Cultural Transmission and the Evolution of Human Behaviour (Philosophical Transactions of the Royal Society series B)

RSTB_363_1509.qxp 9/30/08 5:06 PM Page 1 volume 363 . number 1509 . pages 3467–3603 Cultural transmission and the ...

Author: Kenny Smith | Michael L Kalish | Thomas L Griffiths | Stephan Lewandowsky

56 downloads 1104 Views 3MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

RSTB_363_1509.qxp

9/30/08

5:06 PM

Page 1

volume 363

. number 1509 . pages 3467–3603

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 Introduction. Cultural transmission and the evolution of human behaviour K. Smith, M. L. Kalish, T. L. Griffiths & S. Lewandowsky

3469

Review. Establishing an experimental science of culture: animal social diffusion experiments A. Whiten & A. Mesoudi

3477

Review. The multiple roles of cultural transmission experiments in understanding human cultural evolution A. Mesoudi & A. Whiten Review. Theoretical and empirical evidence for the impact of inductive biases on cultural evolution 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 R. McElreath, A. V. Bell, C. Efferson, M. Lubell, P. J. Richerson & T. Waring

3503

3515

Investigating children as cultural magnets: do young children transmit redundant information along diffusion chains? E. Flynn

3541

The fitness and functionality of culturally evolved communication systems N. Fay, S. Garrod & L. Roberts

3553

Culture, embodiment and genes: unravelling the triple helix M. Wheeler & A. Clark

3563

Exploring gene-culture interactions: insights from handedness, sexual selection and niche-construction case studies K. N. Laland Cultural evolution: implications for understanding the human language faculty and its evolution K. Smith & S. Kirby

Founded in 1660, the Royal Society is the independent scientific academy of the UK, dedicated to promoting excellence in science Registered Charity No 207043

volume 363

number 1509

pages 3467–3603

In this issue

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

3529

3577 3591

journals.royalsociety.org Published in Great Britain by the Royal Society, 6–9 Carlton House Terrace, London SW1Y 5AG

Cultural transmission and the evolution of human behaviour

Review. Studying cumulative cultural evolution in the laboratory C. A. Caldwell & A. E. Millen

3489

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

12 November 2008

ISSN 0962-8436

The world’s longest running international science journal

12 November 2008

RSTB_363_1509.qxp

9/30/08

5:06 PM

Page 2

GUIDANCE FOR AUTHORS

Editor Professor Georgina Mace Publishing Editor Claire Rawlinson Editorial Board Neuroscience and Cognition Dr Brian Billups Dr Andrew Glennerster Professor Bill Harris Professor Atsushi Iriki Professor Trevor Lamb Professor Obaid Siddiqi Professor Marc Tessier-Lavigne Professor Andrew Whiten

Organismal, environmental and evolutionary biology Professor Georgina Mace Professor Yadvinder Malhi Professor Manfred Milinski Professor Peter Mumby Professor Peter Raven Professor Karl Sigmund Professor Monkombu Swaminathan

Cell and developmental biology Professor Julie Ahringer Dr Buzz Baum Dr Anne Donaldson Professor Richard Gardner Professor John Gray Professor Keith Gull Professor Fotis Kafatos Professor Elliot Meyerowitz Professor Anthony Pawson

Health and Disease Professor Zhu Chen Professor Mark Enright Professor Peter Goodfellow Professor Michael Malim Professor Lap-Chee Tsui Professor Nicholas Wald Professor Bob Williamson

Publishing Editor: Claire Rawlinson (tel: +44 (0)20 7451 2602; fax: +44 (0)20 7976 1837; [email protected])

Production Editor: Victoria Brown

6–9 Carlton House Terrace, London SW1Y 5AG, UK publishing.royalsociety.org

Selection criteria The criteria for selection are scientific excellence, originality and interest across disciplines within biology. The Editors are responsible for all editorial decisions and they make these decisions based on the reports received from the referees and/or Editorial Board members. Many more good proposals and articles are submitted to us than we have space to print, we give preference to those that are of broad interest and of high scientific quality. Publishing format Phil. Trans. R. Soc. B articles are published regularly online and in print issues twice a month. Along with all Royal Society journals, we are committed to archiving and providing perpetual access. The journal also offers the facility for including Electronic Supplementary Material (ESM) to papers. Contents of the ESM might include details of methods, derivations of equations, large tables of data, DNA sequences and computer programs. However, the printed version must include enough detail

Conditions of publication Articles must not have been published previously, nor be under consideration for publication elsewhere. The main findings of the article should not have been reported in the mass media. Like many journals, Phil. Trans. R. Soc. B employs a strict embargo policy where the reporting of a scientific article by the media is embargoed until a specific time. The Executive Editor has final authority in all matters relating to publication.

Electronic Submission details For full submission guidelines and access to all journal content please visit the Phil. Trans. R. Soc. B website at publishing.royalsociety.org/philtransb.

AIMS AND SCOPE Each issue of Phil. Trans. R. Soc. B is devoted to a specific area of the biological sciences. This area will define a research frontier that is advancing rapidly, often bridging 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 cluster areas: neuroscience and cognition; organismal and evolutionary biology; cell and developmental biology; and health and disease. As well as theme issues, the journal publishes papers from the Royal Society’s biological discussion meetings. For information on submitting a proposal for a theme issue, consult the journal‘s website at publishing.royalsociety.org/philtransb. 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 suggestions / proposals for unsolicited review articles. Please submit an abstract and a covering letter to the Editorial Office for approval for consideration. 6,000 words is usually a maximum length for reviews.

The Royal Society is an independent scientific academy founded in 1660 and self-governing under Royal Charter. The Society has three roles, as the scientific academy of the United Kingdom, as a learned society, as a funding body. The objectives of the Royal Society are to

• recognise excellence in science • support leading-edge scientific research and its applications

ISBN: 978-0-85403-708-7

• stimulate international interaction

Copyright © 2008 The Royal Society Except as otherwise permitted under the Copyright, Designs and Patents Act, 1988, this publication may only be reproduced, stored or transmitted, in any form or by any other means, with the prior permission in writing of the publisher, or in the case of reprographic reproduction, in accordance with the terms of a licence issued by the Copyright Licensing Agency. In particular, the Society permits the 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 research or private study. SUBSCRIPTIONS 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 our journal fulfilment agent, Portland Customer Services, Commerce Way, Colchester CO2 8HP; tel: +44 (0)1206 796351; fax: +44 (0)1206 799331; email: [email protected] or by visiting our website at publishing.royalsociety.org/subscribers. The Royal Society is a Registered Charity No. 207043.

to satisfy most non-specialist readers. Supplementary data up to 10Mb is placed on the Society's website free of charge. Larger datasets must be deposited in recognised public domain databases by the author.

• further the role of science, engineering

printed version plus electronic access

Europe

USA & Canada

Research Support (UK grants and fellowships) Research appointments: 2547 Research grants: 2539 Conference grants: 2540 Science Advice General enquiries: 2585 Science Communication General enquiries: 2572

and technology in society

• promote the public’s understanding of science • provide independent authoritative advice on matters relating to science, engineering and technology

• encourage research into the history of science Subscription prices 2009 calendar year

For further information on the Society’s activities, please contact the following departments on the extensions listed by dialling +44 (0)20 7839 5561, or visit the Society’s Web site (www.royalsociety.org).

International Exchanges (for grants enabling research visits between the UK and most other countries (except the USA)) General enquiries: 2550 Library and Information Services Library/archive enquiries: 2606

All other countries

£2024/US$3845 £2136/US$4058 £2186/US$4153 /€2631

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.

Cover image: Chinese Whispers I–IX (Series I ) 2006. By Helen Flanagan. Nine carborundum prints on Somerset paper (Edition of five) Each approx 56.544cm.

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

3469

Establishing an experimental science of culture: animal social diffusion experiments A. Whiten and A. Mesoudi

3477

The multiple roles of cultural transmission experiments in understanding human cultural evolution A. Mesoudi and A. Whiten

3489

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

3503

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

3515

Studying cumulative cultural evolution in the laboratory C. A. Caldwell and A. E. Millen

3529

Investigating children as cultural magnets: do young children transmit redundant information along diffusion chains? E. Flynn

3541

The fitness and functionality of culturally evolved communication systems N. Fay, S. Garrod and L. Roberts

3553

Culture, embodiment and genes: unravelling the triple helix M. Wheeler and A. Clark

3563

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

3577

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

3591

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 Humans learn from other humans in a wide variety of domains. Consequently, systems of knowledge and behaviour are culturally transmitted in human populations. 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 programme 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 ([email protected]). One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

2. AN EXPERIMENTAL APPROACH TO CULTURE 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 biological evolution and the cultural evolution of words and languages), and the study of cultural transmission 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 laboratory 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 modifications 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 experiment) 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 miniculture 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 producing similar group-level behaviours (e.g. individual learning). The achievements in this area have been impressive: there is clear evidence for cultural transmission 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 taxonomic 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 individuals; 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, preferentially copying the behaviour of the population majority. Again, in common with the review of the nonhuman 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

Introduction transmitted behaviour changes over time, what is it changing towards? Griffiths et al. adopt a mix of mathematical and experimental diffusion chain techniques 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 transmitted 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 assumptions, 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 phenomenon. Furthermore, these inductive biases can overwhelm 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 simplified, manageable laboratory populations, and establishing 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)

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 payoff-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 individual 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 (preferentially copying high pay-off behaviours, but selecting the most frequently chosen response when the difference in yields is less marked). This use of payoff-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 payoffs 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 opendiffusion 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 achievements. 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 monotonic 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 transmission 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 cumulative 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: crosscultural 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 technological 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 crosscultural 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 transmitted patterns of behaviour in these populations therefore 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 considerably simplified and abstract). This symbolization is attributable to pressure for the participants to minimize their effort in producing graphics, while still maintaining distinct symbols for distinct concepts—in this sense both isolated pair and community systems are highly functional. Fay et al. show, however, that the community systems also simultaneously optimize their transmissibility (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 symbols 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 experimental 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 addressing particular questions. To take an example touched upon in this issue: while there is a general agreement that cumulative cultural evolution is an important subtopic 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 functionality 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

Introduction 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 inheritance (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. Furthermore, 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 widespread 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 coevolutionary 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)

K. Smith et al.

3473

role for extra-genetic transmission and adaptation. Embodied accounts of cognition emphasize the reciprocal relationship between an organism and its environment, 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 incompatibility 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 (potentially 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 process, 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 openended 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) conclusion 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 handedness 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 nicheconstructing 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 transmitted 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 transmission 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 relevant: psychologists concerned with evolutionary explanations 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, understanding 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 development, 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

Introduction 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 realworld 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 experimental 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 experiments 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 mathematical 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 experimental 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)

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.evolhumbehav.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 graphical 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 nicheconstruction 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 frequencydependent 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 cultural 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 identification 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 ([email protected]). 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 experimental 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, experimentally controlled innovations in behaviour are seeded into groups of individuals and the spread (or otherwise) of the innovation is tracked and documented. 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 resume´ 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 experiment temporarily faded from the literature. It was rejuvenated as a tool to study cultural transmission by Jacobs & Campbell’s (1961) explicitly titled ‘perpetuation of an arbitrary tradition through several generations 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 concerned 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, habituation 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 transmission; 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, diffusion studies have appeared with accelerating frequency (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 significance. Below we survey this growing corpus of studies and, most importantly, offer the first systematization 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.)

publication

des. str.

cultural diffusion? (no. of transmissions)

Successive replacements in trios exposed to alarming objects Transmission chains, seeded with alarm calls to arbitrary object Whole groups exposed to novel cues to buried food; spread of discovery in groups documented Three nut-cracking chimpanzees mixed with nine naive ones Models pecked through paper covers, in wild and captive flocks Spread of spontaneously initiated nut-cracking recorded in group Transmission chains seeded with models digging up carrot pieces Replication of 1990 study but incorporating a 24 hour delay Opportunity provided for group to use tools to probe for honey Rearing conciliatory stump-tailed macaques with rhesus, among whom reconciliation is relatively rare Transmission chains seeded with rats preferring different flavours

chimpanzees blackbirds baboons, vervets

Menzel et al. (1972) Curio et al. (1978a,b) Cambefort (1981)

3B 3C 1A

habituation effect stable (17) alarm calling stable (6) ?

chimpanzees pigeons chimpanzees rats rats chimpanzees rhesus macaques rats

Sumita et al. (1985) Lefebvre (1986) Hannah & McGrew (1987) Laland & Plotkin (1990) Laland & Plotkin (1992) Paquette (1992) de Waal & Johanowicz (1993)

2A 4A 2A 4C 4C 1A 5A

Laland & Plotkin (1993)

7C

Nuts cracked elsewhere introduced; spread in group recorded over several years Successive replacements in groups with initial flavour preferences

chimpanzees

1A

rats

Matsuzawa (1994) and Biro et al. (2003) Galef & Allen (1995)

? piercing stable over 55 days ? digging stable (8) stability less, with delay ? effect stable over six-week postmodel phase transmission shown (8) but fidelity variable ?

5B

Langan (1996)

7A

chimpanzees guppies

Tonooka et al. (1997) Laland & Williams (1997)

1A 5B

Similar to 1997 paper but more efficient alternative available

guppies

Laland & Williams (1998)

5B

Groups with mixed naive and experienced fishes created Young cowbirds housed with adults singing either of two different songs; repeated once first cohort became adults (models) Spread of route preference: models in familiarity and experience Spread logged off using either of two routes to escape a threat Flocks exposed to models feeding on blood from mock hen

guppies cowbirds

2A 5A

guppies guppies chickens

Reader & Laland (2000) Freeberg (1998) and Freeberg et al. (2001) Swaney et al. (2001) Brown & Laland (2002) Cloutier et al. (2002)

Wild, banded birds exposed to model using novel foraging method

keas (wild)

Gajdon et al. (2004)

3A

Captive chimpanzees given rough leaves used medicinally in wild

chimpanzees

Huffman & Hirata (2004)

1A

2A 7A 4A

? social transmission not found rapid waning over three transitions no significant evidence of social transmission ? (Continued.)

3479

magpie jays (wild)

A. Whiten & A. Mesoudi

Wild groups seeded with individuals trained to open specific small doors to feed Spread of dipping for honey with specific natural tool documented Founder shoals were seeded with preference for one of two routes

transmission shown, with slight waning (over 14) door opening stable over 3 days, but not door chosen ? differences transmitted (7) but waned by half differences transmitted (7) but waned ? ?

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

species studied

Review. Experimental science of culture

study content

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

6A

7C 4B

Price & Caldwell (2007)

Dindo et al. (2008) Stanley et al. (2008)

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

Two groups each seeded with a different foraging technique (via video) colobus monkeys of related species model Replication of Horner et al. (2006) with appropriately modified task brown capuchins Successive replacements of fishes focused on novel food task guppies, playfish

6A 7A Whiten et al. (2007) Hopper et al. (2007) chimpanzees chimpanzees

7A Bonnie et al. (2006)

As for 2005, but transfer between groups in foraging techniques Replication of Whiten et al. (2005), with one untrained model

7C Horner et al. (2006)

chimpanzees children chimpanzees Two transmission chains; opening artificial ‘fruit’ using alternative methods, plus control condition lacking model Two groups, each seeded with arbitrary convention to obtain food

7A Whiten et al. (2005) chimpanzees Two groups each seeded with model using tool in different way

6A Fragaszy et al. (2004) brown capuchins Juveniles exposed to adults using either of two methods to get juice

des. str. species studied

publication

spread to half group: stable, one corruption stable, spread across three groups minimal evidence of social transmission stable after 5 days with no model, one corruption stable (4) stable (13)

Review. Experimental science of culture

study content

Table 1. (Continued.)

techniques spread but social learning unclear moderate fidelity but stable over two months stable (8)

A. Whiten & A. Mesoudi

cultural diffusion? (no. of transmissions)

3480

(i) Open group diffusion (column A ) Here, a behaviour of interest is introduced into a whole group, as in the study of Lefebvre (1986) described above. Rather than exert experimental control over the potential channels of social learning such as was done in the work of Curio et al. (1978a), the open group approach leaves open for investigation which other individuals might attend to, learn from and possibly adopt the behaviour they see (figure 1). The open 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 likely to be more ‘messy’ than those of more 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 become difficult to disentangle. (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 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 what happens at each step in the diffusion process, and identify, for example, at what point a particular level of corruption occurs, contrasting with the complex 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 some parent–offspring relationships. For these, the linear chain design can be seen as simulating repeated intergenerational transmission, collapsing what in the wild may take decades into a diffusion chain experiment that may occupy only weeks (Horner et al. 2006). Excluding the studies in parentheses (which identified 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 the chain is both comfortable with being isolated from the remainder of their group and compatible with each other. Experience indicates that in primates at least, these can be very exacting requirements (Horner et al. 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)

condition designs (1–7)

A. open group

B. replacement

Curio et al. (1978a) Laland & Plotkin (1990, 1992) (Freeberg (1998), Freeberg et al. 2001 see also Col. A) (Whiten et al. (2007), see also Col. A) Laland & Plotkin (1993), Horner et al. (2006) and Dindo et al. (2008)

A. Whiten & A. Mesoudi

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) following no-model baseline 4. one experimental condition with one trained, seeded Lefebvre (1986) and Cloutier et al. (2002) Stanley et al. (2008) 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 seeded in each Freeberg et al. (2001) & Williams (1997, 1998) 6. two experimental conditions, with alternative actions Fragaszy et al. (2004), Price & Caldwell (2007) and seeded in each, after baseline, no-model control period Whiten et al. (2007) 7. two experimental conditions, with alternative actions Langan (1996), Brown & Laland (2002), Whiten et al. seeded in each, plus third, no-model control condition (2005), Bonnie et al. (2006) and Hopper et al. (2007)

C. linear chain

Review. Experimental science of culture

transmission designs (A–C)

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

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 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.)

3481

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

3482

A. Whiten & A. Mesoudi

(a)

Review. Experimental science of culture

(b)

(i)

A B

C H

E

F

A B C D

(c) (i) B C D E

D

B C D E F

C

A

G D H F (iii) G

B C

B C

H

E

E F G H (ii)

B C D E

E

D

F

A B C D

D

D E F G

G (ii)

(i)

A

E F G H (ii)

A B

C D E F

C D E F G

D E F G H

E F G H

(iii)

G E

B C

D H F

(iii) B D

C G

F H

E

Figure 1. Three principal diffusion experiment designs. (a) Open diffusion with one model seeded in each group and all members free to observe, learn or not; (b) replacement method with an experienced individual replaced by a naive one at each step; (c) linear chain with order of any transmission determined by experimenter. (i,ii) (corresponding to the designs described in row 7 of table 2) Seeding with a model acting in different ways (shaded); (iii) a no-model control group limited to individual learning. Arrows illustrate hypothetical diffusion of information, beginning with seeded model ‘A’.

generations’ that are experimentally imposed, unlike in the open diffusion approach. However, at each step, one naive individual replaces one of a group of experienced individuals, so that in this respect there are resemblances to the open group context; the experimenter will not necessarily know from which of the available models the novice learns, or if it learns from several; and again, the more experienced individuals may be influenced by how the later recruits behave. The replacement method can thus be regarded as something of an intermediate design, lying between the open group and linear chain approaches in our table (table 2 and figure 1). One important aspect of this method is that if the animals being studied are influenced by the number of models they witness, being predisposed to ‘copy the majority’ (a form of conformity, discussed further below), then positive social learning effects might be documented by a replacement approach, yet missed in a linear design that fails to sufficiently stimulate the social learning mechanisms available. In general, the replacement approach provides a good model of natural situations in which there is gradual turnover in a group. A human example is in the present issue (Caldwell & Millen 2008). Phil. Trans. R. Soc. B (2008)

(b) Experimental conditions designs In the seven rows of table 2, we distinguish what we call ‘conditions designs’ on the basis of the experimental and control conditions applied. Each of rows 2–7 includes the introduction of a model, usually a trained one but sometimes capitalizing on the natural emergence of an innovator. Whether we have successfully captured all of the relevant published studies in the present paper or not, the criteria for inclusion of a study in each of these rows appear quite clear. The same cannot be said of row 1, where the experimental approach is the most minimal, simply offering novel learning experiences in such a way that if an innovation occurs, its potential subsequent spread can be systematically tracked. In this row we have included studies that express this intent. For example, Paquette (1992) gave four chimpanzees the opportunity to use tools to dip for honey in an artificial ‘termite mound’ and recorded the emergence and spread of this behaviour. However, it could be argued that many other studies of the spread of behaviour patterns, not included here, share essential features with those listed in row 1. For example, the original Japanese macaque ‘preculture’ studies documented the spread of behaviour patterns elicited by novel learning experiences, such as washing human-supplied foodstuffs (Kawai 1965). Studies in table 2 are differentiated from these by the authors’ intent to conduct an experiment tracing diffusion, but the evidence for social learning remains only of the weakest, circumstantial kind, owing to a lack of any control condition where social learning is not possible. In the Paquette study referred to above, for example, we cannot be sure that the spread of the dipping behaviour was not simply the result of each chimpanzee developing this on its own account, rather than through observing those already dipping for honey. Our principal interest in the present paper is thus in the lower rows of the table. Broadly, as we further descend the rows of table 2, the power of the experimental designs to identify social learning, and in turn cultural transmission in the spread of the behaviour of interest, is enhanced. Row 2 differs from row 1 in that a known model is seeded, providing added focus about what behaviour pattern is to be subsequently tracked. However, the absence of a comparison with a control condition where there is no model means that evidence for social learning here still remains weak (see table S1 in the electronic supplementary material for details). Row 3 is the first where we see the incorporation of a control condition for individual learning, in this case through an initial baseline phase of exposure to the problem of interest, before subjects witness a model in the experimental phase. As in the case of the original ape and avian experiments of Menzel et al. and Curio et al., respectively, where responses changed dramatically between baseline and social learning conditions, compelling evidence of cultural transmission can here be obtained. However, the use of a baseline as the reference condition may remain weaker in some other contexts, wherein the behaviour is more likely to appear through exploration the longer the period of exposure; this may be the case for solving novel foraging problems, for example.

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

Review. Experimental science of culture Row 4 overcomes this limitation through a betweensubjects 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 preferentially 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 replacement 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 highlighting 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 diffusion 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 remainder 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, approximately half the difference between the two experimentally initiated preferences for one route over the other eroded over this period, suggesting that such alternative 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 replacements, 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 information 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 demonstrated 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 preferentially ( 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 transmission. We encourage further studies that experimentally 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 diffusions 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 manipulative 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/s10329003-0065-5) 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 frequencydependent 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 cultural 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 QUESTIONS THAN ANSWERS Cultural transmission is the process by which information 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 experiments 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 ([email protected]). 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’.

persistence of socially learned information in chains or groups that involve larger numbers of individuals. Yet such multiple-individual or multigenerational experimental 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 transmission 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 transmission, while Boyd & Richerson (1985) modelled conformist transmission (preferentially copying the most popular variant) and prestige/indirect bias (preferentially 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 distinctions and findings. A second advantage of adopting a cultural evolutionary 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 generated 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 populationlevel 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 documented 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 predictions, 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 closedgroup 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 material

B chain 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 reintroduce the method as a means of studying cultural change, and have updated the transmission chain method to conform to modern standards of experimental psychology (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 everyday 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 information 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 mathematical 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 representations 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 reconstructive 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 counterintuitive’, 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 thirdparty 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 individuallevel 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 preexisting 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, supernatural 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 generations 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 malefavourable 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 magnitude 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 differential 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 (memelike) 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). Similarly, 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

A. Mesoudi & A. Whiten

Review. Roles of cultural transmission experiments the transmission of minimally counter-intuitive concepts in general, but which specific minimally counterintuitive 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 successive 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)

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 responded 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 morecohesive 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 organization, 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 replacement method to simulate between- and within-group cooperation in the trading of goods. Groups of participants were taught to produce different types of paper models, with pay-offs increased when paper models from different groups were combined. In a ‘voluntaristic’ 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 significantly 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 cooperative 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 onethird of the participants chose the sanctioning societies, indicating an a priori aversion (or at best indifference) to the use of punishment. Despite this initial preference, 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 replacements 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 significantly 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 condition 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/replacement. The somewhat surprising result was that closed societies with no participant migration were significantly more innovative in generating novel interpretations 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 interpretations. 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 experimental 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. Consequently, several closed-group studies have appeared Phil. Trans. R. Soc. B (2008)

A. Mesoudi & A. Whiten

3495

generation 1 social learning

individual learning

2

3

A

B

A

B

A

B

C

D

C

D

C

D

A

B

A

B

A

B

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 replacement 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 ‘information 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 producers) 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: increasing 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 individually 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 modelbased 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). Substantial 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 information 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 pastoralists. 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 conformity). 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 archaeological 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 individual 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-successfulindividuals 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 copysuccessful-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 arrowhead 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 maintaining within-group diversity in arrowhead designs. During periods of cultural learning, different participants 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 psychology 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 nonhuman 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 nonconformists, 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 contrast, 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 landscapes 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 systematically 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 differences 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 supplementary 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. Inspiration 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 processes 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 mathematical 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 understanding 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 Psychology), 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 experiments 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 microand 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 microsocieties including traditions of rule giving and rule following. Evol. Hum. Behav. 25, 305–326. (doi:10.1016/j.evolhumbehav.2004.05.003) Bettinger, R. L. & Eerkens, J. 1999 Point typologies, cultural transmission, and the spread of bow-and-arrow technology 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 replication 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:10.1037/00223514.39.3.431) 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 transmission 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., Happe´, 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 “copysuccessful-individuals” cultural learning strategy: adaptive landscapes, producer-scrounger dynamics, and informational 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 experimental simulation. Am. Antiq. 73, 3–28. Mesoudi, A. & O’Brien, M. J. 2008 The cultural transmission of Great Basin projectile point technology II: an agentbased computer simulation. Am. Antiq 21, 350–363. (doi:10.1016/j.evolhumbehav.2008.04.005) Mesoudi, A. & Whiten, A. 2004 The hierarchical transformation of event knowledge in human cultural transmission. 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 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 ([email protected]). One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

information is passed from one person to another (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. hypothesis data

Review. Inductive biases and cultural evolution hypothesis

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 psychology 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 reconstructing 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. Examining 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 mathematical 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 component. 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 & Claidie´re 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 understanding 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 transmitted 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 experiments 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 compositionality—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 summarize 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 PLA ðhn Z ijdÞPPA ðdjhnK1 Z j Þ; Z

ð3:1Þ

d

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 PðhjdÞ Z P

PðdjhÞPðhÞ ; Pðdjh 0 ÞPðh 0 Þ

ð3:2Þ

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 likelihoods. 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 understanding 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). The learning algorithms we will consider are based on the posterior distribution produced by applying Bayes’ rule. ‘Learning’ in the present context refers 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 Psamp ðhjdÞ Z P

PPA ðdjhÞPðhÞ ; PPA ðdjh 0 ÞPðh 0 Þ

ð3:3Þ

h 0 2H

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 probability 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 consistent with Bartlett’s conclusions about serial reproduction 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 constraints 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 ( 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 relationship 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 importance 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 consistency 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 information 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, Pðh Z 1jd Z 1Þ Z

ð1KeÞp ; ð1KeÞp C eð1KpÞ

Pðh Z 1jd Z 2Þ Z

ep ; 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 effects of inductive biases (expressed in the priors of those agents) on knowledge transmission: the probability 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.

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

10

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 ;

ð3:6Þ

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 mathematical 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)

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 approximated 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 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 difficultyof-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

(iv)

(b) (ii)

(v)

(iii)

(vi)

0.12

(i)

(ii)

1 2 3 4 5 6 7 8 9 10 iteration

1 2 3 4 5 6 7 8 9 10 iteration

3509

0.10 probability of selecting concept

(a) (i)

T. L. Griffiths et al.

0.08 0.06 0.04 0.02 0

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

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

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

3510

T. L. Griffiths et al.

(a) (i)

(ii)

Review. Inductive biases and cultural evolution (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 knowledge transmission. The results support the conclusion that knowledge transmission converges to an equilibrium 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 information when transmitting a sequence of problemsolving moves to an observer in the next generation. Phil. Trans. R. Soc. B (2008)

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

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

Review. Inductive biases and cultural evolution 1998). Let xi denote the proportion of a population of agents entertaining hypothesis i at a given moment t, and qij denote the probability that a learner chooses hypothesis i after seeing the data generated from hypothesis j, as defined in equation (3.1). If we assume that each learner learns from a random member of the population, then the population proportions evolve as X dxi Z fj qij xj Kfxi ; dt j

ð5:1Þ

where fj is the fitness of people who subscribe to P hypothesis j; fZ k fk xk is the mean fitness; and the second term on the right-hand side ensures that P i xi Z 1. In biological evolution, fitness reflects the number of offspring produced by an individual of a particular type. In cultural evolution, it is more natural to interpret fitness as influencing the probability with which an individual chooses an agent from the previous generation as a source of data. If agents are selected with probability determined by fj, the same dynamics hold.4 Equation (5.1) has been extensively applied to cultural evolution for the case of languages, in the form of the ‘language dynamical equation’ explored by Nowak et al. (2001, 2002). In this work, fitness is typically assumed to be a function of how well speakers of a particular language can communicate with the population at large, implementing a selection pressure for communication. If we instead assume that all speakers have equal fitness, fjZ1, equation (5.1) simplifies to X dxi Z qij xj K xi ; dt j

ð5:2Þ

which is a linear dynamical system. This is a ‘neutral’ model, in which there are no selective forces favouring one language or hypothesis over another. A special case of this model was analysed by Komarova & Nowak (2003). The neutral model characterizes the evolution of a population in the absence of selection, and thus provides a valuable null hypothesis against which to evaluate claims about selective forces, as well as a way to study the effects of mutation. It also gives us a way to connect the replicator dynamics to transmission chains. The asymptotic behaviour of this linear dynamical system is straightforward to analyse: it converges towards an equilibrium at the first eigenvector of the transition matrix Q (for details, see Griffiths & Kalish 2007). This means that the neutral form of the replicator dynamics displays asymptotic behaviour that is very similar to that of transmission chains involving discrete generations of single learners. The key difference is in the nature of the quantities that converge: with discrete generations of single learners, it is the probability with which a particular learner entertains hypothesis i that converges to the stationary probability; under the replicator dynamics, it is the proportion of the population that entertains hypothesis i that converges to this probability. The results from the previous sections characterize the consequences of cultural evolution not only for individuals but also for populations. This provides an Phil. Trans. R. Soc. B (2008)

T. L. Griffiths et al.

3511

additional justification for the use of transmission chains in studying cultural evolution: the parallel between the stationary distributions of such chains and the equilibria of the replicator dynamics in populations provides a way to gather clues about the behaviour of populations using a paradigm that is easily simulated in the laboratory. (b) Inductive biases can overwhelm selective pressures In addition to indicating how transmission chains can inform the study of cultural evolution more broadly, this connection provides us with a way to generalize our mathematical results to cases where selective forces also influence the adoption of hypotheses. This can allow us to evaluate whether inductive biases can play a more significant role in cultural evolution than selection, as suggested by Sperber (1996), Boyer (1998) and Atran (2001), or whether selection is the more powerful force, as argued by Henrich & Boyd (2002). While obtaining general analytical results is difficult, we can at least gain an idea of how these forces interact by returning to our example with just two hypotheses. With the two hypotheses, the fact that x 1Cx 2Z1 means that we can work with just one variable. We will use x 1, the proportion of agents choosing hypothesis 1, and denote this x for simplicity. In §3b, we defined the matrix Q as a function of the prior probability of hypothesis 1, p, and the noise rate, e. In the neutral model from §5a, where the fitness of both hypotheses is equal (i.e. each generation chooses an agent to learn from at random from the previous generation with uniform probabilities), the equilibrium of the system is given by finding a value of x such that equation (5.2) is equal to zero. It is straightforward to show that this is equivalent to solving equation (3.6), and thus the equilibrium is given by xZp. The critical question is how this equilibrium is affected by selection, as represented by unequal fitness for the two hypotheses. We will assume that the fitness of hypothesis 1 is f1Zs and hypothesis 2 has constant fitness f2Z1. We are interested in the case where sO1. This higher fitness might reflect higher social status accorded to those who adopt the hypothesis, greater success in solving problems posed by the environment as a consequence of having this belief or some other indicator of success that might make others more likely to try to learn from these ‘fit’ individuals. The equilibrium of the resulting system is given by finding x such that equation (5.1) is equal to zero. Simplification for the case of the two hypotheses reduces this to the quadratic equation dx Z ð1KsÞx2 C ðð1K q21 Þs K q12 K 1Þx C q12 ; dt

ð5:3Þ

which can be solved by standard methods to find an equilibrium for a particular choice of s, q12 and q21. Figure 5a shows how the equilibrium changes as a function of s for pZ0.2 and e2 0:01; 0:05. As might be expected, increasing s increases the representation of hypothesis 1 in the equilibrium solution. We can use equation (5.3) to explore the relative contributions of the prior probability of a hypothesis p and the strength of selection s on the equilibrium of this

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

3512 (a)

T. L. Griffiths et al.

Review. Inductive biases and cultural evolution (b)

1.0 0.9 0.8

100

0.7 0.6

0.5 0.4 0.3 0.2 0.1

10

0.5

0.2 1

10

102

0.4

0.3

0.2

no ise

1 0.5

0.3

lev el,

0.4

0.1

0

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 sZ

1K2q12 ; 1K2q21

ð5:4Þ

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)

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 generations, 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 generations. 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 determining 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 graphical 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. TwentySeventh 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 representations. 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 Population 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 transmission 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 cultural 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. ¨ zyu¨rek, A. 2004 Children creating Senghas, A., Kita, S. & O 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. & Claidie´re, 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, Blu¨mlisalpstrasse 10, 8006 Zu¨rich, 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 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 chimpanzees (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 ([email protected]). One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

serious exchange between the evolutionary anthropology 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, nonhuman 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 experiments. 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 informational 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 humanlike 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), depending 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 anthropology 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 conspecific, 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 positive frequency dependence

0.8

0.8

0.6 0.4

0.6 0.4

0.2

0.2

0

0 pay-off biased

(d)

pay-off biased + frequency dependence 1.0

0.8

0.8

frequency of trait

1.0

0.6 0.4

0.6 0.4

0.2

0.2

0

0 0

0.2 0.4 0.6 0.8 1.0 frequency of trait before imitation

3517

positive frequency dependence 1.0

(c) expected frequency of trait after imitation

(b)

1.0

frequency of trait

expected frequency of trait after imitation

(a)

R. McElreath et al.

5

10

15 time

20

25

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

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 recombination 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

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 dashed curve in figure 1d is the dynamics of a mix of pay-off bias and positive frequency dependence. For different mixes of these and other strategies, different evolutionary dynamics are expected. (e) Modelling integrated pay-off-biased and frequency-dependent social learning While there has been modelling effort devoted to studying linear, unbiased social learning, frequencydependent social learning and pay-off-biased social learning, to our knowledge no theoretical study has simultaneously examined these options in the same context. Therefore, we finish this section by presenting an extension of existing evolutionary theory that includes frequency-dependent bias, pay-off bias and a hierarchical integration of the two. We construct recursions for the dynamics of genes controlling these different learning strategies, as well as for the frequency of adaptive learned behaviour. We then analyse this gene-culture system in order to understand what environments favour different strategies. Consider a large population living in a uniform but temporally varying environment. Each individual faces a choice of two discrete behaviours. One of these choices yields a fitness benefit B, a proportion a of the time, yielding an average of aB. The other yields an average bB!aB. For each generation, there is a chance u that the better behaviour switches to the other option. These changes cannot be observed by individuals. Behaviour is acquired via learning, either individually or socially. Individual learning (I) pays an average learning cost in order to determine which option is better. This makes the fitness of an individual learner: W ðIÞ Z w0 C aBKc; where w0 is baseline fitness from other behaviour and c is the average cost of learning. Social learning can be unbiased (linear, L), frequency dependent (conformist, C), pay-off biased (S) or pay-off conformity (SC). Linear social learning copies a random adult from the previous generation, resulting in average fitness: W ðLÞ Z w0 C aBq C bBð1KqÞ Z w0 C Bðaq C bð1KqÞÞ: The frequency of currently optimal behaviour, q, has its own dynamics, which we define below. The important point here is that linear social learning does not transform this proportion in any direct way. On average, it replicates the frequency of optimal behaviour across generations. Positive frequency dependence, conformity (C), does however transform q. We assume perhaps the simplest conformity heuristic. The learner samples three random adults from the previous generation and then adopts the most common behaviour among these three models. Since the chance that any one model has optimal behaviour is q, the binomial distribution (table 1) allows us to compute the probability of any combination and therefore the probability of the Phil. Trans. R. Soc. B (2008)

Table 1. Probabilities of acquiring optimal behaviour via social learning, for the three nonlinear strategies positive frequency dependence (C), pay-off bias (S) and pay-off conformity (SC). The sample column gives all possible 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 indicates optimal behaviour and B or b indicates non-optimal behaviour. By multiplying each probability of a specific sample occurring by chance in a particular strategy column, one can sum these products to compute an expected probability of acquiring optimal behaviour via a given strategy. In the Pr(sample) column, q is the frequency of optimal behaviour in the entire adult population and a and b are the probabilities of optimal and non-optimal behaviours, respectively, returning large pay-offs.

sample Pr(sample) AAA AAa Aaa aaa AAB AAb AaB Aab aaB aab ABB ABb Abb abb aBb aBB BBB BBb Bbb bbb

q3$a3 q3$3a2(1Ka) q3$3a(1Ka)2 q3$(1Ka)3 3q2(1Kq)$a2b 3q2(1Kq)$a2(1Kb) 3q2(1Kq)$2a(1Ka)b 3q2(1Kq)$2a(1Ka) (1Kb) 3q2(1Kq)$(1Ka)2b 3q2(1Kq)$(1Ka)2 (1Kb) 3q2(1Kq)2$ab2 3qð1KqÞ2 $a2bð1KbÞ 3qð1KqÞ2 $að1KbÞ2 3q(1Kq)2$(1Ka)(1Kb)2 3qð1KqÞ2 $ð1KaÞ2bð1KbÞ 3qð1KqÞ2 $ð1KaÞb2 ð1KqÞ3 $b3 ð1KqÞ3 $3b2 ð1KbÞ ð1KqÞ3 $3bð1KbÞ2 ð1KqÞ3 $ð1KbÞ3

Pr (1rC)

Pr (1rS)

Pr (1rSC)

1 1 1 1 1 1 1 1

1 1 1 1 2/3 1 0 1

1 1 1 1 1 1 0 1

1 1

0 2/3

0 1

0 0 0 0 0 0 0 0 0 0

1/3 1 1 1/3 0 0 0 0 0 0

0 1 1 0 0 0 0 0 0 0

conformist learner acquiring optimal behaviour is qC Z q C qð1KqÞð2q K1Þ: Using this expression gives us a mean fitness for C, W ðCÞ Z w0 C BðaqC C bð1K qC ÞÞ: Pay-off-biased social learning (S) samples three individuals and adopts the behaviour with the highest average observed pay-off. We compute the expected probability of acquiring optimal behaviour through this heuristic in the same fashion as for conformity: each of the three models sampled has a chance q of having optimal behaviour and each model then has a chance either a or b of displaying a pay-off of B. Thus, the probability of any combination of underlying behaviour and displayed pay-offs can be computed from the binomial distribution (table 1). This results in a chance of acquiring optimal behaviour: qS Z q C qð1KqÞðað2 C bð2K3bð1KqÞK4qÞÞ C a2 ð3b K1Þq C bðbð1KqÞK2ÞÞ:

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

Analysis on social learning strategies (i) (a)

0

(ii) I

(iii)

(iv)

R. McElreath et al.

3519

(v)

L

u

0.6 0

(ii)

(iii)

L

C

(ii) L

(iii)

S b

SC

(iv)

(v)

0.5

I

a–b

(b)

C 0 (i)

0.5 (c)

0 (i) 0 I

S b

SC 0.5

(iv)

(v)

u 0.6

C 2

S B

SC 10

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

The complete evolutionary system is very difficult to 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

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 dependence 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 conformity—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 behavioural 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 experimental 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 information, 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 performance. 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 decisionmaking 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

R. McElreath et al.

proportion choosing optimal crop

(a)

Analysis on social learning strategies

1.0

(b)

0.9

proportion neighbours examined

3522

0.8 0.7 0.6

0.65

0.60

0.55

0.5 0.50 2

4

6 8 10 round in farm

12

14

2

4

6 8 10 round in farm

12

14

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.

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 frequencydependent 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) frequencydependent 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 dependence. 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.

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.

Ai;tC1 Z ð1KfÞAi;t C fpi;t ;

(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

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 probabilistic 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

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

(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

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

Analysis on social learning strategies expðl Ai;t Þ ; expðl A1;t Þ C expðl A2;t Þ

where Q indicates a vector of all parameters and l is a parameter that measures the influence of differences between attraction scores on choice. When lZ0, choice is random with respect to attraction scores. As l/N, choice becomes deterministic, in favour of the option with the higher attraction score. (ii) Frequency-dependent social learning To model the family of strategies that use the frequency of behaviour among group members, we modify the learning model above to cue choice by the frequency of options seen. Attractions are updated as before, but choice is given by the rule expðl Ai;t Þ expðl A1;t Þ C expðl A2;t Þ nf C g f i;t f ; n 1;t C n 2;t

PrðijAt ; QÞtC1 Z ð1KgÞ

where ni;t is the count of neighbours observed to have chosen option i in period t; g measures the weight of social information on choice; and f determines how nonlinear frequency dependence is. When fZ1, imitation is unbiased. When f O1, however, more common options have exaggerated chances of being copied, resulting in positive frequency dependence, such as majority rule conformity. When f!1, frequency dependence is negative, and more commonly observed options are less likely to be copied. Since changes in choice feedback to changes in attraction scores, even though this strategy has the same attraction updating recursion as individual learning, reinforcement patterns may be quite different.

0.8 0.6 0.4 0.2 0 –4 –2 0 2 4 difference in observed mean pay-offs

Figure 4. The function that determines the reliance on payoff-biased learning, as a function of the observed difference in means, p 1;t Kp 2;t . See the description of the hierarchical means/conformity strategy in the text. Solid curve, dZ1/10; dashed curve, dZ2.

symmetrical logistic function to model the change in reliance on pay-offs, as the distance between the observed means increases. Let Y ðd; p 1;t ; p 2;t Þ be the proportion of choice that is driven by individual updating, where d is a new parameter that determines how quickly reliance on pay-offs decreases, as the difference in observed means increases, Y ðd; p 1;t ; p 2;t Þ Z

2 hY: 1 C expðdðp 1;t Kp 2;t Þ2 Þ

Figure 4 plots this function for two values of d. The probability of choosing i under the hierarchical compare means/individual strategy is PrðijAt ; QÞtC1 Z ð1KgÞ

(iii) Compare means This pay-off-biased strategy attends to neighbours’ yields and chooses the option with the highest observed mean. It uses the choice rule PrðijAt ; QÞtC1 Z ð1KgÞ

Cg

expðl Ai;t Þ expðl A1;t Þ C expðl A2;t Þ

p 100 i;t ; C p 100 p 100 1;t 2;t

P where p i;t Z j pi; j;t =ni;t is the mean pay-off observed for option i in period t over all group members j, including oneself. Raising these average pay-offs to a large power creates an approximate step function, so that one or the 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. We fix f Z100 in order to force the model to match our theory, i.e. a threshold behaviour. (iv) Hierarchical compare means/individual learning This strategy uses the comparison of choice means and individual updating, but in a manner different from the pure compare means model. Using the distance between estimated means as a cue of uncertainty, the strategy falls back on individual learning (attraction updating) when the means are similar. We use a Phil. Trans. R. Soc. B (2008)

3523

1.0 proportion of pay-off-biased social learning

PrðijAt ; QÞtC1 Z

R. McElreath et al.

expðl Ai;t Þ expðl A1;t Þ C expðl A2;t Þ

C g ð1KY Þ

p 100 i;t C p 100 p 100 1;t 2;t

! expðl Ai;t Þ CY : expðl A1;t Þ C expðl A2;t Þ For similar observed means, the individual learning component will dominate the social learning term. Otherwise, the individual will mainly attend to differences in observed means. However, if d is a very large number, then only a very narrow range of very similar observed means will lead to falling back on individual updating. (v) Hierarchical compare means/frequency-dependent social learning This model is like the previous, but falls back on frequency-dependent social learning, as the difference in observed means increases. PrðijAt ;QÞtC1 Zð1KgÞ

expðl Ai;t Þ expðl A1;t ÞCexpðl A2;t Þ

! ni;tf p 100 i;t CY f Cg ð1KY Þ 100 : f p 1;t C p 100 n1;t Cn2;t 2;t

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 PrðDj jAtK1 ; QÞt ; LðDjx; QÞ Z t

Q 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 log PrðDj jAtK1 ; QÞt : Klog LðDjx; QÞ ZK 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 components 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 information 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 competing 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 experimental 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; where k is the number of free parameters in model x. We use the common sample-size-adjusted version of the above, AICc (Burnham & Anderson 2002), and this is what we display in our results: AICc;x ZK2 log LðDjx; PÞ C 2k C

2kðk C 1Þ ; nKk K 1

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 1 ; wx Z P 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

model

AICc

Akaike weight

f

l

g

d

f

HCMFD HCMINDIV compare means frequency dependent individual

6519.59 6918.89 6924.23 6929.34 7004.25

z1 !0.001 !0.001 !0.001 !0.001

0.6605 0.4620 0.4611 0.4998 0.4382

0.1645 0.1917 0.1921 0.1814 0.1866

0.3365 0.1400 0.1239 0.1349 n.a.

3.210 4.982 n.a. n.a. n.a.

1.953 n.a. n.a. 3.396 n.a.

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

1.0 probability of copying choice

Table 2 presents the AICc, Akaike weight and parameter estimates for each model, sorted from best to worst fitting model. The bulk of evidence favours model 5, hierarchical compare means/frequency dependence. While there is no doubt about heterogeneity 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 dependence or conformity (figure 5). The maximumlikelihood 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 likelihoods 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. Therefore, 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 frequency of choice

2.5

3.0

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

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 consequences 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 themselves are culturally transmitted, because the population 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 counterintuitive 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 payoffs—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 information, 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 microsocieties including traditions of rule giving and rule following. Evol. Hum. Behav. 25, 305–326. (doi:10.1016/j.evolhumbehav.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 multimodel 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. Monographs 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 behavioral 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 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 understanding 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 explaining 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 considerably 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 ([email protected]). One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

only creative invention but also, and just as importantly, 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 modifications 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 evolutionary 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: ‘cumulative 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 accumulation 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 modifications of socially learned behaviour. One of these is nutcracking 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 chimpanzees, 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 manufacture, 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 ‘observational 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 enhancement’; 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 foundations 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. Understanding 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 conspecifics 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 influence 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 therefore 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 forward 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 cumulative 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 findings. 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 mathematical 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 Mesopotamians 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 circumstances 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 structural 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 assumptions, 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 replacement 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 individuals 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 replacement 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 replacement 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 participants 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 participants 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 increasingly 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 generations, 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 modifications 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 minutes

participants present in test group

00.00 1 2 3 02.30 1 2 3 4 2 3 4 5 05.00 3 4 5 6 07.30 4 5 6 7 10.00 5 6 7 8 12.30 6 7 8 9 15.00 7 8 9 10 17.30 8 9 10 20.00 9 10 22.30 10 25.00 Figure 1. Group composition over time in the microsociety design. Generational succession is simulated through the repeated removal of experienced participants and introduction 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, indicating 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 participants, 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 improvement 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 investigating 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 modification). 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

C. A. Caldwell & A. E. Millen (a)

1250

dist. flown by plane (cm)

3536

1000

Review. Cumulative culture in the laboratory

(i)

(ii)

(i)

(ii)

750 500 250 0

(b)

100

height of tower (cm)

80 60 40 20 0 1

2

3

4 5 6 7 position in chain

8

9

10

1

2

3

4 5 6 7 position in chain

8

9

10

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. Convergent 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 convergently 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 convergence 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)

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 convergence 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

(a)

5

mean similarity rating

Review. Cumulative culture in the laboratory

4 3 2

(b)

5

mean similarity rating

1

4 3 2 1 1

2

3

4 5 6 7 position in chain

8

9

10

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 nonetheless illustrative of phenomena that we believe to be operating over much longer time scales, in many realms of behaviour.

C. A. Caldwell & A. E. Millen

3537

Using our paper aeroplane task detailed previously, we have run a variety of different experimental conditions, in which certain sources of social information are either available or unavailable. While in our previous experiment (Caldwell & Millen 2008) participants could observe the two previous participants in the chain, and discuss the task with them, as well as see the results of their efforts in the form of the completed plane and information about its flight distance, in this study we separated these sources of information. In a series of different experimental conditions, participants had access to information in the form of either: actions only; results only; teaching only; actions plus results; actions plus teaching; results plus teaching; or actions, results and teaching. If imitation (in terms of ‘learning to do an act by seeing it done’, e.g. Whiten & Ham 1992) is indeed crucial to cumulative cultural evolution, then conditions in which this not possible (i.e. those with no information from actions) ought to show much weaker trends towards 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 scaleddown 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 (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 cumulative 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. Phil. Trans. R. Soc. B (2008)

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 microsocieties including traditions of rule giving and rule following. Evol. Hum. Behav. 25, 305–326. (doi:10.1016/ j.evolhumbehav.2004.05.003)

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 Christiansen & 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.evolhumbehav.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:10.1037/00223514.44.5.977) Jacobs, R. C. & Campbell, D. T. 1961 The perpetuation of an arbitrary tradition through several generations of laboratory 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 transmission 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 experimental 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 cultural 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 compositional structure. Adv. Complex Syst. 6, 537–558. (doi:10. 1142/S0219525903001055) Sugiyama, Y. 1997 Social tradition and the use of toolcomposites 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 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 understanding 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 regarding the definition of key terms within social learning. *[email protected] One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

In the present study, definitions for two critical social 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)

(b)

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 irrelevant 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)

systematically imitating the method they have witnessed used to reach the same end state, instead they are emulating the goal. The task used in the present study was the glassceiling 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 experiment each tube was baited with a reward, a Velcrobacked 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 demonstration, 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 irrelevant 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 mechanical 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 demonstrator (Uzˇgiris 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 something (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 concentrated 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 adultexperimenter 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 transmission 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 individual 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 generations’ 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 investigate 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 transmitted 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 observational 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.

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.

(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

(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.

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

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-yearolds, 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.

(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 Velcrobacked 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 demonstrated 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 meaningful 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 condition 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 significantly 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 irrelevant 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 transmitted 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 dooropening 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 opaque

3-year-olds

(db)(db)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDR)(RR) (db)(db)(t)(t)(t)(SDR)(RR)

(pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR)

(db)(db)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR)

(pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR)

(pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)(SDR)(RR)

(SDL)(RR) (SDL)(RR) (SDL)(RR) (SDL)(RR)

(SDL)(tu)(RR) (SDL)(tu)(RR) (SDL)(tu)(RR) (SDL)(tu)(RR)

(pb)(SDR)(RR) (pb)(pb)(SDL)(RR) (pb)(SDL)(RR) (pb)(SDL)(RR)

(pb)(SDR)(RR) (pb)(SDR)(RR) (pb)(SDR)(RR) (pb)(SDL)(RR)

(SDL)(RR) (SDL)(RR) (SDL)(RR) (SDL)(RR)

(SDL)(tu)(RR) (SDL)(tu)(RR) (SDL)(tu)(RR) (SDL)(tu)(RR)

(SDL)(RR) (SDL)(RR) (SDL)(RR) (SDL)(RR)

(pb)(pb)(SDR)(RR) (tb)(SDR)(RR) (hp)(hp)(SDR)(RR) (hp)(hp)(SDR)(RR)

(SDL)(RR) (SDR)(RR) (SDL)(RR) (SDL)(RR)

(SDR)(SDL)(tu)(RR) (SDR)(SDL)(tu)(RR) (SDR)(SDL)(tu)(RR) (SDL)(tu)(RR)

(SDL)(RR) (SDR)(RR) (SDR)(RR) (SDL)(RR)

(tb)(SDR)(RR) (tb)(SDR)(RR) (SDR)(RR) (tb)(SDR)(RR)

(SDL)(RR) (SDL)(RR)

(SDL)(SDR)(x20tu)(RR) (SDR)(SDL)(x11tu(RR) (SDL)(SDR)(x4tu)(RR) (SDL)(SDR)(SDL)(tu)((RR))

(SDL)(RR) (SDL)(RR) (SDL)(RR) (SDL)(RR)

(SDL)(SDR)(SDL)(tu)(RR) (SDL)(SDR)(SDL)(tu)(RR)

(SDL)(RR) (SDL)(RR)

(db)(db)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR) (db)(db)(t)(t)(t)(SDL)(RR)

(pb)(pb)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR) (pb)(pb)(t)(t)(t)(SDL)(RR)

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

(SDL)(RR) (tb)(SDL)(RR) (SDL)(RR) (SDL)(RR)

(pb)(pb)(hp)(hp)(t)(t)(t)(SDR)(RR) (hp)(hp)(t)(t)(t)(SDR)(RR) (pb)(pb)(t)(t)(t)(SDR)(RR) (pb)(pb)&&(t)(t)(t)(SDR)(RR)

(pb)(t)(t)(t)(SDL)(RR) (pb)(t)(t)(t)(SDL)(RR) (pb)(t)(t)(t)(SDL)(RR) (pb)(t)(t)(t)(SDL)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(pb)(pb)(SDL)(RR) (pb)(pb)(SDL)(RR) (pb)(pb)(SDL)(RR) (pb)(pb)(SDL)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(pb)(pb)(SDL)(RR) (pb)(pb)(SDL)(RR) (pb)(pb)(SDL)(RR) (pb)(pb)(SDL)(RR)

(SDL)(RR) (SDL)(RR) (SDL)(RR) (SDL)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(SDR)(RR) (SDR)(RR) (SDR)(RR) (SDR)(RR)

(tb)(SDL)(tu)(RR) (SDL)(tu)(RR) (SDL)(RR) (SDL)(RR)

(SDR)(RR) (SDR)(RR)

(SDL)(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 followed a binomial distribution from which the expected distribution of scores could be calculated. The distribution of scores in the actual diffusion chains Phil. Trans. R. Soc. B (2008)

was compared to the expected distribution of scores using a chi-squared statistic. The goodness-of-fit test proved to be significant (c232 Z 53:63, p!0.001; Spiegel 1961). It was found that there were more cases with

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-yearolds, 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 overimitation 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 transmission 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 importantly, 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), comparisons 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 transmission 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 understanding 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 Fourteento 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 tooluse 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. & Kira´ly, 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 troglodytes) 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 cultural 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.14390310.2006.01269.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) Uzˇgiris, 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 Like everything else in the natural world, communication 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 ([email protected]). One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

Dediu & Ladd 2007), cultural transmission, the historical transmission of languages across generations of learners, could optimize language via linguistic 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 functionalist 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. Technological evolution offers a compelling historical example of cumulative cultural evolution. Employing an organic–mechanical analogy, Basalla (1988) proposed 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 improvement 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 laboratory conditions, where performance in producing artefacts improves, or ratchets up, over successive generations (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 biological 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 presentday 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 parameters 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 graphical 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) 1

(d )

(c) 2

(e) 2

1

(f) 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 recurring 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 representations 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 microsocieties, 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. Participants 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

Culturally evolved communication systems 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 2 3 4 5 6 7

1 1 1 1 1 1 1

and 2 and 4 and 6 and 8 and 3 and 5 and 7

3 3 3 3 2 2 2

and and and and and and and

4 2 8 6 4 6 8

5 5 5 5 5 3 3

and 6 and 8 and 2 and 4 and 7 and 7 and 5

7 7 7 7 6 4 4

and 8 and 6 and 4 and 2 and 8 and 8 and 6

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)

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 meaningfulness); encoding of learned signs addresses the effort required to discriminate between the signs themselves (low meaningfulness); and detection determines the ease with which the signs are perceived (no meaning). Clearly an effective sign should be easily detected, efficiently encoded into memory and its meaning should be accurately and efficiently derived from the sign. However, it may be that community evolved signs are 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 participated 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 community 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. postinteraction; 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 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. Phil. Trans. R. Soc. B (2008)

The timed decoding task required participants to 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

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

Culturally evolved communication systems

N. Fay et al.

3557

Microwave 500 ms

40 ms

match or mismatch

50 ms

Figure 3. A sample trial sequence from experiment 1.

Results All participants successfully completed the training phase before progressing to the reaction time task. A majority learned the image–label pairing to the predetermined 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). Performance at round 1 is equivalent. At round 7, community evolved signs are more accurately recognized 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 calculated 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)

1100 1050 response time (ms)

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.

83%

83%

82%

1000 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 iconicitybased 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). Experiment 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 participated 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

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). Individual 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 complexity 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 inspection time (ms)

3558

30 25 20 15 10 5 0

1

round

7

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 participated 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

N. Fay et al.

3559

60

100 90 80 70 60 50 40 30 20

50

R1

R7 10

R1

R7 20

R1

R7 30

per cent correct

hit rate (%)

Culturally evolved communication systems

40 30 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 community 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 participated 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)

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 presentation 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 communication 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 glossogeny, 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 increasingly 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, becoming 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 promoting 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, communication 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 community). 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. Interestingly, 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 noncommunity participants indicates that diversity alone cannot account for the benefits of community evolved linguistic description schemes. In conclusion, like the progressive adaptation characteristic 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 community 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 graphical 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 characteristics on user performance: the role of icon concreteness, 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. Harmondsworth, 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 School of Philosophy, Psychology, and Language Sciences, David Hume Tower, George Square, Edinburgh EH8 9J7, UK

2

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; Noe¨ 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 ([email protected]). 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 embodied cognition generate the first image of human cognitive 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 components 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 landscape 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 organisms collectively build structures that persist beyond their own lifetime. A familiar example is the communally 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 practices resulting from individual lifetime learning, when combined with the physical persistence of artefacts, yields yet another source of potentially selectionimpacting 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 corresponding 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 associating each shape with a kind of drink. The bartender, by creating persisting spatially arrayed stand-ins for the drinks orders, actively structures the local environment 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 transform 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 computation’, 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 developmental 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 sophisticated 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 alongside (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 ‘architectural 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 cognitive 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 internalization. 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 practices, 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 adaptations (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 particular 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. According to the evolutionary-psychological picture, the modules that comprise our innate cognitive endowment 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 psychologists typically argue that the last time any significant modifications were made by selection to the human brain’s functional architecture was during the Pleistocene 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 fitnessdecreasing 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 important 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 psychologists 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 psychological 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 developmental trajectories demonstrates is that the suite of cognitive modules possessed by humankind is not strictly universal and so cannot constitute our specieswide human nature. What might then? The answer, nicely isolated by Buller (2005), is an evolved specieswide set of genetically specified developmental programs 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 variability 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 acquisition device) may be exposed to different developmental 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 domaingeneral 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 sophisticated 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 knowledge’. 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 intrinsically 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 environment. Moreover, this perspective provides one platform 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 wandering; landmark detection; and map learning and path planning. What is theoretically interesting about Toto’s map-learning and path-planning system is that navigation-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 informationally 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. Crucially, 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 direction) and (ii) thought and action (in the mind-to-bodyto-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 environmental structures) and thereby of horizontally extended modularity.9 In spite of these positive steps towards a reconciliation between an approach that emphasizes cognitive modularity and one that emphasizes cultural transmission and the embodied–extended mind, an important 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 unhelpfully 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 evolutionarypsychological 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 kickstarted 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 development 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 rhythmic 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. important 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 indicated 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 comprehension is kick-started by a system that is primed with dispositions to note salient inputs and their likelihood of occurring together. Functional modularization, in the form of a domain-specific, informationally encapsulated system for language comprehension, develops progressively through interactive environmental 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 contributes. 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 organismic and/or wider environmental factors made explanatorily non-negligible contributions to phenotypic form. That is the general picture on offer from approaches that emphasize cultural evolution, cognitive 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 antispecification 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 tempting, 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 developmental 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 developmental 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 likebegets-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 inheritance 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 nongenetic 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 liberality, we would potentially have access to such an account. Allied with the concept of emergent modularity, 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 environment, 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 environments. 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 development-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 selftransformation, artefact-based expansion and a snowballing/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 modularity resulting from trajectories marked by multiple bouts of culturally scaffolded experience and the selfselection 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 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 communication) 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 domaingeneral mechanisms in isolation, i.e. without assistance from domainspecific mechanisms, would not be able to solve the adaptive problems confronted by the brain, or at least that any domaingeneral 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 Wheeler (2006). 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 downstream 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 inconsistency 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.14680017.2004.00252.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 Neuroconstructivism: 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 neurobiologically 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) Noe¨, 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 Neuroconstructivism: 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 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 reasonable 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 represented (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), *[email protected] One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

olfactory receptors (OR4C13 and OR2B6 ), synapseassociated proteins (RAPSN ) and a number of brainexpressed 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 adaptations, which were regarded as products of resistantto-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

Exploring gene–culture interactions

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 wellresearched 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)

t

gene pool

t +1

gene pool

development culture modified selection cultural inheritance

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 environments with different climatic regimes, devising agricultural 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 construction has been a co-director of recent human evolution is essentially the conclusion reached by the geneticists analysing the human genome:

genetic inheritance

K. N. Laland

time

3578

development culture 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 hypertension (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 inheritance, 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 information, 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 conforming 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 circumstances 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 conventional 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 preference, 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 experimental 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 frequencydependent 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 overall 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 frequencies 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) constructed a gene–culture coevolutionary model of handedness 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 handedness). While this assumption would be contested by some researchers (Annett 1985), simulations reveal that this assumption, made for mathematical convenience, 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 handedness 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 righthandedness 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, contentious (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 irrespective 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 righthanded standard. While simulations reveal that selective regimes such as heterozygote advantage and frequencydependent selection could preserve genetic variation, as mentioned above, such explanations are inconsistent with the observation that handedness has low heritability. 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 right!left left!left

1/2CrCa 1/2CrCb 1/2CrKa

1/2Ch1rCa 1/2Ch1rCb 1/2Ch1rKa

1/2Ca 1/2Cb 1/2Ka

explanation: human populations may have reached the equilibrium specified by (2.1), such that no genetic variation underlies variation in handedness, but lefthanders 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 maximumlikelihood 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 maximumlikelihood 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 righthanded 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 proportion, 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 monozygotic 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 interaction. 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 handednessdistorting 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 increasingly 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 Pleistocene 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 righthanders 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 transmitted 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 invertebrates) 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

Exploring gene–culture interactions

instance, mate-choice copying has been observed in 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 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 of cultural variants in a population over time. Laland 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 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 disequilibrium 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 transmission bias, large a, small or negative s), traits may be Phil. Trans. R. Soc. B (2008)

(a) 1

genetic trait (T2)

K. N. Laland

0 (b) 1

trait (T2)

3582

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 conventional 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 compensates 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.

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 transmission 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 preferences. 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 behavioural 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; OdlingSmee 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 perspective 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 organisms inherit from their ancestors. Organisms transmit to their offspring, and subsequent descendents physically altered selective environments, both through actions on their biological and non-biological environments 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 environments that can be altered by niche construction (Laland et al. 1996, 1999, 2001; Odling-Smee et al. 2003). These theoretical analyses suggest that this ‘selfimposed’ 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 evolutionary 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 nicheconstruction traits can drive themselves to fixation by simultaneously generating selection that favours ‘recipient’ 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 construction 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 (CavalliSforza & 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 theoretical 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 construction. 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) representing 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 component 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) W11Za1h1C3R pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 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 nicheconstruction models with unbiased cultural transmission 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 construction 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 nicheconstructing behaviour can also generate interesting evolutionary scenarios. For instance, if transmission bias results in a change in frequency of cultural nicheconstructing 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 consequence 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 semiisolated 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 accelerates 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 constructors. 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 selfconstructed 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 selfimposed 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 construction, but cultural transmission biases favouring particular 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 construction 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 recipients 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 mechanisms, 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, stronger, 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 transmission 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 environment is uniquely powerful, largely due to our capacity for culture. Theoretical analyses regularly reveal coevolutionary 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 environment 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 consequences 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 environment (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 longstanding, 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 coevolutionary 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:10.1016/00405809(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. Princeton, 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 Mathematical 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. & CavalliSforza, 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 preferences 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 Geneculture coevolution and sex ratios: the effects of infanticide, 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 Geneculture 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 longterm 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 frequencydependent 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 reconsidered: 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 populations. 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 domestication. 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 righthandedness 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 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 dimensions, 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 & Szathma´ry (1995) described language as the most recent major evolutionary 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 ([email protected]). One contribution of 11 to a Theme Issue ‘Cultural transmission and the evolution of human behaviour’.

human capacity for language. Specifically, as we will argue in §4, cultural evolution potentially shields the human language faculty from selection, ruling out the evolution of a strongly constraining and domainspecific 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 transmitted 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 underpinned 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 Arbitrariness allows those basic units to be mapped onto the world in a flexible fashion, without additional constraints 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 alarmcalling 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 nonhumans—indeed, Hauser et al. (2002) argue that it is entirely absent from the cognitive repertoires of all nonhuman 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 expressivity. 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 combination 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 consideration 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 compositionality. 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 transmitted 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 conundrum 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 learnability, 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 compositional) 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 compositional 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 biological 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 compositionality 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 predispositions 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 acquisition. A learner’s confidence that a particular grammar h accounts for the linguistic data d that they have encountered is given by PðhjdÞ Z P

PðdjhÞPðhÞ 0 0 ; h 0 Pðdjh ÞPðh Þ

ð4:1Þ

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 determining 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 transmission 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 ). Furthermore, they allow us to explore, using a single framework, 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 parametrizable 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 PðhÞ Z

k Y GðkaÞ Gðnj C aÞ; GðaÞk Gðm C kaÞ jZ1

ð4:2Þ

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 consistency: 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 & Vita´nyi 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 bottleneck on language transmission) being produced by an individual with grammar h is PðdjhÞ Z

Y hxyi2d

Pð yjx; hÞ

1 ; m

ð4:3Þ

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 proportional 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

K. Smith & S. Kirby

Human language faculty and its evolution

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.) Q, sampler

P(h)

Q, maximizer

h

u

bw

bs

u

bw

bs

u

bw

bs

aaa aab abc

0.0370 0.0370 0.0370

0.0389 0.0370 0.0361

0.1 0.0333 0.0167

0.0370 0.0370 0.0370

0.0389 0.0370 0.0361

0.1 0.0333 0.0167

0.0370 0.0370 0.0370

0.2499 0.0135 0.0014

0.2499 0.0135 0.0014

0.25

ð4:5Þ

where H is the set of hypotheses for which P(hjd ) is at a maximum. A model of cultural transmission follows straightforwardly 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 Qi . 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 Qh ; ð4:7Þ CZ h

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

communicative accuracy, C

grammars with the highest posterior probability: ( 1=jHj if h 2 H ; PL ðhjd Þ Z 0 otherwise

0.20 0.15 0.10 0.05 0

1

10

20 a

30

40

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 contrast, cultural evolution in sampling populations

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

Human language faculty and its evolution

(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 the evolvability of those features. A more satisfying solution is to evaluate the evolutionary stability of hypothesis selection strategies and priors. In order to do this, we require a slightly more detailed means of evaluating how communication influences reproductive success. We make the following initial assumptions (see below for a slightly different set of assumptions): (i) a population consists of several subpopulations, (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 Griffiths et al. 2008, §5a), and (iii) natural selection favours learners who arrive at the same grammar6 as their peers in a particular subpopulation, where peers are other learners exposed to the language of the subpopulation. Given these assumptions, the communicative accuracy between two individuals A and B is given by XX B caðA; BÞ Z QA ð4:8Þ hh 0 $Qhh 0 $Qh 0 ; h

(a) 30 minority a

returns the unmodified prior distribution. Amplifying the differences between languages provided by the prior increases the likelihood that two individuals from a population will share the same language, and therefore yields higher C. This analysis therefore suggests that evolution should favour MAP learning over sampling, leading to selective neutrality with respect to the strength of prior biases: strength of prior bias is shielded in MAP populations.

20

10

(b) 30 minority a

K. Smith & S. Kirby

20

10

(c) 30 minority a

3598

20

10

h0

where the superscripts on Q indicate that learners 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 individuals become common, at which point type I individuals are positively selected for and the dominance of behaviour I is re-established. Phil. Trans. R. Soc. B (2008)

10

20 30 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

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.

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 majority (a) minority

(b) minority

u

bs

bw

u bs bw

— 0.98 0.99998

0.81 — 0.99

0.9997 0.82 —

u bs bw

— 1 1

0.45 — 1

0.45 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 individuals with weaker priors. As shown in table 2, this tendency for weaker majority priors to reduce the extent to which strengthened 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 populations, 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)

strong weak flat

minority

sampling

maximizing

sampling maximizing sampling maximizing sampling maximizing

— 1.39 — 1.14 — 1.12

0.6 — 0.38 — 0.88 —

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 datagenerating 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 homogeneous 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 rewarding 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.)

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

majority

minority

u bs bw

u

bs

bw

— 1 1

0.79 — 0.79

0.9997 1.008 —

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 populations 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 subpopulation’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)

minority

sampling maximizing

sampling

maximizing

— 1

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 ; ð4:9Þ ca 0 ðA; BÞ Z 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 nonlinguistic 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 outlined 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; Go´mez 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 communicating 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 inventories 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 productive 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 constraining, 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 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 communication. 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 communication 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. & Vita´nyi, 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/S13646613(00)01494-7) Fitch, W. T. & Hauser, M. D. 2004 Computational constraints 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 songbirds. 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) Go´mez, R. L. 2002 Variability and detection of invariant structure. Psychol. Sci. 13, 431–436. (doi:10.1111/14679280.00476) 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 encyclopaedia 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. & Szathma´ry, 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 nonadjacent 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 chimpanzees (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 compositional 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 Campbell’s monkeys, Cercopithecus campbelli. Behav. Ecol. Sociobiol. 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

volume 363

. number 1509 . pages 3467–3603

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 Introduction. Cultural transmission and the evolution of human behaviour K. Smith, M. L. Kalish, T. L. Griffiths & S. Lewandowsky

3469

Review. Establishing an experimental science of culture: animal social diffusion experiments A. Whiten & A. Mesoudi

3477

Review. The multiple roles of cultural transmission experiments in understanding human cultural evolution A. Mesoudi & A. Whiten Review. Theoretical and empirical evidence for the impact of inductive biases on cultural evolution 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 R. McElreath, A. V. Bell, C. Efferson, M. Lubell, P. J. Richerson & T. Waring

3503

3515

Investigating children as cultural magnets: do young children transmit redundant information along diffusion chains? E. Flynn

3541

The fitness and functionality of culturally evolved communication systems N. Fay, S. Garrod & L. Roberts

3553

Culture, embodiment and genes: unravelling the triple helix M. Wheeler & A. Clark

3563

Exploring gene-culture interactions: insights from handedness, sexual selection and niche-construction case studies K. N. Laland Cultural evolution: implications for understanding the human language faculty and its evolution K. Smith & S. Kirby

Founded in 1660, the Royal Society is the independent scientific academy of the UK, dedicated to promoting excellence in science Registered Charity No 207043

volume 363

number 1509

pages 3467–3603

In this issue

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

3529

3577 3591

journals.royalsociety.org Published in Great Britain by the Royal Society, 6–9 Carlton House Terrace, London SW1Y 5AG

Cultural transmission and the evolution of human behaviour

Review. Studying cumulative cultural evolution in the laboratory C. A. Caldwell & A. E. Millen

3489

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

12 November 2008

ISSN 0962-8436

The world’s longest running international science journal

12 November 2008

Neuroeconomics (Philosophical Transactions of the Royal Society series B)

The Neurobiology of Social Recognition, Attraction and Bonding (Philosophical Transactions of the Royal Society series B)

Mechanisms and Functions of Brain and Behavioural Asymmetries (Philosophical Transactions of the Royal Society series B)

Predictions in the brain: using our past to prepare for the future (Philosophical Transactions of the Royal Society series B)

The Nature of Mathematical Proof (Philosophical Transactions of the Royal Society A)

Scientific Discourse in Sociohistorical Context: The Philosophical Transactions of the Royal Society of London, 1675-1975 (Rhetoric, Knowledge, and Society)

Cultural Transmission and the Evolution of Human Behaviour (Philosophical Transactions of the Royal Society series B)

Neuroeconomics (Philosophical Transactions of the Royal Society series B)

Evolution and Human Behavioural Diversity (Philosophical Transactions of the Royal Society series B)

Evolution of Society (Philosophical Transactions of the Royal Society series B)

Cooperation and Deception: From Evolution to Mechanisms (Philosophical Transactions of the Royal Society series B)

Cultural and Linguistic Diversity: Evolutionary Approaches (Philosophical Transactions of the Royal Society series B)

The Neurobiology of Social Recognition, Attraction and Bonding (Philosophical Transactions of the Royal Society series B)

Mechanisms and Functions of Brain and Behavioural Asymmetries (Philosophical Transactions of the Royal Society series B)

Rationality and Emotions (Philosophical Transactions of the Royal Society series B)

Group decision making in humans and animals (Philosophical Transactions of the Royal Society series B)

Culture Evolves (Philosophical Transactions of the Royal Society series B) issue 1567

Transactions of the American Philosophical Society

The Sapient Mind: Archaeology Meets Neuroscience (Philosophical Transactions of the Royal Society B)

Social intelligence: from brain to culture (Philosophical Transactions of the Royal Society B)

Predictions in the brain: using our past to prepare for the future (Philosophical Transactions of the Royal Society series B)

The Nature of Mathematical Proof (Philosophical Transactions of the Royal Society A)

Scientific Discourse in Sociohistorical Context: The Philosophical Transactions of the Royal Society of London, 1675-1975 (Rhetoric, Knowledge, and Society)

Babylonian Horoscopes (Transactions of the American Philosophical Society)

Isaiah Berlin's Counter-Enlightenment (Transactions of the American Philosophical Society)

Evolution of the Human Diet: The Known, the Unknown, and the Unknowable (Human Evolution Series)

The Evolution of Human Language: Scenarios, Principles, and Cultural Dynamics

The Evolution of Human Sexuality

The Philosophy of Human Evolution

Handbook of Evolution: The Evolution of Human Societies and Cultures

Handbook of Evolution: The Evolution of Human Societies and Cultures

Handbook of Evolution: The Evolution of Human Societies and Cultures

Communication and the Evolution of Society

Communication and the Evolution of Society

The New Economics of Human Behaviour

The New Economics of Human Behaviour

Prehistoric Settlement of the Pacific (Transactions of the American Philosophical Society)

Cultural Transmission and the Evolution of Human Behaviour (Philosophical Transactions of the Royal Society series B)

Neuroeconomics (Philosophical Transactions of the Royal Society series B)

Evolution and Human Behavioural Diversity (Philosophical Transactions of the Royal Society series B)

Evolution of Society (Philosophical Transactions of the Royal Society series B)

Cooperation and Deception: From Evolution to Mechanisms (Philosophical Transactions of the Royal Society series B)

Cultural and Linguistic Diversity: Evolutionary Approaches (Philosophical Transactions of the Royal Society series B)

The Neurobiology of Social Recognition, Attraction and Bonding (Philosophical Transactions of the Royal Society series B)

Mechanisms and Functions of Brain and Behavioural Asymmetries (Philosophical Transactions of the Royal Society series B)

Rationality and Emotions (Philosophical Transactions of the Royal Society series B)

Group decision making in humans and animals (Philosophical Transactions of the Royal Society series B)

Culture Evolves (Philosophical Transactions of the Royal Society series B) issue 1567

Transactions of the American Philosophical Society

The Sapient Mind: Archaeology Meets Neuroscience (Philosophical Transactions of the Royal Society B)

Social intelligence: from brain to culture (Philosophical Transactions of the Royal Society B)

Predictions in the brain: using our past to prepare for the future (Philosophical Transactions of the Royal Society series B)

The Nature of Mathematical Proof (Philosophical Transactions of the Royal Society A)

Scientific Discourse in Sociohistorical Context: The Philosophical Transactions of the Royal Society of London, 1675-1975 (Rhetoric, Knowledge, and Society)

Babylonian Horoscopes (Transactions of the American Philosophical Society)

Isaiah Berlin's Counter-Enlightenment (Transactions of the American Philosophical Society)

Evolution of the Human Diet: The Known, the Unknown, and the Unknowable (Human Evolution Series)

The Evolution of Human Language: Scenarios, Principles, and Cultural Dynamics

The Evolution of Human Sexuality

The Philosophy of Human Evolution

Handbook of Evolution: The Evolution of Human Societies and Cultures

Handbook of Evolution: The Evolution of Human Societies and Cultures

Handbook of Evolution: The Evolution of Human Societies and Cultures

Communication and the Evolution of Society

Communication and the Evolution of Society

The New Economics of Human Behaviour

The New Economics of Human Behaviour

Prehistoric Settlement of the Pacific (Transactions of the American Philosophical Society)

Recommend Documents