaptive problems when it comes to finding,
holding onto and reproducing with a mate. Thus, men
and women instantiate different, sex-relative psycho-
logical adaptations in the mating game. Since sex
determination is under the control of a genetic switch,
so are these alternative psychological architectures.
What the existence of such alternative develop-
mental trajectories demonstrates is that the suite of
cognitive modules possessed by humankind is not
strictly universal and so cannot constitute our species-
wide human nature. What might then? The answer,
nicely isolated by Buller (2005), is an evolved species-
wide set of genetically specified developmental pro-
grams that (i) determine how the emerging human
phenotype responds to critical environmental triggers
and (ii) control processes such as genetic switching. It is
at that level that strict universality (allegedly) holds,
and at which our evolved human nature is (allegedly) to
be found.
Now, if all developmentally normal human beings
share a set of genetically specified developmental
programs and, as a result, at least a very large number
of innately specified psychological adaptations meshed
with ancestral environments, what explains the varia-
bility of human behaviour across contemporary
cultures? Here, we can draw a lesson from the example
of ordinary digital computer programs. As in such
programs, our cognitive information processing
modules may respond differentially to variations in
the inputs that they receive, inputs that are supplied
largely by the particular cultural environments in which
the bearers of those modules are embedded. A
developmental version of this process is equally
important. In certain cases, a particular innately
specified module (e.g. a Chomskyan language acqui-
sition device) may be exposed to different develop-
mental environments (different linguistic communities
providing different developmental inputs), leading
ultimately to cognitive variation (different speakers
learning and producing different languages).
Our second vision has now emerged fully. It is at root a
vision of the evolved human brain as a locus of relatively
static, genetically based forms of neural encoding and
processing, executing a restricted set of pre-specified
adapted functions in response to the triggers provided by
variable cultural inputs. This certainly seems to suggest
a very different view of what it is to be a natural human
thinker from the one evoked by our synthesis of
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3568 M. Wheeler & A. Clark Culture, embodiment and genes
embodied and extended cognition, cultural evolution
and cognitive niche construction. But just how much of
an intellectual chasm really exists between these
apparently divergent views? In other words, along
which dimensions, and to what extent, are our two
visions in genuine competition with each other? It is to
this issue that we shall now turn.
8. REMOULDING MODULARITY
What seems clear is that there is no necessary tension
between, on one hand, an approach that foregrounds
cultural evolution and, on the other, the kind of
cognitive modularity favoured by the evolutionary
psychologists. This might seem an odd claim to make
at first, given that the fans of cultural evolution often
place an emphasis on psychological mechanisms that
exhibit a robust kind of domain generality. For
example, drawing on Boyd & Richerson’s (1985) dual
inheritance model, a model that (as Sterelny’s
approach sketched earlier) stresses cultural as well as
genetic transmission in evolution, Coultas (2004)
provides experimental evidence that individual
human beings have an essentially domain-general
tendency to conform in social groups, a tendency that
can be adaptive for the individual when information
gathering by that individual would be costly. And
Tomasello (1999), in a treatment that also stresses dual
inheritance, argues that evolution has endowed us with
a set of basic cognitive capacities, including shared
attention and the imitation of other humans’
behaviours and intentions, that allow us to take
developmental advantage of a kind of accumulated
species-specific knowledge made available through
human cultural environments. At the heart of this
process, and the capacity that sets human beings apart
from other species, is our ability to identify intentions
in others. It is this uniquely human, essentially domain-
general ability, argues Tomasello, that allows us to
build on foundational capacities that we share with
other animals (such as the capacities for tool use and
signalling), in order to become vastly more sophis-
ticated thinkers in specific domains (e.g. vastly more
sophisticated tool users and signallers) than have our
evolutionary cousins. Finally, as we have seen already,
Sterelny (2003) offers an account of our capacity to
interpret others as intentional agents, according to
which basic perceptual adaptations are bootstrapped
up to a full-blown mind-reading ability via cognitive
niche construction. This contrasts sharply with the
evolutionary-psychological idea of an innate ‘folk
psychology’ module, in the form of a domain-specific
adaptation for mind-reading.
That said, Atran (2001, p. 8) presents an alternative
view of the relationship between cultural transmission
and cognitive modularity in which the latter underlies
the former, with certain modules serving as ‘as a
principled basis for transmission and acquisition of
more variable and extended forms of cultural know-
ledge’. For example, he argues that the widespread
anthropological phenomenon of totemism—religious
systems in which generic species spiritually represent
social groups (e.g. an animal that spiritually represents
a clan)—piggybacks on a genetically specified folk
Phil. Trans. R. Soc. B (2008)
biology module. That module latches onto generic
species (and groups of generic species) whose intrinsi-
cally well-structured character renders them apt for
memorability and cultural transmission between
minds. These underlying categories supply cognitive
hooks onto which our minds subsequently hang beliefs
about intrinsically less well-structured social groups. In
sum, according to Atran (2001, p. 8):
modularized structures—such as those which produce
folkmechanical, folkpsychological and folkbiological
concepts—are special players in cultural evolution.
Their native stability derivatively attaches to more
variable and difficult-to-learn representational forms,
thus enhancing the latter’s prospects for regularity and
recurrence in transmission within and across cultures.
The availability of these alternative positions within
the evolution-of-cognition research programme
suggests strongly that one cannot infer that a cognitive
architecture will be non-modular, or indeed that it will
be modular, simply from the existence or otherwise of
cultural transmission in the inheritance system.
Cognitive modularity is also compatible with the
other partner in our symbiotic dyad, an embodied–
extended approach to mind. A powerful illustration of
how an embodied–extended modularity might go is
provided by the field of situated robotics (e.g. Brooks
1991; Mataric 1991; Pfeifer & Bongard 2007). With
the goal of building complete agents that are capable of
integrating perception and action in real time so as to
generate fast and fluid embodied adaptive behaviour,
researchers in situated robotics shun the classical
cognitive-scientific reliance on detailed internal world
models, on the grounds that such structures are
computationally expensive to build and keep up to
date. Instead they adopt a design strategy according
to which the robot regularly senses its environment to
guide its action. It is this specific behaviour-generating
strategy that marks out a robot as situated (Brooks
1991). Against this background, one of the key ideas
from the field is that much of the richness and flexibility
of intelligence is down not to general-purpose processes
of reasoning and inference, but rather to integrated
suites of special-purpose adaptive couplings that realize
distributed or extended behaviour-generating strategies
by combining non-trivial causal contributions from
three constituencies: the brain (or its robotic
equivalent); the non-neural body; and the environ-
ment. Moreover, this perspective provides one plat-
form for the previously mentioned refusal to
conceptualize perception and action as interfaces
between mind and world. As Brooks (1991, p. 173)
puts it, one of the guiding principles of the approach is
that: ‘There is no separation into perceptual system,
central system, and actuation system. Pieces of the
network [the distributed robotic control system] may
perform more than one of these functions. More
importantly, there is intimate intertwining of aspects
of all three of them.’
A classic example of such work is provided by Maja
Mataric’s sonar-driven mobile robot, Toto (Mataric
1991). Toto wanders around its office environment
following walls and avoiding obstacles. As it proceeds,
it constructs an internal map based on landmarks,
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culture, embodiment and genes M. Wheeler & A. Clark 3569
which then enables it to navigate between locations.
Toto is controlled by three main layers of situated
special-purpose adaptive coupling: collision-free wan-
dering; landmark detection; and map learning and path
planning. What is theoretically interesting about Toto’s
map-learning and path-planning system is that naviga-
tion-related information is encoded in it in terms of
patterns of embodied sensorimotor activity. For
example, if, as Toto moves, it keeps detecting
proximally located objects on its right-hand side,
while its compass bearing remains unchanged, then a
‘right wall’ is encoded in its inner map, not as some
agent-independent objectively specified entity, but in
terms of its sensorimotor ‘experience’ at the time.
These structured sensorimotor experiences (Toto’s
landmarks) are stored as connected nodes in a
distributed graph, and this record of the robot’s own
embodied sensorimotor history constitutes its inner
map of the spatial environment.
Crucially, given our interests, Toto’s strategy of
encoding spatial paths as internally represented
sequences of past, current and expected embodied
sensorimotor experiences is a domain-specific solution,
one tailored to the particular navigational context for
which the robot is designed. The action-oriented
structures in question presumably would not be much
good for a vast range of other space-related purposes,
such as ordering correctly sized carpets for the
corridors or determining the precise distance to the
snack bar. Moreover, the navigation system is informa-
tionally encapsulated, in just the way required by the
modularity hypothesis. (Of course, the map-learning
and path-planning system depends on the successful
functioning of the other layers of coupling, but
informational encapsulation does not rule out such
inter-systemic dependencies.) What all these suggest is
that the sorts of situated special-purpose adaptive
couplings promoted within situated robotics are
illuminatingly understood as cognitive modules. Cru-
cially, however, these modules have (what we might
call) a horizontally extended character, in that their
functional boundaries are no longer constrained by the
orthodox transitions that remain in force in mainstream
evolutionary psychology, between (i) perception and
thought (in the world-to-body-to-mind input direc-
tion) and (ii) thought and action (in the mind-to-body-
to-world output direction).
To develop further this notion of horizontally
extended cognitive modularity, consider Ziemke
et al.’s (2004) coevolutionary experiment involving
two sets of simulated robots—scouts and drones—
whose cooperation-demanding task is to enable the
drones to find a spatially located goal. Both sets of
agents are controlled by simple fixed topology neural
networks under artificial evolutionary control. The task
is posed in a grey-walled environment, in which each
junction requiring a left turn to reach the goal is marked
with a white stripe, while each junction requiring a right
turn is marked with a black stripe. Scouts have cameras
and so, in principle, can find their way to the goal
autonomously using the turn-signalling stripes. By
contrast, the drones have no cameras, only light
sensors, so they cannot see the stripes. Their only
hope, beyond random search, is to evolve to respond
Phil. Trans. R. Soc. B (2008)
correctly to light sources that are deposited by the
scouts as the latter traverse the environment. So the
scouts need to evolve a cognitive niche-construction
strategy, one in which they place the light sources in
such a way that they produce an increase in (what
Ziemke et al. call) the cognitive congeniality of the
environment inherited by the drones.8
Under the experimental conditions described,
scouts evolve to drop light sources in response to the
white stripe on the wall—thereby constructing a niche
that simplifies the problem task for the drones—and
drones evolve to exploit these ‘road signs’, by turning
left while sensing the light, but right at the other
junctions. This niche-construction scenario once
again displays the distinctive hallmarks of situated
special-purpose adaptive coupling (e.g. tight linkages
between particular embodied sensorimotor capacities
and task-dedicated action-generating strategies that
factor in the reliable presence of specific environ-
mental structures) and thereby of horizontally
extended modularity.9
In spite of these positive steps towards a reconcilia-
tion between an approach that emphasizes cognitive
modularity and one that emphasizes cultural trans-
mission and the embodied–extended mind, an import-
ant issue remains to be addressed. As we have seen,
evolutionary psychologists explain the development of
cognitive modules in terms of a species-wide set of
genetically specified developmental programs that
orchestrate the journey from genotype to phenotype,
and in particular from genes to massive modularity.
This genocentric stance might seem to clash unhelp-
fully with an account of development that routinely
appeals to the bootstrapping up of basic capacities via
cultural transmission and cognitive niche construction,
and which thereby shifts the centre of explanatory
gravity away from genetic specification and towards a
distributed matrix of co-determining genetic and
environmental factors. Even here, however, there is
some hope that the tension may be relieved, if we
combine the thought that progressive modularization
may emerge during development and learning
(e.g. Karmiloff-Smith 1992), with an account of the
conditions under which, within the sort of distributed
developmental matrix just highlighted, genes may
rightly be said to code for phenotypic traits (Wheeler &
Clark 1999; Wheeler 2003). Each of these ideas
warrants discussion.
9. EMERGENT MODULARITY
Karmiloff-Smith (1992) provides a compelling account
of how, given the plasticity of early neural development,
a progressive functional modularization may be
realized by the brain as part of the developmental
process. Evidence from cases of early brain damage
indicate a degree of baseline neural plasticity that goes
well beyond that suggested by the evolutionary-
psychological image of a set of genetically specified
modules installed in response to environmental
triggers. The mind is not pre-structured at birth to be
modular. Instead, a process of modularization is kick-
started by a limited range of multilevel domain-specific
predispositions that focus the young infant’s attention
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3570 M. Wheeler & A. Clark Culture, embodiment and genes
on certain proprietary inputs. The progressive develop-
ment of emergent modular structures then proceeds
interactively as these proprietary inputs in turn affect
the development of the brain.
A rich example of how functional modularization
may be the outcome of constrained dynamic
interaction during development is provided by
Hirsh-Pasek & Golinkoff ’s (1996) three-phase
coalition model of language comprehension (see also
Hollich et al. 2000). According to this model, infants
in the first phase build on rudimentary language
comprehension achieved during the second half of the
first year of life to perform an initial segmentation of
the flux of their acoustic and visual environments. On
the basis of dispositions to note certain acoustic and
visual cues, alongside a capacity for distributional and
correlational analysis across phonological and rhyth-
mic patterns of speech, the infant’s task is to parcel up
the flow of speech around her into acoustic units that
will later become linguistically relevant, and to use
these acoustic units to help her uncover highly
significant structures in her environment (e.g. import-
ant events and objects). The second phase involves the
interpretation of the acoustic units as components that
correlate with linguistic categories (such as subject,
verb and object), plus the mapping of individual word
units onto their referents. In this way, semantics
dislodges sound as the primary regulator of emerging
language comprehension. Although during this phase
children are beginning to comprehend multiword
sentences and the role of word order in determining
grammatical relations, such advanced comprehension
is fragile, in that it depends on all the relevant social,
semantic and syntactic cues being present. Thus, a
supporting coalition of environmental factors forms a
developmentally crucial cognitive scaffold. In the third
phase, this dependency is overcome. The child’s
syntactic system becomes fully established, as indi-
cated by the late onset ability to understand linguistic
constructions that violate word-order assumptions
(e.g. the English passive).
For Hirsh-Pasek & Golinkoff, then, language com-
prehension is kick-started by a system that is primed
with dispositions to note salient inputs and their
likelihood of occurring together. Functional modular-
ization, in the form of a domain-specific, information-
ally encapsulated system for language comprehension,
develops progressively through interactive environ-
mental engagement. So cognitive modularity may
result from distributed developmental bootstrapping
that potentially involves cultural transmission and
cognitive niche construction. There seems to be no
reason to think that there could not be a large number of
such modules, so in that sense at least the human
cognitive system may be a locus of massive emergent
modularity—an emergent cognitive Swiss army knife!
But now what about the evolutionary-psychological
claim that cognitive modules are genetically specified?
One might think of this as a key component of the
evolutionary-psychological vision. What remains of
this claim in the alternative story? The answer, we
suggest, is: rather more than you might expect.
Phil. Trans. R. Soc. B (2008)
10. GENES, CODES AND EXPLANATORY SPREAD
There is a generic phenomenon that the present
authors once dubbed explanatory spread ( Wheeler &
Clark 1999). Mameli (2005, p. 388) gives a clear
exposition of what it entails.
Causal spread occurs when we discover some new
factor causally involved in the occurrence of a
phenomenon. Explanatory spread occurs when we
realize that some factor that was not considered to be
necessary in the explanation of a phenomenon is
instead explanatorily necessary for that phenomenon.
Or, to put it differently, explanatory spread occurs
when we realize that some factor that was not taken to
be part of a sufficient explanation of a phenomenon
needs to be included in such explanation. Since the
fact that something is causally required does not
entail that it is also explanatorily required, causal
spread does not necessarily lead to explanatory
spread. But in cases where the newly discovered
causal factor is deemed to be an important one, causal
spread is likely to generate the inclusion of the newly
discovered factor in any sufficient explanation of a
phenomenon to which this factor causally contrib-
utes. That is, in these cases, causal spread leads to
explanatory spread.
Where the phenomenon of interest is phenotypic
form, the received position is that such structure is
down to genetic specification. So one would have
explanatory spread where one discovered a distributed
developmental system in which non-genetic organis-
mic and/or wider environmental factors made expla-
natorily non-negligible contributions to phenotypic
form. That is the general picture on offer from
approaches that emphasize cultural evolution, cogni-
tive niche construction and (we can now add)
emergent modularity.
So what? Crucially, some authors have argued that a
proper recognition of developmental explanatory
spread should lead us to reject the claim that genes
specify phenotypic traits. Cognitive modules are, of
course, examples of phenotypic traits, so if this anti-
specification argument is sound, it would undermine
the claim that such modules are genetically specified,
and so re-establish a conflict between our two visions.
But is that argument sound? To answer that question,
let us consider a specific statement of it:
We have often heard it said that genes contain the
‘information’ that specifies a living being. [but] when
we say that DNA contains what is necessary to specify a
living being, we divest these components. of their
interrelation with the rest of the network. It is the
network of interactions in its entirety that constitutes
and specifies the characteristics of a particular cell, and
not one of its components. That modifications in the
components called genes dramatically affect the
structure is very certain. The error lies in confusing
essential participation with unique responsibility. By
the same token one could say that the political
constitution of a country determines its history. This
is obviously absurd. The political constitution is an
essential component in any history but it does not
contain the ‘information’ that specifies that history.
(Maturana & Varela 1987, p. 69)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culture, embodiment and genes M. Wheeler & A. Clark 3571
What is going on here? The first thing to note (as the
opening sentence of the above passage indicates) is that
to conceive of genes as trait specifiers is to conceive of
genes as developmental information carriers, i.e. as
coding for phenotypic traits. Thus, much here turns on
how one understands the nature of that coding
relationship. It seems to us that Maturana and Varela’s
argument depends implicitly on a deceptively tempt-
ing, but ultimately flawed, view of coding talk that we
call strong instructionism (Wheeler & Clark 1999; see
also Wheeler 2003, 2006). Strong instructionism is
the claim that what it means for some element to code
for an outcome is for that element to fully specify the
distinctive features of that outcome, where ‘full
specification’ requires that those distinctive features
may be predicted purely on the basis of what may be
known about the putatively coding factor. In the
present context, strong instructionism amounts to the
claim that what it means for a gene (or a complex of
genes) to code for a phenotypic trait is for that gene (or
complex of genes) to fully specify the form of that trait.
It is this kind of picture that is seemingly suggested by
the classic Lorenzian image of the non-genetic material
causes in development as the bricks and mortar out of
which the organism is assembled according to a genetic
blueprint (Lorenz 1965). However, given the presence
of developmental explanatory spread (what Maturana
and Varela call ‘the network of interactions in its
entirety’), the fact is that knowing the entire sequence
of an organism’s DNA will not be sufficient to predict
phenotypic form. It is this point that underwrites
Maturana and Varela’s observation that the fan of
genetic information mistakenly confuses ‘essential
participation with unique responsibility’. So, if the
understanding of genes as coding for phenotypic traits
is tied to strong instructionism, then, given develop-
mental explanatory spread, that understanding is false.10
The trick, then, is to free coding talk about genes
from strong instructionism. Fortunately, there is plenty
of evidence that coding talk in other domains does not
impose the full-specification condition. Indeed, in
familiar cases of algorithms, programs, instruction
sets and other such coding elements, those states and
processes are able to perform their outcome-generating
functions only given some assumed backdrop of other
causally active states and processes (e.g. working
operating systems) that themselves bear some of the
responsibility for the exact form of the outcome
produced. In other words, strong instructionism is a
spectre without much of a haunting pedigree. That
said, a word of warning: we need to avoid falling into
the opposite trap of giving an account of genetic coding
so excessively liberal, that where explanatory spread is
present, too many developmental factors qualify as
coding for phenotypic outcomes. For then the claim
that a certain gene (or complex of genes) codes for
some trait will simply fail to single out that gene (or
complex of genes) as performing a distinctive develop-
mental function.11
To take just one example (for several others, see
Wheeler 2006), say we adopted the superficially
attractive view that genes code for traits insofar as
they are what is passed on from one generation to the
next in evolution. If we define inheritance without an
Phil. Trans. R. Soc. B (2008)
antecedent pro-gene prejudice, as the biological like-
begets-like phenomenon, and so as to fix on elements
that are robustly and reliably replicated in each
generation of a lineage, and that persist long enough
to be the target of cumulative selection, then the fact
seems to be that genes are not all that organisms inherit.
For example, there are the so-called epigenetic inheri-
tance systems, such as the inheritance of methylation
patterns via a separate (i.e. from the genetic) copying
system; and there is inheritance through host imprinting,
as when parasitic birds, born in the nest of a host
species, imprint on that nest as chicks, and then later
lay their own eggs in the nest of that species; and then
there is inheritance via our old friend niche construction,
as when beaver offspring inherit both the dam that was
communally constructed by the previous generation
and the altered river flow that that physical structure
has produced. What this indicates is that if being
inherited is sufficient for some developmental factor to
qualify as coding for a phenotypic trait, then non-
genetic factors will regularly count as coding elements,
which violates our excessive liberality constraint.
There is, of course, much more to be said about this
issue. In the present treatment, we have done little
more than sketch the form that an account of coding
talk would have to take, if it is to allow genes to code for
(and thus, in a robust sense, specify) phenotypic traits
(including cognitive modules), even in the midst of an
explanatory spread that involved cultural transmission
and cognitive niche construction. But, if we can
successfully navigate between the Scylla of strong
instructionism and the Charybdis of excessive liberal-
ity, we would potentially have access to such an
account. Allied with the concept of emergent mod-
ularity, that result would do much to effect a
rapprochement between our alternative visions of
evolved human cognition.12
11. HUMAN NATURE RECONSIDERED
It is time to revisit the evolutionary psychologist’s notion
of an evolved universal human nature, conceived as a
species-wide set of genetically specified developmental
programs that orchestrate the journey from genotype to
phenotype. According to this view, a maturing human
being, embedded in a normal developmental environ-
ment, will end up with a particular, species-wide set of
cognitive modules (allowing for some branching
pathways, e.g. between the sexes). Significant challenges
to this view are posed by the powerful role assigned, by
cognitive niche-construction models, to stacked
sequences of training environments in the emergence
of specific functional modules.13 While the early stages
of such key developmental trajectories may, as we saw,
be rather predictably determined by small native biases,
the later stages often reflect both the cumulative effects
of cultural evolution and transmission, and the potent
effects of the ongoing self-selection of training environ-
ments. A child whose early experience is shaped by the
special environments provided by books and software
programs, and whose own emerging cognitive profile
favours certain elements within that culturally enabled
nexus over other elements, will end up with a cognitive
system that is not just superficially, but profoundly,
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3572 M. Wheeler & A. Clark Culture, embodiment and genes
different from that of a differently encultured child. Such
a view finds expression in, for example, Schlesinger &
Parisi’s (2007, p. 153) notion of an emergent constraint
according to which:
the outcome of a developmental process need not be
programmed in by maturation but instead may occur as
the result of successive learning experiences that the
organism determines or selects for itself.
The neuroroboticist Olaf Sporns describes the larger
situation well, noting that:
[the] architecture of the brain. and the statistics of the
environment [are] not fixed. Rather, brain-connectivity
is subject to a broad spectrum of input-, experience-,
and activity-dependent processes which shape and
structure its patterning and strengths ( Johnson 2001).
These changes, in turn, result in altered interactions
with the environment, exerting causal influences on
what is experienced and sensed in the future.
(Sporns 2007, p. 179)
This kind of ‘neuroconstructivist’ framework (for a
compelling array of worked examples, see Mareschal et al.
2007a,b) helps locate a potential challenge for any notion
of an evolved human nature that ties that nature too
closely to the properties and features of the EEA. For
what is special about human brains, and what best
explains the distinctive features of human intelligence,
may be precisely their ability (courtesy of extended
development and extensive neural plasticity) to enter into
deep, complex and ultimately architecture-determining
relationships with an open-ended variety of culturally
transmitted practices, endowments and non-biological
constructs, props and aids. Perhaps it is because our
brains, more than those of any other animal on the planet,
are primed to seek and consummate such intimate
relations with non-biological resources that we end up
as bright and as capable of abstract thought as we are. If
so, our distinctive universal human nature, insofar as it
exists at all, would rather be a nature of biologically
determined openness to deep, learning- and develop-
ment-mediated, change.
It is at this point that we locate a potential challenge
to the evolutionary psychologists’ specific vision of a
universal human nature. For that vision, as we saw
earlier, commits them to a restricted range of potential
cognitive modules, with that range determined by a
suite of genetically specified developmental programs.
As a result, the range of possible normal variation
among cognitive modules is strictly and endogenously
limited. By contrast, the constructivist vision of
horizontally extended and emergent cognitive modules
places no such clean limits upon the range of variation.
Insofar as there is something worth calling a universal
human nature on this alternative view, that nature lies
precisely in our continual openness to radical cognitive
change. Our fixed nature is thus a kind of meta-nature:
the suite of capacities, practices and proclivities that
enable the development, use and propagation of a
much more open-ended set of horizontally extended
and emergent cognitive modules.
Such openness, as stressed by recent works on
embodied and extended cognition, adds important
complexity to accounts that emphasize the EEA. For
Phil. Trans. R. Soc. B (2008)
we must now take into account a plastic evolutionary
overlay that yields a constantly moving target, an
extended cognitive architecture whose constancy lies
mainly in its continual openness to change. Even
granting that the biological innovations that got
this ball rolling may have consisted only in some
small tweaks to an ancestral repertoire, the upshot
of this subtle alteration would be a sudden, massive
leap in cognitive-architectural space: the emergence
of a cognitive machine intrinsically geared to self-
transformation, artefact-based expansion and a snow-
balling/bootstrapping process of computational and
representational growth. The machinery of human
reason (the environmentally extended apparatus of our
distinctively human intelligence) could thus turn out to
be rooted in a biologically incremental progression
while simultaneously existing on the far side of a
precipitous cliff in cognitive-architectural space.
12. CONCLUSIONS: THE SPACE BETWEEN
Such, at least, would be the most radical model, one
that indeed locates some genuine tension between
the evolutionary psychologist’s emphasis on hard
modules and the EEA, and the cognitive niche
constructivist emphasis on emergent modularity as
reflecting the complex ratchet effects made available
by the interplay of neural plasticity, learning and
embodied activity involving inherited or self-created
environmental structure.
But between these poles of human nature as highly
reflective of the specific features of the EEA, and
human nature as one of extensive openness to training
and input-based modification, lies the full and inviting
cognitive space structured by the triple helix of culture,
embodiment and genes. Triple helix models of mind
recognize the role of genetic biases in sculpting key
developmental trajectories, and the resulting space
both for strong forms of genetically specified cognitive
modularity and for weaker forms of emergent mod-
ularity resulting from trajectories marked by multiple
bouts of culturally scaffolded experience and the self-
selection of environments. But crucially, the triple helix
template also invites us to consider, pretty much on a
case-by-case basis, all points and stations in between.
Understanding this spectrum, and unravelling the
complex interplay between genes, environments and
embodied action, will surely be one of the great
intellectual adventures of the 21st century.
This paper was prepared in part thanks to support granted to
A.C. by the AHRC, under the ESF Eurocores CNCC
scheme, as part of the CONTACT (Consciousness in
Interaction) project AH/E511139/1, and to M.W. by the
AHRC as part of project AH/F002963/1. Some sections have
been adapted from Clark (2003, in press, ch. 4) and Wheeler
(2006, in press). Many thanks to John Protevi, Kenny Smith
and John Sutton for their constructive critical feedback on
an earlier version of this paper.
ENDNOTES
1
In line with much contemporary usage, we shall take the term
‘evolutionary psychology’ to signal not simply any psychological
science that takes its cues from evolutionary biology, but rather a
specific research paradigm centred on the work of Cosmides & Tooby
(1987), Buss (1994) and Pinker (1997), among others.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culture, embodiment and genes M. Wheeler & A. Clark 3573
2
The idea of a triple helix in evolution was originally developed by
Richard Lewontin (2000), who identified its components as genes,
organism and environment. Our usage makes contact with
Lewontin’s own, but adapts the latter two components so as to
focus on the especially potent and intriguing dimensions provided by
embodiment and culture.
3
For a host of other examples, see Laland et al. (2000) and
Odling-Smee et al. (2003). See also Dawkins (1982), Lewontin
(1983) and Turner (2000).
4
The basic idea of human beings as cognitive niche constructors is
familiar within cognitive science. Richard Gregory (1981) spoke of
‘cognition amplifiers’, Don Norman (1993) of ‘things that make us
smart’, Kirsh & Maglio (1994) of ‘epistemic actions’ and Daniel
Dennett (1996) of ‘tools for thought’.
5 Wheeler (2006).
It is worth noting that nothing in this view commits us to the notion
of a single ‘abstract’ human subject rather than a population of
subjects with different traits and nuances. Instead, it is best to think of
a range of subjects displaying, as a result of genetic, cultural and
environmental influences, a spread of different traits and capacities.
For each such trait and capacity, taken in its local context, there will
be a correlated pattern of empowerment and constraint. The most
successful human groups will then be those in which the spread itself
(which will include differences in affect and affective response) is
mutually beneficial. Thanks to John Protevi (personal communi-
cation) for drawing these issues to our attention.
6
For a compelling analysis of how involvement in a particular kind of
narrative practice may explain the developmental path to an
understanding of other minds, an understanding which itself turns
on the construction of narratives, see Hutto (2008).
7
Here, we shall not be concerned with assessing the positive
conceptual arguments or the experimental data that are supposed
to take us from the Darwinization of information processing
psychology to the massive modularity of the adapted mind. In
general, the conceptual arguments turn on the thought that domain-
general mechanisms in isolation, i.e. without assistance from domain-
specific mechanisms, would not be able to solve the adaptive
problems confronted by the brain, or at least that any domain-
general mechanism in the evolving population will typically have been
systematically outperformed by any competing domain-specific
mechanisms, such that it is the latter kind of mechanism that will
have been selected for. For critical discussion of the arguments and
evidence here that typically finds them wanting, see, for example,
Samuels (1998), Sterelny & Griffiths (1999), Atkinson & Wheeler
(2004) and Buller (2005).
8
The fitness scores that determine the survival and reproduction
prospects in the evolutionary scenario are calculated as follows: in
each trial, an individual scout is rewarded (i) for finding the goal itself
and (ii) if an associated follower drone also reaches the goal, while
each drone is rewarded (iii) for finding the goal itself and (iv) if an
associated leader scout also reaches the goal. Thus, it is the
achievement of the goal state that is rewarded directly, and not the
specific strategies for reaching that state.
9
One might see the drones as constituting a limit case in which the entire
control system implements a single functionally identified module.
10
It might seem that strong instructionism about genes is a straw
position that no one seriously holds. However, the fact is that the idea
remains insidiously at work behind commonplace metaphors for
understanding the relationship between genes and traits. As John
Dupre, a philosopher of biology and the director of a centre for research
on genomics in society (Egenis, http://www.genomicsnetwork.ac.uk/
egenis), comments: ‘It is still common to hear the genome described,
for instance, even by eminent experts, as a blueprint for the organism.
Perhaps not many people will defend the blueprint metaphor very far
these days, if pushed, however. A common retreat is to the metaphor of
a recipe. But this metaphor is still quite inadequate. With due
allowance for an element of assumed common knowledge, the recipe is
a complete set of instructions for how to make the cake.’ (Dupre 2005,
p. 198, our emphasis)
11
A longer justification for why such liberality is excessive goes like this.
If the primary goal of introducing the concept of genetic coding is to
single out genes as privileged causal elements in the developmental
process, then it might well seem that any successful account of coding
talk must have the consequence that, of the many causal factors that
Phil. Trans. R. Soc. B (2008)
combine causally during development, it is the genes alone that end up
coding for phenotypic traits. Elsewhere one of us has dubbed this the
uniqueness constraint ( Wheeler 2006). Griffiths & Knight (1998; see also
Griffiths 2001) introduce what is essentially the same constraint in
terms of what they call the ‘parity thesis’. The uniqueness constraint will
not be met if either (i) the account of genetic coding under
consideration fails to deliver the result that genes code for traits, since
if genes do not code for traits then they can not do so uniquely, or
(ii) that account does deliver the result that genes code for traits, but its
conditions for what it is to do this are met by other elements in the
extended developmental system, since then genes will not be the only
developmental elements that code for traits. Condition (ii) gives
expression to the excessive liberality problem. For discussion and
a more careful formulation of the uniqueness constraint, see
12
Note that the extended character of certain embodied and situated
modules is no barrier to this project. Dawkins’ (1982) influential notion
of the extended phenotype already shows us how genes may be
understood as coding for traits that are located outside the skin of the
organism (e.g. the genes that code for the spider’s web). Beyond that,
however, the waters between our sea monsters are exceptionally
turbulent. For example, the present authors have argued in the past
( Wheeler & Clark 1999; Wheeler 2003) that what we need is an
account of genetic coding based on two features of protein synthesis: the
arbitrariness of the mappings from particular nucleotide triplets to
particular amino acids, and the way in which information is consumed
by the subsystems that implement translation. However, one of us
( Wheeler 2006) has subsequently argued that once the details of this
account are filled in, it turns out that, strictly speaking, it is not the
molecules of DNA that code in development, but rather the down-
stream nucleotide triplets out of which molecules of mRNA are
constructed. It may be that the final route between the dual dangers of
strong instructionism and excessive liberality is still to be found.
13
For a different way of criticizing the evolutionary-psychological
conception of human nature, one that identifies an alleged incon-
sistency between that conception and the population-thinking
foundations of contemporary neo-Darwinian biology, see
Buller (2005).
REFERENCES
Atkinson, A. & Wheeler, M. 2004 The grain of domains: the
evolutionary-psychological case against domain-general
cognition. Mind Lang. 19, 147–176. (doi:10.1111/j.1468-
0017.2004.00252.x)
Atran, S. 2001 The case for modularity: sin or salvation? Evol.
Cogn. 7, 46–55.
Ballard, D., Hayhoe, M., Pook, P. & Rao, R. 1997 Deictic
codes for the embodiment of cognition. Behav. Brain Sci.
20, 723–767.
Barkow, J. H., Cosmides, L. & Tooby, J. (eds) 1992 The
adapted mind: evolutionary psychology and the generation of
culture. New York, NY: Oxford University Press.
Beach, K. 1988 The role of external mnemonic symbols in
acquiring an occupation. In Practical aspects of memory (eds
M. M. Gruneberg & R. N. Sykes), pp. 342–346. New
York, NY: Wiley.
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: University of Chicago Press.
Brooks, R. A. 1991 Intelligence without reason. In Proc. 12th
Int. Joint Conf. on Artificial Intelligence, pp. 569–595.
San Mateo, California: Morgan Kauffman.
Buller, D. J. 2005 Adapting minds: evolutionary psychology and
the persistent quest for human nature. Cambridge, MA: MIT
Press.
Buss, D. M. 1994 The evolution of desire: strategies of human
mating. New York, NY: Basic Books.
Buss, D. M. 2005 The handbook of evolutionary psychology.
Hoboken, NJ: Wiley.
Clark, A. 1997 Being there: putting brain, body and world
together again. Cambridge, MA: MIT Press.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3574 M. Wheeler & A. Clark Culture, embodiment and genes
Clark, A. 2001 Mindware: an introduction to the philosophy of
cognitive science. New York, NY: Oxford University Press.
Clark, A. In press. Supersizing the mind: embodiment, action,
and cognitive extension. New York, NY: Oxford University
Press.
Clark, A. & Chalmers, D. 1998 The extended mind. Analysis
58, 7–19. (doi:10.1111/1467-8284.00096)
Collins, S. H., Ruina, A. L., Tedrake, R. & Wisse, M. 2005
Efficient bipedal robots based on passive-dynamic walkers.
Science 307, 1082–1085. (doi:10.1126/science.1107799)
Cosmides, L. & Tooby, J. 1987 From evolution to behavior:
evolutionary psychology as the missing link. In The latest
on the best: essays on evolution and optimality (ed. J. Dupre),
pp. 227–306. Cambridge, MA: MIT Press.
Coultas, J. C. 2004 When in Rome. an evolutionary
perspective on conformity. Group Processes and Intergroup
Relations 7, 317–331. (doi:10.1177/1368430204046141)
Crawford, C. 1998 Environments and adaptations: then
and now. In Handbook of evolutionary psychology: ideas,
issues, and applications (eds C. Crawford & D. L. Krebs),
pp. 275–302. Mahwah, NJ: Lawrence Erlbaum.
Dawkins, R. 1982 The extended phenotype. Oxford, UK:
Oxford University Press.
Dennett, D. C. 1996 Kinds of minds: towards an understanding
of consciousness. London, UK: Weidenfeld & Nicolson.
Dupre, J. 2005 Are there genes? In Philosophy, biology and life
(ed. A. O’Hear), pp. 193–210. Cambridge, MA:
Cambridge University Press.
Feldman, M. W. & Cavalli-Sforza, L. L. 1989 On the theory
of evolution under genetic and cultural transmission with
application to the lactose absorption problem. In
Mathematical evolutionary theory (ed. M. W. Feldman),
pp. 145–173. Princeton, NJ: Princeton University Press.
Gaulin, S. J. C. & McBurney, D. H. 2001 Psychology: an
evolutionary approach. Upper Saddle River, NJ: Prentice
Hall.
Gould, S. J. 2000 More things in heaven and earth. In Alas
poor Darwin: arguments against evolutionary psychology
(eds H. Rose & S. Rose), pp. 101–126. New York, NY:
Harmony Books.
Gregory, R. 1981 Mind in science: a history of explanations in
psychology. Cambridge, MA: Cambridge University Press.
Griffiths, P. E. 2001 Genetic information: a metaphor in
search of a theory. Philos. Sci. 68, 394–412. (doi:10.1086/
392891)
Griffiths, P. E. & Knight, R. D. 1998 What is the
developmentalist challenge? Philos. Sci. 65, 253–258.
(doi:10.1086/392636)
Hirsh-Pasek, K. & Golinkoff, R. M. 1996 The origins of
grammar: evidence from early language comprehension.
Cambridge, MA: MIT Press.
Hollich, G., Hirsh-Pasek, K., Tucker, M. L. & Golinkoff, R. M.
2000 A change is afoot: emergentist thinking in language
acquisition. In Downward causation (eds P. Anderson,
C. Emmeche, N. O. Finnemann & P. V. Christiansen),
pp. 143–178. Oxford, UK: Aarhus University Press.
Hurley, S. L. 1998 Consciousness in action. Cambridge, MA:
Harvard University Press.
Hutto, D. 2008 Folk psychological narratives: the sociocultural
basis of understanding reasons. Cambridge, MA: MIT Press.
Jablonka, E. & Lamb, M. J. 2005 Evolution in four dimensions:
genetic epigenetic, behavioral, and symbolic variation in the
history of life. Cambridge, MA: MIT Press.
Johnson, M. 2001 Functional brain development in humans.
Nat. Rev. Neurosci. 2, 475–483. (doi:10.1038/35081509)
Kamiloff-Smith, A. 1992 Beyond modularity: a developmental
perspective on cognitive science. Cambridge, MA: MIT Press.
Kirby, S. 2002 Learning, bottlenecks and the evolution of
recursive syntax. In Linguistic evolution through language
Phil. Trans. R. Soc. B (2008)
acquisition: formal and computational models (ed. E. Briscoe),
pp. 173–204. Cambridge, MA: Cambridge University
Press.
Kirsh, D. 1995 The intelligent use of space. Artif. Intell. 73,
31–68. (doi:10.1016/0004-3702(94)00017-U)
Kirsh, D. & Maglio, P. 1994 On distinguishing epistemic
from pragmatic action. Cogn. Sci. 18, 513–549. (doi:10.
1016/0364-0213(94)90007-8)
Laland, K. N., Odling-Smee, J. & Feldman, M. W. 2000
Niche construction, biological evolution and cultural
change. Behav. Brain Sci. 23, 131–146. (doi:10.1017/
S0140525X00002417)
Lewontin, R. C. 1983 Gene, organism, and environment.
In Evolution from molecules to men (ed. D. S. Bendall),
pp. 273–285. Cambridge, MA: Cambridge University
Press.
Lewontin, R. C. 2000 The triple helix: gene, organism
and environment. Cambridge, MA: Harvard University
Press.
Lorenz, K. 1965 Evolution and the modification of behaviour.
Chicago, IL: University of Chicago Press.
Mameli, M. 2005 The inheritance of features. Biol. Philos. 20,
365–399. (doi:10.1007/s10539-004-0560-0)
Mareschal, D., Johnson, M., Sirois, S., Spratling, M.,
Thomas, M. & Westermann, G. 2007a Neuroconstructi-
vism: vol. 1, how the brain constructs cognition. New York,
NY: Oxford University Press.
Mareschal, D., Sirois, S., Westermann, G. & Johnson, M.
2007b Neuroconstructivism: vol. 2, perspectives and prospects.
New York, NY: Oxford University Press.
Mataric, M. 1991 Navigating with a rat brain: a neurobio-
logically inspired model for robot spatial representation.
In From animals to animats: proceedings of the first
international conference on simulation of adaptive behavior
(eds J.-A. Meyer & S. Wilson), pp. 169–175. Cambridge,
MA: MIT Press.
Maturana, H. & Varela, F. J. 1987 The tree of knowledge: the
biological roots of human understanding. Boston, MA:
New Science Library.
Namy, L., Smith, L. & Gershkoff-Stowe, L. 1997 Young
children’s discovery of spatial classification. Cogn. Dev. 12,
163–184. (doi:10.1016/S0885-2014(97)90011-3)
Noë, A. 2004 Action in perception. Cambridge, MA: MIT
Press.
Norman, D. 1993 Things that make us smart. Cambridge, MA:
Perseus Books.
Odling-Smee, J., Laland, K. & Feldman, M. 2003
Niche construction. Princeton, NJ: Princeton University
Press.
Pfeifer, R. & Bongard, J. 2007 How the body shapes the way we
think. Cambridge, MA: MIT Press.
Pinker, S. 1997 How the mind works. New York, NY: Norton.
Rowlands, M. 1999 The body in mind. Cambridge, MA:
Cambridge University Press.
Samuels, R. 1998 Evolutionary psychology and the massive
modularity hypothesis. Br. J. Philos. Sci. 49, 575–602.
(doi:10.1093/bjps/49.4.575)
Schlesinger, M. & Parisi, D. 2007 Connectionism in an
artificial life perspective: simulating motor, cognitive, and
language development. In Neuroconstructivism: perspectives
and prospects (eds D. Mareschal, S. Sirois, G. Westermann
& M. H. Johnson), pp. 129–158. Oxford, UK: Oxford
University Press.
Smith, E. A., Borgerhoff Mulder, M. & Hill, J. 2001
Controversies in the evolutionary social sciences: a guide
for the perplexed. Trends Ecol. Evol. 16, 128–135. (doi:10.
1016/S0169-5347(00)02077-2)
Sperber, D. 1996 Explaining culture: a naturalistic approach.
Oxford, UK: Blackwell.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Culture, embodiment and genes M. Wheeler & A. Clark 3575
Sporns, O. 2007 What neuro-robotic models can teach us
about neural and cognitive development. In Neurocon-
structivism: perspectives and prospects (eds D. Mareschal, S.
Sirois, G. Westermann & M. H. Johnson), pp. 179–204.
Oxford, UK: Oxford University Press.
Sterelny, K. 2003 Thought in a hostile world: the evolution of
human cognition. Oxford, UK: Blackwell.
Sterelny, K. & Griffiths, P. E. 1999 Sex and death: an
introduction to philosophy of biology. Chicago, IL: University
of Chicago Press.
Tomasello, M. 1999 The cultural origins of human cognition.
Cambridge, MA: Harvard University Press.
Tooby, J. & Cosmides, L. 1990 The past explains the present:
emotional adaptations and the structure of ancestral
environments. Ethol. Sociobiol. 11, 375–424. (doi:10.
1016/0162-3095(90)90017-Z)
Turner, S. J. 2000 The extended organism: the physiology of
animal-built structures. Cambridge, MA: Harvard
University Press.
Varela, F. J., Thompson, E. & Rosch, E. 1991 The embodied
mind: cognitive science and human experience. Cambridge,
MA: MIT Press.
von Frisch, K. 1975 Animal architecture. London, UK:
Hutchinson.
Wheeler, M. 2003 Do genes code for traits? In
Philosophical dimensions of logic and science: selected
contributed papers from the 11th International Congress
Phil. Trans. R. Soc. B (2008)
of Logic, Methodology, and Philosophy of Science (eds A.
Rojszczak, J. Cachro & G. Kurczewski), pp. 151–164.
Dordrecht, The Netherlands: Kluwer.
Wheeler, M. 2005 Reconstructing the cognitive world: the next
step. Cambridge, MA: MIT Press.
Wheeler, M. 2006 Traits, genes and coding. In Handbook of
the philosophy of biology, pp. 381–411. Amsterdam, The
Netherlands: Elsevier.
Wheeler, M. In press. Evolutionary models in psychology. In
The Routledge companion to the philosophy of psychology (eds
P. Calvo and J. Symons). Oxford, UK: Routledge.
Wheeler, M. & Clark, A. 1999 Genic representation:
reconciling content and causal complexity. Br. J. Philos.
Sci. 50, 103–135. (doi:10.1093/bjps/50.1.103)
Wilson, R. A. 1994 Wide computationalism. Mind 103,
351–372. (doi:10.1093/mind/103.411.351)
Wilson, R. A. 2004 Boundaries of the mind: the individual in the
fragile sciences. Cambridge, MA: Cambridge University
Press.
Wilson, R. & Clark, A. In press. How to situate cognition:
letting nature take its course. In Cambridge handbook of
situated cognition (eds M. Aydede and P. Robbins).
Cambridge, MA: Cambridge University Press.
Ziemke, T., Bergfeldt, N., Buason, G., Susi, T. & Svensson,
H. 2004 Evolving cognitive scaffolding and environment
adaptation: a new research direction for evolutionary
robotics. Connect. Sci. 16, 339–350. (doi:10.1080/095
40090412331314821)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3577–3589

doi:10.1098/rstb.2008.0132
Published online 17 September 2008

Exploring gene–culture interactions: insights

from handedness, sexual selection and
niche-construction case studies
Kevin N. Laland*
School of Biology, University of St Andrews, St Andrews, Fife KY16 9TS, UK
Genes and culture represent two streams of inheritance that for millions of years have flowed down
the generations and interacted. Genetic propensities, expressed throughout development, influence
what cultural organisms learn. Culturally transmitted information, expressed in behaviour and
artefacts, spreads through populations, modifying selection acting back on populations. Drawing on
three case studies, I will illustrate how this gene–culture coevolution has played a critical role in
human evolution. These studies explore (i) the evolution of handedness, (ii) sexual selection with a
culturally transmitted mating preference, and (iii) cultural niche construction and human evolution.
These analyses shed light on how genes and culture shape each other, and on the significance of
feedback mechanisms between biological and cultural processes.
Keywords: gene–culture coevolution; niche construction; handedness; sexual selection;
human evolution; evolutionary psychology

1. GENE–CULTURE COEVOLUTION olfactory receptors (OR4C13 and OR2B6 ), synapse-

With the human genome sequenced, attention has
been focused on analyses of the genetic data that have
been generated. One such set of analyses are attempts,
by mathematically minded geneticists, to detect
statistical signatures in the genome of recent, rapid
selection—genes favoured by natural selection over the
last 100 000 years (Sabeti et al. 2006, 2007; Voight et al.
2006; Wang et al. 2006; Nielsen et al. 2007; Williamson
et al. 2007). Such signals include high-frequency alleles
in linkage disequilibrium, unusually long haplotypes of
low diversity, and a variety of other signatures. While
relatively sensitive statistical tests for positive selection
have been developed, such methods are in their infancy
( Wang et al. 2006). Rather than giving absolute
numbers of selected genes, in definitive terms, the
analyses specify the likelihood that specific genes have
been subject to a recent selective sweep, which means
that it is difficult to give a clear answer as to precisely
how many genes are involved. Nonetheless, a reason-
able reading of the data suggests that, thus far,
somewhere between a few hundred and a couple of
thousand human genes have been identified, which
show signals of very strong and recent selection. The
best-known cases are alleles that provide resistance to
diseases such as malaria, and alleles that allow the
metabolism of lactose in cow’s milk.
One of the more intriguing categories, well rep-
resented (more than 15%) in inferred selective events,
is neuronal function ( Wang et al. 2006), including the
serotonin transporter (SLC6A4), glutamate and
glycine receptors (GRM3, GRM1 and GLRA2),
*knl1@st-andrews.ac.uk
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
associated proteins (RAPSN ) and a number of brain-
expressed genes with largely unknown function (ASPM
and RNT1). There is evidence that the evolution of
nervous system genes has been accelerated in humans
(Dorus et al. 2004), with faster evolution of gene
expression in the human brain compared with other
primates ( Wang et al. 2007) and with an increased rate
of changes in the genomic regions responsible for the
regulation of brain development in the human genome
(Pollard et al. 2006). In other words, a substantive
proportion of recently favoured genes are expressed in
the human brain, which has undergone significant
recent remodelling.
Humans possess approximately 25 000 genes, so
researchers should not be surprised that a small
proportion shows signs of recent selection. Moreover,
a substantial fraction (perhaps even a quarter) of
human genes are expressed in the brain so, even
allowing for slower evolution in brains than elsewhere
as molecular insights from comparison of human and
chimpanzee genome imply (Hill & Walsh 2005), we
have every reason to expect recent evolution of the
human brain. Minimally, a small subset of neural
genes, and perhaps many more, have been targets of
positive selection ( Hill & Walsh 2005). Yet the
dominant view within North American evolutionary
psychology has been that our species has undergone
comparatively little evolutionary change in recent
millennia, particularly with respect to mental adap-
tations, which were regarded as products of resistant-
to-change gene complexes (Cosmides & Tooby 1987).
I suggest that the large numbers of human genes
now known to have been subject to recent positive
selection, including those expressed in the brain and
behaviour, are an embarrassment to this evolutionary
psychology viewpoint.

3577 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3578 K. N. Laland Exploring gene–culture interactions
Nonetheless, frequent signs of recent selection make
a lot of sense when one considers the dramatic changes
in selection pressures that our species has experienced.
Among other challenges, in the last 100 000 years,
humans have spread from East Africa around the globe,
experienced an ice age, begun to exploit agriculture,
witnessed rapid increases in densities and, by keeping
animals, experienced a new proximity to animal
pathogens. They have also domesticated hundreds of
species of plants and animals (Smith 2007). What
is immediately striking about these major challenges is
that all except one (the ice age) have been self-imposed:
that is, human activities have modified selection
pressures, for instance by dispersing into new environ-
ments with different climatic regimes, devising agri-
cultural practices or domesticating livestock.
These activities are instances of human ‘niche
construction’ (the modification of environments by
organisms), which, I suggest, have precipitated
evolutionary responses in the human genome (Laland
et al. 2001; Odling-Smee et al. 2003). However, the
capacity for culture is clearly a critical factor underlying
the potency of human niche construction: agriculture
was not independently invented by each farmer, nor is
its presence an unlearned maturational outcome of
human gene expression. Moreover, even in the case
of climatic regimes, beyond human control, human
‘cultural niche construction’ would have strongly
affected the intensity of selection, for instance, by
manufacturing clothes or shelters, or controlling fire.
The argument that human cultural niche construc-
tion has been a co-director of recent human evolution is
essentially the conclusion reached by the geneticists
analysing the human genome:
Homo sapiens have undoubtedly undergone strong
recent selection for many different phenotypes..
Given that most of these selective events likely occurred
in the last 10,000–40,000 years.it is tempting to
speculate that gene–culture interactions directly or
indirectly shaped our genomic architecture
( Wang et al. 2006, p. 140, my italics)
This perspective is also supported by some well-
researched cases of gene–culture coevolution. For
instance, there are several examples of culturally
induced genetic responses to human agriculture
(Odling-Smee et al. 2003). The best known is the
coevolution of the gene for lactose absorption and dairy
farming (Durham 1991). There is now compelling
theoretical and empirical evidence that dairy farming
spread prior to the allele for lactose absorption,
generating a selection pressure favouring this gene
in some human pastoralist societies ( Feldman &
Cavalli-Sforza 1989; Holden & Mace 1997; Burger
et al. 2007). Another is provided by a population of
Kwa-speaking yam cultivators in West Africa (Durham
1991). These people cut clearings in forests to grow
crops, with a cascade of consequences. The clearings
increased the amount of standing water, which
provided better breeding grounds for mosquitoes
and increased the prevalence of malaria. This, in
turn, modified natural selection pressures in favour
of an increase in the frequency of the sickle-cell
S allele because, in the heterozygous condition, the
Phil. Trans. R. Soc. B (2008)
gene development
t culture
pool modified
selection
inheritance
inheritance
cultural
genetic
time
gene development
t +1 culture
pool modified
selection
Figure 1. Gene–culture coevolution. Genes and culture are
two interacting forms of inheritance. Genetic propensities,
expressed throughout development, influence what cultural
organisms learn. Culturally transmitted information,
expressed in behaviour and artefacts, modifies selection
acting back on the genome.
S allele confers protection against malaria. The fact
that other Kwa speakers, whose agricultural practices
are different, do not show the same increase in the
S allele frequency supports the conclusion that cultural
practices can drive genetic evolution (Durham 1991).
It is not just yam cultivation that generates this pattern
of selection: modern Asian tyre manufacturing is
having the same effect, with mosquitoes infesting
pools of rainwater that collect in tyres stored outside,
and tyre export contributing to the spread of malaria
and dengue (Hawley et al. 1987). Malaria became a
major health problem only after the invention of
farming, a human cultural niche-constructing practice,
yet there are several additional genes that appear to
have been favoured by selection because they provide
resistance to malaria. These include G6PD, TNFSF5
and alleles coding for haemoglobin C and Duffy
blood groups (Balter 2005; Wang et al. 2006). There
is also evidence that genes have been selected because
they confer resistance to other modern diseases,
including AIDS and smallpox (CCR5) and hyperten-
sion (AGT, CYP3A; Balter 2005). In all these cases,
human modifications of the environment triggered or
modified selection on human genes.
The view that genes and culture coevolve was first
suggested by pioneers of the field of ‘gene–culture
coevolution’ nearly 30 years ago (Cavalli-Sforza &
Feldman 1981; Boyd & Richerson 1985; see Laland &
Brown (2002) for an overview). These researchers view
genes and culture as two interacting forms of inherit-
ance, with offspring acquiring both a genetic and a
cultural legacy from their parents and, in the latter case,
other conspecifics too (figure 1). Genetic propensities,
expressed throughout development, influence what
cultural organisms learn. Culturally transmitted infor-
mation, expressed in behaviour and artefacts, spreads
through populations, modifying selection acting back on
populations. Mathematical gene–culture coevolutionary
models have shown how our views of human evolution
change when both inheritance systems are taken
into account (Feldman & Cavalli-Sforza 1976, 1989;
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Exploring gene–culture interactions
Boyd & Richerson 1985; Richerson & Boyd 2005).
Culture is not just a property of humans, it is a
fundamental cause of how humans got to be the way
they are, a dynamic process that shapes psychological and
material worlds (Boyd & Richerson 1985; Richerson &
Boyd 2005). Human minds have evolved specifically to
exploit the cultural realm.
Gene–culture coevolutionary analyses typically
build on conventional population genetic theory. In
addition to tracking how allele or genotype frequencies
change in response to evolutionary processes such as
selection and drift, the analyses also incorporate
cultural transmission (by, for instance, learning from
parents, or from the previous generation, or conform-
ing to the majority view) into the models, and explore
how learned characters coevolve with genetic variation
that either affects its expression or acquisition, or whose
fitness is affected by the cultural environment, or both.
The theory has deployed in a variety of different
ways. First, and primarily, it has been used to explore
the adaptive advantages of reliance on learning and
culture, for instance, by asking under what circum-
stances natural selection favours reliance on social
learning (Boyd & Richerson 1985; Rogers 1988;
Feldman et al. 1996; Enquist et al. 2007), and what
kinds of learning biases are adaptive (Boyd & Richerson
1985; see McElreath et al. 2008). Second, it has been
deployed to investigate the inheritance of behavioural
and personality traits (Cavalli-Sforza & Feldman 1973;
Otto et al. 1995), frequently finding lower heritabilities
and higher influence of social learning than conven-
tional human behaviour genetics twin studies. Third, it
has been applied to investigate specific instances of
human evolution, including cultural group selection
(Boyd & Richerson 1985), and the emergence of incest
taboos (Aoki & Feldman 1997).
I will not attempt here to provide a summary of the
entire field of gene–culture coevolution, a challenging
task given recent growth in this domain of research (see
Feldman & Laland (1996) and Richerson & Boyd
(2005) for overviews). Rather, in this article, I will
restrict myself to presenting work carried out by me and
my collaborators, and provide a personal account of
what I believe the principal take-home messages of this
small body of theory. In §2 I present what are designed
to be accessible verbal summaries of three case studies
exploring gene–culture interactions through the use of
gene–culture coevolutionary models. These studies
explore (i) the evolution of handedness, (ii) sexual
selection with a culturally transmitted mating pref-
erence, and (iii) cultural niche construction and human
evolution. In §3 I attempt to synthesize insights from
these case studies into a coherent general statement
concerning how genes and culture have interacted
throughout recent evolution, and what the implications
of this interaction are for understanding human
behaviour and society.
2. CASE STUDIES OF GENE–CULTURE
COEVOLUTION
(a) The evolution of handedness
Why is not everyone right-handed? Extensive experi-
mental studies reveal that approximately 90 per cent of
Phil. Trans. R. Soc. B (2008)
K. N. Laland 3579
humans are right-handed (Corballis 1991). This
estimate is loosely consistent across the world, but
does vary to some degree between cultures (Corballis
1991). But there are no cultures in the world in which
left-handers are the majority, and this has led researchers
to conclude that right-handedness must have been
favoured by selection during the course of recent
human evolution. But that begs the question, if it is
advantageous to be right-handed, why is not everybody?
What processes might be preserving left-handers in
human populations? The most commonly given answer
to this question is genetic variation, preserved through
some selective regime such as heterozygote advantage
(Annett 1985; McManus 1985) or frequency-
dependent selection Faurie & Raymond (2005).
There are two major problems for exclusively
genetic models of handedness, and genetic models are
currently the leading models of handedness (Annett
1985; McManus 1985). First, such models would
predict that concordance for handedness would
increase with relatedness, but as Morgan & Corballis
(1978) stated: ‘knowledge of a person’s handedness
tells us virtually nothing of the handedness of that
person’s twin or sibling’ (p. 273). This statement
remains entirely valid in 2008. One is given no insight
into the likely handedness of an individual if one knows
that of its siblings. Moreover, genetic models would
predict that identical twins would be more alike than
fraternal twins, yet they have essentially the same
concordance rates for handedness: 0.772 for MZ and
0.771 for DZ twins (data based on a meta-analysis of
14 twin studies from McManus 1985). While isolated
studies (e.g. Warren et al. 2006) have reported positive
heritabilities for some handedness measures, the over-
all picture across multiple studies remains that
handedness, at least as measured in the vast majority
of questionnaire and performance studies, does not
exhibit strong heritability (McManus 1985; Neale
1988; Su et al. 2005).
Second, purely genetic accounts of handedness fail
to explain the well-established cultural influences on
handedness. Left-handers are found at lower frequen-
cies in societies that associate it with clumsiness, evil,
dirtiness or mental illness, such as some middle and far
eastern countries (Harris 1980; Corballis 1991).
Studies of school children in China and Taiwan report
only 3.5 and 0.7 per cent used their left hand
for writing, compared with a 6.5 per cent estimate for
Oriental school children living in the USA (Hardyck
et al. 1976; Teng et al. 1976; Hung et al. 1985).
As the worldwide dominance of right-handers
strongly suggests a genetic influence or constraint, yet
the cross-cultural variation reveals a cultural influence,
handedness appears to be well suited to a gene–culture
coevolutionary analysis. Laland et al. (1995a) con-
structed a gene–culture coevolutionary model of hand-
edness that made the following assumptions. First,
there are two phenotypic states: that is, individuals are
characterized as right- or left-handed (there are no
ambidextrous individuals and no degrees of handed-
ness). While this assumption would be contested
by some researchers (Annett 1985), simulations reveal
that this assumption, made for mathematical con-
venience, does not greatly affect our conclusions, and
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3580 K. N. Laland Exploring gene–culture interactions
other researchers have argued compellingly that hand-
edness data are strongly bimodal in distribution
(McManus 1985). Second, following McManus’s
notation, we assume that the probability of becoming
right- or left-handed is influenced by alleles D and C at a
single locus. This is not to suggest that we believed only
a single gene influences handedness, but rather we
focused on a single hypothetical gene as a means of
exploring how any autosomal genetic variation is likely
to respond. Below I argue that our model implies a series
of selective sweeps of handedness genes throughout
human evolution, each ratcheting up the proportion of
right-handers. Third, we assumed that culturally
transmitted biases also affect handedness, primarily
through a parental influence. This assumption is
justified by the observation that handedness is usually
fully developed by the age of 2–3 (Bishop 1990).
Hence, an individual’s handedness depends on its
genotype and the handedness of its parents. The
probability of a right-handed child being born to parents
with various different patterns of handedness, given the
three possible offspring genotypes, is given in table 1.
Here, the parameter r represents the dextralizing effect
of genotype DD, a represents the increase in right-
handedness caused by having two right-handed parents
(or the decrease caused by two left-handed parents) and
b represents the change in handedness affected by
parents of mixed-handedness. Since non-human
primates may exhibit individual hand preferences, but
evidence for population-level biases is, at best, conten-
tious (Palmer 2002), we assume as a starting point for
our analysis an ancestral population in which individuals
were not genetically predisposed towards either hand
(a CC population). We consider two forms of selection,
favouring either right-handedness directly or allele D,
the latter representing cases in which handedness is
favoured owing to selection on some other lateralized
structure or function. The analysis found that irrespec-
tive of the starting frequency of right-handedness, the
magnitude of the selective advantage to right-handers or
the degree of dominance of the alleles, all genetically
variable populations converge on a single evolutionary
trajectory, and continue to evolve until allele D is fixed,
and the frequency of right-handers is given by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2a C 2b K1 C 4a2 K4a C 4b2 C 1 C 8br
PDR Z
4b
ð2:1Þ
Given this finding, how can variation in handedness be
reconciled? Clearly, the gene–culture interaction has not
preserved variation in handedness. The hypothesis that
human populations are currently evolving towards the
equilibrium specified in (2.1) is inconsistent with data
revealing a decreasing trend in right-handedness in the
USA and Australia over the last century (Corballis
1991), data generally interpreted as reflecting a
relaxation in the social pressure to conform to a right-
handed standard. While simulations reveal that selective
regimes such as heterozygote advantage and frequency-
dependent selection could preserve genetic variation, as
mentioned above, such explanations are inconsistent
with the observation that handedness has low herit-
ability. However, the analysis suggests an alternative
Phil. Trans. R. Soc. B (2008)
Table 1. The probability of a right-handed child being born to
parents with various different patterns of handedness, given
the three possible offspring genotypes. (Here, r represents the
dextralizing effect of genotype DD, a represents the increase
in right-handedness caused by having two right-handed
parents (or the decrease caused by two left-handed parents),
b represents the change in handedness affected by parents of
mixed-handedness and h1 is a parameter specifying the
dominance of D and C alleles.)
parental mating DD DC CC
right!right 1/2CrCa 1/2Ch1rCa 1/2Ca
right!left 1/2CrCb 1/2Ch1rCb 1/2Cb
left!left 1/2CrKa 1/2Ch1rKa 1/2Ka
explanation: human populations may have reached the
equilibrium specified by (2.1), such that no genetic
variation underlies variation in handedness, but left-
handers would nonetheless remain in the population if
aCr!1/2.
We explored this possibility by collating data on
patterns of handedness in families. We found 17 studies
that gave the frequencies of right- and left-handed
offspring born to two right-handed parents, one right
and one left, and two left-handed parents, which give
rise to decreasing proportions of right-handed offspring.
(The datasets derive from western Europe and North
America, for which the incidence of left-handedness is
relatively consistent.) We then carried out a maximum-
likelihood analysis in which we used the familial dataset
to estimate the best-fit values of a, b and r, the three
remaining free parameters in our model at equilibrium.
In the first instance, b came out very close to zero, so we
eliminated it from the model and reconducted the
analysis, which gave values of aZ0.14 and rZ0.28.
With these values, the model gives a good fit to 16 out of
the 17 studies, and across all studies combined
(GZ44.33, d.f.Z32, pO0.05). Similar maximum-
likelihood analyses to the same kind of data applied to
the leading genetic models give a poorer fit—our model
gives a good fit to more studies and a poor fit to fewer
studies than any other model.
The analysis suggests that all humans are born with a
predisposition to be right-handed of (1/2)CrZ0.78;
that is, all other factors being equal, 78 per cent of
: people would be right-handed. However, all other
factors are not equal, since parents exert a bias on
patterns of handedness. Two right-handed parents
increase the probability that their child will be right-
handed by a further 14 per cent (aZ0.14), to give an
overall probability of 0.92, while two left-handed
parents decrease the probability by the same pro-
portion, leaving the probability of a right-hander at
0.64. Parents of mixed-handedness cancel out each
others’ influence (bZ0). The exact nature of the
parental influence is not clear, but we assume that it
represents a combination of imitation, inadvertent
shaping and direct instruction (see Laland et al.
(1995a) for discussion).
Three independent tests of our model were
performed. First, we plugged the values of a, b and r
into equation (2.1), to derive an overall expected
frequency of right-handers of 0.88, very close to the
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Exploring gene–culture interactions
observed value. Second, we collated studies giving the
frequency of right–right, right–left and left–left pairs of
monozygotic and dizygotic twins, and, compared with
the observed data, our predictions of expected
proportions in each category given the overall number
of twins in the study. Using the same parameter values
derived from the familial dataset, our model generated
expectations that were a good fit to 27 out of the 28
twin datasets we could find, and across all studies
combined (GZ35.76, d.f.Z28, pZ0.38). Once again,
our model outperforms all other models subjected to
this analysis. Third, we used the values of a, b and r to
compute expectations for the degree of concordance
for handedness in unrelated individuals and mono-
zygotic twins, which gave values of 0.79 and 0.8,
respectively. This explains Morgan & Corballis’s (1978)
observation that knowledge of a person’s handedness
tells us virtually nothing of the handedness of that
person’s twin or sibling. These three independent tests
of the model lend confidence in our conclusions.
In summary, patterns of inheritance and variation in
handedness are the outcome of a gene–culture inter-
action. A history of selection on handedness has
created a universal genetic predisposition towards
right-handedness; our genes load the die to favour the
right, but in a facultative rather than obligate manner.
However, patterns of variation in handedness within
families and across societies are the product of a
cultural influence—specifically, a parental bias leading
individuals to shape their child’s handedness to
resemble their own. In this respect, I anticipate
variation between societies will correspond to different
values of a (and possibly b), a hypothesis that is open to
testing. Since our model assumes no genetic variation
underlying current variation in handedness, it is
consistent with humans possessing many handedness-
distorting genes of small effect, reflecting multiple
selective sweeps over the course of human evolution,
and, in this respect, is consistent with human genetic
data. (By contrast, those models reliant on genetic
variation underlying variation in handedness typically
assume that a single gene of major effect influences
handedness, and such strong single-gene effects on
behaviour are extremely rare). It is plausible that
selection for right-handers may have occurred over
millions of years, and may perhaps even have begun in
a common ancestor of humans and chimpanzees.
Studies of handedness in chimpanzees provide increas-
ingly compelling evidence for a population-level
handedness bias to the right, although Palmer (2002)
notes effect sizes shrink as the number of recordings per
individual and sample size increase. However, if this
pattern is confirmed, it is clear that the bias is small—
56 per cent of hand use by common chimpanzees is
right-handed (Palmer 2002). Archaeological data,
based on patterns of flint knapping or skeletal data,
provide evidence for increasingly strong biases in
Lower Pleistocene hominids (0.57), Middle Pleisto-
cene hominids (0.61) and Neanderthals (0.8–0.9)
(Toth 1985; Uomini in press). Thus, the comparative
data, weak though it is, support the suggestion that
handedness distorters have been repeatedly favoured
by selection over hundreds of thousands, and perhaps
even millions of years. With each selective sweep
Phil. Trans. R. Soc. B (2008)
K. N. Laland 3581
favouring a dextralizing allele, the proportion of right-
handers would be ratcheted up, not just owing to the
immediate effect of the gene, but also because, by
increasing the frequency of right-handed parents, the
proportion of children exposed to a cultural bias
favouring right-handedness increases. Although the
extent to which culture shaped selection pressures is
unknown, I suspect that both directly, by constructing
an environment suited to the right-handed majority,
and indirectly, by introducing new behaviour patterns
that benefited from hand specialization, hominin
cultural processes increasingly reinforced selection
favouring right-handedness.
(b) Sexual selection with a culturally
transmitted mating preference
The field of evolutionary psychology is dominated by
experimental and questionnaire studies of human
mating preferences and behaviour, for which sexual
selection interpretations are rife (Buss 1994; Barrett
et al. 2001). By contrast, theoretical analysis of human
sexual selection is relatively understudied. Certainly,
there is a well-established general body of theory
investigating the interaction between genetically trans-
mitted traits and preferences (e.g. Kirkpatrick 1982),
but it is not clear to what extent human mating
preferences are influenced by genetic variation. In their
classic book, Gould & Gould (1989, p. 254) wrote:
Much of our thinking about the role of sexual selection
in shaping modern human behaviour is paralyzed by
the difficulty of separating the effects of nature and
nurture.
The clear implication of this statement is that
learning and culture may shape human mating
behaviour, obscuring understanding of how sexual
selection has acted. Similarly, social science critics of
human sociobiology and evolutionary psychology
frequently argue that sexual selection explanations for
human mating behaviour are implausible given the
cultural influence on human preferences (Ford &
Beach 1951; Tan 1979; Aronson 1995).
Contrary to these positions, here I show that the
interaction of cultural and selective processes can itself
result in sexual selection. That is, even if human mating
preferences are learned, socially transmitted, and
culture-specific, sexual selection will still result; indeed,
culturally generated sexual selection may be even more
potent than its conventional gene-based counterpart.
Laland (1994) combined sexual selection and gene–
culture coevolutionary theory to explore the impact of a
culturally transmitted mating preference favouring
genetically inherited traits in the opposite sex. Gene–
culture interactions are likely to be important here for
several reasons. First, evidence for the cultural
transmission of human preferences is pervasive in
human societies (Cavalli-Sforza & Feldman 1981;
Boyd & Richerson 1985; Hewlett & Cavalli-Sforza
1986). Second, as an increasing number of species
(currently many hundreds, including some inver-
tebrates) are found to exhibit a capacity for social
transmission (see Galef & Laland (2005) for a review),
the possibility emerges that gene–culture interactions
may have shaped selection in other species. For
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3582 K. N. Laland Exploring gene–culture interactions

instance, mate-choice copying has been observed in (a) 1

birds and fishes (Dugatkin 1992; White & Galef 2000),
and theoretical models of mate-choice copying reveal
that learned preferences could plausibly coevolve with
gene-based traits (Kirkpatrick & Dugatkin 1994).
Third, gene–culture interactions may significantly

genetic trait (T2)

affect evolutionary rates, speeding them up and slowing
them down under different circumstances (Laland
1992; Laland et al. 1995b, 2001).
Laland’s (1994) analysis was based on Kirkpatrick’s
(1982) sexual selection model, but incorporated vertical
cultural transmission, although an oblique transmission
model gave qualitatively similar results. Cultural biases,
variously termed ‘cultural selection’ (Cavalli-Sforza &
Feldman 1981) or ‘biased transmission’ (Boyd &
Richerson 1985), in the form of differential social
learning of behavioural alternatives, affect the frequency 0
of cultural variants in a population over time. Laland (b) 1
explored the consequences of biased and unbiased
cultural transmission of mating preferences (expressed
in either sex) on the sexual selection of gene-based traits
in the opposite sex. Both haploid and diploid models
were constructed, reliant on either uniparental or
biparental inheritance of preferences. In the simplest
case, members of one sex (here males) exhibit one of
trait (T2)

two traits, T1 and T2, the latter exhibiting a viability

deficit of 1Ks relative to the former, and the other sex
(here females) exhibit one of two culturally learned
preferences, P1 and P2, for traits in their mating
partners, with P1 (females) unbiased and P2 (females)
preferring to mate with T2 (males) a times more
frequently than T1 (males). The principal finding is
general to all models: sexual selection is the outcome of
this interaction.
Indeed, when cultural transmission is unbiased the
haploid system is formally equivalent to Kirkpatrick’s
(1982) classic model of sexual selection, and exhibits
the same familiar curve of neutrally stable equilibria
(shown as the thick line in figure 2a). As in
Kirkpatrick’s model, for the trait allele to have any
non-zero equilibrium frequencies, it is required that
s!1K1/a. If a population is on the curve, cultural drift
(Cavalli-Sforza & Feldman 1981; Boyd & Richerson
1985) or individual learning could change the
frequency of the preference, and hence indirectly alter
the frequency of the trait. As with genetic models, a
statistical association equivalent to linkage disequi-
librium builds up between genetic trait and cultural
preference, as the offspring of P2!T2 matings inherit
both characteristics. If the covariance between trait and
preference and the frequency of P2 is sufficiently high,
P2 generates selection favouring T2 in spite of the trait’s
viability deficit and hitch-hikes to fixation on the back
of it; that is, P2 and T2 exhibit runaway sexual selection.
As with the genetic models, the observed curves of
neutrally stable equilibria are structurally unstable, and
disappear with selection on the preference. Here, with
any degree of bias in favour of P2 during cultural
transmission, there is only one stable equilibrium point,
with P2 and T2 fixed if s!1K1/a (figure 2b) and with P1
and T1 fixed if sO1K1/a. Strong biases quickly result
in the fixation of P2, and a subsequent rapid increase in
the frequency of T2. At the extreme (strong trans-
mission bias, large a, small or negative s), traits may be
0 1
cultural preference (P2)
Figure 2. Sexual selection resulting from a culturally
transmitted mating preference (P1 or P2) in one sex for a
genetically transmitted trait (T1 or T2) expressed in the other,
where P1 individuals are unbiased and P2 individuals prefer
T2 mates. (a) Unbiased vertical cultural transmission and
(b) biased transmission favouring P2.
taken from low to high frequency in just a handful of
generations. Even weak biases typically bring about
more rapid patterns of genetic change than conven-
tional gene-based models, since cultural preference
frequencies typically increase faster than genetic
preferences. The findings hold for both biparental
and uniparental inheritance of preferences, for haploid
and diploid genetics, and for both ‘Fisherian’ and ‘good
genes’ scenarios (positive and negative s). Oblique
transmission (learning from non-relatives) weakens the
covariance between trait and preference, but compen-
sates by inducing more rapid spread of the preference,
such that strong sexual selection is again the outcome.
In summary, the analysis reveals that a culturally
transmitted mating preference that reaches a significant
frequency through drift, asocial or social learning can
under most circumstances generate selection that takes
a preferred trait in the opposite sex to fixation, or to
non-zero frequencies, even if that trait is costly, and
frequently with the preference hitch-hiking along.

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Exploring gene–culture interactions
Given the pervasiveness of cultural influences on
human mating preferences (Darwin 1871; Tan 1979;
Gould & Gould 1989; Aronson 1995), social trans-
mission may exert a powerful influence on the selection
of secondary sexual characteristics, and other physical
and personality traits that affect human mate choice.
The hypothesis could plausibly apply to many human
traits, including skin colour, facial features, facial and
body hair, body shape, height, degree of character
symmetry, degree of neoteny, level of aggressiveness,
emotionality and a variety of personality traits.
The analysis leads to several predictions. First, it
suggests that we should expect to see mate-choice
copying and, second, the social transmission of mating
preferences, in humans. Third, it predicts society-wide
correlations between culturally transmitted preferences
and gene-based traits (in both sexes). Evidence is now
starting to accumulate that supports these predictions.
Jones et al. (2007) conducted experiments in which
images of male faces were presented to females
adjacent to images of females that were either smiling
or looking impassively at the males. Females rated the
smiled-at faces as more attractive than the alternatives,
indicative of mate-choice copying. A similar study by
Little et al. (2008) revealed that this process can
generate preferences for particular characteristics of the
smiled-at male that are expressed in other males,
indicative of the social transmission of mating pref-
erences. Several recent studies of human mating
preferences report differences in the perception of
female attractiveness in different cultural groups, and
preferences that change rapidly over time (e.g.
Furnham & Baguma 1994; Craig et al. 1996; Yu &
Shepard 1998; Wetsman & Marlowe 1999; Marlowe &
Wetsman 2001; Tovee et al. 2006), again strongly
suggestive of culture-specific and culturally transmitted
mating preferences. Indeed, Darwin (1871) devotes an
entire chapter (XIX) of The Descent of Man to
documenting cross-cultural differences in human
mating preferences and points out that these coincide
with physical characteristics in the opposite sex. He
writes (p. 353): ‘It is certainly not true that there is
in the mind of man any universal standard of beauty
with respect to the human body’. Assuming he is
correct, this mechanism could be a major source
of cross-cultural variation in anatomical and beha-
vioural traits.
(c) Cultural niche construction and human
evolution
Niche construction is the very general process whereby
organisms modify their own and/or each others’ niches,
through their metabolism, their activities and their
choices (Odling-Smee et al. 2003). It is far from
restricted to humans: numerous animals manufacture
nests, burrows, holes, webs and pupal cases; plants
change levels of atmospheric gases and modify nutrient
cycles; fungi and bacteria decompose organic matter;
bacteria fix nutrients (Lewontin 1982, 1983; Odling-
Smee 1988; Odling-Smee et al. 2003). The defining
characteristic of niche construction is the modification
of the relationship between an organism and its
environment (Odling-Smee 1988), and hence niche
construction subsumes habitat selection, dispersal and
Phil. Trans. R. Soc. B (2008)
K. N. Laland 3583
migration. Advocates of the niche-construction per-
spective within evolutionary biology stress the active
role that organisms play in driving evolutionary and
coevolutionary events.
The niche-construction perspective differs from the
conventional one in recognizing two major adaptive
processes in evolution, natural selection and niche
construction, and two general forms of inheritance,
genetic and ecological inheritance (Odling-Smee
1988). Ecological inheritance refers to the modified
environments (e.g. nests, burrows), incorporating
modified selection pressures, which descendant organ-
isms inherit from their ancestors. Organisms transmit
to their offspring, and subsequent descendents phys-
ically altered selective environments, both through
actions on their biological and non-biological environ-
ments and by their habitat choices.
Many researchers have explored the evolutionary
ramifications of niche construction by developing and
analysing mathematical models (Laland et al. 1996,
1999, 2001; Odling-Smee et al. 2003; Ihara & Feldman
2004; Borenstein et al. 2006; Silver & Di Paolo 2006).
All such analyses conclude that niche construction is
evolutionarily consequential. Typically, population
genetic models investigate the dynamics of the joint
evolution of environment-altering, niche-constructing
traits in organisms and ‘recipient traits’, whose fitness
depends on feedback from natural selection in environ-
ments that can be altered by niche construction (Laland
et al. 1996, 1999, 2001; Odling-Smee et al. 2003).
These theoretical analyses suggest that this ‘self-
imposed’ selection resulting from niche construction
will often override external sources of selection (i.e.
selection acting on the population independent of their
niche-constructing activities) to create new evolution-
ary trajectories, which will lead to the fixation of
otherwise deleterious alleles, the support of stable
equilibria where none are expected and the elimination
of what would otherwise be stable polymorphisms.
Among the most significant analyses is Silver &
Di Paolo’s (2006) study, which found that niche-
construction traits can drive themselves to fixation by
simultaneously generating selection that favours ‘re-
cipient’ trait alleles and linkage disequilibrium between
niche-construction and recipient trait alleles.
Frequently, the evolution of the recipient trait
depends on the frequency of the niche-constructing
trait over several generations—that is, on ecological
inheritance. Processes that carry over from past
generations can change the evolutionary dynamic in a
number of ways, generating time lags in response to
selection of the recipient trait, momentum effects
(populations continuing to evolve in the same direction
after selection has stopped or reversed), inertia effects
(no noticeable evolutionary response to selection for a
number of generations), opposite responses to selection
and sudden catastrophic responses to selection
( Feldman & Cavalli-Sforza 1976; Kirkpatrick &
Lande 1989; Laland et al. 1996, 1999, 2001).
Niche construction also provides a non-Lamarckian
route by which acquired characteristics can influence
the selective environment. While the information
acquired by individuals through ontogenetic processes
cannot be inherited because it is lost when they die,
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3584 K. N. Laland Exploring gene–culture interactions
processes such as learning can nonetheless still be of
considerable importance to subsequent generations
because learned knowledge can guide niche construc-
tion in ways that do modify natural selection. This
route is considerably enhanced by social learning,
which allows animals to learn from each other.
Hundreds of species of mammals, birds and fishes are
now known to learn socially (Zentall & Galef 1988;
Heyes & Galef 1996), allowing novel learned traits to
sweep through populations and exposing individuals
to novel selection pressures. This process is further
amplified with stable trans-generational culture, and it
is now widely believed that such characters were
probably important to hominid evolution (Cavalli-
Sforza & Feldman 1981; Richerson & Boyd 2005). In
humans, culture has greatly amplified our capacity for
niche construction and our ability to modify selection
pressures. This highlights the requirement for theoreti-
cal analyses exploring the evolutionary ramifications of
human cultural niche construction.
Laland et al. (2001) combined niche-construction
and gene–culture coevolutionary models to explore the
evolutionary consequences of cultural niche construc-
tion. Our models were based on three key assumptions.
First, a population’s capacity for niche construction is
influenced by the frequency of a cultural trait (E or e),
where the two traits represent the presence and
absence, more and less, or different forms, of niche
construction, respectively. Second, the amount of some
resource R in the environment is dependent on the
niche-constructing activities of past and present
generations. This resource could be an artefact (e.g.
shelter, tools) that the population constructs, some
manufactured or accrued commodity (e.g. food,
water), or a modified environmental condition (e.g.
temperature). Third, the amount of the resource in the
environment influences the pattern and strength of
selection acting on alleles (A and a) at a genetic locus.
For illustration, in the aforementioned Kwa example,
the cultural trait E represents yam cultivation, the
resource R is the amount of standing water and
the recipient allele is the sickle-cell S allele.
The fitnesses of individuals with cultural traits E or e,
and genotypes AA, Aa and aa, are shown in table 2.
Two components of selection are represented by these
fitness functions: a fixed-fitness component (a, h
terms) representing selection acting on the population
independent of their niche construction, and a
frequency-dependent component (3, R terms) re-
presenting the selection brought about or modified
through niche construction, where 3 is a constant that
weights the relative importance of the two components.
Two classes of model were constructed, in which the
amount of the resource depended exclusively on prior
niche construction, and where additional processes of
resource accrual and depletion were acting. The model
assumed vertical cultural transmission of the cultural
trait (learning from parents) of unbiased, biased or
incomplete forms, although, once again, simulations
introducing oblique transmission gave qualitatively
similar results.
The analysis provided ample evidence that cultural
niche construction could plausibly affect human
genetic evolution, in a multitude of ways. As with the
Phil. Trans. R. Soc. B (2008)
Table 2. The fitnesses of individuals with cultural traits E or e
and genotypes AA, Aa and aa (phenogenotypes). (Each
phenogenotype fitness is composed of two terms: the first
(with a and h terms) representing the fixed-fitness com-
ponent of selection acting on the population that stems from
independent sources in the environment, and the second
(with 3 and R terms) representing the frequency-dependent
component of selection resulting from the population’s niche
construction. Here, 3 (K1%3%1) is a parameter that weights
the relative importance of the two sources of selection and
R (0%R%1) is the frequency of the resource altered through
niche construction.)
E(a1) e(a2)
AA(h1) W11Za1h1C3Rpffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi W12Za2h1C3R
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Aa(1) W21 Z a1 C 3 ðRð1KRÞÞ W22 Z a2 C 3 ðRð1KRÞÞ
aa(h2) W31Za1h2C3(1KR) W32Za2h2C3(1KR)
above sexual selection model, gene–culture niche-
construction models with unbiased cultural trans-
mission simplify to the equivalent purely genetic
niche-construction models. However, the case of
unbiased cultural transmission with overdominance at
the A locus, has some interesting features: for example,
curves of stable fully polymorphic equilibria are found
that exhibit monotonic relationships between the
frequencies of the cultural trait and the alleles at the
A locus, similar to those found with the purely genetic
models when selection operates at the A locus (Laland
et al. 1999). Such curves might represent situations
similar to that of the effect of yam cultivation (the
cultural niche-constructing trait or E ) on the frequency
of the sickle-cell allele (the allele maintained through
overdominance or A) and illustrate the sensitivity of
allele frequencies to cultural niche construction.
Biased cultural transmission frequently increased
the range of parameter space over which niche
construction has an impact. For instance, in the face
of external selection favouring allele A, cultural
transmission may generate counter selection that
increases the likelihood of fixation on a. Similarly,
cultural niche construction will increase the chance of
convergence to equilibria polymorphic for A and a, if
cultural transmission favours E when an increase in the
amount of the resource results in a decrement in the
fitness of genotypes containing A (e is negative). In both
cases, cultural niche construction is driving genetic
evolution. Because cultural processes typically operate
on a faster timetable than natural selection, biased
cultural transmission is likely to have a much greater
influence on the consequences of niche construction
than would natural selection on E. These findings
illustrate processes by which cultural niche construc-
tion may have played an instrumental and active role in
hominid evolution, initiating novel evolutionary events
through the creation of novel selection pressures,
and changing the direction of evolution by modifying
established selection pressures. Moreover, they
confirm the hypothesis that the hominid capacity for
niche construction is likely to have been greatly
enhanced by, and coevolved with, a capacity for
cultural transmission.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Exploring gene–culture interactions
Weak transmission biases favouring a cultural niche-
constructing behaviour can also generate interesting
evolutionary scenarios. For instance, if transmission
bias results in a change in frequency of cultural niche-
constructing traits, then selection at the A locus may be
modified or even reversed, as R may have increased or
decreased beyond the RZ0.5 switch point. In the case
of weak biases, there may be many more generations of
selection favouring one of the alleles at the A locus than
would be the case for strong biases before selection
switches to favour the other allele, and as a conse-
quence one or other allele may reach a very low
frequency before increasing in frequency again. In
reality, small populations that follow this trajectory may
lose genetic variation at the A locus before selection
could favour the allele that had previously been
selected against. This type of process could easily
create and maintain genetic differences between semi-
isolated populations, and in hominids may have played
a role in biological speciation events.
If cultural transmission and natural selection on E
conflict, there are circumstances under which cultural
transmission can overwhelm selection. If the two
processes act in concert, cultural transmission accel-
erates the rate at which the cultural trait spreads.
When the amount of the resource is a function of
more than one generation of niche construction, the
analysis reveals time lags at the A locus in response to a
change in selection pressures caused by the spread of
the E trait, as were observed in the population genetic
models. Typically, the time lags are shorter than in the
case of the purely genetic systems, principally because
the cultural trait reaches equilibrium faster than an
analogous genetic trait. It is only if there is no selection
and weak transmission bias that time lags of the order
seen in the genetic models are observed. With
incomplete transmission, neither E nor e goes to
fixation, but provided a cultural transmission bias
favours trait E, A will eventually fix. Here, a cultural
niche-constructing trait only has to spread through the
population enough to increase the frequency of the
resource R above 0.5 before it can generate selection
that will fix A.
As with gene-based niche construction (Laland et al.
1996, 1999), these models demonstrate that cultural
niche construction will commonly generate counter
selection that compensates for, or counteracts, a
natural selection pressure in the environment. A
reasonable inference from such findings would be that
competent niche constructors should be more resistant
to genetic evolution in response to autonomously
changing environments than less able niche construct-
ors. As culture enhances the capacity of humans to alter
their niches, it would seem plausible to infer that
hominid niche construction, in general, has been more
flexible than that of other mammals.
This finding can be used to develop a number of
predictions about human evolution. For instance, one
might expect hominids to show less of an evolutionary
response in morphology to fluctuating climates than
other mammals, assuming that the latter must have
been less well equipped than the former to invest in
counteractive niche construction. Similarly, more
technologically advanced hominids should exhibit less
Phil. Trans. R. Soc. B (2008)
K. N. Laland 3585
of a response to climates than less technologically
advanced hominids.
Moreover, it should also be possible to reverse this
inference and use the fossil record to draw conclusions
about the niche-constructing capabilities of animals,
including hominids. Here, the greater the phenotypic
(as opposed to extended phenotypic) response to
environmental change by hominids, the more restricted
must have been their capacity for niche construction. If
hominids have evolved more in response to self-
constructed selection pressures than other mammals
and less in response to selection pressures that stem
from independent factors in their environment, then
hominid populations may have become increasingly
divorced from local ecological pressures. Support for
this line of reasoning comes from Guglielmino et al.’s
(1995) study of variation in cultural traits among 277
contemporary African societies in which most traits
examined correlated with cultural (e.g. linguistic)
history rather than ecology.
In the light of these findings, the view that modern
human populations are adapted to an ancestral
Pleistocene habitat, or environment of evolutionary
adaptedness, is likely to be misleading because it treats
humans as passive victims of selection rather than as
potent niche constructors (Laland & Brown 2006).
Our recent evolutionary history may well reflect our
capacity continuously to create solutions to self-
imposed problems caused by prior niche construction.
This adaptability may mean that, rather than being
adapted to a particular environment, humans adapted
to a broad range of potential environments that they
and their ancestors were involved in modifying.
In summary, the analysis suggests that where
cultural traits are transmitted in an unbiased fashion
from parent to offspring, cultural niche construction
will have a similar effect to gene-based niche construc-
tion, but cultural transmission biases favouring parti-
cular cultural traits may increase the range of parameter
space over which niche construction has an impact.
The analysis also reveals circumstances under which
cultural transmission can overwhelm natural selection,
accelerate the rate at which a favoured gene spreads,
initiate novel evolutionary events and trigger hominid
speciation. Because cultural processes typically operate
faster than natural selection, cultural niche construc-
tion probably has more profound consequences than
gene-based niche construction, and is likely to have
played an important role in human evolution.
It can be seen that niche construction changes the
evolutionary process in fundamental ways, by creating
an ecological inheritance, by modifying phenotypes,
norms of reaction and heritabilities, and by allowing
acquired characters to play a significant role in
evolution. While the niche-construction perspective is
controversial (Laland et al. 2004), and could not yet be
regarded as mainstream opinion, there are reasons to
anticipate that it will be less contentious and more
readily acceptable to human and social scientists than
the conventional perspective. After all, it is quite
apparent that human niche construction is highly
potent. Moreover, social scientists are rarely content
to describe human behaviour as fully determined by
naturally selected genes, and typically view humans as
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3586 K. N. Laland Exploring gene–culture interactions
active, constructive agents rather than passive recipi-
ents of selection. To be aligned with this viewpoint,
evolutionary biology must explicitly recognize the
changes that humans bring about in their world to be
drivers of evolutionary events.
3. CONCLUSIONS
The three case studies, together with other gene–culture
coevolutionary analyses from my laboratory (Kumm
et al. 1994; Laland et al. 1995b; Mesoudi & Laland
2007), and the aforementioned genetic data, provide a
number of general insights into human behaviour and
evolution. Collectively, these present a very different
view of human evolution from the prevailing perspective
within the dominant school of evolutionary psychology
(Barkow et al. 1992; Pinker 1997).
(a) Genes and culture coevolve
Genes and culture can, and do, coevolve. Theoretical
models, such as the handedness, sexual selection and
niche-construction case studies, illustrate the mechan-
isms, while genetic, anthropological and archaeological
data demonstrate that the coevolutionary dynamic
is not just a hypothetical possibility, but a reality.
Genetic and cultural change can occur on similar time
scales. Analysis of the human genome implies that
gene–culture coevolutionary interactions are likely to
be pervasive.
(b) The gene–culture leash tugs both ways
Edward Wilson famously claimed ‘the genes hold
culture on a leash’ (1978, p. 172), by which he meant
that genetic propensities shape the acquisition of
cultural knowledge. In fact, Wilson failed to emphasize
that, for our species at least, the leash tugs both ways.
Culture may be shaped by genes, but the architecture
of the human genome has been profoundly shaped by
culture, as the aforementioned genetic data attest.
Human culture and technology are amply manifest in
our species’ extraordinarily potent capacity for niche
construction, and have shaped the selective landscape
of human evolution.
(c) Gene–culture coevolution may be the
dominant form of evolution for our species
Theoretical gene–culture models consistently find that
the gene–culture dynamics are typically faster, strong-
er, operate over a broader range of conditions and are
more potent than conventional evolutionary dynamics.
Gene–culture coevolution is likely to be the dominant
form of evolutionary adaptation for our species. By
modifying selection pressures and increasing the
intensity of selection, cultural processes can speed up
evolution; by providing an alternative means of
responding to ecological and social challenges, cultural
processes can damp out selection and slow down the
evolutionary response. Extensive evolutionary
responses to cultural niche construction in our species
are likely to mean that human minds are specifically
adapted for culture.
Phil. Trans. R. Soc. B (2008)
(d) Culture is a potent co-director of
evolutionary events
Cultural processes are every bit as influential as genetic
processes in gene–culture coevolution. Theoretical
analyses reveal many instances where cultural trans-
mission overwhelms, or reverses, natural selection.
Moreover, the observed patterns of selection often
depend intimately on the cultural details. For instance,
whether female-biased infanticide selects for male- or
female-biased sex ratio distorters depends on the
culturally transmitted rules that individuals adopt
(Kumm et al. 1994; Laland et al. 1995b).
(e) Humans are active constructors of their
selective environments
Humans are not passive victims of natural selection, but
active constructors of major components of their
selective environments. While niche construction is
universal (Odling-Smee et al. 2003), our species’
capacity to control, regulate and transform the environ-
ment is uniquely powerful, largely due to our capacity
for culture. Theoretical analyses regularly reveal coevo-
lutionary dynamics in which human cultural processes
can hitch-hike to fixation on the selection they generate
(Laland 1994; Silver & Di Paolo 2006). It may be no
coincidence that humans, the species most reliant on
culture, have the most potent capability for niche
construction (Laland et al. 2000); autocatalytic and
runaway effects may have fuelled ever more powerful
niche construction in our lineage. As reliance on social
learning covaries with relative brain size in primates
(Reader & Laland 2002), such autocatalytic dynamics
may have played a critical role in brain evolution.
(f ) Humans do not have stone-age minds
Humans are not primarily adapted to ancestral rather
than current environments, as some evolutionary
psychologists suggest. When humans engage in niche
construction they do not do so randomly; in the same
way as other animals, they build structures and have
other impacts on their world that are often ‘extended
phenotypes’ (Dawkins 1982), adaptations that enhance
fitness. Animals also deplete resources and pollute
environments, but this too increases fitness in the short
term and is often tied to life-history strategies that take
account of this activity, for instance through dispersal or
migration when resource levels are low or the environ-
ment becomes uninhabitable. While niche construction
can have negative effects on fitness, Odling-Smee et al.
(2003) are explicit about their expectation that most
niche construction will increase the short-term fitness of
the constructor, although it may have negative con-
sequences for other species. This is hardly contentious:
the fitness benefits of animal artefacts are well
documented. Niche construction is typically functional
and adaptive because it is informed, but not determined,
by genes, and sometimes also by learning and culture.
Humans largely construct their world to suit themselves,
leaving human behaviour largely adaptive in spite of the
transformations they have brought about in the environ-
ment (Laland & Brown 2006).
This adaptiveness is reinforced by two further
processes (Laland & Brown 2006). First, humans
frequently buffer any adaptive lag through further
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Exploring gene–culture interactions
cultural niche construction; for instance clothing, fires
and air-conditioning buffer extremes of temperature,
while new agricultural practices and innovations
alleviate food shortages. Such cultural responses
eradicate any mismatch between human selective
environments and human genotypes. Second, where
this does not occur, natural selection ensues, and
recent estimates of typical rates of selection suggest that
significant human evolution could occur in hundreds of
years (Kingsolver et al. 2001). Among the malaria-rife
regions of the Kwa homeland, being a heterozygote for
the sickle-cell S allele is adaptive. Similarly, in dairying
societies, genes expressed in high lactase activity pay
fitness dividends.
Humans do not walk the streets of the urban
metropolis left hopelessly unable to cope by the
ancestral primate or stone-age brains in their heads,
as evolutionary psychologists (and others) have
claimed (Morris 1967; Buss 1999). Human minds
and human environments have been engaged in a long-
standing, intimate exchange of information, mediated
by reciprocal bouts of niche construction and natural
selection, leaving each beautifully fashioned in the
other’s image.
I am indebted to the Human Frontier Science Programme,
the BBSRC and the Royal Society for providing financial
support for the above projects, and to John Odling-Smee,
Gillian Brown and two anonymous reviewers for their helpful
comments on earlier drafts.
REFERENCES
Aoki, K. & Feldman, M. W. 1997 A gene-culture coevolu-
tionary model for brother-sister mating. Proc. Natl Acad.
Sci. USA 94, 13 046–13 050. (doi:10.1073/pnas.94.24.
13046)
Annett, M. 1985 Left, right, hand and brain: the right shift
theory. London, UK: Earlbaum.
Aronson, E. 1995 Social psychology. New Jersey, NJ: Prentice
Hall.
Balter, M. 2005 Are humans still evolving? Science 309,
234–237. (doi:10.1126/science.309.5732.234)
Barkow, J. H., Cosmides, L. & Tooby, J. 1992 The adapted
mind: evolutionary psychology and the generation of culture.
Oxford, UK: Oxford University Press.
Barrett, L., Dunbar, R. & Lycett, J. 2001 Human evolutionary
psychology. London, UK: Macmillan.
Bishop, D. V. M. 1990 Handedness and developmental disorder.
Hove, UK: Earlbaum.
Borenstein, E., Kendal, J. & Feldman, M. 2006 Cultural
niche construction in a metapopulation. Theor. Popul. Biol.
70, 92–104. (doi:10.1016/j.tpb.2005.10.003)
Boyd, R. & Richerson, P. J. 1985 Culture and the evolutionary
process. Chicago, IL: Chicago University Press.
Burger, J., Kirchner, M., Bramanti, B., Haak, W. & Thomas,
M. G. 2007 Absence of the lactase-persistence-associated
allele in early Neolithic Europeans. Proc. Natl Acad. Sci.
USA 104, 3736–3741. (doi:10.1073/pnas.0607187104)
Buss, D. M. 1994 The evolution of desire: strategies of human
mating. New York, NY: HarperCollins.
Buss, D. M. 1999 Evolutionary psychology. The new science of
the mind. London, UK: Allyn & Bacon.
Cavalli-Sforza, L. L. & Feldman, M. W. 1973 Models for
cultural inheritance. I. Group mean and within group
variation. Theor. Popul. Biol. 4, 42–55. (doi:10.1016/0040-
5809(73)90005-1)
Phil. Trans. R. Soc. B (2008)
K. N. Laland 3587
Cavalli-Sforza, L. L. & Feldman, M. W. 1981 Cultural
transmission and evolution: a quantitative approach. Prince-
ton, NJ: Princeton University Press.
Corballis, M. C. 1991 The lopsided ape. Oxford, UK: Oxford
University Press.
Cosmides, L. & Tooby, J. 1987 From evolution to behavior:
evolutionary psychology as the missing link. In The latest on
the best: essays on evolution and optimality (eds L. Cosmides
& J. Tooby), pp. 277–306. Cambridge, MA: MIT Press.
Craig, P. L., Swinburn, B. A., Matenga-Smith, T., Matangi,
H. & Vaughn, G. 1996 Do Polynesians still believe that big
is beautiful? Comparison of body size and preferences of
Cook Island Maori and Australians. New Zeal. Med.
J. 109, 200–203.
Darwin, C. 1871 The descent of man and selection in relation to
sex. London, UK: John Murray. (1st edn. Reprinted by
Princeton University Press, Princeton, NJ 1981)
Dawkins, R. 1982 The extended phenotype. Oxford, UK:
Oxford University Press.
Dorus, S., Vallender, E. J., Evans, P. D., Anderson, J. R.,
Gilbert, S. L., Mahowald, M., Wyckoff, G. J., Malcom,
C. M. & Lahn, B. T. 2004 Accelerated evolution of
nervous system genes in the origin of Homo sapiens. Cell
119, 1027–1040. (doi:10.1016/j.cell.2004.11.040)
Dugatkin, L. A. 1992 Sexual selection and imitation: females
copy the mate choice of others. Am. Nat. 139, 1384–1489.
(doi:10.1086/285392)
Durham, W. H. 1991 Coevolution: genes, culture and human
diversity. Stanford, CA: Stanford University Press.
Enquist, M., Eriksson, K. & Ghirlanda, S. 2007 Critical social
learning. A solution to Roger’s paradox of nonadaptive
culture. Am. Anthropol. 109, 727–734. (doi:10.1525/aa.
2007.109.4.727)
Faurie, C. & Raymond, M. 2005 Handedness, homicide and
negative frequency-dependent selection. Proc. R. Soc. B
272, 25–28. (doi:10.1098/rspb.2004.2926)
Feldman, M. W. & Cavalli-Sforza, L. L. 1976 Cultural and
biological evolutionary processes, selection for a trait
under complex transmission. Theor. Popul. Biol. 9,
238–259. (doi:10.1016/0040-5809(76)90047-2)
Feldman, M. W. & Cavalli-Sforza, L. L. 1989 On the theory of
evolution under genetic and cultural transmission with
application to the lactose absorption problem. In Mathemat-
ical evolutionary theory (ed. M. W. Feldman), pp. 145–173.
Princeton, NJ: Princeton University Press.
Feldman, M. W. & Laland, K. N. 1996 Gene-culture
coevolutionary theory. Trends Ecol. Evol. 11, 453–457.
(doi:10.1016/0169-5347(96)10052-5)
Feldman, M. W., Aoki, K. & Kumm, J. 1996 Individual
versus social learning. Anthropol. Sci. 104, 209–232.
Ford, C. S. & Beach, F. A. 1951 Patterns of sexual behavior.
New York, NY: Harper.
Furnham, A. & Baguma, P. 1994 Cross-cultural differences in
the evaluation of male and female body shapes. Int.
J. Eating Disord. 15, 81–89. (doi:10.1002/1098-108X(19
9401)15:1!81::AID-EAT2260150110O3.0.CO;2-D)
Galef Jr, B. G. & Laland, K. N. 2005 Social learning in
animals: empirical studies and theoretical models.
Bioscience 55, 489–499. (doi:10.1641/0006-3568(2005)
055[0489:SLIAES]2.0.CO;2)
Gould, J. L. & Gould, C. G. 1989 Sexual selection. New York,
NY: Scientific American Library.
Guglielmino, C. R., Viganotti, C., Hewlett, B. & Cavalli-
Sforza, L. L. 1995 Cultural variation in Africa: role of
mechanism of transmission and adaptation. Proc. Natl
Acad. Sci. USA 92, 7585–7589. (doi:10.1073/pnas.92.16.
7585)
Hardyck, C., Petriovich, L. & Goldman, R. 1976 Left
handedness and cognitive deficit. Cortex 12, 266–278.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3588 K. N. Laland Exploring gene–culture interactions
Harris, L. J. 1980 Left handedness: early theories, facts and
fancies. In Neuropsychology of left handedness (ed. J. Herron),
pp. 3–78. London, UK: Academic Press.
Hawley, W. A., Reiter, P., Copeland, R. S., Pumpuni, C. B. &
Craig Jr, G. B. 1987 Aedes albopictus in North America:
probable introduction in used tires from northern Asia.
Science 236, 1114–1116. (doi:10.1126/science.3576225)
Hewlett, B. S. & Cavalli-Sforza, L. L. 1986 Cultural
transmission among Aka pygmies. Am. Anthropol. 88,
922–934. (doi:10.1525/aa.1986.88.4.02a00100)
Heyes, C. M. & Galef, B. G. 1996 Social learning in animals:
the roots of culture. London, UK: Academic Press.
Hill, R. S. & Walsh, C. A. 2005 Molecular insights into
human brain evolution. Nature 437, 64–67. (doi:10.1038/
nature04103)
Holden, C. & Mace, R. 1997 Phylogenetic analysis of the
evolution of lactose digestion in adults. Hum. Biol. 69,
605–628.
Hung, C. C., Tu, Y. K., Chen, S. H. & Chen, R. C. 1985 A
study of handedness and cerebral speech dominance in
right-handed Chinese. J. Neuroling. 1, 143–163. (doi:10.
1016/S0911-6044(85)80007-5)
Ihara, Y. & Feldman, M. W. 2004 Cultural niche construction
and the evolution of small family size. Theor. Popul. Biol.
65, 105–111. (doi:10.1016/j.tpb.2003.07.003)
Jones, B. C., DeBruine, L. M., Little, A. C., Burriss, R. P. &
Feinberg, D. R. 2007 Social transmission of face prefer-
ences among humans. Proc. R. Soc. B 274, 899–903.
(doi:10.1098/rspb.2006.0205)
Kingsolver, J. G., Hoekstra, H. E., Hoekstra, J. M., Berrigan,
D., Vignieri, S. N., Hill, C. E., Hoang, A., Gilbert, P. &
Beerli, P. 2001 The strength of phenotypic selection in
natural populations. Am. Nat. 157, 245–261. (doi:10.
1086/319193)
Kirkpatrick, M. 1982 Sexual selection and the evolution
of female choice. Evolution 36, 1–12. (doi:10.2307/
2407961)
Kirkpatrick, M. & Dugatkin, L. A. 1994 Sexual selection and
the evolutionary effects of copying mate choice. Behav.
Ecol. Sociobiol. 34, 443–449. (doi:10.1007/BF00167336)
Kirkpatrick, M. & Lande, R. 1989 The evolution of maternal
characters. Evolution 43, 485–503. (doi:10.2307/2409054)
Kumm, J., Laland, K. N. & Feldman, M. W. 1994 Gene-
culture coevolution and sex ratios: the effects of infanti-
cide, sex-selective abortion, and sex-biased parental
investment on the evolution of sex ratios. Theor. Popul.
Biol. 46, 249–278. (doi:10.1006/tpbi.1994.1027)
Laland, K. N. 1992 A theoretical investigation of the role of
social transmission in evolution. Ethol. Sociobiol. 13,
87–113. (doi:10.1016/0162-3095(92)90020-5)
Laland, K. N. 1994 On the evolutionary consequences of
sexual imprinting. Evolution 48, 477–489. (doi:10.2307/
2410106)
Laland, K. N. & Brown, G. R. 2002 Sense and nonsense.
Evolutionary perspectives on human behaviour. Oxford, UK:
Oxford University Press.
Laland, K. N. & Brown, G. R. 2006 Niche construction,
human behaviour and the adaptive lag hypothesis. Evol.
Anthropol. 15, 95–104. (doi:10.1002/evan.20093)
Laland, K. N., Kumm, J., Van Horn, J. D. & Feldman, M. W.
1995a A gene-culture model of handedness. Behav. Genet.
25, 433–445. (doi:10.1007/BF02253372)
Laland, K. N., Kumm, J. & Feldman, M. W. 1995b Gene-
culture coevolutionary theory: a test case. Curr. Anthropol.
36, 131–156. (doi:10.1086/204346)
Laland, K. N., Odling-Smee, F. J. & Feldman, M. W. 1996
On the evolutionary consequences of niche construction.
J. Evol. Biol. 9, 293–316. (doi:10.1046/j.1420-9101.1996.
9030293.x)
Phil. Trans. R. Soc. B (2008)
Laland, K. N., Odling-Smee, F. J. & Feldman, M. W. 1999
Evolutionary consequences of niche construction and
their implications for ecology. Proc. Natl Acad. Sci. USA
96, 10 242–10 247. (doi:10.1073/pnas.96.18.10242)
Laland, K. N., Odling-Smee, J. & Feldman, M. W. 2000
Niche construction, biological evolution, and cultural
change. Behav. Brain Sci. 23, 131–175. (doi:10.1017/
S0140525X00002417)
Laland, K. N., Odling-Smee, F. J. & Feldman, M. W. 2001
Cultural niche construction and human evolution.
J. Evol. Biol. 14, 22–33. (doi:10.1046/j.1420-9101.2001.
00262.x)
Laland, K. N., Odling-Smee, F. J. & Feldman, M. W. 2004
Causing a commotion. Niche construction: do the
changes that organisms make to their habitats transform
evolution and influence natural selection? Nature 429, 609.
(doi:10.1038/429609a)
Lewontin, R. C. 1982 Organism and environment. In
Learning, development and culture (ed. H. C. Plotkin),
pp. 151–170. New York, NY: Wiley.
Lewontin, R. C. 1983 Gene, organism, and environment.
In Evolution from molecules to men (ed. D. S. Bendall),
pp. 273–285. Cambridge, UK: Cambridge University
Press.
Little, A. C., Burriss, R. P., Jones, B. C., DeBruine, L. M. &
Caldwell, C. C. 2008 Social influence in human face
preference: men and women are influenced more for long-
term than short-term attractiveness decisions. Evol. Hum.
Behav. 29, 140–146. (doi:10.1016/j.evolhumbehav.2007.
11.007)
Marlowe, F. & Wetsman, A. 2001 Preferred waist-to-hip ratio
and ecology. Pers. Indiv. Differ. 30, 481–489. (doi:10.1016/
S0191-8869(00)00039-8)
McElreath, R., Bell, A. V., Efferson, C., Lubell, M.,
Richerson, P. J. & Waring, T. 2008 Beyond existence and
aiming outside the laboratory: estimating frequency-
dependent and pay-off-biased social learning strategies.
Phil. Trans. R. Soc. B 363, 3515–3528. (doi:10.1098/rstb.
2008.0131)
McManus, I. C. 1985 Handedness, language dominance and
aphasia. Psychological Medicine Monograph Supplement 8.
Cambridge, UK: Cambridge University Press.
Mesoudi, A. & Laland, K. N. 2007 Culturally transmitted
paternity beliefs and the evolution of human mating
behaviour. Proc. R. Soc. B 274, 1273–1278. (doi:10.1098/
rspb.2006.0396)
Morgan, M. J. & Corballis, M. C. 1978 The inheritance of
laterality. Behav. Brain Sci. 2, 270–277.
Morris, D. 1967 The naked ape. London, UK: Vintage.
Neale, M. C. 1988 Handedness in a sample of volunteer twins.
Behav. Genet. 18, 69–79. (doi:10.1007/BF01067076)
Nielsen, R., Hellmann, I., Hubisz, M., Bustamante, C. &
Clark, A. G. 2007 Recent and ongoing selection in the
human genome. Nat. Rev. Genet. 8, 857–868. (doi:10.
1038/nrg2187)
Odling-Smee, F. J. 1988 Niche constructing phenotypes.
In The role of behavior in evolution (ed. H. C. Plotkin),
pp. 73–132. Cambridge, MA: MIT Press.
Odling-Smee, F. J., Laland, K. N. & Feldman, M. W. 2003
Niche construction. The neglected process in evolution.
Monographs in Population Biology. 37. Princeton, NJ:
Princeton University Press.
Otto, S. P., Christiansen, F. B. & Feldman, M. W. 1995
Genetic and cultural inheritance of continuous traits. Morrison
Institute for Population and Resource Studies. Paper no. 64.
Stanford, CA: Stanford University Press.
Palmer, A. R. 2002 Chimpanzee right-handedness recon-
sidered: evaluating the evidence with funnel plots.
Am. J. Phy. Anthropol. 118, 191–199. (doi:10.1002/ajpa.
10063)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Exploring gene–culture interactions
Pinker, S. 1997 How the mind works. London, UK: Penguin.
Pollard, K. S. et al. 2006 An RNA gene expressed during
cortical development evolved rapidly in humans. Nature
443, 167–172. (doi:10.1038/nature05113)
Reader, S. M. & Laland, K. N. 2002 Social intelligence,
innovation, and enhanced brain size in primates. Proc. Natl
Acad. Sci. USA 99, 4436–4441. (doi:10.1073/pnas.
062041299)
Richerson, P. J. & Boyd, R. 2005 Not by genes alone. Chicago,
IL: Chicago University Press.
Rogers, A. R. 1988 Does biology constrain culture? Am.
Anthropol. 90, 819–831. (doi:10.1525/aa.1988.90.4.02a
00030)
Sabeti, P. C. et al. 2006 Positive natural selection in the
human lineage. Science 312, 1614–1620. (doi:10.1126/
science.1124309)
Sabeti, P. C. et al. 2007 Genome-wide detection and
characterization of positive selection in human popu-
lations. Nature 449, 913–918. (doi:10.1038/nature06250)
Silver, M. & Di Paolo, E. 2006 Spatial effects favour the
evolution of niche construction. Theor. Popul. Biol. 20,
387–400. (doi:10.1016/j.tpb.2006.08.003)
Smith, B. 2007 Human niche construction and the
behavioural context of plant and animal domesti-
cation. Evol. Anthropol. 16, 188–199. (doi:10.1002/
evan.20135)
Su, C. H., Kuo, P. H., Lin, C. C. H. & Chen, W. J. 2005 A
school-based twin study of handedness among adolescents
in Taiwan. Behav. Genet. 35, 723–733. (doi:10.1007/
s10519-005-6189-1)
Tan, A. S. 1979 TV beauty ads and role expectations of
adolescent female viewers. J. Quart. 56, 283–288.
Teng, E. L., Lee, P., Yang, P. C. & Chang, P. C. 1976
Handedness in a Chinese population: biological, social
and pathological factors. Science 193, 1148–1150. (doi:10.
1126/science.986686)
Toth, N. 1985 Archaeological evidence for preferential right-
handedness in the Lower and Middle Pleistocene, and its
possible implications. J. Hum. Evol. 14, 607–614. (doi:10.
1016/S0047-2484(85)80087-7)
Phil. Trans. R. Soc. B (2008)
K. N. Laland 3589
Tovee, M. J., Swami, V., Furnham, A. & Mangalparsad, R.
2006 Changing perceptions of attractiveness as observers
are exposed to a different culture. Evol. Hum. Behav. 27,
443–456. (doi:10.1016/j.evolhumbehav.2006.05.004)
Uomini, N. T. In press. Handedness in Neanderthals. In
Neanderthal lifeways, subsistence and technology (eds N. J.
Conard & J. Richter).
Voight, B. F., Kudaravalli, S., Wen, X. & Pritchard, J. K.
2006 A map of recent position selection in the human
genome. PLoS Biol. 4, e72. (doi:10.1371/journal.pbio.
0040072)
Wang, E. T., Kodama, G., Baldi, P. & Moyzis, R. K. 2006
Global landscape of recent inferred Darwinian selection
for Homo sapiens. Proc. Natl Acad. Sci. USA 103, 135–140.
(doi:10.1073/pnas.0509691102)
Wang, H. Y. et al. 2007 Rate of evolution of brain-expressed
genes in humans and other primates. PLoS Biol. 5,
e130335–e130342. (doi:10.1371/journal.pbio.0050013)
Warren, D. M., Stern, M., Duggirala, R., Dyer, T. D. &
Almasy, L. 2006 Heritability and linkage analysis of hand,
foot, and eye preference in Mexican Americans. Laterality
11, 508–524. (doi:10.1080/13576500600761056)
Wetsman, A. & Marlowe, F. 1999 How universal are
preferences for female waist-to-hip ratios? Evidence from
the Hadza of Tanzania. Evol. Hum. Behav. 20, 219–228.
(doi:10.1016/S1090-5138(99)00007-0)
White, D. J. & Galef Jr, B. G. 2000 ‘Culture’ in quail: social
influences on mate choices of female Coturnix japonica.
Anim. Behav. 59, 975–979. (doi:10.1006/anbe.1999.1402)
Williamson, S. H., Hubisz, M. J., Clark, A. G., Payseur,
B. A., Bustamante, C. D. & Nielsen, R. 2007 Localizing
recent adaptive evolution in the human genome. PLoS
Genet. 3, e90. (doi:10.1371/journal.pgen.0030090)
Wilson, E. O. 1978 On human nature. Cambridge, MA:
Harvard University Press.
Yu, D. W. & Shepard, G. H. 1998 Is beauty in the eye
of the beholder? Nature 396, 321–322. (doi:10.1038/
24512)
Zentall, T. R. & Galef, B. G. (eds) 1988 Social learning:
psychological and biological perspectives. Hillsdale, NJ:
Erlbaum.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Phil. Trans. R. Soc. B (2008) 363, 3591–3603

doi:10.1098/rstb.2008.0145
Published online 19 September 2008

Cultural evolution: implications for understanding

the human language faculty and its evolution
Kenny Smith1,* and Simon Kirby2
1
Cognition and Communication Research Centre, Division of Psychology, Northumbria University,
Northumberland Building, Northumberland Road, Newcastle NE1 8ST, UK
2
Language Evolution and Computation Research Unit, School of Philosophy, Psychology and Language
Sciences, University of Edinburgh, Dugald Stewart Building, 3 Charles Street, Edinburgh, EH8 9AD, UK
Human language is unique among the communication systems of the natural world: it is socially
learned and, as a consequence of its recursively compositional structure, offers open-ended
communicative potential. The structure of this communication system can be explained as a
consequence of the evolution of the human biological capacity for language or the cultural evolution
of language itself. We argue, supported by a formal model, that an explanatory account that involves
some role for cultural evolution has profound implications for our understanding of the biological
evolution of the language faculty: under a number of reasonable scenarios, cultural evolution can
shield the language faculty from selection, such that strongly constraining language-specific learning
biases are unlikely to evolve. We therefore argue that language is best seen as a consequence of
cultural evolution in populations with a weak and/or domain-general language faculty.
Keywords: language; communication; language faculty; cultural evolution; biological evolution

1. INTRODUCTION human capacity for language. Specifically, as we will

When compared with other animals, humans strike us
as special. First, we are highly cultural—while culture
appears not to be unique to humans (Whiten 2005), its
ubiquity in human society and human cognition is
highly distinctive. Second, humans have a unique
communication system, language. Language differs
from the communication systems of non-human
animals along a number of fairly well-defined dimen-
sions, to be elucidated below.
The first main contention of this article is that the
co-occurrence of these two unusual properties is not
coincidental: they are causally related. The view that
language facilitates human culture is not a new one: it is
for this reason that Maynard Smith & Szathmáry (1995)
described language as the most recent major evolution-
ary transition in the history of life on Earth. However, we
will argue that the relationship works in the other
direction as well—at least some of the distinctive features
of human language are adaptations to its cultural
transmission, and cultural evolution potentially plays a
major role in explaining why the human communication
system has the particular features that it does. The most
extreme form of this argument, to be expounded in §5, is
that a communication system looking a lot like human
language is a natural and inevitable consequence of
cultural evolution in populations of a particular sort of
social learner.
Our second main contention is that a serious
consideration of cultural evolution radically changes
the kinds of evolutionary stories we can tell about the
* Author for correspondence (kenny.smith@northumbria.ac.uk).
One contribution of 11 to a Theme Issue ‘Cultural transmission and
the evolution of human behaviour’.
argue in §4, cultural evolution potentially shields the
human language faculty from selection, ruling out the
evolution of a strongly constraining and domain-
specific language faculty in our species.
2. LANGUAGE DESIGN
What is special about language? In an early attempt to
answer this question, Hockett (1960) identified 13
design features of language. Of particular relevance are
the following three features, whose conjunction (to a
first approximation) distinguish language from the
communication systems of all other animals.
—Semanticity: ‘there are relatively fixed associations
between elements of messages (e.g. words) and
recurrent features or situations of the world around
us’ (Hockett 1960, p. 6).
—Productivity: ‘[language provides] the capacity to say
things that have never been said or heard before and yet
to be understood by other speakers of the language’
(Hockett 1960, p. 6).
—Traditional (i.e. cultural) transmission: ‘Human genes
carry the capacity to acquire a language, and probably
also a strong drive towards such acquisition, but the
detailed conventions of any one language are trans-
mitted extragenetically by learning and teaching’
(Hockett 1960, p. 6).
In isolation, each of these features is not particularly
rare. Limited non-productive semantic communication
systems are common in the natural world, notably in the
alarm-calling behaviours of various species (e.g. diverse
species of bird and monkey; Marler 1955; Cheney &
Seyfarth 1990; Evans et al. 1993; Zuberbuhler 2001).

3591 This journal is q 2008 The Royal Society

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3592 K. Smith & S. Kirby Human language faculty and its evolution
However, such semantic communication systems are
not traditionally transmitted: the consensus view is that
there is no vocal learning of the form of alarm calls in
such systems, although there may well be a role for
learning in establishing the precise situations under
which calls must be produced, or how calls must be
responded to (usage and comprehension learning,
respectively: Slater 2005).
Traditionally transmitted communication systems
also exist in non-humans—in mammals (seals, bats,
whales and dolphins) and in birds (most notably the
oscine passerines, the songbirds). Aspects of the
production of communicative signals in all of these
groups show sensitivity to input and (particularly in the
case of songbirds and whales) patterns of local dialects
characteristic of cultural transmission. Furthermore,
some bird song (e.g. that of the starling, willow warbler
and Bengalese finch; Eens 1997; Gil & Slater 2000;
Okanoya 2004) is productive, to the extent that these
songs are constructed according to rules that generate
several possible songs with the same underlying
structure. However, none of these systems rise above
more than a superficial degree of semanticity: this is
clearest in the songbirds, where song seems to serve a
dual function as a means of attracting mates and
repelling rivals (Catchpole & Slater 1995). Indeed, the
same song being sung by a particular individual can be
differentially interpreted depending on the identity of
the listener, with female listeners interpreting it as
sexual advertisement and males interpreting it as
territory defence.
Finally, traditionally transmitted and semantic
communication systems seem to exist, to a limited
extent, in the gestural communication systems of our
nearest extant relatives, the apes. Pairs of apes develop,
through repeated interactions, meaningful gestures
that can be subsequently used to communicate about
various situations (feeding, play, sex, etc.; Call &
Tomasello 2004). However, such systems are not
productive: each sign is underpinned by a rather
laborious history of interaction (through a process
known as ontogenetic ritualization; Tomasello 1996)
and consequently such systems cannot be expanded to
include meaningful novel signals.
While there are, in principle, several ways of
designing a productive semantic communication
system, in human language these features are under-
pinned by four subsidiary design features (two from
Hockett 1960, the others implicit in his account).
—Arbitrariness: ‘the ties between the meaningful
message elements and their meanings can be arbitrary’
(Hockett 1960, p. 6). Arbitrary signals of this sort can be
contrasted with, for example, signals that have meaning
by resemblance (icons) or signals that are causally
related to their meanings (indexes, such as the gestural
signals established by ontogenetic ritualization).
—Duality of patterning: ‘The meaningful elements in
any language (e.g. words).constitute an enormous
stock. Yet they are represented by small arrangements
of a relatively very small stock of distinguishable sounds
which are in themselves wholly meaningless’ (Hockett
1960, p. 6). Languages have a few tens of phonemes
that are combined to form tens of thousands of words.
Phil. Trans. R. Soc. B (2008)
—Recursion: a signal of a given category can contain
component parts that are of the same category, and this
embedding can be repeated without limit. For example,
‘I think he saw her’ is a sentence of English, which
contains an embedded sentence, ‘he saw her’. This
embedding can be continued indefinitely (‘I think she
said he saw her’, ‘You know I think she said he saw her’,
and so on), yielding an infinite number of sentences.
—Compositionality: the meaning of a complex signal is a
function of the meaning of its parts and the way in which
they are combined (Krifka 2001). For example, the
sentence ‘She slapped him’ consists of four component
parts—‘she’, ‘him’, ‘slap’ and ‘-ed’ (the past tense
marker). The meaning of the sentence is determined by
the meaning of these parts and the order in which they
are combined—‘She slapped him’ differs predictably in
meaning from ‘They slapped him’ and ‘She kicked him’,
but also from ‘He slapped her’, where the order of the
male and female pronouns has been reversed.
The combination of these four subsidiary features
results in a system that is productive and semantic.
Duality of patterning allows for the generation of an
extremely large set of basic communicative units from
a small inventory of discriminable sounds.1 Arbitrari-
ness allows those basic units to be mapped onto the
world in a flexible fashion, without additional con-
straints of iconicity or indexicality. Recursion allows
that large inventory of basic units to be combined to
form a truly open-ended system. Finally, compositional
structure makes the interpretation of novel utterances
possible—in a recursive compositional system, if you
know the meaning of the basic elements and the effects
associated with combining them, you can deduce the
meaning of any utterance in the system, including
infinitely many entirely novel utterances.
Again, we can ask to what extent these subsidiary
design features are realized in the communication
systems of non-human animals. Perhaps, as we might
expect, the productive systems highlighted above (e.g.
the song of certain species of bird) appear to adopt
rudiments of duality of patterning, in that they consist of
recombinable subunits (notes or syllables). Semantic
systems show limited amounts of arbitrariness (in alarm-
calling systems) and compositionality (in the boom
alarm combination call in Campbell’s monkeys, where
the preceding boom serves to change the meaning of the
subsequent alarm call, or reduces its immediacy;
Zuberbuhler 2002). Finally, there is little evidence
for recursion in the communicative behaviour of non-
humans—indeed, Hauser et al. (2002) argue that it is
entirely absent from the cognitive repertoires of all non-
human species, although this point is still a matter for
debate (Kinsella in press).
The foregoing discussion suggests that language is
unusual owing to the particular bundle of features it
possesses, and the pervasiveness of those features in
language, rather than possession of any individual
unique element. Why is language designed like this?
This at first seems like a fairly straightforward question
to answer: language is designed like this because these
design features make for a useful communication
system, specifically a system with open-ended expressiv-
ity. A population of individuals sharing such a system
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Human language faculty and its evolution
can in principle communicate with each other about
anything they chose, including survival relevant issues
such as where to find food and shelter, how to deal with
predators and prey, how social relationships are to be
managed, and so on. Furthermore, each individual only
requires a finite (and fairly small) set of cognitive
resources to achieve this expressive range—a few tens
of speech sounds, a few tens of thousands of meaningful
words created from those speech sounds and a few
hundred grammatical rules constraining the com-
bination of those words into meaningful sequences.
However, the fact that human language makes good
design sense in various ways does not explain how
language came to have these properties. To truly
answer the question ‘why is language designed like
this?’ we need to establish the mechanisms that explain
this fit between function and form: how did the
manifest advantages of such a linguistic system become
realized in language as a system of human behaviour?
We will review two potential mechanisms below:
one that explains the fit between function and form as
arising from the biological evolution of the human
language faculty (in §3a), and the other that views it as a
consequence of the cultural evolution of language itself
(in §3b). In both cases we will see how these contrasting
explanations have been applied to the specific question
of the evolution of compositionality, one of the language
design features subserving productivity. The evolution
of compositionality has received a great deal of attention
in the literature, which is why we focus on it here.
Ultimately, all these design features will require such
explanation, and a similar research effort aimed at
understanding the evolution of duality of patterning (in
both biological and cultural terms) is underway (see, e.g.
Nowak & Krakauer 1999; Nowak et al. 1999; Oudeyer
2005; Zuidema & de Boer in press). In §4, we will turn
to the issue of interactions between biological and
cultural evolutionary accounts.
3. TWO EXPLANATIONS FOR LANGUAGE
DESIGN
(a) An explanation from evolutionary biology
The uniqueness of human language must have some
biological basis—there must be some feature of human
biology that results in this unusually rich, expressive
system of communication in our species alone. One
obvious biological adaptation for vocal communication
is the unusual structure of the human vocal tract, which
provides a wide range of highly discriminable sounds
(see Fitch 2000, for review). However, language is
more than just speech: the design features picked out
above are ambivalent as to the modality of the system in
question, and while adaptations for speech are
important in that they provide a good clue as to the
age, modality and selective importance of language in
human evolution history, they are peripheral to what
we see as the fundamental design features of language.
Moving beyond the productive apparatus, then,
what is our biological endowment for language?
Linguists have approached this question via a consider-
ation of the problem of language acquisition: how do
children acquire a complex and richly structured
language with apparent ease at a relatively early stage
Phil. Trans. R. Soc. B (2008)
K. Smith & S. Kirby 3593
in their lives? The difficulty of reconciling the precocity
of language acquisition with the complexity of language
leads to the fairly prominent view that humans must be
endowed with a highly structured and constraining
language-specific mental faculty that, to a large degree,
prefigures much of the structure of language. One of
the main cornerstones of this argument is known as
the argument from the poverty of the stimulus, a term
introduced by Chomsky (1980): the data language must
be learned from lack direct evidence for a number of
features of the grammars that children must learn. If
children end up with grammars that contain features for
which they received no evidence in their input, then (the
argument goes) those features of language must be
prefigured in the acquisition device. Rather than a
flexible and open-ended process of social learning,
language acquisition is ‘better understood as the growth
of cognitive structures along an internally directed
course under the triggering and partially shaping effect
of the environment’ (Chomsky 1980, p. 34). Design
features of language are then naturally explained as
features of the internally directed course of acquisition.
This account has been so successful that it now
constitutes a scientific orthodoxy, at least in some form
(and despite a lack of empirical support for stimulus
poverty arguments; Pullum & Scholz 2002). However,
even if we accept that language is the way it is because
the language faculty forces it to be that way, this simply
pushes the question back one step: why is the language
faculty designed the way it is?
A well-established solution to such questions in
biological systems is that of evolution by natural
selection. In a highly influential article, Pinker &
Bloom (1990) argued that this solution can be applied
to language, assuming (following the argument above)
that language is to a non-trivial extent a biological
capacity: ‘It would be natural, then, to expect everyone
to agree that human language is the product of
Darwinian natural selection. The only successful
account of the origin of complex biological structure
is the theory of natural selection’ (Pinker & Bloom
1990, p. 707). To support this claim, Pinker and
Bloom provided a number of arguments that language,
in general, offers reproductive pay-offs. Some of these
arguments have been rehearsed in brief above—for
example, the open-ended communicative possibilities
that language affords might plausibly have reproductive
consequences. The compositionality of language should
then be viewed in the context of the functional
advantages that compositionality affords: language
must be compositional because the language faculty
forces it to be that way, and the language faculty must be
designed like that because a compositional language
offers reproductive advantages (what could be more
useful than the ability to produce entirely novel
utterances about a wide range of situations, and have
them understood?).
Nowak et al. (2000) develop this argument in a
mathematical model of the evolution of composition-
ality. They assume two types of language learner: those
who learn a holistic (non-compositional) mapping
between meanings and signals, and those who learn a
simple compositional system. They consider the case of
populations of such learners converged on stable
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3594 K. Smith & S. Kirby Human language faculty and its evolution
languages and find that, as expected, populations of
compositional learners have higher within-population
communicative accuracy than learners who learn in a
holistic fashion, assuming that the number of events
that individuals are required to communicate about is
not small. Under conditions where there are a large
number of fitness relevant situations to communicate
about, the productivity advantage of compositional
language pays off in evolutionary terms.
(b) An explanatory role for cultural evolution
The biological account of the evolution of language
sketched by Pinker & Bloom (1990) is an attractive
one. However, their assertion that it is the only possible
explanation for the evident adaptive design of language
is based on a false premise: while we would accept that
‘[t]he only successful account of the origin of complex
biological structure is the theory of natural selection’,
we would dispute that language is solely a biological
structure. As already discussed in §2, language is
manifestly a socially learned, culturally transmitted
system. Individuals acquire their knowledge of language
by observing the linguistic behaviour of others, and go
on to use this knowledge to produce further examples of
linguistic behaviour, which others can learn from in turn
(see, e.g. Andersen 1973; Hurford 1990; Kirby 1999).
We have previously termed this process of cultural
transmission iterated learning: learning from the
behaviour of another, where that behaviour was itself
acquired through the same process of learning. The fact
that language is socially learned and culturally trans-
mitted opens up a second possible explanation for the
design features of language: those features arose through
cultural, rather than biological, evolution. Rather than
traditional transmission being another design feature
that a biological account must explain, traditional
transmission is the feature from which the other
structural properties of language spring.
One of the primary objections to this account has
already been stated: the argument from the poverty of
the stimulus suggests that language learning should be
impossible, and therefore the apparent social learning
of language must be illusory. However, cultural
transmission offers a potential solution to this con-
undrum that does not require an assumption of
innateness—while the poverty of the stimulus poses a
challenge for individual learners, language adapts over
cultural time so as to minimize this problem, because
its survival depends upon it. We have dubbed this
process cultural selection for learnability:
in order for linguistic forms to persist from one
generation to the next, they must repeatedly survive
the processes of expression [production] and induction
[learning]. That is, the output of one generation must
be successfully learned by the next if these linguistic
forms are to survive.
(Brighton et al. 2005, p. 303).
Cultural selection for learnability offers a solution to
the conundrum posed by the argument from the
poverty of the stimulus:
Human children appear preadapted to guess the rules of
syntax correctly, precisely because languages evolve so as
Phil. Trans. R. Soc. B (2008)
to embody in their syntax the most frequently guessed
patterns. The brain has coevolved with respect to
language, but languages have done most of the adapting
(Deacon 1997, p. 122).
As a consequence of cultural selection for learn-
ability, the poverty of the stimulus problem induces a
pressure for languages that are learnable from the kinds
of data learners can expect to see: ‘the poverty of the
stimulus solves the poverty of the stimulus’ (Zuidema
2003), but only if we look at language in its proper
context of cultural transmission.
How does cultural selection for learnability explain
compositionality? As discussed above, languages are
infinitely expressive (due to the combination of
recursion and compositionality). However, such
languages must be transmitted through a finite set of
learning data. We call this mismatch between the size
of the system to be transmitted and its medium of
transmission the learning bottleneck (Kirby 2002a; Smith
et al. 2003). Compositionality provides an elegant
solution to this problem: to learn a compositional
system, a learner must master a finite set of words and
rules for their combination, which can be learned from a
finite set of data but can generate a far larger system.
This fit between the form of language (it is compo-
sitional) and a property of the transmission medium
(it is finite but the system passing through it is infinite) is
suggestive, and computational models2 of the iterated
learning process (known as iterated learning models) have
repeatedly demonstrated that cultural evolution driven
by cultural selection for learnability can account for this
goodness of fit.
In their simplest form, iterated learning models
consist of a chain of simulated language learner/users
(known as agents). Each agent in this chain learns their
language by observing a set of utterances produced by
the preceding agent in the chain, and in turn produces
example utterances for the next agent to learn from.
The treatment of language learning and language
production varies from model to model, with similar
results having been shown for a fairly wide range of
models (see Kirby 2002b, for review). In all cases, the
crucial feature is that these agents are capable of
learning both holistic and compositional languages: they
can memorize a holistic meaning–signal mapping,
but they can also generalize to a partially or wholly
compositional language when the data they are
learning from merit such generalization. In other
words, compositionality is not hard-wired into the
language learners.
These models show that the presence of a learning
bottleneck is a key factor in determining the evolution
of compositionality. In conditions where there is no
learning bottleneck, the set of utterances produced by
one agent for another to learn from covers (or is highly
likely to cover) the full space of expressions that any
agent will ever be called upon to produce. This is of
course impossible for human language, where the set
of expressions is extremely large or infinite. On the
other hand, where there is a learning bottleneck, there
is some (typically high) probability that learner/users
will subsequently be called upon to produce a novel
expression and will therefore be required to generalize.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Human language faculty and its evolution
It is this pressure to generalize that leads to the
evolution of compositional languages: whereas compo-
sitional languages or compositional parts of language
are generalizable and can therefore be successfully
transmitted through a bottleneck, holistic (sub)systems
are by definition not generalizable and will be subject to
change. This adaptive dynamic has been used to
explain the putative transition from a hypothesized
holistic protolanguage (see, e.g. Wray 1998) to a
modern compositional system.
Compositionality can therefore be explained as a
cultural adaptation by language to the problem of
transmission through a learning bottleneck. Note that
throughout this explanation we are appealing to a
different notion of function to that discussed in §3a:
while Pinker & Bloom (1990) and Nowak et al. (2000)
appealed to compositionality as an adaptation for
communication, it is also a potential adaptation for
cultural transmission.
(c) Explanations for language design:
conclusions
Based on the preceding discussion, we believe there is
at least as good a case for explaining one design feature
of language as a cultural, rather than biological,
adaptation. Of course, cultural and biological evolution
can happily work in the same direction, and in some
cases (such as the evolution of duality of patterning),
the distinction between cultural and biological
mechanisms seems rather minor: in both cases, the
linguistic system is being optimized for its ability to
produce numerous maximally discriminable signals
(see Zuidema & de Boer in press). For the case of
compositionality, there is a difference in the function
being optimized by the two alternative adaptive
mechanisms: communicative usage under the bio-
logical account and learnability under the cultural
account. However, given that the conclusions in all
cases are the same—compositionality and duality of
patterning are good ideas—the temptation might be
to gloss over the differences in mechanism leading to
these adaptations.
However, the two competing explanations make
rather different predictions about the structure of the
human language faculty, which, as linguists, is our
ultimate object of study. Biological accounts suggest a
fairly direct mapping between properties of the
language faculty and properties of language, whereas
the cultural accounts propose a rather more opaque
relationship. This opaque relationship between the
language faculty and language design significantly
complicates evolutionary accounts of the language
faculty, as we will see in §4.
4. IMPLICATIONS FOR UNDERSTANDING THE
EVOLUTION OF THE LANGUAGE FACULTY
Acknowledging a role for cultural processes in the
evolution of language might not actually have any
consequences for our understanding of the human
language faculty in its present form. For example, it
would be perfectly consistent to accept that compos-
itionality was initially a cultural adaptation by language
to maximize its own transmissibility, but to argue that
Phil. Trans. R. Soc. B (2008)
K. Smith & S. Kirby 3595
this feature has subsequently been assimilated into the
language faculty. Arguments of this sort, appealing to
the mechanism known as the Baldwin effect or genetic
assimilation (Baldwin 1896; Waddington 1975) are in
fact reasonably common in evolutionary linguistics.
Pinker & Bloom (1990) appeal directly to the notion
that initially learned aspects of language will gradually
be assimilated into the language faculty, with any
remaining residue of learning being a result of
diminishing returns from assimilation. More detailed
arguments and formal models of this effect are
presented by Briscoe (2000, 2003). The prevalence of
assimilational accounts in the evolutionary linguistics
literature can be neatly characterized in a parenthetical
remark from Ray Jackendoff: ‘I agree with practically
everyone that the ‘Baldwin effect’ had something to do
with it’ (Jackendoff 2002, p. 237).
We believe this picture is fundamentally wrong, for
at least two reasons. First, we have previously argued
that the coevolution of culturally transmitted systems
and biological predispositions for learning those system
is somewhat problematic, due to time lags introduced
by the cumulative and frequency-dependent nature of
cultural evolution (Smith 2004). Our focus here will be
on a second problem: social transmission and cultural
evolution can shield the language-learning machinery
of individuals from selection, such that two rather
different language faculties end up being behaviourally
equivalent and selectively neutral. This neutrality rules
out evolution of more strongly constraining language
faculties via assimilational processes. Furthermore,
selection itself can drive evolution precisely into these
conditions—a plausible evolutionary scenario sees
natural selection acting on the language faculty
selecting for conditions where the language faculty is
shielded from selection.
(a) The link between language structure and
biological predispositions
A useful vehicle to develop this argument is provided by
a series of recent papers that seek to explicitly address
the link between language structure, biological predis-
positions and constraints on cultural transmission
(Griffiths & Kalish 2005, 2007; Kirby et al. 2007;
Griffiths et al. 2008). These papers are based around
iterated learning models where learners apply the
principles of Bayesian inference to language acqui-
sition. A learner’s confidence that a particular grammar
h accounts for the linguistic data d that they have
encountered is given by
PðdjhÞPðhÞ
PðhjdÞ Z P 0 0 ; ð4:1Þ
h 0 Pðdjh ÞPðh Þ
where P(h) is the prior belief in each grammar and
P(djh) gives the probability that grammar h could have
generated the observed data. Based on the posterior
probability of the various grammars, P(hjd ), the learner
then selects a grammar and produces utterances that
will form the basis, through social learning, of language
acquisition in others. This learning model provides a
transparent division between the contribution of the
learning bias of individuals prior to encountering data
(the prior) and the observed data in shaping behaviour.
We will equate prior bias with the innate language
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3596 K. Smith & S. Kirby Human language faculty and its evolution
faculty of individuals—while Griffiths and Kalish
rightly point out that the prior need not necessarily
take the form of an innate bias at all (e.g. it might be
derived from non-linguistic data), ours is a possible and
(we believe) natural interpretation.
Within this framework, Griffiths & Kalish (2005,
2007) showed that cultural transmission factors (such
as noise or a learning bottleneck imposed by partial
data) have no effect on the distribution of languages
delivered by cultural evolution: the outcome of cultural
evolution is solely determined by the prior biases of
learners, given by P(h). In other words, only the
structure of the language faculty matters in determin-
ing the outcomes of linguistic evolution. Griffiths &
Kalish (2007) and Kirby et al. (2007) demonstrated
that this result is a consequence of the assumption that
learners select a grammar with probability proportional
to P(hjd )—if learners instead select the grammar that
maximizes the posterior probability (known as MAP
learners), then cultural transmission factors play an
important role in determining the distribution of
languages delivered by cultural evolution: while the
distribution of languages produced by cultural
evolution will be approximately centred on the
language most favoured by the prior, different trans-
mission bottlenecks (for example) lead to different
distributions. Furthermore, and crucially for our
arguments here, for MAP learners the strength of the
prior bias is irrelevant over a wide range of the
parameter space (Kirby et al. 2007)—it matters which
language is most favoured in the prior, but not how
much it is favoured over alternatives.
These models suggest two candidate components of
the innate language faculty: first, the prior bias, P(h),
and second, the strategy for selecting a grammar based
on P(hjd )—sampling proportional to P(hjd ), or
selecting the grammar that maximizes P(hjd ). Further-
more, they allow us to explore, using a single frame-
work, the evolution of the language faculty under two
rather different conceptions of what cultural evolution
does: either it ensures that the distribution of languages
ultimately delivered by cultural evolution reflects
exactly the biases of language learners, regardless of
transmission factors such as the learning bottleneck
(if learners sample from the posterior), or it allows such
transmission factors to play a significant role in
determining the languages we see, and obscures
differences in the nature of individual language
faculties (if learners maximize). We can therefore
straightforwardly extend models of this sort to ask
whether cultural evolution alters the plausibility of
scenarios regarding the biological evolution of the
language faculty and, if so, under what conditions.
(b) The model of learning and cultural
transmission
We adopt Kirby et al.’s (2007) model of language and
language learning. Full details are given below, but to
briefly summarize: languages are mappings between
meanings and signals, where learners have a paramet-
rizable preference for regular languages. Implicit in this
model of learning is a model of cultural transmission,
which can be used to calculate the stable outcomes of
cultural evolution.
Phil. Trans. R. Soc. B (2008)
A language consists of a system for expressing m
meanings, where each meaning can be expressed using
one of k means of expression, called signal classes. In a
perfectly regular (or systematic) language the same
signal class will be used to express each meaning—for
example, the same inflectional paradigm will be used
for each verb, or the same compositional rules will be
used to construct an utterance for each meaning. By
contrast, in a perfectly irregular system each meaning
will be associated with a distinct signal class—each verb
an irregular, each complex utterance an idiom.
We will assume two types of prior bias. For unbiased
learners, all grammars have the same prior probability:
P(h)Z1/km. Biased learners have a preference for
languages that use a consistent means of expression,
such that each meaning is expressed using the same
signal class. Following Kirby et al. (2007), this prior is
given by the expression
GðkaÞ Yk
PðhÞ Z Gðnj C aÞ; ð4:2Þ
GðaÞk Gðm C kaÞ jZ1
where GðxÞZ ðx K1Þ! when x is an integer; nj is the
number of meanings expressed using class j; and aR1
determines the strength of the preference for consist-
ency: low a gives a strong preference for consistent
languages and higher a leads to a weaker preference for
such languages.3 Kirby et al. (2007) justify the use of
this particular prior distribution on the basis that
Bayesian inference with this prior can be viewed as
hypothesis selection based on minimum description
length principles, which has been argued (convincingly,
to our minds) to be relevant to cognition in general and
language acquisition in particular (see, e.g. Brighton
2003; Chater & Vitányi 2003). However, the precise
details of this prior are unimportant for our purposes
here: any prior that provides a (partial) ordering over
hypotheses supports our conclusions.
The probability of a particular dataset d (consisting
of b observed meaning–form pairs: b gives the bottle-
neck on language transmission) being produced by an
individual with grammar h is
Y 1
PðdjhÞ Z Pð yjx; hÞ ; ð4:3Þ
hxyi2d
m
where all meanings are equiprobable; hxyi is a meaning–
signal pair consisting of a meaning x and a signal class y;
and Pð yjx; hÞ gives the probability of y being produced
to convey x given grammar h and noise e:
8
> 1Ke if y is the class
>
>
>
< corresponding to x in h
Pð yjx; hÞ Z : ð4:4Þ
>
> e
>
> otherwise
: k K1
Bayes’ rule can then be applied to give a posterior
distribution over hypotheses, given a particular set of
utterances. This posterior distribution is used by a learner
to select a grammar, according to one of two strategies.
Sampling learners simply select a grammar propor-
tional to its posterior probability: PL ðhjd ÞZ Pðhjd Þ.
MAP learners select randomly from among those
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

Human language faculty and its evolution K. Smith & S. Kirby 3597

Table 1. P(h) for three grammars given various types of bias (unbiased, weak bias (aZ40) and strong bias (aZ1), denoted by u,
bw and bs, respectively), and the frequency of those grammars in the stationary distribution for sampling and MAP learners.
(Grammars are given as strings of characters, with the first character giving the class used to express the first meaning and so on:
aaa is a perfectly regular language and abc is perfectly irregular. All results here and throughout the paper are for mZ3, kZ3,
bZ3 and eZ0.1. For MAP learners, qualitatively similar results are obtainable for a wide range of the parameter space. For
sampling learners, as shown by Griffiths & Kalish (2007), qualitatively similar results are obtainable for any region of the
parameter space where eO0.)

P(h) Q
, sampler Q
, maximizer

h u bw bs u bw bs u bw bs

aaa 0.0370 0.0389 0.1 0.0370 0.0389 0.1 0.0370 0.2499 0.2499
aab 0.0370 0.0370 0.0333 0.0370 0.0370 0.0333 0.0370 0.0135 0.0135
abc 0.0370 0.0361 0.0167 0.0370 0.0361 0.0167 0.0370 0.0014 0.0014

grammars with the highest posterior probability: 0.25

(
1=jHj if h 2 H

PL ðhjd Þ Z ; ð4:5Þ
0 otherwise
where H is the set of hypotheses for which P(hjd ) is at a
maximum.
A model of cultural transmission follows straight-
forwardly from this model of learning: the probability
of a learner at generation n arriving at grammar hn given
exposure to data produced by grammar hnK1 is simply
X
Pðhn Z ijhnK1 Z j Þ Z PL ðhn Z ijd ÞPðdjhnK1 Z j Þ:
d ð4:6Þ
The matrix of all such transition probabilities is
known as the Q matrix ( Nowak et al. 2001): entry Qij
gives the transition probability from grammar j to i. As
discussed in Griffiths & Kalish (2005, 2007), the stable
outcome of cultural evolution (the stationary distribution
of languages) can be calculated given this Q matrix and
is proportional to its first eigenvector.4 We will denote
the probability of grammar i in the stationary
distribution as Q
i .
Table 1 gives some example prior probabilities and
stationary distributions, for various strengths of prior
and both selection strategies. As shown in table 1,
strength of prior determines the outcome of cultural
evolution for sampling learners, but is unimportant for
MAP learners as long as some bias exists.
(c) Evaluating within-population communicative
accuracy
In order to calculate which selection strategies and
priors are favoured by biological evolution, we need to
define a measure that determines reproductive success
and therefore predicts the evolutionary trajectory of the
language faculty. One possibility (following Nowak et al.
2000) is that the relevant quality is the extent to which
members of a genetically homogeneous population can
communicate with one another: the probability that any
two randomly selected individuals drawn from a
population at equilibrium (at the stationary distribution
provided by cultural evolution) will share the same
language.5 This quantity, C, is simply
X

2
CZ Qh ; ð4:7Þ
h contrast, cultural evolution in sampling populations
communicative accuracy, C
0.20
0.15
0.10
0.05
0
1 10 20 30 40
a
Figure 1. Fitness of sampler (filled squares) and MAP (open
squares, bZ3; circles, bZ6; triangles, bZ9) learners. Results
for MAP learners are for eZ0.1 and for various bottlenecks
(values of b). C for sampler populations is the same regardless
of noise level and amount of data, as these factors have no
impact on the stationary distribution.
where the sum is over all possible grammars. The
within-population communicative accuracies of various
combinations of strength of prior and hypothesis
selection strategy are given in figure 1. Three results
are apparent from this figure. First, in sampling
populations, stronger priors (lower values of a) yield
higher communicative accuracy. Stronger priors make
for a less uniform stationary distribution, with more
regular languages being over-represented. This skewing
away from the uniform distribution results in greater
within-population coherence. By contrast, strength of
prior is irrelevant in populations of MAP learners: it has
no impact on the stationary distribution, and as a result
there is no communicative advantage associated with
stronger priors.
Finally, MAP populations always have higher
within-population communicative accuracy than
sampling learners. As shown by Kirby et al. (2007),
and illustrated in table 1, in MAP populations the
differences between languages are amplified by cultural
evolution, and the extent of the amplification is
inversely proportional to the size of the transmission
bottleneck: tight bottlenecks give greater amplification,
such that the a priori most likely language is greatly
overrepresented in the stationary distribution. By

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010

3598 K. Smith & S. Kirby Human language faculty and its evolution

returns the unmodified prior distribution. Amplifying (a)

the differences between languages provided by the prior
increases the likelihood that two individuals from a
population will share the same language, and therefore 30
yields higher C. This analysis therefore suggests that

minority a
evolution should favour MAP learning over sampling,
leading to selective neutrality with respect to the 20
strength of prior biases: strength of prior bias is
shielded in MAP populations.
10

(d) Evaluating evolutionary stability

While the preceding analysis tells us which priors and
hypothesis selection strategies are objectively best from
a communicative point of view, it tells us nothing about (b)
the evolvability of those features. A more satisfying
solution is to evaluate the evolutionary stability of 30
hypothesis selection strategies and priors. In order to

minority a
do this, we require a slightly more detailed means of
evaluating how communication influences reproduc- 20
tive success. We make the following initial assumptions
(see below for a slightly different set of assumptions):
(i) a population consists of several subpopulations, 10
(ii) each subpopulation has converged on a single
grammar through social learning, with the probability
of each grammar being used by a subpopulation given
by that grammar’s probability in the stationary
distribution (as suggested by the analysis provided in (c)
Griffiths et al. 2008, §5a), and (iii) natural selection
favours learners who arrive at the same grammar6 as 30
their peers in a particular subpopulation, where peers
minority a

are other learners exposed to the language of the

subpopulation. Given these assumptions, the commu- 20
nicative accuracy between two individuals A and B is
given by
XX 10


hh 0 $Qhh 0 $Qh 0 ;
QA B
caðA; BÞ Z ð4:8Þ
h h0

where the superscripts on Q indicate that learners 10 20 30

A and B may have different selection strategies and
priors. The relative communicative accuracy of a single
learner A with respect to a large and homogeneous
population of individuals of type B is therefore given by
rcaðA; BÞZ caðA; BÞ=caðB; BÞ. Where this quantity is
greater than 1, the combination of selection strategy
and prior (the learning behaviour) of individual A offers
some reproductive advantage relative to the population
learning behaviour, and may come to dominate the
population. Where relative communicative accuracy
is less than 1, learning behaviour A will tend to be
selected against; and when relative communicative
accuracy is 1, both learning behaviours are equivalent
and genetic drift will ensue. Following Maynard
Smith & Price (1973), the conditions for evolutionary
stability for a behaviour of interest, I, are therefore:
(i) rcaðJ; IÞ! 1 for all JsI (populations of type I
resist invasion by all other learning behaviours) or
(ii) rcaðJ; IÞZ 1 for some JsI, but in each such case
rcaðI; JÞO 1. The second condition covers situations
where the minority behaviour J can increase by drift
to the point where encounters between type J indi-
viduals become common, at which point type I
individuals are positively selected for and the dom-
inance of behaviour I is re-established.
majority a
Figure 2. (a,b) Relative communicative accuracy of various
combinations of strengths of prior in sampling populations
((a) bZ3; (b) bZ9). Black cells indicate the minority a-value
will be selected against (rca!1), white cells indicate the
minority a-value will be selected for (rcaO1) and grey cells
(seen on the diagonal) indicate selective neutrality (rcaZ1).
(c) An example result given the global evaluation of
communicative accuracy (bZ9); see text for details.
Again, we can use this model to ask several
evolutionary questions. First, what does the evolution
of the strength of prior preference look like in sampling
and MAP populations? Do different learning models
(and their associated differences in the outcomes of
cultural evolution) change the evolutionary stability of
different strengths of prior? Figure 2a,b shows the
results of numerical calculations of evolutionary
stability in sampling populations.
As can be seen in figure 2, in sampling populations
there is selection against weaker priors and selection for
a stronger prior bias than the population norm. The
stronger the strength of the prior bias of the population,
the more dominant languages favoured by that prior
bias (i.e. the regular languages) will be, and the greater

Phil. Trans. R. Soc. B (2008)

 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Human language faculty and its evolution
Table 2. Relative communicative accuracy of various strengths
of prior in (a) sampling and (b) maximizing populations.
majority
u bs
(a) minority u — 0.81
bs 0.98 — 0.82
bw 0.99998 0.99
(b) minority u — 0.45
bs 1 —
bw 1 1
the advantage in being biased in favour of acquiring
those languages. In maximizing populations, strength
of prior is always selectively neutral—regardless of the a
of the majority and minority, rcaZ1. These results are
therefore not plotted. In accordance with the analysis
given in §4c, the evolutionary stability of different
strengths of prior bias differs markedly across sampling
and maximizing populations.
The degree to which stronger priors are favoured in
sampling populations is somewhat sensitive to the
strength of the population prior, however. For example,
populations with a very weak prior bias in favour of
regularity (aZ40) resist invasion by mutants with
much stronger prior preferences for regularity (az1)—
given the relatively uniform distribution of languages in
these populations, the added ease of acquiring highly
regular languages provided by a very strong preference
for such languages is counteracted by the reduced
likelihood of being born into populations speaking such
languages. This phenomenon becomes more marked
given larger bottlenecks (higher b, meaning that
learners observe more data during learning), as this
reduces the likelihood of misconvergence for individ-
uals with weaker priors.
As shown in table 2, this tendency for weaker
majority priors to reduce the extent to which strength-
ened priors can invade also pertains for populations
where the majority have a flat, unbiased prior: in such
populations every language is equally probable, and any
bias to acquire a particular language is penalized due to
the consequent decreased ability to acquire the other
languages. Consequently, there are two evolutionarily
stable strategies (ESSs) in sampling populations: the
strongest possible prior (aZ1) or a completely
unbiased prior.
In contrast with the situation in sampler popu-
lations, selection in MAP populations is neutral with
respect to bias strength. This follows naturally from the
insensitivity of MAP learners to the strength of their
prior biases—all that matters is the ranking of those
languages under the prior. Consequently, strength of
the prior makes no difference to the stationary
distribution provided by cultural evolution (as shown
in table 1), nor to the ability of learners to acquire those
languages. Consequently, strength of prior is selectively
neutral. However, as shown in table 2, this neutrality
with respect to strength of prior bias only applies when
there is some prior bias—a completely flat prior is not
evolutionarily stable (through the second clause of the
definition given above). Consequently, the set of all
Phil. Trans. R. Soc. B (2008)
K. Smith & S. Kirby 3599
Table 3. Relative communicative accuracy of the minority
hypothesis selection strategy for strong (aZ1), weak (aZ40)
and flat priors.
majority
bw
minority sampling maximizing
0.9997
strong sampling — 0.6
— maximizing 1.39 —
weak sampling — 0.38
0.45 maximizing 1.14 —
1 flat sampling — 0.88
— maximizing 1.12 —
non-flat priors constitutes the evolutionarily stable set
(Thomas 1985) for MAP learners.
These two models of learning, which make different
predictions about the outcomes of cultural evolution,
also make different predictions about the coevolution of
the language faculty and language. In particular, the
learning model that predicts an opaque relationship
between bias and outcomes of cultural evolution (MAP
learning) also predicts extensive shielding of the
language faculty from selection, and certainly no
positive selection for increasingly constraining
language faculties.
We can ask a final evolutionary question: is it better
to be a sampling learner or a maximizing learner? As
shown in table 3, maximizing is the ESS, regardless
of bias strength. MAP learners boost the probability
that the most likely grammar will be learned, and are
consequently more likely to arrive at the same grammar
as some other learner exposed to the same data-
generating source. Given the fitness function that
emphasizes convergence on the same language as
other members of the population, maximizing is
obviously the best thing to do.
It is worth noting that this pattern of results is not
dependent on the assumption that learners are
rewarded for arriving at the same grammar as other
individuals reared in the same linguistically homo-
geneous subpopulation. While this strikes us as a
reasonable fitness function, it is not the only possible
one. For example, we could make the following rather
different set of assumptions: (i) rather than consisting
of several subpopulations, there is a single well-mixed
(meta-)population, (ii) the probability of each grammar
being used by any individual is given by that grammar’s
probability in the stationary distribution (again, as
suggested by the analysis provided in Griffiths et al.
2008, §5a) and (iii) natural selection favours learners
who arrive at the same grammar as any randomly
selected peer. This global fitness function evaluates
learners on their ability to communicate with any
randomly selected individual from a linguistically
heterogeneous metapopulation, as opposed to reward-
ing communication within a local subpopulation.
Under this alternative global fitness regime, the
same pattern of results emerges: stronger priors are
favoured in sampling populations; prior strength is
selectively neutral in maximizing populations; and
maximizing is favoured over sampling. There are
three relatively minor qualitative differences.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3600 K. Smith & S. Kirby Human language faculty and its evolution
Table 4. Some results for the alternative global evaluation of
fitness. (Relative communicative accuracy of various
strengths of prior in sampling populations. Unbiased priors
are no longer evolutionarily stable.)
majority
u bs bw
minority u — 0.79 0.9997
bs 1 — 1.008
bw 1 0.79 —
First, in sampling populations, the flat prior no longer
constitutes an ESS: the sole ESS is the strongest possible
prior, aZ1. This is shown in table 4 (cf. table 2a).
Populations converged on a flat prior are invasible by
drift by learners with a non-flat prior: in a well-mixed
unbiased population, any randomly selected unbiased
individual will be equally likely to have each grammar;
therefore, no matter what grammar a biased learner
converges on, they will successfully communicate with
an identical proportion of the population. Note the
difference with the local population case, where biased
learners are penalized whenever they are born into a
subpopulation using a language disfavoured by the
biased learner’s prior.
Second, and following the same reasoning, in
sampling populations, the zone of selection where
much stronger priors than the population norm are
selected against disappears (figure 2c, cf. figure 2b)—
again, global mixing removes the risk associated with
strong priors of converging on the wrong language
within a homogeneous subpopulation.
Finally, and again by the same reasoning, in popu-
lations with a flat prior bias, sampling and maximizing are
equivalent: under the global evaluation, the selective
advantage to maximizing disappears in such populations
(but not in biased populations, where maximizing is still
preferred). This is illustrated in table 5 (cf. table 3,
subtable for flat priors). Given the completely uniform
distribution over languages and global mixing, samplers
(who would normally be disproportionately penalized
in situations where they fail to converge on their sub-
population’s language) will successfully communicate
with an equal proportion of the population regardless of
which grammar they (mis)learn.
Notwithstanding these minor differences, the results
of the evolutionary analysis seem to be fairly robust
under different conceptions of how communicative
success within a population should be defined. It is an
open question whether the same conclusion pertains
given more complex fitness functions—for example,
one that favours the acquisition of a specific language
rather than the (locally or globally) most frequent, or
favours the acquisition of a frequent but not too
frequent language.
To summarize: this model therefore suggests that
selection for communication acting on the language
faculty should select learners who maximize over the
posterior distribution rather than sampling from it.
Given this initial selective choice, selection should
consequently be neutral with respect to the strength of
the prior preference in favour of particular linguistic
Phil. Trans. R. Soc. B (2008)
Table 5. Some results for the alternative global evaluation of
fitness. (Relative communicative accuracy of sampling versus
maximizing for flat priors. Under these circumstances,
maximizing is no longer preferred over sampling.)
majority
sampling maximizing
minority sampling — 1
maximizing 1 —
structures: strongly constraining language faculties are
no more likely than extremely weak biases. This is the
first main conclusion we would like to draw from this
model: the prediction that selection will favour less
flexible rather than more flexible learning is not always
warranted, because learning and cultural evolution can
shield the language faculty from selection.
Furthermore, there are conditions under which
selection will favour the weakest possible prior biases.
Cost can be integrated into the model by assuming that
natural selection favours learners who arrive at the
same grammar as their peers and minimize cost as a
function of a: for example, individuals might reproduce
proportionally to their weighted communicative
accuracy, where the weighted communicative accuracy
between two individuals A and B is given by
caðA; BÞ
ca 0 ðA; BÞ Z ; ð4:9Þ
cðaA Þ C cðaB Þ
where ca(A, B) is as given in equation (4.8); c(a) is a cost
function, with higher values corresponding to greater
cost; and aA gives the strength of prior of individual A. If
we assume that strong prior biases have some cost (i.e.
c(a) is inversely proportional to a: perhaps stronger
priors require additional, restrictive but costly cognitive
machinery), there are conditions under which only weak
bias would be evolutionarily stable. There will be some
high value of a, which we will call a
, for which: (i) the
prior is sufficiently weak that its costs relative to the
unbiased strategy are low enough to allow a
individuals
to invade unbiased populations and (ii) the prior
remains sufficiently strong that a
populations resist
invasion by unbiased individuals. Under such a scenario,
the extremely weak a
prior becomes the sole ESS:
evolution will favour maximization and the weakest
possible (but not flat) prior.
The evolutionary argument sketched above only
applies if we assume that the only selective advantage to
a particular prior bias arises from its communicative
function: were particular priors to offer some non-
linguistic advantage, less prone to being shielded by
learning and cultural evolution, then it could be
selected for (or against) based on those more selectively
obvious functions. This means that language-learning
strategies that are strongly constraining but domain
general are more likely to evolve than constraining
domain-specific strategies. In other words, the
traditional transmission of language means that the
most likely strongly constraining language faculty
will not be a language faculty in the strict sense at all,
but a more general cognitive faculty applied to the
acquisition of language.
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Human language faculty and its evolution
(e) Conclusions on the evolution of the
language faculty
Cultural transmission alters our expectations about
how the language faculty should evolve. Different
language faculties may lead to identical outcomes for
cultural evolution (as shown by Kirby et al. 2007),
resulting in selective neutrality over those language
faculties. If we believe that stronger prior biases (more
restrictive innate machinery) cost, then selection can
favour the weakest possible prior bias in favour of a
particular type of linguistic structure. This result, in
conjunction with those provided in Smith (2004), leads
us to suspect that a domain-specific language faculty,
selected for its linguistic consequences, is unlikely to be
strongly constraining—such a language faculty is the
least likely to evolve. Any strong constraints from
the language faculty are likely to be domain general, not
domain specific, and owe their strength to selection for
alinguistic, non-cultural functions.
5. CONCLUSIONS
How do the predictions of this evolutionary modelling
relate back to the uniqueness question posed in §2:
what is it about humans that gives us our unusual
communication system? Based on the argument out-
lined in §4, we think that a strongly constraining
domain-specific language faculty is unlikely to have
evolved in humans. It strikes us as more likely that
humans have a collection of domain-general biases that
they bring to the language-learning task (see also
Christiansen & Chater in press)—while these biases
might be rather weak, their cumulative application
during language learning will lead to strong universal
patterns of linguistic structure, such as the design
features identified in §2. This prediction is consonant
with a recent body of work in developmental linguistics
that seeks to identify how domain-general statistical
learning techniques can be used to acquire language
from data (e.g. Saffran et al. 1996; Gómez 2002), what
the biases of those statistical processes are (e.g.
Newport & Aslin 2004; Endress et al. 2005; Hudson
Kam & Newport 2005) and whether non-linguistic
species share the same capacities and biases (Hauser
et al. 2001; Fitch & Hauser 2004; Newport et al. 2004;
Gentner et al. 2006). The comparative aspect of this
work strikes us as particularly important, in that it
sheds light on the extent to which humans actually have
any specializations for the acquisition of language,
rather than specializations for sequence learning or for
social learning more generally.
We would not be surprised if species-specific
specializations for the acquisition of linguistic structure
turn out to be rare or even non-existent—this is what
the evolutionary argument in §4 suggests. Rather, we
suspect that the uniqueness of language arises from the
co-occurrence of a number of relatively unusual
cognitive capacities that constitute preconditions for
the cultural evolution of these linguistic features.
The modelling literature on the cultural evolution of
communication is a useful tool in identifying what these
preconditions might look like. The components
required for cultural evolution to produce a simple,
Phil. Trans. R. Soc. B (2008)
K. Smith & S. Kirby 3601
traditionally transmitted, semantic and productive
system seem to be fairly minimal:
(i) ability to modify own produced signal forms
based on observed usage;
(ii) ability to learn to associate meanings with signals;
(iii) ability to infer communicative intentions
(meanings) in others;
(iv) these meanings are drawn from a reasonably large
and structured meaning space.
At a first approximation, cultural evolution in a
population of social learners meeting these four
preconditions should yield a productive and semantic
communication system. The first three conditions are
simply the component parts required for a traditionally
transmitted semantic communication system. The
fourth stipulation relates to the requirement for a
learning bottleneck (a bottleneck is more likely if the
system contains lots of meanings to be communicated),
but also the possibility of generalizing from meaning to
meaning, which requires some similarity structure
among meanings—when this similarity structure is
reduced, the learnability advantage of compositional
language is reduced (Brighton 2002; Smith et al. 2003).
As discussed in §2, some of these abilities are present
in non-human species. The first is present in all vocal
learners, and also in non-human primates capable of
sustaining ritualized gestural communication systems.
While the second is not present to any meaningful
extent in songbirds and other vocal learners (at least
with reference to their vocally learned communication
systems), it is again present in gesturally communi-
cating apes. However, the third ability seems to be rare,
including among other primates. The ability to infer
the communicative intentions of others underpins the
learning of large sets of arbitrary meaning–form
associations: in order to learn such associations, you
must be able, somewhat reliably, to infer the meaning
associated with each utterance you encounter. While
some other primates are able to learn to infer the
meaning of a communicative signal, the laborious
process of ontogenetic ritualization by which meaning
for signals become established stands in stark contrast
to the human ability to rapidly and accurately infer
communicative intentions (see, e.g. Bloom 1997,
2000). There is no good evidence that apes are able
to acquire communicative signs by more streamlined
observational processes (and see Tomasello et al. 1997,
for a negative result).
This difference in efficiency in inferring the meaning
associated with a signal may be important for the
following reasons. First, the requirement for a shared
history of interaction reduces the potential spread
of any signal, limiting it to the individuals able to
devote time to its establishment. Second, it restricts the
ultimate size of the repertoires of signals that can be
learned, which directly links to the fourth requirement
listed above, for a large and structured meaning space
to associate signals with. The (limited) success of apes
trained on artificial communication systems (see, e.g.
Savage-Rumbaugh et al. 1986), with modest inven-
tories of communicative tokens and meaningful
combinations of those tokens, suggests that there may
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
3602 K. Smith & S. Kirby Human language faculty and its evolution
be no fundamental cognitive barrier to the acquisition
of a small productive communication system in apes, if
only sufficient scaffolding is provided for its acquisition.
Such communication systems may not exist in the wild
(and we offer this as no more than a tentative
suggestion) due to the limits on repertoire size imposed
by the ritualization process, which prevents the
establishment of large and systematically related sets
of conventional meaning–signal pairings and therefore
removes any cultural evolutionary pressure for pro-
ductive structure.
Of course the question then becomes: why are these
preconditions for the cultural evolution of language
only met in humans? Why have humans evolved the
unusual abilities to vocally learn, infer meaning
efficiently, and so on? We have at present no answer
to these questions, but we suspect that exploring the
evolution of the preconditions for the cultural evolution
of language is likely to be more profitable, and more
amenable to the comparative approach, than seeking to
establish the evolutionary history of a strongly con-
straining, domain-specific and species-unique faculty
of language.
K.S. was funded by a British Academy Postdoctoral Research
Fellowship. The development of the model described in §4
was facilitated by our attendance at the NWO-funded
Masterclass on Language Evolution, organized by Bart de
Boer and Paul Vogt.
ENDNOTES
1
Or some other discriminable unit, e.g. handshape, orientation and
location in sign language (Stokoe 1960).
2
See Kirby et al. (2008) for an experimental treatment of the same
phenomenon.
3
We will only consider the case where a takes integer values.
4 Fitch, W. T. & Hauser, M. D. 2004 Computational constraints
Provided the Markov process described by the Q matrix is ergodic
(see Griffiths & Kalish 2007, p. 446).
5
This assumes that identical grammars are required for communi-
cation. We could instead evaluate communicative accuracy based on
the degree of similarity between grammars or the degree of similarity
between datasets produced by those grammars. These more graded
notions of communication produce qualitatively similar results to
those presented here, the quantitative change being a reduction in the
magnitude of the selection differentials in some cases.
6
Again, more graded notions of match between grammars produce
qualitatively similar results.
REFERENCES
Andersen, H. 1973 Abductive and deductive change.
Language 40, 765–793. (doi:10.2307/412063)
Baldwin, J. M. 1896 A new factor in evolution. Am. Nat. 30,
441–451. (doi:10.1086/276408)
Bloom, P. 1997 Intentionality and word learning. Trends Cogn.
Sci. 1, 9–12. (doi:10.1016/S1364-6613(97)01006-1)
Bloom, P. 2000 How children learn the meanings of words.
Cambridge, MA: MIT Press.
Brighton, H. 2002 Compositional syntax from cultural
transmission. Artif. Life 8, 25–54. (doi:10.1162/106454
602753694756)
Brighton, H. 2003 Simplicity as a driving force in linguistic
evolution, PhD thesis, The University of Edinburgh.
Brighton, H., Kirby, S. & Smith, K. 2005 Cultural selection
for learnability: three principles underlying the view that
Phil. Trans. R. Soc. B (2008)
language adapts to be learnable. In Language origins:
perspectives on evolution (ed. M. Tallerman), pp. 291–309.
Oxford, UK: Oxford University Press.
Briscoe, E. 2000 Grammatical acquisition: inductive bias and
coevolution of language and the language acquisition
device. Language 76, 245–296. (doi:10.2307/417657)
Briscoe, E. 2003 Grammatical assimilation. In Language
evolution (eds M. H. Christiansen & S. Kirby), pp. 295–316.
Oxford, UK: Oxford University Press.
Call, J. & Tomasello, M. (eds) 2004 The gestural communi-
cation of apes and monkeys. Hillsdale, NJ: Lawrence
Earlbaum Associates.
Catchpole, C. K. & Slater, P. J. B. 1995 Bird song: biological
themes and variations. Cambridge, UK: Cambridge
University Press.
Chater, N. & Vitányi, P. M. B. 2003 Simplicity: a unifying
principle in cognitive science? Trends Cogn. Sci. 7, 19–22.
(doi:10.1016/S1364-6613(02)00005-0)
Cheney, D. & Seyfarth, R. 1990 How monkeys see the world:
inside the mind of another species. Chicago, IL: University of
Chicago Press.
Chomsky, N. 1980 Rules and representations. London, UK:
Basil Blackwell.
Christiansen, M. & Chater, N. In press. Language as shaped
by the brain. Behav. Brain Sci.
Deacon, T. 1997 The symbolic species. London, UK: Penguin.
Eens, M. 1997 Understanding the complex song of European
starling: an integrated ethological approach. Adv. Study
Behav. 26, 355–434.
Endress, A. D., Scholl, B. J. & Mehler, J. 2005 The role of
salience in the extraction of algebraic rules. J. Exp. Psychol.
Gen. 134, 406–419. (doi:10.1037/0096-3445.134.3.406)
Evans, C. S., Evans, L. & Marler, P. 1993 On the meaning of
alarm calls: functional reference in an avian vocal system.
Anim. Behav. 46, 23–38. (doi:10.1006/anbe.1993.1158)
Fitch, W. T. 2000 The evolution of speech: a comparative
review. Trends Cogn. Sci. 4, 258–267. (doi:10.1016/S1364-
6613(00)01494-7)
on syntactic processing in a nonhuman primate. Science
303, 377–380. (doi:10.1126/science.1089401)
Gentner, T. Q., Fenn, K. M., Margoliash, D. & Nusbaum,
H. C. 2006 Recursive syntactic pattern learning by song-
birds. Nature 440, 1204–1207. (doi:10.1038/nature04675)
Gil, D. & Slater, P. J. B. 2000 Song organisation and singing
patterns of the willow warbler, Phylloscopus trochilus.
Behaviour 137, 759–782. (doi:10.1163/15685390050
2330)
Gómez, R. L. 2002 Variability and detection of invariant
structure. Psychol. Sci. 13, 431–436. (doi:10.1111/1467-
9280.00476)
Griffiths, T. L. & Kalish, M. L. 2005 A Bayesian view of
language evolution by iterated learning. In Proc. 27th
Annual Conf. of the Cognitive Science Society (eds B. G.
Bara, L. Barsalou & M. Bucciarelli), pp. 827–832.
Mahwah, NJ: Erlbaum.
Griffiths, T. L. & Kalish, M. L. 2007 Language evolution by
iterated learning with Bayesian agents. Cogn. Sci. 31,
441–480.
Griffiths, T. L., Kalish, M. L. & Lewandowsky, S. 2008
Theoretical and empirical evidence for the impact of
inductive biases on cultural evolution. Phil. Trans. R. Soc.
B 363, 3591–3603. (doi:10.1098/rstb.2008.0146)
Hauser, M. D., Newport, E. L. & Aslin, R. N. 2001
Segmentation of the speech stream in a non-human
primate: statistical learning in cotton-top tamarins.
Cognition 78, B53–B64. (doi:10.1016/S0010-0277(00)
00132-3)
 Downloaded from rstb.royalsocietypublishing.org on May 15, 2010
Human language faculty and its evolution
Hauser, M. D., Chomsky, N. & Fitch, W. T. 2002 The faculty
of language: what is it, who has it, and how did it evolve?
Science 298, 1569–1579. (doi:10.1126/science.298.5598.
1569)
Hockett, C. F. 1960 The origin of speech. Sci. Am. 203,
88–96.
Hudson Kam, C. L. & Newport, E. L. 2005 Regularizing
unpredictable variation: the roles of adult and child
learners in language formation and change. Lang. Learn.
Dev. 1, 151–195. (doi:10.1207/s15473341lld0102_3)
Hurford, J. R. 1990 Nativist and functional explanations in
language acquisition. In Logical issues in language acquisition
(ed. I. M. Roca), pp. 85–136. Dordrecht, The Netherlands:
Foris.
Jackendoff, R. 2002 Foundations of language: brain, meaning,
grammar, evolution. Oxford, UK: Oxford University Press.
Kinsella, A. R. In press. Language evolution and syntactic
theory. Cambridge, UK: Cambridge University Press.
Kirby, S. 1999 Function, selection and innateness: the emergence
of language universals. Oxford, UK: Oxford University
Press.
Kirby, S. 2002a Learning, bottlenecks and the evolution of
recursive syntax. In Linguistic evolution through language
acquisition: formal and computational models (ed. E. Briscoe),
pp. 173–203. Cambridge, UK: Cambridge University
Press.
Kirby, S. 2002b Natural language from artificial life. Artif.
Life 8, 185–215. (doi:10.1162/106454602320184248)
Kirby, S., Dowman, M. & Griffiths, T. L. 2007 Innateness
and culture in the evolution of language. Proc. Natl Acad.
Sci. USA 104, 5241–5245. (doi:10.1073/pnas.060822
2104)
Kirby, S., Cornish, H. & Smith, K. 2008 Cumulative cultural
evolution in the laboratory: an experimental approach to the
origins of structure in human language. Proc. Natl Acad. Sci.
USA 105, 10 681–10 686. (doi:10.1073/pnas.0707835105)
Krifka, M. 2001 Compositionality. In The MIT encyclopae-
dia of the cognitive sciences (eds R. A. Wilson & F. Keil),
pp. 152–153. Cambridge, MA: MIT Press.
Marler, P. 1955 Characteristics of some animal calls. Nature
176, 6–8. (doi:10.1038/176006a0)
Maynard Smith, J. & Price, G. R. 1973 The logic of animal
conflict. Nature 146, 15–18. (doi:10.1038/246015a0)
Maynard Smith, J. & Szathmáry, E. 1995 The major transitions
in evolution. Oxford, UK: Oxford University Press.
Newport, E. L. & Aslin, R. N. 2004 Learning at a distance I:
statistical learning of non-adjacent dependencies. Cogn.
Psychol. 48, 127–162. (doi:10.1016/S0010-0285(03)00
128-2)
Newport, E. L., Hauser, M. D., Spaepan, G. & Aslin, R. N.
2004 Learning at a distance II: statistical learning of non-
adjacent dependencies in a non-human primate. Cogn.
Psychol. 49, 85–117. (doi:10.1016/j.cogpsych.2003.12.002)
Nowak, M. A. & Krakauer, D. C. 1999 The evolution of
language. Proc. Natl Acad. Sci. USA 96, 8028–8033.
(doi:10.1073/pnas.96.14.8028)
Nowak, M. A., Krakauer, D. C. & Dress, A. 1999 An error
limit for the evolution of language. Proc. R. Soc. B 266,
2131–2136. (doi:10.1098/rspb.1999.0898)
Nowak, M. A., Plotkin, J. B. & Jansen, V. A. A. 2000 The
evolution of syntactic communication. Nature 404,
495–498. (doi:10.1038/35006635)
Phil. Trans. R. Soc. B (2008)
K. Smith & S. Kirby 3603
Nowak, M. A., Komarova, N. L. & Niyogi, P. 2001 Evolution
of universal grammar. Science 291, 114–117. (doi:10.
1126/science.291.5501.114)
Okanoya, K. 2004 The bengalese finch: a window on the
behavioral neurobiology of birdsong syntax. Ann. N. Y.
Acad. Sci. 1016, 724–735. (doi:10.1196/annals.1298.026)
Oudeyer, P.-Y. 2005 The self-organization of speech sounds.
J. Theor. Biol. 233, 435–449. (doi:10.1016/j.jtbi.2004.
10.025)
Pinker, S. & Bloom, P. 1990 Natural language and natural
selection. Behav. Brain Sci. 13, 707–784.
Pullum, G. K. & Scholz, B. C. 2002 Empirical assessment of
stimulus poverty arguments. Linguist. Rev. 19, 9–50.
(doi:10.1515/tlir.19.1-2.9)
Saffran, J. R., Aslin, R. N. & Newport, E. L. 1996 Statistical
learning by 8-month-old infants. Science 274, 1926–1928.
(doi:10.1126/science.274.5294.1926)
Savage-Rumbaugh, S., McDonald, K., Sevcik, R. A.,
Hopkins, W. D. & Rubert, E. 1986 Spontaneous symbol
acquisition and communicative use by pygmy chimpan-
zees (Pan paniscus). J. Exp. Psychol. Gen. 115, 211–235.
(doi:10.1037/0096-3445.115.3.211)
Slater, P. J. B. 2005 Animal communication: vocal learning.
In The encyclopedia of language and linguistics (ed.
K. Brown), pp. 291–294, 2nd edn. Amsterdam, The
Netherlands: Elsevier
Smith, K. 2004 The evolution of vocabulary. J. Theor. Biol.
228, 127–142. (doi:10.1016/j.jtbi.2003.12.016)
Smith, K., Brighton, H. & Kirby, S. 2003 Complex systems
in language evolution: the cultural emergence of compo-
sitional structure. Adv. Complex Syst. 6, 537–558. (doi:10.
1142/S0219525903001055)
Stokoe, W. C. 1960 Sign language structure. Silver Spring,
MD: Linstok Press.
Thomas, B. 1985 On evolutionarily stable sets. J. Math. Biol.
22, 105–115. (doi:10.1007/BF00276549)
Tomasello, M. 1996 Do apes ape? In Social learning in
animals: the roots of culture (eds C. Heyes & B. Galef ),
pp. 319–436. San Diego, CA: Academic Press.
Tomasello, M., Call, J., Warren, J., Frost, G., Carpenter, M. &
Nagell, K. 1997 The ontogeny of chimpanzee gestural
signals: a comparison across groups and generations. Evol.
Commun. 1, 223–259.
Waddington, C. H. 1975 The evolution of an evolutionist.
Ithaca, NY: Cornell University Press.
Whiten, A. 2005 The second inheritance system of
chimpanzees and humans. Nature 437, 52–55. (doi:10.
1038/nature04023)
Wray, A. 1998 Protolanguage as a holistic system for social
interaction. Lang. Commun. 18, 47–67. (doi:10.1016/
S0271-5309(97)00033-5)
Zuberbuhler, K. 2001 Predator-specific alarm calls in Camp-
bell’s monkeys, Cercopithecus campbelli. Behav. Ecol. Socio-
biol. 50, 414–422. (doi:10.1007/s002650100383)
Zuberbuhler, K. 2002 A syntactic rule in forest monkey
communication. Anim. Behav. 63, 293–299. (doi:10.
1006/anbe.2001.1914)
Zuidema, W. H. 2003 How the poverty of the stimulus solves
the poverty of the stimulus. In Advances in neural information
processing systems 15 (Proceedings of NIPS ‘02) (eds S. Becker,
S. Thrun & K. Obermayer), pp. 51–58. Cambridge, MA:
MIT Press.
Zuidema, W. H. & de Boer, B. In press. The evolution of
combinatorial phonology. J. Phon.
RSTB_363_1509.qxp 9/30/08 5:06 PM Page 1

Phil. Trans. R. Soc. B | vol. 363 no. 1509 pp. 3467–3603 | 12 Nov 2008
ISSN 0962-8436

volume 363

12 November 2008 number 1509

volume 363 . number 1509 . pages 3467–3603
pages 3467–3603
Cultural transmission and the evolution
of human behaviour In this issue

Papers of a Theme Issue compiled and edited by Kenny Smith, Michael L. Kalish, Thomas L. Griffiths Cultural transmission and the evolution
and Stephan Lewandowsky
of human behaviour
Introduction. Cultural transmission and the evolution of human behaviour 3469
K. Smith, M. L. Kalish, T. L. Griffiths & S. Lewandowsky
Papers of a Theme Issue compiled and edited by Kenny Smith, Michael L. Kalish, Thomas L. Griffiths
Review. Establishing an experimental science of culture: animal social diffusion experiments 3477 and Stephan Lewandowsky
A. Whiten & A. Mesoudi
Review. The multiple roles of cultural transmission experiments in understanding human
cultural evolution 3489
A. Mesoudi & A. Whiten
Review. Theoretical and empirical evidence for the impact of inductive biases on cultural evolution 3503
T. L. Griffiths, M. L. Kalish & S. Lewandowsky
Beyond existence and aiming outside the laboratory: estimating frequency-dependent and
pay-off-biased social learning strategies 3515
R. McElreath, A. V. Bell, C. Efferson, M. Lubell, P. J. Richerson & T. Waring
Review. Studying cumulative cultural evolution in the laboratory 3529
C. A. Caldwell & A. E. Millen

Cultural transmission and the evolution of human behaviour

Investigating children as cultural magnets: do young children transmit redundant information
along diffusion chains? 3541
E. Flynn
The fitness and functionality of culturally evolved communication systems 3553
N. Fay, S. Garrod & L. Roberts
Culture, embodiment and genes: unravelling the triple helix 3563
M. Wheeler & A. Clark
Exploring gene-culture interactions: insights from handedness, sexual selection and niche-construction
case studies 3577
K. N. Laland
Cultural evolution: implications for understanding the human language faculty and its evolution 3591
K. Smith & S. Kirby

The world’s longest running international science journal

Founded in 1660, the Royal Society
is the independent scientific academy
of the UK, dedicated to promoting
excellence in science
journals.royalsociety.org
Registered Charity No 207043 Published in Great Britain by the Royal Society, 12 November 2008
6–9 Carlton House Terrace, London SW1Y 5AG