Salikoko S Mufwene - Christophe Coupé - François Pellegrino - Complexity in Language - Developmental and Evolutionary Perspectives-Cambridge University Press (2017)

Complexity in Language
Developmental and Evolutionary Perspectives
The question of complexity, as in what makes one language more complex

than another, is a long-established topic of debate among linguists. Recently,
this issue has been complemented with the view that languages are complex
adaptive systems, in which emergence and self-organization play major roles.
However, few students of the phenomenon have gone beyond the basic assess-
ment of the number of units and rules in a language (what has been char-
acterized as bit complexity) or shown some familiarity with the science of
complexity. This book reveals how much can be learned by overcoming these
limitations, especially by adopting developmental and evolutionary perspec-
tives.
The contributors include specialists of language acquisition, evolution and

ecology, grammaticization, phonology, and modelling, all of whom approach
languages as dynamical, emergent, and adaptive complex systems.
salikoko s. mufwene is Professor of Linguistics and a member of the

Committee on Evolutionary Biology and the Commitee on the Conceptual
and Historical Studies of Science at the University of Chicago. He is the
author of The Ecology of Language Evolution (Cambridge University Press,
2001), Créoles, écologie sociale, évolution linguistique (2005), and Language
Evolution: Contact, Competition and Change (2008). He has edited several
books, including Iberian Imperialism and Language Evolution in Latin Amer-
ica (2014). He is the founding editor of Cambridge Approaches to Language
Contact. His research includes the emergence of creoles, the phylogenetic
emergence of language, and globalization and language vitality.
christophe coupé is a researcher in cognitive science at the Centre
National de la Recherche Scientifique (CNRS) and the University of Lyon,
France. With a background in computer science, cognitive science, and psy-
chology, he has been involved in several multidisciplinary programs focusing
on language complexity, language origins, and language change. His contri-
butions have mostly consisted in the design and analyses of databases, and
in statistical or computational models of linguistic evolution and diversity. In
2003, he received the Prize of the Young Researcher of the city of Lyon for his
Ph.D. dissertation on the origins of language. In addition to other papers on
phonological complexity, information rate and functional load, he has co-
edited the collective volume Approaches to Phonological Complexity (2009).
françois pellegrino is a senior researcher in linguistics and cognitive
science at the Centre National de la Recherche Scientifique (CNRS) and the
University of Lyon, France. He has coordinated several projects on language
complexity and has been the coordinator of the “Laboratory of Excellence,”
Advanced Studies on Language Complexity (ASLAN) since 2011. For more
than ten years, his research has focused on the structure and dynamics of
phonological systems in the light of the science of complexity and of Shan-
non’s information theory. He has co-edited the collective volume Approaches
to Phonological Complexity (2009) and, over the last fifteen years, he has
authored or co-authored about 80 journal articles, book chapters or confer-
ence papers.
Cambridge Approaches to Language Contact
Founding Editor
Salikoko S. Mufwene, University of Chicago
Co-Editor
Ana Deumert, University of Cape Town
Editorial Board
Robert Chaudenson, Université d’Aix-en-Provence
Raj Mesthrie, University of Cape Town
Lesley Milroy, University of Michigan
Shana Poplack, University of Ottawa
Michael Silverstein, University of Chicago
Cambridge Approaches to Language Contact is an interdisciplinary series bring-

ing together work on language contact from a diverse range of research areas. The
series focuses on key topics in the study of contact between languages or dialects,
including the development of pidgins and creoles, language evolution and change,
World Englishes, code-switching and code-mixing, bilingualism and second lan-
guage acquisition, borrowing, interference and convergence phenomena.
Published Titles
Salikoko S. Mufwene, The Ecology of Language Evolution
Michael Clyne, Dynamics of Language Contact
Bernd Heine and Tania Kuteva, Language Contact and Grammatical Change
Edgar W. Schneider, Postcolonial English
Virginia Yip and Stephen Matthews, The Bilingual Child
Bernd Heine and Derek Nurse (eds.), A Linguistic Geography of Africa
J. Clancy Clements, The Linguistic Legacy of Spanish and Portuguese
Umberto Ansaldo, Contact Languages
Jan Blommaert, The Sociolinguistics of Globalization
Carmen Silva-Corvalán, Bilingual Language Acquisition
Lotfi Sayahi, Diglossia and Language Contact
Emanuel J. Drechsel, Language Contact in the Early Colonial Pacific
Enoch Oladé Aboh, The Emergence of Hybrid Grammars
Zhiming Bao, The Making of Vernacular Singapore English
Ralph Ludwig, Peter Mühlhäusler, and Steve Pagel (eds.), Linguistic Ecology and
Language Contact
Braj B. Kachru, World Englishes and Culture Wars
Bridget Drinka, Language Contact in Europe
Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino (eds.), Complexity
in Language: Developmental and Evolutionary Perspectives
Further Titles Planned for the Series
Rakesh Bhatt, Language Contact and Diaspora
Gregory D. S. Anderson Language Extinction
Kingsley Bolton, Samuli Kaislaniemi, and Anna Winterbottom (eds.), Language
Contact and the East India Company
Sarah Roberts, The Birth of a Language
Ellen Hurst and Rajend Mesthrie (eds.), Youth Language Varieties in Africa
Cecile Vigouroux, Migration, Economy, and Language Practice
Complexity in Language
Developmental and Evolutionary Perspectives
Salikoko S. Mufwene
University of Chicago
Christophe Coupé
University of Lyon
François Pellegrino
University of Lyon
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
4843/24, 2nd Floor, Ansari Road, Daryaganj, Delhi - 110002, India
79 Anson Road, #06-04/06, Singapore 079906
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107054370
DOI: 10.1017/9781107294264

C Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino 2017
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2017
A catalog record for this publication is available from the British Library
Library of Congress Cataloging-in-Publication Data
Names: Mufwene, Salikoko S., editor. | Pellegrino, Franpcois, 1971– editor. |
Coupbe, Christophe, 1977– editor.
Title: Complexity in language : developmental and evolutionary perspectives /
[edited by] Salikoko S. Mufwene, Franpcois Pellegrino, Christophe Coupbe.
Description: Cambridge ; New York : Cambridge University Press, [2016] |
Series: Cambridge Approaches to Language Contact
Identifiers: LCCN 2016041226 | ISBN 9781107054370
Subjects: LCSH: Complexity (Linguistics) | Discourse analysis. |
Linguistic analysis (Linguistics)
Classification: LCC P128.C664 C65 2016 | DDC 410–dc23 LC record available at
https://lccn.loc.gov/2016041226
ISBN 978-1-107-05437-0 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents
Figures page vii

Tables ix
Contributors xi
Acknowledgments xii
1 Complexity in Language: A Multifaceted Phenomenon 1

salikoko s. mufwene, christophe coupé,
and françois pellegrino
2 How to Explain the Origins of Complexity in Language:
A Case Study for Agreement Systems 30
luc steels and katrien beuls
3 Complexity in Speech: Teasing Apart Culture and Cognition 48
bart de boer
4 A Complex-Adaptive-Systems Approach to the Evolution of
Language and the Brain 67
p. thomas schoenemann
5 Evolutionary Complexity of Social Cognition,
Semasiographic Systems, and Language 101
william croft
6 To What Extent Are Phonological Inventories
Complex Systems? 135
christophe coupé, egidio marsico,
and françois pellegrino
7 A Complexity View of Ontogeny as a Window on Phylogeny 165
barbara l. davis
v
vi Contents
8 Language Choice in a Multilingual Society: A View from

Complexity Science 187
lucía loureiro-porto and maxi san miguel
9 Complexity and Language Contact: A Socio-Cognitive
Framework 218
albert bastardas-boada
Index 245
Figures
2.1 Results of an agent-based simulation page 39

2.2 Comparison of a population that uses lateral inhibition and one
that does not 40
2.3 Experiment comparing the learning efficiency Sg between a
strategy that reuses existing words as meaningful markers and one
that does not 42
2.4 (a) Overall reduction of form complexity in a population of
agents over 50,000 games
(b) The -naeamo marker (expressing feature v-2–1) 43
2.5 Reduction in inventory complexity 45
3.1 The different time scales in language and their interactions 54
3.2 French vowels and Hindi plosives illustrating feature economy 56
3.3 Schematic illustration of vertical transmission (top) and
horizontal transmission (bottom) 57
3.4 Three processes through which experimental participants create
new signals 60
4.1 Increase in ratio of possible combinations of areas for human
vs. ape 76
4.2 Comparison of semantic network density for ape vs. human
corpora 86
5.1 The logical relationship among types of common ground,
its basis, and coordination devices 127
6.1 Distributions of segmental associations according to the value
of the binomial test 150
6.2 Graphical representation of the relations between vowels 152
6.3 Distributions of feature associations according to the value of
the binomial test 157
7.1 Consonant place assimilation patterns 180
7.2 Consonant manner assimilation patterns 180
7.3 Labial, coronal, and dorsal assimilation pattern changes
over time 181
8.1 One-dimensional regular lattice of degree k = 2 199
vii
viii Figures
8.2 Square regular lattice of degree k = 4 199

8.3 Simulation tool designed to visualize the effects of bilingualism,
prestige and volatility 200
8.4 Bilingual speakers on the borders of monolingual domains
(regular lattice) 201
8.5 Islands of monolingual speakers linked by a small number
of bilinguals 202
8.6 Abrams-Strogatz Model vs. Bilinguals Model, when a = 0.1
(high volatility) 204
8.7 Snapshots of the Abrams-Strogatz (top) and the Bilinguals Model
(bottom) in a small-world network (bilinguals = white) 207
8.8 Abrams-Strogatz Model (left) and Bilinguals Model (right) in a
network with community structure (t = time) 208
9.1 Main principles of the complexity perspectives in contrast with
the more traditional scientific ones 223
Tables
3.1 Languages from UPSID451 (Maddieson, 1984; Maddieson &

Precoda, 1990) with inventory sizes smaller than or equal to 16
phonemes. Although there appears to be an important influence
of area or language family, nevertheless, languages with small
phoneme inventories are spoken in different, linguistically
unrelated regions of the world page 49
4.1 Nouns used in sentences Kanzi responded to correctly 81
4.2 Verbs used in sentences Kanzi responded to correctly 82
4.3 Proper names used in sentences Kanzi responded to correctly 82
4.4 Pronouns used in sentences Kanzi responded to correctly 83
4.5 Adjectives, adverbs and prepositions used in sentences Kanzi
responded to correctly 83
4.6 Comparative semantic web density for select words in
human vs. chimp 87
5.1 Evolutionary complexity based on the evolution of
semasiographic systems 125
6.1 Observed frequencies of voiced labial fricatives and voiced
coronal fricatives across languages in UPSID 142
6.2 Observed frequencies of /a/ and /ã/ across languages in UPSID 143
6.3 Categorization of the different pair associations of components
of phonological inventories according to our statistical approach 146
6.4 Comparison of four different approaches to multiple testing for
vocalic features 148
6.5 Comparison of four different approaches to the statistical issue
of multiple testing for consonants 149
6.6 Estimation of pair interactions at the segmental level 149
6.7 The 20 vowel-vowel associations with the smallest p-values 150
6.8 The 20 consonant-consonant associations with the smallest
p-values 151
6.9 Vowel-consonant significant associations in the inventories 154
6.10 Distribution of the segments involved in the significant
associations between a vowel and a consonant 154
ix
x Tables
6.11 Estimation of pair interactions at the feature level 155

6.12 The 20 associations of vocalic features with the smallest p-values 156
6.13 The 20 associations of consonantal features with the smallest
p-values 157
6.14 Pairs of consonantal features characterized by mutual attraction
and represented in the 20 most significant associations 158
7.1 Comparison of consonant vowel association levels within
syllables 176
7.2 Intersyllabic infant-language matching results for intrasyllabic
properties 177
8.1 Typical features of complex systems 189
Contributors
katrien beuls, AI Laboratory, Vrije Universiteit Brussel

albert bastardas boada, Universitat de Barcelona
bart de boer, Vrije Universiteit Brussel
christophe coupé, Centre National de la Recherche Scientifique and
University of Lyon
william croft, University of New Mexico
barbara davis, University of Texas at Austin
lucía loureiro-porto, University of the Balearic Islands
maxi san miguel, University of the Balearic Islands and IFISC (CSIC-UIB)
salikoko s. mufwene, University of Chicago
françois pellegrino, Centre National de la Recherche Scientifique and
University of Lyon
p. thomas schoenemann, Indiana University
luc steels, ICREA/Institu de Biologia Evolutiva (UPF-CSIC), Barcelona
xi
Acknowledgments
This book is largely the fruition of the 10-month fellowship that the lead editor
Salikoko S. Mufwene enjoyed at the Collegium de Lyon from October 2010
to July 2011. It enabled him to obtain funding from the same institution to co-
host, in collaboration with the other two editors and with Jean-Marie Hombert
and Egidio Marsico, all at University of Lyon and CNRS, a workshop with the
same title as the present book. We express our hearty thanks to the then Col-
legium’s director Alain Peyraube for his interest in the subject matter, for his
encouragements and advice during the preparations, and for ultimately mak-
ing it possible for this book to materialize. We are also deeply indebted to the
Collegium’s then administrative coordinator Marie-Jeanne Barrier for single-
handedly managing the logistics that enabled the Workshop from the time the
funding was allocated, through actually coordinating the details of the meet-
ing (for instance, making sure that we had all the equipment and the meals on
time), to the bookkeeping details she graciously attended to after the event. We
likewise thank the Ecole Normale Supérieure de Lyon for lending us facilities
for the Workshop.
We are equally grateful to other participants who underscored the signifi-
cance of the Workshop but did not submit their papers for publication: William
S.-Y. Wang (Joint Research Centre for Language and Human Complexity, Chi-
nese University of Hong Kong), Vittorio Loreto (Sapienza University of Rome)
and Francesca Tria (Institute for Scientific Interchange, Turin), Jean-Marie
Hombert (CNRS and Université de Lyon), Fermin Moscoso del Prado Martín
(then a CNRS fellow at Université de Lyon), and Ramon Ferrer-i-Cancho (Uni-
versitat Politècnica de Catalunya, Spain). Our exchanges would have certainly
been less productive without their engaged contributions.
Additionally, Christophe Coupé and François Pellegrino are grateful to the
LABEX ASLAN (ANR-10-LABX-0081, French program “Investissements
d’Avenir” ANR-11-IDEX-0007) of the University of Lyon and to the Labo-
ratoire Dynamique du Langage (UMR5596) for their support.
Lucía Lureiro Porto and Maxi San Miguel thank Vincent Blondel and IOP
Publishing for permission to reproduce Figure 8.5.
xii
1 Complexity in Language: A Multifaceted
Phenomenon
Salikoko S. Mufwene, Christophe Coupé,

and François Pellegrino
1 Complexity in Linguistics
1.1 Linguistics and the Science of Complexity

Complexity has attracted a great deal of attention in linguistics since 2001, at
a rate that proportionally far exceeds its invocations in the field since Ferdi-
nand de Saussure, the father of our discipline, in the early twentieth century.
The number of books bearing complexity in their title is remarkable, suggest-
ing that there may be an emergent research area whose focus is complexity
in Language. The dominant question that the relevant linguists have addressed
is the following: To what extent does complexity as observed in different lan-
guages or in different modules of the language architecture display both cross-
systemic variation and universal principles? This has entailed asking whether
there are languages that are more complex than others and explaining the nature
of differences.
One is struck by the sheer number of book-length publications alone,1 and
even more when the numerous journal articles and chapters in edited volumes
are added to the total count, regardless of whether or not they include com-
plex(ity) in their title. On the other hand, one is also shocked by the scarcity of
works that explain what complexity is, apparently because it is assumed to be
known.2 This is quite at variance with publications outside linguistics, which
are devoted to explaining various ways in which the notion can be interpreted.
1 Book titles containing the term language or linguistic(s) include: Dahl (2004), Hawkins (2005),
Risager (2006), Larsen-Freeman (2008), Miestamo et al. (eds., 2008), Sampson et al. (eds.,
2009), Givón (2009), Givón & Shibatani, (eds., 2009), Pellegrino et al. (eds., 2009), Aboh &
Smith (eds., 2009), Faraclas & Klein (eds. 2009), Cyran (2010), Trudgill (2011), Robinson
(2011), McWhorter (2012), Kortmann & Szmrecsanyi (eds., 2012), Housen & Kuiken (2012),
Blommaert (2013), Culicover (2013), Massip-Bonet & Bastardas-Boada (eds., 2012), Newmeyer
& Preston (eds., 2014), Berlage (2014), Kretzchmar (2015), and Baerman, Brown & Corbett
(2015).
2 A noteworthy exception is Ellis and Larsen-Freeman’s (2009) “Language as a complex adaptive
system,” which is derived from the lead and seminal chapter by Beckner et al. (also identified as
“The Five Graces Group”), “Language is a complex adaptive system: A position paper.”
1
2 Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino
The subject matter has actually also evolved into what is identified by some as
“complexity theory” or the “science of complexity” (see footnote citations).
Thus, a convenient starting point for this chapter and this book is to explain
what is meant by complexity as it applies both to linguistics and other research
areas. Etymologically, the term complexity, as a nominalization from complex,
can ultimately be traced to Latin complexus, a past participle of the deponent
verb complecti ‘embrace, comprise,’ according to Webster’s Collegiate Dic-
tionary, and also confirmed by the French Petit Robert, which translates it as
contenir ‘contain.’ According to the Online Etymology Dictionary, the adjec-
tive complex ‘composed of parts’ was borrowed from French complexe ‘com-
plicated, complex, intricate’ (seventeenth century), from Latin complexus ‘sur-
rounding, encompassing,’ past participle of complecti ‘to encircle, embrace.’
In transferred use, the verb meant ‘to hold fast, master, comprehend’, from
com- ‘with’ and plectere ‘to weave, braid, twine, entwine.’ The noun complex
evolved to mean ‘a whole comprised of parts.’
This etymological definition remains very generic. Beyond it, it appears that
no strong consensus has emerged in the science of complexity itself about what
complexity means (see, e.g., Strogatz 2003; Gershenson, ed. 2008; Mitchell
2009). There are nonetheless some common themes and properties that recur
in the relevant literature. They include the following, which overlap in some
ways:
(1) Complexity arises from the coexistence of components that interact with
each other, not necessarily from the fact that a space or a system is popu-
lated with several components or members; it is therefore interactional.
(2) Complexity arises from the dynamics of activity coordination or synchro-
nization that integrate individuals as members of a population (e.g., ant
colonies, bird flocks, and fish schools); thus, it is dynamical.
(3) Complexity emerges from nonlinear evolution, which is driven by multiple
factors whose significance may vary at different stages of the evolutionary
process; its effects are not constant, but subject to the changing values of
the relevant variables.
(4) Complexity lies in what brings order out of chaos,3 through what is also
known as “self-organization” and was formerly referred to as an “invisible
hand” (Smith, 1776).4
(5) There is complexity in any system where the properties of the whole do not
amount to the sum of the properties of the components.
3 “Chaos” is used here as in “chaos theory,” which studies systems whose behavior is highly sensi-
tive to initial conditions, in the sense that small differences in initial conditions may produce quite
divergent evolutions or outcomes. It also seeks to capture emergent patterns from the interplay
between order and disorder, from which complexity arises.
4 For a more linguistic take on the “invisible hand,” see Keller (1994).
Complexity in Language: A Multifaceted Phenomenon 3
(6) Finally, complexity is the peculiarity of emergent patterns in a system in

constant state of flux between disorder and transient order (or equilibrium).
In other words, complexity arises from the dynamics of coexistence and
interaction or cooperation of components toward generating the properties
of whole.
As we are reminded by Loureiro-Porto and San Miguel (Chapter 8), com-
plex should not be confused with complicated (pace the etymology of com-
plex(ity) cited earlier). For instance, airplanes are complicated rather than com-
plex pieces of engineering. Despite the very large number of parts, each part has
a clear function that makes it possible – and to some extent easier – to predict
its contribution to the whole. On the contrary, it is not evident which role each
component of a true complex system (such as an ant colony or a flock of birds)
plays in the behavior or function of the whole system, or what it specifically
contributes as a unit to the larger, integrated whole.
Connecting these interpretations to Language, the idea that a system consists
of interacting components is not new. It was indeed at the core of the struc-
turalist program, in which phonemes, words, and other linguistic units were
primarily considered as components of structures. An important peculiarity of
this tradition is that a language was typically construed as an autonomous sys-
tem, independent of its speakers and the wider ecology in which it and the
speakers evolve. Thus, internal forces and their interactions were paid much
more attention than external ones. Although this suits the etymology of the
term complexity and some of the earlier points, it understandably omits other
interpretations made evident in the science of complexity, especially regarding
the dynamical aspects. Complexity arises not just from how the different parts
interact with each other but also from how they respond to external pressures of
the environment, or the external ecology (Mufwene 2001), outside the system.
However, decades later, despite the increasingly interdisciplinary nature of
the relevant scholarship in other research areas, most linguists deal with com-
plexity almost as if hypotheses in those other areas couldn’t possibly apply to
languages. This observation does not include modelers, often coming from the
field of artificial intelligence, who have invoked multi-agent systems or net-
work theory to investigate language emergence and change (e.g., de Boer &
Zuidema (2010), Ke et al. (2008), Kirby (2000), Steels (1998, 2011a). Aside
from them, others such as Massip-Bonet and Bastardas-Boadas, eds. (2012)
or Kretzschmar (2015) have also highlighted the dynamic aspects of language
behavior. Overall, linguists still have to ask themselves what interpretation of
Language they subscribe to, what complexity may mean under that particular
interpretation, whether the architecture of Language and linguistic behavior are
exceptional in relation to the common properties of complexity that have been
observed in other aspects of nature, and how they contribute to this expanding,
more inclusive research endeavor.
1.2 Components, Structures, and Domains in Linguistic Complexity

Likely because most of them have ignored work in the science of complexity,
linguists have remained faithful to the interpretation of a complex system as ‘a
whole consisting of several parts.’ Thus, the more parts the whole consists of,
the more complex it is assumed to be, regardless of how the parts interact with
each other. This explains why an approach commonly found in the literature
consists in counting the number of elements of a linguistic system in order to
evaluate its complexity. Depending on the specific study, the focus may be the
number of phonemes, morphemes, or words, but also relations among variants
of such units (allophones, allomorphs, or near-synonyms), or yet the number of
categories, rules, or constraints that can be posited in a system. This has come
to be called “bit complexity” and has been criticized as uninformative (DeGraff
2001, 2009).
Indeed, this approach does not pay attention to possible relationships
between the components, and goes counter to the well-known idea that “simply
more does not mean more complex.” For example, a set of five bodies moving
randomly – in the absence of any interaction force – is not as complex as, let
alone more than, a set of five bodies, or even three or four, moving according
to gravitational forces they exert on each other (Poincaré 1891).
The possibility that a whole with fewer parts engaged is several multilateral
interactions can generate more interactive complexity cannot be accounted for
in a bit-complexity approach. A good example is when an item generates dif-
ferent interpretations depending on what other item it is combined with. This
is illustrated with the particle up in combinations with various verbs such as
in pick up, give up, show up, and look up. While the item up is basically the
same particle in all these constructions, its contribution to the meaning of each
phrase appears to vary. This variation suggests that the particular dynamics of
each combination produce the meaning of the whole phrase. The overall mean-
ing of each phrase is not the sum of the meanings of its parts (see, for instance,
Victorri 1994 on the dynamics of such constructions).
To be sure, some linguists have shifted from counting elements to assess-
ing how they make a system together. These linguists have first attempted to
identify the patterns of interactions between the components and then infer lin-
guistic complexity from the interactions. Their approach has involved building
mathematical graphs and then quantifying their “structural complexity” with
an appropriate measure. The vertices of the graph correspond to the compo-
nents of the system, while the edges connecting them reflect how they can be
related meaningfully. Another strategy is presented in Coupé et al. (Chapter 6),
in which patterns of co-occurrences of phonetic segments are evaluated with
respect to the individual occurrences of these segments, in order to detect sig-
nificant interactions.
Rather than focusing on (mostly pairwise) interactions between elements,

another systemic view seeks to describe the linguistic system in terms of reg-
ularities and irregularities. A classical concept here is Kolmogorov complex-
ity, which is the length of the shortest program that can produce the descrip-
tion of the system (given a programming language). Behind it is the central
idea that the more compressible a piece of information is, the lower its com-
plexity is as well. This algorithmic approach to complexity (Dahl 2004:42) is
more processual than the previous one, because the algorithm has to be run
in order to get the description. While it has been proved that the Kolmogorov
complexity cannot be computed, reasonable approximations can be obtained
with standard compression algorithms and be applied to compare complexity
between objects. The size of an archive containing the compressed version of
the initial description of the system and the means to decompress it (i.e., a self-
extracting archive) is inversely proportional to the complexity of the system.
Rissanen’s (1978) “minimum description length” is another possible approxi-
mation to Kolmogorov complexity.
Such approaches have been applied especially to measuring the complex-
ity of morphological systems (e.g., Bane 2008, Walther & Sagot 2011). Some
linguists have also echoed Gell-Mann’s (2003) concern that Kolmogorov com-
plexity is highest in the case of totally random expressions, while we intuitively
do not see totally unordered systems as complex. His counterproposal for mea-
suring complexity effectively, named effective complexity, is the length of the
shortest description of the set of regularities of the system. Along these lines,
Newmeyer and Preston (2004:182) also state that “the more patterns a linguistic
entity contains, the longer its description, and then the greater its complexity.”
Although these quantitative approaches offer more refined considerations of
linguistic complexity, they rely on the descriptions that linguists can provide
of a linguistic system. When different options compete in this regard, the quan-
tification methodologies themselves do not help. This echoes Edmonds’ (1999)
statement that complexity lies before all in the eye of the interpreter of the sys-
tem. Another way of considering this is that a descriptive account of complex-
ity can be at odds with a more functional approach: Does the complexity of the
description of an utterance always correspond to the amount of difficulty the
hearer experiences in processing it? Does the description capture adequately
the nature of the neural and psychological encoding, and its consequences in
terms of processing? Is such an approach to complexity informative about the
overall complexity of a language?
Attempts to assess the complexity of a whole language present an additional
difficulty with the same endeavor in other, physical or cultural systems. For
example, ferromagnetic materials, a well-known physical system in which self-
organization of microscopic magnets can occur in the absence of strong external
magnetic field, are only composed of identical elements without hierarchical
structure. By contrast, in the case of languages, different modules and levels

of analysis – phonology, morphology, syntax, semantics, and so on – can be
distinguished, and the hierarchical integration of their elements (starting from
meaningless sound units) into phrases, sentences, and discourse raises a number
of issues. Complexity can be investigated in each module independently but
also between these modules, raising research questions such as whether the
lesser complexity of a module will be balanced by the greater complexity of
another.
Such a conception of equilibrium can also be considered within a domain,
if, for example, one attempts to check whether the greater complexity of the
consonant or vowel system is counterbalanced by the lesser complexity of the
syllabic structures (Maddieson 2011). Technically, comparing distinct domains
such as phonology and morphosyntax is uneasy beyond simply counting ele-
ments, which, as remarked earlier, typically disregards the interactions between
them.
On a more theoretical level, what is obviously missing from the relevant
literature is the interactional complexity that arises from the division of labor
and cooperation between different components, including members of the same
module. However, it is not evident how many modules must ultimately be
posited to account for how the production and interpretation of utterances work
in a language. As a matter of fact, Lieberman (2012) goes as far as rejecting
the idea of modules, arguing that the neurons of the brain are connected in
a way similar to (though more complicated than) the parts of an automobile
engine. However, if one subscribes to the modular architecture of language, it
becomes important to understand how the modules interface with each other
during the production and interpretation of utterances, certainly not in a linear
way (McCawley 1998). According to the latter, the modules work concurrently
rather than sequentially, as is made evident by, for instance, the correction of
false starts while speaking and self-corrections of the interpretations of utter-
ances as the discourse evolves.
Let’s assume that the materials of a language fall in one or another mod-
ule (viz., phonology, morphology, syntax, etc.), each of which makes a clear
contribution to the overall system, while its components (such as individual
sounds in a phonemic system) do not. We may then have to wonder whether
languages do not fall in between complicated and complex systems, consistent
with Loureiro-Porto and San Miguel’s distinction (Chapter 8). The question is
difficult to answer within the bit-complexity approach.
On the other hand, determining whether a language is complex or compli-
cated becomes rather pointless without reference to something that it can be
compared with. This explains why linguistic complexity has typically con-
jured up cross-linguistic research. Any measure of complexity presupposes or
entails some cross-linguistic comparison of phonological inventories, morpho-

logical systems, and so on, as is evident, for instance, in discussions about
whether or not creoles’ grammars are simpler than those of other languages
(e.g., McWhorter 2001, DeGraff 2001). This approach also suggests that a lan-
guage may exist that has no, or very little, complexity built in it. However, from
the point of view of the interaction of modules, McWhorter’s (2001) and Gil’s
(2001, 2009) claims about the simplicity of the grammars of creole vernaculars
and Riau, Indonesia, respectively, beg the question. However, see Gil’s (2009)
reaction discussed below.
1.3 From Static to Dynamic Linguistic Systems

The complexity of a linguistic system can be assessed synchronically, relative
to a given time, regardless of what the system was like before. However, lan-
guages are constantly changing and being adapted to satisfy various communi-
cation pressures, including those that index speakers and the circumstances of
their interactions. Beyond structures that may be assumed to be a static response
to a fragile assemblage of structural constraints – in the spirit of a saying usu-
ally attributed to Ferdinand de Saussure, “la langue est un système où tout se
tient” – linguistic systems are in a constant state of flux, with new components
appearing and older ones evolving or disappearing. It thus makes sense to ask
how complexity evolves under these ecological pressures, and see languages or
their subsystems as complex dynamical systems (Bruckner et al. 2009).
Self-organization and emergence express how order and regularities arise
from an initially chaotic state (as defined in Chaos Theory). They are fun-
damental processes in the study of physical and biological complex systems,
for instance, how ants may build optimal paths between their nest and a food
source, how microscopic dipoles can align to create magnetic domains, and how
traders’ activity at a market can result in macroscopic events such as economic
bubbles or crashes. These concepts can be invoked to account for linguistic phe-
nomena such as the emergence of new language varieties. For example, both
self-organization and emergence can be invoked to explain how elements
from several languages have been selected, in varying proportions, into a new
variety, called creole. Linguistic systems can merge – with some features being
selected and possibly modified, and others rejected – into what appears to be a
new dynamic equilibrium.
Unlike in the science of complexity, linguistics stands out also by the lim-
ited attention that has been given to how complexity emerges, that is, from
a diachronic perspective, relative to language development and to the phylo-
genetic emergence of language. Exceptions include Wang et al. (2004), Givón
(2009), Lee et al. (2009), Mufwene (2012), and some of the authors contributing
to this volume, especially Bart de Boer, P. Thomas Schoenemann, and Luc

Steels, the first and the last based on modeling the emergence of language. It is
to this fold of linguistic complexity that they were invited to contribute.
Different time scales relate to different evolutionary processes: from the
ontogeny of complexity in child language acquisition, to its modifications in
language change and the evolution of languages, to its rise during the phylo-
genetic emergence of Language. In ontogenetic and phylogenetic evolution,
the focus is especially on how a system develops/evolves from architecturally
poorer to richer structures (see, e.g., Dahl 2004, Givón 2009, Givón & Shi-
batani, eds. 2009). Regarding historical language changes, phenomena such
as grammaticalization can be studied with a focus on whether they increase
or decrease the complexity of the system (e.g., Heine & Kuteva 2007). Other
attempts yet derive diachronic models from synchronic constraints, and observe
how system coherence and complexity evolve between lower and upper bounds
(Coupé et al. 2009). From a phylogenetic point of view, one should not dodge
the question of how complexity arose during the transitions from vocalizations
to naming and the rise of phonetic systems, to predication and the emergence
of simple sentences, all the way to modern linguistic systems (e.g., Mufwene
2013).
The linguistics discourse has generally overlooked the dynamics of the lin-
guistic elements in relation to each other, such as what may happen when a new
sound is added to the phonetic inventory of a language; or when a preposition
is used as the syntactic head of the predicate phrase (like in dis buk fuh you ‘this
book [is] for you’ in Gullah), whereas a verb has traditionally been required in
this position in English (Mufwene 1996). That is, while speakers/signers mod-
ify the extant system with their innovations, the latter may trigger other adjust-
ments in the system. This is the case in the sentence You bin fuh come ‘you
had/were expected to come’ in Gullah, where, because it can function as head
of a predicate phrase, the preposition fuh has also been coopted as a marker of
obligation, the counterpart of a modal verb in English. (As head of the predi-
cate phrase it can also be modified by the anterior tense marker bin, regardless
of whether it functions as a preposition or as a modal marker.) Such a change
by cooption of extant materials is undoubtedly true of other cultural systems,
which are also adaptive but depend primarily on the activities that shape them,
those of the practitioners of the culture. Future research should return to this
issue, which arises also from some of the contributions to the present book.
An important question in such systemic adaptations is: What are the forces
or constraints responsible for linguistic change? Answering this question offers
complementary and enriching perspectives on linguistic complexity. Indeed,
the previous approaches can all fit a framework where linguistic structures
are considered in isolation and studied on the basis of their internal (possibly
dynamic) patterns of occurrences or interactions. But considering the various
dimensions – social, cognitive, and pragmatic – of what may be called the ecol-
ogy of language (Mufwene 2001, 2008; Coupé 2016) opens further avenues
toward a more complete understanding of linguistic complexity.
1.4 Complexity and Language Ecology

Language ecology has usually been invoked in relation to social factors
(Mufwene 2001, Lupyan & Dale 2010).5 Indeed, a language does not exist out-
side its social environment; it is a communal creation, with structures shaped
through speakers’ communicative acts. It displays emergent patterns, which
linguists have attempted to capture in the form of rules and constraints, from
a synchronic perspective. However, there are ecological factors that arise from
within the system itself that also influence the evolution of a language (e.g.,
frequency, transparency, regularity, and length of particular variants). They
determine which variants will prevail and which ones will remain minority
alternatives or will be given up.6 Innovations and their replications (or copies)
compete among themselves, subject to these and other ecological factors, social
and otherwise (Mufwene 2001, 2008; Blythe & Croft 2009). This is especially
noticeable in cases of language contact, when a new variety (such as a creole)
emerges and retains only a subset of the variants in the prevailing language
(called lexifier) and only some of the competing substrate features are selected
into the emergent language variety.
Equally, if not more, interesting are cases where the competition7 is not
resolved. For instance, in (standard) English, the primary stress in the word
exquisite may be placed on the first or second syllable; a relative clause may
start with a null complementizer, with the complementizer that, or with a
5 Mufwene actually applies the term ecology to a wider range of factors, both internal and external
to particular languages, some direct and others indirect, that influence the evolution of a lan-
guage, including its vitality. He applies the term to any factor that may be considered as (part
of the) environment relative to a language (variety) or a linguistic feature being discussed. Rel-
ative to language evolution, some ecological factors are economic and historical. Relative to the
phylogenetic emergence of Language, Mufwene (2013) singles out the human anatomy and the
brain/mind as critical ecological factors.
6 Linguists such as Weinreich et al. (1968), McMahon (1994), and Labov (2001) have invoked
actuation (similar to but not exactly the same as actuator in physics) in reference to the particular
combination of factors, which are indeed ecological, that produce particular changes at specific
places and at specific points in time. This tradition of course underscores the need to approach
language evolution from the point of view of complexity and emergence, as these notions are
construed in the science of complexity, in dynamical terms.
7 As explained in Mufwene (2008), “competition” is used here in the same sense as in evolution-
ary biology, applying to variants, organisms, or species that ecology may not sustain equally,
favoring one or some but disadvantaging the other(s). In languages, variants for the same func-
tion (including languages spoken in the same community) are often rated differently by their
speakers or signers, a state of affairs that explains why some disappear. From an evolutionary
perspective, and even that of language ontogenetic development (influenced by who the learner
interacts with), different speakers/signers may not rate the variants in identical ways.
relative pronoun (a wh form); and some speakers say I want you not to come,
whereas others prefer I want you to not come. There are indeed a host of similar
examples not only in English but in other languages too. What also appears evi-
dent in such cases is not only the number of distinctions that are made but also
the ways in which the distributions of the competing expressions are articu-
lated. No assessment of complexity in a language should ignore this interactive
aspect of the system, which is consistent with Saussure’s notion of opposition
between forms or between constructions: the expressions derive their meanings
from how they are opposed to or distinguished from each other.
How is variation managed in a language or speech community? Can it remain
free, in the sense that a speaker/signer may use any variant or another with-
out any communicative or social consequences? Or is it constrained by other
factors that are social, such as age, gender, ethnicity, profession, and level of
education, or those that stem from the precise context of interaction? Are the
constraints rigid or flexible? The interfacing of systems (consisting of struc-
tural units and rules) and social constraints emanating from the communities
using and shaping the languages appear to foster alternative interpretations of
linguistic complexity, which also explains variation in the way that linguists
discuss it, as is evident in, for instance, Sampson et al. (eds., 2009) and Massip-
Bonet and Bastardas-Boada (eds., 2012). To the extent that languages can be
construed as communal systems, complexity arises at least as much from the
dynamics of interaction within the population associated with the language, as
from the actual system hypothesized by the linguist (or any analyst). Linguis-
tic complexity therefore conjures up complexity of linguistic structures and
external constraints exercised by ecological factors, including specific kinds of
social interactions and the particular business or social networks in which one
operates.
No speaker/signer has complete knowledge of their communal language
as an ensemble of idiolects (Mufwene 2001), while they all use it and adapt
their respective idiolects relative to other users. The speakers’/signers’ mutual
accommodations and their respective responses to novel communicative pres-
sures (which are similar to adaptive responses of elements of better understood
complex adaptive systems) drive change or evolution. This peculiarity explains
the claim that languages as both practices and systems are in a constant state of
flux, hardly staying in equilibrium, and are therefore emergent phenomena.
These dynamic aspects of complexity are hardly quantifiable. They also
make it obvious that, as stated by Beckner et al. (2009), the agents of the emer-
gence of complexity are the speakers/signers who manipulate the system. They
are the ones that modify it, innovate new forms and structures, introduce new
dynamics of interaction among the different components of the system on dif-
ferent levels, and therefore modify the patterns of complexity in one way or
another. On the other hand, this agency also sets the discourse on linguistic
complexity at odds with epistemological models of complex systems in the

science of complexity, such as ant colonies and flocks of birds, in which the
agents are parts of the system rather than its users/manipulators.
Speakers and signers largely differ from birds flying in flocks or ants living
in colonies in that they do not reposition themselves physically but modify their
idiolectal characteristics to approximate those of the other speakers or signers
they wish to align with socially or professionally. In other words, a bird situates
itself as part of a system, whereas a speaker or signer realigns his or her linguis-
tic productions and uses these modifications to situate him or herself socially
or professionally but not in the linguistic system itself. Also, while a bird never
changes the innate rules that dictate how, while flying, it adapts its speed and
direction to the speeds and directions of neighboring birds (see, e.g., Hilden-
brandt et al. 2010), humans are much more flexible when it comes to adapting
their behaviors – that is, linguistic strategies – which are not what they are as
individuals.
At the root of this behavioral flexibility obviously lies highly developed cog-
nitive capacities, which allow us to internalize part of the complexity of the
linguistic system we are immersed into. Crucially, they enable us as speak-
ers to anticipate the possible effects of our words on the hearer’s mind, and to
reconstruct as hearers what was in the speaker’s mind when they produced the
message we just received.
Beyond social aspects, cognitive aspects also shed light on linguistic com-
plexity. As much as languages are social, communal constructions, they are also
processed and stored in individual minds. In a way similar to the aforemen-
tioned algorithmic complexity, linguistic complexity can be estimated based
on the cognitive efforts required by the mind (conceived of as the state of the
brain in activity) to produce or process a message. The question of how diffi-
cult specific items or subsystems are to learn especially points to a learnability
complexity.
Along these lines, Dahl (2009) makes a distinction between, on the one hand,
“absolute complexity,” ascribed to the mechanics of the system, and, on the
other, “relative complexity,” based on how much difficulty a learner experi-
ences in learning a language. What does this “relative complexity” tell us about
the inherent complexity to learn a given language? The answer appears to be
negative, because “relative complexity” is predicated on the fact that speak-
ers of particular languages find some other languages harder to learn for vari-
ous reasons, for instance, the tone contrasts are too difficult to replicate faith-
fully, or there are too many morphological distinctions (especially inflections)
to remember accurately, or there are too many constraints regulating when par-
ticular variants can or should not be used.
This complexity must be assessed variably, depending on which speaker of
which language is learning which other language. However, all languages are
effectively learned and spoken by their speakers, regardless of their very differ-
ent strategies when it comes to conveying information. There is no evidence that
it takes speakers of, say, tonal and agglutinating languages longer to develop
native competence in them than speakers of toneless and English-like languages
that are not agglutinating or polysynthetic.
What is generally overlooked in this respect is also inter-individual variation
in learning skills, even in the acquisition of one’s mother tongue. For instance,
while the English article system (including usage of bare nouns in the singular
form) and distinction between the preterit and present perfect seem too illusive
to many non-native speakers, there are also many others who have no serious
problem with them.
Going further, we must also consider the fact that, beyond moderate inter-
individual differences, all living humans are endowed, through their common
biological phylogeny, with the capacity to learn modern language(s), identi-
fied by Noam Chomsky as the “biological endowment for language.” Assum-
ing that this capacity is unique to the human mind, the cognitive aspects of
linguistic complexity are relevant when it comes to the phylogenetic evolu-
tion of human societies and communication. For instance, Mufwene (2013)
argues that, as the hominine mental capacity increased and hominines devel-
oped larger social organizations in which they had to manage their modes of
coexistence and norms of interactions, the pressure for more informative com-
municative systems grew. According to MacNeilage (1998), neurobiological
changes and displacements of communicative functions in the brain supported
these changes.
Mufwene (2013) also assumes that languages as communicative technol-
ogy developed incrementally without foresight of the emergent structures, with
every speaker and signer contributing in different ways to the process (some
more successfully than others), exapting the current system not necessarily
according to the same principles. Thus, languages appear to have evolved
“chaotically” (i.e., without a master plan, according to the science of complex-
ity) toward more complexity both in the architecture and size of the system and
in the quantity of interactions between components within and between their
different modules.8 The modules themselves may have instantiated complexity,
in that they may be assumed to have emerged by self-organization, with the
mind dividing the labor to be run concurrently, for faster production and pro-
cessing of utterances (Mufwene 2012, 2013).
8 This is not a denial of regularities within the different modules of languages. Exaptations are
based on analogies between, on the one hand, the new meanings to be conveyed or the new
function to be played and, on the other, some meanings or functions currently in use, except that
different speakers/signers do not necessarily perceive the same ones when they innovate. Thus,
although they introduce variants that compete with each other, they also introduce regularities
(Mufwene 2008). Because the resolution of competitions between variants depends on various
factors that are not so predictable, the situation is comparable to chaos in complexity.
A last overarching ecological factor bearing on linguistic complexity is the

pressure to communicate adequately or accurately. Most of the contributions
to Ellis and Larsen-Freeman (2009, best articulated by Beckner et al.) and
to Massip-Bonet and Bastardas-Boada (2012) stand out in subscribing to the
idea that languages are adapted to the communicative needs of their speak-
ers/signers, thus they reflect changes in the latter’s cognition and adaptations
to their social and other ecologies. They keep changing over time, also reflect-
ing the accommodations that the speakers/signers make to each other toward
the emergence of some communal norms (however transient these may be),
including how they respond selectively to each other’s innovations (see also
Mufwene 2001, 2008).
Along this line, some usage- and agent-based investigations in artificial intel-
ligence attempt to shed light on the emergence of complexity in linguistic sys-
tems. They focus on how individual communicators (often identified by mod-
elers as “agents”) exapt current structures for novel communicative needs and
generate new structures (e.g., Lee et al. 2009). However, very little of this (e.g.,
Wang et al. 2004) is integrated in ways that also address the role of ecolog-
ical factors identifiable in the anatomical, mental, and social aspects of com-
munication (as discussed above) that contribute to the emergence of various
interpretations of linguistic complexity.
Different individuals may have (noticeably) different communicative needs.
How exaptations are made depends on the context of use of the extant commu-
nication system and on individual speakers’/signers’ skills, influenced by their
ontogenetic trajectories. But the evolution of languages by exaptation applies
to both the phylogenetic and the ontogenetic development of linguistic sys-
tems. We assume that our hominine ancestors from one million years ago,
for example, needed less expressive communication systems, owing to their
less developed cognitive/mental capacities and less complex social organiza-
tions. Similarly, young infants have fewer communicative needs than adults,
and they make fewer nuances, especially regarding the pragmatics of utter-
ances. Their cognitive capacities and social skills are definitely less developed
in the sense that they have not yet developed a full “theory of mind” or mind-
reading capacity, which forms the basis of the inferential adult communica-
tion (Sperber & Wilson 2002). They do not do as well as adults regarding, for
instance, metaphors, irony, implicatures, and the significance of connotations
(in addition to denotations) in linguistic communication. Their range of skills
for mature competence is not fully in place by puberty and will certainly con-
tinue to expand until adult life.
Hawkins (2009) introduces the concept of effective complexity,9 accord-
ing to which an utterance with more words may be easier or faster to process
9 Not to be confused with Gell-Mann’s “effective complexity” introduced in the beginning of this
chapter.
than an alternative with fewer words. This appear to be puzzling, as the longer
an utterance is, the more taxing it is on the short-term memory for processing,
in part because the number of (morpho-)syntactic relations within the utterance
increases. Or, to repeat our quotation from Newmeyer and Preston (2004:182),
“The more patterns a linguistic entity contains, the longer its description, and
then the greater its complexity.” However, according to Hawkins (2009:259),
“Complexity in form processing is matched by simplicity with respect to the
processing functions performed by rich case marking and definite articles.”
That is, what is structurally more complex is not necessarily more compli-
cated. The position suggests that we should not overlook the computational
aspect of processing utterances, as one may have to infer more when less is
said explicitly, such as in the case of the so-called “restricted code.” However,
Gil (2009:24) denies that there is evidence of such “hidden complexity” in Riau,
Indonesia, which, he claims, has very little syntax.
Koster (2009), Mufwene (2013), and Dor (2015) see “language[s] as tech-
nology” developed to meet humans’ communicative needs. However, as there
are alternative ways of solving the same problems, different populations have
not developed their languages in identical ways, which account for typologi-
cal variation. As with other technologies, it is indeed legitimate to ask whether
some languages are simpler than others. However, as observed by Hawkins
(2009), among others, what must one measure in assessing complexity: only the
physical, mechanical parts of the language technologies or also their abstract
aspects involving various structural and pragmatic rules/principles? The par-
ticular way in which one answers this question should help determine whether
comparing different languages relative to their complexity, as opposed to trying
to understand how complexity arises in language and how to explain it, is an
intellectually rewarding exercise.
Moving closer to practitioners of the science of complexity (viz., outside
linguistics), how variably does complexity arise from the different ways in
which different populations shape their languages through their communica-
tive acts? From an evolutionary perspective, languages are perceived not as
designed deliberately by particular populations and in different ways, but rather
as arising spontaneously from attempts by different populations to communi-
cate using phonetic or manual means (Mufwene 2013). This piecemeal emer-
gence and evolution, in unpredictable, “chaotic” ways, favor treating them as
complex adaptive systems. This is also the kind of position that arises from
proponents of usage-based or construction grammar, including Croft (2000,
2001, 2009), Steels (2011b, 2012), and Kretzschmar (2015), among others.
From this perspective, students of linguistic complexity must explain the con-
sequences of thinking of languages as complex adaptive systems. For instance,
is the addition of a phoneme to the phonemic inventory of a language, or the
deletion of one from it, as adaptive as the addition a new meaning or word to the
lexicon of a language, or even the loss of one meaning or word? Are the sys-
temic consequences the same in both cases? What about grammatical rules?
Overall, the various folds of linguistic systems call for an approach to the
global complexity of languages that is distinct from assessing the complexity
of particular linguistic structures. That is, the adaptation of linguistic struc-
tures to their ecologies (i.e., the social, cognitive, and interactional contexts in
which speakers evolve) may justify positing a usage complexity, rooted in the
actual use of language rather than in more theoretical structural considerations.
A linguistic system that fully responds with its structures to the communicative
needs and the context of use may be seen of little complexity, while a complex
linguistic system may be characterized by structures that do not reflect these
needs. For example, in different languages of the North Pacific Rim, dominant
winds play a significant role in the linguistic description of space and spatial
directions (Fortescue 2011).10 An adequate coupling exists between these lin-
guistic systems expressing space and their context of use, which can relate to
a low complexity. But due to climatic change in these regions, wind patterns
have recently been changing (Gearheard et al. 2010) and some linguistic sys-
tems may therefore now be at odds with their context of use. In such cases, the
coupling is now characterized by a greater complexity, and change is needed to
restore a better adequacy of the linguistic system (e.g., move away from winds
to describe directions). Such a complexity seems to echo a cognitive complex-
ity, in the sense that non-adapted forms, given their discrepancy with the actual
physical and mental world of the individual, may require more cognitive efforts
to be processed or learned.
Because speakers’ natural and socioeconomic ecologies constantly change,
as do their communication needs, a language always has to be adapted to these
changes. This coupling, which is typically unplanned and ad hoc, cannot be per-
fect, partly because of the time needed for the linguistic system to be adapted
to new social and pragmatic evolutions (and possibly cognitive evolutions in
a phylogenetic perspective); additionally, the coupling is imperfect because
speakers/signers have no foresight of the ultimate consequences of their cur-
rent communicative behaviors for the overall system of their language.11
10 Expressing directions according to winds can also be found in other languages, for example in
the Oceanic family (Palmer 2007).
11 As made evident by the literature on child language development and on the phylogenetic emer-
gence of language, the changes affecting speakers/signers have to do with the increase of their
mental capacity and the richer experiences they develop with their social and other aspects
of their ecologies, including the climates of their settings, the fauna, and the vegetation. If
coordinating their social lives is the primary function of language (Chapter 5 in this volume),
then all these factors exert constant pressure on language to keep adapting to new communica-
tion needs, which Mufwene (2013) characterizes as adaptive responses to changing ecological
pressures.
1.5 The Debate Over the Equal Complexity of Languages

A classic issue in linguistics pervades the various approaches to linguistic
complexity discussed in the previous sections: it is the question whether one
(type of) language is more complex than another (Sampson 2009, Gil 2009,
Trudgill 2009). Some of the recent publications have specifically been driven
by McWhorter’s (2001) claim that “the world’s simplest grammars are creole
grammars.” The debate is most evident in Sampson et al. (eds., 2009), in which
the contributors articulate different views.
As is made more obvious by especially Bisang (2009) and Deutcher (2009),
most, if not all, claims that one (type of) language is more, or less, complex
than another depend on what modules of a language are considered as (most)
representative of the architecture of language and sufficient to justify one’s con-
clusion that a (type of) language is more complex than another: morphology
and syntax only or also the phonology and semantics? For the proponents of
equal complexity, an argument is that compensations may take place between
these modules: if morphosyntax is more complex in language A than in lan-
guage B, semantics in B will be less complex than in A. However, this liter-
ature has typically focused on “bit complexity” (i.e., a system having more
units than another), which again does not do justice to the interactions between
components.
If one considers the issue more globally, what does one make of the prag-
matic considerations that help us decide whether, say, drop the ball must be
interpreted literally or as an idiom? Does a language boil down to the mechan-
ics that help us code and decode information (Hawkins 2009)? Or does it also
include the principles that guide the encoding and decoding processes, includ-
ing the choices that one must make among competing variants and the extent
to which one must rely on context during these processes? Both Bisang (2009)
and Deutcher (2009) conclude that there is no well-articulated measure of what
the global complexity of a language is; therefore claims that some languages
are more, or less, complex than others amount to what Deutcher compares to
“urban legends.”
This is not to say that the relevant literature has not taught us anything about
complexity. Some studies shed light on interesting ways in which some lan-
guages vary in the mechanics of their forms and constructions, regardless of
whether or not these are considered as compensations: for example, languages
that are spoken at a faster syllabic rate – Spanish and Japanese compared to
Mandarin and English – tend to need more syllables to convey a given seman-
tic content, and vice versa (Pellegrino et al. 2011). Introducing speech rate –
and thus language usage – in the equation, this study enabled the authors to
shift the debate from a putative (un)equal complexity of languages to whether
languages have an equal communicative capacity.
If languages are complex adaptive systems relative to the communication

needs of their practitioners (according to Mufwene 2013, in ways similar to
technologies with varying designs and developed incrementally to solve sim-
ilar problems),what is to be gained from comparing, for instance, the overall
complexity of a pidgin to that of a non-pidgin? After all, they are not used in the
same ecologies of interactions nor have they evolved to meet identical commu-
nication needs: one is used strictly as a lingua franca for limited communication
needs (typically basic informal trade transactions), whereas the other is used as
a vernacular for a broader range of communicative functions. Is the structure
of a pidgin less modular? Aren’t pidgins modular and generative like other
languages? Don’t utterances in a pidgin involve compositionality and therefore
some form of syntax, although this may not be as elaborate as in non-pidgin lan-
guages? Haven’t pidgins emerged in a nonlinear fashion? Don’t they respond
to novel communicative pressures in the same way as non-pidgin languages?
Besides, Gil (2001, 2009), for instance, argues that limited syntax is not an
exclusive peculiarity of pidgins. We must ask whether traditional discussions
do not simply suggest that there are alternative ways of developing or evolving
a language relative to the communication needs it is intended to satisfy.
In other words, while it is indeed legitimate to question the assumption that
all languages are equally complex, it is not clear that most of the current scholar-
ship can help linguists answer questions such as the following: Why are human
languages (including incipient pidgins and child language) more complex than
animal means of communication, and in what specific ways? How do human
languages as emergent phenomena or complex adaptive systems vary among
themselves from the point of view of complexity? Aside from the obvious fact
that they differ typologically in various ways, do they display different pat-
terns of complexity? In terms of the complexity of the overall linguistic system,
are there informative ways of articulating why a language may be claimed
to be more complex than another and how? Do these challenges call for an
operational definition of Language, which can be assumed by all who engage
in measuring the overall complexity in particular languages or some of their
modules?
2 The 2011 Workshop on Language Complexity

The following short history will help the reader put the contributions in this
book in the relevant perspective. The book evolved out of a successful workshop
also titled Complexity in Language: Developmental and Evolutionary Perspec-
tives that we the editors, along with Jean-Marie Hombert and Egidio Marsico,
hosted at the Collegium de Lyon in May 2011, when the lead editor was a fel-
low there. We had invited a select slate of experts in evolutionary linguistics
and child language development, in paleontology, and in artificial intelligence
to assess the state of the art, focusing on ontogeny and phylogeny, without nec-
essarily overlooking synchrony. The invitation stated the following goals:
For the “Complexity” workshop, our aim is to sort out things about the most useful
way(s) to conceive of complexity in language. Need there be only one way or can there
be several ways specific to particular research objectives? For instance, should the inter-
pretation in relation to the phylogenetic emergence of language be the same as in com-
parisons of structures of modern languages? Indeed, can one claim that one language
is more complex than another? If the answer is affirmative, how does he/she go about
demonstrating it? If the answer is negative, what are the arguments in support of the
position? Does the scholarship on language complexity measure up to the current schol-
arship on complexity theory? Can one discuss complexity without discussing emergence
as understood in complexity theory? We would like to address some of these questions
and/or any others that you may think of.
To be sure, some authors such as Östen Dahl and Talmy Givón discuss some
of these aspects of complexity briefly, the former especially in relation to
emergentism and the latter in relation to the phylogenetic evolution of lan-
guage. Nonetheless, we think that modern linguistics may benefit from more
exchanges of ideas, especially those also engaging colleagues from other dis-
ciplines who are modeling various dynamical (systemic and social) aspects of
language. We want to emphasize that our goal is not to downplay the relevance
of those approaches that focus on different aspects of structural complexity.
Rather, it is to shed more light on the other, interactional/dynamical and emer-
gentist aspects of complexity that deserve just as much attention and provide
us a better sense of how linguistic communicative systems differ from their
nonlinguistic counterparts both systemically and socially.
The contributors to this book address complexity from the perspectives of
both the evolution and the ontogenetic development of language. They focus
on social dynamics involving decisions that speakers or signers make (not nec-
essarily consciously) during their interactions with others and on the dynam-
ics that produce systems out of the different units or constructions they use
frequently in their utterances. This approach helps us address the question of
whether, say, pidgins (leaving creoles alone) still exhibit some complexity and
remain generative, in the sense that they can generate new structures and thus
be adapted to the expanding communication needs of their speakers, as is evi-
denced by expanded pidgins such as Cameroon and Nigerian Pidgin Englishes,
Tok Pisin, and Bislama.
3 The Chapters
The body of the book starts with the chapter by Luc Steels and Katrien Beuls.
Focusing on the origins and evolution of grammatical agreement as a case study,
they use multi-agent modeling to explore how various aspects of complexity
(especially in the system and in forms, among others) emerge in language.

Their working assumption is that complexity arises gradually from innova-
tions produced by interactants largely to meet their new communication needs.
It arises also from the competing variants (phonetic, lexical, syntactic, and
semantic) they introduce during the process, owing largely to imperfect copy-
ing. On the other hand, this communal form of complexity decreases as the
speakers’ emergent idiolects converge toward some norm (which maintains
less variation), the outcome of their repeated successful interactions. According
to Mufwene (2001), the mutual accommodations that speakers/signers make
to each other are indeed among the mechanisms that drive selection in lan-
guage evolution, in particular the emergence of new language varieties such as
creoles.
Steels and Beuls illustrate another fold of complexity by discussing, with
some examples, the way in which ambiguity (in simpler forms or structures)
increases complexity in processing. This arises from the fact that the hearer
has to eliminate references that may be associated with particular construc-
tions or interdependences between constituents that are not relevant in a multi-
word utterance. They show what a critical role agreement markers play in dis-
ambiguating utterances. This suggests that, although they have typically been
interpreted as adding complexity to the structure of a language, agreement
markers actually decrease complexity in processing. If complexity is assessed
in terms of processing time (Newmeyer & Preston 2014), morphological com-
plexity does not appear to be particularly costly when it enables speakers to
express more information compacted in a short form, as with fusional markers
such the Latin –arum inflected on a noun to indicate that it is plural, femi-
nine, and in the genitive/possessive. This appears to corroborate Hawkins’
(2009:259) position that “complexity in form processing is matched by simplic-
ity with respect to the processing functions performed by rich case marking and
definite articles.”
Likewise, the cooption of some current forms for new grammatical functions,
such as in grammaticalization, is said to be a case of “damping complexity,”
as the strategy reduces guesswork in figuring out the new meaning or function.
So is the erosion of forms or constructions that follows for ease of production,
supporting their hypothesis that speakers tolerate just the necessary amount of
complexity they need to communicate efficiently in their language; otherwise
they dampen it. Steels and Beuls’ discussion highlights the fact that the archi-
tecture of a language is multi-modular (though it is not evident how many mod-
ules one must posit) and that complexity can be assessed differently, depending
on the work that the module is assumed to do. Though we need not subscribe
to the traditional position that all languages are of equal level of complexity,
we may need a multi-dimensional metric for assessing the extent to which a
language is more, or less, complex than another.
Chapter 3 focuses on the following question: When a language can function

with just a dozen contrasting sounds (e.g., Hawaian), why does the average
phonetic inventory of the world’s languages amount to 29 sounds? This leads
Bart de Boer to start with the observation, “Languages are more complex than is
strictly necessary for their communicative function.” Assuming that this makes
it possible to tell which language is more complex than another phonetically,
he focuses on determining “which aspects of linguistic complexity are due to
cultural processes, and which aspects are due to cognitive biases.”
One may want to entertain the question of whether the emergence of lan-
guages can really be attributed to cultural processes. If culture is understood
roughly as the particular ways in which members of a particular population
behave and do things, is this question well formulated? Is a linguistic system
not a cultural system enabled by the particular evolutionary trajectory of its
practitioners and shapers? Not only is cultural evolution not mutually exclu-
sive with biological evolution, it also presupposes it. Only animals endowed
with uniquely generative and highly adaptive mental/cognitive capacities (viz.,
the hominine species) have produced human cultures, aspects of which include
culture-specific languages (Mufwene, in press).
We want to clarify that de Boer does not want to exclude the role of biologi-
cal evolution in language evolution cum cultural evolution. What he means by
“cultural processes” appears to be related to the fact that nobody really builds
a language with foresight and based on a master plan. If we can borrow from
William Croft (this volume, Chapter 5), a linguistic system emerges in the same
way as other “emergent phenomena” (the way systems are seen in the science
of complexity), through the addition or disuse of the strategies that the interac-
tants develop in the here and now of their communicative acts, as they integrate
gradually into a system. De Boer concludes tentatively that the “complexity of
phonological systems is due to cognitive mechanisms that re-use and generalize
building blocks.” This appears to be the consequence of transmission through
learning by inference, which replicates unfaithfully and introduces (more) vari-
ation, as well as of the nonlinear way in which linguistic systems evolve.
In Chapter 4, P. Thomas Schoenemann argues, in ways consistent with Bart
de Boer, that “the complexity of a language is the result of the evolution of
complexity in brain circuits underlying our conceptual awareness.” According
to him, modern languages have evolved from the complex interactions of bio-
logical evolution, cultural evolution, and successions of ontogenetic develop-
ment in several generations of individuals in particular populations. Linguistic
systems, with their patterns, have been facilitated by humans’ “socially-
interactive existence,” which is reminiscent of Steels and Beuls’ discussion
of how communal norms emerge. From this perspective, Schoenemann argues
that language complexity can best be understood when it is grounded in an
evolutionary perspective, focused on interactions of the biological evolution

with the changes in the ecologies in which the interacting agents evolve.
Highlighting differences and similarities between humans and chimpanzees
in particular, notably in the ways they conceptualize about the world, Schoen-
emann also concludes that the differences can be correlated with differences
in the anatomies of their brains. However, some of the similarities also sug-
gest that humans’ ability to conceptualize is pre-linguistic, suggesting that the
emergence of Language and the complexification of its architecture are the con-
sequence of the further complexification of the human mind, beyond the chim-
panzee’s under the conditions of “social-interaction existence.” He observes
that ontogenetically, “the development of expressive grammatical complexity
appears to be an exponential function of the size of the lexicon.”
Assuming that phylogenetic language evolution proceeded gradually,
William Croft argues in Chapter 5 that “at least some elements of the structural
complexity of modern human languages are the consequence of the cognitive
complexity of the conceptual structures being communicated.” He also argues
that “it is only in its social interactional context that the evolution of linguis-
tic complexity can be understood,” thus, that “the evolution of social-cognitive
complexity (in terms of joint action) is a prerequisite for the evolution of struc-
tural complexity of linguistic signals.” Language as a communal phenomenon
is the product of joint action; it is more than the sum of the actions and systems
of its practitioners. Thus it satisfies the characterization of a complex system
according to practitioners of the science of complexity, especially since it can
work only if every member of the community cooperates toward its successful
behavior/practice.
From an ecological perspective (Mufwene 2001, 2013), Croft also highlights
the role played by the material in determining the shape of the emergent sema-
siographic system (“encod[ing] information in a lasting, visual medium”), for
instance, the role of clay in reducing the number of iconic signs. If this hypoth-
esis is correct, one may assume that the shapes of the Chinese logographic char-
acters were largely influenced by the use of papyrus and ink. Overall, Croft’s
general observation is that writing systems, which have evolved from simpler
nonlinguistic and more iconic semasiographic conventions, emerged gradually,
becoming more arbitrary as they were increasingly being used to represent
speech. It is, of course, debatable whether the Chinese system has evolved to
serve speech, as the same graphic representation can be read equally in any
Chinese variety (e.g., Mandarin or Cantonese).
Croft also argues that “writing did not express grammatical elements until
centuries after its first emergence. In other words, substantial common ground
between author and reader was required to interpret just the linguistic form
encoded by early writing.” Illustrating how exaptation works in cultural
evolution, Croft shows through his discussion of musical notations how ele-
ments of the current system are recruited for new functions, to help expand it
(to the satisfaction of the performers/practitioners). This is indeed reminiscent
of the exaptation that takes place during grammaticalization.
Based on how semasiographic systems have evolved, gradually and restricted
to a wider range of functions, Croft hypothesizes that “language began in highly
restricted functional domains, and its extension to become a general-purpose
communication system was a long and gradual process in human prehistory.”
He further concludes that, like the semasiographic systems, “language initially
functioned simply as a coordination device for joint action,” conveying minimal
information.12 It is only later that it became more explicit, conveying richer
information, and developed more complexity in its architecture, especially as it
developed “displacement” (Hockett 1959), the capacity to convey information
about entities and states of affairs that are not present.
In Chapter 6, Christophe Coupé, Egidio Marsico, and François Pellegrino
start with a historical synopsis of the interest of linguists in complexity since
Ferdinand de Sausure’s (1916) characterization of a language as a system con-
sisting of interacting parts. Then they consider the particular ways in which
scholarship in complexity theory, as practiced in especially physics, mathe-
matics, and cybernetics, has inspired some of the current research in phonol-
ogy. They also underscore the fact that “a language is an aggregate of individual
idiolects” (comparable to Mufwene’s 2001 idea that it is “extrapolation from
idiolects”). As spoken of in linguistics, individual languages are reductions of
convenience, which overlook interidiolectal variation, which is more closely
matched by Michel Breal’s (1897) idea that every idiolect is somewhat a sepa-
rate language. Both interidiolectal variation and the breakdown of sounds into
(articulatory and acoustic) features add complexity in the ways phonological
systems can be thought of.
Coupé, Marsico, and Pellegrino also highlight the difference between
acknowledging that a system is complex and assessing the extent or level of
complexity, while citing some studies that have proposed particular metrics.
They warn against importing uncritically hypotheses developed by physicists
and mathematicians, which are typically based on simplified models of reality,
although they are useful research tools. This is consistent with their basic posi-
tion that any research field (including linguistics) can contribute to the science
of complexity.
Applying the statistical method to 451 phonetic inventories, the authors
address the question of whether phonological systems worldwide present evi-
dence of preferred interactions among segments that may be based on manner
12 Croft uses coordination in a way related to cooperation in theories of human and cultural evo-
lution, in reference to members of a population engaging in joint actions.
or point of articulation, nasality and orality for vowels, or any other phonetic
features. That is, are there any particular features that are more significant than
others in the emergence of phonological systems? Their conclusions include
the observation that, from an evolutionary perspective, “it [is] ( …) difficult to
conclude in favor of strongly non-linear interactions between either features
or segments.” They also note that it is difficult to assess the overall complex-
ity of a language using tools developed for physics and biology, as they do not
transfer faithfully to language. While it is evident that traditional, typologically
oriented discussions of linguistic complexity do not capture the full picture, lin-
guists should consider a more informative metric for addressing the question of
whether or not different languages display the same level of global complexity.
Barbara Davis focuses in Chapter 7 on the ontogenetic development of the
phonological component of language to explore the kind of light the analy-
sis may shed on the phylogenetic evolution of language. Like Schoenemann,
she grounds her discussion in the interpretation of complexity in the science
of complexity. According to her, “Within the tenets of complexity science,
phonological knowledge and behavioral patterns can be seen as emerging from
connections enabled by general-purpose child capacities such as learning and
cognition as opposed to language-dedicated modular mechanisms.” The emer-
gence of a complex phonological system in the child is driven by both their
cognitive-neural capacities and the production system capacities that work in
cooperation. She refers to the complex interaction between the environment
and the child’s brain in the gradual emergence of his or her phonological sys-
tem, which appears to call for an approach similar to the analysis of emergent
phenomena in the science of complexity, which may presuppose only the dis-
position of a mind sensitive to complex interactions and ready for complex
systems rather than specifically for language.
Davis’ observations are similar to those of Schoenemann. She terms
this approach “biological-functional approach to phonological acquisition,”
according to which “outcomes of phonological acquisition result from mul-
tiple interactions between heterogeneous aspects of a complex system.” She
moves on to explain the significance of change in both the ontogenetic devel-
opment and the phylogenetic emergence and evolution of language, especially
in introducing complexity throughout the adaptations that the emergent sys-
tem undergoes to satisfy current communicative pressures. The acquisition
of phonology is thus characterized as “‘change’ in infant output capacities.”
Thus, the “progressive diversification in the inventory of sound types and how
they are produced in sequences relative to ambient language patterns is usually
considered a critical index of increasing complexity toward mature phonolog-
ical capacities.”
In Chapter 8, Lucía Loureiro-Porto and Maxi San Miguel approach com-
plexity in language practice from the point of view of language choice in a
multilingual setting, especially those that may result in language loss. That is,
they focus on complexity that arises not from interactions between the differ-
ent components or modules of language as a system but from various factors
external to the system that influence speakers’ choices in their discourses, espe-
cially when they have to alternate between languages in a bilingual setting.
Their study involves modeling as a simplified tool for addressing certain spe-
cific questions regarding linguistic behavior in this particular case.
From the outset, the authors articulate the distinction they make between a
complex system from a complicated system (such as an airplane), which
is “composed of many parts, each [of which] has a clear, identifiable function
which makes prediction possible.” Complexity has to do largely with the unpre-
dictability of the properties of the whole from those of the parts. From the point
of view of language practice, the whole regards the vitality of a language, as
it depends on language choices that speakers make when they interact with
each other, without foresight of the ultimate consequences of these decisions
regarding the languages in competition. It is the whole ecosystem in which the
language belongs, in coexistence with other languages, which is of concern.
What are the factors that individually or in combinations determine the choice
of one or another language (variety) on specific occasions of social interac-
tions? Things are made more unpredictable by the fact that speakers are not
necessarily coordinated about their decisions in the typically dyadic or triadic
interactions they are most often engaged in.
Loureiro-Porto and San Miguel’s modeling reveals the significance of local
interactions in bilingual settings, regarding how they reduce the chances of
sustaining the vitality of both of the languages in competition. Other interest-
ing questions arise too, as language evolution is not uniform from one bilin-
gual setting to another. One of them regards when multilingualism spells the
endangerment of the less prestigious language(s) and when it does not. In the
real world, the explanation can be found in differences between the population
structures of the multilingual settings: for instance, those fostering assimilation
also favor endangerment, whereas those that are socially segregated according
to language groups do not.
Because it simplifies reality, modeling helps us become more aware of the
complexity of factors that influence the linguistic behavior of (members of) a
population and thus bear on language vitality. Underscoring the complexity of
actuating factors is the fact that even those population structures that are assim-
ilationist do not endanger the disadvantageous languages at the same speed
either. Loureiro-Porto and San Miguel’s modeling reveals differences between
small-world networks, regular lattices, and networks with community structure.
As the authors conclude, “The kind of network in which interactions take place
is a strong influencing factor on language dynamics, as it plays a central role in
the potential survival or disappearance of one of the languages in competition.”
Closing the book, Albert Bastardas-Boada approaches linguistic complexity

both from the parts to the whole and from the whole to the parts. The aspect of
complexity he focuses on is that which arises from the interaction of the system
with its ecology, including the socio-conceptual matrix of the speakers’ inter-
actions, economic pressures, the distribution of political power, and the effects
and language policies. Complexity increases as a consequence of the fact that
populations are not uniform and foster variation, which obtains not only inter-
individually and between groups, but also intergenerationally. According to the
author, a language must be conceived of “as a historical and, therefore, temporal
phenomenon, with earlier events playing a major role in how the phenomenon
evolves.” History shapes and may provide some explanation for the present,
including current linguistic behavior.
There are indeed other aspects of complexity that this book, like the dominant
literature in linguistics, still does not tackle, despite our focus on developmental
and evolutionary perspectives. One of these is the extent to which increase in
population size affects complexity in the communal language, perhaps more
in the pragmatic and social aspects of its usage than in its structures. Another
is whether contact with (an)other language(s) reduces or increases structural
complexity, and under what specific conditions. Is contact the only explanation
for why major world languages such as Modern English and Modern French
have lost most of the inflections of Old English and Old French, respectively?
On the other hand, hasn’t contact also increased complexity in their systems in
other ways, such as in introducing alternative grammatical rules or changing
some of the rules while preserving some exceptions? These are all interesting
topics for future studies.
REFERENCES
Aboh, Enoch O. & Norval Smith, eds. 2009. Complex Processes in New Languages.
Amsterdam: John Benjamins.
Baerman, Matthew, Dunstan Brown, & Greville Corbett, eds. 2015. Understanding
Morphosyntactic Complexity. Oxford: Oxford University Press.
Bane, Max. 2008. Quantifying and Measuring Morphological Complexity. In Proceed-
ings of the 26th West Coast Conference on Formal Linguistics, ed. by Charles
B. Chang & Hannah J. Haynie, 69–76. Somerville, MA: Cascadilla Proceedings
Project.
Beckner, Clay, Richard Blythe A., et al. 2009. Language is a Complex Adaptive System:
Position Paper. In Ellis & Larsen-Freeman (eds.), 1–26.
Berlage, Eva. 2014. Noun Phrase Complexity in English. Cambridge: Cambridge Uni-
versity Press.
Bisang, Walter. 2009. On the Evolution of Complexity: Sometimes Less is More in East
and Mainland Southeast Asia. In Sampson et al. (eds.), 34–49.
Blommaert, Jan. 2013. Ethnography, Superdiversity and Linguistic Landscapes: Chron-
icles of Complexity. Bristol, UK: Multilingual Matters.
Blythe, Richard A. & William Croft. 2009. A Usage-Based Account of Constituency

and Reanalysis. In Ellis & Larsen-Freeman (eds.), 47–63.
Bréal, Michel Jules Alfred. 1897. Essai de Sémantique: Science des Significations. Paris:
(Original publisher unknown).
Coupé, Christophe. 2016. Commentary: Defining and Assessing Constraints on Lin-
guistic Forms. Journal of Language Evolution 1(1): 52–55.
Coupé, Christophe, Egidio Marisco, & François Pellegrino. 2009. Structural Complexity
of Phonological Systems. In Pellegrino et al. (eds.), 141–169.
Croft, William. 2000. Explaining Language Change: An Evolutionary Approach.
London: Longman.
Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typological
Perspective. Oxford: Oxford University Press.
Croft, William. 2009. Towards a Social Cognitive Linguistics. In New Directions in
Cognitive Linguistics, ed. by Vyvyan Evans & Stéphanie Pource, 395–420. Ams-
terdam: John Benjamins.
Culicover, Peter. 2013. Grammar and Complexity: Language at the Intersection of Com-
petence and Performance. Oxford: Oxford University Press.
Cyran, Eugeniusz. 2010. Complexity Scales and Licensing in Phonology. Berlin:
Mouton de Gruyter.
Dahl, Östen. 2004. The Growth and Maintenance of Linguistic Complexity. Amsterdam:
John Benjamins.
Dahl, Östen. 2009. Testing the assumption of complexity invariance: The case of Elfa-
dian and Swedish. In Sampson et al. (eds.), 50–63.
DeGraff, Michel. 2001a. On the Origin of Creoles: A Cartesian Critique of neo-
Darwinian Linguistics. Linguistic Typology 5: 213–310.
DeGraff, Michel. 2009. Language Acquisition in Creolization (and Language Change):
Some Cartesian-Uniformitarian Guidelines. Language and Linguistics Compass
3&4: 888–971.
Deutcher, Guy. 2009. “Overal Complexity”: A Wild Goose Chase? In Sampson et al.
(eds.), 243–251.
Dor, Daniel. 2015. The Instruction of Imagination: Language as a Social Communica-
tion Technology. Oxford: Oxford University Press.
Edmonds, Bruce. 1999. What is Complexity? The Philosophy of Complexity per se with
Application to some Examples in Evolution. In The Evolution of Complexity, ed. by
Francis Heylighen, Johan Bollen, & Alexander Riegler, 1–18. Dordrecht: Kluwer.
Ellis, Nick C. & Diane Larsen-Freeman, eds. 2009. Language as a Complex Adaptive
System. Malden, MA: Wiley-Blackwell.
Faraclas, Nicholas & Thomas B. Klein, eds. 2009. Simplicity and Complexity in Creoles
and Pidgins. London: Battlebridge Publications.
Fortescue, Michael. 2011. Orientation Systems of the North Pacific Rim. Copenhagen:
Museum Tusculanum Press.
Gearheard, Shari, Matthew Pocernich, Ronald Stewart, Joelie Sanguya, & Henry P.
Huntington. 2010. Linking Inuit Knowledge and Meteorological Station Obser-
vations to Understand Changing Wind Patterns at Clyde River, Nunavut. Climatic
Change 100(2): 267–294.
Gell-Mann, Murray. 2003. Effective Complexity. In Nonextensive Entropy – Interdisci-
plinary Applications, ed. by Murray Gell-mann & Constantino Tsallis, 387–398.
Oxford: Oxford University Press.
Gershenson, Carlos, ed. 2008. Complexity: 5 Questions. Denmark: Automatic Press.

Gil, David. 2001. Creoles, Complexity, and Riau Indonesian. Linguistic Typology 5:
325–371.
Gil, David. 2009. How Much Grammar Does it Take to Sail a Boat? In Sampson et al.
(eds.), 19–33.
Givón, T. 2009. The Genesis of Syntactic Complexity. Amsterdam: John Benjamins.
Givón, T. & Masayoshi Shibatani, eds. 2009. Syntactic Complexity: Diachrony, Acqui-
sition, Neuro-Cognition, Evolution. Amsterdam: John Benjamins.
Hawkins, John A. 2005. Efficiency and Complexity in Grammars. Oxford: Oxford Uni-
versity Press.
Hawkins, Hohn A. 2009. An Efficiency Theory of Complexity and Related Phenomena.
In Sampson et al. (eds.), 252–268.
Heine, Bernd & Kuteva, Tania. 2007. The Genesis of Grammar: A Reconstruction.
Hockett, Charles F. 1959. Animal “Languages” and Human Language. Human Biology
31: 32–39.
Housen, Alex, Folkert Kuiken, & Ineke Vedder, eds. 2012. Dimensions of L2 Perfor-
mance and Proficiency: Complexity, Accuracy and Fluency in SLA. Amsterdam:
John Benjamins.
Keller, Rudi. 1994. On Language Change: The Invisible Hand in Language. London,
New York: Routledge.
Kortmann, Bernd & Benedikt Szmrecsanyi, eds. 2012. Linguistic Complexity. Berlin:
Mouton de Gruyter.
Koster, Jan. 2009. Ceaseless, Unpredictable, Creativity. Biolinguistics 3: 61–92.
Kretzchmar, William A, Jr. 2015. Language and Complex Systems. Cambridge: Cam-
bridge University Press.
Labov, William. 2001. Principles of Linguistic Change: Social Factors. Malden, MA:
Blackwell.
Larsen-Freeman, Diane. 2008. Complex Systems and Applied Linguistics. Oxford:
Oxford University Press.
Lee Namhee, Lisa Mikesell, Anna Dina L. Joaquin, Andrea W. Mates, & John H. Schu-
mann. 2009. The Interactional Instinct. Oxford: Oxford University Press.
Liebertman, Philip. 2012. The Unpredictable Species: What Makes Humans Unique.
Princeton: Princeton University Press.
Lupian, Gary & Rick Dale. 2010. Language Structure is Partly Determined by Social
Structure. PLoS ONE 5(1): e8559.
MacNeilage, Peter F. 1998. The Frame/Content Theory of Evolution of Speech Produc-
tion. Behavioral and Brain Sciences 21(4): 499–511.
Maddieson, Ian 2011. Phonological Complexity in Linguistic Patterning. In Proc. of the
17th International Congress of Phonetic Sciences, 17–21.
Massip-Bonet, Angels & Albert Bastardas-Boada, eds. 2012. Complexity Perspectives
on Language, Communication and Society. Heidelberg: Springer.
McCawley, James D. The Syntactic Phenomena of English. Chicago: University of
Chicago Press.
McMahon, April. 1994. Understanding Language Change. Cambridge: Cambridge
University Press.
McWhorter, John H. 2001. The World’s Simplest Grammars are Creole Grammars.
Linguistic Typology 5: 125–166.
McWhorter, John H. 2012. Linguistic Simplicity and Complexity: Why do Languages

Undress. Berlin: Mouton de Gruyter.
Miestamo, Matti, Kaius Sinnemäki, & Fred Karlsson. 2008. Language Complexity:
Typology, Contact, Change. Amsterdam: John Benjamins.
Mitchell, Melanie. 2009. Complexity: A Guided Tour. Oxford: Oxford University Press.
Mufwene, Salikoko S. 1996. Creolization and Grammaticization: What creolistics
Could Contribute to Research on Grammaticization. In Changing Meanings,
Changing Functions, ed. by Philip Baker and Anand Syea, 5–28. London: Uni-
versity of Westminster Press.
Mufwene, Salikoko S. 2001. The Ecology of Language Evolution. Cambridge: Cam-
bridge University Press.
Mufwene, Salikoko S. 2008. Language Evolution: Contact, Competition and Change.
London: Continuum Press.
Mufwene, Salikoko S. 2009. Postscript: Restructuring, Hybridization, and Complexity
in Language Evolution. In Aboh & Smith (eds.), 367–399.
Mufwene, Salikoko S. 2012. The Emergence of Complexity in Language: An Evolu-
tionary Perspective. In Massip-Bonet & Bastardas-Boada (eds.), 197–218.
Mufwene, Salikoko S. 2013. Language as Technology: Some Questions that Evolution-
ary Linguistics Should Address. In In Search of Universal Grammar: From Norse
to Zoque, ed. by Terje Lohndal, 327–358. Amsterdam: John Benjamins.
Mufweme, Salikoko S. in press. The Evolution of Language as Technology: The
Cultural Dimension, in Beyond the Meme, ed. by William Wimsatt & Alan Love.
Minneapolis, MN: University of Minnesota Press.
Newmeyer, Frederick J. & Laurel B. Preston, eds. 2014. Measuring Grammatical Com-
plexity. Oxford: Oxford University Press.
Newmeyer, Frederick J. & Laurel B. Preston. 2014. Introduction. In Newmeyer & Pre-
ston (eds.), 1–13.
Palmer, Bill. 2007. Pointing at the Lagoon: Directional Terms in Oceanic Atoll-Based
Languages. In Language Description, History and Development. Linguistic Indul-
gence in Memory of Terry Crowley, ed. by Jeff Siegel, John Lynch, & Diana Eades,
101–118. Amsterdam/Philadelphia: John Benjamins.
Pellegrino, François, Christophe Coupé & Egidio Marsico. 2011. A Cross-Language
Perspective on Speech Information Rate. Language 87(3): 539–558.
Pellegrino, François, Egidio Marsico, Ioana Chitoran, & Christophe Coupé, eds. 2009.
Approaches to Phonological Complexity. Berlin: Mouton de Gruyter.
Poincaré, Henri. 1891. Sur le problème des trois corps et les équations de la dynamique.
Bulletin Astronomique 8: 12–24.
Risager, Karen. 2006. Language and Culture: Global Flows and Local Complexity.
Clevdon: Multilingual Matters.
Rissanen, Jorma. 1978. Modeling by Shortest Data Description. Automatica 14(5): 465–
471.
Robinson, Peter. 2011. Second Language Task Complexity: Researching the Cognition
Hypothesis of language learning and performance. Amsterdam: John Benjamins.
Sampson, Geoffrey. 2009. A Linguistic Axiom Challenged. In Sampson et al. (eds.),
1–18.
Sampson, Geoffrey, David Gil, & Peter Trudgill, eds. 2009. Language Complexity as
an Evolving Variable. Oxford: Oxford University Press.
Smith, Adam. 1776. An Inquiry into the Nature and Causes of the Wealth of Nations.
London: W. Strahan and T. Cadell.
Sperber, Dan. 1994. Understanding Verbal Understanding. In What is Intelligence? ed.
by Jean Khalfa, 179–198. Cambridge: Cambridge University Press.
Sperber, Dan & Deidre Wilson. 2002. Pragmatics, Modularity and Mind-Reading. Mind
and Language 17: 3–23.
Steels, Luc. 1998. Synthesising the Origins of Language and Meaning Using Co-
evolution, Self-Organisation and Level Formation. In Approaches to the Evolution
of Language: Social and Cognitive bases, ed. by James R. Hurford, Chris Knight,
& Michael Studdert-Kennedy, 384–404. Cambridge: Cambridge University Press.
Steels, Luc. 2011a. Modeling the Cultural Evolution of Language. Physics of Life
Reviews 8: 339–56.
Steels, Luc, ed. 2011b. Design Patterns in Fluid Construction Grammar. Amsterdam:
John Benjamins.
Steels, Luc, ed. 2012. Experiments in Cultural Language Evolution. Amsterdam: John
Benjamins.
Strogatz, Steven. 2003. Sync: The Emerging Science of Spontaneous Order. New York:
Theia.
Trudgill, Peter. 2009. Sociolinguistic Typology and Complexification. In Sampson et al.
(eds.), 98–109.
Trudgill, Peter. 2011. Sociolinguistic Typology: Social Determinants of Linguistic Com-
plexity. Oxford: Oxford University Press.
Victorri, Bernard. 1994. La Construction Dynamique du Sens. In Passion des Formes:
Dynamique Qualitative, Sémiophysique et Intelligibilité, ed. by Michèle Porte,
733–747. Fontenay-Saint Cloud: ENS Editions.
Walther, Géraldine & Benoît Sagot. 2011. Modélisation et Implémentation de
Phénomènes Non-Canoniques. In Traitement Automatique des Langues, 52(2)
(Vers la morphologie et au-delà), N. Hathout & F. Namer (eds.), p. 91–122.
Wang, William Shi-Yuan, Jinyun Ke, & James W. Minett. 2004. Computational Studies
of Language Evolution. In Computational Linguistics and Beyond, ed. by C. R.
Huang, & W. Lenders, 65–106. Academia Sinica: Institute of Linguistics.
Weinreich, Uriel, William Labov, & Marvin I. Herzog. 1968. Empirical Foundations for
a Theory of Language Change. In Directions for Historical Linguistics: A Sympo-
sium, ed. by Winfred P. Lehman & Yakov Malkiel, 97–195. Austin: University of
Texas Press.
2 How to Explain the Origins of Complexity in
Language: A Case Study for Agreement Systems
Luc Steels and Katrien Beuls
1 Introduction
It is useful to make a distinction between five different types of complexity in
language:
1. Inventory complexity is about the number of conceptual, phonological, lexi-
cal and grammatical building blocks in use by an individual or by a particular
language community. It pertains to the number of phonemes, the number of
concepts (e.g., color categories, action categories, spatial relations, tempo-
ral relations, etc.), the number of lexical items (words and morphological
elements), the number of grammatical categories (syntactic classes, cases,
classifiers, grammatical functions and types of phrase structures), and the
number of grammatical constructions.
2. Form complexity is measured in terms of statistics over the length of words
and the length of utterances.
3. Processing complexity is defined as the cognitive effort involved in pars-
ing and producing utterances: How much memory is needed? How many
processing steps are required? How much combinatorial search is unavoid-
able? How much ambiguity is left before semantic interpretation? Process-
ing complexity depends on the architecture of the language processing sys-
tem, the nature and complexity of the grammar, the complexity of forms,
and the ecological complexity of the environment in which language users
operate.
4. Learning complexity refers to the amount of uncertainty that learners face
when acquiring new words or constructions. Is it always possible to uniquely
guess unknown meanings and functions, or does a search space get gener-
ated? How big is this search space?
5. Population-level complexity is concerned with the properties of the evolving
communal language as a whole: How much variation is there in the popu-
lation with respect to the usage or knowledge of linguistic forms? Or, con-
versely, what is the coherence, that is, the degree of sharing? What is the
resilience of particular constructions in the process of cultural transmission?
How intense is language change?
30
How to Explain the Origins of Complexity in Language 31
The central thesis of the present chapter is that human languages evolve in
such a way as to minimize complexity at all these levels while providing
enough expressive power to handle all the meanings relevant to the community.
Minimizing complexity is necessary to keep the language viable; otherwise it
would become too complex for regular usage and would no longer be learnable
(Hawkins 2005).
Minimizing complexity does not happen by intelligent design. Language
processing is mostly subconscious and even linguists have no total explicit
understanding of the systems underlying a language, let alone that they could
rationally expand it. There is no central committee that designs or has total con-
trol over the use of English, Chinese or any other language. Speakers are aware
of dialectal differences but in general they do not have the power to directly
influence others.
The most plausible alternative to intelligent design is a Darwinian selection-
ist process mapped to the cultural level (Mufwene 2001; Steels 2012b). Lan-
guage users unavoidably generate variation, often because they do not know
which alternatives already exist in the population. They also unavoidably gen-
erate complexity, for example because they do not know of a more compact
way to express their message or they are unknowingly using a non-standard
variant. But each language user balances this with strategies to minimize com-
plexity, for example, by simplifying grammatical structure, leaving out some of
the phonemes of a word, adopting the most frequently occurring variant, and so
on. Variants for concepts, words and constructions compete and those that lead
to higher communicative success and less cognitive effort have a higher chance
of survival, simply because speakers and hearers want to understand each other
and they do not want to spend more effort than is necessary. There is no optimal
solution, and so languages keep moving on a linguistic fitness landscape (van
Trijp 2014), sometimes optimizing one aspect (e.g., minimizing form complex-
ity) while relaxing another one (e.g., processing complexity).
Let us look in more detail at the many ways in which language users mini-
mize the different types of complexity introduced earlier:
1. Damping inventory complexity. Language is an open system. New mean-
ings constantly come up and need to become expressible. Routine expressions
tend to lose their force and are then replaced by new ones. Consequently, there
is a steady renewal in the inventory of a language: New words are invented or
existing words are used for new meanings. New grammatical constructions are
introduced as well, sometimes based on new paradigms or on the application of
a paradigm to a new domain. Existing constructions may also be coerced into
new uses or words may be reanalyzed to fit with existing constructions.
At the same time, inventories must be kept within bounds because larger
inventories take up more linguistic memory, increase access time, and require
longer time for learning. How are the inventories kept in check? Many words
32 Luc Steels and Katrien Beuls
or constructions simply get out of fashion after a while. They disappear from
usage and are forgotten, particularly after a few generations. There is also a
general tendency to reuse existing forms as much as possible even if they
have already existing functions, a process similar to exaptation in biology. If
a new invention is based on the exaptation of an existing word or construc-
tion in a slightly different context, then there is a higher chance that the hearer
might guess this new meaning than if a radically new invention is made. Hence
the exapted invention has a higher chance to propagate and survive in the
communal language and it contributes to keeping the language inventory in
check.
2. Damping form complexity. Novel words tend to be pronounced in full,
but once introduced, their production gets eroded over time due to articulatory
optimization or errors, reducing both the length of words and the length of utter-
ances in which they appear. Some words end up being morphemes, then clitics
and later affixes. A similar form of optimization process happens at the level of
constructions. The first time new meanings are expressed it is usually done in
an elaborate, circumscriptive way. But with routine usage, the combination of
constructions that was used to build the more elaborate phrase gets collapsed
into a single construction to achieve faster processing, a process usually called
chunking. Words within a chunk may then start to disappear and the phrase
may progressively become idiomatic. For instance, the English collocation by
name as in I know all of my students by name is a shorter version of the full
noun phrase by the/their name.
3. Damping processing complexity. Strategies for reducing inventory size
and form complexity may lead to syncretism (the same form gets multiple
meanings) and hence syntactic ambiguity. Syntactic ambiguity leads in turn
to combinatorial search, which needs to be kept in check because it causes
an exponential increase in memory and processing time and hence fluent rapid
speaking and listening becomes more and more difficult. Inventory reduction by
reuse of existing forms may also lead to semantic ambiguity because the forms
in the utterance no longer unequivocally communicate the intended meaning.
Processing complexity can be dampened by adding additional syntactic struc-
ture. For example, as discussed later in this chapter, grammatical agreement
is ways in which language users signal which words belong together in the
same sentence and hence it reduces the combinatorial complexity of parsing.
More generally, we argue that the reduction of cognitive effort in parsing and
interpretation is one of the main motivating forces for the origins of syntactic
structure.
4. Damping learning complexity. Language learners need to guess the mean-
ing and function of unknown words or constructions. The more possibilities
there are, the more hypotheses the learner has to consider. The learning task
gets more manageable if the learner can make strong use of context, which
reduces the set of possible meanings, and if the speaker, acting as tutor, helps
by scaffolding the complexity of utterances and by correcting wrong uses of
words and constructions. The structure of language itself can also help. For
example, syntactic structure may not only decrease the complexity of parsing
but also help to constrain the possible meanings of an unknown word, lead-
ing to syntactic bootstrapping. For example, the adjectives in a nominal phrase
are typically ordered based on semantic classes and hence this ordering can
help constrain the meaning of an unknown adjective, thus reducing learning
complexity.
5. Damping population-level complexity. Individuals in a population will
never share the same inventory because language learners have to indepen-
dently acquire the language based on their own history of interactions with
others and their own needs, and each language user has in principle the right to
extend the language. Moreover, language is a social tool that is used by speakers
to stress their identity as belonging to a certain community or to prevent them-
selves from adopting ongoing changes (Labov 1966). Hence we see tremendous
variation. On the other hand, language variation is detrimental to the efficiency
of a language because language users need to be able to parse and interpret
expressions that deviate from their own usage and possibly even store these
variations. Instead of learning a single language they have to learn a multitude
of overlapping languages. So we need a strong force that dampens population-
level complexity and ensures that idiolects become shared into a common lan-
guage. We will see later that appropriate learning strategies (such as a lateral
inhibition learning rule) can act as such a force.
2 Agent-Based Modeling
The far from exhaustive list of suggestions provided in the previous section
certainly sounds plausible, but how can we make them more concrete and test
them with scientific methods? There are several ways to study language dynam-
ics: linguistic analysis deconstructs the kinds of rules that must intervene in the
parsing and production of utterances, studies in historical linguistics show how
a language has been shaped and reshaped over time by its users, research into
language typology and sociolinguistics maps the diversity and variation in lan-
guage, psychological experiments measure delays in response to utterances or
difficulties of learning, and neuroimaging experiments measure brain activa-
tion while parsing, producing or learning utterances. All of these approaches
are valuable. There is also another approach, which is gaining momentum since
the mid-1990s, namely agent-based modeling (Steels 1995; Smith et al. 2003;
Loreto et al. 2011; Steels 2011).
Agent-based approaches model a population of linguistic agents engaging

in situated interactions, either embodied or simulated, called language games,
and test the behavior of the model through computer simulations or robotic
experiments. An agent-based model defines a set of possible situations and an
interaction pattern between agents that contains both verbal and non-verbal
aspects (such as pointing). It also contains a set of mechanisms that the agents
in a population can use for producing and understanding utterances, such as cat-
egorization, pattern matching, associative retrieval, and so on. The agents are
also endowed with a Language Acquisition Device in the form of strategies they
can use to build up (i.e., learn or extend) their language systems. Then a series
of games is being played between randomly chosen agents from the popula-
tion. Each agent can play both the roles of speaker and listener. As a side effect
of a game, the participating agents may expand their conceptual or linguistic
inventory, or adapt it, based on the outcome of the game. We can track different
measures, both for individual agents and for populations of agents, in order to
tell us whether a model works, in the sense of whether the desired linguistic
structures indeed arise in model simulations. If they do not, the mechanisms
and strategies initially put into the agents are insufficient and they are changed
for the next experiment until we know what causal mechanisms generate the
phenomena we want to explain.
For example, experiments trying to explain how a set of color terms and
color concepts could emerge, would involve agents that are initially endowed
with concept formation mechanisms and strategies for inventing, acquiring or
aligning associations between words and color concepts (Steels & Belpaeme
2005). At the start of the experiment there is neither a shared color language
nor any color concept inventory, but if we have been able to come up with the
necessary and sufficient mechanisms and strategies we expect to see after a
number of games that a shared color language emerges (which indeed happens
in these computer simulations). If that is the case, we have evidence that the
given mechanisms and strategies (plus the interaction patterns and ecological
conditions) explain the emergence of color terms and of the color concepts they
express.
An agent-based model typically focuses on one aspect of language; for exam-
ple, color terms, names for actions, tense-aspect systems, quantifiers, expres-
sion of information structure, phrase structure, case grammar, or grammatical
function. (See Steels 2012a for a representative sample.) Each time a particular
language game is chosen that brings out the aspect of language being studied.
For instance, if we want to study the emergence and evolution of color terms we
obviously need an environment with objects of different colors and the color
should matter in the interaction between the agents.
Agent-based models have numerous advantages compared to verbal theoriz-
ing; it is therefore surprising that this methodology is still so controversial, even
though it is widely accepted in sociology, biology, economics and many other

scientific disciplines (de Landa 2011):
r Agent-based models require a comprehensive mechanistic model of language
processing, which is usually lacking in linguistic or psychological research.
Most often grammars are not formally represented, and if they are, the for-
malization is often not in a form that can be used by processing algorithms.
Various learning strategies are assumed but they are only described vaguely
and not operationalized, so that we cannot know whether they are effective.
Agent-based models can only be built with effective computational models of
language processing and because these models need to cope with emerging
and evolving languages, they challenge several dogmas in formal and com-
putational linguistics (Steels 2012b). In that sense, agent-based models push
not only the state of the art in language evolution but also in computational
linguistics and linguistic theory in general.
r When doing simulations using agent-based models it is possible to measure
precisely the complexity of the languages generated with a particular strat-
egy at each of the different levels of complexity described earlier. We can do
repeatable experiments and examine the statistical distribution of the results.
It is possible to selectively add or take out mechanisms or change a strat-
egy so that we can determine the causal relationships between mechanisms,
strategies or parameters of the model, and the structure of an emergent lan-
guage system. These methods are standard practice in science and there is no
reason why the study of language evolution should not use them.
Agent-based models are occasionally criticized because the strategies and
processing mechanisms are stated at an abstract, mathematical/computational
level, but abstraction is a characteristic of every scientific model. They are also
criticized because such models are not validated by psychological observations
or neuroimaging data. However, we are still far removed from effective mod-
els of language processing and even if we would have them, we would not
be able to validate them, given the precision of current neuroimaging instru-
ments. Instead, agent-based models should be seen by psychologists and neu-
roscientists as theoretical hypotheses that they can test using their observational
techniques. Indeed, some cognitive psychologists have already been replicating
similar conditions as used in agent-based models with results and phenomena
comparable to what we see in agent-based simulations (Galantucci 2005; Sel-
ten & Warglien 2007; Pickering & Garrod 2013). The observations made by
historical linguists remain the ultimate empirical target of agent-based models
of cultural language evolution. Historical linguists have identified many of the
processes that are active in language emergence and change, such as lexical
category shifts, phonological erosion of recruited word forms, damping of syn-
onymy, and so on (Heine & Kuteva 2007). Similar processes have been identi-
fied in the formation of creoles (Mufwene 2008). Agent-based models explicate
the mechanisms and strategies behind these proposals and test their explanatory
power. Of course we cannot directly replicate precisely the historical evolution
of Old English into Middle English because there are so many contingencies,
but, just like in evolutionary biology, we should be able to demonstrate what
strategies and cognitive mechanisms have been playing a role. A good example
is agent-based models about the changes to the German article system from the
cleaner old German system to contemporary German (van Trijp 2013).
3 Case Study for Agreement Systems

The best way to illustrate both the theory of language evolution by linguistic
selection and the methodology of agent-based modeling is to look at concrete
examples. We will do this now for grammatical agreement, building further on
agent-based models reported earlier (Beuls & Steels 2013). Grammatical agree-
ment occurs when two linguistic units (typically a word or a phrase) receive a
marker that indicates their relatedness (Lehmann 1988). For instance, in the
French utterance une belle fille – “a pretty girl” – both the article and the adjec-
tive have a feminine form, marked by the final vowel -e, which is governed by
gender of the head noun fille, which is specified as feminine. Agreement sys-
tems can be very complex and messy and many linguists (and second language
learners) have been puzzled as to why they are there. We now look at the dif-
ferent types of complexity discussed earlier and study the origins of agreement
systems from that point of view.
3.1 The Language Game

In the current experiment, agents play a game of reference (known as the Nam-
ing Game), which has been used in many earlier investigations of language
dynamics (e.g., Steels 1995; Dall’Asta et al. 2006). All agents can play the role
of speaker or hearer and have an equal chance of playing a game. The agent
chosen as speaker goes through the following steps: (i) The speaker selects a
subset of the objects in the current situation to act as the topic of the utterance.
(ii) The speaker looks for a distinctive combination of properties for each of
the objects in the topic, where a distinctive combination is a set of properties
that are true for the object but not for any other object in the current situation.
(iii) The speaker retrieves the minimal set of words in the vocabulary that cover
the chosen properties, which implies that words with the largest coverage are
preferred. The speaker then utters these words. Although there is unavoidably
a sequential ordering to the words, this does not carry any meaning, that is,
agents use a word-order free language.
Next, the agent chosen as hearer goes through the following steps: (i)
The hearer looks up the words in the vocabulary and thus reconstructs what
properties have been communicated by the speaker. (ii) The hearer identifies
which objects in the present situation satisfy these properties and points to
them. The game is a success if the objects pointed to by the hearer are those
initially chosen by the speaker. The game fails if this is not the case or if the
utterance remains semantically ambiguous, that is, if there is more than one
possible interpretation that fits with the current situation model.
3.2 Processing Complexity

Although this language game looks deceptively simple, there is the potential
for a combinatorial explosion and semantic ambiguity. The hearer does not
know how many objects the speaker is talking about and the utterance does
not communicate which words are about the same object. Hence, all possible
combinations must be tried by the hearer to find those that fit with the current
situation. This obviously implies a very high degree of processing complexity.
Processing complexity can be measured in different ways and it always
depends on the architecture of the language processing system that one has
implemented. It could be the nodes in the search tree during production or
parsing of a particular sentence, the number of constructions needed to build or
understand an utterance, or the accessibility of the constructions that are used to
parse or produce the utterance, for example, frequently used constructions can
be made more accessible (i.e., primed). All of these measures tell us something
about the cognitive effort of the speaker or the hearer.
In the case of agreement the most relevant complexity measure is the number
of hypotheses that remain after parsing a particular utterance. Suppose there is
neither word order nor any other kind of grammar. Then the utterance brown
blue cat chair big, could have meant:
r (cat) (big blue brown chair);
r (brown cat) (big blue chair);
r (big cat) (blue brown chair);
r (blue cat) (big brown chair);
r (big brown cat) (blue chair);
r (big blue cat) (brown chair); or
r (big blue brown cat) (chair) and so on.
And this still assumes that there are only two objects. If an utterance can be
about more than one object, then the set of possible combinations is still much
larger. More generally, the number of possibilities Bn is equal to the number of
partitions of the set D of words in an utterance of size n. Bn is known as the
Bell number and can be defined using the following equation (Bell 1938):
n
n
Bn+1 = Bk
k
k=0
with B0 = 0 and B1 = 1. Bn grows exponentially with the number of words. It

means that the sentence you are now reading (which contains 20 words) gen-
erates 51,724,158,235,372 partitions and hence possible interpretations. This
definition is only true if all words in an utterance are semantically equivalent
(i.e., can be grouped together in every possible way), which is not the case in
real language use where ontological selection criteria help to determine which
words should be grouped together. Yet, selection restrictions reduce the number
of solutions but this number remains exponential as the length of the utterance
increases (Garcia Casademont & Steels 2014).
However, now suppose that the agents adopt a strategy to use agreement
markers, that is, morphemes are added to every word that refers to the same
object. An example utterance could now be brown-ki blue-ba cat-ki chair-ba
big-ba. It is now directly clear which words belong together and the combinato-
rial explosion disappears. Agreement markers are therefore an efficient means
to dampen processing complexity as all the combinations that are evoked by
the Bell number are hereby reduced to a single one.
We can demonstrate this with an agent-based simulation, in which a popu-
lation of agents starts without an agreement system but each agent is endowed
with a strategy to invent new markers when needed (as speaker) or to adopt them
(as hearer). The markers are in this case purely formal (i.e., they don’t express
number, gender, case or any other semantic feature). We see in Figure 2.1 that
the number of hypotheses gets drastically reduced to a single one as soon as
the agents share a sufficient set of markers.
Clearly an agreement strategy leads to a language that requires less cogni-
tive effort in parsing and interpreting utterances; hence this explains, from a
functional point of view, why we find agreement systems in human languages.
3.3 Population-Level Complexity: Language Variation

When speakers invent new markers, there is unavoidable variation, because
individuals are not supposed to have a telepathic general overview of all inter-
actions happening throughout the population. Agents end up with a working
system, but there are many more markers than is absolutely necessary and hence
agents have to learn many more markers. This in turn slows down the emer-
gence of the agreement system and makes it harder for new agents coming into
the population to acquire the existing set of markers.
This unnecessary language variation can be damped when agents use lateral
inhibition as part of the marker strategy discussed in the previous paragraph
(de Vylder & Tuyls 2006; Steels 1998). Agents maintain a score σm between
0.0 and 1.0 for every marker m. The initial score of a new marker is σ = 0.5.
The hearer (but not the speaker) increments this score whenever a marker mi
Figure 2.1 Results of an agent-based simulation, with a population of 10

agents playing 200 language games in total, which means an average of 40 per
agent. The x-axis shows the number of games played; the left y-axis the num-
ber of hypotheses left after parsing and the right y-axis the number of markers.
When agents use the agreement marker strategy, the number of hypotheses
decreases rapidly thus damping drastically the processing complexity. (Error
bars indicate standard deviation of 10 runs.)
appears in an utterance and decreases the score of all other non-used markers
mj using the following equations with alignment rate γ = 0.2:
σm ← σm (1 − γ ) + γ
i i
σm ← σm (1 − γ )
j j
When choosing which marker to use, the speaker should prefer the marker with
the highest score that was not yet used in the same utterance. This establishes
a positive feedback loop between usage and marker preference, which leads
to a shared minimal marker system. The resulting process is similar to many
other self-organizing processes found in natural systems in which large-scale
structures arise from local interactions through random behavior influenced by
positive feedback loops (Camazine et al. 2001).
The effect of using lateral inhibition is illustrated in Figure 2.2. When agents
simply invent markers or acquire them from others, the number of markers
becomes stable with an inventory of 10 markers. When they use lateral inhibi-
tion as part of the agreement strategy, they end up with the optimal number of
three markers, because utterances in this experimental run have been limited
Figure 2.2 Comparison of a population that uses lateral inhibition and one that
does not. The use of lateral inhibition gives a selectionist advantage because
it leads to a smaller marker inventory at the population level, that is, less
unneeded variation.
to contain maximally three objects. When there are more than three objects,
the population would settle on more markers but always the minimal number
needed.
This simulation demonstrates again that the adoption of a particular strategy
can lead to a more efficient language system in the sense that it dampens com-
plexity along a particular dimension. In this case, lateral inhibition dampens
variation in the population, so that each agent needs to keep fewer markers in
memory and new agents need to learn fewer markers.
3.4 Learning Complexity

So far the strategies that we investigated use only formal markers, that is,
symbols without any meaning. A formal marker such as –ki in the previous
example purely functions as a grouping label. It indicates that words carrying
this marker refer to the same object in the world. Formal markers are used in
some languages, including sign language. One region in the space of the signer
(and it can be any region) is temporarily designated to stand for an object and
other signs that later refer to the same object are then produced in the same
region (Aranoff, Sandler et al. 2004). However, spoken human languages only
rarely use formal markers. For example, the Swahili marker ki- used as an
agreement marker in the phrase ki-kapu ki-kubwa ki-moja (ki.sg-basket ki-large
ki-one) (Welmers 1973) is used when the referent belongs to the class of inani-
mate objects, which contains artifacts such as baskets. The Latin marker -arum
expresses plural, feminine and genitive.
Meaningful markers lead unavoidably to a larger inventory of markers,
because a marker has to fit with the object being referred to. For example, if
gender is used as a meaning distinction, then we need at least two markers (one
for feminine and the other for masculine – and possibly a third one for neuter).
And if both objects being referred to are masculine we need markers along
another dimension, for example for number (with two markers for singular and
plural). On the other hand, a marker can introduce additional meanings on top
of the meaning supplied by words. This is the case with the Latin -arum, which
not only carries out the agreement function but also conveys that the referent
is plural and genitive. Hence meaningful markers make it possible to express
more meaning with fewer linguistic forms, which is a way to dampen form
complexity.
Studies in grammaticalization have shown that in the historical record, agree-
ment systems initially arise by reusing existing words as markers (Fuss 2005;
Givón 1976). Why would that be? We argue that this dampens learning com-
plexity. If an arbitrary label is used as a meaningful marker, then the meaning of
this marker needs to be guessed and agreed upon, and there is a higher chance
that different agents introduce different markers. But if an existing word is used,
then the hearer can make a much better guess of the possible meaning of the
marker and hence the search space for learning the marker is much smaller. The
learner only has to become aware that the word is now used not only lexically
but also grammatically.
We have conducted another experiment in the context of agreement to
explore this phenomenon. Learning efficiency is measured as follows. Let the
inventory size Ig be the total number of distinct marker constructions invented
by the whole population up till game g and Ug be the number of markers being
in used for the same window, then the learning efficiency Sg= Ug/Ig captures
how well superfluous inventions could be avoided within this interval. This
gives an indication of the learning efficiency because the fewer constructions
learners add to their inventory (as a possible hypothesis of the shared language),
the lower Sg will be. Figure 2.3 compares two strategies: one in which new
markers are just random symbols (the no-reuse strategy) and a second strat-
egy where new markers are based on existing words (the reuse strategy). We
see that the learning efficiency is considerably different: only 20 percent of the
constructions are still circulating with the reuse strategy, whereas 60 percent of
the constructions survive in the case of no-reuse (Figure 2.3). This is another
example where a particular strategy allows agents to dampen complexity, in
this case along the learning dimension.
1
0.9
0.8
learning efficiency
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 500 1000 1500 2000 2500
games
no reuse reuse
Figure 2.3 Experiment comparing the learning efficiency Sg between a strat-

egy that reuses existing words as meaningful markers and one that does not.
We see that the first strategy is much more efficient in the sense that learners
generate fewer hypotheses about the possible meanings of markers that are
then later not used.
3.5 Form Complexity

Another phenomenon that is observed quite often in language evolution is that
the form of a word gets shortened and thus its form complexity reduced. This
has also been observed with agreement markers, which are often shortened so
that the original word is no longer recognizable. We have therefore run addi-
tional experiments in which agents use this strategy, in addition to the strategies
discussed earlier (see also Beuls 2014 for a more elaborate discussion). Speak-
ers optimize articulation by leaving out the last consonant or vowel of a marker
with a certain probability ε = 0.1. Hearers are flexible enough in their parsing
of markers to recognize that a truncated form is a variant of an existing marker,
as long as it deviates for only one consonant or vowel. This maintains an ade-
quate level of communicative success and does not diminish the effectiveness
of markers to cut down combinatorial complexity and semantic ambiguity. But
how can we explain that a variant might itself become the norm and in turn
become the object of further optimizations?
Figure 2.4 shows the outcome of a computer simulation of this phonologi-
cal reduction strategy. An agreement system based on meaningful markers is
first emerging using the meaningful marker strategy. But after agents reach a
stable level of performance (in the experiment this is typically after 200 games
per agent), they occasionally introduce phonological reductions with probabil-
ity ε = 0.1 and this leads to the erosion of the original markers. Figure 2.4(a)
shows that the average marker length decreases from an average of seven to two
5
marker length
0
0 5000 10000 15000 20000 25000 30000 35000 40000 45000 50000
games
marker length
(a)
0.8
0.6
0.4
0.2
0
0 10000 20000 30000 40000 50000
-naeamo-(v-1-2) -naea-(v-1-2) -na-(v-1-2)
-naeam-(v-1-2) -nae-(v-1-2) -n-(v-1-2)
(b)
Figure 2.4 (a) Overall reduction of form complexity in a population of agents

over 50,000 games. (b) The -naeamo marker (expressing feature v-2–1) starts
off with six letters and erodes down to a single letter “-n.” Longer versions of
the original marker (“-naea” and “-nae”) are still occasionally used.
consonants and vowels, without affecting performance. There is greater vari-

ation Vg in the population because there are always different variants of the
same marker in use, but this generally does not have an impact, because agents
are able to recognize them as such, that is, variants within their own norm.
3.6 Inventory Complexity

Agents need a bigger inventory in the case of meaningful markers because a
marker can only be used for certain types of referents. The question now is: how
can this inventory be further reduced? One possibility is to ‘bleach’ the seman-
tic features of a marker. For example, it is widely attested that the markers for
gender (like masculine/feminine) are initially entirely biologically motivated
(noun phrases referring to females get feminine and to males masculine) but
once this is adopted, genders must be assigned by convention to non-animate
beings as well if one wants to avoid additional markers. For example, table
in French is assigned to be feminine. The assignment is conventional because
another language might use another gender. Indeed, in German, Tish ‘table’ is
masculine.
We have constructed an agent-based model of this as well, based on the
process of coercion: when a marker cannot be used with a given word group
because the meaning of the marker is incompatible with the referent of the
word group, then the controller1 of the word group can be coerced to become
compatible by assigning it the features of the marker. Agents use again a lat-
eral inhibition dynamics to settle on which conventionalization to adopt. Figure
2.5(b) shows the winner-take-all dynamics for a single controller. This strategy
is discussed in more detail in Beuls and Steels (2013).
Figure 2.5(a) shows the results of our simulations. If agents are allowed to
coerce controllers to be compatible with the agreement markers they already
have, then the inventory size is reduced by 73 percent, which means that fewer
marker constructions need to be consulted in parsing and production and a
smaller inventory needs to be learned.
4 Conclusion
The theory of cultural language evolution through linguistic selection
(Mufwene 2001, 2008; Steels 2012a) argues that language users are able to
come up with different strategies for building their language systems but that
they will tend to prefer those strategies that help them construct a better lan-
guage system. ‘Better’ means first of all that it satisfies their communicative
needs, that is, that there is enough expressive power to convey the meanings
that need to be expressed. Second, it means that language complexity is damped
as much as possible along the different dimensions discussed here: inventory
complexity, form complexity, processing complexity, learning complexity and
population-level complexity (i.e., language variation). Damping complexity is
needed to keep the language viable and culturally transmittable.
1 The controller is the element that determines the agreement, most often the head noun of the
noun phrase or the subject noun phrase in subject-verb agreement (Corbett 2006).
(a)
(b)
Figure 2.5 A reduction in inventory complexity from 15 to 4 markers (a) is
possible thanks to the feature coercion in controllers that are marked conven-
tionally for one of the four possible feature values (b).
This chapter did not discuss how new strategies get formed or how agents
choose themselves between different strategies based on assessment criteria
that they potentially compute themselves. Steps in this direction are shown in
other research (e.g., Bleys & Steels 2001; van Trijp 2013). Instead, we focused
on giving concrete examples of agent-based models to demonstrate that cer-
tain universal features occur in human languages because they help to dampen
complexity along various dimensions. In particular, we developed explanations

for the use of agreement markers, the use of meaningful as opposed to formal
markers, the recruitment of existing words as markers, the erosion of marker-
forms and the conventionalization of markers.
5 Acknowledgments
The authors acknowledge the following financial support: Luc Steels received
an ICREA fellowship to work at the Institut de Biologia Evolutiva (UPF-CSIC)
and a Marie Curie Integration grant to finance collaboration with the VUB AI
Lab. Katrien Beuls received a grant from the IWT (Belgium) for her Ph.D.
research at the VUB AI Lab in Brussels.
REFERENCES
Aronoff, Mark; Irit Meir; Carol Padden; and Wendy Sandler. 2004. Morphological uni-
versals and the sign language type. Yearbook of Morphology, ed. by Geert Booij
and Jaap Van Maerle, 19–39. Dordrecht: Kluwer.
Bell, Eric Temple. 1938. The iterated exponential integers. The Annals of Mathematics.
39, 539–557.
Beuls, Katrien, and Luc Steels. 2013. Agent-based models of strategies for the emer-
gence and evolution of grammatical agreement. PLoS ONE, 8(3), e58960, 03 2013.
doi: 10.1371/journal.pone.0058960.
Beuls, Katrien. Spirals in language evolution. Proceedings of the Tenth International
Conference on the Evolution of Language, Vienna, 389–390.
Bleys, Joris and Luc Steels. 2001. Linguistic selection of language strategies. Advances
in AI – LNCS, 5778, 150–157.
Corbett, Greville. 2006. Agreement. Cambridge: Cambridge University Press.
Camazine, Scott; Jean-Louis Deneubourg; N. Franks; J. Sneyd; Eric Bonabeau; and
Guy Theraulaz. 2001. Self-Organization in Biological Systems. Princeton: Prince-
ton University Press.
Dall’asta, Luca; Andrea Baronchelli; Alain Barrat; and Vittorio Loreto. 2006. Nonequi-
librium dynamics of language games on complex networks. Physical Review E,
74(3), 036105.
Delanda, Manuel. 2011. Philosophy and simulation. The emergence of synthetic rea-
soning. New York: Continuum.
De Vylder, Bart; and Karl Tuyls. 2006. How to reach linguistic consensus: A proof of
convergence for the naming game. Journal of Theoretical Biology, 242(4), 818–
831.
Fuss, Eric. 2005. The Rise of Agreement: A Formal Approach to the Syntax and Gram-
maticalization of Verbal Inflection. Amsterdam: John Benjamins.
Galantucci, Bruno. 2005. An experimental study of the emergence of human communi-
cation systems. Cognitive Science, 29(5), 737–767.
Garcia Casademont, Emília; and Luc Steels. 2014. Strategies for the emergence of first-
order constituent structure. Proceedings of the Tenth International Conference on
the Evolution of Language, Vienna, 50–57.
Givón, Talmy. 1976. Topic, pronoun and grammatical agreement. Subject and Topic, ed.
by Charles N. Li, 149–188. New York: Academic Press.
Hawkins, John A. 2005. Efficiency and Complexity in Grammars. Oxford: Oxford Uni-
versity Press.
Heine, Bernd; and Tanja Kuteva. 2007. The Genesis of Grammar. Oxford / New York:
Labov, William. 1966. 2006, Second Edition. The Social Stratification of English in New
York City. Cambridge: Cambridge University Press.
Lehmann, Christian. 1988. On the function of agreement. Agreement in Natural Lan-
guage: approaches, theories, descriptions, ed. by Michael Barlow, and Charles
Albert Ferguson, 55–65. Stanford: Center for the Study of Language and Informa-
tion, Stanford University.
Loreto, Vittorio; Andrea Baronchelli; Animesh Mukherjee; Andrea Puglisi; and
Francesca Tria. 2011. Statistical physics of language dynamics. Journal of Sta-
tistical Mechanics, 4, P04006.
Mufwene, Salikoko. 2008. Language Evolution: Contact, Competition and Change.
New York: Continuum.
Mufwene, Salikoko. 2001. Competition and selection in language evolution. Selection,
3(1), 45–56.
Pickering, Martin and Simon Garrod. 2013. An integrated theory of language compre-
hension and production. Behavioral and Brain Sciences 36, 329–392.
Selten, Reinhard and Massimo Warglien. 2007. The emergence of simple languages in
an experimental coordination game. Proc. Natl Acad. Sci. USA, 104(18), 7361–
7366.
Smith, Kenny; Simon Kirby and Henry Brighton. 2003. Iterated learning: a framework
for the emergence of language. Artificial Life, 9(4), 371–386.
Steels, Luc. 1995. A self-organizing spatial vocabulary. Artificial Life, 2(3), 319–332.
Steels, Luc. 1998. The origins of ontologies and communication conventions in multi-
agent systems. Journal of Agents and Multi-Agent Systems, 1(2), 169–194.
Steels, Luc. 2011. Modeling the cultural evolution of language. Physics of Life Reviews,
8(4), 330–356.
Steels, Luc. 2012a. Experiments in Cultural Language Evolution. Amsterdam: John
Benjamins.
Steels, Luc. 2012b. Self-organization and selection in language evolution. Experiments
in Cultural Language Evolution, ed. by Luc Steels, 1–37. Amsterdam: John Ben-
jamins.
Steels, Luc and Tony Belpaeme. 2005. Coordinating perceptually grounded categories
through language. a case study for colour. Behav. Brain Sci., 24, 469–529.
Van Trijp, Remi. 2013. Linguistic assessment criteria for explaining language change:
A case study on syncretism in German definite articles. Language Dynamics and
Change, 3(1), 105–132.
Van Trijp, Remi. 2014. Fitness landscapes in cultural language evolution: A case study
on German definite articles. Proceedings of the Tenth International Conference on
the Evolution of Language, Vienna, 334–341.
Welmers, Wim E. 1973. African Language Structures. Berkeley: University of Califor-
nia Press.
3 Complexity in Speech: Teasing Apart Culture
and Cognition
Bart de Boer
1 Introduction
Languages are more complex than is strictly necessary for their communica-
tive function. For example, the smallest repertoires of speech sounds found in
language have about a dozen different contrasting phonemes (see Table 3.1)
whereas in the sample used for the UCLA Phonological Segment Inventory
Database (UPSID451 , Maddieson, 1984; Maddieson & Precoda, 1990) the
median number of speech sounds is 29 (with a first quartile of 23 and a third
quartile of 36). This indicates that in general, languages tend to use more than
twice the number of phonemes that appears to be minimally necessary for full
language. This makes it possible to define a notion of relative complexity: one
language is more complex than another if a language user needs to learn more
in order to use the one language than the other.1 . The question then arises of why
languages differ in relative complexity and how they become more complex.
This chapter focuses on the second of these questions (as it is felt that it
is more susceptible to empirical investigation) and more specifically on how
one determines which aspects of linguistic complexity are due to cultural pro-
cesses, and which aspects are due to cognitive biases. Cultural processes that
shape language are the way language is used for communication and the way
it is transferred from generation to generation or from group to group. Func-
tional pressures, such as constraints on production and perception or informa-
tion theoretical issues, may play a role here as well. Some linguistic structures
may be easier to understand or to produce and will therefore be preserved and
transmitted better. Proponents of such a usage-based theory of language (Gold-
berg, 2003; Tomasello, 2003) explain linguistic structure from these cultural
processes, while the underlying cognitive mechanisms are generally consid-
ered less important. In other words, even if the cognitive mechanisms would be
rather different, still the same universals would be found.
1 It should be noted that in this definition complexity of languages can only be compared for a
given aspect of the language, such as the phoneme inventory size in the example.
48
Complexity in Speech: Teasing Apart Culture and Cognition 49
Table 3.1 Languages from UPSID451 (Maddieson, 1984; Maddieson &

Precoda, 1990) with inventory sizes smaller than or equal to 16 phonemes.
Although there appears to be an important influence of area or language
family, nevertheless, languages with small phoneme inventories are spoken
in different, linguistically unrelated regions of the world
Phonemes UPSID Name ISO 639–3 UPSID Family
11 Pirahã Myp South American, Paezan

11 Rotokas Roo East Papuan
13 Hawai’ian Haw Polynesian
13 Nasioi Nas East Papuan
14 Roro Rro Malayo-Polynesian
14 Taoripi Tqo Trans-New Guinea
15 Gadsup Gaj Trans-New Guinea
15 Ekari Ekg Trans-New Guinea
16 Ainu Ain Isolate
16 Gugu-Yalanji Gvn Pama-Nyungan
16 Bandjalang Bdy Pama-Nyungan
16 Yidiny Yii Pama-Nyungan
16 Dyirbal Dbl Pama-Nyungan
16 Koiari Kbk Trans-New Guinea
Other linguists do explain properties of language from language-specific

cognitive constraints. In their view, the way language is used plays a secondary
role, and the structure of language is mainly due to properties of the brains
of language users. An example of a model that explains structure of language
from very language-specific cognitive biases is the principles and parameters
model (Chomsky, 1981; Baker, 2002, 2003). In this chapter, we will not argue
for or against these positions, but rather address the issue of how one can inves-
tigate empirically what aspects of language may be due to cultural processes
and what aspects of language may be due to cognitive biases. This in turn may
help answer the question of how languages become complex: is it the result of
cultural and functional processes, or do we have cognitive biases that tend to
make language more complex?
The chapter addresses this issue from an evolutionary perspective. It adopts
Dobzhansky’s (1973) point of view that “nothing in biology makes sense,
except in the light of evolution.” What this means is that if one is interested in
the ultimate causes of biological facts, one has to investigate how and why they
have evolved. An evolutionary perspective has always been implicit in the study
of language from a cognitive point of view: the properties of the human brain
that linguists tend to focus on are the ones that are different from other animals,
50 Bart de Boer
and that are considered essential for language. This is an implicit way of
investigating which cognitive adaptations have evolved for language. Over the
last twenty years or so, the evolutionary perspective on language has become
increasingly explicit and influential (Pinker & Bloom, 1990; Hauser et al.,
2002; Fisher & Marcus, 2005; Fitch, 2005; Mufwene, 2008; Fitch, 2010).
The evolutionary perspective is doubly important for language: in order to
understand potential biological adaptations for language, one has to study why
and how they evolved biologically. In addition, if one wants to understand the
properties of a particular language, one has to understand their history, that is,
how they evolved culturally. Cultural evolution is thus responsible for language
change and usage-based emergence of structure, while biological evolution is
responsible for the cognitive constraints that help humans learn language. The
evolutionary perspective sees cultural processes and cognitive biases as two
sides of the same coin, rather than as mutually exclusive positions. However,
when observing modern language, one only sees the final product of the inter-
action between culture and cognition. It is therefore difficult to tell which factor
is responsible for observed linguistic phenomena.
This chapter therefore introduces a recently emerging experimental
paradigm – experimental cultural learning (Kirby et al., 2008; Galantucci,
2009; Scott-Phillips & Kirby, 2010). Experimental cultural learning attempts
to re-create cultural evolution in a laboratory setting, and in this way to tease
apart the effects of cultural processes and the effects of cognitive biases. It will
be illustrated how this can be applied to the question whether complexity in
phonemic systems is created by cognitive biases (feature economy) or rather
by cultural processes (self-organization for distinctiveness).
In the next section, the recent debate about the role of biologically evolved
cognitive constraints versus culturally evolved properties of human language
is described in more detail. This puts the experiments that are described in the
section after that in perspective. Finally we discuss how these experiments can
be refined and how they may be applied to other aspects of language.
2 The Controversy Between Biology and Culture

Every linguist is aware that there is both a cultural and a biological compo-
nent to language. Because there are so many different languages, and because
children who are not exposed to a language do not spontaneously develop a
language on their own,2 it is clear that languages are the products of culture.
On the other hand, because only humans (and not, chimpanzees, elephants or
dolphins, to name a few intelligent animals) learn language, it is clear that one
2 Interestingly, children not exposed to a language do appear to develop a new language if they
grow up in a group, as illustrated by the case of Nicaraguan Sign Language (Polich, 2005).
needs a human brain to learn language. As in many areas of the study of human
behavior, there is a debate about the role of nature (biology) versus nurture (cul-
ture) in linguistics. However, because the previous observations incontestably
show that both culture and biology are important this debate is about the relative
importance of both.
There are two independent questions in this debate: how specific to language
are the cognitive mechanisms that are used in language and how much of the
complexity and diversity of language is due to cultural factors versus biologi-
cal factors. These questions are independent, because whether cognitive mech-
anisms are specific to language or not does not determine whether cultural pro-
cesses are most important in shaping language. Thus Chomsky (1957, 2007)
proposes both highly language-specific cognition and an important role of these
cognitive factors in shaping language. Tomasello (2003) on the other hand pro-
poses highly general cognitive mechanism and an important role for culture.
Baker (2003) proposes highly language-specific cognitive adaptations, but an
important role for culture nevertheless. Griffiths and Kalish (2007) investigate
a model that has a very general learning mechanism, and show that the proper-
ties of the language directly reflect the properties of the learning mechanism,
thus giving less influence to cultural processes.
Even so, the position that there are highly language-specific cognitive adap-
tations tends to go together with the position that complexity in language is
mostly determined by cognitive factors, while the position that language is
based on general cognitive mechanisms often goes together with the position
that complexity in language is mostly determined by cultural processes. This
can perhaps be explained from the fact that general linguistics attempts to derive
properties of cognition from observation of language(s) (as first widely advo-
cated by Chomsky, 1957). It is then not extremely fruitful to look for evidence
for language-specific cognition in language if one does not believe that lan-
guages are determined by those cognitive properties. On the other hand, if one
does not believe in language-specific cognition, then one is also less worried
about the fact that properties of language may not be determined by cognitive
biases. Although these combinations of points of view are therefore method-
ologically understandable, they are in fact independent, and which combination
obtains in reality is an empirical question. Nonetheless, linguistics does seem
to oscillate between these two positions.
Recently, the position that there are no language-specific cognitive adapta-
tions and that languages are shaped by cultural processes appears to be gain-
ing in popularity. Evans and Levinson (2009) have investigated many proposed
universals of human languages and find exceptions for almost all the univer-
sals they investigate. They conclude that language universals are a “myth” and
therefore that they cannot reflect strict innate restrictions of the human capac-
ity for language. They do find that some patterns occur (much) more often than
52 Bart de Boer
other possible patterns, but they explain this as the result of cultural evolution
under functional pressure. In reaction to their article, Tomasello (2009) has
said, “Universal grammar is, and has been for some time, a completely empty
concept.”
To give further arguments against the possibility of evolved language-
specific cognitive adaptations, Christiansen, Chater and Reali (Christiansen &
Chater, 2008; Chater et al., 2009) have shown that non-functional biological
adaptations for language cannot become fixed through biological evolution,
because language changes too quickly through cultural evolution. Because cul-
tural evolution works much faster than biological evolution, language is a mov-
ing target. Although in biology it is possible for acquired behavior to influence
the biological makeup of an organism through the Baldwin effect (Baldwin,
1896) this requires the environment in which the behavior is acquired to be
stable for a very long time. The Baldwin effect has been proposed by lin-
guists as a way in which genetic adaptations to language can emerge (Pinker &
Bloom, 1990) but Christiansen and Chater (2008) argue that language changes
too quickly for this to work.
It must be noted that Christiansen, Chater and Reali (Christiansen & Chater,
2008; Chater et al., 2009) do allow for the possibility that properties of language
that tend to recur because they are functionally useful may be stable enough
that they influence the biological evolution of the brain. However, in the way
their work is cited, for example by Evans and Levinson (2009), it often appears
that this nuance is lost, and that it is impossible for language-specific cognitive
adaptations to evolve at all. This leads to the impression that biological evolu-
tion has not played an important role in the evolution and complexification of
language.
Interestingly, a very similar point of view is advocated by Chomsky (2007)
when writing about evolution, even though (as pointed out earlier) Chomsky
has advocated the importance of language-specific innate properties of the brain
for understanding language. His point of view is that recursion is a crucial and
language-specific property of human cognition (as pointed out in Hauser et al.,
2002). However, because there is not such a thing as partial recursion (accord-
ing to Chomsky, 2007), there is no evolutionary scenario by which recursion
could have evolved through incremental improvements. As Chomsky (at least
from his minimalist perspective) considers other aspects of language as periph-
eral, he therefore does not see an important role for biological evolution in the
emergence of language, rather he proposes that language may be the result of an
evolutionary accident or of exaptation from different functions (e.g., complex
cognition). The unexpected convergence of viewpoints of researchers starting
from very different perspectives about the lack of importance of biological evo-
lution for understanding language has been remarked upon by Berwick (2009).
Of course, many linguists have a more nuanced position, and see language as
the result of complex co-evolution between culture and biology (e.g., Mufwene
2013) and this is the position taken in this chapter as well.
In fact, discarding biologically determined and language-specific cognitive
mechanisms risks throwing the baby out with the bathwater. If there are no
language-specific cognitive mechanisms, then why would one study language
from a cognitive point of view at all? It would then be much more efficient to
investigate the general cognitive mechanisms separately, and leave language to
descriptive linguists. However, Chater et al. (2009) leave open the possibility
that functionally relevant cognitive adaptations can evolve. In addition, it may
be possible that linguistic cognition is less directly reflected in properties of
language than Evans and Levinson (2009) appear to assume, or that there are
properties of language that are universal, but that have been considered unin-
teresting by most linguists (e.g., de Boer, 2014). Therefore, it may be too early
to declare language-specific biological adaptations irrelevant.
One problem is that the search for language-specific cognitive adaptations
appears to have focused on properties of language that do not have a clear
function. This is understandable if one searches for language-specific cogni-
tive adaptations: universals of language that cannot be explained from func-
tional factors must be attributed to cognitive factors. However, this does not
mean that properties of language that have a clear function, such as a large lex-
icon, are not associated with corresponding cognitive adaptations. On the con-
trary, Christiansen, Chater and Reali’s (Christiansen & Chater, 2008; Chater
et al., 2009) work appears to indicate that these are the only aspects of lan-
guage that are stable enough over time to influence biological adaptation. And
indeed, coming back to the example of large lexicons, it would seem plausible
that humans have adaptations to learning large lexicons rapidly. However, the
fact that we need to look for properties of language that are functional, makes
it even more difficult to tell the effects of cultural and biological evolution
apart.
Another problem is that the interaction between cultural evolution and bio-
logical evolution makes language one of the most complex phenomena in biol-
ogy (Steels, 2000; Smith et al., 2003; Beckner et al., 2009). In fact, in order to
understand language completely, one must take into account that language is
influenced by phenomena on three time scales (Kirby & Hurford, 1997). The
shortest time scale is that of individual language learning, which takes place
on the time scale of a decade. Then there is the time scale of cultural evolution
and language change, which takes place on a scale ranging from decades to
millennia. Finally, the slowest time scale is that of biological evolution, which
takes place on time scales of the order of thousands to tens of thousands of
years. These three processes are not independent, but influence each other, as
illustrated in Figure 3.1. This means that it can be hard to tell apart whether
properties of language are due to functional constraints, cultural evolution or
54 Bart de Boer
Stabilizes or destabilizes
targets for
A Phylogeny Biological evolution B adaptation
Cultural
Glossogeny Cultural evolution Sets boundary evolution
conditions Keeps
language
Ontogeny Learning Biological learnable
Determines
Good communication cultural
evolution helps survival fitness
Determines
biases and
mechanisms Language
learning
Synaptic weights
Language in the brain
Language and genes are the only

physical correlates of
these processes
genes
Figure 3.1 The different time scales in language and their interactions. Panel
A illustrates the time scales, and the terminology coined by Kirby and Hurford
(1997). Panel B illustrates the effects the different processes have on each
other.
biological adaptations. This is one of the root causes of the enormous disagree-
ment about these issues among linguists.
The newly emerging experimental paradigm for studying cultural evolution
in a laboratory setting may help to address these issues empirically. This may
help re-interpret linguistic data, which by its very nature is always the result of
a long process of cultural evolution under biological constraints. Even in the
very few cases where emergence of a new language could be observed when it
happened (Senghas et al., 2004; Polich, 2005; Sandler et al., 2005), the situation
was insufficiently controlled to be able to draw general conclusions. The next
section will introduce this new paradigm and illustrate it with a case study.
3 Culture and Cognition in Speech

In order to illustrate how effects of cultural processes and cognitive biases can
be disentangled, an experiment on the emergence of structure in acoustic sig-
nals (Verhoef & de Boer, 2011; Verhoef et al., 2011a; Verhoef et al., 2011b)
and its theoretical background will be presented in some detail.
3.1 Cultural and Cognitive Explanations of Speech Sounds

Human language is characterized by the use of a limited number of building
blocks in order to produce an unlimited number of utterances. Interestingly,
the inventories of building blocks that are used are not random. The building
blocks appear to be constructed using a set of basic distinctive features (Jakob-
son & Halle, 1956). For example, the consonants [b] and [p] differ in the feature
voicing: [b] is voiced and [p] is voiceless, but they are the same for other
features, for example they are both labial consonants. The vowels [i] and [u]
differ in the features backness and rounding where [i] is front and unrounded,
while [u] is back and rounded. They are the same however for the feature
height, being both high vowels. Some building blocks and some feature dis-
tinctions are very common in human languages, others are very rare. Rare fea-
tures and speech sounds are called “marked” by linguists. Ever since the con-
ception of the idea of distinctive features, there has been a debate about whether
this is due to cognitive constraints or functional factors.
Jakobson and Halle (1956) and Chomsky and Halle (1968) among others
have argued for the position that distinctive features reflect cognitive biases.
They use observations from language acquisition and aphasia to argue that there
are learning biases that favor certain speech sounds and features over others.
Other researchers argue that there are functional reasons why sound systems are
the way they are. Liljencrants and Lindblom (1972) show that optimization of
distinctiveness leads to realistic vowel systems for small vowel system sizes.
Stevens (1972) argues that certain speech sounds are inherently more robust
and less subject to small errors in articulation than others, and are therefore pre-
ferred in language. More recent work using agent-based computer simulation
has shown that optimization for distinctiveness in vowel systems can emerge
through self-organization in a population (Berrah et al., 1996; de Boer, 2000).
Apparently, innate distinctive features are not necessary to explain small inven-
tories of speech sounds. However, because the models work less well for larger
inventories, other factors must also play a role.
One factor appears to be feature economy (Clements, 2003, who also gives
a historical overview of the concept) or the maximal use of available distinc-
tive features (Ohala, 1980). If one analyzes the sound systems of many lan-
guages, one finds that the features that distinguish speech sounds are gener-
ally not used to distinguish just two speech sounds in the repertoire, but whole
series of sounds. For example when nasalization is used in vowel systems, it
is generally not used for only one vowel, but for a series of vowels. Similarly,
consonants tend to occur in series: when voiced and voiceless plosives are used,
generally these occur for all places of articulation used in the plosive system.
Examples of the vowel system of French and the plosive system of Hindi are
given in Figure 3.2. Thus distinctive features tend to be used in a maximally
economic way to create large systems of speech sounds.
It is possible that feature economy is the result of cultural processes, but
it is equally possible that it is the result of a cognitive bias. Lending support
to the interpretation of feature economy as the result of self-organization are
the computer simulation results of Berrah and Laboissière (1999). These show
emergent economic use of features in a population of simulated agents that have
to communicate as effectively as possible. There is no explicit mechanism for
economic use of features, and the authors therefore conclude that it is an emer-
gent effect of functional constraints and cultural evolution. In a very different
56 Bart de Boer
Figure 3.2 French vowels and Hindi plosives illustrating feature economy.
French vowels have different levels of height, but at most levels of height
three vowels occur. In addition, all low vowels are nasalized. Hindi plosives
occur at three places of articulation, but at each place of articulation, a plosive
can be voiced or voiceless and aspirated or unaspirated.
line of work however, Maye et al. (2008) show that infants appear to general-
ize features from familiar to unfamiliar contrasts. This appears to indicate that
infants do have cognitive mechanisms for extracting distinctive features from
speech data and for generalizing these features across different speech sounds.
Because Maye et al. did not investigate feature economy, one should be care-
ful with extending their results to this area. Nevertheless, the mechanisms they
uncover are an important prerequisite for an explanation of feature economy
based on cognitive factors. There is therefore evidence that both cultural and
cognitive processes may play a role in the explanation of feature economy. The
challenge is therefore to experimentally tease apart the effect of culture and
cognition.
3.2 Experimental Cultural Learning

The experimental paradigm of iterated (cultural) learning that has been devel-
oped over the last few years (Galantucci, 2005; Griffiths et al., 2008; Kirby
et al., 2008; Smith & Kirby, 2008; Scott-Phillips & Kirby, 2010) may help to
unravel the relation between culture and cognitive biases in language. Iterated
learning is the repeated learning of culturally transmitted information in a group
of agents, and usually focuses on human learning of language. The idea of
studying the effect of repeated learning in a group was originally developed by
computer modelers of the evolution of language. Two variants were proposed:
Steels (1995, 1997, 1998, 2011) initially focused on interactions within a group
of agents that in this way negotiate a shared communication system. This kind
of cultural learning is alternatively called a language game, social coordination
or horizontal transmission. Kirby and Hurford (Hurford, 1989; Kirby & Hur-
ford, 1997; Kirby, 1999) initially focused on transfer of linguistic information
across generations, which usually consisted of one agent. In their model a par-
ent agent produces linguistic utterances that are learned by a child agent. After
Vertical Transmission
Horizontal Transmission
Figure 3.3 Schematic illustration of vertical transmission (top) and horizontal

transmission (bottom). In vertical transmission, participants are either speaker
or hearer sequentially, and only have one role at the same time. In horizon-
tal transmission, each participant plays the role of both speaker and hearer
simultaneously.
a while the parent is removed, the child becomes the new parent, and a new
child is added to the population. This approach is alternatively called3 “iterated
learning,” “vertical transfer” or a “diffusion chain.” Both types are illustrated
in Figure 3.3. After initial successes with these computer models, an increasing
desire to link the simulations to real-world phenomena led to recreating them
in a laboratory setting with human participants (Galantucci, 2005).
In such experimental iterated learning tasks, participants have to learn sets of
signals produced by other participants in the experiment. The fact that stimuli
are produced by other participants in the experiments instead of coming from
a carefully set crafted by the experimenter is an important distinction between
experimental iterated learning and more classical artificial language learning
experiments. In most experimental iterated learning experiments, participants
are required not just to learn and (passively) recognize a set of signals, but also
to reproduce these signals.
Many variants are possible: experiments can either implement horizontal or
vertical transmission (cf. Garrod et al., 2010 for a critical comparison). Sig-
nals can be used communicatively or not and they can have meaning or not.
Experiments can be about strings of discrete symbols (letters) and thus address
3 As experimental iterated learning is a newly emerging paradigm, there is confusion about termi-
nology. Kirby et al. prefer to call all these approaches iterated learning and distinguish between
diffusion chains (studying vertical transmission) and social coordination (studying horizontal
transmission). Other authors equate iterated learning with vertical transmission and call the gen-
eral approach cultural learning.
58 Bart de Boer
questions related to syntax or about continuously variable signals and thus

address questions about phonology and phonotactics. These and other variants
are reviewed in several recent papers (Kirby et al., 2008; Smith et al., 2008;
Galantucci, 2009; Scott-Phillips & Kirby, 2010).
The results from iterated learning experiments can be analyzed in two ways:
by looking at either the system of signals that emerges from the experiments
or by looking at the ways in which participants solve the problem of learning
and reproducing sets of signals. Analyzing the system of signals that emerges
is comparable to analyzing existing real languages: they are the product of cul-
tural evolution, language-specific cognitive processes, and general cognitive
processes combined. The ways in which participants learn and generalize utter-
ances, however, make it possible to directly observe the individual behaviors
that shape cultural processes. By manipulating the experimental setup, or by
complementing iterated learning experiments with specially crafted artificial
language learning experiments, it is possible to zero in on whether these behav-
iors are driven by functional factors or by cognitive factors and whether they
are driven by language-specific or domain general cognition. These methods
will be illustrated with a small case study discussed next.
3.3 Iterated Learning of Speech

The work presented here is from Verhoef and de Boer (2011) and Verhoef
et al. (2011a, 2011b, 2014). These iterated learning experiments investigate the
emergence of combinatorial structure in systems of acoustic signals. Combina-
torial structure is the ability to construct an unlimited number of utterances from
a limited set of speech sounds. In human languages this happens according to a
set of learned rules that are generally referred to as the language’s phonotactic
structure. As explained earlier, some researchers have claimed that combinato-
rial structure is the result of cultural processes, while others have proposed that
it is due to cognitive biases related to language learning and language use.
The hypothesis that structure is (mainly) due to cultural processes makes rad-
ically different predictions from the hypothesis that structure is (mainly) due
to cognitive processes. A mainly cultural process, as modeled for example in
de Boer (2000) and Zuidema and de Boer (2009) results in slow changes that
spread in the language community because the systems containing the changed
sounds are better at avoiding confusion. A mainly cognitive process, such as
feature economy discussed by Clements (2003) and maximal use of available
distinctive features discussed by Ohala (1980), results in rapid emergence of
(sets of) new building blocks and re-use of these building blocks in the utter-
ances that make up the language. Verhoef et al.’s (Verhoef & de Boer, 2011;
Verhoef et al., 2011a; Verhoef et al., 2011b; Verhoef et al., 2014) experiments
directly investigate this issue.
These experiments investigate vertical transmission of small systems of ini-

tially unstructured signals. Participants are exposed to a set of 12 different sig-
nals and are then asked to reproduce them four times. Their last set of repro-
ductions is used as training data for the next participant in the chain. If one of
the signals that the participants produce is too close to one they have already
made, this is flagged, and they are asked to retry to produce a signal. It should be
noted that no meaning is associated with the signals. In order to avoid interfer-
ence from preexisting linguistic knowledge as much as possible, the signals are
not spoken sounds, but are short whistles produced by means of a slide whis-
tle. Because most participants are not familiar with the use of a slide whistle,
they can practice for a short period before the actual experiment. Four different
chains of ten participants each have been investigated so far.
It turns out that systems of signals become significantly more learnable over
the generations (as measured by Page’s trend test on the recall error of the par-
ticipants, Verhoef & de Boer, 2011). Also, the signals in the systems become
more similar (as measured by Page’s trend test on the average distance between
a signal and its nearest neighbor in the set, Verhoef & de Boer, 2011). In addi-
tion, it appears that the entropy of the building blocks as used in the set of
signals also decreases over the generations (Verhoef et al., to appear 2012).
The first observation would be expected for both the cultural and cognitive
hypotheses: more distinctiveness predicted by the cultural hypothesis would
make it easier for participants to keep signals apart, whereas more structured
sets of signals predicted by the cognitive hypothesis would be more easily
learned. The second observation is in contradiction with the cultural hypothe-
sis, which predicts emergence of more distinctive signals, and therefore a larger
distance between signals. The cognitive hypothesis does not really make a pre-
diction here, although increased re-use of building blocks could make signals
constructed out of these building blocks more similar. The third observation
is predicted by the cognitive hypothesis. However, it has been shown that cul-
tural processes can also result in sets of signals that appear to reuse building
blocks (de Boer & Zuidema, 2010). Taken together, the observations do appear
to indicate a problem for the cultural hypothesis, but it is necessary to look
more closely at what happens during learning and reproduction in order to find
out which hypothesis (if any) can be rejected.
When we look at individual reproductions of learned stimuli, it is much
clearer that this involves a process of re-use and generalization of building
blocks. Participants have a hard time learning the initial, unstructured set. How-
ever, when reproducing signals, they are forced to produce 12 distinct signals.
This process results in subconscious generalization and re-use of observed pat-
terns. Three processes by which participants do this are illustrated in Figure
3.4. The first process is the combination of (parts of) one signal into a new sig-
nal. Verhoef and de Boer call this “borrowing.” The second process is reversing
60 Bart de Boer
1s 1 octave
generation n generation n + 1 n n+1 m m+1

“borrowing” “mirroring” “duplication”
Figure 3.4 Three processes through which experimental participants create

new signals in their imperfect reproductions of a learned set of signals. In the
left panel, borrowing is illustrated through the recombination of one signal of
the learned set with the second half of another signal from the learned set. In
mirroring, a signal of the learned set appears both in its original form and in
a temporally mirrored form. In duplication, a signal from the learned set is
analyzed as consisting of two elements. One of these elements is duplicated
in reproduction. Signals were redrawn from (Verhoef & de Boer, 2011).
the pitch pattern of a signal and including this as a new signal in the new set
of signals. Verhoef and de Boer call this “mirroring.” The third process creates
new signals by repeating elements from existing signals. Verhoef and de Boer
call this “duplication.” These processes are clearly indicative of (conscious or
subconscious) manipulation of elements of learned signals and must therefore
be based on cognitive processes. They are very different from the small and
random shifts that are used in models of cultural processes, such as the simula-
tions by Zuidema and de Boer (Zuidema & de Boer, 2009; de Boer & Zuidema,
2010). It must therefore be concluded that in this experiment, cognitive pro-
cesses play a leading role in the emergence of structure.
This does not mean that cultural processes do not play a role whatsoever. In
fact, structure emerges only gradually over the experimental generations, indi-
cating that structure is amplified by cultural transmission of the system of sig-
nals. In addition, the different chains of transmission from participant to partic-
ipant appear to converge on differently structured sets of signals. It is currently
tested experimentally whether human listeners can identify this culturally trans-
mitted structure (Verhoef, 2013, ch. 5). In these experiments, participants (dif-
ferent from the ones who participated in the original experiment) are exposed
to signals that emerged from a particular chain. Then they are tested with a mix-
ture of signals that emerged in the same chain they were trained with (but that
they have not heard before) and signals that emerged from different chains. It
is measured whether participants can reliably distinguish between signals that
emerged from their training chain and those that did not. These experiments
appear to indicate that chains of culturally transmitted signals converge to dis-
tinguishable sets of signals (thus stressing that cultural effects are important in
addition to the cognitive processes described previously).
4 Discussion
This chapter has attempted to illustrate a possible method, with origins in the
study of language evolution, of teasing apart the role of cultural and cognitive
processes in the emergence of complexity in language. It is a central question
in linguistics which aspects of language are due to cultural processes, which
aspects are due to language-specific cognitive processes and which aspects are
due to general cognitive processes. In the case of general cognitive mechanisms,
the evolutionary pressures that shaped these mechanisms have not been specific
to language, whereas in the case of language-specific cognitive mechanisms,
there must have been evolutionary pressures specific to language. It is difficult
to assess the role of these different mechanisms, as existing languages are the
product of continuous interaction between the three mechanisms.
The paradigm of experimental cultural (or iterated) learning was developed
by researchers of the evolution of language in order to study the effects of
cultural processes in a laboratory setting. It consists of repeated interactions
between participants in which language-like stimuli are learned, reproduced
and (possibly) invented. In this way it can be studied directly how language-
like systems change due to interactions between users of the system. Human
behavior can then be compared with predictions both of models that are based
on cultural processes and of those that are based on cognitive processes.
This approach has been illustrated for the case of emergence of complex
combinatorial structure in acoustic signals. It was shown that human behav-
ior corresponds more closely to behavior predicted by models based on cog-
nitive processes (comparable to feature economy or maximal use of available
distinctive features) than by models based on purely cultural processes (self-
organization under pressure of distinctiveness). However, as the different learn-
ing chains converge to sets of signals that are structured according to different
rules, cultural processes must play an important role as well. If only cogni-
tive processes would play a role, then one would expect the learning chains
to always converge to very similar sets of utterances. These experiments both
illustrate the interaction of cognitive and cultural processes, as well as the abil-
ity to tease them apart.
The experiments conform to observations of systems of speech sounds in
existing languages and the way they change: people have a tendency to apply
the same processes (duplication and borrowing in the experiments; phonolog-
ical processes in real languages) in different utterances and to re-use existing
building blocks. In addition, humans appear to apply similar processes to the
building blocks themselves. In the experiments this is observed in the case of
mirroring of building blocks, and in real languages it happens when one feature
spreads over different phonemes or when utterances are re-analyzed in terms of
new sets of building blocks. In this way, whole sets of new speech sounds can
62 Bart de Boer
appear (or disappear) in a language in a single operation. We can therefore ten-

tatively conclude that complexity of phonological systems is due to cognitive
mechanisms that re-use and generalize building blocks.
Due to the cognitive biases to learn and generalize complex sets of speech
sounds, change of (systems of) speech sounds is not a completely random pro-
cess. This does not alter the fact that language change can be viewed as a pro-
cess of cultural evolution, just that the operations that introduce variation are
not as random as those of biological evolution. This is a well-known potential
difference between cultural evolution and biological evolution (e.g., Dennett,
2006) but it is one that is sometimes overlooked by linguists applying evolution-
ary theory to language change (e.g., Blevins, 2004). Experimental investigation
of language creation, change and complexification as illustrated in this chapter,
may help to clarify how variation is introduced and how complexity is created
in language.
A discrepancy between the experiments and real language is that in the
experiments rather radical changes happen quickly and frequently, whereas
real languages are more stable over generations. This is most likely due to a
much stronger conformity bias in the case of real language: language users are
exposed to a much larger set of linguistic stimuli, and therefore learn much
more accurately than in the experiments. Moreover, language users cannot
change their language too radically as otherwise they would be unable to com-
municate with others in the community.
The work described here is just the beginning of the experimental investi-
gation of how culture and cognition interact in the emergence of phonologi-
cal complexity, and a rather rough beginning at that. The work does form part
of a movement to apply similar techniques to different aspects of language,
most notably syntax and morphology, but also pragmatics and semantics (Kirby
et al., 2008; Galantucci, 2009; Scott-Phillips & Kirby, 2010). There are a large
number of issues that still need to be addressed, including: What is the effect of
adding meaning to the tasks? What is the effect of interaction between speak-
ers? Which phenomena are due to learning and reproduction and which phe-
nomena are due to invention of utterances? Which processes are due to general
cognition and which processes are due to language-specific cognition? Finally,
there is the issue of the extent to which it is possible to exclude all influence
from the linguistic knowledge that the participants of the experiments already
have. All these questions notwithstanding, the approach of experimental cul-
tural learning promises to be a useful new way to investigate old questions.
Acknowledgments
This work was financed by the NWO vidi grant 276–75–007 “modelling the
evolution of speech” and the ERC starting grant project 283435, “ABACUS.”
The author wishes to thank Tessa Verhoef and the editors for comments on the
manuscript.
REFERENCES
Baker, Mark C. (2002). The atoms of language. Oxford: Oxofrd University Press.
Baker, Mark C. (2003). Linguistic differences and language design. Trends in Cognitive
Sciences, 7(8), 349–353.
Baldwin, James M. (1896). A new factor in evolution. The American Naturalist, 30,
441–451, 536–553.
Beckner, Clay, Blythe, Richard, Bybee, Joan, Christiansen, Morten H., Croft, William,
Ellis, Nick C., Holland, John, Ke, Jinyun, Larsen-Freeman, Diane & Schoenemann,
Tom (2009). Language is a complex adaptive system: Position paper. Language
Learning, 59(Suppl. 1), 1–26.
Berrah, Ahmed-Redah, Glotin, Hervé, Laboissière, Rafael, Bessière, Pierre, & Boë,
Louis-Jean. (1996). From form to formation of phonetic structures: An evolution-
ary computing perspective. In Terry Fogarty & Gilles Venturini (Eds.), ICML ‘96
workshop on evolutionary computing and machine learning, bari (pp. 23–29).
Berrah, Ahmed-Redah, & Laboissière, Rafael (1999). Species: An evolutionary model
for the emergence of phonetic structures in an artificial society of speech agents.
In Dario Floreano, Jean-Daniel Nicoud & Francesco Mondada (Eds.), Advances
in artificial life, lecture notes in artificial intelligence (Vol. 1674, pp. 674–678).
Berlin: Springer.
Berwick, Robert C. (2009). What genes can’t learn about language. Proceedings of the
National Academy of Sciences, 106(6), 1685–1686.
Blevins, Juliet (2004). Evolutionary phonology. Cambridge: Cambridge University
Press.
Chater, Nick, Reali, Florencia, & Christiansen, Morten H. (2009). Restrictions on bio-
logical adaptation in language evolution. Proceedings of the National Academy of
Sciences, 106(4), 1015–1020.
Chomsky, Noam (1957). Syntactic structures (second, 2002 ed.). Berlin: Mouton de
Gruyter.
Chomsky, Noam (1981). Lectures on government and binding. Dordrecht: Foris Publi-
cations.
Chomsky, Noam (2007). Of minds and language. Biolinguistics, 1(1), 9–27.
Chomsky, Noam, & Halle, Morris (1968). The sound pattern of English. Cambridge,
MA: MIT Press.
Christiansen, Morten H., & Chater, Nick (2008). Language as shaped by the brain.
Behavioral and Brain Sciences, 31, 489–558.
Clements, George N. (2003). Feature economy in sound systems. Phonology, 20, 287–
333.
de Boer, Bart (2000). Self organization in vowel systems. Journal of Phonetics, 28(4),
441–465.
de Boer, Bart (2014). Biological Adaptation to Cultural Traits. In: Cartmill, Erica A.,
Roberts, Seàn, Lyn, Heidi & Cornish, Hannah (Eds.), The Evolution of Language,
Proceedings of the 10th International Conference (EVOLANG10) (pp. 419–420),
New Jersey: World Scientific.
64 Bart de Boer
de Boer, Bart, & Zuidema, Willem (2010). An agent model of combinatorial phonology.
Adaptive Behavior, 18(2), 141–154.
Dennett, Daniel (2006). From typo to thinko: When evolution graduated to semantic
norms. In Stephen C. Levinson & Pierre Jaisson (Eds.), Evolution and culture
(pp. 133–145). Cambridge, MA: MIT Press.
Dobzhansky, Theodoszius (1973). Nothing in biology makes sense, except in the light
of evolution. The American Biology Teacher, 35, 125–129.
Evans, Nick, & Levinson, Stephen C. (2009). The myth of language universals: Lan-
guage diversity and its importance for cognitive science. Behavioral and Brain
Sciences, 32, 429–492.
Fisher, Simon E., & Marcus, Garry F. (2005). The eloquent ape: Genes, brains and the
evolution of language. Nature Reviews Genetics, 7, 9–20.
Fitch, W. Tecumseh (2005). The evolution of language: A comparative review. Biology
and Philosophy, 20, 193–230.
Fitch, W. Tecumseh (2010). The evolution of language. Cambridge: Cambridge Univer-
sity Press.
Galantucci, Bruno (2005). An experimental study of the emergence of human commu-
nication. Cognitive Science, 29, 737–767.
Galantucci, Bruno (2009). Experimental semiotics: A new approach for studying com-
munication as a form of joint action. Topics in Cognitive Science, 1, 393–410.
Garrod, Simon Fay, Nicolas, Rogers, Shane, Walker, Bradley, & Swoboda, Nik (2010).
Can iterated learning explain the emergence of graphical symbols? Interaction
Studies, 11(1), 33–50.
Goldberg, Adele E. (2003). Constructions: A new theoretical approach to language.
Trends in cognitive sciences, 7(5), 219–224.
Griffiths, Thomas L., & Kalish, Michael L. (2007). Language evolution by iterated learn-
ing with Bayesian agents. Cognitive Science, 31(3), 441–480.
Griffiths, Thomas L., Kalish, Michael L., & Lewandowsky, Stephan (2008). Theoretical
and empirical evidence for the impact of inductive biases on cultural evolution.
Philosophical Transactions of the Royal Society of London B, 363(1509), 3503–
3514.
Hauser, Marc D., Chomsky, Noam, & Fitch, W. Tecumseh (2002). The faculty of lan-
guage: What is it, who has it, and how did it evolve? Science, 298, 1569–1579.
Hurford, James R. (1989). Biological evolution of the saussurean sign as a component
of the language acquisition device. Lingua, 77(2), 187–222.
Jakobson, Roman, & Halle, Morris (1956). Fundamentals of language. The Hague:
Mouton & Co.
Kirby, Simon (1999). Function, selection and innateness: The emergence of language
universals. Oxford: Oxford University Press.
Kirby, Simon, Cornish, Hannah, & Smith, Kenny (2008). Cumulative cultural evolution
in the laboratory: An experimental approach to the origins of structure in human
language. Proceedings of the National Academy of Sciences, 105(31), 10681–
10686.
Kirby, Simon, & Hurford, James R. (1997). Learning, culture and evolution in the origin
of linguistic constraints. In Phil Husbands & Inman Harvey (Eds.), Fourth Euro-
pean conference on artificial life (pp. 493–502). Cambridge, MA: MIT Press.
Liljencrants, J., & Lindblom, Björn (1972). Numerical simulations of vowel quality
systems. Language, 48, 839–862.
Maddieson, Ian (1984). Patterns of sounds. Cambridge: Cambridge University Press.

Maddieson, Ian, & Precoda, Kristin (1990). Updating UPSID. UCLA Working Papers
in Phonetics, 74, 104–111.
Maye, Jessica, Weiss, Daniel J., & Aslin, Richard N. (2008). Statistical phonetic learning
in infants: Facilitation and feature generalization. Developmental Science, 11(1),
122–134.
Mufwene, Salikoko S. (2008). Language evolution: Contact, competition and change.
London: Continuum International Publishing Group.
Mufwene, Salikoko S. (2013) Language as technology. In: Lohndal, Terje (Ed.) In
Search of Universal Grammar: From Old Norse to Zoque. (pp. 327–358). Ams-
terdam / Philadelphia, PA: John Benjamins Publishing Company.
Ohala, John J. (1980). Moderator’s introduction to symposium on phonetic universals
in phonological systems and their explanation. In Proceedings of ICPHS IX, vol 3
(pp. 181–185). Copenhagen: Institute of Phonetics.
Pinker, Steven, & Bloom, Paul (1990). Natural language and natural selection. Behav-
ioral and Brain Sciences, 13(4), 707–784.
Polich, Laura (2005). The emergence of the deaf community in Nicaragua. Washington,
DC: Gallaudet.
Sandler, Wendy, Meir, Irit, Padden, Carol, & Aronoff, Marc (2005). The emergence
of grammar: Systematic structure in a new language. Proceedings of the National
Academy of Sciences, 102(7), 2661–2665.
Scott-Phillips, Thom C., & Kirby, Simon (2010). Language evolution in the laboratory.
Trends in Cognitive Sciences, 14, 411–417.
Senghas, Ann, Kita, Sotaro, & Özyürek, Aslı (2004). Children creating core properties
of language: Evidence from an emerging sign language in Nicaragua. Science, 305,
1779–1782.
Smith, Kenny, Brighton, Henry, & Kirby, Simon (2003). Complex systems in language
evolution: The cultural emergence of compositional structure. Advances in Com-
plex Systems, 6(4), 1–22.
Smith, Kenny, Kalish, Michael L., Griffiths, Thomas L., & Lewandowsky, Stephan
(2008). Introduction. Cultural transmission and the evolution of human behaviour.
Philosophical Transactions of the Royal Society of London, 363(1509), 3469–
3476.
Smith, Kenny, & Kirby, Simon (2008). Cultural evolution: Implications for understand-
ing the human language faculty and its evolution. Philosophical Transactions of
the Royal Society of London B, 363, 3591–3603.
Steels, Luc (1995). A self-organizing spatial vocabulary. Artificial Life, 2(3), 319–332.
Steels, Luc (1997). The synthetic modelling of language origins. Evolution of Commu-
nication, 1(1), 1–34.
Steels, Luc (1998). Synthesising the origins of language and meaning using co-
evolution, self-organisation and level formation. In James R. Hurford, Michael
Studdert-Kennedy & Chris Knight (Eds.), Approaches to the Evolution of Lan-
guage (pp. 384–404). Cambridge: Cambridge University Press.
Steels, Luc (2000). Language as a complex adaptive system. In M. Schoenauer, K. Deb,
G. Rudolph, X. Yao, E. Lutton, J. Merelo & H.-P. Schwefel (Eds.), Parallel problem
solving form nature, PPSN VI (Vol. 1917, pp. 17–26). Berlin: Springer.
Steels, Luc (2011). Modeling the cultural evolution of language. Physics of Life Reviews,
8(4), 339–356.
66 Bart de Boer
Stevens, Kenneth N. (1972). The quantal nature of speech: Evidence from articulatory-
acoustic data. In E. E. J. David & P. B. Denes (Eds.), Human communication: A
unified view (pp. 51–66). New York: McGraw-Hill.
Tomasello, Michael (2003). Constructing a language: A usage-based theory of language
acquisition. Cambridge, MA: Harvard University Press.
Tomasello, Michael (2009). Universal grammar is dead. Behavioral and Brain Sciences,
32(5), 470–471.
Verhoef, Tessa (2013). Efficient coding in speech sounds: Cultural evolution and
the emergence of structure in artificial languages, Ph.D. thesis, University of
Amsterdam.
Verhoef, Tessa, & de Boer, Bart (2011). Cultural emergence of feature economy in an
artificial whistled language. Paper presented at the International Conference on
Phonetic Sciences, Hong Kong.
Verhoef, Tessa, de Boer, Bart, Del Giudice, Alex, Padden, Carol, & Kirby, Simon
(2011a). Cultural evolution of combinatorial structure in ongoing artificial speech
learning experiments. CRL Technical Report UCSD, 23(1), 3–11.
Verhoef, Tessa, de Boer, Bart, & Kirby, Simon (2012). Holistic or synthetic protolan-
guage: Evidence from iterated learning of whistled signals. In The evolution of lan-
guage: Proceedings of the 8th international conference (EVOLANG8) (368–375).
Hackensack, NJ: World Scientific.
Verhoef, Tessa, Kirby, Simon, & Padden, Carol (2011b). Cultural emergence of combi-
natorial structure in an artificial whistled language. In L. Carlson, C. Hölscher & T.
Shipley (Eds.), Proceedings of the 33rd annual conference of the cognitive science
society (pp. 483–488). Austin, TX: Cognitive Science Society.
Verhoef, Tessa, Kirby, Simon & de Boer, Bart (2014). Emergence of combinato-
rial structure and economy through iterated learning, Journal of Phonetics 43C,
57–68.
Zuidema, Willem, & de Boer, Bart (2009). The evolution of combinatorial phonology.
Journal of Phonetics, 37(2), 125–144.
4 A Complex-Adaptive-Systems Approach to the
Evolution of Language and the Brain
P. Thomas Schoenemann
1 Introduction
Language has arguably been as important to our species’ evolutionary success
as has any other behavior. Understanding how language evolved is therefore
one of the most interesting questions in evolutionary biology. Part of this story
involves understanding the evolutionary changes in our biology, particularly
our brain. However, these changes cannot themselves be understood indepen-
dent of the behavioral effects they made possible. The complexity of our inner
mental world – what will here be referred to as conceptual complexity – is one
critical result of the evolution of our brain, and it will be argued that this has in
turn led to the evolution of language structure via cultural mechanisms (many of
which remain opaque and hidden from our conscious awareness). From this per-
spective, the complexity of language is the result of the evolution of complexity
in brain circuits underlying our conceptual awareness. However, because indi-
vidual brains mature in the context of an intensely interactive social existence –
one that is typical of primates generally but is taken to an unprecedented level
among modern humans – cultural evolution of language has itself contributed
to a richer conceptual world. This in turn has produced evolutionary pressures
to enhance brain circuits primed to learn such complexity. The dynamics of lan-
guage evolution, involving brain/behavior co-evolution in this way, make it a
prime example of what have been called “complex adaptive systems” (Beckner
et al. 2009).
Complex adaptive systems are phenomena that result from the interaction of
many individual components as they adapt and learn from each other (Holland
2006). They have in common the emergence of interesting higher-level patterns
that do not initially appear obvious given the behavior of individual actors in
the system. Classic examples of complex adaptive systems are ant colonies, in
which the behavior of the colony as a whole is highly intelligent, flexible and
adaptive, even though the colony itself is composed exclusively of individuals
with very simple, rigidly stereotyped behavior (Holland 1998). Those that take
the view that “the mind is what the brain does” are essentially making the same
kind of argument: the action of each neuron is very simple, and no single neuron
67
68 P. Thomas Schoenemann
“knows” anything about the mind, yet the sum of their actions is the mind.
From a complex adaptive systems perspective, it is a mistake to assume that the
patterns and behaviors of the whole system are in each individual agent. Instead,
the patterns emerge from the sum of actions of sets of adaptively interacting
agents and are therefore more properly understood as existing between agents.
The argument here will be that language is similarly not the result of neural
circuits innately coded in the mind of each individual, but instead that these
patterns are the result of complex interactions at three levels: biological evo-
lution, cultural evolution, and the ontogenetic development of individuals. The
complexity of language is the result of a biocultural evolutionary process, or
one that is neither exclusively biological nor exclusively cultural (Christiansen
& Chater 2008; Evans & Levinson 2009). In particular, it will be argued that
the patterns of language use – grammar and, more specifically, syntax – are
more properly understood as being emergent characteristics of increasing con-
ceptual complexity of individuals who are embedded in an intensely socially
interactive (i.e., communicative) existence (Savage-Rumbaugh & Rumbaugh
1993; Schoenemann 1999). This intense social interactivity is a legacy of
our being primates and long predates the origin of our species – let alone
language.
The complexity of language has been studied from a variety of perspectives
in linguistics. One dominant view, identified with Noam Chomsky and follow-
ers (e.g., Chomsky 1972; Jackendoff 2002; Pinker & Jackendoff 2005), has
been that linguistic complexity is best understood as the result of some formal
(logical) model, with the assumption being that a correctly described model will
be innately instantiated in the brain in some way. There is no doubt that regular-
ities exist in specific languages. Whether there are truly any language universals
across languages, however, has been called into question (Evans & Levinson
2009). The complexity and fuzziness of the phonological systems within and
between languages is so great that even the basic assumption that speech can
appropriately be divided into discrete packages (called phonemes) has been
called into question (Port 2010; Port & Leary 2005). Furthermore, those who
believe in Universal Grammar do not agree about exactly what formal model
best describes language. Chomsky’s own models have notoriously changed a
great deal over the past half-century, with the latest incarnation emphasizing a
so-called “minimalist” view of the underlying cognitive mechanisms (Chom-
sky 1995). However, the minimalist program has been criticized by other for-
malists (e.g., Pinker & Jackendoff 2005), highlighting the lack of agreement
about what such a model of language should look like even among linguists
subscribing to Chomsky’s general approach.
In any case, the goal in some areas of linguistics has been to try to under-
stand language complexity on its own terms, assuming that this could be
validly studied independent of how the brain actually works and that language
A Complex-Adaptive-Systems Approach to the Evolution of Language 69
structures so derived would then necessarily be evident in the brain in some

fashion. Language complexity has also usually been studied without a deep
understanding of how the evolutionary process has led to various other kinds
of behavioral change. This lack of evolutionary grounding is particularly unfor-
tunate, because language did in fact evolve, and, as a consequence, any model of
language complexity that is not easily explained from an evolutionary perspec-
tive cannot then be considered a legitimate model of language itself (Schoen-
emann 1999). Understanding the complexity of language necessitates under-
standing how the brain works in creating meaning, within the context of how
the evolutionary process would mold both brain and language.
2 Constraints on Language Adaptation

The evolutionary changes that occurred in our lineage to make language uni-
versal are presumably the result of natural selection (Pinker & Bloom 1990).
For natural selection to operate, however, there must be some environmental
influence favoring these changes. The environment relevant to language ability
is the social environment, which is of course made up of interacting individuals.
Evolutionary changes relevant to language will only be beneficial to an individ-
ual to the extent that they help that individual communicate better with others.
This in turn presumes that these other individuals already have similar enough
abilities to begin with (before selection operates), such that any new changes
introduced by one individual will actually increase this individual’s communi-
cation ability with others. This dynamic constrains the possible ways in which
evolutionary change can occur, constantly biasing changes toward modifica-
tions of pre-existing abilities. This in turn predicts that language will have been
built on a cognitive foundation that we share with other species, and further-
more that these foundations should still be evident. One critical component
of this foundation is conceptual understanding: how we perceive, experience
and make sense of the world. Aspects of this conceptual understanding are, of
course, exactly what we are trying to share and communicate with others using
language.
3 Types of Complexity
In order to assess complexity in language systems, it is important to define what
we mean by “complexity.” One place to look for such a definition is in the
field of Complexity Theory, where a number of definitions have been proposed
(e.g., Horgan 1995; Mikulecky 2001). A system in a high degree of disorder
(i.e., having high entropy) is, in some sense, complex, but it is not necessarily
complex in a particularly interesting way (Horgan 1995). One popular notion in
the field has been that the most interestingly complex systems exist at “the edge
of chaos” (Langton 1990), somewhere between complete chaos and complete

order. Langton argued that systems in such a state are the most potentially useful
for computation (Langton 1990), though subsequent work has not particularly
supported this idea (Mitchell et al. 1993).
One intriguing suggestion is that complexity should be understood as “the
property of a real world system that is manifest in the inability of any one for-
malism being adequate to capture all its properties” (Mikulecky 2001: p. 344).
This definition fits nicely with the observation that it has so far been impossible
to find a single formalism, even among committed formalists, that describes the
patterns of language grammar to everyone’s satisfaction (Pinker & Jackendoff
2005). If instead we view language as the result of a complex adaptive system,
in which interacting biological and cultural evolutionary systems – each with
their own constraints, influences, and partly interdependent histories – conspire
over evolutionary time to produce a system of communication, the problem of
language evolution becomes tractable.
Mikulecky (2001) suggests that the phenomenon of “emergence” seen in
complex adaptive systems may simply be “a result of the limits of a domi-
nant formalism” (p. 344). In other words, perhaps it is our fascination with
trying to box complex systems into single formal models that leads to our sur-
prise at the “emergent” behavior of these systems. If we view language evolu-
tion as the result of a complex interplay of influences of different kinds (each
described, imperfectly, by their own unique formalisms), the emergence of lan-
guage becomes much less miraculous.
For the purposes of this chapter, we will emphasize the following senses of
complexity:
1) number of different kinds of individual things (actions, objects, etc.);
2) number of individual interactions between things;
3) number of types of interactions between things; and
4) levels of hierarchical interaction between sets of things.
4 Conceptual Complexity
What exactly is meant by conceptual complexity? It can be understood to be a
function of: (1) the number of different dimensions the brain can meaningfully
distinguish, and (2) the number of possible interactions between these dimen-
sions (Schoenemann 2010). “Meaningfully distinguish” in this context may be
defined as any internally detectable difference in pattern(s) of brain activation
(whether caused by external stimuli or by internal neural activity); and “dimen-
sions” may be defined as aspects of reality that the brain is sensitive to (e.g.,
wavelengths of light, types of molecules, temperature, etc.) or internally cre-
ates (e.g., emotions, patterns of thought, etc.). One can imagine that there are
organisms that have a much simpler conceptual understanding of the world than
we do. Some invertebrates such as jellyfish, for example, do not have eyes that
produce images, but rather have very simple light-sensitive eyespots (“pigment
spot ocelli”) that simply detect light coming from a particular direction (Hudson
2010). This suggests a very simple conceptual awareness of visual information,
vastly simpler than what humans (and most vertebrates) have available.
5 The Neural Basis of Concepts

To understand why a comparative cross-species assessment of brain structure
could imply something important about likely degrees of conceptual complex-
ity, it is important to understand how concepts are instantiated in the brain.
Neural processing is thought to be the result of networks of neurons in differ-
ent states or temporal patterns of activation (Baars & Gage 2007). The brain
is organized with a degree of regional specialization of function. The cortex
itself, which is the seat of conscious awareness (in humans at least), is divided
up into different regions that specialize, to one degree or another, in particu-
lar types of processing (Baars & Gage 2007). The characteristics of the neural
circuitry in different regions (“cytoarchitecture”) differ enough to be reason-
ably identifiable across individuals, forming the basis for the identification of
specific brain regions. These cytoarchitectural characteristics led the early neu-
roanatomist Brodmann (1909) to suggest a classification scheme for cortical
areas (called “Brodmann areas”) that is still used to this day.
The details of what exactly each of these cytoarchitectural areas does, and
how they interact, is by no means completely understood, but there are well-
studied pathways that are known to specialize in different kinds of informa-
tion (Baars & Gage 2007). For example, it is possible to distinguish separate
areas and neural pathways specialized for visual, auditory, somatosensory (i.e.,
touch, temperature, pain), olfactory, and taste information. Within these, there
are typically further subdivisions of function. For example, cortical processing
of visual information starts in the primary visual cortex in the occipital lobe
(at the far back of the brain), but then divides approximately into (1) a dorsal
“where” pathway (extending superiorly into the parietal lobe) that specializes
in movement and spatial aspects of visual information, and (2) a ventral “what”
pathway (extending anteriorly into the temporal lobe) that specializes in object
identification (Baars & Gage 2007).
To be sure, there are additional areas and pathways that integrate this infor-
mation in various ways, and basic sensory processing in one sensory domain
can affect basic sensory processing in another domain, such that these primary
areas are not completely independent of each other. One very nice example
of this is the McGurk effect (McGurk & MacDonald 1976), where the basic
auditory perception of a syllable (e.g., ba) is fundamentally changed by concur-
rent visual input of someone saying a different syllable (e.g., ga). Nevertheless,
there are areas known as primary sensory areas that are known to specialize in
the processing of specific types of sensory information.
Very simple concepts, like colors, smells or tastes, are thought to be the
result of particular patterns of neural processing within different primary sen-
sory areas that combine different types of signals from external sensors. The
retina of the eye, for example, has cells that are tuned to respond to particular
wavelengths of light. Different colors are distinguished by different patterns
of stimulation of sets of these basic retinal cells. More complex concepts are,
at some level, based on patterns of activation of different networks subserving
often very different kinds of information. The concept of coffee, for example,
binds together a number of sensory components, involving not only taste and
smell, but also – for many people – visual components (e.g., shape and color of
coffee beans), somatosensory components (e.g., warmth of the fluid, physical
features of the cup typically used to drink it), and even more abstract compo-
nents, such as the sense of well-being that many feel as a result of drinking it
(Damasio & Damasio 1992).
There remains some question over how concepts are actually represented in
the brain at the neural level. One suggestion is that there are specific individual
neurons that represent specific concepts. Such hypothetical neurons are usu-
ally referred to as “grandmother cells” (after a tongue-in-cheek parable by the
neurobiologist Jerry Lettvin that introduced the term; Gross 2002). The idea of
individual concept-specific neurons is a logical extension of work on how the
visual system identifies objects (Gross 2002). At the level of the retina, individ-
ual ganglion cells are sensitive to the activity of specific, very simple patterns
of photoreceptor cells: for example, a small spot of light surrounded by dark-
ness. A straight line can therefore be detected as the coincident activation of
a unique set of these “spot detector” ganglion cells. This coincident activation
can be detected by a single neuron that fires only when the unique set of gan-
glion cells are also active (for a basic discussion see Goldsmith & Zimmerman
2000). Such a neuron would be the neural representation of a line (a simple
concept).
However, note that for such a “line detector” neuron to be activated, a net-
work of retinal ganglion neurons must also have been activated. For increas-
ingly complicated, subtle, and interesting concepts, larger and larger networks
of neural activity will be involved, connecting regions specializing in pro-
cessing different kinds of information relevant to those concepts. Whether
unique “grandmother cells” (or even specialized networks that we might call
“grandmother circuits”) actually exist for more complicated concepts, and if
so, whether these should be considered the neural instantiations of concepts, is
not critical for the argument here. The important point is that networks of brain
activation form the foundation for concepts in the brain.
Congruent with this, Barsalou (2010) has argued that the “core representa-
tions in cognition” – what we are calling here “concepts” – are not “amodal
data structures that exist independently of the brain’s modal [basic sensory]
systems,” but are instead fundamentally grounded in “the environment, situ-
ations, the body, and simulations in the brain’s modal systems” (p. 717). He
refers to this as “grounded cognition,” and notes a number of perspectives that
support this idea. Lakoff and Johnson’s (1980) analysis of linguistic metaphors
led them to argue that abstract concepts are specifically grounded in bodily
experience. Gibson’s (1979) work on visual perception led him to argue the
external environment plays a fundamental role in perception. Work by Paivio
(1971), Shepard and Cooper (1982), and Kosslyn (1980) suggested that mental
imagery was important to perceptual representations. Studies of brain activ-
ity while subjects are simply imagining an object (that is not actually present)
have shown that the same areas are activated as when the object is actually
being viewed (Damasio et al. 1993; Kosslyn et al. 1993; Kosslyn et al. 2003).
All of this is consistent with the idea that conceptual understanding is grounded
in basic perceptual information.
It is also important to note that at least some basic conceptual understanding
seems to be instantiated in brains independent of one’s language. For example,
Le Clec’H et al. (2000) showed that for bilingual subjects the same specific
brain networks for particular concepts (i.e., number processing vs. body-part
processing) appeared to be activated regardless of the language used. This is
consistent with the idea that languages map onto underlying conceptual net-
works in the brain.
Because of the difficulties of doing brain scanning in awake primates, little
work has been done outside of humans with respect to the neural instantiation
of concepts. However, there is evidence that grounded cognition holds across
species as well (Barsalou 2005). A study of brain activation in Rhesus macaques
(Macaca mulatta) showed that species-specific social calls resulted in increased
brain activation in auditory processing areas, visual areas, areas of the tem-
poral lobe associated with facial expression and visual motion, and emotion-
processing areas (ventro-medial prefrontal cortex, amygdala, hippocampus),
as compared to activation recorded while hearing unfamiliar sounds (Gil-da-
Costa et al. 2006). Thus, simply hearing socially relevant auditory input led to
increased activation in monkeys in a whole range of other areas known to be
involved in social information processing.
All of this suggests that the more complex a concept is, the greater the num-
ber of distinct brain networks will be activated either concurrently or temporally
in a causal manner. Thus, we may understand conceptual complexity as a func-
tion of the number of differentiable network activation states of a brain (with
the understanding that some sets of physically unique network activation states
may be functionally equivalent – i.e., will not be differentiable – because they
do not make a difference to the organism’s awareness). This notion is congru-
ent with Tononi’s definition of consciousness as being simply the integration
of information from different areas of the brain, and therefore a function of
effective connectivity (Tononi 2010), as well as general models of cognition

that emphasize network connectivity (e.g., McIntosh 2000).
This leads to the likelihood that the degree of network complexity of a
species’ brain is proportional to the complexity of their conceptual universe,
or their understanding of the world.
6 Species Differences in Network Complexity

Given that concepts are in some fundamental way dependent on brain net-
work activation states, comparisons of the neuroanatomical differences across
species, in particular between humans and our closest ape relatives, suggest sig-
nificant differences in their respective degrees of conceptual complexity. Dif-
ferences in brain size are one obvious feature of potential relevance to network
complexity, but there are a number of correlates of brain size that also sug-
gest important ape/human differences in network complexity, and therefore in
conceptual complexity.
The human brain is about three times the size of a chimpanzee brain, in abso-
lute terms (Jerison 1973). Although it is true that brain size is correlated with
body size across mammals, it is not clear that relative brain size (i.e., brain size
corrected for body size) is a better marker of behavioral ability than is simply
absolute brain size (Schoenemann 2006). Empirically, in fact, absolute brain
size is a better predictor of general cognitive ability than is relative brain size,
for primates at least (Deaner et al. 2007). For one thing, larger brains have
greater numbers of neurons (Haug 1987, Herculano-Houzel 2012), leading to a
greater total number of connections. Interestingly however, the connections do
not increase at a rate fast enough to maintain the same degree of interconnec-
tivity between regions (Ringo 1991). In other words, as brain size increases,
neural processing in given areas becomes increasingly independent of the pro-
cessing in other regions (though never completely so, of course). This leads
to the increased likelihood of specialization of function of areas, simply as a
consequence of increasing brain size. Empirically, the size of a species’ brain
predicts the number of anatomically distinct brain regions (which are assumed
to be functionally distinct as well, Changizi & Shimojo 2005; Northcutt & Kaas
1995). Estimating from brain size, the number of distinct areas for humans is
approximately 150, compared to the estimated number for apes of approxi-
mately 100 (Changizi & Shimojo 2005).1 In addition, the number of connec-
tions between areas appears to increase as a function of the square of the total
number of areas (Changizi & Shimojo 2005). Gibson (2002) has argued such
1 During production of this book, a new article was published arguing that the human brain actually
has 180 distinct cortical areas (Glasser et al. 2016). This occurred too late to allow for recalcula-
tions of the estimates discussed in this chapter, but would in any case only magnify the expected
human/chimp difference in conceptual complexity as proposed here.
evolutionary changes in our brain led to a significant increase in the ability to

mix “actions, perceptions, and concepts” (p. 10), which she sees as important
for language evolution.
To give a sense of what a difference in the number of distinct processing areas
might mean for differences in degrees of conceptual complexity, assuming it is
in some manner a function of degree of the number and types of network acti-
vation states, we can look at the increases in possible network activation states
made possible by the additional 50 or so cortical areas in humans. First, con-
sider that some concepts, such as individual colors, are likely the function of
processing at single areas, rather than interactions among more than one area.
The difference between human and ape complexity for concepts like this would
be a function of the ratio of distinct cortical areas (using estimates derived from
Changizi & Shimojo 2005), or 1.5 times (i.e., 150/100). In other words, every-
thing else being equal, and assuming individual areas in the two species are
equally complex in their internal processing (which is not necessarily likely –
see later discussion), we would expect humans to have about 1.5 times as many
of these basic concepts as do chimpanzees. We can think of this ratio as an
estimate of possible increased network complexity, and therefore of possible
increased conceptual complexity. Note that a ratio is probably a safer compari-
son for these purposes than is the absolute increase in numbers of areas, because
our knowledge of how processing in individual areas is connected to specific
concepts is limited. The use of a ratio for comparison just assumes the process
of concept formation is essentially the same across species.
Not all concepts are likely the result of processing in single cortical areas,
however. Several concepts likely rely on the interaction of at least two dis-
tinct areas. For example, our perception of flavor is actually a mix of olfac-
tory (smell) and gustatory (taste) processing, though additional neural systems
often contribute to our sensation of flavor as well (Shepherd 2006). To get a
crude estimate of the possible complexity of conceptual awareness that could
result from combinations of two different areas, we can calculate the possible
combinations of sets of two areas given either 150 (human) or 100 (ape) total
distinct areas.2 The corresponding ratio of increased complexity in this case
would be around 2.3x (11175/4950). This ratio continues to rise as we consider
sets of 3, 4, 5, and so on combinations of these areas (Figure 4.1). If concepts
for both species can be a function of as many as 11 distinct areas, this would
2 This is calculated in the following way: For a given number of areas assumed per concept, the
number of possible combinations (ignoring the order) equals n!/[(k!)(n-k)!], where n = total
number of individual areas, and k = number of areas allowed per combination. As one considers
larger possible subset sizes, the total number of all possible combinations is the sum of combi-
nations for each value of k up to and including the largest subset size of interest: [total number
of possible combinations]=[combinations for k=1]+[combinations for k=2]+ …[combinations
for largest subset size].
100
ratio of distinct combinations for human compared to ape
90
80
70
60
50
40
30
20
10 y = 0.9651e0.4186x
0
0 1 2 3 4 5 6 7 8 9 10 11
maximum number of distinct processing areas used to create a given concept
Figure 4.1 Increase in ratio of possible combinations of areas for human vs.
ape. Figure 4.1 illustrates how many more combinations of areas there could
be, in theory, for humans vs. apes, assuming humans have 150 distinct areas,
and apes have only 100. The ratio advantage for human brains increases expo-
nentially as the maximum numbers of areas possible per unique combination
of areas is increased (X-axis). These are illustrations of possible differences
given a simple model of concept formation, and are not estimates of actual
ratios of conceptual complexity in humans vs. apes. See text for details and
references.
translate to more than a 100-fold advantage for humans in number of theoret-

ically possible concepts. Caution should be taken in assessing these calcula-
tions, of course, as they are based on a number of simplifying assumptions.
For example, we don’t know how the maximum number of different areas that
could potentially contribute to a concept, so exactly how far out on the X axis of
Figure 4.1 we should consider is not clear. It is also entirely possible that larger
brains are actually better at integrating increasingly larger numbers of areas,
and thus, that some human concepts might be formed from a larger number of
areas than occurs in apes. This would mean that we should not be comparing
equivalent cortical area set sizes between them, as Figure 4.1 assumes. How-
ever, note that this would magnify the probable difference in conceptual com-
plexity between humans and apes. These calculations are not meant to be taken
as definitive measures, but simply to highlight the possible effects of anatomical
differences between humans and apes on conceptual complexity.
In addition to the increased number of cortical areas found in human com-
pared to ape brains, individual areas themselves appear to be larger in humans.
Calculating from Changizi and Shimojo’s (2005) regressions of average per-
centage size of a neocortical area as a function of absolute brain size across
mammals, one can calculate that human areas are on average approximately
2.3 times larger, in absolute terms, than what we would expect for correspond-
ing areas in apes. There is substantial variation across particular regions, of
course. The human primary visual cortex (V1, or Brodmann area 17) is only
1.6 times as large as that found in chimpanzees (data from Stephan et al. 1981),
while Brodmann area 10 of the prefrontal is 6.3 times as large in absolute terms
(data from Semendeferi et al. 2001).
Does this necessarily translate into increased complexity of processing? One
reason to suspect it does comes from Penfield’s (1950) pioneering work map-
ping the primary motor and somatosensory cortices. He found that there were
substantial differences in the amount of cortex devoted to different areas of the
body, with areas demonstrating greater sensitivity or degree of motor control
having correspondingly larger cortical representations. He strikingly illustrated
this with images in which body parts are drawn in proportion to the size of the
cortical representation for each area. For example, the lips of these homunculi
are very large, consistent with the fact that we have much greater sensitivity to
our lips than to many other parts of our body. This indicates that the size of a
cortical area does in fact have a functional consequence for primary motor and
somatosensory areas within humans. Additionally, some studies have reported
that the size of localized areas of the cortex correlate with degrees of ability for
behaviors mediated by those areas (e.g., Schoenemann et al. 2000; Thompson
et al. 2001). This all suggests that larger cortical areas do in fact correspond to
more complex processing of information mediated by those areas in humans.
Across species there also appear to be associations between the proportion
of cortex that mediates a particular behavior and the degree of elaboration of
function of that behavior. For example, Star-nosed moles (Condylura cristata)
that live most of their life underground and consequently have very poor eye-
sight have correspondingly small primary visual cortices, whereas ghost bats
(Macroderma gigas) that depend heavily on echolocation devote about half
their cortex to auditory processing (Krubitzer 1995). Thus, both across species
and within humans, the size of a cortical area appears to be associated with the
degree of function of that area (Schoenemann 2010).
All of this suggests not only that humans have a greater number of distinct
cortical areas with more connections between areas, but also that individual
areas are likely capable of more complex types of processing. This would fur-
ther magnify the expected difference in complexity, subtlety, and richness of
concepts in humans compared to apes – beyond what one might expect, based
on differences in the number of distinct areas alone.
Finally, during human evolution there appears to have been a biased increase
specifically in areas known to be relevant to language processing. For example,
the temporal lobe, which is known to mediate connecting concepts to words
(Damasio & Damasio 1992), is about 23 percent larger than expected, based
on primate scaling trends (Rilling & Seligman 2002). In addition, other areas
that participate in language processing, including prefrontal cortex, appear to
have increased disproportionately during human evolution (see Schoenemann
2012 for a review).
All of this leads to the conclusion that there was a dramatic increase in
conceptual complexity during our evolutionary lineage. Given that all of this
increase occurred within the context of an intensely socially interactive group
existence, it is hard to imagine that this increase in conceptual complexity was
not fundamentally important to driving the evolution of language.
7 Conceptual Awareness in Other Species

The behavior of other species is consistent with the notion that they also have
concepts, and further, that many of their concepts overlap in important ways
with our own. However, an organism’s conceptual understanding of the world
will inevitably be influenced by the kinds of sensory information they have
evolved to be aware of. Dogs do not have the same range of color information
as humans do, but they appear to have better discrimination in low light and bet-
ter differentiation of shades of grey (Miller & Murphy 1995; Neitz et al. 1989).
Echolocating bats can hear frequencies far in excess of humans, and they can
use echoes from these sounds to recreate spatial relationships between them-
selves and their flying insect prey (Jones & Holderied 2007). Elephants can
produce and respond to sounds much lower than humans can hear (Garstang
2010). To the extent that species differ in their sensory awareness, their con-
ceptual understanding of the world will likely be different.
There is, however, substantial overlap in the types of sensory information
that humans have with other species, and the ways in which this information is
categorized into conceptual information often appears to match that shown in
humans. Experiments with categorization of images by pigeons, for example,
suggest they organize visual information into categories corresponding essen-
tially to people and trees that are very close to our own (Herrnstein 1979).
Pigeons and monkeys have been shown to be able to correctly categorize still
pictures of animals vs. non-animals, which is a fairly abstract concept (Roberts
& Mazmanian 1988). Wasserman et al. (1988) found that pigeons learn to
categorize images much faster if they are organized into sets that correspond to
human language categories (e.g., cats, flowers, cars and chairs), as com-
pared to arbitrary sets. This suggests that human languages and cultures make
use of categories that are “real” to a wide variety of animals (Schoenemann
2005).
Monkeys appear to have several conceptual categories that correspond at
least partly to those found for humans. Monkeys and apes are able to recog-
nize faces of individuals in their groups (Parr 2011). Monkeys also act as if
they understand complex social relationships, including hierarchical matrilin-
eal kin groupings, not only with respect to their own positions but also that of
others in their social group (Bergman et al. 2003; Seyfarth & Cheney 2000).
Several species of monkeys and at least one small ape species (gibbon) have
been shown to have different alarm calls for different predators (Clarke et al.
2006; Seyfarth et al. 1980; Zuberbuhler 2000).
In order for these primate species to have such calls, they must have sep-
arate concepts that relate differentially in some way to each type of predator
(Schoenemann 2005). Cheney and Seyfarth (2005) argue these alarm calls are
more properly thought of as propositions, as opposed to simple nouns identify-
ing predators. Monkeys and apes (at least) do seem to have concepts that corre-
spond to more than just things. For example, individuals respond differently to
others in their social group depending on the context. Apes have been reported
to hug other individuals when they are distressed (de Waal 2008). Monkeys
respond with reconciliatory behavior after an aggressive encounter with a dom-
inant individual if that individual gives a specific kind of vocalization (Cheney
& Seyfarth 1997). These kinds of behaviors show that they are sensitive to
the behaviors displayed by others, which suggest that primates have concep-
tual understanding that differentiates actions from actors (or things). This
is important part of the conceptual distinction marked by the verbs vs. nouns
in human language grammar.
The clearest evidence that apes have concepts comes from ape language stud-
ies, which have fairly convincingly shown that apes have the ability to use hand
signs or lexicons to represent concepts of various kinds. Double-blind tests
show that chimpanzees (Pan troglodytes) can correctly name objects with hand
signs (Gardner & Gardner 1984; Gardner et al. 1989). Premack and Premack
(1972) showed that chimpanzees could use lexigrams to answer questions about
concepts related to objects. For example, when the subject Sarah was asked –
via lexigrams – to name the color of “apple” (represented by a blue triangle in
her sign system), she responded with the lexigram for “red.” Perhaps the most
extensive and impressive work so far with apes has been with Kanzi, a bonobo
(Pan paniscus), by Savage-Rumbaugh and colleagues (Savage-Rumbaugh et al.
1993). In one study, Kanzi was presented with more than 660 novel and unusual
sentences in controlled blind tests, including things like: “Pour the Coke in the
lemonade,” “Go get the can opener that’s in the bedroom,” and “Kanzi is going
to chase Rose,” and he responded correctly to 72 percent of these, which is
far above chance given the complexity of the sentences. While there is con-
troversy over the extent to which these show evidence of incipient grammar
knowledge (e.g., Wynne 2008; but see Savage-Rumbaugh et al. 2009), Kanzi’s
understanding of a few hundred spoken English words – and by extension the
concepts underlying them – is not seriously debated.
We can get some idea of the range of kinds of concepts that Kanzi appears
to understand by looking more closely at the words in the sentences that he
responded immediately and correctly to, as listed in Savage-Rumbaugh et al.
(1993). Note that this leaves out cases where the researchers nevertheless
believe Kanzi understood the sentence but responded imperfectly to it. For
example, when Kanzi was told: “Put the carrot in the water,” he picked up a
carrot, made a vocalization, took a bite of the carrot, and then put it in the
water. The researchers scored this “not immediately correct” because he ate
some of the carrot beforehand (Savage-Rumbaugh et al. 1993: p. 162). Restrict-
ing our analysis only to sentences Kanzi immediately and correctly responds to
also leaves out cases where he was only partially incorrect. For example, when
instructed: “Give the big tomato to Liz,” he picked up both the big and the
little tomato and gave them to the researcher (Savage-Rumbaugh et al. 1993:
p. 163). However, he didn’t pick up any of the numerous other objects that
are in front of him, and he didn’t interact with any other researchers present,
thereby suggesting that he only misunderstood a single word in the sentence,
that is, “big,” and not all the other words in the sentence. Restricting the anal-
ysis here just to those sentences to which Kanzi responded immediately and
correctly is therefore highly conservative and strongly biased against over-
interpreting Kanzi’s understanding of spoken English. In the view of those
who work with him, Kanzi almost certainly understands more than is indicated
here.
In total, Kanzi responded correctly and immediately to 368 of the novel sen-
tences in this study, involving a total of 2,354 word tokens. Parts of speech from
these sentences were identified using tools from LingPipe, using the Brown cor-
pus (Alias-i 2008), and then checked by visual inspection. Tables 4.1–4.5 list all
the individual words in these sentences, divided into parts of speech categories,
with the number of times each word appeared in the sentences also indicated. In
total, there were 225 different words used across the sentences in this particular
test. This included 119 nouns, 49 verbs, 32 adjectives/adverbs/prepositions, 11
pronouns, and the names of 14 different individuals. (Note that these counts col-
lapse similar words into one instance, e.g., hid, hide, and hiding, are counted as
one verb.) The fact that Kanzi responded correctly and immediately to all these
sentences is of course not evidence that he necessarily understood every single
one, or that he knows the concept adverb, for example.
Table 4.1 Nouns Used in Sentences Kanzi Responded to Correctly
Number of Occurrences Nouns (119)
51 ball
32 outdoors
27 doggie
19 bedroom
18 refrigerator
17 water
16 room
14 rock
13 shot
12 microwave
12 oil
12 potty
11 milk, needles, snake, tomato, toothpaste
10 gorilla, pine, umbrella
9 TV
8 banana, orange, sparklers
7 bowl, collar, colony, juice, lighter, mask, raisins, stick, toy
6 backpack, bunny, can, carrot, melon, monster, mushrooms,
opener, play, soap, vacuum
5 apple, cereal, ice, keys, rubber, yard, wipie(s)
4 band, blanket, clay, hat, hose, picture, pillow, potato, teeth,
telephone
3 balloon, cabinet, coke, dog, door, hand, head, hotdogs, Jello,
lemonade, money, paint, peas, phone, pineapple, shoe
2 book, cleaner, coffee, food, hotdog, knife, orang, paper, tickle
1 bananas, bands, bark, bottle, bubbles, butter, cane, car, clovers,
diaper, egg, fire, flashlight, Fourtrax, grapes, hair, hammer, hug,
lettuce, mouth, mushroom, onions, outside, oven, Perrier,
potatoes, shirt, straw, sugar, surprise, tape, yogurt
However, Kanzi’s performance on these sentences is strong evidence that,

at a cognitive level, he not only understands that arbitrary sound patterns ema-
nating from the researchers’ lips can refer to specific concepts, but also that his
underlying conceptual understanding is quite a bit broader than this vocal com-
munication would otherwise suggest. His conceptual world appears not to be
limited to individual items or people (nouns), but also includes concepts relat-
ing to actions (verbs), and even location (e.g., on, with) and temperature
(hot). His performance on tasks like this depends on some non-trivial degree
of shared conceptual understanding of the world, along with a deep desire to
be a part of a social group in which communication between individuals is
important.
Table 4.2 Verbs Used in Sentences Kanzi Responded to Correctly
Number of Occurrences Verbs (49)
97 put
73 get
62 go/going
58 take
52 can
48 give
15 pour
13 show
12 open, tickle
11 is/are
8 bite
7 hid/hide/hiding
6 could, knife [cut with a knife], make, wash
5 brush, scare
4 chase, grab, see, slap/slapping
3 do, eat, hit, need, tell, throw, want
2 drink, gonna, groom, hug, keep, play, stab
1 bring, carry, close, feed, hammer, let, sit, squeeze, start, think,
turn, would
While it is true that Kanzi and other ape language subjects are in highly
unique circumstances (for non-human apes) and are not representative of what
apes are doing in the wild, they do nevertheless allow us to understand what
an ape brain is capable of, given the right sociocultural developmental circum-
stances. They are our best guess about what the cognitive capacities of our com-
mon ape ancestors would have been like. This research shows that animals, and
in particular our closest primate relatives, have concepts and can use arbitrary
signs to refer to these concepts in interesting and important ways.
Table 4.3 Proper Names Used in Sentences Kanzi

Responded to Correctly
Number of Occurrences Proper Names (14)
31 Rose
21 Kelly
17 Liz
16 Kanzi
6 Linda, Sue
4 Matata
3 Panbanisha
2 Karen, Panzee
1 Austin, Krista, Panban, Sherman
Table 4.4 Pronouns Used in Sentences

Kanzi Responded to Correctly
Number of Occurrences Pronouns (11)
63 you
45 your
15 me
14 it
5 I
3 her, him, them, we
1 my
8 The Human Elaboration of Conceptual Complexity

While the evidence outlined here suggests that other animals, and in particular
apes, have concepts, they nevertheless appear to be limited in the diversity,
subtlety and complexity of their conceptual understanding. In his review of ape
language studies, Snowdon (1990) remarks:
Although the abilities of Kanzi and his companions are remarkable and come very close
to some of the capacities shown by young children, there still appear to be limitations.
Bonobos [pygmy chimpanzees] and chimps appear to be more limited in the topics that
they find interesting to communicate about (p. 222, italics added).
This intuitive assessment is consistent with simple numerical differences in the

number of signs that apes appear to be capable of learning compared to the
number of words that humans typically understand. The previous analysis of
Table 4.5 Adjectives, Adverbs and Prepositions Used in Sentences Kanzi

Responded to Correctly
Number of Occurrences Adjectives/Adverbs/Prepositions (32)
451 the
92 to
79 in
54 on
40 and
38 some
22 that, with
19 a
4 down, out
3 hot, now, of
2 away, big, by, for, this
1 an, at, back, good, hide, if, new, off, over, real, somewhere,
sweet, there
the sentences that Kanzi immediately and correctly responded to, for exam-
ple, suggests he broadly understands the meaning behind at least 225 differ-
ent words. This is broadly similar to what other ape language studies have
reported. For instance, the orangutan Chantek reportedly learned 127 different
signs (Miles 1990).
By contrast, humans have a working vocabulary that is several orders of mag-
nitude larger. Miller and Gildea (1991) estimate that the average high school
student knows the meanings of about 40,000 dictionary entries, and that adding
proper names would likely double this number. This suggests there is at least
a 100-fold increase in the ability to use arbitrary signs to refer to underlying
concepts in humans compared to apes.
However, to what extent should we expect these apparent differences in
the sheer number of lexical items (or communicative signs) to actually reflect
underlying differences in the richness, subtlety, and complexity of conceptual
understanding, as opposed to simply reflecting a difference in the ease or ability
of attaching arbitrary signs to the underlying conceptual meanings? It is true
that Kanzi’s use of signs for communicative purposes reportedly increased at
a slower rate than is typical for normal human children (Savage-Rumbaugh &
Rumbaugh 1993). However, both Kanzi and his half-sister Panbanisha contin-
ued to learn new words into adulthood, in a social context in which words were
used by their caregivers (Lyn & Savage-Rumbaugh 2000). Slower learning of
communicative signs in apes might reflect a difference in their ability to infer
the possible meaning from a complex environment, rather than some specific
difficulty associating signs with concepts. However, given the differences in
brain anatomy outlined earlier, and their relevance to the possible richness of
conceptual understanding, it is likely that a large part of the difference in num-
ber of signs for communication highlighted here does reflect a difference in
underlying conceptual complexity.
To get a visual sense of the possible difference in complexity of the human
semantic network compared to that for Kanzi, semantic similarities were cal-
culated on the corpus of Kanzi’s correct sentences (described earlier) and com-
pared to those calculated on an adult human speech corpus: the Charlotte
Narrative and Conversation Collection. This is a corpus of 95 narratives,
conversations and interviews from residents of Mecklenburg County, North
Carolina, and surrounding North Carolina communities, from the American
National Corpus Consortium (www.americannationalcorpus.org). The relative
semantic similarity between words was calculated using the BEAGLE method
(Bound Encoding of the Aggregate Language Environment), which relies on
statistical redundancies within a corpus to build a semantic space representa-
tion of meaning (Jones & Mewhort 2007). Because the human corpus used here
contains around 8,700 words, a single figure depicting the estimated semantic
network is too dense to assess; only a subset of the human network can be
displayed. As a simple comparison, corresponding human/ape semantic net-

works were plotted connecting all words estimated by the BEAGLE method to
be semantically closer to the word “milk” that is the word “oil.” There were 16
such words in Kanzi’s corpus (Figure 4.2(a), compared to 108 for the human
corpus (Figure 4.2(b). As is immediately obvious, the semantic network for
these broadly equivalent subsets are dramatically different in the two species,
with the human network much richer and covering a much larger portion of
possible semantic meaning space.
The density of the semantic space implied by these corpora, as estimated by
the BEAGLE method, varies quite a bit in different areas, however. Table 4.6
illustrates the relative densities of the same areas of semantic space for Kanzi
vs. humans. The numbers represent the total number of words estimated to
be closer to the target word than is the comparison word for Kanzi’s corpus
and the human corpus. As can be seen from these examples, in some regions
the semantic density seems to be much greater in the human corpus than the
corresponding area in Kanzi’s corpus (e.g., the space around the target word
“apple”), but for the others listed the difference is smaller, and in one case even
reversed (e.g., Kanzi’s estimated semantic space has 25 words closer to “eat”
than is “get,” but for the human corpus “get” is the closest word to “eat”).
This comparison is imperfect for a variety of reasons, and should only be
seen as a suggestive first attempt at a numerical comparison of semantic rich-
ness between the species. Kanzi’s corpus was taken from sentences specifi-
cally selected to assess his understanding of spoken English, and as such were
thought to be sentences he had never encountered before. The human corpus
represents a sample of spoken English, but obviously does not consist of sen-
tences designed to test knowledge of spoken language. Nevertheless, Kanzi’s
corpus is one of the broadest that has been published for apes, and as such
this comparison is the best look at species differences in richness of semantic
meaning that is currently available. Future work exploring the semantic space
between species will hopefully refine the comparisons, but this initial assess-
ment is at least consistent with a dramatic difference in the degree of concep-
tual complexity, as is predicted by comparisons of brain structure and function
between humans and apes.
Finally, note that the basic argument here is not inconsistent with Deacon’s
(1997; 2012) view that there is a fundamental qualitative difference between
humans and all other animals in the types of signs we are able to use. Deacon
relies on Peirce’s (1867) taxonomy of types of signs, in which “icons” are signs
that directly resemble their referents, “indexes” are signs that correlate reliably
with their referents, and “symbols” are signs that refer to their referents only
in non-iconic, non-indexical ways, such that they are completely arbitrary con-
ventionalizations. Deacon believes that only humans and language-trained apes
like Kanzi truly understand and use signs that are symbols in this sense, and for
Figure 4.2 Comparison of semantic network density for ape vs. human cor-
pora. These figures show a subset of the estimated semantic meaning space
for: (a) Kanzi’s corpus of sentences he responded to immediately and cor-
rectly, and (b) the Charlotte Narrative and Conversation Collection corpus
(see text). These figures plot the meaning space that contains all the words
estimated to be as close to, or closer than, “milk” is to “oil.” There were 17
such words for Kanzi’s corpus and 110 in the Charlotte Narrative and Con-
versation Collection corpus. Semantic meaning space was calculated using the
BEAGLE method (see text). Lines between word nodes represent estimates
of some level of shared semantic meaning.
Table 4.6 Comparative Semantic Web Density for Select

Words in Human vs. Chimp
Number of Words Closer to Target Word than

the Comparison Word for
Target Word Comparison Word Human Kanzi
apple raisin(s) 6747 7

milk paper 354 40
milk oil 108 16
carrot flashlight 267 120
eat get 0 25
Numbers in the columns represent the total number of words estimated to be

closer semantically to the target word than is the comparison word. Semantic
network was estimated using the BEAGLE method. See text for details on
the human and Kanzi corpora used for this analysis.
him this is the key difference. But because he agrees that apes can learn to use
symbols in particular developmental circumstances, the transition to symbolic
behavior is for him fundamentally a behavioral one. Deacon also argues that
symbols are built up dynamically from indexical relationships (Deacon 2012).
The more complex the underlying conceptual structure, the more potential there
would likely be for symbolic behavior in Deacon’s sense.
9 Emergence of Syntax
The evidence laid out here supports the contention that there is a large dif-
ference between humans and our closest relatives in the degree of underlying
conceptual complexity. Considered in the context of primate interactive social-
ity, it is hard to believe this would not have had a fundamental influence in
driving the evolution of language. As an enhanced communication system, the
usefulness of language is partly a function of the usefulness of the underlying
conceptual system it is used to convey. While some theorists have suggested
that language should not be thought of as primarily a communication mecha-
nism (e.g., Berwick et al. 2013), this is really just a reflection of these theorists’
conflation of underlying conceptual structure (as viewed here) with a definition
of language itself. A significant part of the underlying conceptual structure is
almost surely pre-linguistic, and even pre-hominin. This is not to say that it
is identical in non-human primates, but rather that it formed the basis for lan-
guage (as a communication mechanism) in the first place. There would not
have been any reason to evolve any kind of enhanced communication system
(leading ultimately to language) absent the development of a rich underlying

conceptual structure to begin with. What would be the point?
It is also sometimes claimed that monkeys have very limited communication
systems (e.g., Berwick et al. 2013). Typically, what is actually meant by this
is that they only have a few identifiable vocal calls. However, this ignores the
tremendous subtlety of primate nonverbal communication (Suomi 1997) and,
as a result, is vastly too simplistic a view of primate communication. The reason
non-human primate vocal communication systems are so simple is likely that
individual survival is not enhanced by extensive vocal communication beyond
a few calls specific to their most dangerous predators, as well as the variety of
calls that play important roles in social signaling. It is quite clear that primates
have very rich understanding of social relationships (Cheney & Seyfarth 2005;
de Waal 1982). Cheney and Seyfarth (2005) specifically argue that “upon hear-
ing vocalizations, listeners acquire information about their social companions
that is referential, discretely coded, hierarchically structured, rule-governed,
and propositional” (p. 135). Given what Kanzi and other ape sign communica-
tion projects have demonstrated, it appears to be possible for much (if not most)
of this rich underlying conceptual understanding to be coded into some more
elaborate vocal communication system, given the right developmental environ-
ment, and if it were specifically adaptive to do so.
It seems likely therefore that at some point our conceptual system became
complex enough that – in the context of an intensely socially interactive exis-
tence, and in the relaxation of strong selective pressures against overt signal-
ing (i.e., because risk of predation became sufficiently reduced) – an enhanced,
complicated communication system involving grammar and syntax would have
been inevitable. This would not, however, have required the evolution of dedi-
cated, innately specified grammar circuits. The complexity of the grammatical
system can be seen as an emergent feature of this process (Savage-Rumbaugh &
Rumbaugh 1993; Schoenemann 1999). But what evidence is there that human
language grammar could actually result simply from increasing conceptual
complexity? The details of how this might have happened have not been worked
out in detail, but several additional areas of study point in this direction.
First, it should be noted that the claim of a clear distinction between grammar
and underlying conceptual understanding has been disputed. Some linguists
have specifically emphasized the fundamental interconnectedness of grammar
and semantics (e.g., Haiman 1985; Langacker 1987; O’Grady 1987). To the
extent that these alternative models of language are correct, we would expect
increasing conceptual complexity to, in effect, essentially mean the same thing
as increasing grammatical complexity. These models fit an evolutionary frame-
work much more elegantly than do those from the formalist tradition.
Second, proposed substantive features of Universal Grammar actually look
suspiciously like descriptions of how we conceptualize the world, rather than
specific rules about how components of language are structured (Schoenemann

1999). For example, Pinker and Bloom (1990) produce a list of “uncontrover-
sial facts about substantive universals, the building blocks of grammars that all
theories of universal grammar posit …” (p. 713). One they list is: “Phrase struc-
ture rules (e.g., “X-bar theory” or “immediate dominance rules”) force con-
catenation in the string to correspond to semantic connectedness in the under-
lying proposition, and thus provides linear clues of underlying structure …” (p.
713). This amounts to simply saying that all languages have rules about how
to translate complex, multidimensional conceptual meaning to a linear string
of sounds. What is not universal is a particular, specific set of rules that all lan-
guages adhere to. Instead, languages differ, sometimes substantially, in what
these particular rules look like (which is why Pinker and Bloom are not able to
list specific rules shared universally). What this substantive universal amounts
to is simply that all languages have some way of coding complex, multidimen-
sional internal conceptual meaning into a serial channel (usually an auditory
signal) in a conventionalized way. However, this is exactly what one would
expect to occur purely through cultural evolutionary processes alone, if the pur-
pose of language is to communicate meaning between individuals who share a
basic conceptual understanding of the world.
To take another example, Pinker and Bloom (1990) posit that in all languages
“Verb affixes signal the temporal distribution of the event that the verb refers
to (aspect) and the time of the event (tense) …” (p. 713). This amounts to say-
ing that all languages have rules that help express when some action occurred
(tense), and how it was/is occurring (aspect). If the function of language is to
allow communication between individuals sharing the same underlying con-
ceptual understanding, then given that humans share a similar conceptualiza-
tion of the passage of time, and given that time is highly relevant to humans,
one would specifically expect conventionalized rules to emerge through cul-
tural evolutionary processes that would allow speakers to mark this important
information for listeners.
Assuming one important function of language has been the communication
of conceptual understanding, we should expect that the structure of this con-
ceptual understanding would have molded the structure of the grammar. This
would be true whether or not the original purpose of language was thinking
(e.g., Berwick et al. 2013). Language clearly has been used for communica-
tion for a long time and therefore its structure would necessarily be expected
to have been significantly molded by shared conceptual understanding among
speakers.
What is lacking in arguments for the evolution of grammar from the for-
malist tradition (e.g., Jackendoff 2002; Pinker & Bloom 1990) is the demon-
stration that cultural evolutionary processes cannot account for the emer-
gence of shared grammatical rules from a common foundation of shared
conceptualizations across individuals played out in the context of individuals

in a socially interactive existence trying to share information.
Third, empirical studies of human children learning language suggest the
development of grammatical knowledge is not actually independent of the
development of lexical knowledge. Bates and Goodman (1997) have shown
that complexity of grammatical knowledge in children’s speech production
is tightly correlated with the development and complexity of their (non-
grammatical) lexicon. If grammar were truly an independent system, it should
show evidence of being highly decoupled in development, yet it does not.
In fact, linguistic development in children has been argued to be essentially
item-based: young children use language in a way that suggests they do not
understand and use abstract grammatical categories correctly until they have
learned the holistic meanings of many isolated phrases first: the “verb-island”
hypothesis (Tomasello 2000; Tomasello 2003). For example, in an experi-
mental setting, two-year-old children rarely used novel verbs transitively if
they had only been introduced to them in intransitive sentences, and vice
versa, even though they understood the holistic meanings of particular tran-
sitive and intransitive sentences (Tomasello & Brooks Patricia 1998). This
is consistent with the view that children construct their grammatical knowl-
edge by first learning the meanings of whole structures, and then deduce the
grammatical rules later based on repeated patterns they experience over being
exposed to many independent tokens. This is not consistent with the idea of
abstract innate grammatical categories independent of meaning. If this were
the case, once a child had knowledge of transitive and intransitive sentences
of any kind, they should be able to use all verbs in both ways; yet they
do not.
There remains debate in the field of language learning, however, about
whether or not children could learn all of syntax without at least some of it in
some sense being specifically built-in (e.g., Universal Grammar) independent
of the lexicon. Gleitman and colleagues (Gleitman 1990; Gleitman & Gleitman
1997; Gleitman et al. 2005), for example, have argued that children make use
of syntactic knowledge to help them learn the meanings of words through a
process they call “syntactic bootstrapping.” Although they believe that some
non-trivial amount of syntactic knowledge is innate in children, this claim is of
course logically independent of whether or not syntactic bootstrapping occurs.
Such bootstrapping would be expected to occur even if syntactic structure is
simply a reflection of underlying shared conceptual understanding. From an
evolutionary perspective, it makes more sense to suppose that conceptual under-
standing has molded language syntax, rather than believing that syntax has an
independent evolutionary origin.
Lidz, Gleitman and colleagues have published a number of experiments that
they believe demonstrate that children have expectations about what syntax
looks like that they could not have learned from input. Their studies have
focused specifically on learning verb-argument structure (Lidz et al. 2003a;

Lidz & Gleitman 2004a; Lidz & Gleitman 2004b) and anaphoric reference
(Lidz et al. 2003b). However, an array of non-innatist criticisms have been
offered for their findings, including questioning whether the studies actually
demonstrate something about innate syntax rather than simply their understand-
ing of word meanings (Tomasello 2004), whether structure is really absent from
the linguistic input children are exposed to (Foraker et al. 2009; Regier & Gahl
2004), and whether simple pragmatic constraints would be sufficient (Goldberg
2004).
Whether or not some aspects of syntactic knowledge are specifically coded
innately, there is broad general agreement among child language learning
researchers that syntactic knowledge and lexical knowledge develop together
in children. For example, Lidz et al. (2003a) state: “As is now well attested,
the verbs of the exposure language are acquired in lockstep with acquisi-
tion of those features of the clause-level grammar having to do with the rela-
tion between a verb’s semantic argument structure and its syntactic structure”
(p. 152; see also Berwick et al. 2013). Thus, both major models of language
learning in children agree that lexical and grammatical knowledge are funda-
mentally linked.
Of particular interest to the central thesis of this chapter, Bates and Goodman
(1997) review empirical evidence that the development of expressive grammat-
ical complexity appears to be an exponential function of the size of the lexicon,
such that grammatical complexity increases very slowly up to a vocabulary of
around 200 words, and then begins to accelerate beyond that (Bates & Good-
man 1997). In fact, children in the tenth percentile of grammatical complexity
for vocabulary sizes of between 200–300 averaged a grammatical complexity
score of zero (Bates & Goodman 1997). Note that ape sign communication
studies typically report that subjects know about this many signs. It is thus not
uncommon for children with vocabulary sizes matching those claimed for apes
to similarly also show limited evidence of grammar in their production. A study
of Kanzi’s productive communicative sequences (as opposed to his comprehen-
sion abilities) when he was five years old showed that they were mostly limited
to two words, thus limiting the complexity that could be expected (Greenfield &
Savage-Rumbaugh 1991). Even so, some simple patterns were evident. Further-
more, language comprehension also precedes language production in human
children, which in turn precedes evidence of grammar usage (Bates & Good-
man 1997). Thus, the fact that Kanzi’s comprehension precedes his production,
and that his use of grammatical structure in expressions is limited compared to
his comprehension of grammatical structure (Greenfield & Savage-Rumbaugh
1991; Savage-Rumbaugh et al. 1993) is not a valid reason to suspect a qualita-
tive difference between apes and humans with respect to language. All of this
is consistent with the emergence of grammatical complexity from increasing
underlying conceptual understanding.
Last, the possibility of a purely cultural evolutionary transition to grammat-

ical structure as a consequence of expanding lexicon has also been explored in
computer simulations. Such simulations necessarily make a number of simpli-
fying assumptions, but they serve as important tests of proof-of-concept, which
is particularly important in the case of evolutionary arguments because human
intuitions about evolutionary dynamics are often incorrect. Computer simula-
tions of language evolution have generally shown that a surprising degree of
emergent structure can occur from cultural evolution alone (for a review, see
Gong et al. 2014; Steels 2011). Using an iterated learning agent-based model,
Smith et al. (2003) showed that compositionality in communication systems
(a precondition of grammar where the meaning of an entire signal sequence
is a function of the meanings of subparts of the sequence and their order in
the sequence) occurs only when the meaning space that agents are trying to
communicate itself exhibits structure. In other words, structure in the com-
munication system emerges from structure in the underlying meaning space
(equivalent to conceptual complexity as it is used here). Earlier agent-based
simulations (Batali 1998; Goroll 1999) had reported emergent compositional-
ity deriving entirely from cultural evolution, though the effects of the structure
of the meaning space were not investigated (e.g., Batali 1998). More recently,
Gong (2009; 2011) using an elegant model of interacting agents endowed only
with general learning abilities, an ability to create signals of arbitrary type,
and a simple interactive social environment, was able to simulate not only the
emergence of compositionality but also consistent sequential order of lexical
items, again, solely through cultural evolutionary mechanisms. Spranger and
Steels (2012) further report the emergence of a communication system dis-
playing incipient hierarchical structure and grammatical marking for spatial
information in a simulation of robots playing a cooperative spatial identifica-
tion guessing game. This occurred even though these features had not been built
into the system to begin with.
Although no simulation studies have yet reported the emergence of anything
approaching the complexity of natural language syntax, the exploration of the
effects of increasing conceptual complexity on emergent syntax through cul-
tural evolution has only begun. These initial results are very promising, and
suggest a great deal is yet to be learned about the possibility of the evolution-
ary emergence of grammar from changes in the underlying conceptual system.
10 Conclusions
The model of language evolution presented here suggests that increasing com-
plexity of the meaning space during human cognitive evolution drove the devel-
opment of syntax and grammar through cultural evolutionary processes. Lan-
guage complexity is seen as a result of the complexity of the conceptual world
it evolved to map and communicate. This model is not only more parsimo-
nious than those requiring separate, independent genetic evolutionary scenar-
ios for both our conceptual system and our linguistic grammatical system, but
is also specifically supported by research in a number of areas of study. Our
understanding of human brain evolution, when placed in the context of how
concepts are instantiated in brains, leads to the conclusion that there has likely
been a dramatic increase in conceptual complexity during our evolutionary his-
tory. Apparent differences in primate cognition, deriving from both studies of
animals in the wild as well as those of captive animals, suggest this is true
as well. In particular, it is apparent that other animals (non-human primates
in particular) have concepts that meaningfully overlap with some of our own.
The difference appears to be one of degree, not of kind. The tight connection
between the size of the lexicon and the degree of grammatical complexity in
speech production in human children actually fits very well with the data from
studies of ape sign communication. Furthermore, several (perhaps all?) univer-
sal features of grammar found in languages around the world can be seen as
inevitable cultural-evolutionary conventions resulting from a common under-
lying conceptual system, rather than requiring independent language-grammar-
specific genetic constraints. Computer agent-based simulations have begun to
investigate possible ways in which this model could work. Future work stands
to flesh out these ideas, and demonstrate their full power in explaining the evo-
lution of language.
This process highlights language as a complex adaptive system (Beckner
et al. 2009). Grammar evolved through cultural evolution, making use of pre-
existing, non-linguistic general cognitive abilities, and was driven by increasing
complexity of underlying conceptual understanding played out in the context
of an intensely socially interactive existence. Each of the parts of this equation
were (and continue to be) influenced by the others, in a complex interactive
feedback system, thus forming a complex adaptive system. Increasing concep-
tual complexity was itself presumably driven by an increasingly complicated
social and technological existence. As social lives got more complicated, lead-
ing to new emergent social patterns (such as elaborated kinship systems and
social institutions), new forms of conceptual understanding of these emergent
patterns followed. Similarly, new technologies created new conceptual under-
standing. Technological advances, in addition to simply adding conceptually
new devices to be named, also changed the social dynamics themselves. Agri-
culture, for example, dramatically increased population density, which in turn
had profound effects on human sociality. More recently, technology applied in
the social domain (social media) appears to be in the process of further elabo-
rating and redefining sociality in a variety of interesting ways.
The usefulness of language, and grammar in particular, was likely also cen-
tral to these social and technological developments. Thus, language, conceived
in the broadest sense, itself facilitated further increases in conceptual complex-

ity. Grammar and conceptual understanding influenced each other’s evolution-
ary trajectories synergistically. They adapted to each other.
Although the model outlined here is consistent with what we know about
the evolution of brain structure and function, comparative studies of primate
cognition, child language learning, and with theories about language itself, the
details of how increasing conceptual complexity itself could have led to com-
plexity of language grammar remains to be described in detail. No doubt part
of the story involves understanding how grammar itself evolves (culturally),
as this suggests ways in which grammar could have emerged in the first place
(Bybee 1998). Further work on exactly how children form grammatical con-
cepts from the constructions they hear, and how their conceptual understanding
might guide this process, will also be critical. Additionally, if this model is cor-
rect, it must be possible to instantiate and probe its dynamics with agent-based
computer models of language evolution. The model predicts that elaborating
the underlying conceptual understanding of the agents should have profound
effects on the grammatical systems that emerge from the simulations. Such
simulations would not prove language evolution occurred exactly this way, but
simply provide proof-of-concept, and allow for a better understanding of what
is possible from this perspective.
Unraveling the mystery of language evolution is central to understanding the
origin and development of our species. The goal should be to explain as much as
possible using cultural evolutionary mechanisms, since this change will occur
faster, and will therefore be favored at each evolutionary time-step, compared
to biological adaptation (Christiansen & Chater 2008). Because shared concep-
tual understanding is the foundation that language communication is based on,
recognizing the importance of evolutionary changes in this system, and how
it would play out over evolutionary time in a socially interactive existence, is
likely to be fundamentally important to the explanation. Taking seriously lan-
guage evolution as a complex adaptive system, involving interactions among
many components, is a critical first step.
Acknowledgments
The ideas in this chapter have benefited from numerous conversations over the
years with William S.-Y. Wang, Vincent Sarich, James Hurford, Morten Chris-
tiansen, John Dolan, and John Holland. I thank Salikoko S. Mufwene for ini-
tially inviting me to the Workshop on Complexity in Language: Developmental
and Evolutionary Perspectives, Lyon, France, May 23–24, 2011, where a sketch
of these ideas was presented. Comments by the editors have helped improve the
text, though any errors that remain are of course my own.
REFERENCES
Alias-i. 2008. LingPipe 4.1.0. http://alias-i.com/lingpipe.
Baars, Bernard J. & Nicole M. Gage. 2007. Cognition, Brain, and Consciousness: Intro-
duction to Cognitive Neuroscience Amsterdam: Elsevier.
Barsalou, L. W. 2005. Continuity of the conceptual system across species. Trends Cogn
Sci 9, 309–11.
Barsalou, Lawrence W. 2010. Grounded Cognition: Past, Present, and Future. Topics in
Cognitive Science 2, 716–24.
Batali, J. 1998. Computational Simulations of the Emergence of Grammar. Approaches
to the Evolution of Language: Social and Cognitive Bases, ed. by J. R. Hurford, M.
Studdert-Kennedy & C. Knight, 405–26. Cambridge: Cambridge University Press.
Bates, Elizabeth & Judith C. Goodman. 1997. On the inseparability of grammar and the
lexicon: Evidence from acquisition, aphasia, and real-time processing. Language
& Cognitive Processes 12, 507–84.
Beckner, Clay, Richard Blythe, Joan Bybee, Christiansen Morten H., William Croft,
Nick C. Ellis, John Holland, L. C., Jinyun Ke, Diane Larsen-Freeman & Tom
Schoenemann. 2009. Language Is a Complex Adaptive System: Position Paper.
Language Learning 59, 1–26.
Bergman, Thore J., Jacinta C. Beehner, Dorothy L. Cheney & Robert M. Seyfarth. 2003.
Hierarchical classification by rank and kinship in baboons. Science 302, 1234–36.
Berwick, Robert C., Angela D. Friederici, Noam Chomsky & Johan J. Bolhuis. 2013.
Evolution, brain, and the nature of language. Trends in Cognitive Sciences 17, 89–
98.
Brodmann, K. 1909. Vergleichende Lokalisationslehre der Grosshirnrinde in ihren
Prinzipien dargestellt auf Grund des Zellenbaues Leipzig: Johann Ambrosius
Barth Verlag.
Bybee, Joan. 1998. A functionalist approach to grammar and its evolution. Evolution of
Communication 2, 249–78.
Changizi, M. A. & S. Shimojo. 2005. Parcellation and area-area connectivity as a func-
tion of neocortex size. Brain, Behavior and Evolution 66, 88–98.
Cheney, D. L. & R. M. Seyfarth. 1997. Reconciliatory grunts by dominant female
baboons influence victims’ behaviour. Animal Behaviour 54, 409–18.
Cheney, Dorothy L. & Robert M. Seyfarth. 2005. Constraints and preadaptations in the
earliest stages of language evolution. The Linguistic Review 22, 135–59.
Chomsky, Noam. 1972. Language and Mind New York: Harcourt Brace Jovanovich,
Inc.
1995. The Minimalist Program Cambridge, M A: The MIT Press.
Christiansen, Morten H. & Nick Chater. 2008. Language as shaped by the brain. Behav-
ioral and Brain Sciences 31, 489–509.
Clarke, E., U. H. Reichard & K. Zuberbuhler. 2006. The syntax and meaning of wild
gibbon songs. PLoS ONE 1.e73.
Damasio, Antonio R. & Hanna Damasio. 1992. Brain and language. Scientific American
267, 89–95.
Damasio, H., T. J. Grabowski, A. Damasio, D. Tranel, L. Boles-Ponto, G. L. Watkins
& R. D. Hichwa. 1993. Visual recall with eyes closed and covered activates early
visual cortices. Society for Neuroscience Abstracts 19, 1603.
de Waal, Frans. 1982. Chimpanzee Politics Baltimore: Johns Hopkins University Press.
de Waal, Frans B. M. 2008. Putting the Altruism Back into Altruism: The Evolution of
Empathy. Annual Review of Psychology 59, 279–300.
Deacon, Terrence William. 1997. The symbolic species: the co-evolution of language
and the brain. New York: W. W. Norton.
Deacon, Terrence W. 2012. Beyond the Symbolic Species. The Symbolic Species
Evolved, ed. by T. Schilhab, F. Stjernfelt & T. Deacon, 9–38. Netherlands: Springer.
Deaner, R. O., K. Isler, J. Burkart & C. van Schaik. 2007. Overall brain size, and
not encephalization quotient, best predicts cognitive ability across non-human pri-
mates. Brain, Behavior and Evolution 70, 115–24.
Evans, Nicholas & Stephen C. Levinson. 2009. The myth of language universals: Lan-
guage diversity and its importance for cognitive science. Behavioral and Brain
Sciences 32, 429–48.
Foraker, Stephani, Terry Regier, Naveen Khetarpal, Amy Perfors & Joshua Tenenbaum.
2009. Indirect Evidence and the Poverty of the Stimulus: The Case of Anaphoric
One. Cognitive Science 33, 287–300.
Gardner, R. A. & B. T. Gardner. 1984. A vocabulary test for chimpanzee (Pan
Troglodytes). Journal of Comparative Psychology 98, 381–404.
Gardner, R. A., B. T. Gardner & T. E. van Cantfort. 1989. Teaching Sign Language to a
Chimpanzee Albany: State University of New York Press.
Garstang, Michael 2010. Elephant infrasounds: long-range communication. Handbook
of Mammalian Vocalization; An Integrative Neuroscience Approach, ed. by S. M.
Brudzynski, 57–67. London: Academic Press.
Gibson, J. J. 1979. The ecological approach to visual perception New York: Houghton
Mifflin.
Gibson, K. R. 2002. Evolution of human intelligence: the roles of brain size and mental
construction. Brain, Behavior and Evolution 59, 10–20.
Gil-da-Costa, R., A. Martin, M. A. Lopes, M. Munoz, J. B. Fritz & A. R. Braun. 2006.
Species-specific calls activate homologs of Broca’s and Wernicke’s areas in the
macaque. Nature Neuroscience 9, 1064–70.
Gleitman, L. 1990. The structural sources of verb meanings. Language Acquisition 1,
3–55.
Gleitman, Lila & Henry Gleitman. 1997. What is a language made out of? Lingua:
International Review of General Linguistics 100, 29–55.
Gleitman, Lila R., Kimberly Cassidy, Rebecca Nappa, Anna Papafragou & John C.
Trueswell. 2005. Hard words. Language Learning and Development 1, 23–64.
Goldberg, Adele E. 2004. But do we need universal grammar? Comment on Lidz et al.
(2003). Cognition 94, 77–84.
Goldsmith, Timothy H. & William F. Zimmerman. 2000. Biology, Evolution, and
Human Nature New York: John Wiley & Sons.
Gong, T. 2009. Computational simulation in evolutionary linguistics: a study on lan-
guage emergence Taipei: Institute of Linguistics, Academia Sinica.
Gong, Tao. 2011. Simulating the coevolution of compositionality and word order regu-
larity. Interaction Studies 12, 63–106.
Gong, Tao, Lan Shuai & Menghan Zhang. 2014. Modelling language evolution: Exam-
ples and predictions. Phys Life Rev 11, 280–302.
Goroll, Nils. 1999. Master’s thesis, (The Deep Blue) Nile: Neuronal Influences on Lan-
guage Evolution. Edinburgh, England: University of Edinburgh.
Greenfield, Patricia M. & E. Sue Savage-Rumbaugh. 1991. Imitation, grammatical

development, and the invention of protogrammar by an ape. Biological and
Behavioral Determinants of Language Development, ed. by N. Krasnegor, D. M.
Rumbaugh, R. L. Schiefelbusch & M. Studdert-Kennedy, 235–58. Hillsdale, NJ:
Lawrence Erlbaum.
Gross, C. G. 2002. Genealogy of the “Grandmother Cell.” The Neuroscientist 8, 512–18.
Haiman, John (ed.) 1985. Iconicity in Syntax (Typological Studies in Language). Ams-
terdam: John Benjamins.
Haug, H. 1987. Brain sizes, surfaces, and neuronal sizes of the cortex cerebri: A stere-
ological investigation of man and his variability and a comparison with some
mammals (primates, whales, marsupials, insectivores, and one elephant). Ameri-
can Journal of Anatomy 180, 126–42.
Herculano-Houzel, S. 2012. The remarkable, yet not extraordinary, human brain as
a scaled-up primate brain and its associated cost. Proceedings of the National
Academy of Sciences, 109 (Supplement 1), 10661–10668.
Herrnstein, R. J. 1979. Acquisition, generalization, and discrimination reversal of a nat-
ural concept. Journal of Experimental Psychology: Animal Behavior Processes 5,
116–29.
Holland, John H. 1998. Emergence: From chaos to order. Oxford: Oxford University
Press.
2006. Studying complex adaptive systems. Journal of Systems Science and Complex-
ity 19, 1–8.
Horgan, John. 1995. From complexity to perplexity. Scientific American, 104–09.
Hudson, Arthur J. 2010. The evolution of the eye from algae and jellyfish to humans:
How vision adapts to environment. Lewiston, NY: Edwin Mellen Press.
Jackendoff, Ray. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution
New York: Oxford University Press.
Jerison, H. J. 1973. Evolution of the Brain and Intelligence New York: Academic Press.
Jones, G. & M. W. Holderied. 2007. Bat echolocation calls: adaptation and convergent
evolution. Proc Biol Sci 274, 905–12.
Jones, M. N. & D. J. Mewhort. 2007. Representing word meaning and order information
in a composite holographic lexicon. Psychological Review 114, 1–37.
Kosslyn, S. M. 1980. Image and mind. Cambridge, MA: Harvard University Press.
Kosslyn, S. M., N. M. Alpert, W. L. Thompson, V. Maljkovic, S. B. Weise, C. F. Chabris,
S. E. Hamilton, S. L. Rauch & F. S. Buonanno. 1993. Visual mental imagery acti-
vates topographically organized visual cortex: PET investigations. Journal of Cog-
nitive Neuroscience 5, 263–87.
Kosslyn, S. M., G. Ganis & W. L. Thompson. 2003. Mental imagery: against the nihilis-
tic hypothesis. Trends Cogn Sci 7, 109–11.
Krubitzer, Leah. 1995. The organization of neocortex in mammals: are species differ-
ences really so different? Trends in Neurosciences 18, 408–17.
Lakoff, George & Mark Johnson. 1980. Metaphors we live by. Chicago: University of
Chicago Press.
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Stanford: Stanford
University Press.
Langton, Chris G. 1990. Computation at the edge of chaos: Phase transitions and emer-
gent computation. Physica D 42, 12–37.
Lidz, Jeffrey, Henry Gleitman & Lila Gleitman. 2003a. Understanding how input mat-
ters: Verb learning and the footprint of universal grammar. Cognition 87, 151–78.
Lidz, Jeffrey & Lila R Gleitman. 2004a. Yes, we still need Universal Grammar. Cogni-
tion 94, 85–93.
Lidz, Jeffrey & Lila R. Gleitman. 2004b. Argument structure and the child’s contribution
to language learning. Trends in Cognitive Sciences 8, 157–61.
Lidz, Jeffrey, Sandra Waxman & Jennifer Freedman. 2003b. What infants know about
syntax but couldn’t have learned: experimental evidence for syntactic structure at
18 months. Cognition 89, 295–303.
Lyn, Heidi & E. Sue Savage-Rumbaugh. 2000. Observational word learning in two
bonobos (Pan paniscus): ostensive and non-ostensive contexts. Language & Com-
munication 20, 255–73.
McGurk, H. & J. MacDonald. 1976. Hearing lips and seeing voices. Nature 264,
746–8.
McIntosh, A. R. 2000. Towards a network theory of cognition. Neural Networks 13,
861–70.
Mikulecky, D. C. 2001. The emergence of complexity: science coming of age or science
growing old? Computers and Chemistry 25, 341–8.
Miles, H. Lyn White. 1990. The cognitive foundations for reference in a signing
orangutan. “Language” and Intelligence in Monkeys and Apes: Comparative
Developmental Perspectives, ed. by S. T. Parker & K. R. Gibson, 511–39. Cam-
bridge: Cambridge University Press.
Miller, George A. & Patricia M. Gildea. 1991. How children learn words. The Emer-
gence of Language: Development and Evolution, ed. by W. S.-Y. Wang, 150–58.
New York: W. H. Freeman.
Miller, P. E. & C. J. Murphy. 1995. Vision in dogs. Journal of the American Veterinary
Medical Association 207, 1623–34.
Mitchell, Melanie, Peter T. Hraber & James P. Crutchfield. 1993. Revisiting the Edge of
Chaos: Evolving Cellular Automata to Perform Computations. Complex Systems
7, 89–130.
Neitz, J., T. Geist & G. H. Jacobs. 1989. Color vision in the dog. Visual Neuroscience
3, 119–25.
Northcutt, R. G. & J. H. Kaas. 1995. The emergence and evolution of mammalian neo-
cortex. Trends in Neurosciences 18, 373–9.
O’Grady, William. 1987. Principles of Grammar & Learning Chicago: University of
Chicago Press.
Paivio, A. 1971. Imagery and verbal processes New York: Holt, Rinehart & Winston.
Parr, Lisa A. 2011. The evolution of face processing in primates. Philosophical Trans-
actions of the Royal Society B: Biological Sciences 366, 1764–77.
Peirce, C. S. 1867. On a new list of categories. Proceedings of the American Academy
of Arts and Sciences 7, 287–98.
Penfield, Wilder & Theodore Rasmussen. 1950. Cerebral Cortex of Man, a Clinical
Study of Localization of Function. New York: Macmillan.
Pinker, S. & R. Jackendoff. 2005. The faculty of language: what’s special about it?
Cognition 95, 201–36.
Pinker, Steven & Paul Bloom. 1990. Natural language and natural selection. Behavioral
and Brain Sciences 13, 707–84.
Port, Robert F. 2010. Language as a Social Institution: Why Phonemes and Words Do
Not Live in the Brain. Ecological Psychology 22, 304–26.
Port, Robert F. & Adam P. Leary. 2005. Against Formal Phonology. Language 81, 927–
64.
Premack, Ann James & David Premack. 1972. Teaching language to an ape. Scientific
American 227, 92–99.
Regier, Terry & Susanne Gahl. 2004. Learning the unlearnable: the role of missing evi-
dence. Cognition 93, 147–55.
Rilling, J. K. & R. A. Seligman. 2002. A quantitative morphometric comparative anal-
ysis of the primate temporal lobe. Journal of Human Evolution 42, 505–33.
Ringo, James L. 1991. Neuronal interconnection as a function of brain size. Brain,
Behavior and Evolution 38, 1–6.
Roberts, William A. & Dwight S. Mazmanian. 1988. Concept learning at different levels
of abstraction by pigeons, monkeys, and people. Journal of Experimental Psychol-
ogy: Animal Behavior Processes 14, 247–60.
Savage-Rumbaugh, E. Sue, Jeannine Murphy, Rose A. Sevcik, Karen E. Brakke, Shelly
L. Williams & Duane M. Rumbaugh. 1993. Language comprehension in ape and
child. Monographs of the Society for Research in Child Development 58, 1–222.
Savage-Rumbaugh, E. Sue & Duane M. Rumbaugh. 1993. The emergence of language.
Tools, Language and Cognition in Human Evolution, ed. by K. R. Gibson & T.
Ingold, 86–108. Cambridge: Cambridge University Press.
Savage-Rumbaugh, Sue, Duane Rumbaugh & William M. Fields. 2009. Empirical
Kanzi: The Ape Language Controversy Revisited. Skeptic 15, 25–33.
Schoenemann, P. T., T. F. Budinger, V. M. Sarich & W. S. Wang. 2000. Brain size does
not predict general cognitive ability within families. Proceedings of the National
Academy of Sciences of the United States of America 97, 4932–7.
Schoenemann, P. Thomas. 1999. Syntax as an emergent characteristic of the evolution
of semantic complexity. Minds and Machines 9, 309–46.
2005. Conceptual complexity and the brain: Understanding language origins. Lan-
guage Acquisition, Change and Emergence: Essays in Evolutionary Linguistics,
ed. by W. S.-Y. Wang & J. W. Minett, 47–94. Hong Kong: City University of Hong
Kong Press.
2006. Evolution of the Size and Functional Areas of the Human Brain. Annual Review
of Anthropology 35, 379–406.
2010. The meaning of brain size: the evolution of conceptual complexity. Human
Brain Evolving: Papers in Honor of Ralph Holloway, ed. by D. Broadfield, M.
Yuan, K. Schick & N. Toth. Bloomington, IN: Stone Age Institute Press.
2012. Evolution of brain and language. Progress in Brain Research, ed. by M. A.
Hofman & D. Falk, 443–59. Amsterdam, The Netherlands: Elsevier.
Semendeferi, K., E. Armstrong, A. Schleicher, K. Zilles & G. W. Van Hoesen. 2001.
Prefrontal cortex in humans and apes: a comparative study of area 10. American
Journal of Physical Anthropology 114, 224–41.
Seyfarth, Robert M. & Dorothy L. Cheney. 2000. Social Awareness in Monkeys. Amer-
ican Zoologist 40, 902–09.
Seyfarth, Robert M., Dorothy L. Cheney & Peter Marler. 1980. Monkey Responses to
Three Different Alarm Calls: Evidence of Predator Classification and Semantic
Communication. Science 210, 801–03.
Shepard, R. N. & L. A. Cooper. 1982. Mental images and their transformations New
York: Cambridge University Press.
Shepherd, Gordon M. 2006. Smell images and the flavour system in the human brain.
Nature 444, 316–21.
Smith, K., S. Kirby & H. Brighton. 2003. Iterated learning: a framework for the emer-
gence of language. Artificial Life 9, 371–86.
Snowdon, Charles T. 1990. Language capacities of nonhuman animals. Yearbook of
Physical Anthropology 33, 215–43.
Spranger, Michael & Luc Steels. 2012. Emergent functional grammar for space. Exper-
iments in Cultural Language Evolution. Amsterdam: John Benjamins.
Steels, L. 2011. Modeling the cultural evolution of language. Phys Life Rev 8, 339–56.
Stephan, Heinz, Heiko Frahm & Georg Baron. 1981. New and revised data on volumes
of brain structures in Insectivores and Primates. Folia Primatologica 35, 1–29.
Suomi, Stephen J. 1997. Nonverbal communication in nonhuman primates: Implications
for the emergence of culture. Nonverbal Communication: Where Nature Meets Cul-
ture, ed. by U. C. Segerstråle & P. Molnár, 131–46. Mahwah, NJ: Lawrence Erl-
baum Associates.
Thompson, P. M., T. D. Cannon, K. L. Narr, T. van Erp, V. P. Poutanen, M. Huttunen,
J. Lonnqvist, C. G. Standertskjold-Nordenstam, J. Kaprio, M. Khaledy, R. Dail,
C. I. Zoumalan & A. W. Toga. 2001. Genetic influences on brain structure. Nature
Neuroscience 4, 1253–8.
Tomasello, M. 2000. The item-based nature of children’s early syntactic development.
Trends Cogn Sci 4, 156–63.
Tomasello, Michael. 2003. Constructing a language: a usage-based theory of language
acquisition Cambridge, MA: Harvard University Press.
2004. Syntax or semantics? Response to Lidz et al. Cognition 93, 139–40.
Tomasello, Michael & J. Brooks Patricia. 1998. Young children’s earliest transitive and
intransitive constructions. Cognitive Linguistics 9, 379.
Tononi, G. 2010. Information integration: its relevance to brain function and conscious-
ness. Archives Italiennes de Biologie 148, 299–322.
Wasserman, E. A., R. E. Kiedinger & R. S. Bhatt. 1988. Conceptual behavior in pigeons:
Categories, subcategories, and pseudocategories. Journal of Experimental Psychol-
ogy: Animal Behavior Processes 14, 235–46.
Wynne, Clive. 2008. Aping language: A skeptical analysis of the evidence for nonhuman
primate language. Skeptic 13, 10–14.
Zuberbuhler, K. 2000. Interspecies semantic communication in two forest primates. Pro-
ceedings of the Royal Society of London Series B-Biological Sciences 267, 713–18.
5 Evolutionary Complexity of Social Cognition,
Semasiographic Systems, and Language
William Croft
1 Introduction
After a long period in which the study of the evolutionary origins of the human
language capacity was avoided, some linguists now speculate or use compu-
tational modeling to try to infer the evolutionary process that led to modern
human language. It is assumed that the evolution of language represents an
increase in some sort of linguistic complexity: “pre-language” was less com-
plex than modern human language in some ways, and – probably gradually –
acquired the sort of complexity that modern human languages display.
Complexity in language is frequently measured in two ways: structural com-
plexity of communicative signals, and the social-cognitive complexity of the
interactional situations in which language or language-like communication is
used. Some linguists are now attempting to measure structural complexity of
contemporary human languages, after a period in which it was assumed that
all modern languages were equal in overall structural complexity. There are
undoubtedly significant differences among contemporary human languages in
terms of the obligatory expression of certain grammatical semantic categories,
and the formal morphological complexity of that expression. Nevertheless,
all attested human languages display similar degrees of structural complex-
ity in “design features” of language (Hockett 1960). These design features
include the combinability and recombinability of meaningful units; some kind
of hierarchical structure; recursion (but see Everett 2005, 2009), and duality of
patterning.
Many of the design features of modern human languages, which presumably
evolved from a lesser degree of complexity, are not specific to human language,
but rather iconically reflect certain aspects of human cognition, which may or
may not be specific to our species (see also Schoenemann 1999). In fact, even
the categories of obligatorily expressed grammatical inflections and construc-
tions, although not always obligatorily expressed in every language, are the
result of the processes by which we verbalize our experience (Chafe 1977a,
1977b; Croft 2007a); these processes will not be discussed here for reasons of
space.
101
102 William Croft
Morphosyntactic complexity largely reflects conceptual complexity. Kirby

(2002) argues that in an iterated learning model of language evolution, a gram-
mar made up of recombining, hierarchical, and recursive units can emerge,
given only a predicate-calculus-like semantic representation and a preference
for an iconic mapping between form and meaning. But it is really the seman-
tic representation and semantics-morphosyntax mapping that brings about the
characteristic syntactic structures of modern human languages. The cogni-
tive or conceptual capacities that lead to these morphosyntactic structures are
an ability for conceptual analysis and compositionality; conceptual grouping
(Langacker 1997); and a preference for diagrammatic iconicity in symbolic
(form-meaning) mapping (Bybee 1985; Haiman 1980).
Recursive grammatical structures, another type of structural complexity
found in most, if not all, modern human languages, commonly occurs in two
semantic contexts. Noun-phrase and adpositional-phrase recursion is found in
nested figure-ground relations specifying spatial locations (and metaphorical
extensions thereof), for example, the bowl in the cabinet above the stove. Sen-
tence recursion is found in certain constructions when events are functioning
as modifiers (relative clauses), arguments (verb complements), and adjuncts
(adverbial clauses). In particular, mental space builders (Fauconnier 1985) such
as believe, want, and hope introduce states of affairs in the “built” mental
space – the space of beliefs, desires, hopes, and so on – and these are usu-
ally expressed structurally in terms of embedded clauses. So the emergence
of recursion presupposes the emergence of these relatively complex cognitive
structures. There are in fact other means of packaging these conceptual struc-
tures in sentences that do not necessarily involve embedding. But presumably
the conceptualization of displaced experience and of other people’s minds is
a basic human cognitive ability (see Corballis 2011, esp. chs. 7–9). Also, it
should be noted that recursion is not necessarily infinite, as assumed in formal
syntactic theories. Spoken language rarely exceeds two levels of embedding
(e.g., Croft 1995, 2007b). So the emergence of syntactic embedding may be
gradual.
On the other hand, duality of patterning does not have a basis in the more-
or-less iconic expression of cognitive or conceptual structures. Duality of pat-
terning is defined here as the fact that utterances are made up of combina-
tions of sounds and simultaneously made up of combinations of symbolic
units. (See Mufwene 2013 for discussion of the different ways that duality of
patterning has been defined.) Duality of patterning is presumably a response
to selectional pressures brought about by the expansion of the inventory of
signals. Unfortunately it is unclear at what point in language evolution the
increase in the number of signals might lead to duality of patterning. I argue
in §4.5 that language probably began as a restricted, domain-specific commu-
nication system but then evolved to a general-purpose communication system.
Evolutionary Complexity of Social Cognition, Semasiographic Systems 103
A general-purpose communication system has to have the ability to express

any situation that arises and any type of entity that comes along. But duality
of patterning is a response to the means to express a large number of situa-
tions. Duality of patterning could have emerged at a pre-language stage lack-
ing word combination (a one-word stage). Phonological recombination would
facilitate production (and remembering) of a larger vocabulary. But duality
of patterning could also have occurred at a stage when multiword utterances
occur. Multiword utterances reduces the need for a larger vocabulary, but the
continued expansion of language to a general-purpose communication system
still requires a large vocabulary, and perhaps at this point duality of patterning
emerged.1
I have argued here that at least some elements of the structural complexity
of modern human languages are the consequence of the cognitive complex-
ity of the conceptual structures being communicated. The leap from concep-
tual structures to linguistic structures is not an automatic step. The dependence
of linguistic structures on conceptual structures only means that the relevant
cognitive/conceptual structures must be in place for the formal linguistic struc-
tures to emerge. There still remains the mechanism for conceptual structures
to be verbalized in language or a pre-language communication system. That
is, there has to be a selectional pressure to lead humans to express concepts
publicly, with something like language. That selectional pressure is provided
by language’s supporting role in the achievement of joint action, which is the
foundation of human culture and society. It is only in its social interactional
context that the evolution of linguistic complexity can be understood. (Indeed,
some might argue that selectional pressures from social interaction led to the
evolution of the conceptual structures described earlier, as well as leading to
the linguistic structures to communicate them.) That is, the evolution of social-
cognitive complexity (in terms of joint action) is a prerequisite for the evolution
of structural complexity of linguistic signals.
The nature of joint action and language’s role in joint action is described
in §2. Joint action presupposes a rich and complex social structure, even if it
is “merely” the most basic structure for the emergence of human culture. In
§3 I offer some cautionary examples to suggest that the evolution of linguistic
complexity may be both gradual and slow. In §4, we look at the nearest sort of
communicative system to language whose evolution is documented, namely the
evolution of semasiographic systems. Commonalities among the evolutionary
paths of different semasiographic systems provide further suggestions about
1 It has been argued that individual speakers do not have that large a vocabulary; e.g., Cheng (2000)
has suggested that individuals have knowledge of around 8,000 words. A speech community’s
lexical stock is quite a bit larger, but it may imply that there are other selectional pressures leading
to duality of patterning than simply the communicative range of a general-purpose communica-
tion system.
104 William Croft
the evolutionary path of modern human language. In §5, I offer speculations

as to how modern human language may have gradually evolved, based on the
observations in §§2–4.
2 Language as Joint Action

Language is fundamentally a joint action. Actually, language is just a part of the
process of joint action, but it is language as joint action that determines most
of the fundamental design features of language. To describe language as joint
action is another way of saying that language can only be properly understood
in its social interactional context. The nature of language follows from the part
it plays in joint action.
The structure of joint actions has been analyzed by social psychologists such
as Clark (1996) and philosophers of action such as Bratman (1992); I follow
Bratman’s analysis here. Loosely, what makes a joint action joint is that it is
more than just the sum of individual actions performed by separate persons;
in particular each individual involved must take into consideration the other
individual’s beliefs, intentions, and actions in a way that can be described as
cooperative. It is this level of cooperation, and the features it entails that are
described later, that appears to be a distinctive characteristic of human behav-
ior and has allowed for the emergence of human culture and society (Tomasello
2008). Although great apes engage in some degree of sharing, reciprocity, and
collaboration, it appears to be only the first steps beyond essentially compet-
itive social interactions (see Tomasello 2011; Tomasello and Vaish 2013, and
references cited therein).
Human beings of course engage in all sorts of actions, including actions
that involve other human beings. However, an action involving more than one
person must satisfy certain conditions in order to count as a joint action – in
Bratman’s terms, shared cooperative activity. Each participant, for example
each of two people performing a musical piece, intends to perform the joint
action, that is, their intention extends beyond just their own individual actions
as part of the joint action. The joint action is intended by each person to be per-
formed in accordance with and because of their meshing subplans: that is, the
overall intention of performing a musical piece requires each individual to have
subplans for contributing to the overall joint action that mesh with the other’s
subplans. Neither individual is coercing the other; the cooperation must be vol-
untary. Each participant is also committed to mutual support: that is, if the other
participant falters in some way, the first participant will step in to help the other
perform her part and thereby allow the joint action to be successfully achieved.
Two other conditions are essential. All of these intentions and commitments
on the part of the participants are shared knowledge among them, or as I will say
following Clark (1996), it is part of the participants’ common ground. Finally,
in the execution of the joint action, there is mutual responsiveness of the partic-
ipants. That is, in executing the joint action we will coordinate our individual
actions in order to ensure that they mesh with each other in execution and hence
the joint action will be successfully carried out (to the best of our abilities).2
Coordination is essential in carrying out joint actions successfully. Yet coor-
dination of our individual actions to achieve a joint action is a problem, because
of another fundamental characteristic of human beings: we cannot read each
other’s minds directly or literally; we can only interpret the external behaviors
of others. This is not always a bad thing, as a moment’s thought will demon-
strate. But much of the time, we want or need to carry out joint actions, so we
must find a way to overcome this problem. The means to solve coordination
problems are coordination devices (Lewis 1969).
There are a number of coordination devices that people use to solve coordi-
nation problems. One of the most effective coordination devices is communi-
cation: we communicate to each other our intentions, thereby making them part
of our common ground and allowing us to carry out our joint action. What pre-
cisely is communication? According to Grice, communication happens when
one of us recognizes the other’s intentions. Communication is itself a joint
action, however, as its Latinate etymology (commūnicāre ‘to make common’)
implies. One person has to signal her intention and the other person has to rec-
ognize the intention. In other words, communication solves one coordination
problem, but poses another coordination problem in order for communication
itself to succeed.
Fortunately, there is an effective coordination device for communication,
although it must satisfy specific conditions in order to be usable. This is con-
vention, which has been analyzed by both Lewis (1969) and Clark (1996). The
Lewis and Clark definition of convention has five parts: (i) a regularity in behav-
ior, (ii) which is partially arbitrary (iii) and is common ground in a community,
(iv) which serves as a coordination device (v) for a recurrent coordination prob-
lem. For example, shaking right hands is a regularity in behavior that is partially
arbitrary (one could shake left hands, or kiss on the cheek, etc.), and it is com-
mon ground in the American community that shaking hands serves as a coor-
dination device for the recurrent coordination problem of greeting someone.
Many human behaviors are conventional. In addition to conventions for
greeting someone, there is the convention of driving on the right, or left, side
of the road and the convention of paying for a meal before, or after, eating it.
And language is a vast inventory of conventions. Every word and construction
is a regularity in behavior – producing those words and constructions in utter-
ances – that solves a recurrent coordination problem, namely communicating
2 The concept of coordination in joint action is unrelated to the concept of coordination in syntactic
structure.
106 William Croft
a particular intention or meaning. So language involves conventions for com-

munication for the achievement of joint actions.
Powerful as convention is as a coordination device, it is not enough. Con-
vention can work only under certain circumstances. A convention must become
established in a community, that is, become common ground in that community.
Hence it cannot be the first use of this coordination device (the particular lin-
guistic form) for this coordination problem (the particular meaning intended).
For a convention to be established, the behavior must be used first to solve the
coordination problem via nonconventional coordination devices (Lewis 1969).
One type of nonconventional coordination device is precedent: a behavior
serves as a coordination device because it served before and was successful.
For example, a new word was coined by a commentator in The Economist in its
Lexington column in the May 10, 2003, issue: “Laugh at Bill Bennett, the erst-
while virtuecrat, but don’t forget his message. …” Assuming that the reader
understood what the commentator meant, he uses the new word again later in
the column: “Who needs satire when you have the social conservatives? … Now
Bill Bennett, the capo di tutti capi of the virtuecrats, has been caught … with
his hand glued to the slot machines of Las Vegas and Atlantic City.”
Precedent is not yet convention. It operates on the principle of “Hey, it
worked once, let’s try it again!” Although it is logically possible that precedent
does not even require shared knowledge of the precedent between the individu-
als, the definitions of precedent found in the literature on coordination devices
(Lewis 1969: 36; Clark 1996: 81) are formulated as shared precedent: it worked
for us once, so it might work for us again. I believe that the concept of shared
precedent is the one that is relevant to human behavior and human evolution
(see §5).
Coordination via precedent, of course, requires the precedent, and that prece-
dent has to succeed via some other coordination device. And at this point we
reach the grounding of all joint action: joint salience, made possible by the exis-
tence of joint attention among human beings (Tomasello 1999). For example,
while I was driving in the rain a few years ago, my mother, who was a passen-
ger, said, “You need more wiper” – a novel use of the count noun wiper in a
mass noun construction, not a convention of English. Her novel coordination
device (more wiper) succeeded because our joint attention was focused on the
contextually salient fact that the windshield was becoming obscured because
the wiper wasn’t going fast enough, and on our shared knowledge that one can
adjust the speed of the wiper.
Joint salience is required for all utterances, for at least two fundamental
properties of utterances (Clark 1996). First, almost every word in every utter-
ance has multiple senses, leading to denotational ambiguity (if this is doubted,
just examine a comprehensive dictionary like the Oxford English Dictionary),
and that ambiguity has to be resolved “in context,” that is, via joint salience.
Second, almost every utterance is communicating a specific event happening to

a specific person or thing: I broke the teapot is about a specific speaker, a spe-
cific teapot, and a specific event at a specific time (expressed by the past tense
verb form). Yet the words used – I, break, teapot – are linguistic conventions for
general types, not specific tokens of the type. Using a word for a general type
to refer to one specific individual and not another only succeeds “in context,”
that is, via joint salience.
Linguistic convention is not yet enough in another way. Words and construc-
tions must be realized in a jointly perceptually salient form: auditory, as in
speech, or visual, as in signed language or writing. Only at this point is language
fully grounded in joint salience. Thus, the full joint action using language has
four levels (Clark 1996) (the paired actions are a reminder that all levels are
joint):
r proposing and taking up a joint action, via
r signaling and recognizing the communicative intention, via
r formulating and identifying a conventional proposition, via
r producing and attending to a perceivable utterance.
Central to joint action and to convention, including linguistic convention, is
common ground: shared knowledge, beliefs, and attitudes of members of the
community. Common ground requires a shared basis (Clark 1996). Some com-
mon ground is shared between persons because of common experiences in their
face-to-face interaction: things they have seen or done together. The shared
basis for that common ground is joint perceptual salience. Other common
ground is shared between persons because of what they have communicated
to each other over the course of their interactions, even if it wasn’t experienced
together. The shared basis for that common ground is the actional basis – their
communicative acts, including linguistic acts. Like the shared basis of joint
perceptual salience, the actional basis of shared communication requires direct
interaction between the persons. Clark calls common ground founded on per-
ceptual and actional basis personal common ground.
A third basis for common ground does not require face-to-face interaction.
We share some knowledge simply by being members of the same community:
chess players, birders, environmentalists, Mormons, linguists, University of
New Mexico employees, and so on. By virtue of membership in these commu-
nities, we have shared expertise that serves as a communal basis for common
ground. For instance, as chess players, we can assume knowledge of the rules
of chess and many strategies and moves in playing the game. But that is stating
it a bit too glibly: where does that shared expertise come from?
Shared expertise is also not a state, but a process as well: the communi-
ties we belong to are communities of shared practices. Wenger (1998) defines
communities of practice as possessing mutual engagement (i.e., joint action),
a joint enterprise (i.e., the purpose for the joint action), and a shared repertoire
108 William Croft
(i.e., commonly performed joint actions to carry out the joint enterprise).
Shared expertise emerges from communities of practice. Conversely, com-
munities persist and expand via the transmission of shared practices and
shared expertise to new members. Even personal common ground is ultimately
grounded in a process: the shared experiences and communications between
the persons.
Language is joint action, but it is really part of a joint action that extends
far beyond what linguists usually think of as language. Two persons wish to
engage in a joint action. This requires a shared intention, meshing subplans,
absence of coercion, and mutual support. Above all, it requires coordination
of the individual actions of the persons. Coordination of a joint action can be
achieved at least in part by joint salience. For example, two musicians look
at each other to coordinate their playing. The joint action itself, like its coor-
dination, requires common ground, which also emerges from joint salience
but also a history of shared practice in the community to which the partici-
pants belong. Much joint action, certainly many sophisticated ones, use com-
munication for coordination. Again, joint salience can facilitate communica-
tion. But convention, particularly linguistic convention, is highly effective and
widely used – though only after joint salience and precedent have allowed
the establishment of the convention in the speech community, and only when
supplemented by precedent and joint salience. Finally, linguistic convention
must ultimately be realized in a perceivable utterance, whose joint percep-
tual salience allows the linguistic convention to successfully serve its purpose.
Language is only one part – albeit a very useful part – serving a much broader
process.
3 The Evolution of Linguistic Complexity

The description of joint action and social cognition in §2 describes joint action
in modern human social interaction. It looks very much like a process that is
tightly integrated, where each element depends on every other element. It seems
that once the basic elements of joint action have evolved – common ground,
helpfulness, and so on – all the rest will emerge automatically, and modern
human language will therefore emerge with it. It is a little easier to conceive
of a gradual evolution of the complexity of the signal described in §1 (and
perhaps this is why so much attention has been focused on it). But one cannot
separate the evolution of the complexity of the signal from the evolution of
social cognitive complexity.
Nevertheless, we shouldn’t overestimate how rapidly or suddenly a complex
suite of social cognitive skills can evolve. Examples from the evolution of sema-
siographic systems – which encode information in a lasting, visual medium, of
which more is discussed in §4 – and the ontogenetic development of social
cognition show that hindsight is too powerful to form the basis of an evolution-
ary scenario.
For example, it is universally agreed that humans had fully modern language
by fifty thousand years ago, if not much earlier. Around that same time (fifty
thousand years ago if not earlier; see for example d’Errico et al., 2001), repre-
sentational art appears in the archaeological record, including artistically quite
sophisticated representations. One might think that putting the two together –
auditory human language and visual representation – is not an enormous next
step. Yet although all human societies have modern human language and repre-
sentational art, it took tens of thousands of years before the first writing systems
emerged; and the evolution of writing systems happened independently twice
or at most four times, out of the thousands of human societies that must have
existed five thousand years ago. Clearly, putting the two together for the first
time was not easy at all. The enormous time lag was probably due to the absence
of selection pressure, which probably arrived in the form of large-scale strati-
fied societies where writing emerged. Even so, the actual emergence of writing
appears to have been a gradual process where it is documented (see §4.2).
Another far more mundane example illustrates the difficulty and gradual-
ness of evolution of semasiographic systems, which probably is characteristic
of all human invention. Musical notation, another semasiographic system that
will be described in §4.3, had evolved considerably by the seventeenth cen-
tury. This included representation of notes of different time values (longer and
shorter), conventionally described as half-notes, fourth-notes, and so on. By the
seventeenth century musical notation represented irregular note lengths, such
as a note lasting for three-sixteenths followed by a single sixteenth note (3 +
1), by adding a dot after the longer note: in this case, an eighth note followed
by a dot denoted three-sixteenths (“one and a half eighth notes”), and it was
followed by a sixteenth note. At the same time, a larger note value could be
divided into three shorter notes instead of two notes or four notes. But it wasn’t
until after the middle of the nineteenth century that a notation was devised
for describing a 2 + 1 combination, even though composers wrote music with
2 + 1 note values. Instead the 3 + 1 notation was used (for example, in music
as late as Chopin’s Polonaise-Fantaisie, published in 1846). Even though there
was selectional pressure to fix this notational problem – composers produced
both the 2 + 1 and 3 + 1 note value combinations frequently in their composi-
tions – it still took some two centuries for it to be solved.
The moral is that even when the cognitive parts are there – that is, they are
part of human cognitive ability at that point in human history – putting the parts
together to produce a new behavior does not happen automatically. In fact, it
may take a long time before it even starts to happen at all. There has to be a
reason for it to happen, that is, a selectional pressure in evolutionary terms – and
even then, it takes a long time, even multiple human generations, for the new
110 William Croft
behavior to emerge and be propagated in the society. The selectional pressure

is most likely to be social, not purely cognitive.
An example that suggests that social cognitive evolution is also gradual is the
ontogenetic development of the human conceptualization of others. Tomasello
argues that in ontogenetic development, a “revolution” occurs at around nine
months of age, which leads to the ability of the infant to achieve joint attention
with other human beings (Tomasello 1999: 61–70). But other types of mental
reasoning that are built on the notion of other minds having beliefs other than
our own (called the “theory of mind”) take several further years to develop (e.g.,
Tomasello, Kruger, and Ratner 1993: 499–500).
The conclusion we can draw from these various examples is that it is emi-
nently plausible to infer that the evolution of social cognition as described in §2
took some time, possibly a very long time – hundreds of thousands if not some
millions of years. Tomasello suggests that although primates and some other
animal species appear to engage in joint cooperative activity, the apparently
“coordinated” behavior may be due to individual responses to causal infer-
ences based on practical reasoning about conspecifics’ intentions and goals
(Tomasello 2008: 44–49, 173–85). Hence, it may have taken up to five million
years (from the branching off of the hominin line to no later than fifty thousand
years ago) for modern human social cognition to have evolved.
Unfortunately, there is no direct empirical evidence regarding intermediate
stages of complexity in human cognition between primate group activities and
signals and modern human joint actions and language. This is the reason that
scholars avoided the question of modern human origins until recently. In the
next section, I will look at the evolution of semasiographic systems (the near-
est language-like systems for which we have direct empirical evidence) to see
what they might tell us about stages in the evolution of language (cf. Coupé
2012, who compares the development of language to the development of pho-
tography). After taking a glance at language acquisition I will speculate on the
evolution of language from pre-language in §5, using the logical priority of
different elements of the model of joint action described in §2 in combination
with the insights from the evolution of semasiographic systems.
4 The Evolution of Semasiographic Systems

Semasiographic systems are used for “the communication of relatively spe-
cific ideas in a conventional manner by means of permanent, visible marks”
(Boone 1994: 15; 2004: 313). Semasiographic systems include images, numer-
als (counting), maps, calendars, bookkeeping systems, tables, mathematical
notation, music and dance notation, graphs, and, as a special case, writing. As
noted earlier, semasiographic systems have been around for at least fifty thou-
sand years, with the first production of images and statues. Semasiographic
systems are similar to language in several respects. As noted before, they

encode information, as linguistic expressions do. More specifically, they
encode information in such a way that it can be shared, just as language involves
sharing individuals’ intentions. Semasiographic systems employ, or evolve to
employ, relatively complex information encoded by recombinable units, also
like modern human language. Most useful for the goals of this chapter, sema-
siographic tokens survive and leave a record, allowing us to observe stages
of their evolution in at least some cases. In this section, I will briefly describe
salient properties of the evolution of number systems, writing systems, musical
notation, and dance notation, and offer some generalizations over the evolution
of these semasiographic systems.
4.1 Numbers
Number appears to be the earliest semasiographic system after images. Arti-
facts from ten to twenty thousand years ago, such as the Ishango Bone
(Marshack 1991: 22; Rudman 2007: 62) appear to involve tallies, the earliest
type of number notation.3 In tallies, the number of repeated symbols (such as
strokes or notches) denotes the equivalent number of entities. That is, tallies
represent an additive number representation system (Pettersson 1996: 796).
Additive number representations are essentially iconic: the number of signs
reflects the number of objects being counted.
Another example of an additive numeral system is represented by the undec-
orated tokens found in Middle Eastern sites dating back to 8000 BC and
apparently used to count goods (Englund 1998: 46). A later development
is the employment of distinct signs for larger numbers, such as 10 and 60,
which effectively function as bases in the numerical semasiographic system.
This development occurs in Mesopotamia at around the same time as proto-
cuneiform around 3200–3100 BC (see later; Nissen et al., 1993: 25). Petters-
son (1996: 796) calls these sign-value systems, the earliest of which are also
additive, repeating each sign to express multiples of the sign value. Familiar
examples are the later Roman numerals, such as XXXIII for 33. However, the
signs for each value are largely arbitrary: although signs for larger values are
sometimes larger in size, or require more strokes, than signs for smaller values,
the actual numerical value denoted by the sign is arbitrary.
One striking fact (to modern eyes) about the Mesopotamian sign-value nota-
tion is that the numerical signs varied depending on what is being counted or
measured, and the same sign may have a different numerical value depend-
ing on what is being counted or measured (Nissen et al., 1993: 25–27). For
3 Claims that the marks represent more complex number notation are highly speculative (Rudman
2007: 62–65).
112 William Croft
example, the sign resembling a dot represents 10 of the smaller units for dis-
crete objects, 6 units for dry measures of cereals, and 18 units for surface mea-
sures (ibid., 131). This is an instance of the high degree of context-dependence
that is commonly found in the earliest stages of semasiographic systems.
A still later development of number notation is a positional system (Pet-
tersson 1996: 796), in which the position of the sign determines its value by
specifying a base amount that it multiplies. In a sign-value system, one must
keep creating new symbols for larger quantities, but in a positional system,
the only symbols necessary are the values up to the base (e.g., 10 in the mod-
ern decimal system); the position of the symbol determines the magnitude of
the number. In Mesopotamia, a positional system evolved around 2000 BC
(Pettersson 1996: 798) – several thousand years after the appearance of undec-
orated tokens, and more than a thousand years after the emergence of a sign-
value notation system. The counting board or abacus, which dates back at least
to seventh century B.C. Greece (Menninger 1969: 301), and the quipu system
of knots, used in the Inca empire (fifteenth century AD; Menninger 1969: 252),
also use the positional method to represent numbers. (Interestingly, the con-
temporary ancient Greek written numerals did not make use of the positional
system found in its counting boards.)
The evolution of Mayan numeral notation also apparently proceeded through
the same stages: additive, then sign-value (with different glyphs for powers of
20), then positional (Rudman 2007: 118; he uses the term “additive” for sign-
value systems). The Ancient Greeks used a sign-value system adopting written
letters for signs (Pettersson 1996: 803), illustrating another characteristic of
semasiographic evolution that we will encounter in other systems, namely, the
adoption of signs from one semasiographic system for use in another semasio-
graphic system. Hindu-Arabic numerals appear not to have been adopted from
another semasiographic system. The earliest system (ca. 200 BC) was partly
additive (1–3 were one to three strokes), and partly sign-valued; a positional
system emerged around 600 AD (Pettersson 1996: 804). Again, the modern
Hindu-Arabic numeral system is the result of a gradual evolutionary process.
4.2 Writing
Writing essentially involves the adaptation of a semasiographic system in order
to represent spoken language. Writing evolved independently at least twice.
Undoubtedly independent are its evolution in Mesopotamia and Mesoamerica.
Egyptian writing evolved at the same time as Mesopotamian writing. It was
earlier thought to be inspired by Mesopotamian writing, but its structure is so
different, and the timing so uncertain, that it is now thought more likely to
have evolved mostly if not entirely independently (Baines 2004: 175–76). The
extant fragments of earliest Chinese writing – oracle bone inscriptions and clan
names on bronze vessels from ca. 1200 BC – are a full-fledged writing system in
structure (Bagley 2004: 198). There are no clear precursors in the archaeologi-
cal record (ibid., 190); so the early evolution of Chinese writing is unknown. (It
is possible, though now considered less likely, that Chinese writing developed
via diffusion from the Near East.)
By far the best documented evolution of writing is from Mesopotamia,
because writing was mostly done on clay. The emergence of the writing system
probably ultimately began with undecorated stone tokens referred to earlier.
By 3300–3200 BC, these undecorated tokens were enclosed in clay envelopes
impressed with seals, presumably used for accounting purposes (Englund
1998: 48). Seals were semasiographic representations of specific persons (or
institutions such as a temple), that is, they denoted names of individuals. Thus
the clay envelopes combined two semasiographic systems, the tokens counting
goods and the seal impression designating the person or institution responsible
for the goods. This is an example of a multimodal semasiographic system.
Some of the clay envelopes were also impressed with signs representing the
number of tokens in the envelope and their form. There also occur numeri-
cal tablets, tablets with numeral signs (and seal impressions) but not enclosing
tokens that they designate. These devices – sealed clay envelopes containing
tokens, envelopes containing tokens but impressed with the number and type
of tokens, and numerical tablets – are attested at around the same time and may
reflect either a sequence of developments or alternative systems for accounting
for goods. The envelopes bearing external signs illustrate another characteristic
in the emergence of semasiographic systems: the often direct association of the
signifier with the signified.
The next step occurred around 3200–3100 BC (the Late Uruk period and
Uruk IV writing phase), in both Uruk and Susa (the seat of the ancient Elamite
kingdom). In this period are found tags – small tablets – with ideographic
signs that may represent personal names (Nissen et al., 1993: 20), though some
ideographs may denote the goods (Englund 1998: 57). In addition, there are
numero-ideographic tablets, containing seal impressions, numeral signs, and
ideographic signs that designate different types of goods (Englund 1998: 51–
53). The numero-ideographic tablets mark the beginning of proto-cuneiform.
This step involved the combination of a largely preexisting set of numeri-
cal signs combined with a largely preexisting set of ideographic conventions
(Cooper 2004: 77).4 The number of ideographic signs quickly expanded to
4 Citing Buccellati (1981), Cooper (2004: 77) considers the ideographs to derive from glyptic art.
Schmandt-Besserat (1996) argues that the proto-cuneiform ideographs derive from decorated
tokens, which contrast with the undecorated tokens referred to earlier. However, the decorated
tokens are not found enclosed in the clay envelopes whose markings appear to have evolved into
the precursors of proto-cuneiform, namely the numerical tablets (Englund 1998: 47). Englund
suggests that the iconic nature of the proto-cuneiform pictographs and the decorated tokens are
instances of convergent evolution, rather than derivation of the former from the latter (Englund
1998: 47 [fn. 92], 53).
114 William Croft
some 900 signs (Englund 1998: 68; Cooper 2004: 68), while the context-
sensitive system of sign-valued numerals described previously constituted
around 60 signs. The proto-cuneiform signs are almost all pictographic, that
is, iconic, in nature (Englund 1998: 71) or at least iconic-indexical (i.e., pic-
tographic representation via metonymy and synedoche; Cooper 2004: 84),
although some are abstract representations that had been in use for some time.
It is doubtful whether the proto-cuneiform tablets represent “writing” in the
sense of reflecting in graphic form a spoken language. Both numerals and
ideograms do not denote words in a specific language in the way that a writ-
ing system that records phonetic content, even only partly so, does. The proto-
cuneiform tablets do not match spoken language syntax but “the ‘grammar’ of
the archaic accountants’ syntax” (Englund 1998: 63; 79). The information is
structured instead by the shape of the tablet, the ordering of cells in the tablet,
and other nonlinguistic properties (Cooper 2004: 80–81). In other words, writ-
ing originated as a semasiographic system for purposes other than to record
spoken language:
…no writing system was invented, or used early on, to mimic spoken language or to
perform spoken language’s functions. Livestock or ration accounts, land management
records, lexical texts, labels identifying funerary offerings, offering lists, divination
records, and commemorative stelae have no oral counterparts. Rather, they represent the
extension of language use into areas where spoken language cannot do the job (Cooper
2004: 83).
One did not find written royal inscriptions in Mesopotamia until around 2700
BC, in literature until around 2600 BC, and in letters until around 2400 BC
(Cooper 2004: 83). In other words, the semasiographic systems that evolved
into writing originated in specific domains of use, such as the accounting func-
tions of early Mesopotamian writing. The signs then came to be construed
as denoting word-like concepts (the semasiologographic systems of Trigger
2004: 47), and eventually words in a specific language. Thus, it is not surpris-
ing that proto-cuneiform did not have a language-driven syntax, and did not
express grammatical morphemes. Indeed, it has been suggested that perhaps
proto-cuneiform was not language-specific because ancient Uruk was a polity
inhabited by speakers of different languages, although the Sumerians were the
dominant group at first (Englund 1998: 73–81). It was not until around 2400
BC that the “spoken language determined the order of the script” (Nissen et al.,
1993: 123), and it took until the end of the third millennium BC for grammati-
cal morphemes to be written (ibid., 116): “Full writing was the byproduct of a
gradual discovery of new applications” (Bagley 2004: 233).
Over time, the signs evolved from a more iconic to a more arbitrary sym-
bolic form (i.e., no longer iconically representing the concept denoted by a
word, not its phonetic form). This was in part due to the Mesopotamian writing
technology, namely the use of the stylus on soft clay, which made it difficult to
draw curved lines.
The origins of writing in Egypt have some similarities to and some dif-
ferences from the origins of writing in Mesopotamia (Baines 2004; see also
Stauder 2010). Its earliest known uses do not seem to be entirely economic. The
earliest known ideographic objects are bone tags found in Tomb U-j in the Umm
al-Qa’ab cemetery near Abydos (no later than the Naqada III era, ca. 3200 BC;
Baines 2004: 153, 154). The tags bear numerals or “representationally based
signs,” but normally not both at once, unlike the numero-ideographic tablets
of contemporary Mesopotamia (ibid., 157). The tags were probably attached
to goods, that is, the signifier is associated with the signified. The tags (apart
from the numeral tags) are very likely to denote names, perhaps of prestigious
entities (ibid., 164). A small number of pictographic (iconic) signs are attested,
mostly different from later hieroglyphics. There are also jars, probably contain-
ing valuable oils or fats, with large painted signs similar to those on the tags
(but no numerals; ibid., 157, 159). There is nothing language-specific about
this semasiography, not unlike proto-cuneiform. The slightly later Hunters’
Palette (Naqada IIIa) and Scorpion Macehead (Dynasty 0) have similar signs
(ibid., 168–69). In Dynasty 0 (before the beginning of the third millennium
BC) the number of signs greatly expands to more than a thousand, again not
unlike proto-cuneiform, before being consolidated to a few hundred by the mid-
dle of the third millennium (ibid., 172). In Dynasty 0, one finds larger tags
that combine pictorial representation and linguistic (or at least semasiologo-
graphic) forms (ibid., 173) – another sort of multimodality. The function of the
writing system expanded to other administrative functions in the 1st Dynasty
(ca. 2950–2775 BC), but “continuous discourse and full syntax were not
notated for another couple of centuries” (ibid., 174), again like Mesopotamian
writing.
As noted earlier, little is known about the early evolution of the Chinese writ-
ing system. The earliest writings are already linguistic in the way that the earli-
est Mesopotamian, Egyptian, and Mesoamerican systems are not. For example,
the signs are fairly arbitrary in form, although more iconic than later versions of
the signs; literate Chinese speakers retain an awareness of their original iconic-
ity (Cooper 2004: 89).
Mesoamerica represents an undoubtedly independent origin of writing sys-
tems, but the earliest history of writing in Mesoamerica is very complex and
fragmentary. Part of the complexity is that the best-documented writing sys-
tem, that of the Maya, is in fact a later adaptation of earlier writing system(s)
used for other languages. All of the earliest signs have resisted decipherment,
or their decipherment is highly controversial. The earliest occurrence of glyphs
that look like later writing signs are found on Olmec carved heads dating from
1150–900 BC (Houston 2004: 288). The glyphs on the heads probably represent
116 William Croft
names of the individuals represented by the heads (ibid., 289). The direct asso-
ciation of signifier and signified was not decoupled until 900–600 BC, when
a stela is found with the glyph next to, not directly on, the headdress of the
head (ibid., 290). By ca. 500 BC there is a monument, La Venta Monument 13,
which contains writing, defined by Houston as representing language, detached
from its signified, and occurring in linear sequences (ibid., 292). Nevertheless,
the writing occurs on stelae denoting individuals who are probably named by
the writing, as in the case of San José Mogote Monument 3 (ibid.), where a
captive appears to be named by a calendrical date. This last monument illus-
trates a common Mesoamerican association of numerical, specifically calen-
drical, information with a logographic (or semasiologographic) system – not
unlike the origins of Mesopotamian writing in a combination of numerical and
semasiologographic signs (viz. the numero-ideographic tablets). The Isthmian
(or “Epi-Olmec”) texts from 36 BC–162 AD are also associated with calendri-
cal information (hence the precise dating; ibid., 296–97).
The Mayan scripts are adapted from the earlier Mesoamerican writing sys-
tems for the Mayan language(s). The scripts evolved over time, which makes
decipherment of the fragmentary earliest texts difficult even though much is
known about the more widely attested Classic Maya writing system. The ear-
liest Mayan texts are lists of deities and name tags (ibid., 304), not unlike the
earliest Mesopotamian and Egyptian systems. Again, grammatical morphemes
are not encoded until the beginning of writing phase 1B (150 AD; ibid., 305),
around 250 years after the earliest Mayan texts.
We can make some general observations about the evolutionary emer-
gence of writing from the origins of Mesopotamian, Egyptian, Chinese, and
Mesoamerican scripts. The emergence of writing is a gradual process, both in
the transition from pre-writing semasiographic systems to writing as a represen-
tation of spoken language, and in the further development of writing to express
grammatical structure as well as words (syntax and grammatical morphemes).
The signifier is often directly associated with the signified, or a representation
thereof (e.g., the image of a ruler and his name).
Pre-writing is often multimodal, involving some combination of numerals,
ideograms, seals, and pictorial representations. Some of this multimodality
may have been a trigger for the graphical representation of language, that is,
the combination of discrete linguistic units. In pre- or proto-writing phases,
the encoding of proper names appears to have been an important step. Proper
names are indexical in function (picking out an individual) but do not oper-
ate via deixis. The earliest ideographs and logographs were pictographic, that
is, iconic, or iconic-indexical (via metonymy and synecdoche). Even the devel-
opment of phonetic representation originates in iconic signs metonymically via
the rebus principle (a sign originally standing for a concept is extended to repre-
sent the associated phonetic form of the word denoting that concept). The form
of signs later becomes more arbitrary. Another aspect of pictographic repre-

sentation for language is that it adapts the signs of another semasiographic sys-
tem, namely that of images. In this context it is not surprising that later writing
systems emerged by adapting existing signs from writing systems that origi-
nally evolved for other languages, even using signs for different phonetic val-
ues than the originals, such as the Greeks adapting consonant signs to represent
vowels.
The domains of use for writing were at first very limited, often for centuries,
even though spoken language of course is a general-purpose communication
system for all sorts of joint actions. This is because writing originates to solve
specific information storage and communication problems, not to faithfully
reflect spoken language. Even when the domains of use of writing expanded
to include narrative, writing left much of the language out. Writing did not
express grammatical elements until centuries after its first emergence. In other
words, substantial common ground between author and reader was required to
interpret just the linguistic form encoded by early writing (this is still true of
abjads, a writing system that indicates only the consonants; one must know the
language to fill in the vowels).
4.3 Musical Notation

Musical notation appears to have been invented independently or partly inde-
pendently several times in the Near East and Europe. The earliest known
fragment of musical instruction is an Old Babylonian text that gives the incipit
of a hymn to Ishtar. The notation names the sixth string of the Babylonian lyre
or harp, certain intervals that involve the sixth string, and the mode in which
the hymn is to be played (Kilmer and Civil 1986: 96). This fragmentary text
appears to be a linguistic description of how to play the hymn rather than any
specific musical notation.
Later Babylonian texts, from 1250–1200 BC (West 1994) are less fragmen-
tary but remain difficult to interpret. The notation was designed for a nine-
stringed instrument, probably the bovine lyre (West 1994: 166), and for the
seven standard tunings of the instrument (ibid., 164). The texts bearing notation
are in Hurrian but apparently in a Babylonian musical tradition. The notation
consists of names of intervals followed by a numeral sign (ibid., 171). The inter-
pretation of the numeral is disputed, but may indicate the repetition of notes.
Nevertheless, we may make some general observations. The notation is adapted
from other semasiographic systems – writing and numbers. As such, it is also
multimodal. The music is also meant to be sung, and the lyrics are included.
This provides another degree of multimodality, because sung music is also mul-
timodal (words and melody). The notation is very domain-specific: it only can
represent music for the lyre, and only in one of the seven modes. Finally, the
118 William Croft
rhythm and other details of performance are not specified (ibid., 176); that is,
common ground about the songs is required to perform it properly.
Musical notation was reinvented by the Greeks by the third century BC, and
lasted no later than the fourth or fifth century AD (West 1992: 254). It is gener-
ally agreed that the musical notation developed gradually. Pitches are notated by
letters borrowed from the Greek alphabet for use in instrumental music. Some
of the pitch letters are Ionian, but other letters appear to derive from the Argive
script, a local Greek script of the sixth and fifth centuries BC (ibid., 261). The
notation appears to have originated for the aulos (a flute). Vocal notation based
on the Ionian alphabet was developed probably in the late fifth or fourth cen-
tury BC (ibid., 263). Rhythm came to be notated in some detail in the second
century AD, employing a separate set of symbols indicating length (ibid., 268).
The history of Greek musical notation illustrates the gradual evolution of the
notation over time. It is restricted at first to the aulos. Finally, it again presup-
poses familiarity with the music itself, and its performance style. The notation
was probably used only by professional musicians; and was rarely used at all,
as most lyric and dramatic texts lack musical notation (West 1994: 270).
Musical notation was again reinvented in Europe in the middle ages. (Appar-
ently Boethius [sixth century] was aware of Greek notation using letters for
pitches, but his writings did not lead to a similar notation in the middle ages;
Taruskin 2005: 17.) The earlier medieval musical notation is in the form of
staffless neumes from the nineth and tenth centuries. Staffless neumes are basi-
cally dots and lines that indicate the overall pitch contour, but without the staves
that allow one to identify musical intervals. Nor do staffless neumes indicate
rhythm. Instead, individual neumes appear to have specified particular melodic
formulas (by their shapes, which were given Latin names) that were associated
with particular chants and modes (Taruskin 2005: 22). Hence staffless neumes
presuppose much common ground: they do not indicate details of melody or
rhythm (let alone other aspects of performance, including polyphonic accompa-
niment). The user of the musical notation must already know the music, which
is still transmitted essentially orally; the notation at best serves as a reminder
or an aid (Taruskin 2005: 17). Musical notation was restricted at first to sacred
music and later extended to elite secular music, namely the courtly songs of the
troubadours, trouvères, and minnesänger.
Neumes were positioned on a staff starting around the early eleventh century
(Taruskin 2005: 16) – that is, two centuries after the first surviving documents of
staffless neumes. This allowed relative pitch (i.e., intervals) to be represented.
(Even so, manuscripts with staffless neumes were produced in the monastery of
St. Gall until the fifteenth century.) But rhythm would not be represented until
the beginning of the thirteenth century, another two centuries later (Taruskin
2005: 176). Rhythm was first notated for the polyphonic music of Nôtre-Dame
of Paris, by recruiting the neume shapes or combinations (known as ligatures)

to express rhythmic patterns instead of melodic formulas. The rhythmic inter-
pretation of ligatures was context-sensitive, depending on the overall pattern of
ligatures (ibid., 177).
By this point, polyphonic music was being notated, but there did not exist
explicitly notated means of aligning the voices such as the later bar lines, apart
from the iconic visual alignment of voices in the manuscripts. (Bar lines first
appeared in keyboard notation in the early sixteenth century; Apel 1953: 9.) In
the middle thirteenth century, so-called Franconian notation was developed for
notating the music of the motet style of the time (namely a style in which all
voices are equally melismatic, but a text or texts were applied to the melismatic
melodies). Again, the existing neume shapes and combinations were recruited,
this time for a less context-sensitive way to represent note length and hence
rhythm (Taruskin 2005: 212–17). While each specific ligature (combination of
note shapes and stems) has a unique rhythmic interpretation in Franconian nota-
tion, the individual note shapes did not, except in the context of the ligature. Ars
Nova notation (fourteenth century) is less context-dependent for its rhythmic
interpretation, and the notation developed by the beginning of the seventeenth
century is essentially the same as modern notation with respect to rhythm (apart
from odd anomalies such as the one described in §3).
Even with the development of more explicit notation of pitch intervals and
values, and rhythm, other aspects of musical performance were left to common
ground (leading to much debate over the “historically accurate” interpretation
of medieval and renaissance music by modern early music performers). From
the seventeenth to the twentieth century, notation of articulation, dynamics and
tempo were gradually added to musical scores. Much was still left to common
ground and an oral tradition, such as how to ornament melodic lines or fill out
a figured bass in Baroque music.
While a general notation system for European music evolved over the past
twelve centuries, specialized notations continued to emerge. These specialized
notations were mostly designed for instrumental music, and hence specific to
an instrument. The most widespread example was lute tabulature, since lute
was the most popular instrument in the Renaissance (Apel 1953: 54–86). Lute
tabulature used staff lines, but the staff lines indicate the strings of the lute, not
specific pitches. Numbers or letters were adapted to indicate which frets to use,
and notes above the staves were adapted from general musical notation to indi-
cate the rhythm. In other words, lute tabulature was multimodal, drawing from
general musical notation, letters, and numbers to indicate strings, frets (which
together specify pitch), and rhythm. A similar tabulature was used for German
organ music through its flowering in the seventeenth and early eighteenth cen-
turies (Apel 1953: 21–47).
120 William Croft
4.4 Dance Notation

Dance notation is not standardized; numerous systems have been invented in the
West. Like spoken language and music, dance requires strict temporal sequence
(and hence linear order). But dance is far more complex a phenomenon to rep-
resent. The notations used in the West vary tremendously; many if not most
choreographers invented their own notation. Nevertheless, the types of nota-
tion used present a historical sequence.
The earliest known dance notation was devised for the Renaissance basse
danse (low dance) (Guest 1989: 1). Letters were adapted to indicate the steps
to perform (letter notations continued for 200 years; ibid., 4). As with other
semasiographic systems, the earliest dance notation was contextually highly
restricted (only basse danse steps can be represented by the letters) and required
much common ground – they can only be interpreted by those who know the
steps to the dance.
The next types of dance notation in Europe (in terms of when it was first
devised) were iconic in form. In the seventeenth century, a track notation
was devised, which iconically represented the choreography as couples moved
around the floor (Guest 1989: 12–27). This notation developed in part because
the dances of this period involved such movements. The notation also imag-
istically represented bodily movements, though it also used arbitrary symbols
to indicate men vs. women (ibid.). Again, the notation was restricted to rep-
resenting Baroque dances. In the nineteenth century, another iconic notation
was devised, namely, stick-figure notation (Guest 1989: 28–68). Stick-figure
notation, unlike the earlier notation, was less restricted in the types of dance
movements it could represent. All of these types of dance notation were mul-
timodal in that the music to which one danced was also notated along with the
dance steps and movements.
In the late nineteenth and early twentieth centuries, a music-based dance
notation was invented. It adapted musical notation for dance steps: note values
indicate time – a dimension of dance not well notated by earlier systems. The
notes were placed on different musical staves that indicated different parts of
the body. Direction of movement was often indicated iconically, while modifi-
cations of the form of the musical note expressed bodily movements, sometimes
iconically and sometimes not.
In the 1920s, purely abstract dance notations developed (Guest 1989: 102–
62). Guest also notes that an earlier abstract system was devised by a choreogra-
pher in 1831 but was an isolated invention. Unlike the Baroque and nineteenth-
century notations, abstract dance notation was a completely symbolic, and in
some instances anti-iconic, representation of dance movements, although some
symbols were iconically motivated in part. On the other hand, the abstract
systems are also highly general in representing human bodily movement,
timing, and direction of dance – possibly a response to the freer dance styles
of twentieth-century dance and the greater variety of movements that they
encompassed.
The evolution of music and dance notation exhibits all of the properties of the
chronologically prior evolution of number and writing. It was gradual (extend-
ing over centuries in the case of Western musical notation); it is multimodal,
and adapts signs from other semasiological systems; its domain of use was
very restricted at first, before expanding; the interpretation of some signs was
context-dependent; and the notation also required much common ground to
interpret, becoming more explicit as it evolved. The chief difference is that
the signifier was not directly associated with the signified, except when using
musical scores in performance. This difference is presumably due to the fact
that these semasiographic systems represent human performances rather than
the conceptual structures represented by mathematics and language.
4.5 Some Commonalities Across the Evolution of

Semasiographic Systems
The evolution of different types of semasiographic systems reveals some
common features, which may offer clues for the evolutionary origin of the
language capacity. To begin with, the evolution of semasiographic systems,
where we have documentary evidence, is gradual and incremental. Contempo-
rary semasiographic systems for number, writing, music, and dance are very
different from their first notation, and the process leading to contemporary
semasiographic systems included many intermediate stages and alternative
notations. This observation reinforces the view expressed in §3, viz., that the
evolution of social cognition and the language that facilitates joint actions was
also gradual and incremental.
The earliest uses of semasiographic systems are very domain-specific and
context-dependent. They are used for only a restricted functional domain of
social (joint) actions. Numbers were used only for counting and for arith-
metical functions, before being elaborated to more and more mathematical
functions. Initial uses of writing were for specific communicative acts, and
only later came to be used in more and more domains. In fact, only with the
advent of electronic media is it increasingly used in the most casual conversa-
tional interactions. Initial uses of musical notation were restricted to specific
types of music, for specific social or religious functions, or for specific instru-
ments, as in the case of the creation of lute notation. Dance notations were
only for specific genres of dance; and even now no general dance notation has
come into widespread use, although the latest notations are highly general in
their representation of bodily movements. Thus we might expect that language
began in highly restricted functional domains, and its extension to become a
122 William Croft
general-purpose communication system was a long and gradual process in

human prehistory.
Another respect in which early semasiographic systems are context-
dependent is the minimal expression of what they signify. The earliest writing
did not encode grammatical inflections, or all of the sounds of a language. Early
European musical notation did not encode exact intervals or rhythms, let alone
dynamics. Early dance notation was even more minimal in the information that
it encoded. In other words, the interpretation of the semasiographic represen-
tation required much common ground; it served to minimally evoke what the
relevant information or action actually was. Likewise, language initially func-
tioned simply as a coordination device for joint action. Thus the “meaning” it
conveyed was probably minimal – just enough for the (restricted) purpose of
coordinating a particular type of joint action. Eventually, language came to be
used to share knowledge, that is, to increase common ground in and of itself.
At that point, language had to become more semantically specific (although as
we saw in §2, language can never be totally “precise”). But this was probably
a late stage in the evolution from pre-language to language.
In some semasiological systems, the signifier is often directly associated with
the signified. This was the case for Mesopotamian tokens and clay envelopes,
which were associated with the objects they enumerated or denoted; and for
Egyptian and Mesoamerican writing, which were often at first names asso-
ciated with representations of the persons or deities that they signified. The
semantic relationship is thus associative at first. However, the situation is some-
what different with performance notation. Musical notation at first was at most
a mnemonic for the melodies that it encoded. In later periods, music was
frequently performed from notation, but it is not clear if that was true at the
beginning. Dance notation is not directly used in performance. Music and dance
notations are more likely to be used for teaching rather than for final perfor-
mance; they serve as a record of how to perform the composition in question.
Still, the cases in which the signifier is directly associated with the signified
suggests that the earliest uses of pre-language or language was to denote ele-
ments of immediate experience, and only later was language (or pre-language)
extended to denote displaced or imagined experience, which must be inferred
from the semantics of the sign. (This last suggestion is a common speculation
about the evolution of language.)
Semasiographic systems are essentially coordination devices: graphic coor-
dination devices. With respect to that function, the coordination device for one
semasiographic system is often parasitic on another semasiographic system.
For example, writing symbols were appropriated for Babylonian and Greek
music notation, for Greek numerals, Renaissance dance, and arithmetic sym-
bols. Writing systems designed for one language were adapted for another
language (Olmec for Maya, Phoenician for Greek, Sumerian for Akkadian,
and then Hurrian, Hittite, Persian, and other ancient languages; Chinese for
Japanese and Korean; Greek for Slavic languages via Cyrillic, and the Roman
alphabet for many languages). And images, the earliest semasiographic system,
were adopted for medieval music, for Baroque and nineteenth-century dance,
and for the earliest writing and counting systems.
Thus we might expect that the earliest linguistic signals might be adopted
from some other communicative system. It has been argued that primate
vocalizations cannot be the source of modern human linguistic vocalizations,
because primate vocalizations and their “meanings” (alarm calls or whatever)
are innate. While modern human language definitely is not innate in the specific
linguistic structures and vocalizations employed – it is extremely diverse across
speech communities – it could still be the case that at some point in the evolu-
tion of pre-language, vocalizations or manual and facial gestures were adapted
from some other function, including perhaps even innate vocal or other ges-
tures. (This is the process known as exaptation, a phenomenon well attested in
contemporary language change; see Lass 1990; Croft 2000, Mufwene 2001.)
Semasiographic systems also appear to evolve to a (more) arbitrary symbol
system. Again, this reflects a common speculation about the evolution from
pre-language to language. Pre-language may originate with indexical or iconic
gestures or vocalizations, before being supplemented with more arbitrary sym-
bols. Of course, modern human language retains indexical and iconic expres-
sions, along with more arbitrary symbolic forms. Indexicals form a central
body of linguistic expressions in modern human languages: demonstratives
are among the first words to be learned, are very ancient (not clearly derived
from other words), are central to structuring discourse, and grammaticalize into
ubiquitous grammatical functions, including definiteness, reference, and sub-
ordination (Diessel 2006). Diagrammatic iconicity is exhibited by the great
majority of syntactic structures, and many signed language gestures are iconic.
In terms of the complexity of the signals themselves, the signs in the ear-
liest semasiographic systems are frequently multimodal. Early writing com-
bined symbols for numbers and for objects. Medieval and later music combined
words and notes (pitches); later music combined notes for pitch and duration
with other types of symbols for dynamics and articulation. All dance notations
combine music with symbols for bodily movements. Baroque, music-based,
and abstract dance notation uses different types of symbols for different types
of bodily movements. Hence it might be suggested that the earliest language
was multimodal. There is an ongoing speculative debate about the priority of
vocalization or manual gesture in the evolution from pre-language to language.
It is perhaps most likely that both were used at first, without the separation that
we currently conceive for the two (Arbib 2012; McNeill 2012). And of course
modern spoken language is accompanied by manual gestures, and some have
argued that intonation represents a vocal gesture that is at least partly separate
124 William Croft
from language in the sense of the latter being a segmentally based symbolic
system of communication (Bolinger 1985).
Finally, semasiographic systems exhibit virtually no duality of patterning.
The closest thing to duality of patterning is the use of determinatives or clas-
sifiers and phonological complements in logographic writing systems (e.g.,
Cooper 2004: 89; Nissen et al., 1993: 115). The absence of duality of pattern-
ing may be due to the restricted domains of expression of most semasiographic
systems, even late in their evolution, in contrast to the general-purpose commu-
nicative function of modern human language (see §3). But it might also sug-
gest that the earliest pre-language did not exhibit duality of patterning at first
either.
One important difference between semasiographic systems and modern
human languages in their spoken or signed medium is that language in use
must be processed in real time, while semasiographic systems, being lasting,
are not. Thus the selectional pressures for semasiographic systems and lan-
guage might be somewhat different, and the evolutionary characteristics may
also differ. However, the properties drawn out from comparison of the evolu-
tion of different semasiographic systems – gradualness, domain-specificity and
context dependence, minimal expression, association of signifier and signified,
function as coordination devices, exaptation of functions, multimodality and
lack of duality of patterning – are independent of the medium of communica-
tion. In fact, some of these properties (in particular context-dependence, mini-
mal expression, association of signifier and signified) are often treated as more
typical of the real-time medium of spoken or signed language than of writing.
Before proceeding to speculation on the stages of the evolution from pre-
language to language, it is worth taking a glance at the process of language
acquisition. Although ontogeny doesn’t always recapitulate phylogeny, there
are some parallels in early language acquisition with the observations made on
the evolution of semasiographic systems that are worth noting in supporting
speculations on language evolution.
Language acquisition is a gradual process. This has been repeatedly demon-
strated by recent research on both phonological and grammatical acquisition
(Vihman 1996; Lieven, Pine, and Baldwin 1997; Tomasello 2003). The earli-
est use of language is also restricted in functional domain, to a restricted set
of activities that the caregiver and child engage in. It is also minimal, initially
consisting of single words that evoke the relevant interaction; productive gram-
matical inflections do not emerge until later. The signifier and signified are
directly associated at first – the first utterances are about immediate experience.
Finally, the earliest communicative gestures are multimodal – both vocaliza-
tions and bodily gestures. Thus, language acquisition also shares many features
with semasiographic systems.
Of course, there are some important differences between child language
acquisition and the evolution of pre-language from language. The caregivers
Table 5.1 Evolutionary Complexity Based on the

Evolution of Semasiographic Systems
Less “complex” More “complex”
Context dependent: Context independent:

specialized function general purpose
minimal coding explicit coding
immediate experience displaced experience
(associative) (inferred)
extensional (indexical) intensional (iconic/symbolic)
multimodality duality of patterning
that children interact with are already users of modern human language. And
children, at least after the “nine month revolution” (see §3), have most of the
social cognitive capacity that underpins modern human language. But the par-
allels to the evolution of semasiographic systems observed here (admittedly,
also created by humans with modern social cognitive abilities) lend support to
the speculations offered in the next section.
5 Speculations on the Evolution of Language

The observations on semasiographic systems, and the parallels with child lan-
guage acquisition, suggest that some functions are likely to be earlier and less
complex in an evolutionary sense, while other functions are likely to be later
and more complex in the same sense (see Table 5.1).
I have used the term “complex” in scare quotes because complexity here is
based on evolutionary sequence. Also, it is not based purely on the structure of
the signifier but on how the signifier expresses the concepts being communi-
cated. Nevertheless, it seems that the phenomena in Table 5.1 do reflect lesser or
greater degrees of cognitive complexity. A context-dependent system is going
to be less complex because it involves a small number of more specific con-
cepts, whereas a context-independent system will be more complex because
it will involve abstracting categories across a wide range of concepts. Explicit
coding will also require more signal complexity (grammatical inflections, more
complex constructions, etc.). Displaced experience is more complex to commu-
nicate than immediate experience because it requires conceptualization of an
experience apart from the speech act situation. Intensional symbols are more
complex than extensional ones in that they require the formation of a category.
Duality of patterning requires two alternative parsings of the same signal struc-
ture (phonological and morphosyntactic), whereas even a multimodal signal
does not operate on two levels at once; both modalities are being used to con-
vey the intended concept.
126 William Croft
These asymmetries in communicative means can be used to suggest possible

stages in the evolution from pre-language to language, as already suggested in
§4.5. But at least as important, the logical structure of the suite of cognitive
abilities described in §2 also suggests the evolutionary priority of some social
cognitive abilities over others.
Joint action, coordination, and communication all presuppose the presence of
common ground. Hence we will take the evolutionary emergence of common
ground – the ability to recognize that others have mental states that can be
shared (Tomasello 2008) – as our starting point.
In §2, we saw that common ground comes in different types that have dif-
ferent bases. Communal common ground emerges from shared practice among
humans in a social group. If so, then personal common ground, based on com-
mon perception and action, seems prior: it is based on direct interpersonal
interaction. But the actional basis presupposes the communal common ground
of shared communicative conventions, in particular a shared language, in the
community. Hence the actional basis must follow, or be part of, the shared
practice that underpins communal common ground. The perceptual basis for
personal common ground can arise from communal common ground result-
ing from shared practice among members of the social group. That is, the way
that we categorize elements of our perceptual experience may be influenced
by shared practice. But there is a more fundamental basis for common ground,
namely a human being’s individual interaction with the world, including one’s
own self (body, bodily functions and needs, etc.). We can call this natural com-
mon ground. Natural common ground is made up of an individual’s interaction
with the world and one’s body. Its basis is a human’s perceptual recognition
of species identity with another human being: this maximally inclusive “com-
munity” allows “natural” knowledge to be shared knowledge, that is, common
ground.
Hence, an ordering of types of common ground based on logical priority
would be something like this:
r natural common ground: individual ways of interacting with the world and
one’s body, shared via recognition of other humans as conspecifics;
r perceptual personal common ground, shared via joint attention to and joint
salience of the interacting individuals’ environment and actions;
r communal common ground, shared via shared practices between the mem-
bers of the community for various social and cultural purposes;
r actional common ground after the emergence of language and other human
communicative conventions.
With respect to coordination devices, a clear logical priority exists, as described
in §2. Convention presupposes precedent: a convention cannot emerge until
a coordination device is repeated due to precedent and then becomes com-
mon ground in the community. And precedent presupposes joint salience: joint
Basis Common Ground Coordination Devices
Being in the World+ Natural

Joint Salience
Conspecific Recognition Common Ground
Shared Environmental Perceptual Shared

Interaction + Personal Precedent
Joint Attention Common Ground
Shared Communal
Convention
Practice Common Ground
Actional Language
Conversation
Personal (Linguistic
(Language Use)
Common Ground Convention)
Figure 5.1 The logical relationship among types of common ground, its basis,
and coordination devices.
salience allows the possibility of first-time (nonrandom) success in coordina-

tion, which can then be followed as a precedent. Joint salience and precedent
require both current and prior interactions between the individuals (i.e., per-
sonal common ground). Convention is shared in a community, that is, conven-
tions are part of a community’s communal common ground. And language is
a conventional system for communication, so it also presupposes communal
common ground.
The logical relationships between types of common ground and types of
coordination devices are given in Figure 5.1 (the dotted line indicates the rela-
tionship between joint attention and joint salience, and the double role it plays
in the achievement of joint actions).
These logical relationships suggest the types of coordination devices that one
might expect with different types of human interactions with the world, one’s
self, and with others, which suggest stages in the evolutionary emergence of
coordination devices including, eventually, language.
An evolutionary account of the emergence of these social cognitive abilities
requires some sort of selectional pressure for hominins to develop each suc-
cessive stage of these abilities. The obvious place to look is in the evolution
of joint actions. In §2, we presented an analysis of joint actions that requires
coordination and common ground, but also a degree of helpfulness, or commit-
ment to the other’s success in carrying out their part of the joint action (Bratman
1992: 336–38; Tomasello 2008, §5.2.1). Tomasello and colleagues have more
128 William Croft
recently developed an evolutionary account of the evolution of human moral-

ity as a basis for the emergence of joint actions (Tomasello 2011; Tomasello
et al., 2012; Tomasello and Vaish 2013). We will use this account to motivate
the emergence of common ground, coordination devices, and thence language.
Tomasello and colleagues argue for two stages in the evolution of human
morality beyond great ape “morality” (their interdependence hypothesis). They
argue that while great ape sociality involves some degree of sharing, reciprocity,
and revenge, they are not committed to helping the other in collaborative activi-
ties in the way that humans are (Tomasello and Vaish 2013: 236). The first stage
in the evolution of human morality is brought on by a selectional pressure for
genuinely collaborative action (rather than just reciprocal altruism; Tomasello
et al., 2012: 673), which they take to be an environmental need to engage in
collaborative foraging in order to survive (ibid., 674). This stage, which they
call second-personal morality (Tomasello and Vaish 2013: 240), involves help-
fulness and commitment – treating the other as an equal, not as a “social tool”
(ibid., 236) – but specifically with individuals known personally to the agent.
Reputation, critical to avoiding free riding, is based on direct experience of
social interactions. The effect is that collaborative joint actions emerge, but
only at a small scale.
The next stage is what Tomasello and colleagues call group-mindedness
(Tomasello et al., 2012: 681) or norm-based morality (Tomasello and Vaish
2013: 245). They argue that this stage emerged as the result of increasing size of
groups, and pressure from competing groups also increasing in size (Tomasello
et al., 2012: 681). The outcome is the ability to engage in collaborative joint
activity with individuals who are not directly known to the agent, and hence do
not have a reputation acquired by direct experience of interaction, but instead
can be trusted by virtue of group membership. At this stage, there come to exist
social norms and institutions that enable large-group collaboration.
Second-personal morality is probably the selectional pressure leading to the
emergence of perceptual personal common ground and shared precedent as a
coordination device (namely, the second level from the top in Figure 5.1). Per-
ceptual personal common ground comes from direct interactions between the
individuals, and shared precedent also requires direct experience. On the other
hand, norm-based or group morality is probably the selectional pressure lead-
ing to communal common ground and convention as a coordination device.
Collaboration is based on group membership; the common ground and coordi-
nation devices also have to be established groupwide, not by direct experience
of interpersonal interaction.
Figure 5.1 implies that logically there may have been two other discrete
stages in the emergence of coordination devices, including language. The last
stage is clearly an elaboration of group-mindedness, where the cultural trans-
mission of knowledge through language (and, eventually, writing) proceeds at
an abstract level and a large scale in a large society. The first stage may be an
initial stage of second-personal morality. It presupposes at least some level of
collaboration that would first occur only in small groups of individuals known
to each other – perhaps a kin group. Hence it may have been kin recognition
rather than conspecific recognition that initially allowed for the emergence of
(natural) common ground.
Finally, we may speculatively map more concrete proposals about pre-
language and language based on the evolution of semasiographic systems
and language development in Table 5.1 onto the logical sequence of common
ground and coordination devices in Figure 5.1.
The first stage involving joint action, as a number of other scholars of the
evolution of language have proposed, is most likely to employ only indexi-
cals. Indexicals rely solely on natural common ground and joint salience or
attention as the coordination device. Indexicals involve association and hence
are restricted to immediate experience; they are context-dependent and typi-
cally multimodal – even linguistic demonstratives in their indexical uses are
multimodal, accompanied by gestures (Diessel 2006: 469–71). Iconic gestures
might also serve as coordination devices at this stage, because they also require
only natural common ground, particularly when used for immediate experi-
ence. They are also likely to be multimodal. The sort of joint actions that are
coordinated by indexicals, icons, and natural common ground would have to
be very simple, immediate actions.
However, it is likely that individuals would engage in joint actions of this
sort only with close associates, perhaps only kin. Second-personal morality
might represent a more extensive group that engages in collaborative joint
actions: known individuals, but not necessarily kin. Precedent as a coordination
device would allow the emergence of more stable signals. One type of signal
might be proper names for individuals, because individual identity is impor-
tant for second-personal morality (proper names are also among the first things
to appear in the evolution of writing). These proper names would be used at
least as vocatives (i.e., in immediate experience) or possibly in displaced con-
texts. Proper names might begin with a gesture iconic of an associated property
of the individual but may evolve to arbitrary symbols. Naming would facilitate
the performance of polyadic joint actions. As such, naming might lead to vocal-
ization as a more prominent or even primary modality: the broadcast nature of
vocalization compared to visually perceived bodily gestures may be favored for
selection in a polyadic joint action situation (cf. Tomasello 2008: 231). Stable
symbols within the group might also lead to the emergence of semantic content
in symbols (in fact, iconic gestures may also be used with semantic content).
The joint actions are everyday activities that involve a number and variety of
objects as well as agents (so are more complex in that sense), requiring explicit
categorizing and labeling.
130 William Croft
The advent of larger groups gives rise to communal common ground and
conventional signals in that community. It seems likely that the evolutionary
success of such groups would require engaging in planned joint actions. These
joint actions are complex activities requiring subplans that need to be made
explicit (i.e., they are not simply part of communal common ground and can-
not be executed by everyone without explicit negotiation or instruction). The
content for communication is not just immediate experience but also a restricted
set of imagined and displaced experience, namely the intended or desired sub-
plans that must occur in the future in order to achieve the total planned joint
action, and the subplans that have just been executed. Successful coordination
of planned joint actions of this sort would require combinations of signals (dif-
ferent combinations may involve gesture + gesture, gesture + vocalization, or
vocalization + vocalization). Combinations are required to construct imagined
experiences, namely the intended subplans. Because these experiences are not
in the here and now and may not have existed before, they need to be evoked
by combining units that do denote objects and actions in the here and now,
but in combinations representing a novel situation. At this stage, combinations
are likely to be structured by order, not grammatical inflection; the sort of fine-
grained conceptual detail that calls for grammatical inflection (see Croft 2007a)
is not needed for coordination of everyday joint actions, even these more com-
plex actions.
The final stage is the emergence of sharing of knowledge. The joint actions
are complex, but now include the joint action of sharing propositional knowl-
edge, beliefs, desires, evaluations, and so on. In other words, they are not coor-
dinating some physical everyday activities but enlarging the stock of knowledge
(including cultural knowledge and attitudes) in the community. The content to
be communicated is now not only immediate experience and imagined expe-
rience geared to executing complex actions, but also recalled experience and
other types of imagined experience not leading to immediate action. At this
point, actional common ground is produced by the joint action. Of course, the
coordination devices include all three of joint attention, precedent, and con-
vention. The signaling system includes the signals from the earlier stages and
combinations of such signals. At this stage, the combinations of signals serve
to communicate displaced experience and the full range of imagined experi-
ence. The generalization of the function of the signaling system to include
sharing of knowledge of all kinds will lead to the emergence of grammatical
inflection and derivation, and the sort of grammatical constructions familiar
from modern human languages. This generalization of function will also lead
to duality of patterning in the structure of the signal, though it is possible that
duality of patterning emerged earlier (see §1 and Mufwene 2013). Finally, this
generalization of function leads to the primacy of vocalization in the signal-
ing system (among hearing persons). In other words, the final stage leads to
modern human language, though I would guess that this stage also undergoes
gradual and incremental evolution, and we might only be able to speak of mod-
ern human language at the end of the fifth stage described here.
This account of stages of language evolution is very speculative (perhaps
modeling would give some clues). Nevertheless, by grounding the account of
the evolution of language in its function to facilitate joint actions among indi-
viduals in a community, we can develop plausible scenarios for the stages it
must have gone through on the way to modern human language.
REFERENCES
Appel, Willi. 1953. The notation of polyphonic music, 900–1600 (5th ed.). Cambridge,
MA: The Medieval Academy of America.
Arbib, Michael. 2012. How the brain got language: the mirror system hypothesis.
Bagley, Robert W. 2004. Anyang writing and the origin of the Chinese writing system.
Houston 2004a, 190–249.
Baines, John. 2004. The earliest Egyptian writing: development, context, purpose. Hous-
ton 2004a, 150–89.
Bolinger, Dwight. 1985. The inherent iconism of intonation. Iconicity in syntax, ed. John
Haiman, 97–108. Amsterdam: John Benjamins.
Boone, Elizabeth Hill. 1994. Introduction: writing and recording knowledge. Writing
without words: alternative literacies in Mesoamerica and the Andes, eds. Elizabeth
Hill Boone and Walter D. Mignolo, 1–26. Durham, NC: Duke University Press.
Boone, Elizabeth Hill. 2004. Beyond writing. Houston 2004a, 313–48.
Bratman, Michael. 1992. Shared cooperative activity. The Philosophical Review
101:327–41.
Buccellati, Giorgio. 1981. The origin of writing and the beginning of history. The shape
of the past: studies in honor of Franklin D. Murphy, eds. Giorgio Buccellati and
Charles Speroni, 3–13. Los Angeles: Institute of Archaeology and Office of the
Chancellor.
Bybee, Joan L. 1985. Diagrammatic iconicity in stem-inflection relations. Iconicity in
syntax, ed. John Haiman, 11–48. Amsterdam: John Benjamins.
Chafe, Wallace. 1977a. Creativity in verbalization and its implications for the nature
of stored knowledge. Discourse production and comprehension, ed. Roy Freedle,
41–55. Norwood, NJ: Ablex.
Chafe, Wallace. 1977b. The recall and verbalization of past experience. Current issues
in linguistic theory, ed. Peter Cole, 215–46. Bloomington: Indiana University
Press.
Cheng, Chin-Chuan. 2000. Frequently used Chinese characters and language cognition.
Studies in the Linguistic Sciences 30:107–17.
Clark, Herbert H. 1996. Using language. Cambridge: Cambridge University Press.
Cooper, Jerrold S. 2004. Babylonian beginnings: the origin of the cuneiform writing
system in comparative perspective. Houston 2004a, 71–99.
Corballis, Michael C. 2011. The recursive mind: the origins of human language,
thought, and civilization. Princeton: Princeton University Press.
132 William Croft
Coupé, Christophe. 2012. Semiotic investigations into early forms of symbolism and
language. The evolution of language: proceedings of the 9th International Confer-
ence (Evolang 9), eds. Thomas C. Scott-Phillips, Monica Tamariz, Erica A. Cart-
mill and James R. Hurford, 80–87. Singapore: World Scientific.
Croft, William. 1995. Intonation units and grammatical structure. Linguistics 33:839–
882.
Croft, William. 2007a. The origins of grammar in the verbalization of experience. Cog-
nitive Linguistics 18:339–82.
Croft, William. 2007b. Intonation units and grammatical structure in Wardaman and
English. Australian Journal of Linguistics 27:1–39.
Croft, William. 2000. Explaining language change: an evolutionary approach. Harlow,
Essex: Longman.
d’Errico, Francesco, Christopher Henshilwood and Peter Nilssen. 2001. An engraved
bone fragment from c. 70,000-year-old Middle Stone Age levels and Blombos
Cave, South Africa: implications for the origin of symbolism and language. Antiq-
uity 75:309–18.
Diessel, Holger. 2006. Demonstratives, joint attention, and the emergence of grammar.
Cognitive Linguistics 17:463–89.
Englund, Robert. 1998. Texts from the late Uruk period. Mesopotamien: Späturuk-
Zeit und Frühdyntastische Zeit (Orbis Biblicus et Orientalis, 160/1), eds. Joseph
Bauer, Robert Englund and Manfred Krebernik, 15–233. Freiburg: Universitäts-
Verlag/Göttingen: Vandenhoeck & Ruprecht.
Everett, Daniel L. 2005. Cultural constraints on grammar and cognition in Pirahã:
another look at the design features of human language. Current Anthropology 76:
621–46.
Everett, Daniel L. 2009. Pirahã culture and grammar: a response to some criticisms.
Language 85:405–42.
Fauconnier, Gilles. 1985. Mental Spaces. Cambridge, MA: MIT Press.
Guest, Ann Hutchinson. 1989. Choreo-graphics: a comparison of dance notation sys-
tems from the fifteenth century to the present. New York: Gordon and Breach.
Haiman, John. 1980. The iconicity of grammar: isomorphism and motivation. Language
54:565–589.
Haiman, John. 1985. Natural syntax: iconicity and erosion. Cambridge: Cambridge Uni-
versity Press.
Hockett, Charles F. 1960. The origin of speech. Scientific American 203:88–96.
Houston, Stephen D. (ed.). 2004a. The first writing: script invention as history and pro-
cess. Cambridge: Cambridge University Press.
Houston, Stephen D. (ed.). 2004b. Writing in Early Mesoamerica. In The first writing:
script invention as history and process. Cambridge: Cambridge University Press,
274–309.
Kilmer, Anne Draffkorn and M. Civil. 1986. Old Babylonian musical instructions relat-
ing to hymnody. Journal of Cuneiform Studies 38:94–98.
Kirby, Simon. 2002. Learning, bottlenecks and the evolution of recursive syntax. In: Ted
Briscoe (ed.), Linguistic evolution through language acquisition, 173–203. Cam-
Langacker, Ronald W. 1997. Constituency, dependency, and conceptual grouping.
Cognitive Linguistics 8:1–32.
Lass, Roger. 1990. How to do things with junk: exaptation in language change. Journal
of Linguistics 26:79–102.
Lewis, David. 1969. Convention. Cambridge, MA: MIT Press.
Lieven, Elena V. M., Julian M. Pine & Gillian Baldwin. 1997. Lexically based
learning and early grammatical development. Journal of Child Language 24:
187–219.
Marshack, Alexander. The roots of civilization: the cognitive beginnings of man’s first
art, symbol and notation, revised and expanded edition. Mount Kisco, NY: Moyer
Bell Limited.
McNeill, David. 2012. How language began: gesture and speech in human evolution.
Cambridge: Cambridge University Press.
Menninger, Karl. 1969. Number words and number symbols: a cultural history of num-
bers, translated from the revised German edition by Paul Broneer. Cambridge, MA:
MIT Press.
Mufwene, Salikoko S. 2001. The ecology of language evolution. Cambridge: Cambridge
University Press.
Mufwene, Salikoko S. 2013. Language as technology: some questions that evolutionary
linguistics should address. In search of Universal Grammar: from Norse to Zoque,
ed. Terje Lohndal, 327–58. Amsterdam: John Benjamins.
Nissen, Hans, Peter Damerow and Robert Englund. 1993. Archaic bookkeeping: early
writing and techniques of economic administration in the ancient Near East.
Chicago: University of Chicago Press.
Pettersson, John Sören. 1996. Numerical notation. The world’s writing systems,
eds. Peter T. Daniels and William Bright, 795–806. Oxford: Oxford University
Press.
Rudman, Peter Strom. 2007. How mathematics happened: the first 50,000 years. New
York: Prometheus Books.
Schmandt-Besserat, Denise. 1996. How writing came about. Austin: University of Texas
Press.
Schoenemann, P. Thomas. 1999. Syntax as an emergent characteristic of the evolution
of semantic complexity. Minds and Machines 9:309–46.
Stauder, Andréas. 2010. The earliest Egyptian writing. Visible language (Oriental Insti-
tute Museum Publications, No. 32), ed. Christopher Wood, 137–47. Chicago: The
Oriental Institute of the University of Chicago.
Taruskin, Richard. 2005. Music from the earliest notations to the sixteenth century.
(Oxford History of Music, 1.) Oxford: Oxford University Press.
Tomasello, Michael. 1999. The cultural origins of human cognition. Cambridge, MA:
Harvard University Press.
Tomasello, Michael. 2003. Constructing a language: a usage-based theory of language
acquisition. Cambridge, MA: Harvard University Press.
Tomasello, Michael. 2008. Origins of human communication. Cambridge, MA: MIT
Press.
Tomasello, Michael. 2011. Human culture in evolutionary perspective. Advances in cul-
ture and psychology, eds. Michele J. Gelfand, Chi-yue Chiu, and Ying-yi Hong,
5–52. New York: Oxford University Press.
Tomasello, Michael, A. C. Kruger, and H. H. Ratner. 1993. Cultural learning. Behav-
ioral and Brain Sciences 16:495–552.
134 William Croft
Tomasello, Michael, Alicia P. Melis, Claudio Tennie, Emily Wyman, and Esther Her-
rmann. 2012. Two key steps in the evolution of cooperation: the interdependence
hypothesis. Current Anthropology 53:673–92.
Tomasello, Michael and Amrisha Vaish. 2013. Origins of human cooperation and moral-
ity. Annual Review of Anthropology 64:231–55.
Trigger, Bruce. 2004. Writing systems: a case study in cultural evolution. Houston
2004a, 39–68.
Vihman, Marilyn May. 1996. Phonological development: The origins of language in the
child. Oxford: Blackwell.
Wenger, Étienne. 1998. Communities of practice: learning, meaning and identity. Cam-
West, M. L. 1992. Ancient Greek music. Oxford: Clarendon Press.
West, M. L. 1994. The Babylonian musical notation and the Hurrian melodic texts.
Music & Letters 75:161–79.
6 To What Extent Are Phonological Inventories
Complex Systems?
Christophe Coupé, Egidio Marsico,

and François Pellegrino
1 Theoretical Background
1.1 Phonological Inventories as Complex Systems

Complex systems are often defined on the basis of the interactions that take
place between their constituents. This approach echoes the founding principles
of structuralism in linguistics more than a century ago. It was indeed Ferdi-
nand de Saussure, in his 1916 Cours de linguistique générale, who defined
any language as a system whose building blocks only exist in the structure
of their relationships, be they of equivalence or opposition. Structuralism as a
theoretical framework in linguistics initially developed more at a synchronic
level, before the dynamical phenomena leading to the emergence, preserva-
tion, or disruption of structures were addressed. Concepts such as retroaction
and self-organization appeared in the wake of cybernetics in the 1950s and
1960s (Ahsby, 1947; Wiener, 1961). Further developments led to notions such
as emergence, dynamical equilibrium, chaos, self-organized criticality, and so
on. These concepts soon spread to various scientific fields, including the life sci-
ences and the humanities; their ubiquity is one of the pillars of today’s generic
“theory of complex systems,” which to some extent succeeded in adding a
dynamical perspective to the time-frozen structures of structuralism.
In the field of linguistics, considering phonological inventories in the light of
the theory of complexity is not new, and they may be the first linguistic struc-
tures that benefited from its explanatory power. The notion of self-organization
appears in a 1983 paper on phonological universals (Lindblom et al., 1983): on
the basis of observed regularities in the inventories of the world’s languages,
the location of vowels in the vocalic triangle was accounted for by a self-
organized process. According to it, vowels found their optimal positions by
maximizing the perceptual distances between them. This maximal dispersion
was related to the idea of maximal perceptual contrast: words whose sounds are
maximally distant at the perceptual level are easier to discriminate. Lindblom
et al.’s approach was further refined along different lines, taking consonants
135
136 Christophe Coupé, Egidio Marsico, and François Pellegrino
into account (Lindblom & Maddieson, 1988), refining the notion of perceptual
distance, and shifting from maximal to sufficient perceptual contrast (Vallée,
1994; Schwartz et al., 1997).
At the turn of the twenty-first century, the theory of complexity further per-
vaded the study of phonological systems. Shifting from variation-less inven-
tories at a communal level to collections of speakers and individual invento-
ries was one of the successful paths explored. Borrowing from Steels’ (1996)
naming games for the emergence of lexical conventions, de Boer (2000) investi-
gated the formation of individual, yet shareable and “synchronized,” phonolog-
ical systems in a population of speakers. Multi-agent simulations were the key
tool for considering self-organization “in motion,” with the gradual shaping of
individual systems through myriads of interactions, in order to reach success-
ful communication, much as one would assume in real situations. First attempts
were enriched with more careful models of human perception and production,
of language acquisition, of the dynamicity of speech, and of the aggregation
of phonemes into higher units of speech (de Boer & Zuidema, 2005; Oudeyer,
2006; Au, 2008).
1.2 Specific Aspects of the Complexity of Phonological Inventories

Complexity theory intersects with fields as diverse as linguistics, epidemiology,
and physics. While properties that transcend disciplinary boundaries have been
found – such as the ubiquity of power laws – different systems have their own
specific properties. Phonological inventories are no exception to the rule, as
described next.
As highlighted in the previous section, each language is an aggregate of indi-
vidual idiolects. Each speaker indeed possesses their own phonological inven-
tory. Therefore, speaking of Mandarin, French, or Tzeltal language does not
reflect well the geographic distribution and the constantly evolving nature of
speakers’ idiolects. That a reductionist approach is needed to characterize a
higher-level object is common practice. However, two specific questions arise:
(1) Does working on a single phonological inventory for a language make
sense? And (2) does comparing different languages on this basis really help
us understand the phenomena at hand?
A second specific question regards the range of constraints that underlie the
organization of phonological inventories and other structures of speech like syl-
lables and words. These constraints refer to at least three domains: (i) speech
perception, (ii) speech production, and (iii) speech cognition – which refers to
properties of representations in cognitive areas like the mental lexicon. More
is known about the first two domains, which exist in partial tension to each
other. Indeed, phonological systems are trade-offs between “ease of articu-
lation,” which minimizes articulatory efforts and favors articulatorily similar
To What Extent Are Phonological Inventories Complex Systems? 137
segments, and “perceptual salience,” which promotes maximum or at least suf-

ficient perceptual distance between segments.
A last characteristic of phonological inventories when considered as com-
plex systems is the possibility to decompose phonemes into features. A vowel
like [i] may be defined by the three features high, front, and unrounded,
in relation to the required articulatory configuration to produce it: the tongue
close to the palate, in the front of the mouth, with lips unrounded. Descrip-
tive sets of features usually rely more on the articulatory side of speech,
with only limited references to perception (e.g., sonorant vs. non-sonorant
consonants).
The imbrication of features into segments may be seen as an advantage: some
phenomena at the level of phonemes may be explained in a simple way by prop-
erties of their features. However, it may also be seen as an additional difficulty:
the relationship between features and segments may be convoluted and add up
to a supplementary layer of complexity. On the one hand, principles like the
“maximal use of available features” (MUAF) (Ohala, 1980) and “feature econ-
omy” (Clements, 2003b) – the tendency for features to be used parsimoniously
to “create” phonemes in an inventory – help to explain observed regularities
such as series of segments differing only in one feature, be it voicing, nasality,
aspiration and so on. On the other hand, such series often present gaps, systems
of the same size include varying features, and the like. Although other complex
systems exhibit a layered structure and may be described at different scales1 ,
features and segments interact in specific ways, which may not be found in
other systems and therefore deserve investigation.
1.3 A Quantitative Approach to Phonological Inventories

as Complex Systems
A first step toward understanding complexity is to rely on qualitative defini-
tions. The various approaches that Edmonds (1999) illustrates include:
– complexity as a holistic feature: some systems are complex and escape tradi-
tional reductionist approaches, whereas others are not; or, from a gradualist
perspective, some systems are more complex than others;
– complexity as relative to the observer of a system is not necessarily inherent
to the system itself;
– complexity reflects the (meta)language adopted to describe the system.
Along these lines, various general concepts such as the size or the inter-
nal variety of a system have been analyzed as inadequate when applied to
1 However, many well-known complex systems (e.g., spin glasses, ant colonies, and even human
collective behaviors) do not require the consideration of causal interactions between several
scales of description, and can be thoroughly analyzed at a single scale.
complexity; they have been complemented with more sophisticated math-

ematical or computational concepts. Although “minimum description
size,” “Kolmogorov complexity,” “Bennett’s logical depth,” “algorithmic
information content,” or “effective complexity” (ibid; Gell-Mann & Lloyd,
2003) do better when it comes to assessing the complexity of a system,
going beyond general considerations and applying them to specific systems
appear to be difficult. A recurring problem is therefore to be able to shift from
qualitative approaches to quantitative ones. Graph or network theory might
today be an interesting framework to this end, given the wide range of fields
in which its tools may be used, from statistical physics to epidemiology to
sociology (Newman, 2010; for examples in linguistics, see Cancho & Solé,
2001; Dorogovtsev & Mendes, 2001; Ke et al., 2008).
The previous considerations have been attested in the field of phonology, and
more generally in linguistics (e.g., Dahl, 2004; Miestamo et al., 2008; Samp-
son, 2009). Efforts to choose how to correctly represent phonological systems
in ways that make it easier to discuss complexity span over decades, from early
generative phonology (Chomsky & Halle, 1968) to auto-segmental phonology
(Goldsmith, 1990) and optimality theory (Prince & Smolensky, 2004). In more
recent years, numerous studies have undertaken to revisit phonological systems
in the light of complexity theory (e.g., Pellegrino et al., 2009). Among the main
approaches, typological studies and interactions between phonology, morphol-
ogy, and syntax have played a significant role (Plank, 1998; Fenk-Oczlon &
Fenk, 2005; Shosted, 2006). Specifically, phonological inventories have been
studied in a typological perspective, taking advantage of datasets covering hun-
dreds of languages (Maddieson & Precoda, 1990; Dryer & Haspelmath, 2011).
Interestingly, these attempts have often augmented with a quantitative dimen-
sion the more qualitative appearance of phonological inventories, which display
an organization that escapes either full randomness or full regularity. Scales of
complexity for consonantal, vocalic, or tonal systems have been defined, con-
trasted, and explored geographically (Maddieson, 2006; 2009; 2011); indices
of the complexity or economy of phonological inventories have been defined
(Marsico et al., 2004), and graph theory has been applied to their structure
(Coupé et al., 2009).2 Different descriptive options of the inventories in terms
of features have also been explored with computational methods (Coupé et al.,
2011).
In this chapter, we try to complement the previous approaches by question-
ing the degree of complexity of phonological inventories. The motives behind
this attempt are as follows: without evading a cross-linguistic framework that
has proven fruitful in recent years, we would like to come back to a more pri-
mary definition of complex systems: complex systems are first of all made of
interacting elements, and their complexity arises from the nature of these
2 The statistical study presented hereafter partly draws on this contribution.

interactions. Linguists familiar with computational studies in the emergence

of language may refer here to Steels (1997)’s definition:
…a complex dynamical system [ …] consists of a set of interacting elements where the
behavior of the total is an indirect, non-hierarchical consequence of the behavior of the
different parts. (p. 2)
This definition is appealing because we wish to rely on a basic approach that,

without ignoring the specificities of phonological inventories in terms of fea-
tures and segments, could also be applied to other inventories of linguistic ele-
ments. More precisely, we wish to propose a deductive methodology that can
quantify the degree of relatedness of the building blocks of a collection of inven-
tories. This would hopefully open the door to new investigations of today’s
available digital linguistic datasets. In particular, we will see that this approach
helps assess simultaneously the strengths of individual constraints.
The previous notion of “indirect, non-hierarchical consequence” has also
been expressed as non-linearity: in a complex system, the behavior of the whole
cannot be linearly derived – that is, by way of superposition or addition – from
the behavior of the different parts. In section 2 of this chapter, we propose a
simple mathematical way to partially estimate the degree of non-linearity of
phonological inventories. To be clear, our way to answer the questions “Are
phonological inventories complex systems?” or “To what extent may phonolog-
ical inventories be considered as complex systems?” will therefore be: “Can the
compositional properties of phonological inventories be linearly derived from
the properties of the segments and features that compose them?”
1.4 Relying on Available Data

We wish our approach to be faithful to available data on languages. Typologists
have made clear that studying languages from a wide range of families leads to
patterns that may be quite different from extrapolations from languages exhibit-
ing a narrow range of structural diversity (Evans & Levinson, 2009). This state-
ment should serve as a caveat when it comes to applying tools and notions of
the theory of complexity. Many of them indeed come from statistical physics,
where a simplification of the properties of the real phenomena at hand is often
accepted as a first step of the analysis. This usually involves reducing the num-
ber of dimensions, considering open systems rather than systems with limits,
and adopting a “mean field” perspective. This approach is often necessary to
turn a mathematically intractable problem into a solvable one. However, one
should beware of not neglecting the reality of available data for the sake of
proposing a unifying framework.
Regarding phonological inventories, this translates into the issue of available
data, and how general principles of organization may be derived from an analy-
sis of the observable diversity based on sound statistical principles. We rely on
the UPSID database, compiled over the course of many years by Ian Maddieson
(Maddieson, 1984; Maddieson & Precoda, 1990), a version of which was inte-
grated in the World Atlas of Linguistic Structures (Dryer & Haspelmath, 2001).
In our specific version of the database, 451 inventories were available in a bal-
anced way across linguistic families and geographic areas. This is less than
one tenth of all languages spoken on Earth today; however, given the sampling
methods adopted, it is likely to be representative of the worldwide diversity.
The sustained efforts required to collect and compile the data are sometimes
met with suspicion regarding their correctness (see Clements, 2003a, for some
of the problems met). One also has to keep in mind that such synchronic collec-
tions are snapshots of diachronic flows that often carry ongoing sound changes
and shape the dynamical equilibriums observed at any moment in time. Never-
theless, large-scale typological studies likely compensate for errors at the level
of individual languages.
In section 2, we will present our mathematical approach to interactions
between segments and features in phonological inventories, and explain how
we take into account the limited amount of evidence offered by the available
datasets.
1.5 Objectives
Given the background introduced in the previous sections, we can summarize
our approach and its objectives as follows:
– we adopt a deductive data mining approach with the UPSID dataset;
– we focus on interactions between constituents of phonological inventories;
– we quantify the degree of non-linearity of these interactions to estimate the
degree of complexity of the inventories;
– we further analyze the interactions to confirm or infirm current hypotheses
on the structuration of the inventories; and
– we hope to discover new facts and formulate new hypotheses on their basis.
In section 2, we describe our methodology and the choice we made in terms
of statistical analysis. We then provide the results of our computations and
how we interpret them. Finally, conclusions and perspectives are given in
section 4.
2 Methodology
2.1 Defining Simple Phonological Inventories and a First Order

of Complexity
For the sake of clarity, we start by defining a case where the complexity of
phonological inventories could be considered as minimal, or at least very low.
A first step towards understanding how the world’s languages build their words
consists in paying attention to phonemes independently from each other, and
estimating their probability of occurrence. It is well known that while some
sounds like [i], [u], [p], or [k] are very frequent, others only appear in small per-
centages of languages – for example around 5 percent for [y]. From the UPSID
dataset, it is easy to compute the individual frequencies of the approximately
900 segments and 100 features that compose the inventories.
Let us assume that we can infer the composition of inventories on the sole
basis of this heterogeneous distribution of segments. The presence or absence
of any given phoneme in any inventory would be the reflection of its overall fre-
quency in the whole set of inventories; it would not depend on other segments
being present or absent in this inventory. In such a case, the compositional prop-
erties of the inventories would directly derive from the individual properties of
segments that compose them. If such a situation were true, we could refute the
idea that phonological inventories are complex systems. On the contrary, esti-
mating the extent to which the composition of an inventory departs from what
can be expected from the overall frequencies of its segments would provide a
measure of its complexity.
Our measure only partially encompasses the notion of complexity for inven-
tories. Indeed, it relies on pairwise interactions between segments or features,
and therefore decomposes potentially very complex relationships into much
simpler ones. Such a reductionist approach directly derives from models in
physics or sociophysics (e.g., Jensen, 2006), and echoes both the basic struc-
ture of graphs with nodes connected by vertices, and interactions in multi-agent
models of language emergence and evolution. We may say that it relates to a
first order of complexity; a “0-order” would relate to the absence of interac-
tions between the components of the inventories, an order higher than one to
complex interactions between more than two components. We may now turn
to the actual evaluation of pairwise interactions.
2.2 Evaluating the Interaction Between Two Components of Inventories

Considering a pair of two components, how can we estimate their degree of
interaction based on their frequencies of appearance in the available invento-
ries? We have adopted Clements’ (2003a; 2003b) assessment of feature econ-
omy in the UPSID inventory. Assuming that speech sounds tend to be composed
of features that are used elsewhere in the system, Clements tested two comple-
mentary hypotheses: (i) a given sound will occur more frequently in systems
in which all of its features are distinctively present in other sounds (mutual
attraction); and (ii) a given sound will have a lower than expected frequency
in systems in which one or more of its features are not distinctively present
in other sounds. To this end, he set up a number of contingency tables (see
Table 6.1 Observed frequencies of voiced labial fricatives and voiced coronal
fricatives across languages in UPSID. Expected frequencies are shown in
parentheses. Reproduced from Clements, 2003a
Voiced Coronal Fricatives

Present Absent Total
Present 110 (57) 37 (90) 147

Voiced labial Absent 65 (118) 239 (186) 304
fricatives Total 175 276 451
Table 6.1) to investigate feature economy for various categories of sounds and
features.
To evaluate the significance of the differences between observed and
expected distributions, Clements used the χ² test.3 For example, Table 6.1 led
him to the conclusion that the association between voiced labial fricatives
and voiced coronal fricatives in inventories was significantly positive (χ² =
119.203, p < 0.0001). This was an argument for the economy of the features
[+continuant] and [+voiced] shared by these consonants.4
We have retained from Clements’ studies both the statistical approach and the
following rationale: if two components appear in inventories together signifi-
cantly more or significantly less than what their individual frequencies would
predict, they may be said to interact, and the deviation from an absence of inter-
action reflects the strength of this interaction. In other words, our null hypothe-
sis (H0) is the following one: there is no interaction between two components,
that is, the frequency of their co-occurrence in inventories can be deduced from
their individual frequencies – more precisely the frequency of co-occurrence
is the product of the two individual frequencies. We investigated cases for which
the null hypothesis could be rejected, and further estimated the confidence one
could have in these rejections.
We tested interactions for all pairs of segments and all pairs of features. Our
findings are summarized in section 3. The test was conducted to evaluate the
first-order complexity of inventories on the basis of the number of significant
interactions. The χ² is actually only an approximate test of the statistical signif-
icance of a deviation from a theoretically expected distribution of observations
(with two categories like “present” or “absent”). Its low computational cost
3 With Yates’ correction for tables with cells containing values of seven or less.
4 Clements also tested for genetic or areal skewing by examining distributions in different groups
of languages.
Table 6.2 Observed Frequencies of /a/ and /ã/ Across Languages in UPSID
(Expected Frequencies in Parentheses)
/ã/
Present Absent Total
Present 82 (72) 310 (320) 392

/a/ Absent 1 (11) 58 (48) 59
Total 83 368 451
and its adequacy for moderately large values explain why it is often preferred
to the binomial test, although the latter is more exact. However, because higher
computational costs were not a serious issue for us, and because we were will-
ing to address cases where very frequent or very rare components were con-
sidered, we chose to rely on the binomial test anyway. We also considered not
only the associations of two components A and B both present in inventories,
but more generally the four possibilities A & B, !A & B, A & !B, !A & !B,
where “!” stands for the absence of a component. This allowed us to consider
additional relevant interactions, as exemplified in Table 6.2.
Out of the four associations /a/ & /ã/, !/a/ & /ã/, /a/ & !/ã/, !/a/ & !/ã/, only the
second one departs significantly from what can be expected given the individual
frequencies of /a/ and /ã/ (p < 0.001, when p > 0.05 for all other associations).
Especially, looking at the positive association of /a/ and /ã/ does not reveal any
significant trend, likely because of the high frequency of /a/. After other pairs
of oral vowels and their nasalized counterparts have been tested, the following
conclusion can be drawn – and only this one: inventories where a nasalized
vowel is present without its oral counterpart are significantly disfavored. This
may be explained by the mechanism of transphonologization through which
nasalized vowels derive from their oral counterparts by acquiring the nasal fea-
ture of an adjacent consonant. Without the oral vowel, the nasalized counterpart
cannot appear, and rare cases like !/a/&/ã/ only occur when the oral vowel sub-
sequently changes or disappears – usually by a change in quality. This example
illustrates why looking at absent components may be interesting. It also shows
how synchronic inventories may reveal relevant diachronic processes, although
in an implicit manner. All in all, we took a binary approach to inventories, as
they were not only described by the set of phonemes they contained, but also
by the set of all the others they did not.
An association of two components may occur significantly more or less
than expected (in terms of individual frequencies). Additionally, several asso-
ciations among the four possibilities may be shown to be statistically signifi-
cant. This makes sense when one thinks of the redundancy of the four possible
cases. Three combinations frequently found can be highlighted. Assuming that

“> 0” means a frequency greater than expected and “< 0” a frequency lower
than expected:
– As in the previous example, !A&B < 0 means that the presence of B implies
the presence of A. This association may appear or not with other associations;
– A&B > 0, !A&!B > 0, !A&B < 0, A&!B < 0 indicate that the two compo-
nents A and B mutually attract each other;
– A&B < 0, !A&!B < 0, !A&B > 0, A&!B > 0 indicate that the two compo-
nents A and B mutually repel each other.
A limitation was the difficulty to automatically take into account what Clements
(2003a) called covert attractors and subset effects. Covert attractors referred to
situations where a significant association between A and B is due to the covert
influence of a third component C, rather than to a direct relation between A
and B. Subset effects pointed to positive associations between A and B due to a
subset of A rather than to A as a whole. These two cases ultimately arise from
decomposing structural complexity into pairwise interactions only. Because we
analyzed all possible pairs of components, it was difficult to detect and test
such situations, where Clements was guided by its theoretical framework and
its restriction to a limited number of cases. We therefore could only keep in
mind that the number of significant associations was likely an overestimation
of the real interactions at play.
2.3 Evaluating the Interaction Between Multiple Pairs of Components

Our approach relied on the application of the previous statistical test to all pos-
sible pairs of components composing the UPSID inventories. We therefore had
to face the problem of multiple comparisons (Hochberg & Tamhane, 1987),
which is known to inflate the type I error rate (i.e., the detection of false posi-
tives). In our case, this meant some p-values for the binomial test smaller than
the significance threshold (the standard 0.05 value) falsely leading to the rejec-
tion of the null hypothesis of no interaction between the two components.
We considered different corrections of the standard test to address this issue,
often referred to as controlling the family-wise error rate (FWER). Among
the better known methods is the Bonferroni correction (Dunn, 1961). It states
that for m tests, rejecting p-values of tests smaller than α/m guarantees a
FWER lower than α. This correction is however very conservative. Although
more powerful and still controlling for the FWER, the Šidák correction and
Hochberg’s “step-up” procedure could not be considered, because they required
the tests to be independent of each other (Šidák, 1979; Hochberg, 1988). On
the other hand, Holm’s so called “step-down” procedure, also called the Holm-
Bonferroni method, does not require independence (Holm, 1979), and guaran-
tees control for the FWER. Finally, because of various theoretical issues, some
recommend to stick to the per comparison error rate (PCER) and ignore the
multiplicity problem (Saville, 1990).
Benjamini and Hochberg (1995) provided an analysis of the previous tests
and of their contexts of use which resonated with our own approach:
– The previous procedures control the FWER in a strong sense. “At levels con-
ventional in single-comparison problems,” e.g., 0.05, they “tend to have sub-
stantially less power than the per comparison procedure of the same levels”
(ibid, p. 290)
– “Often the control of the FWER is not quite needed. The control of the FWER
is important when a conclusion from the various individual inferences is
likely to be erroneous when at least one of them is” (ibid, p. 290).
In our own case, we did not intend to draw conclusions on the basis of at least
one significant association between two components of the inventories. We also
did not want to increase the number of false negatives too much for the sake of
avoiding false positives at all costs. Instead, we tried to optimize the statistical
power overall. We therefore eventually followed Benjamini and Hochberg’s
approach, with controls for a false detection rate (FDR) rather than the FWER.
Rather than taking into account the question whether any error was made, the
authors focused on the number of erroneous rejections. Controlling for the FDR
amounts to controlling the FWER in a weaker sense; indeed, what is contained
is the expected proportion of errors among the rejected hypotheses. Concretely
in our case, the procedure guaranteed that the number of false positives in the
discovered significant associations was lower than a chosen threshold, which
we took to be equal to 5 percent.
The algorithm applied to our study unfolded as follows: if m associations
of components were considered and m p-values p1 , p2 , …pm obtained with the
binomial test,
– The associations A1 , A2 , … Am were ordered in ascending order of their
p-values.
– Starting from the first, that is, smallest p-value, for a given threshold α (which
we considered equal to 0.05), K was computed as the largest k such that
pk ࣘ α x k / m.
– The null hypothesis was rejected for, and only for, the associations A1 ,
A2 , … AK , that is, these associations were accepted as expressing true inter-
actions between the components each was made of.
2.4 Dealing with Limited Amounts of Data

Relying on the limits of statistical tests like the χ² or the binomial test and
applying relevant corrections to multiple tests were the first ways to deal with
the limited amount of data offered by the UPSID database, and to draw conclu-
sions from them and only from them.
Table 6.3 Categorization of the Different Pairwise

Associations of Components of Phonological Inventories
According to our Statistical Approach
Meaningful associations Significant associations

Non-significant associations
Meaningless associations Non-significant associations
In cases where the overall frequency of a component in the inventories was

either very low or very high, checking associations with other components was
likely to result in non-significant results (except in the case of rare segments
most often occurring together in inventories). This explained why very frequent
segments like /a/, /i/, or /u/, overall did not strongly interact with other seg-
ments according to the binomial test. One may see this as frustrating, because
not much can then be said about the relations these very frequent segments
have with others. On the contrary, we see this as an opportunity to reflect the
available amount of information in a neutral way. This refers to the idea of the
aforementioned deductive approach.
To further understand the extent to which our limited information constrained
what could be found, we posited a distinction between two categories of asso-
ciations: meaningful and meaningless ones. On the one hand, all statistically
significant associations were meaningful. On the other hand, non-significant
associations were either meaningful or not, depending on whether the individ-
ual frequencies of the two components could have led to a significant associ-
ation, or whether their values were too extreme for this. In the first case, the
dataset contained enough information to potentially detect a significant associ-
ation, and it turned out not to be the case. In the second case, nothing could be
expected, therefore the term “meaningless.” The distinction allowed us to com-
pute percentages of significant associations on the basis of meaningful asso-
ciations, rather than on the basis of tested associations (see section 3). This
approach is summarized in Table 6.3.
Detecting meaningful non-significant associations once again implied multi-
ple comparisons. For each non-significant association, we computed the small-
est p-value that could be obtained given the individual frequencies of the com-
ponents. We then applied Benjamini and Hochberg’s procedure on all these
p-values to estimate which non-significant associations could be considered
meaningful while controlling for the FDR with a 5 percent error rate.
The approach we considered with pairs of components could potentially be
applied to n-uplets with n > 2. However, in practice, the size of the UPSID
database was too small to correctly estimate meaningful associations but for a
limited number of n-uplets. This would have seriously hampered our approach.
The explosion of the number of combinations would also have made the anal-
ysis and presentation of the results significantly more difficult.
3 Results and Interpretations
3.1 Overview
We considered two levels of analysis for UPSID: segments and features. Our
approach gave quantitative measures and an estimation of first-degree com-
plexity for each level. Each time, we considered vowels and consonants sep-
arately, but also gave figures for all segments or features considered together.
This included the addition of diphthongs, although we chose to exclude them
from our analyses for the sake of clarity and because they are overall less
studied and understood than monophthongs and consonants.
To be clear, a feature was present in a phonological inventory if at least one
segment containing it was part of that inventory. Two features could therefore
co-occur in an inventory whether or not they co-occurred in the same segment.
In each case, we computed, along with a few other figures, the number of tested
associations,5 the number of significant associations, and non-significant mean-
ingful and non-significant meaningless associations. The percentage of signif-
icant associations in meaningful associations revealed the degree of first-order
complexity of the inventories at the chosen level of description and for the set
of components – vowels, consonants, vocalic features, consonantal features, all
segments or all features – considered. Knowing the significant associations, we
could also compute the number of pairs that presented at least one significant
association among the four possible configurations A&B, !A&B, A&!B, and
!A&!B. The percentage of “significant pairs” – with respect to the total number
of pairs of components – could then be computed, and give another estimation
of the first-order complexity.
We also investigated the distribution of significant associations according to
their p-values, in order to go beyond the single figure of their numbers and pos-
sibly observe different distributional configurations. More precisely, pK being
the threshold for single comparisons as defined in section 2.3, we divided the
associations into groups defined by the limits [pK /10n+1 , pK /10n ] (n being an
integer).
Finally, we reported the most significant associations sorted by p-values. This
helped to shift from our somewhat “a-theoretical” approach to phonological
5 The number of tested associations is equal to four times the number of possible pairs of compo-
nents. This value corresponds to the four interactions A&B,A&!B, !A&B, !A&!B; n being the
number of components, the number of possible pairs is equal to C(n,2) = n.(n − 1)/2.
Table 6.4 Comparison of Four Different Approaches to Multiple Testing for

Vocalic Features
PCER B HB FDR
# of components 26
# of tested associations 1300
# of significant associations 35 3 3 5
# of non-significant associations 1,265 1,297 1,297 1,295
theory and to see if the first one could bring new arguments to the second. Asso-
ciations can be partitioned into two main categories, according to whether the
observed number of occurrences is smaller or greater than what was expected.
3.2 Comparing Corrections for Multiple Testing

In connection with the discussion in section 2.3, we compared the figures
obtained with different approaches to the problem of inflated type I error rate
in multiple testing:
– no correction, that is, control for the per comparison error rate (PCER)
– The Bonferroni correction (B)
– The Holm-Bonferroni procedure (HB)
– Benjamini and Hochberg’s control for the false discovery rate (FDR) rather
than for the family-wise error rate (FWER).
Tables 6.4 and 6.5 offer data for these four procedures in the case of vocalic
features and consonants.6 The choice of these two cases was guided by their
clear distinction in terms of number of components and number of tested asso-
ciations.
The results are in agreement with what could be expected in terms of con-
servatism: the order of the four procedures, ranked according to the number
of detected significant associations, is PCER > FDR > HB > = B. Note also
that even with the less conservative approach, PCER, the number of significant
associations is but a small fraction of the total number of tested associations.
In both cases, the Bonferroni and Holm-Bonferroni corrections provided iden-
tical results; the Benjamini-Hochberg procedure returned more associations,
especially in the case of consonants, but was still very far from the low degree
of conservatism of the control for PCER. In the following paragraphs, all data
given are based on Benjamini and Hochberg’s method for estimating significant
associations and meaningful as opposed to meaningless ones.
6 For the sake of clarity, we did not provide data here about meaningful associations, as we always
used Benjamini and Hochberg’s procedure for them.
Table 6.5 Comparison of Four Different Approaches to the Statistical Issue of

Multiple Testing for Consonants
PCER B HB FDR
# of components 582
# of tested associations 676,284
# of significant associations 10,754 249 249 731
# of non-significant associations 665,530 676,035 676,035 675,553
3.3 Complexity of the Interactions Between Segments

Tables 6.6, 6.7, 6.8, and Figure 6.1 provide the main results for the interac-
tions between vowels, between consonants, and between all segments in UPSID
phonological inventories.
It first appears that the number of significant associations is very small com-
pared to the number of meaningful associations – the maximum is 1.8 percent
for vowels – and therefore even smaller compared to the number of tested asso-
ciations. Being careful with what the UPSID database may reliably tell us about
associations of segments, our conclusion is as follows: it is not possible to rule
out that segments interact very little with each other, to the exception of a few
strong interactions that we will further describe later. The percentage of signif-
icant pairs is also very small, always less than 1.5 percent, which corroborates
the previous idea. Once again, it is important to remember here and in what
follows that there might be a significant number of false negatives. We can
note here that even with a more relaxed approach, which is controlling only
Table 6.6 Estimation of Pairwise Interactions at the Segmental Level
Vowels Consonants All segments
# of elements 162 582 833

# of pairs 13,041 169,071 346,528
# of tested associations 52,164 676,284 1,386,112
# of significant associations 292 731 851
# of meaningful pairs 200 554 617
# of non-significant associations 51,872 675,553 1,385,261
# of meaningful non-significant associations 15,621 180,200 370,611
% of meaningful associations (w.r.t. tested ones) 30.5% 26.8% 26.8%
% of significant associations (w.r.t. meaningful ones) 1.8% 0.4% 0.2%
% of significant pairs (w.r.t. total number of pairs) 1.5% 0.3% 0.2%
Table 6.7 The 20 Vowel-Vowel Associations with the Smallest p-values
Log 10
ID Seg. 1 Seg. 2 Assoc. Observed Expected (p-value)
1 ũ ı̃ A&B 69 13 −27,7
2 ã ı̃ A&B 72 15 −27,0
3 ã ũ A&B 67 13 −25,8
4 uː iː A&B 36 3 −25,3
5 oː eː A&B 32 2 −23,9
6 õ ẽ A&B 47 7 −23,4
7 ʊ ɪ A&B 55 10 −21,8
8 ũ ı̃ A & !B 5 60 −21,1
9 ã õ A&B 54 11 −20,0
10 õ ı̃ A&B 53 11 −19,5
11 ɔ ɛ A & !B 26 95 −19,2
12 ã ı̃ !A & B 10 66 −19,0
13 ã ũ !A & B 7 60 −18,9
14 ɔ̃ ɛ̃ A&B 27 2 −18,7
15 ã ı̃ A & !B 11 67 −18,6
16 o ɘ !A & B 25 90 −18,1
17 oː iː A&B 29 3 −18,0
18 u i !A & !B 49 10 −17,9
19 ɪ i A & !B 46 9 −17,4
20 o e !A & !B 116 50 −17,4
300
279
Distributions of Segmental Interactions
250
200
150
130
100 86
81
64
50 45 42 39
32 25
22 20 18
10 8 5 514 13 5 7 9 65 4
3 2 2 33 6 2 1 3 1 3 1 2 1 3 11 1 2 1 2 11 1 1 1
0
pK.10-1 pK.10-5 pK.10-9 pK.10-13 pK.10-17 pK.10-21 pK.10-25
Vowels Consonants
Figure 6.1 Distributions of segmental associations according to the value of

the binomial test.
Table 6.8 The 20 Consonant-Consonant Associations with the Smallest

p-values
Log 10
1 ɡ b A & !B 9 92 −31,7
2 kʰ pʰ A&B 94 23 −30,5
3 d b A & !B 6 76 −27,6
4 kʰ pʰ !A & B 7 77 −27,0
5 ŋɡ mb A&B 43 4 −26,7
6 nd mb A&B 42 4 −26,0
7 kʰ pʰ A & !B 9 79 −25,9
8 ŋɡ nd A&B 39 4 −24,4
9
gb
kp A&B 34 3 −23,9
10 tʰ pʰ A&B 75 18 −23,8
11 kʰ tʰ A&B 75 18 −23,3
12 k t A&B 48 7 −23,0
13 n n̪ !A & B 69 16 −22,9

14 t k A&B 45 6 −21,9
15 n̪ t̪ A&B 73 19 −21,3
16 ɡ b !A & B 43 126 −21,3
17 k p A&B 45 7 −21,2
18 ɡ b !A & !B 155 72 −21,1
19 ɗ ɓ A&B 35 4 −21,0
20 tʰ pʰ A & !B 8 64 −19,8
for PCER and not for FDR, the percentage of significant pairs and significant
associations would still be low, in the order of 5 percent.
The percentage of meaningful associations, whether significant or not, is
around a quarter or a third of tested associations, and this even after controlling
for FDR. This means that the percentage of significant associations and sig-
nificant pairs is established upon a relatively high number of meaningful asso-
ciations, which underscores the assumption that there is a significant amount
of information in the inventories regarding segment-segment associations. A
very low percentage of meaningful associations would have undermined the
credibility of the percentages given in the previous paragraph.
As could have been expected, Figure 6.1 reveals an exponential-like decay
of the number of significant associations as the value of the binomial test
becomes smaller. The long tails of consonant and vowel distributions indicate
a number of very strongly significant associations, which therefore deserve
further examination because they may tell us something important about the
structure of phonological inventories, and potentially what the strongest forces
are at play in their organizations. The exponential-like decay also once again
i u ı̃ ũ i: u:
ɪ ʊ
e o ẽ õ e: o:
ɛ ɔ ɛ̃ ɔ̃
a+ a ã a:
Figure 6.2 Graphical representation of the relations between vowels. Solid

lines correspond to relations of mutual attraction, dotted lines to relations of
repulsion.
reinforces the idea that the interactions between segments are very limited,
because among significant associations, many do not depart much from the
threshold of the statistical test.
For vowels, the most significant associations involve, 14 times out of 20,
a secondary feature: nasality (11) or length (3). The six others include height
and frontness. If one looks more broadly at all significant associations, these
tendencies are confirmed.7 All in all, the main results are the following:
– For most pairs of either nasalized vowels or long vowels, the two phonemes
tend (i) to occur together more than expected and (ii) to be both absent more
than expected; and (iii) situations where one is present and the other one is
absent are much less frequent than expected. These patterns clearly suggest
mutual attraction between nasal vowels and between long vowels.8 In other
words, a vowel with a secondary articulation is seldom isolated in a vocalic
system. This result is in line with the MUAF principle and the notion of
feature economy (see section 1.2).
– As far as height and frontness are concerned, our results confirm the well-
known dispersion theory (Liljencrants & Lindblom, 1972), according to
which front and back dimensions are balanced for a given height, and neigh-
boring height degrees tend to repeal each other.
Figure 6.2 summarizes the relations of mutual attraction or repulsion for pri-
mary vowels as well as nasalized and long vowels. Mutual attraction and
reinforcement are particularly visible for nasalized vowels, and the repulsion
between high and lowered-high primary vowels clearly illustrate the avoidance
of neighboring height degrees in a system.
Interestingly, both MUAF and dispersion theory are systemic princi-
ples recovered here via one-to-one segments interactions, meaning that our
approach can reach the “collective” level, although in a non-straightforward
7 Other secondary features appear, although patterns of mutual attraction or repulsion only clearly
appear for nasality and length.
8 One of the four possible cases is missing for some pairs, but there is no case where mutual
attraction seems contradicted.
way. It adds the benefice of being able to rank these principles according to
their prominence in inventories.
The results are similar for consonants, where most of the twenty top signifi-
cant associations (with the exception of #9, #13, and #15) involve series of stop
consonants. They show attraction between pairs of series of elements based on
some shared feature (voiced, aspirated, ejective or pre-nasalized), and confirm
that phonological systems have a highly productive use of features.
Associations #13 and #15 more directly involve place of articulation, par-
ticularly dental consonants. Association #13 suggests repulsion between the
alveolar nasal /n/ and its dental counterpart, which is confirmed by three fur-
ther associations ranked #36, #50, and #236; /n/ and its dental counterpart tend
to avoid each other. Association #15, on the contrary, suggests that we find
many more systems containing dental /n/ and dental /t/ than expected, and fur-
ther associations #26, #76, and #240 confirm that there is a mutual attraction
between them. Our analysis is that both associations relate to the fact that the
coronal place of articulation is somehow unspecified (Maddieson, 1984) in sys-
tems; the most frequent choice is alveolar, but some languages choose a dental
articulation, and when they do so, it happens across manner of articulation,
hence association #15. Very rarely, languages have both place of articulation
(especially with the same manner); in UPSID, only fourteen languages have
both /n/ and its dental counterpart (which is much less than expected, as high-
lighted by association #36), and among them nine are Australian languages,
which are well known for contrasting consonants on more places of articula-
tion than usual, especially in the coronal region. The other five languages are
scattered over five different families.
When we examine the associations for segments considered altogether, the
most significant ones only concern vowel-to-vowel or consonant-to-consonant
associations. As shown in Table 6.9, only twelve associations involve a vowel
and a consonant, which are ranked moreover between #331 and #815 out of the
851 total significant associations.
Strangely enough, except for the 11th one, they all involve a labial-velar stop
(kp or
gb ) and a lower-mid vowel (ɛ or ɔ), either oral or nasal. Any perceptual
or articulatory explanation for these relationships would probably seem too far-
fetched. In fact, a closer look at the data reveals a family effect. Table 6.10 gives
the distribution of the various segments in both the whole UPSID database and
the Niger-Congo family (which contains 55 languages).
The second column of Table 6.10 gives the numbers and percentages of
occurrence of the segments in the UPSID database. Looking at the third col-
umn, a strong bias in the distribution of labial-velars in favor of the Niger-
Congo family appears. As a matter of fact, around 90 percent of the languages
having one or the other is in the Niger-Congo family. This isolates this fam-
ily as the relevant subset of languages to investigate. As for the more frequent
Table 6.9 Vowel-Consonant Significant Associations in the Inventories
Log 10
1
kp ɔ̃ A&B 15 2 −7.7
2
kp ɛ̃ A&B 15 2 −6.8
3
gb ɔ̃ A&B 15 2 −6.7
4
gb ɛ̃ A&B 15 3 −6.2
5
gb ɔ A & !B 5 24 −6.0
6
gb ɛ A & !B 4 22 −6.0
7
gb ɔ A&B 34 14 −5.5
8
kp ɔ A & !B 5 22 −5.1
9
kp ɛ A & !B 4 20 −5.1
10
kp ɔ A&B 30 12 −4.8
11 m õ !A & B 14 3 −4.7
12
gb ɛ A&B 35 16 −4.7
higher-mid and lower-mid oral vowels, their distribution across linguistic fam-
ilies is broader; they however appear in a large majority of Niger-Congo lan-
guages. Their nasalized counterparts are rarer in UPSID, but still appear in 19
Niger-Congo languages. These figures, when compared to the total size of the
family (55 languages), lead to percentages of occurrence in the family all much
higher than those found in the whole UPSID database.
The Niger-Congo languages thus form a subset where nearly all labial-velar
stops ( kp or
gb ) co-occur with more numerous than average higher-mid or
lower-mid vowels, whether oral or nasal. It is this peculiar distribution in the
family that makes the associations significant for the whole sample.
Table 6.10 Distribution of the Segments Involved in the Significant

Associations Between a Vowel and a Consonant (N-C stands for
Niger-Congo, # for Number, % for Percentage)
# of N-C % of the N-C

# and % of languages languages languages
containing the containing the containing the
Segment segment (in UPSID) segment segment

kp 35 (8%) 32 58%

gb 39 (9%) 34 62%
ɛ 187 (41%) 44 80%
ɔ 163 (36%) 45 82%
ɛ̃ 35 (8%) 19 35%
ɔ̃ 32 (8%) 19 35%
Table 6.11 Estimation of Pairwise Interactions at the Feature Level
Vocalic Consonantal All

Features Features Features
# of elements 26 55 97
# of pairs 325 1,485 4,656
# of tested associations 1,300 5,940 18,624
# of significant associations 5 77 145
# of meaningful pairs 4 53 104
# of non-significant associations 1,295 5,863 18,479
# of meaningful non-significant associations 513 2,679 8,017
% of meaningful associations (w.r.t. tested ones) 39.9% 46.4% 43.8%
% of significant associations (w.r.t. meaningful ones) 1.2% 2.8% 1.8%
% of significant pairs (w.r.t. total number of pairs) 1.5% 3.6% 2.2%
This result indicates that our approach makes it possible to find genetic (and
also areal) traits in the organization of phonological inventories, although again
not straightforwardly. This suggests that a more systematic exploration of these
parameters should be done, to distinguish between broadly distributed signif-
icant co-occurrences and more localized ones. This will be investigated more
thoroughly in future work.
3.4 Complexity of the Interactions Between Features

This section reports our findings at the level of features. As in the previous
section, Tables 6.11, 6.12, and 6.13 and Figure 6.3 provide our main findings for
the interactions between vocalic features, consonant features, and all features
(including diphthong features) in UPSID phonological inventories.
Given the smaller number of features, the number of elements and associ-
ations tested were smaller than for segments. However, the different percent-
ages used to assess first-order complexity, despite being slightly higher, are
of the same magnitude, with values between 1.2 and 3.6 for the percentages
of significant associations and significant pairs. Once again, based on infor-
mation that can reliably be extracted from the UPSID database for associa-
tions of features, it cannot be concluded that features interact with each other,
except for a few pairs. It is tempting to claim that at least for vowels, the fea-
tures that compose the segments of the inventories occur predominantly on
the basis of their individual frequencies, owing to their intrinsic articulatory
or perceptual properties. This statement is limited by the potential existence
of a high number of false negatives. However, the percentages of meaningful
associations are higher than for segments, between 40 and 50 percent. These
values hence favor the reliability of the previous percentages.
Table 6.12 The 20 Associations of Vocalic Features with the Smallest

p-values. Only the first five are significant
Log 10
ID Feature 1 Feature 2 Assoc. Observed Expected (p–value)
1 lowered-high9 high A & !B 21 5 −7,1

2 long velarized A&B 8 1 −5,0
3 lowered-high high !A & !B 4 19 −4,9
4 low raised-low !A & B 8 1 −4,2
5 lip-compressed raised A&B 2 0 −4,1
6 nasalized velarized A&B 9 2 −3,6
7 nasalized velarized !A & B 0 6 −3,0
8 retracted advanced A&B 2 0 −3,0
9 central low !A & !B 4 0 −2,7
10 long velarized !A & B 1 7 −2,5
11 low raised-low !A & !B 1 7 −2,4
12 higher-mid nasalized !A & B 9 19 −2,3
13 central mid !A & B 0 4 −2,2
14 creaky-voiced voiced A & !B 1 0 −2,1
15 lowered-high retracted A&B 4 0 −2,0
16 breathy-voiced voiced A & !B 1 0 −2,0
17 lower-mid lowered-high !A & B 31 44 −1,8
18 retroflexed lip-compressed A&B 1 0 −1,8
19 raised-low10 long A&B 14 7 −1,8
20 rounded high !A & !B 2 0 −1,7
The number of significant associations for vocalic features was extremely

small: only 5 out of 1,300 were detected by the statistical approach. By com-
parison, there were 35 of them when controlling only for the PCER (305 and
649, respectively, for consonantal features and all features; controlling for FDR
gives 77 and 145). The distributions of significant associations according to
the value of the binomial test (Figure 6.3) once again suggest exponential-like
decay, at least for consonantal features (and also for all features – not shown).
At the segmental level, the principle of feature economy could be advocated
as a likely explanation for many of the observed interactions. This is no longer
the case with features, and significant associations may therefore tell us some-
thing different about phonological inventories.
For vowels, close to nothing could be said on the basis of the five signifi-
cant associations. For the sake of analyzing the data a bit further, we looked at
the 35 interactions returned by controlling for PCER. The following proposals
9 This feature is equivalent to near-close.

10 This feature is equivalent to near-open.
Table 6.13 The 20 Associations of Consonantal Features with the Smallest

p-values
Log 10
ID Feature 1 Feature 2 Assoc. Observed Expected (p-value)
1 labial-velar approximant A & !B 0 33 −15,3

2 labial-velar approximant !A & !B 45 11 −14,2
3 palatal approximant !A & !B 29 4 −14,0
4 non-sibilant-fricative sibilant-fricative !A & !B 32 6 −11,9
5 labiodental non-sibilant-fricative A & !B 0 25 −11,5
6 lateral-fricative affricate-lateral A&B 21 3 −10,9
7 postalveolar sibilant-affricate !A & !B 109 57 −10,7
8 ejective affricate-lateral A&B 24 4 −10,6
9 glottal non-sibilant-fricative !A & !B 45 14 −10,4
10 postalveolar sibilant-affricate A & !B 51 102 −9,6
11 postalveolar sibilant-affricate !A & B 52 103 −9,5
12 click affricate A&B 6 0 −9,5
13 ejective lateral-fricative A&B 32 8 −9,3
14 glottal non-sibilant-fricative A & !B 12 42 −7,9
15 ejective affricate-lateral !A & B 2 21 −7,2
16 trill-or-unspecified flap A&B 26 61 −7,2
17 nasalized affricate A&B 5 0 −6,7
18 nasalized click A&B 5 0 −6,7
19 glottalized affricate A&B 4 0 −6,5
20 glottalized click A&B 4 0 −6,5
30
27
Distributions of Feature Interactions
25
20
16
15 14
10
6
5 4 4 4
2 2
1 1 1
0
pK.10-1 pK.10-2 pK.10-3 pK.10-4 pK.10-5 pK.10-6 pK.10-7 pK.10-8 pK.10-9 pK.10-10 pK.10-11 pK.10-12 pK.10-13
Vocalic features Consonantal features
Figure 6.3 Distributions of feature associations according to the value of the

binomial test.
Table 6.14 Pairs of Consonantal Features Characterized by Mutual

Attraction and Represented in the 20 Most Significant Associations
Feature 1 Feature 2
postalveolar Place sibilant-affricate Mode

approximant Mode labial-velar Place
palatal Place approximant Mode
sibilant-fricative Mode non-sibilant-fricative Mode
non-sibilant-fricative Mode labiodental Place
lateral-fricative Mode lateral-affricate Mode
ejective Secondary lateral-affricate Mode
glottal Place non-sibilant-fricative Mode
click Mode affricate Mode
must therefore be treated with great caution: few clear patterns could be dis-
tinguished overall except for the mutual repulsion between the features high
and lowered-high (near-close), which echoed the idea of sufficient perceptual
contrast already mentioned at the segmental level. The presence of velarized
vowels seemed to imply those of nasalized vowels as well as those of long
vowels. Raised-low (near-open) vowels seemed to appear when low vowels
were absent.
Regarding consonantal features, many of the interactions turned out to be
mutual attraction. Analyzing all significant associations while controlling for
PCER, we found in particular that the 16 most significant associations presented
in Table 6.13 were all complemented by other associations supporting mutual
attraction as depicted in section 2.2. Looking only at significant associations
with control for FDR, mutual attraction was clearly suggested, but without all
cues being available: three or four associations were found among the 77 sig-
nificant associations for most of the 16 most significant associations.
On the other hand, only one relation of mutual repulsion was found among
the 77 “FDR-significant” associations, with all four associations confirming it:
it was for the pair trill-or-unspecified and flap.
Table 6.14 shows that different pairs of mutually attracted consonantal fea-
tures may be composed of features relating to: (i) two modes of articulation,
(ii) a mode and a place of articulation, or (iii) a secondary feature and a mode.
None of these pairs can be explained straightforwardly, but they give rise to
interesting hypotheses in many directions, whether at the articulatory, percep-
tual, historical, genetic, or areal level. All these questions are worth investigat-
ing but more work is required to firmly ground any proposal. Finally, we found
very few mutual attractions between pairs of features related to two places of
articulation, one exception being between palatal and labial-velar.
Our study of associations between all features revealed many significant

interactions between features appearing exclusively in diphthongs. Interest-
ingly too, leaving aside secondary features carried by either vowels or conso-
nants, no significant associations were found between consonantal and vocalic
features, or between diphthong features and either consonantal or vocalic fea-
tures. In line with what was found in the study of segments, this argues for
independence between the separate domains of vowels, consonants, and diph-
thongs.
4 Conclusions and Perspectives

We proposed a generic approach to estimate the complexity of phonological
inventories relying on the information provided by a dataset of 451 actual
inventories. This method derived from a simple yet central definition of com-
plex systems stating that they are usually composed of elements whose global
behavior cannot be linearly deduced from their individual properties. We con-
sidered individual frequencies of segments and features to be such individual
properties and developed a statistical model to test the significance of pairwise
interactions, trying as much as possible not to extrapolate from the information
provided by the dataset.
Analyses at both the segment and feature levels revealed that under a mildly
conservative approach (the control for FDR), only a very small number of sig-
nificant pairwise associations could be deduced from the available data, and
that it was therefore difficult to conclude in favor of strongly non-linear inter-
actions between either features or segments. On account of this, our tentative
conclusion is that the degree of complexity of phonological inventories is rather
low. However, one should consider this position with caution, given the likeli-
hood of a significant number of false negatives, which were not prevented by
the statistical test used.
Detailed analyses of the most significant associations supported classi-
cal hypotheses about the structuration of inventories like feature economy
(Clements, 2003b) or maximal use of available features (Ohala, 1980). It also
suggested a ranking of some organizational constraints and specific interac-
tions, including the previous principle and interactions between nasalized vow-
els or between long vowels. Some additional phenomena were revealed, like the
very limited number of interactions between consonants and vowels, whether
at the segment or feature level.
Other paths could be followed to better consider the knowledge that can be
extracted from available datasets like UPSID. Rather than focusing on pairwise
interactions, more powerful methods to learn the regularities of the invento-
ries could be used. Artificial neural networks (ANN) would be one of these
methods.
It would also be interesting to apply our method to collections of invento-

ries from other unrelated domains, and compare the percentages of significant
associations with those obtained in this chapter. Could we find situations with
much higher percentages, or would we always observe weak values, even for
systems that can without doubt be considered as complex? Such comparative
findings would provide a stronger basis for our conclusion regarding the limited
complexity of phonological inventories.
Evolutionary models have also been proposed on the basis of the synchronic
information provided by UPSID (Coupé et al., 2009). Although these models
are built on statistical tests like those presented in this chapter, they did not
include corrections of multiple testing. Fixing this shortcoming would represent
a possible next step.
Our main findings are limited in relation to established theories. However,
two comments can be made: first of all, the statistical tests failed to reveal
that phonological inventories were strongly non-linear in their organization.
This inflects the issues at play, prompting investigators to prove why they think
that phonological inventories are complex systems rather than simple ones. It
can be conceded though that looking only at pairwise interactions is a reduc-
tionist approach, and that more sophisticated structures could be considered.
Clements’ approach was to look for arguments in favor of feature economy.
This led him to consider specific classes of segments to apply his statistical
test. Such an approach can be termed “theory-driven,” when our own initially
follows a theory-less, bottom-up approach. This does not mean that theoretical
phonological questions may not be addressed and answered, but that they do
not play a role in the design of the method. Both approaches are of course inter-
esting in their own sake, and are partially derived from the Kantian distinction
between nomothetic and idiographic tendencies. What is gained in genericity
is lost in terms of specific validations, and vice versa.
Secondly, we want to stress that the main interest of our approach lies in its
capacity to highlight the gap between the concepts and tools commonly used
in complexity theory, especially when applied to physical or biological sys-
tems, and in the theories proposed by scholars in linguistics. It seems that, until
further improvement, these concepts and tools still often lack the power and res-
olution to address specific questions of the field, and may therefore be consid-
ered as irrelevant by some. Furthermore, defining the “complexity of language”
has turned out to be a difficult enterprise, especially when one assesses the
multiple conceptions and measures of complexity. Definitions change depend-
ing on whether the investigator deals with the acquisition of specific struc-
tures, the amount of covert versus overt information in communication, the
respective complexity of syntax versus morphology versus phonology, and so
on. This difficulty may once again be explained in part by the divide between
nomothetic and idiographic approaches. The former apply better to natural sci-
ences, where most of the tools of complexity theory have been developed, and
the latter to the humanities.
Echoes of this argument can be heard in the heated debates that have fol-
lowed recent publications that relied on nomothetic approaches to investi-
gate the distribution of language structures in large samples of languages,
for example, Atkinson’s approach to the worldwide distribution of the size of
phonological inventories (Atkinson, 2011a, 2011b; Bybee, 2011), or Lupyan &
Dale’s (2010) study of the links between morphological complexity and social
structures.
REFERENCES
Ashby, W. Ross. 1947. Principles of the Self-Organizing Dynamic System. Journal of
General Psychology 37: 125–128.
Atkinson, Quentin D. 2011a. Phonemic Diversity Supports Serial Founder Effect Model
of Language Expansion from Africa. Science 332: 346–349.
Atkinson, Quentin D. 2011b. Linking Spatial Patterns of Language Variation to Ancient
Demography and Population Migrations. Linguistic Typology 15(2): 321–332.
Au, Ching-Pong. 2008. Acquisition and Evolution of Phonological Systems. Taipei:
Institute of Linguistics, Academia Sinica.
Benjamini, Yoav, and Yosef Hochberg. 1995. Controlling the False Discovery Rate: A
Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statis-
tical Society Series B (Methodological) 57(1): 289–300.
Bybee, Joan. 2011. How Plausible Is the Hypothesis that Population Size and Dispersal
are Related to Phoneme Inventory Size? Introducing and Commenting on a Debate.
Linguistic Typology 15(2): 147–153.
Cancho, Ramon Ferrer I., and Ricard V. Solé. 2001. The Small-World of Human Lan-
guage. Santa Fe Institute Working Paper 268: 2261–2266.
Chomsky, Noam, and Morris Halle. 1968. The Sound Pattern of English. New York:
Harper & Row.
Clements, G. Nick. 2003a. Feature Economy in Sound Systems. Phonology 20: 287–
333.
Clements, G. Nick. 2003b. Feature Economy as a Phonological Universal. Proceedings
of the 15th International Congress of Phonetic Sciences, Barcelona, Spain, 371–
374.
Coupé, Christophe, Egidio Marsico, and François Pellegrino. 2009. Structural Complex-
ity of Phonological Systems. Approaches to Phonological Complexity, Phonology
& Phonetics Series vol. 16, ed. by François Pellegrino et al., 141–169. Berlin, New
York: Mouton de Gruyter.
Coupé, Christophe, Egidio Marsico, and Gérard Philippson. 2011. How Economical
are Phonological Inventories? Proceedings of the 17th International Congress of
Phonetic Sciences, Hong Kong, China.
Dahl, Osten. 2004. The Growth and Maintenance of Linguistic Complexity. Amsterdam:
Benjamins.
de Boer, Bart. 2000. Self-Organization in Vowel Systems. Journal of Phonetics 28(4):

441–465.
de Boer, Bart, and Willem H. Zuidema. 2005. Multi-Agent Simulations of the Evo-
lution of Combinatorial Phonology. BNAIC 2005 – Proceedings of the 17th
Belgium-Netherlands Conference on Artificial Intelligence, Brussels, Belgium,
66–73.
De Saussure, Ferdinand. 1916. Course in General Linguistics. Edited by Charles Bally
and Albert Sechehaye in collaboration with Albert Riedlinger, translated by Wade
Baskin. New York: McGraw-Hill Book Company, 1966.
Dorogovtsev, Sergey N., and José F. F. Mendes. 2001. Language as an Evolving Word
Web. Proceedings of The Royal Society of London Series B, Biological Sciences
268: 2603–2606.
Dryer, Matthew S., and Martin Haspelmath (eds.). 2011. The World Atlas of Language
Structures Online. Munich: Max Planck Digital Library. Available at http://wals.
info/.
Dunn, Olive J. 1961. Multiple Comparisons among Means. Journal of the American
Statistical Association 56: 52–64.
Edmonds, Bruce. 1999. What is Complexity? The Philosophy of Complexity per se
with Application to Some Examples in Evolution. The Evolution of Complexity,
Dordrecht: Kluwer.
Evans, Nicholas, and Stephen C. Levinson. 2009. The Myth of Language Universals:
Language Diversity and its Importance for Cognitive Science. Behavioral and
Brain Sciences 32: 429–48.
Fenk-Oczlon, Gertraud, and August Fenk. 2005. Cross-linguistic Correlations Between
Size of Syllables, Number of Cases, and Adposition Order. Sprache und Natür-
lichkeit: Gedenkband für Willi Mayerthaler, ed. by Gertraud Fenk-Oczlon and
Christian Winkler, 75–86. Tübingen: Narr.
Gell-Mann, Murray, and Seth Lloyd. 2003. Effective Complexity. Nonextensive
Entropy – Interdisciplinary Applications, ed. by Murray Gell-Mann, & Constantino
Tsallis, 387–398. Oxford: Oxford University Press.
Goldsmith, John. 1990. Autosegmental and Metrical Phonology. Oxford: Blackwell.
Hochberg, Yosef. 1988. A Sharper Bonferroni Procedure for Multiple Tests of Signifi-
cance. Biometrika 75(4): 800–802.
Hochberg, Yosef, and Ajit C. Tamhane. 1987. Multiple Comparison Procedures. New
York: Wiley.
Holm, Sture. 1979. A Simple Sequentially Rejective Multiple Test Procedure. Scandi-
navian Journal of Statistics 6(2): 65–70.
Jensen, Pablo. 2006. Network-Based Predictions of Retail Store Commercial Categories
and Optimal Locations. Phys. Rev. E 74: 035101.
Ke, Jinyun, Tao Gong, and William Shi-Yuan Wang. 2008. Language Change and Social
Networks. Communications in Computational Physics 3(4): 935–949.
Liljencrants, Johan, and Bjorn Lindblom. 1972. Numerical Simulation of Vowel Quality
Systems: The Role of Perceptual Contrast. Language 48: 839–862.
Lindblom, Bjorn, Peter MacNeilage, and Michael Studdert-Kennedy. 1983. Self-
Organizing Processes and the Explanation of Phonological Universals. Explana-
tions for Language Universals, ed. by Brian Butterworth, Bernard, Comrie, and
Östen Dahl, 181–203. Berlin: Mouton.
Lindblom, Bjorn, and Ian Maddieson. 1988. Phonetic Universals in Consonant Systems.
Language, Speech and Mind, ed. by Charles Li and Larry Hyman, 62–78. London:
Routledge.
Lupyan, Gary, and Rick Dale. 2010. Language Structure is Partly Determined by Social
Structure. PLoS ONE 5(1): 1–10.
Maddieson, Ian. 1984. Patterns of Sounds. Cambridge: Cambridge University Press.
Maddieson, Ian. 2006. Correlating Phonological Complexity: Data and Validation. Lin-
guistic Typology 10(1): 106–23.
Maddieson, Ian. 2009. Calculating Phonological Complexity. Approaches to Phonolog-
ical Complexity, Phonology and Phonetics Series, vol. 16, ed. by François Pelle-
grino et al., 85–110. Berlin, New York: Mouton de Gruyter.
Maddieson, Ian. 2011. Phonological Complexity and Linguistic Patterning. Proceed-
ings of the 17th International Congress of Phonetic Sciences, Hong Kong,
China.
Maddieson, Ian, and Kristin Precoda. 1990. Updating UPSID. UCLA Working Papers
in Phonetics 74: 104–111.
Marsico, Egidio, Ian Maddieson, Christophe Coupé, and François Pellegrino. 2004.
Investigating the ‘Hidden’ Structure of Phonological Systems. Berkeley Linguis-
tics Society 30: 256–67.
Miestamo, Matti, Kaius Sinnemäki, and Fred Karlsson (eds.). 2008. Language Com-
plexity: Typology, Contact, Change. Amsterdam: Benjamins.
Newman, Mark. 2010. Networks: An Introduction. Oxford: Oxford University Press.
Ohala, John J. 1980. Chairman’s Introduction to Symposium on Phonetic Universals in
Phonological Systems and Their Explanation. Proceedings of the 9th International
Congress of Phonetic Sciences, 1979, 184–185.
Oudeyer, Pierre-Yves. 2006. Self-Organization in the Evolution of Speech. Oxford:
Pellegrino, François, Egidio Marsico, Ioana Chitoran, and Christophe Coupé (eds.).
2009. Approaches to Phonological Complexity. Phonology & Phonetics Series
vol. 16. Berlin, New York: Mouton de Gruyter.
Plank, Franz. 1998. The Co-Variation of Phonology with Morphology and Syntax: A
Hopeful History. Linguistic Typology 2(2): 195–230.
Prince, Alan, and Paul Smolensky. 2004. Optimality Theory: Constraint Interaction in
Generative Grammar. Oxford: Blackwell.
Sampson, Geoffrey. 2009. Language Complexity as an Evolving Variable. Oxford:
Saville, Dave J. 1990. Multiple Comparison Procedures: The Practical Solution. Amer-
ican Statistician 44: 174–180.
Schwartz, Jean-Luc, Louis-Jean Boë, Nathalie Vallée, and Christian Abry. 1997. The
Dispersion-Focalization Theory of Vowel Systems. Journal of Phonetics 25(3):
255–286.
Shosted, Ryan K. 2006. Correlating Complexity: A Typological Approach. Linguistic
Typology 10(1): 1–40.
Šidák, Zbyněk. 1967. Rectangular Confidence Regions for the Means of Multivariate
Normal Distributions. Journal of the American Statistical Association 62(31): 626–
633.
Steels, Luc. 1996. Self-Organising Vocabularies. Cambridge: MIT Press.
Steels, Luc. 1997. The Synthetic Modeling of Language Origins. Evolution of Commu-
nication 1(1): 1–34.
Vallée, Nathalie. 1994. Systèmes Vocaliques: De la Typologie aux Prédictions. Thèse de
doctorat. Grenoble: Université Stendhal.
Wiener, Norbert. 1961. Cybernetics: or Control and Communication in the Animal and
the Machine, 2nd edition. Cambridge: MIT Press.
7 A Complexity View of Ontogeny as a Window
on Phylogeny
Barbara L. Davis
1 Introduction
One of the goals of this book is to consider the strength of a complexity perspec-
tive on the long time scale origins of language in human speakers and listeners.
In this chapter, a complexity view of the ontogeny of speech acquisition in
human infants will be addressed as it potentially illuminates the phylogenetic
process of language evolution. In the broader context of language origins, the
ethnographer Nikko Tinbergen (1963) has suggested four questions that should
be addressed to characterize the evolution of the human language capacity.
These questions include: (1) What are the stimuli that produce the response in
the organism? (2) How does the behavior contribute to the organism’s survival
and reproductive success? (3) How does the behavior develop in the organism’s
lifetime? and last, (4) How did the behavior arise in the organism’s species?
(Tinbergen, 1963). Cogent to the topic addressed here, the last question con-
cerns ontogeny in organisms, in our case, human infants who are learning lan-
guage. In Tinbergen’s perspective, the ethnographer needs to ask how the young
organism’s behavior changes with age, what early experiences are necessary for
the behavior to be manifest, and which developmental steps and environmental
factors play a role in ontogenesis.
In the spirit of Tinbergen’s ontogenetic question, the phonological com-
ponent of language will be evaluated as it relates to examining the value of
ontogeny for considerations of phylogenetic origins. Phonology uniquely con-
cerns two dimensions of capacity that must be acquired in human infants rel-
ative to mastery of language structure and function. First, it is critical to con-
sider the child’s acquisition of an ambient language phonological knowledge
base. In linguistics, this dimension has often fallen within classical (Chomsky &
Halle, 1968) and contemporary (Prince & Smolensky, 2004) views of phono-
logical acquisition. A second dimension of phonological acquisition that has
been included in understanding phonological acquisition includes the per-
ception and action components of the infant’s suite of biological capacities
available for acquiring phonology. Inclusion of this “embodied” aspect into
165
166 Barbara L. Davis
understanding of the nature of phonological acquisition implies that both the

phonological knowledge base and the use of bodily capacities for implementa-
tion of this knowledge base for communicative function within the environment
are necessary for specifying the nature of the ontogenetic acquisition process
in human infants. In this view, both the neural-cognitive knowledge base for
phonology and the perception-action capacities supporting use of this knowl-
edge base are necessary to specify the acquisition process. Neither alone is suf-
ficient. This multifaceted view of phonological acquisition is congruent with
complexity science, where multiple and diverse components are seen as co-
creating the eventual behavioral product. In an extended consideration of com-
plexity science as a theoretical base for specifying acquisition of phonological
capacities in contemporary humans, this perspective has recently been termed
“Knowing and Doing” (Davis & Bedore, 2013).
Modern children’s acquisition of phonology potentially enables a short time
scale view across ontogeny of the phylogeny of human language capacities
within Tinbergen’s conceptualization. It also supports a consideration of the
value of complexity science in understanding ontogenetic processes and how
they may illustrate important aspects of phylogenetic origins of language. The
fundamental principles of complexity science are compatible with responses
to Tinbergen’s ethnographic questions about ontogeny. Human infants’ intrin-
sic biological capacities and general-purpose enabling mechanisms mature on
an extended developmental timetable when compared with other species. This
extended timetable for development of adult complexity has been character-
ized as neoteny. Because the process of development is extended in human
infants relative to other modern species, the human ontogenetic process affords
an opportunity for examining the emergence of a complex system when it is in
its simplest early phases.
2 Historical Background
One fundamental aspect of the acquisition of language in humans is impor-
tant to illuminating the dialogue on the relevance of complexity constructs to
understanding the acquisition of knowledge as it might relate to phylogenic
considerations. The nature versus nurture debate has been a core aspect of the
continuing controversy regarding the origins of complex capacities in humans
for centuries (viz. Descartes, 1637; Kant, 1924). It has formed a fundamental
area of difference in the epistemological study of the origins of children’s acqui-
sition of knowledge across multiple areas of description, not limited to phonol-
ogy or even to language (see Woolhouse, 1988). Motor, social, perceptual and
cognitive domains of acquisition have been sites of study for this issue. A
brief overview of nature versus nurture as a continuing question in understand-
ing the origins of complex capacities provides a conceptual background for
A Complexity View of Ontogeny as a Window on Phylogeny 167
considering a complexity perspective on phonological acquisition in the present

context.
The earliest competing hypotheses about the origin of knowledge in humans,
particularly human infants, are to be found in antiquity. They formed the basis
for later focus on the origins of language knowledge. Plato was the first philoso-
pher to propose innateness as a solution to the question of the origins of knowl-
edge. He proposed mental “Essences” as necessary building blocks for the
later knowledge a child acquires. Platonic philosophy asserted that humans
are the only living creatures who possess innate ideas. In contrast, Locke, a
seventeenth-century empiricist, argued that the child is a tabula rasa or blank
slate on which knowledge is written. Diverging from Plato’s philosophical
perspective, Locke emphasized the role of experience in the acquisition of
complex knowledge. Descartes (1637), following Platonic philosophy, asserted
that humans are the only living creatures who possess innate ideas. However,
Descartes saw the source of innate human ideas as divine, albeit as an endow-
ment from the God of Christian divinity rather than the Platonic Greek gods.
These early competing hypotheses about the origins of knowledge formed the
basis for the later focus on language knowledge.
In later centuries, de Saussure (1916) focused the debate about the origins
of knowledge in young children on language among the other dimensions of
knowledge that the child was acquiring at the same time. He conceived of
language as a cultural phenomenon that was the result of input from cultural
inventions unrelated to biology. The origin of these complex behaviors resided
outside of the individual, compatible with the empiricist position espoused by
Locke much earlier. De Saussure’s view of language diversity was in contrast
to Platonic philosophy, which placed the language capacity within the child.
Roman Jakobson (1941) focused on children’s phonological acquisition as
a manifestation of mental representations of perceptual contrasts in human
speakers. He proposed a deterministic order of acquisition by children that was
universal across languages and individuals. Following de Saussure, children
were supposed to learn universal feature or contrast hierarchies unrelated to
biological constraints. Jakobson’s proposal was not based on actual data from
children.
Chomsky and Halle’s (1968) early work on linear generative phonology
expressed in The Sound Pattern of English (SPE) formed the first driving force
for proposals of knowledge-based phonological essences in the philosophical
tradition of Plato and Descartes. Chomsky and Halle proposed knowledge-
based essences (termed “distinctive features”) that were present at birth to
guide acquisition and revealed by maturation. More modern constructions of
phonological theory related to the origins of complex knowledge in human chil-
dren can be found more recently in Optimality Theory (OT; Prince & Smolen-
sky, 2004). The Handbook of Phonology (edited by DeLacy, 2007) provides a
comprehensive overview of these perspectives. In these constructions of the

acquisition of phonology, the child learner is seen as moving toward adult
language competence by re-ranking a set of mentally available constraints.
Observed “surface” forms in child output and in languages arise from the res-
olution of conflicts between these ranked grammatical constraints. In some OT
perspectives, constraints are considered innate and universal (e.g., Dekkers,
2000). Other OT approaches do not assume innate or universal status (e.g.,
Boersma, 1998). But the origin of knowledge is seen in a priori Platonic
essences that are revealed by maturation and parameterized by perceptual input
from the environment about ambient phonology.
While Chomsky and Halle’s proposals formed the first basis for the modern
phonological paradigm in which language is considered modular and “special”
in humans, their basis for understanding acquisition of phonology has much in
common with contemporary phonological theory (e.g., OT). In both conceptu-
alizations, phonology is seen as a symbol manipulation system based on com-
putational properties shared by speakers and listeners. The speaker and listener
are viewed as computational devices utilizing a modular system to communi-
cate with formal linguistic symbols (see Scott-Phillips, 2010, for a review of
this perspective). Critically, language is viewed as a computational system that
operates independently of function. Evidence for computational operations is
found in frequency counts of language forms. Description of these forms and
their frequency is equated with explanation of their origins within the under-
lying grammar. Assertions of the “psychological reality” of speech elements
rather than a neural basis for system properties are the norm in this perspective
on the origins of complex language knowledge in human children. Actions of
the body in perceiving and producing speech are not intrinsic to the compu-
tational system. Function of the system in achieving the young child’s social
needs for survival and growth in knowledge is also outside the scope of the
computational system envisioned by classic and contemporary phonological
theories.
3 Alternative Theoretical Proposals

An alternative theoretical model for understanding the origins of phonologi-
cal knowledge in human infants can be found in complexity theory (Davis &
Bedore, 2013; Pellegrino et al., 2009). Complexity science refers to a view of
complex phenomena where local components of a system interact and complex
behavior emerges from those interactions (Holland, 1998). Local interactions
create interconnectivity. Complexity theory has been employed to characterize
complex systems as diverse as weather systems (Prigogine & Stengers, 1984),
developmental biology (Kauffman, 1995), and comparative studies of the bio-
logical structures and functions of diverse organisms (Camazine et al., 2001),
phenomena that had long resisted linear explanations.
It is useful to take into account some of the claims of complexity science

as they are illustrated by emergence of complexity in children’s acquisition
of the phonological component of language. Within the tenets of complex-
ity science, phonological knowledge and behavioral patterns can be seen as
emerging from connections enabled by general-purpose child capacities such
as learning and cognition as opposed to language-dedicated modular mech-
anisms. Heterogeneous inputs to the system from child intrinsic production,
perception, neural cognitive and social interaction capacities are tuned by feed-
back from parent input available to the child. These interactions among system
components are context-dependent in application during social routines across
acquisition. Components operate in a hierarchy (Wilson & Holldöbler, 1988) in
which “higher” (e.g., neural-cognitively instantiated knowledge) and “lower”
(e.g., perceptual and production system processing) levels of the overall com-
plex system are interactive systems that develop interconnectivity in support
of growth in complexity of phonological knowledge and output patterns. Lan-
guage – in particular, the phonological component of language that is the focus
here – can be seen as a communal communication system. This system includes
both the micro (i.e., caregiver interactions) and macro (i.e., the larger cultural
community of the child) dimensions of the social world of the child as well
as the structure(s) that specify spoken language forms used in communicative
interactions.
As an example, local social interactions between mothers and children create
mutually understood communications. These communications are coded with
recognizable sounds as the child utilizes her perception-cognition-production
system abilities in salient social routines with her mother. “Peek-a-boo” rou-
tines in the first year of life require participation in formulaic exchanges using
verbal sequences that have meaning to the participants.
Across development, growth in language capacities must be accompanied
by growth in the phonological system to convey those ideas to others (i.e.,
the /r/ and /s/ sounds in “rhinoceros” for a trip to the zoo). Perceptual capac-
ities are functional in enabling the child to differentiate between salient and
non-essential incoming stimuli. Cognitive-neural capacities support the young
child in gaining and storing increasingly precise phonological knowledge struc-
tures. Production system capacities enable the child to perform more diverse
goal-directed speech movements related to her ambient language require-
ments. All of these strands cooperate in enabling the system to gain com-
plexity. Again, none are considered sufficient alone to the task of acquisi-
tion of phonology; implying a complex system approach to understanding the
process.
A complementary theoretical perspective to complexity theory can be found
in embodiment theory. Prominent representatives of the embodiment per-
spective are found in philosophy (M. H. Johnson, 1987), linguistics (Lakoff
& Johnson, 1999), cognitive science (Varela, Thompson, & Rosch, 1991),
neuroscience (Damasio, 1994; Edelman, 1992), artificial intelligence (Clark,

1997), and developmental cognition (Smith & Breazeal, 2007). In embodiment
perspectives, mental activity and underlying brain activity cannot be under-
stood outside the context of bodily activities. According to Clark (1997), the
goal is to learn to manipulate a mental model in the same way as we originally
manipulated the real world … to internalize cognitive competencies rooted in
manipulations in the external world. (See Iverson, 2010, for an extended treat-
ment of the importance of movement in the development of cognition.)
Applied to phonological acquisition, embodiment approaches support the
assertion that the human speech production and perception systems may form
a primary limiting factor on acquisition of speech output as well as, potentially,
the structure and processing of phonological knowledge (Davis & MacNeilage,
2000). As a biological system, the speech production-perception-neural appa-
ratus actually operates to mold or sculpt the set of complex phonological pat-
terns used for linguistic communication by humans. Sound patterns that utilize
the human production and perception systems in achieving maximal percep-
tual distinctiveness for minimal production difficulty are valuable assets for
efficient and rapid human communication. Those attributes spring from the
tandem operation of the peripheral and physically embodied perception and
production systems in speaking and listening. They are stored in neural tis-
sue based on repeated use in speaker-listener interactions (Lindblom, 2008).
Critically, acquisition of the complex system of phonological knowledge and
behavior is embodied in the sense that the child’s body operates in the world
as a critical factor in the complex system underlying acquisition of her phono-
logical system.
A second complementary theoretical perspective that supports the proposals
of complexity science in understanding phonological acquisition is found in
dynamic systems theory (Thelen & Smith, 1994). A dynamic system is a sys-
tem that is stable, yet far from thermodynamic equilibrium. Equilibrium con-
ditions can only be maintained by a continuous flow of free energy into and out
of the system according to dynamic system perspectives. Biological systems
have been evaluated as prime examples of open systems within this paradigm.
Over time their order and complexity are not only maintained but may actually
increase, as in development. Researchers have employed dynamic systems to
model early acquisition of complex action in the locomotor domain (Thelen,
1995; Thelen & Smith, 1994), in neuro-physiological structures, and in vocal
tract biomechanics in neonates (Steven M. Barlow, 1998). Thelen and Smith
(1994) tested dynamic systems in the domain of infant acquisition of crawl-
ing and walking. Based on this body of work, Thelen and colleagues propose
that the infant’s own system components, including production, perception, and
cognitive capacities, are mutually interactive across development. None can
individually specify the nature of the complex system.
Dynamic systems emphasize the changing nature of a complex system across

time. Diverse inputs from the environment interacting with child output create
this change. Dynamic and qualitative changes occur in the complex system
across the process of acquisition. This construction of the process of acquir-
ing complexity in phonological form is in contrast with linguistic perspectives
(Prince & Smolensky, 2004). These perspectives propose a progressive revela-
tion of child internal preexisting, static phonological knowledge structures via
maturation guided by external perceptual triggering from input.
Multiple causes of observable changes lie in the heterogeneous sources of
input into the complex system. Oyama (2000) has suggested the term con-
structivist interaction, implying that interpenetration between organism and
environment is integral to the acquisition of a complex behavioral and knowl-
edge repertoire. Thelen (1991) asserted that perceptual-motor activities result-
ing in a complex vocal behavior require coordination of multiple subsystem
elements. This coordination produces spatially and temporally organized
behavior with a cohesive structure, the essential defining quality of human
speech production.
Operation of the elements of the production system, including respiratory,
phonatory, and articulatory capacities, and the observable perceptual biases
toward sensitivity to salient acoustic and visual features embody two critical
dimensions of this system. The system is open. The status of the infant’s pro-
duction system, as well as sensory processes, externally available adult speech
and language models, and the socio-cultural milieu requiring function from the
child are crucial inputs driving the nature of the overall system. The system is
dynamically open to inputs from each of these subcomponents.
The confluence of principles available from these complementary theoreti-
cal perspectives enables a proposal that can be termed a biological-functional
approach to phonological acquisition. In this view, the heterogeneously deter-
mined, embodied, and dynamically changing complex system found in the
young human child acquiring phonology supports the proposal that ontogeny
serves as a window on phylogenetic explanation.
These theoretical perspectives ask that we encompass consideration of both
biological and social dimensions of function in understanding phonological
acquisition. Acquisition of increasingly mature behavioral patterns and under-
lying knowledge about phonology are viewed as being accomplished by inter-
actions of biological and social components of a complex system across the
process to achieve maximal function for the young child. In acquiring knowl-
edge patterns, the child is internalizing neural/cognitive competencies rooted in
social manipulations in the external world. To master speech-related behavioral
patterns, the child is assembling functionally adaptive behavioral patterns that
exploit intrinsic dynamics of the production and perception systems to respond
to local social contexts. Both growing complexity in knowledge and growing
complexity in the behavioral repertoire enable the young child to function in the
world around him or her to acquire relevant information for long-term survival
in the world. Thus, in a biological-functional approach, outcomes of phonolog-
ical acquisition result from multiple interactions among heterogeneous aspects
of a complex system.
These explanatory principles from diverse theoretical perspectives provide
a potential window on explanation for phylogeny of human language. They
add the components of function and change over time in an open system that
is heterogeneous and enabled by feedback as tools for explanation of patterns
observed. In terms of Tinbergen’s question about ontogeny, these principles
enable access to “How did the behavior arise in the organism’s species?”
4 Considerations from Data on Acquisition

Data for evaluating a behavioral–functional-based complexity perspective can
be found in the earliest phases of speech-like behavioral output in human
infants; the pre-linguistic babbling period. Perceptually rhythmic speech-like
syllables are apparent to listeners between seven and nine months in infants
who are developing typically. These robustly apparent rhythmically speech-
like vocal outputs give evidence of aspects of the infant movement system
that are available for initial approximation of the serial organization of adult
speech (Davis & MacNeilage, 1995). They do not appear reliably in infants
with diminished perceptual capacities, emphasizing the importance of auditory
perceptual capacities in onset of babbling (Davis et al., 2005; Von Hapsburg &
Davis, 2006). They are also seen as reflecting neural integrity for support-
ing rhythmic behaviors across domains, including other gross and fine-motor
aspects of infant behavior in this period (Thelen & Smith, 1994), And they
appear in reliably more mature form in social conditions where there is mater-
nal contingency (Gros-Lewis et al., 2006) supporting the importance of a social
component for developmental processes. These diverse foundations support a
multifaceted basis for the emergence of complexity in infant speech-like output
related to the mature patterns of the language community. This early foundation
includes embodied biological perception, production, and neural capacities as
well as social contextual components.
The sounds and sequential structures in this earliest period of speech-like
behavior appear quite restricted. A relevant question of interest in considering a
complexity perspective on acquisition concerns the relationships between early
patterns observable in infants during canonical babbling and overall phono-
logical patterns that must be mastered for intelligible speaking and listening
in their ambient language environment. The process of acquisition from the
earliest beginnings to the eventual outcome of a mature complex phonology
highlights the resonance of questions of ontogeny with phylogenetic questions.
Both the process of acquisition and the structural sound system products of that
process are of interest in understanding the emergence of complexity in this
context.
A cross-linguistic approach that includes analysis of diverse languages is of
value to understanding common patterns as well as underscoring the diversity
of language complexity that must be learned by infants to achieve intelligi-
ble message transmission in their own unique language. Cross-language anal-
yses of output patterns have a long history in linguistic study (e.g., Maddieson,
1984). Cross-linguistic profiles of child acquisition patterns are a newer aspect
of study relative to understanding the nature of child-adult similarities and
differences (Kern & Davis, 2009). Comparison of patterns in infant output
and in languages provides a cornerstone for understanding the ambient lan-
guage targets to be achieved by young children in diverse language environ-
ments. It shows, as well, the numerous ways in which languages have exploited
the speech production mechanism for achieving message distinctiveness. As
noted earlier, aspects of the complex system include the physical implementa-
tion mechanisms of the production, perceptual, and neural sub-systems. These
sub-systems are common across child and adult speakers where phonological
implementation is overlaid on the primary and basic life functions of these bod-
ily systems. But the young child presents a dynamic and changing system at
the levels of neural-cognitively instantiated knowledge and implementation of
that knowledge through speech-related actions. In contrast, adult speakers and
listeners could be viewed as embodying a relatively static complexity where
stable properties of a complex system are used in rapid message transmission.
However, the congruence of adults and infants enables the child to begin the
process of mastery of phonology within social interchanges as their growing
system functions to enable communication of meanings using vocal means. As
well, the behavioral phenotypes of consonants, vowels, serial properties of syl-
lable types and within and across syllable patterns must be accounted for as
another dimension of complexity reflected in outcomes of operation of bod-
ily systems. In a complexity perspective these observable structures that serve
functional purposes emerge from the operation of the body in adults and in
children as they interact in social contexts.
4.1 Stability and Change

The behavioral sound patterns that are found to be in common in infant out-
put and in languages within and across diverse ambient language environments
suggest the concept of stability. Stable properties in phonological components
can be considered as those that are available early to infants in the first year of
life who are producing canonical babbling and are also retained by adult speak-
ers using phonological inventories who are fully mature speakers. Within a
dynamic system approach, these stable output properties can be seen as “stable
states” (Thelen & Smith, 1994). In contrast, properties that are modified over
time in acquisition reflect needed change in the system. Change is necessary
in order to maximize the sound properties available to produce rapid message
transmission with the phonological elements available. If some aspects of infant
systems in common with adults connote stability, others change with maturity
and social input to enable increases in message complexity. Again, both sta-
bility and change are necessary components of the system to enable children
and adults to use the system for communication with one another as the child is
simultaneously acquiring new aspects of the system for communicating more
and diverse ideas about the world.
4.2 Stability: Data

First, let us consider data that would connote stability in common across infants
and adult speakers. In this regard, it has been observed that some robust patterns
in babbling preference are retained in adult users of contemporary languages
(MacNeilage et al., 2000a; 2000b). According to Bell and Hooper (1978), there
is a tendency for words in the world’s languages as well as in early infant
productions to begin with a consonant and end with a vowel. The open syl-
lable (CV) is both the only universal syllable type in the world’s languages
(Maddieson, 1984) and the favored syllable type of babbling across a number
of languages studied to date (e.g., Davis & MacNeilage, 1995; Kern & Davis,
2009).
Some sound types occur more frequently both in babbling and in languages.
Simple stop consonants and nasals favored in babbling are highly frequent in
languages of the world (Maddieson, 1984). These sound types tend to dominate
the repertoire of languages with small systems (< 15 phonemes) containing a
few segments characterized as articulatorily “simple” (Lindblom & Maddieson,
1988; Lindblom, Krull, & Stark, 1993). Maddieson (1984) showed that all 317
observed languages he analyzed exhibited a version of [p] and [t], at least one
nasal consonant, and a high frequency of [w] and [j]. Further, 88 percent of
language phonologies included the vowel [a].
Certain within-syllable CV co-occurrence patterns favored in babbling
within and across languages are also retained in mature language users. Studies
of babbling patterns in widely varied languages indicate many infants prefer
within-syllable patterns that show lack of movement between the consonant
and the vowel portion of the syllable (i.e., labials and central vowels, coro-
nals and front vowels and dorsals and back vowels). Studies include children
speaking English (Davis & MacNeilage, 1995; Oller & Steffans, 1993; Giulivi
et al., 2011), Korean (Lee, Davis, & MacNeilage, 2008), French (Vihman,
1992), Mandarin Chinese (Chen & Kent, 2005), and Equadorean-Quichua
(Gildersleeve-Neumann, Davis, & MacNeilage, 2013) as well as in English-

Serbian bilinguals (Zlatic et al., 1997).
In these languages, children most frequently (although not universally) use
their vocal mechanisms for producing rhythmic syllables without articulatory
movement between consonant and vowel within syllables. This relationship
has been termed the Frame-Content Theory (FC; MacNeilage & Davis, 1990).
The FC theory proposes that the aspect of the infant movement system avail-
able for replicating speech like output at the onset of canonical babbling is the
jaw. Rhythmic open-close jaw oscillations without independent movement of
the tongue within syllables result in a listener percept of consonant and vowel
properties dominated by least movement.
Relative to within-syllable patterns retained in languages that are often
present in infant syllable-level inventories, Janson (1986) studied consonant-
vowel relationships derived from written texts of five languages (Finnish, Turk-
ish, Latin, Latvian, and Setswana). Maddieson and Precoda (1991) studied
consonant-vowel relationships derived from dictionary counts in five additional
languages (viz., Hawaiian, Rotokas, Piraha, Kadazan, and Shipibo). In both
studies, there was a significant tendency for coronals to favor front vowels and
disfavor back vowels and for dorsals to disfavor back vowels. There was no
tendency shown in these languages for labial central vowel co-occurrences.
MacNeilage and Davis (1993) also analyzed intrasyllabic trends involving stop
consonants, nasals, and vowels in ten languages other than those studied previ-
ously. Their analysis of English, Estonian, French, German, Hebrew, Japanese,
New Zealand Maori, Quichua, Spanish, and Swahili showed a significant ten-
dency for dentals/coronals to favor front vowels. This front closure portion was
also found to disfavor back vowels. A significant tendency for dorsals to disfa-
vor front vowels was also observed in the data. Additionally, there was a non-
significant trend for dorsals to favor back vowels. In seven languages, labials
were predominantly associated with central vowels, coronals to front vowels.
In eight languages, dorsals were associated with back vowels.
Thus, there appear to be commonalities between infants and adult speakers
in both basic sound and syllable properties that appear with regularity in output
inventories. Questions that should be posed in this regard include the following:
1) what is the relationship of child patterns in pre-linguistic babbling to ambi-
ent language patterns across languages? and 2) Do child patterns vary cross-
linguistically in the babbling period? Study of babbling, as previously noted,
enables consideration of the complex system embodied in perceptually speech-
like behaviors that occur and the early period of onset of these behaviors.
Statistical evaluation of relationships between infants and languages has
been pursued using an infant database available through PhonBank (MacWhin-
ney & Rose, 2005). Participants in the presently reported data include twenty-
six children in Turkish, Dutch, Tunisian, French, Romanian, and English
Table 7.1 Comparison of Consonant Vowel Association Levels Within

Syllables
Co-occurrence Finding Languages Other Languages
Coronal/front =/+ 5 − Romanian

Labial/front = 5 − 1 Romanian
Dorsal/front = 5 + 1 Romanian
Coronal/back − 5 = 1 A-Eng.
Labial/back = 4 − 2 – 1 French, 1 A-Eng
Dorsal/back − 5 = 1 French
Coronal/central + 6
Labial/central + /= 6
Dorsal/central − 3 Dutch, Tunisian, English
+ 3 Romanian, French, Turkish
+ Sig. greater than AL = No sig. diff. from AL; – Sig. less than AL
language environments. Infants were typically developing and between 7 and

12 months of age. These data are part of a larger study (Kern & Davis,
2009). For the larger study, data collection encompassed the canonical babbling
period. Canonical babbling is defined as rhythmic alternations between conso-
nant and vowel-like properties, giving a percept of rhythmic speech that sim-
ulates adult output without conveying meaning (Davis & MacNeilage, 1995;
Oller, 1980).
Spontaneous comfort state vocalizations that were perceptually rhythmic
and contained a consonant-like closure phase + vowel-like open phase within
a single utterance string (viz., CV, CVCV, CVCVCVCV) were audio- and
video-recorded bimonthly for one hour from infants in each language environ-
ment. Language data for comparison included 1,000 words randomly selected
from dictionaries on computers in each of the ambient languages. Data ana-
lyzed that will be reported here from the larger study by Kern and Davis
were sequential consonant-vowel patterns within syllables (viz., consonant-
vowel co-occurrences of consonant labial, coronal, and dorsal place of artic-
ulation). Also reported are across-syllables patterns for reduplication ver-
sus variegation and changes in (1) place versus manner for consonants and
(2) height versus backness for vowels. The standardized residual [observed-
expected/SQRT(expected)] was calculated for statistical comparison of infant
and language properties.
Results for intrasyllabic consonant-vowel co-occurrences are shown in
Table 7.1. Based on FC predictions of lack of independent movement of
articulators separate from the jaw cycle, associations are predicted to be
found between labial closures and central vowels, coronal closures and front
vowels, and dorsal closures with back vowels. As noted, a + in the Finding
Table 7.2 Intersyllabic Infant-Language Matching Results for

Intrasyllabic Properties
Sequence Value Languages Other Patts. Languages
CVCV Reduplication = 6
Variegation = 6
Height = 5 − Tunisian
Backness = 5 + Tunisian
Place = 6
Manner = 6
+ Sig. greater than AL = No sig. diff. from AL; – Sig. less than AL
column indicates that the infants showed significantly more of that CV co-
occurrence than was found in their ambient language environment. An = indi-
cates that the frequency of occurrence was not significantly different than that
in the ambient language. A – displays comparisons where the infants produced
significantly fewer of that CV combination than was found in their ambient
language. As shown, in 5.5 of 9 potential CV co-occurrence environments,
the infant’s frequency of occurrence was equal to the frequency in their own
ambient language. Significantly less frequently occurring in the infants were
coronal-back, dorsal-back, and dorsal-central associations. In every case except
dorsal-central vowel associations, four to six of the language groups followed
the significance trends. Exception language groups included two languages
where the children’s frequency of occurrence was equal to their ambient lan-
guage frequencies, one language where the children produced more than their
ambient language, and three languages where the children produced fewer of
that CV co-occurrence than their ambient language. Importantly, the predicted
CV co-occurrences of labial central, coronal front, and dorsal back were equiv-
alent for six, five, and five of the languages, respectively. While the infants were
not universally producing the within-syllable patterns found in their languages,
in particular, for the patterns predicted by the FC hypothesis, they were pre-
dominantly doing so. There were no consistent counter-trends in the picture
where one type of CV co-occurrence was found to be predominantly different
from the ambient language pattern. The numbers in the final two columns indi-
cate findings that were not consistent with the broader trend. In each case, the
number of children in that language who displayed the diverse trend is shown.
Six of 27 children showed individual patterns that were different from the group
trends overall.
Intersyllabic infant-language matching results for intrasyllabic properties are
displayed in Table 7.2. The use of the +, =, and – signs are the same as for
the within syllable CV co-occurrence results. For reduplication and variegation
frequency comparisons between infants and ambient languages, six out of six
infant groups showed no statistically significant difference from their own
ambient language properties. Thus, these infants can be seen as anchoring early
in their language regarding the frequency of variegation relative to reduplica-
tion shown across syllables. For the comparison of variegation in vowel height
versus vowel backness, infants in five out of six languages showed the same pat-
tern as their own ambient language. The Tunisian infants were the only group
that showed diversity in the patterns of vowel variegation. For consonant var-
iegation, six out of six languages showed no statistically significant difference
from the frequency of variegation in either place or manner dimensions from
their own ambient language patterns for frequency of occurrence.
In summary, there is abundant evidence of stability in these data. Stability
is indicated when phonological patterns that infants are producing in canon-
ical babbling in their first year of life are also found in mature adult speak-
ers in their own ambient language. In the case of within-syllable CV co-
occurrences, infant participants across the six languages showed predominant
matching of the within-syllable characteristics of their ambient language. It
should be noted that only the Dutch language does not show predicted CV co-
occurrences (MacNeilage et al., 2000; Kern & Davis, 2009). Relative to inter-
syllabic reduplication-variegation patterns, there was matching of ambient lan-
guage variegation levels across syllables in all six languages analyzed. Types of
variegation for consonants and for vowels were also remarkably similar across
these languages.
These findings for within- and across-syllable comparisons for infants rela-
tive to their ambient language patterns in these contemporary languages show
remarkable levels of similarity between infants and mature language patterns.
This finding implies that these types of patterns characteristic of mature lan-
guage users in these six languages are available to the movement system of
beginning users of rhythmic syllables in those languages; an early “anchor”
in the language if you will. Although they are also characteristic of these lan-
guages, they are not found in all of the languages. This finding of relative sta-
bility between infants at the onset of syllable use and their language suggests
that these patterns could indicate a route to syllable-like organization of serial
output in early hominids initiating vocal communication based on the jaw cycle
and accompanying phonation co-opted from earlier functional routines. Evolu-
tion may likely have preserved functional aspects of the internal structure of the
jaw cycle (viz., intrasyllabic regularities) elaborating over time with additional
adaptations seen in increase of independent articulator movements within syl-
lables in the service of increase in message complexity. The lack of universal
occurrence of these patterns across languages suggests that individual paths
to language complexity are potentially unique within a landscape of dramatic
regularity in sequential movement patterning.
5 Change
We have reviewed some sites of early occurring stability in matching by
infants in their own output to their ambient language characteristics. Results
reported have been interpreted as early anchors of stability between infants
and languages in the complex system observable in production output across
acquisition. Understanding of acquisition from a complexity perspective must
include progressive diversification across acquisition or “change” in infant out-
put capacities as well. This progressive diversification in the inventory of sound
types and how they are produced in sequences relative to ambient language
patterns is usually considered a critical index of increasing complexity toward
mature phonological capacities.
a Change: Data
One index of change in complexity across acquisition is in the area of con-
sonant assimilation. Consonant assimilation is a type of simplification that is
observed in early child inventories where consonants in a word are changed to
become more like one another (for instance, dog is produced either as “dod”
or as “gog”). Children who are developing typically most often resolve these
kinds of simplification patterns by 36–48 months of age as they progress toward
diversified sound inventories and accuracy using those inventories relative to
ambient language targets they are attempting.
Kim (Kim, 2010; Kim & Davis, 2009) conducted a longitudinal study of
consonant assimilation as an index of required change in phonological com-
plexity in ten children learning English. The children’s word-based output was
analyzed between 12 and 36 months of age. In addition, data were divided into
age time periods to examine change over time: 12–18; 19–24; 25–30; and 31–
36 months. Audio and video data were recorded bimonthly for one hour from
each child. Data gathered were part of the Texas Davis database on PhonBank
(MacWhinney & Rose, 2005). Spontaneously occurring words that could be
identified as such by listeners were analyzed. CV and CVCV forms were ana-
lyzed. This included 1467 assimilated word forms (viz. 1058 CVC and 409
CVCV). The forms that contained consonant assimilations accounted for 7 per-
cent of the entire corpus studied for the ten children. There were 20,522 words
in the corpus overall during the entire period. The results reported here are
illustrative of the data from the larger study (Kim, 2010).
Results for consonant place assimilation showed that assimilations occurred
in the following levels of frequency: Labial > Coronal > Dorsal (see
Figure 7.1). The relative frequencies were calculated in ratios of assimilation
to target words. For example, the ratio of 271 labial assimilations to 335 target
words containing labials accounted for .81 overall. The same trend was found
in both CVC and CVCV word forms. This finding indicates that word forms
1.00
0.81
0.80
Ratios
0.60 0.54 Labial Assimilation
0.40 Coronal Assimilation
0.21 Dorsal Assimilation

0.20
0.00
Place Assimilation
Figure 7.1 Consonant Place Assimilation Patterns.
were most likely to change to labials (e.g., bag → bab) when there were two
different consonant places in the CVC or CVCV word target, next most likely
to change to coronals (e.g., dog → dod), and least likely to change to dorsals
(e.g., pig→ gig).
For consonant manner of articulation, the pattern of assimilations in CVC
and CVCV word forms (Figure 7.2) was as follows: Stop ࣙ Nasal > Fricative
assimilation. Ratios were computed in the same way as previously described for
consonant place. This result indicates a dominant propensity to change toward
oral and nasal stop manner, followed at very low levels by changes toward frica-
tives. Here the same trend was found in any consonant sequences in targets
except for the nasal-stop sequence.
Developmental patterns in resolution of consonant assimilation were also
analyzed. The data for each participant were divided into the following time
periods: 12–18; 19–24; 25–30; and 31–36 months. Results of this analysis are
displayed in Figure 7.3 as raw frequencies. In the figure, it can be seen that
frequency of assimilation was the highest at Time 2 (18–24 months), and then
decreased across the remaining period of analysis. Labial and coronal assimi-
lation persisted as the highest frequency patterns produced by the children in
assimilated forms.
1.00
0.91
0.79
0.80
Ratios
0.60 Stop Assimilation

Nasal Assimilation
0.40 Fricative Assimilation
Other Assimilation
0.20
0.06 0.05
0.00
Manner Assimilation
Figure 7.2 Consonant Manner Assimilation Patterns.

Frequency of Place Assimilation 350
300
48
250
19
200
Dorsal Assimilation
191 114 9
150 Coronal Assimilation
95 Labial Assimilation
100
25
50 19 100
70 66
41
0
Time1 Time2 Time3 Time4
Time periods
Figure 7.3 Labial, Coronal, and Dorsal Assimilation Pattern Changes over
Time.
Overall, these typically developing children followed a Labial > Coronal >
Dorsal pattern in their assimilations of consonants within word forms in this
early period of development. However, in the second time period where assim-
ilation frequency peaked, coronals dominated, indicating a change in relative
frequencies as discrete time periods are considered.
As noted, this perspective aims to take a movement-based approach to under-
standing the acquisition of phonological capacities in young children. Results
can be conceptualized as a movement-based hierarchy relative to movement
patterns that are available to young children. Labials are the most available for
children (Davis et al., 2002). They involve mandibular movement only with-
out engagement of the tongue. Coronals are somewhat less frequent but still
high in frequency relative to dorsals. They involve complete tongue tip contact.
Within this movement-based approach, dorsals are the least available (Locke,
1983). They involve the back of the tongue, which is much less flexible than
the tongue tip relative to movement patterning. It should be observed that these
relative frequencies found in assimilated word forms match the L > C > D fre-
quencies found in early words overall (Davis, MacNeilage, & Matyear, 2002).
In contrast, languages show a pattern of more frequent coronals than labials
(Maddieson, 1984).
Relative to consonant manner, the hierarchy of frequency found in assimi-
lated word forms was: Stop ࣙ Nasal > Fricative. Again, this pattern can be
interpreted within the framework of a movement-based hierarchy. Stops and
nasals are also more available to the young child’s movement system. Both
involve complete closure of the oral tract followed by release during movement
sequences of close-open syllables (Davis et al., 2002). In contrast, fricatives

involve fine adjustments of varied degrees of closure and far more precision;
they are found at very low frequencies in the inventories of young children at
the onset of word use (Gildersleeve-Neumann, Davis, & MacNeilage, 2000).
b Summary
Results from analysis of a frequently attested pattern in early child speech indi-
cate that these children continue to favor labial and coronal forms available to
the movement system from the onset of word use when they produce assim-
ilations. Thus, acquisition of serial complexity in frequently occurring word
forms begins with assimilation to available movements. Resolution of these
constraints guided by perceptual input from ambient language is a process of
overcoming movement constraints to match ambient language word target com-
plexity. Overall, results support a proposal of movement-based principles guid-
ing output patterns in formative periods of phonological development guided
by socially mediated perceptual input to meet functional goals within the envi-
ronment.
6 Ontogeny and Phylogeny

Tinbergen (1963) suggests that consideration of the processes and products of
early speech acquisition in human infants might provide a window into phy-
logenetic processes in early users of the vocal medium for communication.
In considering Tinbergen’s questions, contemporary theoretical conceptualiza-
tions found in complexity science, embodiment theory, and dynamic systems
theory have been proposed as an alternative to the dominant linguistic paradigm
for considering acquisition of the phonological component of language. In these
theoretical approaches, phonological acquisition is seen as founded on a com-
plex and embodied dynamic system that changes over time to support social
function in the environment.
The complex system proposal considered here is founded on multiple het-
erogeneous inputs from the child’s biological makeup (viz., neural-cognitive,
production, and perception systems). It is enabled by socially available phono-
logical input from members of the child’s speech community who “show the
way” toward using the sounds necessary to communication of precise linguis-
tic meanings. The social enabling system is embedded in function, as the forms
illustrated by adults in input and the forms produced by children in output are in
service of functional interactions. These functional interactions help the child
construct and use his or her growing knowledge about the world for survival
and for scaffolding myriad types of learning about relevant world knowledge.
In Tinbergen’s sense, this complex system provides a picture of the putative
process that could have been implemented over historical time in early speak-
ers; emergence of a complex and embodied dynamic system implemented
socially for functional purposes of supporting the organism’s survival and
growth in knowledge.
In considering closely the kinds of processes that might have been apparent
over historical time, we have viewed the products of the complex dynamic sys-
tem of the contemporary human infant as characterized by both stability and
change. Early stability in the system creates a background for needed growth
in complexity to produce diverse messages. We have reviewed data indicating
that within-syllable patterns related to lack of movement of active articulators
independent of the jaw in rhythmic syllables are found in common in young
children and adult speakers across a number of languages. These patterns have
also been found in a putative corpus of early speakers (MacNeilage & Davis,
2000c), supporting a proposal regarding their fundamental status as a base for
growth in production system complexity in service of perceptual distinctiveness
in phylogeny.
In addition to stability in the output system as a foundation for rapid message
transmission, ontogenetic and phylogenetic processes must encompass change
to accomplish the functional goal of perceptual distinctiveness. In ontogeny,
children must develop from highly constrained movement types and movement
sequences to ambient language-specific levels of movement variegation that
accurately match their growing repertoire of word targets. Early on, children
depend on available movement properties for early productions of word-based
sequences. Consonant assimilation is an early structural property of word forms
produced by young children. Data reviewed on resolution of assimilations sug-
gests that children rely on available movement patterns at the onset of word
use in assimilations and emerge into word-target-related movement variega-
tion. The process of change toward language-like variegation for word targets
can be seen as based on the child’s complex, embodied dynamic biological sys-
tem tuned by social input for functional purposes. Consistent with phylogenetic
conceptualizations, the process of change is for functional purposes.
REFERENCES
Barlow, Steven M. (1998). Real time modulation of speech-orofacial motor performance
by means of motion sense. Journal of Communication Disorders, 31(6), 511–534.
Bell, Allen & Hooper, Joan B. (eds.) (1978). Syllables and segments. Amsterdam:
North-Holland.
Boersma, Paul (1998). Functional Phonology: Formalizing the Interactions Between
Articulatory and Perceptual Drives. The Hague: Holland Academic Graphics.
Camazine, Scott, Deneubourg, Jean-Louis, Franks, Nigel, Sneyd, James, Theraulaz,
Guy, & Bonabeau, Eric (2001). Self-Organization in Biological Systems. Prince-
ton: Princeton University Press.
Chen, Lei-Mei & Kent, Raymond D. (2005). Consonant-vowel co-occurrence patterns

in Mandarin-learning infants, J. Child Lang. 32, 507–534.
Chomsky, Noam & Halle, Morris (1968). The Sound Patterns of English. New York:
Harper & Row.
Clark, Andy (1997). Being There: Putting Brain, Body, and World Together Again.
Cambridge, MA: The MIT Press.
Damasio, Antonio R. (1994). Descartes’Error: Emotion, Reason, and the Human Brain.
New York: G. P. Putnam.
Davis, Barbara L. & Bedore, Lisa M. (April, 2013). Knowing and Doing: An Emergence
Approach to Speech Acquisition, New York: Routledge, Psychology Press.
Davis, Barbara L., McCaffrey, Helen, Von Hapsburg, Deborah, & Warner-Czyz, Andrea
(2005). Perceptual Influences on Motor Control: Infants with Varied Hearing Lev-
els, Volta Review, Vol. 105(1), 7–27.
Davis, Barbara L. & MacNeilage, Peter F. (1995). The Articulatory Basis of Babbling.
Journal of Speech and Hearing Research, 38, 1199–1211.
Davis, Barbara L. & MacNeilage, Peter F. (2000). An Embodiment Perspective on the
Acquisition of Speech Perception. Phonetica, 57(Special Issue), 229–241.
Davis, Barbara L., MacNeilage, Peter F., & Matyear, Christine (2002). Acquisition of
Serial Complexity in Speech Production: A Comparison of Phonetic and Phono-
logical Approaches to First Word Production. Phonetica, 59, 75–107.
Dekkers, Joost (2000). Optimality Theory: Phonology, Syntax and Acquisition. Oxford:
de Saussure, Ferdinand (1916). Course in General Linguistics (R. Harris, Trans.). Peru,
IL: Carus Publishing Company.
Descartes, Rene (1637). The Philosophical Works of Descartes (E. S. Haldane &
G. T. R. Ross, Trans. Vol. Volume 1). New York: Cambridge University Press.
Edelman, Gerald M. (1992). Bright Air, Brilliant Fire: On the Matter of the Mind. New
York: Basic Books.
Gildersleeve-Neumann, Christina, Davis, Barbara L., & MacNeilage, Peter F. (2000).
Contingencies governing production of fricatives, affricates and liquids in bab-
bling. Applied Psycholinguistics, (21), 341–363.
Giulivi, Sara, Whalen, Douglas H., Goldstein, Louis M., Nam, Hosung, & Levitt,
Andrea G. (2011). An Articulatory Phonology Account of Preferred Consonant-
Vowel Combinations. Language Learning and Development, 7 (3), 202–225.
Kauffman, Stuart (1995). At Home in the Universe: The Search for the Laws of Self-
Organization and Complexity. New York: Oxford University Press.
Kern, Sophie & Davis, Barbara L. (2009). Emergent complexity in early vocal acquisi-
tion: Cross-linguistic comparisons of canonical babbling. In Chitoran, Iona, Coupé,
Christophe, Marsico, Egideo & Pellegrino, Francoise (eds.), Approaches to Phono-
logical Complexity. Phonology and Phonetics Series, Berlin: Mouton de Gruyter.
353–376.
Kim, Namhee (2010). Consonant Assimilation in Early Phonological Development: A
Phonetic Perspective, Unpublished doctoral dissertation, The University of Texas
at Austin, Austin, Texas.
Kim, Namhee & Davis, Barbara L. (April 2009). Consonant Assimilation in early words,
A Phonetic Perspective, Society for Research in Child Language Development,
Denver: Colorado
Lakoff, George, & Johnson, Mark (1999). Philosophy in the Flesh: The Embodied Mind
and Its Challenge to Western Thought. New York: Basic Books.
Lee, Suyoung, Davis, Barbara L., & MacNeilage, Peter F. (2008). Segmental properties
of input to infants: A study of Korean. Journal of Child Language, 35(3), 591–
617.
Lindblom, Bjorn, & Maddieson, Ian (1988). Phonetic Universals in Consonant Systems.
In L. M. Hyman & C. N. Li (eds.), Language, Speech, and Mind (62–78). London:
Routledge.
Lindblom, Bjorn, Krull, Diana, Stark, Johan (1993). Phonetic Systems and Phonologi-
cal Development. In Boysson Bardies, Benedicte, de Schoen, Jusczyk, Peter, Mac-
Neilage, Peter, Morton (eds.), Developmental Neurocognition: Speech and Face
Processing in the First Year of Life. Kluwer: Dordrecht, 399–409.
Lindblom, Bjorn (2008). The target hypothesis, dynamic specification, and segmen-
tal independence. In Barbara L. Davis & Krisztina Zajdo (eds.), Syllable Devel-
opment: The Frame Content Theory and Beyond (327–354). New York: Rout-
lege/Taylor & Francis.
MacNeilage, Peter F. & Davis Barbara L. (1990). Acquisition of Speech Production:
Frames, then Content. In Jeannerod, Marc (ed.), Attention and Performance XIII;
Motor Representation and Control, Hillsdale, NJ: LEA, 452–468.
MacNeilage, Peter F., Davis, Barbara L., Kinney, Ashlyn & Matyear, Christine (2000a).
The Motor Core of Speech: A Comparison of Serial Organization Patterns in
Infants and Languages. Invited submission to Special Millenium Issue. Child
Development, 71(1), 153–163.
MacNeilage, Peter F., Davis, Barbara L., Matyear, Christine & Kinney, Ashlyn (2000b).
Origin of Speech Output Complexity in Infants and in Languages. Psychological
Science, 10(5), 459–460.
MacNeilage, Peter F. & Davis, Barbara L. (2000c). Origin of the Internal Structure of
Words. Science, 288, 527–531.
MacWhinney, Brian & Rose, Yvan (2005). A Shared Database for the Study of Phono-
logical Development, PhonBank, available at http://childes.psy.cmu.edu/phon/.
Maddieson, Ian (1984). Patterns of Sounds. Cambridge, New York: Cambridge Univer-
sity Press.
Maddieson, Ian & Precoda, Kristin 1991. Syllable Structure and Phonetic Models.
UCLA Working Papers Phonetica, 78, 38–49.
Oyama, Susan (2000). Evolution’s Eye: A System’s View of the Biology-Culture Divide.
Durham, SC: Duke University Press.
Oller, D. K. & Steffans, M. L. (1994). “Syllables and Segments in Infant Vocalizations
and Young Children” (speech). In M. Yavas, (ed.), First and Second Language
Phonology 45–61, San Diego: Singular Publishing Group.
Scott-Phillips, Thomas C. (2010). Evolutionary Psychology and the Origins of
Language. Journal of Evolutionary Psychology, 8(4), 289–307.
Oller, D. K. (1980). The emergence of the sounds of speech in infancy. In G. H. Yeni-
Komshian, J. F. Kavanagh & C. A. Ferguson (eds.), Child Phonology 1: Production.
New York, NY: Academic Press.
Pellegrino, Francoise, Marisco, Egidio, Chitoran, Ioana, & Coupe, Christophe (eds.)
(2009). Approaches to Phonological Complexity, Phonology and Phonetics Series,
Berlin: Mouton de Gruyter.
Prigogine, Ilya, & Stengers, Isabelle (1984). Order Out of Chaos: Man’s New Dialogue
with Nature. New York: Bantam.
Prince, Alan, & Smolensky, Paul (2004). Optimality Theory: Constraint Interaction in
Generative Grammar. Malden, MA: Blackwell Publishing.
Smith, Linda B., & Breazeal, Cynthia (2007). The Dynamic Lift of Developmental
Processes. Developmental Science, 10(1), 61–68.
Thelen, Esther (1991). Motor Aspects of Emergent Speech: A Dynamic Approach. In
N. Krasnegor, D. Rumbaugh, & M. Studdert-Kennedy (eds.), Biological and
Behavioral Determinants of Language Development (339–362). Hillsdale, NJ:
Lawrence Erlbaum.
Thelen, Esther (1995). Motor Development: A New Synthesis. American Psychologist,
50(2), 79–95.
Tinbergen, Nicolas (1963). On Aims and Methods of Ethology. Zeitschrift für Tierpsy-
chologie, 20, 410–433.
Von Hapsburg, Deborah & Davis, Barbara L. (2006). Exploring the Effect of Audi-
tory Sensitivity on Canonical Babbling, Journal of Speech, Language, and Hearing
Research, 49, 809–822.
Varela, Francisco J., Thompson, Evan, & Rosch, Eleanor (1991). The Embodied Mind.
Cambridge, MA: MIT Press.
Vihman, Marilyn May (1992). Early Syllables and the Construction of Phonology
in Ferguson, Menn, Stoel-Gammon (eds.), Phonological Development: Models,
Research, Implications, 393–422, Timonium, MD: York Press.
Wilson, E. O., & Holldöbler, Bert (1988). Dense heterarchies and mass communication
as the basis of organization in ant colonies. Trends in Ecology and Evolution, 3,
65–68.
Woolhouse, R. S. (1988). The Empiricists. Oxford: Oxford University Press.
Zlatic, Larissa, MacNeilage, Peter F., Matyear, Christine, & Davis, Barbara L. (1997).
Babbling of twins in a bilingual environment. Applied Psycholinguistics, 18(4),
453–469.
8 Language Choice in a Multilingual Society:
A View from Complexity Science1
Lucía Loureiro-Porto and Maxi San Miguel
1 Introduction
Complexity Science, which can be defined as the science that studies complex
systems and emergent phenomena through the construction of computational
models, has for decades proved to be useful to the understanding and descrip-
tion of quite diverse scientific phenomena, such as flocks of birds or fish, as
well as the behaviour of the brain as an aggregate of neurons and traffic jams
(Mitchell 2009). Those working in the social sciences, from Thomas Hobbes
(seventeenth c.) onwards (Ball 2004) have felt attracted to the shaping of a sta-
tistical view of social phenomena. Yet despite these early attempts, it is only
recently that Complexity Science has begun to be used as a framework for the
study of social phenomena. Indeed, its relevance is now sometimes said to be
greater than any other scientific perspective, given that the boundaries between
disciplines are fuzzier than ever. As noted by Ball (2012: vii), ‘The major chal-
lenges of the twenty-first century are not ones that can be understood, let alone
solved, from a particular academic perspective.’
This chapter approaches the study of complexity in language from a differ-
ent perspective from the majority of other contributions in this volume. It aims
to provide a model of language choice in a multilingual society by examining
the different factors that might play a role in the survival of one (or all) of the
available languages, and does this by adopting an approach based on Complex-
ity Science. It contributes to the study of the sociology of language, discussing
the possible role played by the interaction of agents (i.e. autonomous entities
1 The authors wish to thank Xavier Castelló, Víctor M. Eguíluz and Federico Vazquez for their
indispensable contributions to earlier studies which made the present chapter possible. In addi-
tion, for financial support, Lucía Loureiro-Porto thanks the Spanish Ministry for Science and
Innovation and the European Regional Development Fund (grant FFI2011–26693-C02–02) and
the Spanish Ministry of Economy and Competitiveness (grants FFI2014-53930-P and FFI2014-
51873-REDT); Maxi San Miguel acknowledges financial support from FEDER and MINECO
(Spain) under project FIS2015-63628-C2-2-R. Thanks are also due to Riitta Toivonen, Jari
Saramäki and Kimmo Kaski for fruitful discussions of the topic, and to an anonymous reviewer
for very insightful suggestions.
187
188 Lucía Loureiro-Porto and Maxi San Miguel
which can make decisions) and the structure of the social network in language
choice. It also presents a model for language choice that encompasses parame-
ters such as the role of bilingual speakers and the effects of the kind of network
in which interaction takes place.
2 Complexity Science, Complex Systems, and the Social Sciences
2.1 Complexity
Some 31 different definitions of a complex system can be found in the literature,
notably in earlier works (Horgan 1995). However, Complexity Science is now
a mature, settled discipline with a central paradigm that reflects the triumph of
emergence over reductionism. According to Anderson (1972) and Gallagher
and Appenzeller (1999), for example, complex systems are those in which the
collective behaviour of the constituent parts exhibits properties that cannot be
inferred from their separate individual properties. In a way, it is a modern phras-
ing of the Aristotelian quote: ‘The whole is greater than the sum of its parts’
and it basically implies that individual behaviour is not a guide for collective
behaviour. This behaviour is said to be ‘emergent’ when an effective theory
of the system at some scale, or level of organisation, is qualitatively different
from a lower-level theory. In other words, the emergent behaviour of the sys-
tem cannot be understood in terms of the simple extrapolation of the properties
of the parts. Thus, for instance, sociology is not to be reduced to the study of
the psychology of individuals. Instead of focusing on the individual features
of the elements of the system, attention should be paid to their interactions, as
Schelling (1978) claimed in his pioneering work exploring self-organisation in
societies.
We usually have to look at the system of interaction between individuals and their envi-
ronment, that is, between individuals and other individuals or between individuals and
the collectivity. And sometimes the results are surprising. Sometimes they are not eas-
ily guessed. Sometimes the analysis is difficult. Sometimes it is inconclusive. But even
inconclusive analysis can warn against jumping to conclusions about individual inten-
tions from observations of aggregates, or jumping to conclusions about the behavior of
aggregates from what one knows or can guess about individual intentions (Schelling
1978: 13–14).
Hence, the collective behaviour of elements in complex systems entails emer-

gence of properties that can hardly, if at all, be expected from properties of
those parts. Examples of complex systems include anthills, ants themselves,
human economies, climate, nervous systems, cells and living things, including
human beings, as well as modern energy and telecommunication infrastruc-
tures (San Miguel et al. 2012). When in a system composed of many parts,
Language Choice in a Multilingual Society 189
Table 8.1 Typical Features of Complex Systems (adapted from San Miguel
et al. 2012)
Feature Example of Complex System
Many heterogeneous parts Cities, companies, climates

Complicated transition laws Markets, disease transmission
Unexpected or unpredictable emergence Chemical systems, accidents
Sensitive dependence on initial conditions Weather systems, investments
Networked hierarchical connectivities Social networks, ecosystems
Interactions of autonomous agents Road traffic, dinner parties
Self-organisation of collective shifts Revolutions, fashion
Adaptivity to changing environments Biological systems, manufacturing design
Co-evolving subsystems Computer virus software
Ill-defined boundaries Pollution, terrorism
Multilevel dynamics Armies, governments, internet
each part has a clear, identifiable function which makes prediction possible,
such systems are said to be ‘complicated systems’ (San Miguel et al. 2012).
For example, an airplane is a ‘complicated system’ in this sense. Complex sys-
tems, by contrast, are characterised by the emergence of collective behaviour
which cannot be reduced to the individual behaviour of each part. Instead, the
properties of its behaviour emerge from the interaction between the parts, as in
the case of the brain. The many constituent parts may be heterogeneous, as in
the case of cities or climates, or homogeneous, as with water molecules form-
ing waves in the ocean. These parts are also sensitive to network connectivity,
which may guarantee the survival or cause the death of an ecosystem. They
may be the result of self-organisation, as in the case of revolutions or fashion
shifts, and they may have co-evolving systems, as with computer virus soft-
ware, which evolves as anti-virus software is developed. Further characteristics
of complex systems are included in Table 8.1 (adapted from San Miguel et al.
2012).
Several steps are considered necessary for a thorough analysis of a complex
system (from San Miguel et al. 2012):
i) Exploration, observation and basic data acquisition
ii) Identification of mechanisms
iii) Modelling
iv) Model validation, implementation and prediction
v) Construction of a theory
Needless to say, not all studies of complex systems achieve the final level here,
and this will not be our intention in this chapter either. For the purposes of this
chapter, we will focus on the first three levels: observation of the problem (lan-
guage choice in multilingual societies), identification of mechanisms (based on
observation and on previous studies), and mathematical modeling (which will

be qualitatively compared to several well-known linguistic situations), as will
be seen in section 3.2
2.2 Social Complexity

Human societies are a prototypical complex system. Made of a high number
of individuals whose interactions determine the evolution of the system (war,
trade, arts, linguistic change, etc.), society is claimed to share many features
with other emergent phenomena occurring in other systems composed of a large
number of interacting individuals. In fact, very early in the history of sociology
scholars tried to devise a mechanistic view of society, from Thomas Hobbes to
David Hume, John Locke, Jeremy Bentham, John Stuart Mill and Karl Marx. To
different degrees, they all tried to apply statistical laws of the natural sciences
to societies (Ball 2004). Although their approaches remain largely part of the
history of the field, the truth is that society conforms broadly to all the features
outlined in Table 8.1. For example, it is made of many heterogeneous parts, it is
network-based, it exhibits ill-defined boundaries, it includes co-evolving sub-
systems, and it is characterised by non-equilibrium dynamics (which explains
why societies never stop developing new features).
In fact, many current social changes can only be properly described in terms
of Complexity Science, such as the Arab Spring revolutions, the 15M move-
ment in Spain (Borge-Holthoefer et al. 2011),3 the Occupy Movement in the
USA, the world financial crisis, international terrorism, and cyber-crime. All
of these can be considered emergent phenomena whose very explanation is a
matter of the interactions between individuals rather than the particular psy-
chology of each human being involved. Nevertheless, in addition to the pro-
totypical features of complex systems, society is also characterised by a high
number of variables which differentiate humans from, say, gas particles, includ-
ing ‘memory, anticipation, emotion, creativity and intention’ (Ball 2012: XI).
Thus, when modelling society, it is also necessary to consider a variety of tech-
nical, financial, ethical and cultural dimensions.
This chapter, then, is a contribution to the development of this new approach
to social phenomena, given that the increasingly connected world in which we
live will undoubtedly have consequences for the survival and disappearance of
languages.
2 Other scholars have indeed dealt with empirical data as a means of validating their models (step
(iv) in the previous list), such as the pioneering work of Abrams and Strogatz (2003), who do so
with data on Scottish Gaelic and Quechua in Peru, and Kandler (2010), who studies the survival
of Celtic languages.
3 See also the 15M Movement video available online at BIFI (Instituto Universitario de Investi-
gación Biocomputación y Física de Sistemas Complejos): http://15m.bifi.es/index_en.php.
2.3 Modelling Society: Bases and Challenges

Mathematical models used within the framework of Complexity Science are
usually considered simple, because they involve few components and thus
avoid dealing with the myriad of variables that general sociological or human-
istic studies typically grapple with. It is not uncommon that these ‘simple’
models end up being used in fields other than the one for which they were
designed. One well-known example of this is the Ising model, developed in
1925 to model magnetism, but has gone on to be used in fields as diverse as
DNA studies and the study of political elections (San Miguel 2012). Accord-
ing to Schelling (1978, chapter 3), simplicity should therefore not be seen as
a drawback but as an advantage, in that the descriptions of different physi-
cal, social or animal phenomena may shed light on the notion, traditionally
concealed or overlooked, that different phenomena may behave according to
recurrent patterns. In this sense, one might ask the question whether, and to
what extent, a model might predict the outcome of such disparate phenomena?
The answer is that our interest in modelling, rather than being aimed exclusively
at a quantitative prediction, can be more about understanding the behaviour of
the system, describing mechanisms involved, and developing concepts which
may lead to forecasting probable scenarios.4 Our approach, thus, is very much
in Nettle’s (1999a: 103) line, who claims that ‘A simulation, however rudi-
mentary, is thus an improvement over a merely verbal argument, in deciding
what general conditions must obtain for languages to evolve in the way that
they do.’
The basic elements upon which the modelling of complex collective social
phenomena is based are the following:
a) Characterisation of the agents. As mentioned earlier, the agents are
autonomous entities that can make decisions and which are the mathemati-
cal equivalent of social individuals. Their characterisation includes the dif-
ferent states which describe the agents and, therefore, they may exhibit
some degree of heterogeneity, as in the case of Nettle’s (1999a, 1999b)
agents, which differ according to age and degree of influence on society,
and the case of Ke et al.’s (2008) paper on language change in which agents
differ in two aspects, namely age and type of learner. Regarding language
choice, the focus of this chapter, agents are nevertheless homogeneous and
will be attributed one of three possible values: they may be speakers of
language A, of language B, or of both A and B.
b) Interaction rules. These rules determine the way in which agents change
their state through interaction among themselves. Different models include
4 In current complex systems research an important distinction is made between prediction and
forecasting. Prediction is considered to be quantitatively precise, while forecasting describes
possible future general scenarios.
different interaction rules. For example, the Voter Model is based on the
agent’s imitation of another randomly chosen agent, while in Granovet-
ter’s (1978) Threshold Model an agent will change its status only when the
majority of the agents which interact with it share that different status. On
the other hand, Axelrod’s Model of cultural dynamics (Axelrod 1997) takes
into account homophily as a factor which determines the degree to which
an agent may influence another (and make it change a given feature).
c) Patterns of activity. Normally, computational approaches and simulation
models update the state of the agent at a constant rate. The underlying
assumption is that there is a well-defined mean inter-event time in human
interactions. However, human activity patterns are not characterised by a
constant rate of interactions: Beyond circadian rhythms, there is now ample
evidence in data of communication technologies that interactions between
two people occur by bursts of activity (Wu et al. 2010), so that the distribu-
tion of inter-event times is heavy tailed with no characteristic time. The dis-
tribution of time intervals between two interactions becomes an important
element in determining collective phenomena (Barabási 2005). In terms
of modelling, the updating rule must be considered as part of the model
itself and not as a technical simulation detail (Fernández-Gracia et al.
2011).
d) Complex networks. Determining which agents interact with which others
will be the key to configuring the network. In a fully connected society,
each agent interacts with all other agents. However, because this rarely hap-
pens in real situations, further types of networks have been modelled, such
as small world networks, random networks, or networks with community
structure. All of these networks are described in section 2.4.
There are many examples of simple models that have contributed to the under-
standing of the mechanisms underlying social phenomena. For reasons of
space, we will focus on some of the studies arising from Axelrod’s Model of
cultural diffusion (Axelrod 1997). This model was born with the intention
of providing an answer to the following question: If people tend to share more
and more cultural features as they interact, why is it that cultural diversity per-
sists? The model, which defines culture as a series of individual features that
may change owing to social interactions, is based on two principles. The first
is the homophily principle, according to which interaction is more likely to
take place between similar agents (i.e. agents which share several cultural fea-
tures). The second principle is the principle of social influence, according
to which social interaction favours cultural similarity. The model illustrates how
an interaction mechanism of local convergence can generate global polarisation
(i.e. persistence of cultural diversity). It also explains the transition from cul-
tural globalisation to cultural polarisation, depending on the number of cultural
traits allowed for each cultural feature. In addition, culturally polarised states
are shown to be unstable against cultural drift in a fixed social network, but
cultural diversity can be stabilised by a dynamic co-evolution of the state of
the agents and the network of interaction; the social network evolves in tandem
with the collective action that it makes possible, that is, circumstance makes
men as much as men make circumstance (Centola et al. 2007).
This model has also been used to describe the effects of propaganda and
mass media effects (e.g. González-Avella 2007, 2010). These approaches to the
model describe two interesting mechanisms: (i) rather than leading to cultural
homogeneisation, excessively direct propaganda causes social polarisation, as
interactions among agents sharing few features tend not to take place; and (ii)
a given social group may converge in a state of cultural globalisation different
from the message of the propaganda.
2.4 Complex Networks and Social Networks

The use of networks in the study of social behaviour has been reported for the
last eight decades (Borgatti et al. 2009). Thus, the psychiatrist Moreno (1934)
explained the epidemic of runaways at the Hudson School for Girls in upstate
New York by using sociometry, a technique that allowed him to graphically
represent individuals’ feelings for each other by means of networks. Another
example is Travers and Milgram’s (1969) often-quoted study that illustrates the
well-known small world phenomenon. By randomly selecting some 300 indi-
viduals in the United States, all of whom were asked to direct letters to a certain
target person, the authors showed that the average distance between two ran-
domly chosen individuals was about six (the well-known ‘six degrees of separa-
tion’). The experiment, which was criticised on the grounds that relatively few
letters actually reached their intended recipients, was later successfully repli-
cated in two studies by Leskovec and Horvitz (2007, 2008), who used hundreds
of thousands of individuals and replaced letters with online instant messages.
An extensive network-based study is that by Onnela et al. (2007), who analysed
a mobile phone network with more than four million users and substantiated
the theory of weak ties formulated by Granovetter (1973).
With these examples we can see how, although networks have been stud-
ied both from mathematical and social perspectives for several decades, ‘Net-
work Science’ can only be said to have emerged as a discipline in the 1990s
(see, as general references Barabási 2002 and Watts 2003), when, owing to the
advances in statistical physics, large networks came to be studied for differ-
ent purposes. Thus, network science has allowed for the comparison of natural
ecosystems, social networks, electric systems, the internet, and so on, identify-
ing universal common properties of all these. This universality and the ability to
compare different systems by means of mathematically created models allows
for current network science to now be considered ‘one of the most vibrant,
diverse and insightful areas of complex-systems research, which has cast a

new light on many areas of the social, natural and engineering sciences’ (Ball
2012: 23).
Complex networks (or graphs, a term preferred in mathematics) can be said
to be the backbone of complex systems (see, as overviews of this topic, Albert
and Barabási 2002; Dorogovtsev and Mendes 2002; Bocaletti et al. 2006; Doro-
gotsev et al. 2008). In such studies of society, networks have come to be mod-
elled by nodes (representing individuals) linked by representations of social
ties, very much in the line with studies carried out in the humanities following
Milroy’s (1980) pioneering work on interaction networks (see, for example, the
special issue of the International Journal of the Sociology of Language on the
subject matter, edited by de Bot and Stoessel 2002, and Ke et al. 2008). Despite
this common basis, different types of networks have been proposed from net-
work science, some of which are described in what follows.
Firstly, in random networks (Erdös and Rényi 1959) every node is ran-
domly connected to another. This obviously does not exactly match reality, in
which individuals choose their links on the basis of their common interests, geo-
graphical situation, and so on. Nonetheless, random networks yield interesting
results, drawing them closer to real networks, such as the fact that the distance
between connected nodes is small and grows slowly with the system’s size.
Secondly, small world networks (Watts and Strogatz 1998) came to be mod-
elled in such a way that they reflected the short average path length between two
randomly chosen individuals (which had been empirically observed in Travers
and Milgram 1969). In addition, small world networks also introduce a large
clustering coefficient, something which is also observed in real networks, as
seen in human society where new acquaintances often already share a friend or
a friend of a friend. This has been seen empirically on many occasions, such as
in the collaboration networks of movie actors, as in the so-called Kevin Bacon
Game at www.oracleofbacon.org, which shows different ways in which movie
actors are connected.
Thirdly, when all the nodes of a network are connected all to all (i.e. when
all the nodes of a networks have only one degree of separation between them),
the network is said to be fully connected. We will also highlight in the present
chapter networks with community structure, that is, those in which some
groups of nodes are much more connected with one another than with other
nodes in the network, leading to a modular structure which very much resem-
bles the type of topology we find in real networks. Several modellers from sta-
tistical physics have recently tried to find a satisfying algorithm for the mod-
elling of networks with community structure (e.g. Girvan and Newman 2002),
but the network used in the current chapter is based on the algorithm proposed
by Toivonen et al. (2006). These four types of complex networks have been
analysed in detail so as to assess their role in language choice in a multilingual
community, and results are summarised in section 3.4.
2.5 Complexity Science and Linguistic Studies

The study of language dynamics from the point of view of Complexity Science
is a growing academic field, which tries to answer questions related to language
evolution, language and cognition, and language competition (or the dynam-
ics of language use in multilingual communities), as shown by Castelló et al.
(2011). This approach to language sets itself apart from Saussurean’s struc-
tural linguistics and from the Chomskyan tradition of generative grammar in
that Complexity Science considers language to be a complex adaptive system
(Ellis and Larsen-Freeman 2009). As such, language, with its major social func-
tion, is a system consisting of multiple agents interacting with one another (The
“Five Graces” Group 2009).
In this approach, Complexity Science has a lot to contribute, because inter-
actions among individuals give rise to global emergent properties, such as con-
sensus or fragmentation (Loreto and Steels 2007). Complexity Science tries
to identify the mechanisms that favour consensus or fragmentation among
the members of a social group, how opinion is formed in societies, how a
shared language emerges (Naming Game, for example, in Steels 1995), and
how network structure affects interactions among individuals in addition to
all the previous phenomena, as well as others. The applications are so var-
ied and promising that scholars from different disciplines have been attracted
by the fresh air of Complexity Science, as seen, for example, in the special
issue of the journal Advances in Complex Systems, edited by Baronchelli et al.
(2012), where language dynamics are studied from the perspective of Com-
plexity Science by linguists, cognitive scientists, psychologists, philosophers,
physicists, mathematicians, computer scientists and Artificial Intelligence (AI)
researchers.
A main tool used in Complexity Science is computer simulations, one of
whose advocates in linguistic studies is Nettle (1999a, 1999b). He makes clear
that computer simulations are not aimed at formulating predictions, but at
exploring the general effects of the main relationships of a real situation across a
range of conditions (Nettle 1999a: 103). His work had quite an impact on stud-
ies of language change and of the spread of linguistic innovations; it inspired,
for example, the work by Ke et al. (2008), who address his notion of threshold
problem. This notion intends to account for an important difference between
biological and cultural evolution: genetic mutations and cultural changes do not
spread in the same way. Accordingly, Ke et al. (2008) propose a model for the
spread of language change in different networks (regular, small-world, random
and scale-free networks). A precedent of such ideas of threshold and the spread
of innovations in the sociological literature is the classical paper by Granovetter
(1978).
In addition to Nettle (1999a, 1999b) and Ke et al.’s (2008), other rele-
vant computational studies of language dynamics include Schulze and Stauffer
(2006) and Schulze, Stauffer and Wichmann (2008). We follow this trend of
computational studies. However, while Nettle (1999a, 1999b) and Ke et al.
(2008) study language change, we focus on social competition between two lan-
guages. We do this from a cross-disciplinary perspective, because, like Blythe
and Croft (2010: 19), we also believe that ‘there are benefits to collabora-
tion between empirical researchers in the humanities, who generally have not
been trained in mathematical analysis of quantitative data, and modelers from
physics or other strongly mathematical fields, whose models are not always
based on empirical data from the domains being modeled’. Our findings are
qualitatively compared with real-life situations in section 3.
3 Models of Language Choice in Multilingual Societies5

Language contact is a phenomenon which has long been of interest to linguists
from different perspectives. Some focus on language contact and its conse-
quences on a language’s structure or on linguistic ecology (e.g. Weinreich 1953;
Hamers and Blanc 1989; Romaine 1989; Grenoble and Whaley 1998, 2006;
Nettle and Romaine 2000; Mufwene 2001, 2008), while others look at the way
in which speakers might favour one available language over another, leading
to the potential endangerment of the other language or languages (e.g. Fish-
man 1991, 2001; Grenoble and Whaley 1998, 2006; Crystal 2000; Nettle and
Romaine 2000; Hinton and Hale 2001; Bradley and Bradley 2002; UNESCO
2003; Wölck 2004; Tsunoda 2005; Mufwene 2003, 2004, 2008; Ehala and
Niglas 2007; Ehala 2010). Indeed, this is our own area of interest. We build
on two models of language change (Abrams-Strogatz 2003, and the Bilinguals
Model by Wang and Minett 2005 and Minett and Wang 2008) and study the
different phenomena that may emerge from projecting these models on differ-
ent complex networks, as well as by changing the value of the parameters. In a
way, our study could be compared to that of Ke et al. (2008), although our focus
is language competition, rather than language change. The following sections
summarise the models we rely on and our results.
3.1 Models
As noted in section 2.1, complex systems need not be complicated, but the truth
is that in real life they often are. That is the case with a multilingual society,
and for this reason sociolinguistic studies from Labov (1972) onwards have
paid attention to as many variables as possible (density of the population, age,
5 This section summarises the main results obtained from a six-year research project which has
been partially published elsewhere (Castelló et al. 2006, 2007a, 2007b, 2008, Toivonen et al.
2009, Vazquez et al. 2010 and Castelló et al. 2013). We hereby acknowledge the work of all the
authors of the previous publications.
gender, social class, etc.). On the contrary, the virtue of modelling, in general,
is that models are designed to be as simple as possible, because, in San Miguel
et al.’s (2012) words:
[t]he aim of simple models is to get [a] better understanding of the so-called ‘stylized
facts’ of the system, i.e. to make simplified, abstracted, or typical observations – in other
word[s, to] capture some ‘essence’ of the real system (San Miguel et al. 2012: 7).6
Then, with the aim of capturing the ‘essence’ of language choice in a mul-
tilingual society, mathematical models of language competition designed in
Complexity Science include few variables. That is the case with both Abrams-
Strogatz Model (2003) and the Bilinguals Model (Wang and Minett 2005,
Minett and Wang 2008), on which we base our work here.
In the model proposed by Abrams and Strogatz (2003), agents may be speak-
ers of either of the two languages available: A or B. The main factors that con-
dition the probability (P) that one agent changes its choice of language from B
to A (PB→A )7 are the prestige (s) of the languages and the relative number of
neighbouring speakers (σ) using each language in the community (σA and σB ),
as seen in the following equations:
pB−A = s · (σA )a
pA−B = (1 − s) · (σB )a
(1) Abrams-Strogatz (2003) Model of Language Competition

As (1) shows, in addition to s (prestige) and σ (density of the population), there
is a third variable that conditions the probability that a speaker changes their
status, namely a, which we label ‘volatility’ and which represents the speaker’s
willingness to shift language. With these simple equations Abrams and Strogatz
(2003: 900) manage to fit their model to the data found for Scottish Gaelic in
Sutherland and Quechua in Huanuco.
6 The dilemma regarding the ideal number of variables and parameters to be included in a model
parallels that of the ideal size of a map, as found in Lewis Carroll’s Silvie and Bruno Concluded,
illustrated with this passage: ‘What do you consider the largest map that would be really useful?’
‘About six inches to the mile.’ ‘Only six inches!’ exclaimed Mein Herr. ‘We very soon got six
yards to the mile. Then we tried a hundred yards to the mile. And then came the grandest idea of
all! We actually made a map of the country, on the scale of a mile to the mile!’ ‘Have you used
it much?’ I enquired. ‘It has never been spread out, yet,’ said Mein Herr: ‘The farmers objected:
they said it would cover the whole country, and shut out the sunlight! So now we use the country
itself, as its own map, and I assure you it does nearly as well.’
7 Note that in these models speakers may switch from A to B and vice versa (both A > B and B >
A are possible). One important difference regarding our study and Ke et al.’s (2008) pioneering
model of language change is that in their model the change is unidirectional (only U > C is
possible, namely Unchanged > Changed or adoption of an innovation in a process of innovation
spreading).
The simplicity of the Abrams-Strogatz Model, however, results in one impor-

tant aspect being missed: the fact that speakers may be users of both the
languages available in the community, which proves to be a key element of
language modelling. In Wang and Minett (2005) (later published with some
mathematical modifications as Minett and Wang 2008), bilinguals are given an
important role since they represent the bridge between two monolingual stages
as a necessary step for a monolingual speaker of language A to become a mono-
lingual speaker of language B, or vice versa. The mathematical formulation of
the Bilinguals Model departs from (1), including transitions from using lan-
guage A or B to become bilingual (AB), so that there is no direct change from
one monolingual state to the other, as seen in (2):
pA→AB = (1 − s) · (σB )a
pB→AB = s · (σA )a
pAB→A = s · (1 − σB )a
pAB→B = (1 − s) · (1 − σA )a
(2) Bilinguals Model (Minett and Wang 2005)

Whenever there is more than one model designed to account for the same
phenomenon, the best model is identified by testing the results obtained from
model-based simulations against real data: the higher the degree to which model
results and data match, the higher the validity of the model. This, for exam-
ple, is the procedure followed in the natural sciences, but things become more
complex (or complicated) when modelling social phenomena, since, as Ball
(2012: xi) observes:
It is extremely difficult – practically, financially and ethically – to conduct experiments
on human social systems. And even when this is possible, the number of parameters of
variables describing the system is commonly very large, and some of them may be hard
or impossible to quantify. [ …] In such cases, we need not despair of the value of models
or theories. Rather, it might be necessary to accept that a particular phenomenon has no
unique explanation, no ‘best’ model that accounts for it (Ball 2012: xi).
Therefore, in this chapter we will not strive to arrive at firm conclusions regard-
ing which of the two models is better: Abrams-Strogatz’s or Minett and Wang’s.
We will simply compare them with the aim of exploring the different param-
eters they include, namely monolingualism versus bilingualism (section 3.2)
and the roles of prestige (s) and volatility (a) (section 3.3), as well as a very
important element in Complexity Science, namely the role played by agent
interaction in different topological networks (section 3.4).
3.2 Effects of Bilingualism (Versus Monolingualism)

The effects of monolingualism versus bilingualism have been the subject of dif-
ferent studies which we have published elsewhere (Castelló et al. 2007a, 2013;
Figure 8.1 One-dimensional regular lattice of degree k = 2.
Castelló 2010) by comparing the Abrams-Strogatz Model and the Bilinguals

Model in two kinds of networks: regular networks and small-world networks.
A regular network is one in which each agent has the same number of imme-
diate neighbours (degree, k) and is connected only to them (see Figure 8.1 and
Figure 8.2).
In a simulation tool available online (http://ifisc.uib.es/research/complex/
APPLET_LANGDYN.html) we have used the network represented in Figure
8.2 to compare both models, as can be seen in Figure 8.3.
Lattice Network
Figure 8.2 Square regular lattice of degree k = 4.

Figure 8.3 Simulation tool designed to visualise the effects of bilingualism,

prestige and volatility.
The tool shows that the default value for s (prestige) is 0.5, which means
that both languages are socially equivalent. Likewise, the default value for a
(which we have termed volatility) is 1. The population density of each option
(monolinguals A and B in Abrams-Strogatz Model, and monolinguals A and B
and bilinguals in the Bilinguals Model) is proportionally given (notice that the
sum of the density of speakers is 1 and that bilingual agents, only found in the
right-hand square, are white dots).
With the aim of assessing the role of bilingualism, the values of s and a have
not been altered for either the regular lattice (simulation tool) or small-world
network (see Castelló et al. 2007a, 2013, section 3.2). The qualitative results
we have obtained have the following implications:
(a) The presence of bilinguals (right-hand part of the tool) gives rise to well-
defined borders between the A and B domains, with the bilinguals form-
ing these boundaries. These well-defined domains grow in time much faster
than when there are no bilingual agents (Abrams-Strogatz Model, left-
hand part of the tool). On the other hand, except for a minority of real-
isations leading to dynamical traps, the presence of bilinguals acceler-
ates language death, that is, in the Bilinguals Model one of the languages
disappears earlier than in the Abrams-Strogatz Model (Castelló et al.
2006).
Figure 8.4 Bilingual speakers on the borders of monolingual domains (regular

lattice).
(b) Once the interaction among individuals starts (‘start’ button on Figure 8.3),
the number of bilinguals, which was one third in the initial conditions,
decreases dramatically to ca. 3 percent.
(c) The reduced numbers of bilingual speakers are located on the borders of
the monolingual domains, as seen in Figure 8.4. This can be interpreted
as bilinguals choosing language A or language B depending on the inter-
locutor. They could be considered bridging agents. In fact, a recent study
of a Belgian mobile phone network (Blondel et al. 2008) shows that this is
indeed the location of bilinguals, as shown in Figure 8.5.
(d) Finally, the dynamics lead inevitably to language death, both in the case
of Abrams-Strogatz and the Bilinguals Model. It is, however, in this third
finding that we observe the most pronounced differences between the syn-
ergistic effects of bilingualism and the type of network. Although language
death is the final scenario in both regular lattices and small-world networks,
the latter accelerate the process (see Castelló et al. 2007a, and also section
3.4). Qualitatively this is a highly relevant result in an increasingly glob-
alised world, where distance among speakers is much less than one would
expect. Perhaps the rapid erosion of grammatical, lexical and phonologi-
cal features that gives way to the emergence of linguistic varieties such as
world standard spoken English (Crystal 1997) can be explained in terms of
a small-world effect, just like the disappearance of language A or B in the
simulations here.
Figure 8.5 Islands of monolingual speakers linked by a small number of

bilinguals. [Figure from Blondel et al. 2008]
3.3 Prestige and Volatility: Different Values and Effects

The simulation tool described earlier (see Figure 8.3) allows the user to alter
the values of the parameters s (prestige) and a (volatility) in real time. Let us
first consider the effects of prestige.
As described in detail in Castelló et al. (2013; section 3.3), the values of
s have two basic implications: (a) languages may be socially equivalent (s =
0.5) and (b) one of the languages is socially more prestigious than the other
(bearing any other value). The results obtained in the simulation tool confirm
this: a situation in which the two languages exhibit different degrees of prestige
accelerates the disappearance of the less prestigious language. There are hun-
dreds of examples of the effects of prestige on language communities around
the world (e.g. the threat to Native American languages in America). Like-
wise, there are many real examples of languages which have increased their
probability of survival once the prestige associated with them has been actively
enhanced, as is the case of Quechua in Cuzco (see Manley 2008). Nevertheless,
although prestige is traditionally considered a powerful factor in the survival or
death of a language, dissenting voices have also been heard against the overes-
timation of prestige (e.g. Mufwene 2003). For this reason, we have modelled
another important factor in the survival or death of a language, namely speak-
ers’ willingness to shift languages, parameter a in the model.
As far as parameter a is concerned, although it is not often used in linguistics,
it has its precedent in social impact theory (Nowak et al. 1990) and it is also
a main ingredient of the work on language change by Nettle (1999a, 1999b).
In Nowak et al.’s (1990) theory, the impact of a group of agents on a given
individual is proportional to a power a of their number. The departure from
a linear relation (a = 1) is argued to allow for the fact that persuasiveness is
not just proportional to the size of the group. While for matters of public opin-
ion empirical studies show that a is ca. 0.5 (Nowak et al. 1990), Nettle (1999a,
1999b) fixed a = 0.8 in his simulations. However, Abrams and Strogatz (2003),
with no reference to social impact theory, fitted a = 1.3 to reproduce data on
language competition between Spanish and Quechua, English and Welsh, and
English and Scottish Gaelic. We will use a as a free parameter to study different
scenarios of language competition and finding different qualitative behaviour
for a < 1 and a > 1. Thus, in our simulation tool (which, it must be remem-
bered, is based on a regular lattice), parameter a (or volatility) has a default
value 1 which can be changed to any value between 0.1 and 5. The lower the
value of a, the higher the degree of volatility, and vice versa. This means that
the most volatile system is that in which a = 0.1, and the least volatile one,
that in which a = 5. The simulation tool shows how the model responds to the
expected result: the effects of a low prestige will be minimised by a system
of low volatility.8 In other words, if speakers are unwilling to shift languages
(for instance, because it functions as a strong ethnic identity marker), they
are less influenced by prestige. This explains, for example, the survival of
minority languages for centuries, even if they are apparently despised by
socially more prestigious languages (H-languages). In addition, the compari-
son between the Abrams-Strogatz Model and the Bilinguals Model shows how
8 Even if one might be tempted to think that prestige is one of the factors underlying volatility (i.e.
that speakers will be more willing to shift to the most prestigious language), it must be noted that
in the model the two parameters work independently.
Figure 8.6 Abrams-Strogatz Model vs. Bilinguals Model, when a = 0.1 (high
volatility).
the presence of bilinguals constitutes a factor slowing down the death of the
less prestigious language. Elsewhere Castelló et al. (2013: section 3.3) estab-
lished a qualitative comparison between this computer simulation and a real
social situation, namely that exhibited by the ‘competition’ between Spanish
and Galician in NW Spain under dictator Franco’s regime (1936–1975). In this
period, the overt prestige of Spanish was clearly higher than that of Galician,
the L-language, but the latter was not threatened by the former, because of an
attested lack of volatility of its speakers, both in linguistic and social terms. In
a nearly feudal society in which class-shifting was virtually impossible, lan-
guage shift was pointless, as it would isolate speakers from their home group,
while integration in the community speaking the other language was prevented
by the social rules. Speakers of Galician were bilingual; they used Spanish for
specific purposes (Ayestaran and de la Cueva 1974), but the diglossic structure
in which languages coexisted (without competition) appears to have delayed
the possible threat from Spanish, the more prestigious language.
The last interesting effect of volatility can be seen if a is lowered to 0.1, its
lowest possible value in the simulation tool, which represents a highly volatile
system, that is, one in which all speakers are willing to shift languages (as might
be the case, for example, at a party with Erasmus students anywhere in Europe).
The results are worth noting, because, counterintuitively, this does not lead to
language death or to the imposition of one language over the other. Rather,
the situation obtained is quite the opposite: no language ever dies. Irrespective
of the number of interactions which are allowed (note that in Figure 8.6 more
than 8,000 interactions have taken place), the density of speakers of both lan-
guages available remains without significant alteration. Even more interesting
is the result obtained for the Bilinguals Model. Contrary to the situation pre-
viously described (in which, once interactions take place, the number of bilin-
guals becomes very low because they are restricted to the borderline areas),
bilinguals persist in representing nearly a third of the agents in the lattice and
are placed throughout the network (with no clear domain of monolingual speak-
ers emerging), as can be seen in Figure 8.6.
High volatility, as we have argued elsewhere (Castelló et al. 2013; section
3.3), may be responsible for the emergence of contact varieties in which speak-
ers master both languages available and switch often and comfortably. Evi-
dence for this claim may be found in code-switching varieties such as Span-
glish in the United States and Yanito (or Llanito) in Gibraltar, a British colony
on the south of the Iberian Peninsula, which has for 300 years experienced
constant contact between Spanish and English. In the words of a Gibraltarian
interviewed by Levey (2008): ‘I’ve got about two or three different groups of
friends and some of them speak just English and some of them speak a mixture
but I think that does tend to happen a lot [ …] I have no problems with Spanish
or English. I just adapt’ (2008: 80). This natural ability to adapt is what we call
volatility.
From a general complexity perspective the change from low to high volatil-
ity at a = 1 is a transition point. For high volatility (a < 1) the system does
not evolve to an ordered state and there is dynamical coexistence of two lan-
guages, while for low volatility (a > 1) the system orders with growing spatial
domains of the two languages until one of them takes over in what is known
as spontaneous symmetry breaking (for s = 0.5). The parameter a is the ana-
logue of temperature in thermodynamic phase transitions, with a = 1 being the
analogue of the critical point. A main difference in that analogy is that here the
individual changes of state (language shift by an agent) only occur at the spatial
borders between domains, not in the bulk of the domains, with these borders
being only well defined for a > 1.
3.4 Effects of the Topological Structure: Different Complex Networks

Having now qualitatively addressed the roles of bilingualism, prestige and
volatility in regular lattices, in this section we compare the results obtained in
the simulation tool with those from mathematical approximations to the types
of networks described in section 2.4, namely, fully connected networks, ran-
dom networks, small-world networks and networks with community structure.
Our aim is to assess whether the topological structure of the network has any
effect on the dynamics and, therefore, on the survival or disappearance of the
languages in competition. Various previous studies have included a quantitative
approach to the differences among the networks (Castelló et al. 2006, 2007b,
2011; Vazquez et al. 2010). Here, for the sake of clarity and brevity, we will
start by comparing fully connected networks and random networks on the one
hand, and small-world networks and networks with community structure on the
other.
Fully connected networks, it must be recalled, are those in which each agent
is connected to all other agents in the network. Random networks, in turn, are
those in which each agent is randomly connected to more than one other agent,
with some agents being connected to more agents than some others, which have
only a handful of neighbours. In fully connected networks we have observed
an abrupt behaviour in both the Abrams-Strogatz Model and the Bilinguals
Model. That is, in both models the situation shifts from language maintenance
to language death discretely, rather than smoothly for a threshold value of a. In
addition, in neither model does parameter s (prestige) have any effect on lan-
guage maintenance, as the only factor that conditions the transition from lan-
guage coexistence to language extinction is parameter a (volatility). However,
an important difference is found between both models: for the two languages
to survive in the Bilinguals Model, agents need to be more volatile than in the
Abrams-Strogatz alternative. That is, in the presence of bilinguals language
coexistence is possible only if speakers are highly volatile, if they are willing
to shift languages.9
If we compare these results from fully connected networks with those
obtained for random networks, we observe both a qualitative and a quantitative
difference between them, as in the latter prestige does play a role in language
maintenance or death. While in fully connected networks the value of param-
eter s is irrelevant, in random networks both a and s condition the survival of
one or the two languages. Obviously, the most prestigious language has more
chances to survive, but, one may wonder, which difference between these two
types of networks puts prestige in such a relevant position? The answer lies in
the effects of local interactions. While in fully connected networks there are
no local interactions at all (since agents are connected all to all); in random
networks each agent has a finite, small number of neighbours. The number of
neighbours of each agent is called k (degree), and random networks tend to
become fully connected networks as the mean degree increases. Thus, if the
mean degree of a random network is low, local interactions take place and the
effect of prestige is higher. At the same time, as the mean degree increases (i.e.
as the random network becomes more and more similar to a fully connected
one), interactions become global and the effect of prestige is diminished.
To sum up the comparison between fully connected and random networks,
we could say that the lower the mean degree of a random network, the higher
the number of local interactions, and, therefore, the stronger the role of prestige.
The role of local interactions, then, must be considered as a relevant factor in
the modelling of language choice, as it reduces the chances of language coex-
istence.
9 For the Abrams-Strogatz Model the value of a in which this transition takes place is 1, while for
the Bilinguals Model it is 0.63.
Figure 8.7 Snapshots of the Abrams-Strogatz Model (top) and the Bilinguals
Model (bottom) in a small-world network (bilinguals = white). (Figure from
Castelló et al. 2013)
Having assessed the role of local interactions, let us now focus on two types
of networks which do allow for local interactions, namely small-world net-
works and networks with community structure. Figure 8.7 illustrates the com-
parison between the Abrams-Strogatz Model and the Bilinguals Model in a
small-world network (for a = 1, s = 0.5) and reveals another important piece of
information regarding modelling language maintenance: the presence of bilin-
guals accelerates language death in this type of network, as, after the same
number of interactions the minority language is about to disappear in the Bilin-
guals Model, while it is not still in danger of extinction in the Abrams-Strogatz
Model. One could think, then, that bilinguals play a negative role in language
maintenance, but the comparison between small-world networks and networks
with community structure prevents us from jumping to conclusions.
Figure 8.8 compares the Abrams-Strogatz Model and the Bilinguals Model in
a network with community structure and shows how the presence of bilinguals
prevents language death. After 50 interactions (t = 50) in the Abrams-Strogatz
Model, the number of black agents is much smaller than after 100 interactions
(t = 100) in the Bilinguals Model. This difference between both models in this
type of network goes on for different time values, until the minority language
Figure 8.8 Abrams-Strogatz Model (left) and Bilinguals Model (right) in a

network with community structure (t = time). (Figure from Castelló et al.
2007b)
dies out in the Abrams-Strogatz Model, while in the Bilinguals Model the path
to extinction may be simply blocked by a bilingual agent being the single con-
nection between an isolated community and the remainder of the network (bot-
tom, right-hand side of Figure 8.8). After 1,000 interactions among the agents
(t = 1000) an isolated community persists nonetheless, owing to the communi-
cation held with the other communities by means of the (white) bilingual agent,
which may be considered an agent with a privileged situation in the network
and which plays the role of a social bridge. The situation in Figure 8.8 is not
counterintuitive at all, and there are plenty of examples of this sort in the real
world of language contact, such as the often-cited case of the Pennsylvania
Dutch, an old variety of German which has survived for centuries, owing to its
use in a community that is not just geographically but also socially isolated.
A final finding regarding the role played by the topological structure of the
network concerns the different times needed for extinction in the presence and
in the absence of bilinguals. We have quantitatively measured this time for
regular lattices, small-world networks and networks with community structure
when a = 1 and s = 0.5 (socially equivalent languages). Our findings show that
the extinction speed of one of the languages in the Abrams-Strogatz Model is
similar in the three types of networks, while important differences are observed
for the Bilinguals Model. In the latter, the extinction of one of the languages
is faster in small-world networks, slower in regular lattices and even slower in
networks with community structure (small-world networks > regular lattices
> networks with community structure).
Summing up this section, it can be said that, without question, the kind of net-
work in which interactions take place is a strong influencing factor on language
dynamics, as it plays a central role in the potential survival or disappearance of
one of the languages in competition.
4 Summary and Conclusions

This chapter represents an approach to the study of language competition and
selection in a multilingual population (Mufwene 2001, 2008) from the point
of view of Complexity Science. Such a framework departs from traditional
sociolinguistic methods that consider many individual features of the poten-
tial speakers of a language, focusing instead on interactions among individuals
and the different possible outcomes of this. With the aim of guiding the reader
through the basic aims and methods of Complexity Science, section 2 intro-
duced the topic (2.1) and discussed the various ways in which this interdis-
ciplinary framework has proved useful to understanding different social phe-
nomena (2.2 and 2.3). It also concentrated on the crucial concept of network
(section 2.4) and described the different complex networks used in various
proposals for modelling society. Finally, it provided a brief summary of the
different approaches to language which have been made from the point of view
of Complexity Science (section 2.5).
Section 3 included the presentation of two models of language choice
Abrams-Strogatz (2003) Model and the Bilinguals Model by Wang and Minett
(2005; Minett and Wang 2008). As stated in section 3.1, our aim was not to
determine which of the two models is better. We find both of them useful as
points of departure for our study of the different factors that appear to play the
most central role in the survival or disappearance of a language: (a) bilingual
speakers, (b) prestige of the languages and volatility of the speakers, and (c)
topological structure of the network. The following are our main findings:
(1) Bilinguals have been found to play a central role in language competition;
the results for the Abrams-Strogatz and the Bilinguals Models show sig-
nificant differences. Firstly, the presence of bilingual agents reduces the
possibility of language coexistence in fully connected networks, because
a higher degree of volatility is required for both languages to survive than
in the Abrams-Strogatz Model. Secondly, bilinguals accelerate language
death in small-world networks (Figure 8.7). Thirdly, bilinguals prevent lan-
guage death in networks with community structure (Figure 8.8). The role
of bilinguals, then, is highly dependent on the topological structure of the
network.
(2) The prestige of the languages (parameter s) has usually been claimed to
be determinant in the survival of a language, although dissenting voices
have been heard, such as Mufwene (2003). The volatility of the speak-
ers (parameter a) is probably not so often used in sociolinguistic studies,
but, as we have shown, it has its origins in social impact theory (Nowak
et al. 1990) and it has been included in pioneering attempts to model lan-
guage change (Nettle 1999a,1999b). Both parameters have been studied
in detail for regular lattices (section 3.3) and have also been qualitatively
approached for fully connected and random networks (section 3.4). Our
findings are in line with Mufwene’s (2003) as they question the overwhelm-
ing weight accorded to prestige as a factor affecting language maintenance.
The volatility of the speakers proves itself to be a more determining feature:
low volatility slows down language death in regular lattices (for socially
inequivalent languages), whereas high volatility prevents it from taking
place (for s = 0.5, as seen in Figure 8.6). In fully connected networks we
have observed that volatility is the only factor that conditions the transition
from language coexistence to language extinction, regardless of the value
of s (i.e. regardless the prestige of the two languages). Finally, in both fully
connected and random networks a high volatility of the speakers is required
for both languages to survive. Thus, the models presented here question the
almighty role played by prestige and militate instead for volatility as a fac-
tor of paramount importance in resolving language competition.
(3) The topological structure of the network has been proved to play a central
role in language competition too. In the first place, we have found that in
networks that do not give way to local interactions (fully connected net-
works), the role of prestige is simply irrelevant, while it does interact with
volatility in networks that give way to local interactions, that is, those in
which agents have a finite, small number of neighbours (random networks,
small-world networks and networks with community structure). Thus, local
interactions have been considered responsible for an increase in the role
played by prestige. Secondly, the topological structure of the network also
has been found to play a role in the speed of extinction when bilingual
agents are present: extinction takes place earlier in small-world networks
than in regular lattices, and earlier in regular lattices than in networks with
community structure.
Apart from these general results, it should be noted that the combination of
bilingual speakers and networks with community structure (the most likely
in real situations) is a powerful factor in the survival of threatened languages
(Figure 8.8) and it actually helps to understand real-life situations of language
maintenance. This leads us to two important conclusions. Firstly, even if our
intention was not to assess the degree of representativeness of the models pro-
vided by Abrams and Strogatz (2003) and by Wang and Minett (2005), we can
affirm that the introduction of bilinguals provides a more accurate approach to
reality. Secondly, the detailed assessment of different complex networks allows
us to state that those with community structure (based on the algorithm by
Toivonen et al. 2006) do provide a closer representation of many social net-
works. Our study, then, goes beyond the results by Ke et al. (2008), who do not
consider this type of topological structure, although they recognise that their
modelling is necessary and desirable (2008: 947). These two general conclu-
sions constitute a relevant contribution to the study of language dynamics, with
special reference to language competition.
Nevertheless, we are aware that further research must include improvements
in the models, such as the possibility of modelling the use of two languages by
the same speaker depending on the interlocutor. That is, in the models, language
could be considered a feature of each link between agents rather than a feature
of each agent, in line with what has been generally modelled by Fernández-
Gracia et al. (2012). A second improvement of the model could be to include
dynamical networks, in which links are not fixed, but change over time (just
like human relationships), as studied by Carro et al. (2014). Following these
two general studies, one could derive dynamical models of language competi-
tion considering, for example, that the preference to use a language (including
its knowledge) is a property of the agent, while the use of a language is a prop-
erty of the link. In a different line, future developments of the work could also
involve the modelling of heterogeneous agents, by introducing parameters such
as the age or the progressive acquisition of a language, as done in other studies

(e.g. Ke et al. 2008).
As for the contribution to Complexity Science, our approach to the study of
language dynamics, as presented in this chapter, fulfils the three initial steps
acknowledged to be taken in this field, following San Miguel et al. (2012),
described and discussed in section 2.1. Thus, we have observed reality about
language coexistence and possible competition and reviewed some literature on
language endangerment in order to identify some of the mechanisms involved
in language choice. What we learned enabled us to design models that may
contribute to a better understanding of the dynamics involved in language shift
or maintenance. We are aware, nonetheless, that there are still two further nec-
essary steps in order to meet the aims of Complexity Science: the validation of
the model and the construction of a theory.
The validation of the model is only possible through comparison with
real ethnographic data. Although prior attempts to model language shift have
resorted to real data (e.g. Abrams and Strogatz 2003; Blondel et al. 2008; Kan-
dler et al. 2010), this is a step we have not yet taken, given that our primary
aim has not been to predict but to forecast by trying to understand the mech-
anisms that underlie language shift from a strictly theoretical point of view,
leaving aside real data. This chapter shows that this can indeed be done and
reveals the role played by different factors in language competition, such as
bilingual speakers, prestige of the languages, speakers’ volatility and topo-
graphical structure of the network in a way that the mere analysis of data would
not allow. In the same vein, we would like to point out that blind data analy-
sis need not lead us to the formulation of a theory, but to a simple correlation
among variables without establishing cause-relation implications. In this sense,
we believe that non-data-based theoretical models may contribute to question-
ing the observer’s commonsense beliefs (Watts 2011) at the same time as new
questions open up in a given field. To sum up, then, even if future data-based
studies may help calibrate and validate the models presented here, we think that
the approach developed in this chapter reveals important information about the
factors that play a role in language choice and which might have gone unnoticed
in a study employing a discursive or a data analysis approach.
REFERENCES
Abrams, Daniel M., and Steven H. Strogatz. 2003. Modeling the dynamics of language
death. Nature 424, 900.
Albert, Réka, and Albert-Lászlo Barabási. 2002. Statistical mechanics of complex net-
works. Review of Modern Physics 74, 47–97.
Anderson, Philip. 1972. More is different. Science 177, 393–6.
Axelrod, Robert. 1997. The dissemination of culture: A model with local convergence
and global polarization. The Journal of Conflict Resolution 41(2), 203–26.
Ayestaran Aranaz, Margarita, and Justo de la Cueva Alonso. 1974. Las familias de la
provincia de Pontevedra (Galleguidad y conflicto lingüístico). Sevilla: Instituto de
Ciencias de la Familia.
Ball, Philip. 2004. Critical Mass: How One Thing Leads to Another. London: Heine-
mann.
Ball, Philip. 2012. Why Society is a Complex Matter. Meeting Twenty-First Century
Challenges with a New Kind of Science. Berlin: Springer Verlag.
Barabási, Albert-Lászlo. 2002. Linked: The new science of networks. Cambridge:
Perseus.
Barabási, Albert-Lászlo. 2005. The origin of bursts and heavy tails in human dynamics.
Nature 435, 207–11.
Baronchelli, Andrea; Vittorio Loreto; and Francesca Tria. 2012. Language dynamics.
Advances in Complex Sytems 15(3–4), 1203002.
Blondel, Vincent; Jean-Loup Guillaume; Renaud Lambiotte; and Etienne LeFebvre.
2008. Fast unfolding of communities in large networks. Journal of Statistical
Mechanics P10008.
Blythe, Richard and William Croft. 2010. Can a Science–Humanities Collaboration Be
Successful? Adaptive Behavior 18, 12–20.
Boccaletti, Stefano; Vito Latora; Yamir Moreno; Martin Chavez and D.-U. Hwang.
2006. Complex Networks: Structure and Dynamics. Physics Reports 424, 175–
308.
Borgatti, Stephen P.; Ajay Mehra; Daniel J. Brass; and Giuseppe Labianca. 2009. Net-
work analysis in the social sciences. Science 323, 892–5.
Borge-Holthoefer Javier; Alejandro Rivero; Íñigo García; Elisa Cauhé; Alfredo Fer-
rer; Darío Ferrer; David Francos; David Iñíguez; María Pilar Pérez; Gonzalo Ruiz;
Franciso Sanz; Fermín Serrano; Cristina Viñas; Alfonso Tarancón; and Yamir
Moreno. 2011. Structural and Dynamical Patterns on Online Social Networks:
The Spanish May 15th Movement as a Case Study. PLoS ONE 6(8), e23883.
doi:10.1371/journal.pone.0023883.
de Bot, Kees, and Saskia Stoessel (eds.). 2002. Language change and social networks.
[Special issue]. International Journal of the Sociology of Language 153.
Bradley, David, and Maya Bradley (eds.). 2002. Language Endangerment and Lan-
guage Maintenance. London/New York: Routledge Curzon.
Carro, Adrian; Federico Vazquez; Raúl Toral; and Maxi San Miguel. 2014. Fragmenta-
tion transition in a coevolving network with link-state dynamics. Physical Review
E, 89(6): 062802.
Castelló, Xavier. 2010. Collective Phenomena in Social Dynamics: Consensus prob-
lems, ordering dynamics and language competition. Ph.D. dissertation. IFISC
(Institute for Cross-Disciplinary Physics and Complex Systems, CSIC-UIB),
Palma de Mallorca, Spain. Available online at http://ifisc.uib-csic.es/publications/
publication-detail.php?indice=2103, accessed 8 November 2012.
Castelló, Xavier; Víctor M. Eguíluz; and Maxi San Miguel. 2006. Ordering dynamics
with two non-excluding options: Bilingualism in language competition. New Jour-
nal of Physics 8, 308–22.
Castelló, Xavier; Lucía Loureiro-Porto; Víctor M. Eguíluz; and Maxi San Miguel.
2007a. The fate of bilingualism in a model of language competition. Advancing
social simulation. The First World Congress, ed. by Shingo Takahashi, David Sal-
lach and Juliette Rouchier, 83–94. Berlin: Springer Verlag.
Castelló, Xavier; Riitta Toivonen; Víctor M. Eguíluz; Jari Saramäki; Kimmo Kaski; and
Maxi San Miguel. 2007b. Anomalous lifetime distributions and topological traps
in ordering dynamics. Europhysics Letters 79, 66006–1–6.
Castelló, Xavier; Riitta Toivonen; Víctor M. Eguíluz; Lucía Loureiro-Porto; Jari Sar-
mäki; Kimmo Kaski; and Maxi San Miguel. 2008. Modelling language competi-
tion: Bilingualism and complex social networks. The Evolution of Language. Pro-
ceedings of the 7th International Conference (EVOLANG7), ed. by Andrew D. M.
Smith, Kenny Smith and Ramon Ferrer i Cancho, 59–66. Singapore: World Scien-
tific Publishing Co.
Castelló, Xavier; Federico Vazquez; Víctor M. Eguíluz; Lucía Loureiro-Porto; Maxi San
Miguel; Laetitia Chapel; and Guillaume Deffuant. 2011. Viability and Resilience
in the Dynamics of Language Competition. Viability and Resilience of Complex
Systems. Concepts, Methods and Case Studies from Ecology and Society, ed. by
Gillaume Deffuant and Nigel Gilbert, 49–84. Boston/Dordrecht/London: Kluwer
Academic Publishers.
Castelló, Xavier; Lucía Loureiro-Porto; and Maxi San Miguel. 2013. Agent-based mod-
els of language competition. International Journal of the Sociology of Language
221, 21–51.
Centola, Damon; Juan Carlos González-Avella; Víctor M. Eguíluz; and Maxi San
Miguel. 2007. Homophily, Cultural Drift and the Co-Evolution of Cultural Groups.
Journal of Conflict Resolution 51, 905–29.
Crystal, David. 1997. English as a Global Language. Cambridge: Cambridge University
Press.
Crystal, David. 2000. Language Death. Cambridge: Cambridge University Press.
Dorogovtsev, Sergey, and José Fernando Ferreira Mendes. 2002. Evolution of networks.
Advances in Physics 51, 1079.
Dorogovtsev, Sergey N.; Alexander V. Goltsev; and José Fernando Ferreira, Mendes.
2008. Critical phenomena in complex networks. Review of Modern Physics 80,
1275.
Ehala, Martin. 2010. Ethnolinguistic vitality and intergroup processes. Multilingua 29,
203–21.
Ehala, Martin, and Katrin Niglas. 2007. Empirical evaluation of a mathematical model
of ethnolinguistic vitality: The case of Võro. Journal of Multilingual and Cultural
Development 28(6), 427–44.
Ellis, Nick C., and Diane Larsen-Freeman. 2009. Language as a complex adaptive sys-
tem. Oxford: Wiley-Blackwell.
Erdős, Paull, and Alfréd Rényi. 1959. On Random Graphs. I. Publicationes Mathemat-
icae 6, 290–97.
Fernández-Gracia, Juan; Víctor M. Eguíluz; and Maxi San Miguel. 2011. Update rules
and interevent time distributions: Slow ordering vs. no ordering in the Voter Model.
Physical Review E 84, 015103.
Fernández-Gracia, Juan; Xavi Castelló; Víctor M. Eguíluz; and Maxi San Miguel. 2012.
Dynamics of link states in complex networks: The case of a majority rule. Physical
Review E 86.066113(1–8).
Fishman, Joshua A. 1991. Reversing language shift: Theoretical and empirical founda-
tions of assistance to threatened languages. Clevedon: Multilingual Matters.
Fishman, Joshua A. (ed.). 2001. Can Threatened Languages Be Saved? Reversing Lan-
guage Shift Revisited: A 21st-Century Perspective. Clevedon: Multilingual Matters.
Gallagher, Richard, and Tim Appenzeller. 1999. Beyond reductionism. Science 284,
79.
Girvan, M. and M. E. J. Newman. 2002. Community structure in social and biological
networks. Proceedings of the National Academy of Science 99(12), 7821–6.
González-Avella, Juan Carlos; Mario G. Cosenza; Konstantin Klemm; Víctor M.
Eguíluz; and Maxi San Miguel. 2007. Information Feedback and Mass Media
Effects in Cultural Dynamics. Journal of Artificial Societies and Social Simula-
tion 10, 1–17.
González-Avella, Juan Carlos; Mario G. Cosenza; Víctor M. Eguíluz; and Maxi San
Miguel. 2010. Spontaneous ordering against an external field in nonequilibrium
systems. New Journal of Physics 12, 013010.
Granovetter, Mark S. 1973. The strength of weak ties. The American Journal of Sociol-
ogy 78, 1360–80.
Granovetter, Mark S. 1978. Threshold Models of Collective Behavior. The American
Journal of Sociology 83–1420–43.
Grenoble, Lenore A., and Lindsay J. Whaley (eds.). 1998. Endangered Languages:
Current Issues and Future Prospects. Cambridge: Cambridge University Press.
Grenoble, Lenore A., and Lindsay J. Whaley. 2006. Saving Languages: An Introduction
to Language Revitalization. Cambridge: Cambridge University Press.
Hamers, Josiane F., and Michel H. A. Blanc. 1989. Bilinguality and Bilingualism.
Cambridge: Cambridge University Press.
Hinton, Leanne, and Kenneth Locke Hale. 2001. The green book of language revitaliza-
tion in practice. San Diego: Academic Press.
Horgan, John. 1995. From complexity to perplexity. Scientific American 272(6), 104–9.
Kandler, Anne; Roman Unger; and James Steele. 2010. Language shift, bilingualism
and the future of Britain’s Celtic languages. Philological Transactions of the Royal
Society B 365, 3855–64.
Ke, Jinyun; Tao Gong; and William S-Y Wang. 2008. Language Change and Social
Networks. Communications in Computational Physics 3(4), 935–49.
Labov, William. 1972. Sociolinguistic Matters. Philadelphia: University of Pennsylva-
nia Press.
Leskovec, Jure, and Eric Horvitz. 2007. Worldwide Buzz: Planetary-Scale Views on an
Instant-Messaging Network. Microsoft Research Technical Report MSR-TR-2006–
186, Microsoft Research, June 2007. Available online at http://research.microsoft
.com/apps/pubs/default.aspx?id=70389, accessed 8 November 2012.
Leskovec, Jure, and Eric Horvitz. 2008. Planetary-Scale Views on a Large Instant-
Messaging Network. Proceedings of WWW 2008 Beijing, China, ed. by Jin-
peng Huai and Robin Chen 695–704. Available online at wwwconference.org/
www2008/papers/Proceedings.html, accessed 8 November 2012.
Levey, David. 2008. Language Change and Variation in Gibraltar. Amsterdam and
Philadelphia: John Benjamins.
Loreto, Vittorio, and Luc Steels. 2007. Emergence of language. Nature Physics 3, 758–
60.
Manley, Marilyn S. 2008. Quechua language attitudes and maintenance in Cuzco, Peru.
Language Policy 7, 323–44.
Milroy, Lesley. 1980. Language and Social Networks. Oxford: Blackwell.
Minett, James W., and William S-Y. Wang. 2008. Modelling endangered languages: The
effects of bilingualism and social structure. Lingua 118, 19–45.
Mitchell, Melanie. 2009. Complexity: A Guided Tour. Oxford: Oxford University Press.
Moreno, Jacob L. 1934. Who shall survive? Washington DC: Mental Disease Publishing
Company.
Mufwene, Salikoko. 2001. The Ecology of Language Evolution (Cambridge Approaches
to Language Contact 1). Cambridge: Cambridge University Press.
Mufwene, Salikoko. 2003. Language Endangerment: What Have Pride and Prestige Got
to Do With It? When languages collide. Perspectives on Language Conflict, Lan-
guage Competition and Language Coexistence, ed. by Brian Joseph, 324–46. Ohio:
Ohio State University Press.
Mufwene, Salikoko. 2004. Language birth and death. Annual Review of Anthropology
33; 201–22.
Mufwene, Salikoko. 2008. Language Evolution. Contact, Competition and Change.
London/New York: Continuum International Publishing Group.
Nettle, Daniel. 1999a. Using social impact theory to simulate language change. Lingua
108, 95–117.
Nettle, Daniel. 1999b. Is the rate of linguistic change constant? Lingua 108, 119–36.
Nettle, Daniel, and Suzanne Romaine. 2000. Vanishing Voices: The Extinction of the
World’s Languages. Oxford: Oxford University Press.
Nowak, Andrzej; Jacek Szamrej; and Bibb Latané. 1990. From private attitude to public
opinion: A dynamical theory of social impact. Psychological Review 97, 362–76.
Onnela, Jukka-Pekka; Jari Saramäki; Jörkki Hyvönen; Gábor Szabó; David Lazer;
Kimmo Kaski; Janos Kertész; and Albert-Lászlo Barabási. 2007. Structure and
tie strengths in mobile communication networks. Proceedings of the National
Academy of Science 104, 7332–36.
Romaine, Suzanne. 1989. Bilingualism. Oxford/New York: Blackwell.
San Miguel, Maxi. 2012. Fenómenos colectivos sociales. Revista Española de Física
26, 56–63.
San Miguel, Maxi; Jeffrey H. Johnson; Janos Kertesz; Kimmo Kaski; Albert Díaz-
Guilera; Robert S. MacKay; Vittorio Loreto; Peter Erdi; and Dirk Helbing. 2012.
Challenges in Complex Systems Science. European Physical Journal Special Top-
ics 214, 245–71.
Schelling, Thomas. 1978. Micromotives and Macrobehaviour. New York: Norton.
Schulze, Christian, and Dietrich Stauffer. 2006. Recent developments in computer sim-
ulations of language competition. Computing in Science and Engineering 8(3), 60–
7.
Schulze, Christian; Dieter Stauffer; and Soeren Wichmann. 2008. Birth, survival and
death of languages by Monte Carlo simulation. Communications in Computational
Physics 3, 271–94.
Steels, Luc. 1995. A self-organizing spatial vocabulary. Artificial Life 2, 319–32.
The “Five Graces” Group (Clay Beckner, Richard Blythe, Joan Bybee, Morten H. Chris-
tiansen, William Croft, Nick C. Ellis, John Holland, Jinyun Ke, Diane Larsen-
Freeman and Tom Schoenemann). 2009. Language is a complex adaptive system:
A position paper. In Ellis & Larsen-Freeman, 1–26.
Toivonen, Riitta; Jukka-Pekka Onnela; Jari Saramäki; Jörkki Hyvönen; and Kimmo
Kaski. 2006. Model for social networks. Physica A 371, 851–60.
Travers, Jeffrey, and Stanley Milgram. 1969. An experimental study of the Small World
problem. Sociometry 32, 425–43.
Tsunoda, Tasaku. 2005. Language endangerment and language revitalisation.

Berlin/New York: Mouton de Gruyter.
UNESCO. 2003. Language vitality and language endangerment. UNESCO (Ad Hoc
Expert Group on Endangered Languages). Available online at www.unesco.org/
culture/ich/doc/src/00120-EN.pdf, accessed 7 July 2014.
Vazquez, Federico; Xavier Castelló; and Maxi San Miguel. 2010. Agent-based models
of language competition: Macroscopic descriptions and order-disorder transitions.
Journal of Statistical Mechanics: Theory and Experiment, P04007.
Wang, William S.-Y., and James W. Minett. 2005. The invasion of language: emergence,
change and death. Trends in Ecology and Evolution 20(5), 263–9.
Watts, Duncan. 2003. Six degrees: The science of a connected age. New York: Norton.
Watts, Duncan. 2011. Everything Is Obvious: How Common Sense Fails Us. New York:
Crown Business.
Watts, Duncan J., and Steven H. Strogatz. 1998. Collective dynamics of ‘small-world’
networks. Nature 393(6684), 409–10.
Weinreich, Uriel. 1953. Languages in contact. Findings and Problems. New York:
Linguistic Circle.
Wölck, Wolfgang. 2004. Universals of language maintenance, shift and change.
Collegium Antropologicum 28(1), 5–12.
Wu, Ye; Changsong Zhou; Jinghua Xiao; Jürgen Kurths; and Hans Joachim Schellnhu-
ber. 2010 Evidence for a bimodal distribution in human communication. Proceed-
ings National Academy of Sciences (USA), 107, 18803–8.
9 Complexity and Language Contact:
A Socio-Cognitive Framework
Albert Bastardas-Boada
1 Introduction
Throughout most of the twentieth century, analytical and reductionist
approaches have dominated in biological, social, and humanistic sciences,
including linguistics and communication. We generally believed we could
account for fundamental phenomena in invoking basic elemental units.
Although the amount of knowledge generated was certainly impressive, we
have also seen limitations of this approach. Discovering the sound formants
of human languages, for example, has allowed us to know vital aspects of the
‘material’ plane of verbal codes, but it tells us little about significant aspects
of their social functions. I firmly believe, therefore, that alongside a linguistics
that looks ‘inward’ there should also be a linguistics that looks ‘outward’, or
even one that is constructed ‘from the outside’, a linguistics that I refer to else-
where as ‘holistic’ (Bastardas 1995), though it could be identified by a different
name. My current vision is to promote simultaneously the perspective that goes
from the part to the whole and goes from the whole to the parts, that is, both
from the top down and from the bottom up (see Bastardas 2013).
This goal is shared with other disciplines which recognise that many phe-
nomena related to life are interwoven, self-organising, emergent and proces-
sual. Thus, we need to re-examine how we have conceived of reality, both the
way we have looked at it and the images we have used to talk about it. Several
approaches now grouped under the label of complexity have been elaborated
towards this objective of finding new concepts and ways of thinking that better
fit the complex organisation of facts and events.
1.1 Complexity/Complexities
The use of the term complexity in science poses serious difficulties if we do
not first clarify the sense in which we are using it. The reason is that this label
has been taken up by a variety of disciplines and schools of thought. Thus, it
has been attached to perspectives, phenomena and aspects of reality that do not
dovetail very well with one another, which creates a good deal of confusion
that must be dispelled at the outset.
218
Complexity and Language Contact: A Socio-Cognitive Framework 219
The most frequent initial confusion is the use of the term complexity to refer
not to a given scientific approach, but to an intrinsic quality of many of the
phenomena of reality. In layman’s language, complexity or complex is custom-
arily used to signify complication, confusion, intricacy, diversification or a large
number of units and rules in play. For instance, we speak of the complexity of
a country or of a society, and the complexity of the human body or brain, or of
a specific language. Complexity, in these sorts of uses, does not refer so much
to a particular way of conceiving reality as to a feature of specific phenomena
of the world that we wish to understand.
In linguistics, the application of the term complexity has largely focused on
the structural and grammatical features of human languages, particularly the
comparative study of their grammatical systems. In other words, the focus has
been on the existing diversity in the number of units used and their character-
istics and combinations at different levels of verbal codes as well as how they
change and develop over time (see, for example, Sampson et al. 2009; Nichols
2009; Emmert-Streib F. 2010; McWorther 2011).
As previously observed, from a scientific point of view, and starting from
different fields and distinct lines of research, several authors have been con-
structing a paradigm that has also come to be known as the sciences (or theory)
of complexity (e.g. Morin 1980, 1991, 1992, 1994, 1999, 2001; Wagensberg
1985; Gell-Mann 1996; Cilliers 1998; Gershenson & Heylighen 2005; Castel-
lani & Hafferty 2009; Mitchell 2009; Jörg 2011). This is only one and perhaps
the most appropriate name from among others currently available, including
cybernetics (Wiener 1948; Ashby 1956; Bateson 1972; Heylighen 2001), sys-
temics (Von Bertalanffy 1981; Capra 1982), ecology (Margalef 1991), chaosol-
ogy (Flos 1995; Bernárdez 1995), autopoiesis or self-organisation (Matu-
rana & Varela 1996; Maturana 2002; Solé & Bascompte 2006), emergentism
(Holland 1998; Johnson 2002), and networks science (Newman et al. 2006;
Solé 2009).
In this approach, the terms complexity and complex are used to refer in gen-
eral to a characteristic typical of a great number of elements and phenomena of
reality: an organisation comprised of interwoven units that give rise to new and
emergent levels of organisation and (inter)actions, with properties and capabil-
ities that are distinct from those of the initial constituent elements. With this
type of phenomena, there is a high degree of awareness that any application of
a more or less mechanistic picture is fraught with difficulty, because of the vast
number of dynamic inter-retro-actions produced, which prompt the emergence
of new organisational levels that have different functions and, at the same time,
integrate functionally with other coexisting phenomena.
However, even within the theory or science of complexity approach itself,
there are important distinctions to make. On the one hand, we have a perspective
that focuses more on modelling and computation. On the other, the focus is
220 Albert Bastardas-Boada
more epistemological and philosophical, that of the pensée complexe, inspired

primarily by Edgar Morin. Morin (2007) called these two sets of approaches,
respectively, ‘restricted’ (complexité restreinte) and ‘general’ complexity.
Turning now to the more methodological and formalist approaches, the sci-
ence of complexity has made important contributions connected to the poten-
tialities of computing and to the appearance of new forms of mathematical rea-
soning that better suit complex and dynamic phenomena with a high degree
of interactivity and mutual emergent feedback. In recent decades, physicists,
mathematicians, computer scientists and some biologists and sociologists, prin-
cipally, have been the driving force in important lines of research and thought
devoted to studying the formal properties, potentialities and characteristics of
‘complex systems’. This ‘synthetic’ method – as it has been called by Luc Steels
(1995) to distinguish it from inductive and deductive methods – offers us some-
thing different from what we have seen so far: a chance to understand the gen-
esis and development of phenomena (see e.g. Abrams & Strogatz 2003; Ball
2005; Newman et al. 2006).
This method involves simulating complex processes with agent-based pro-
grams that model behaviours generally governed by simple rules that are them-
selves usually based on ‘if stimulus, then response’ formulations (cf. Wolfram
1983, 2002; Epstein 2006; Castelló et al. 2011). Such ‘complex adaptive sys-
tems’ (CAS)1 can also learn from their relationships with the context in order
to adapt better (by more adequately making use of the environment in order to
take full advantage of it for their own purposes).
1.2 Agents and Models in Language Contact

Some researchers have already started to apply the computational tools to lan-
guage contact and bilingualism (e.g., Loureiro et al., chapter 8; Castelló 2010;
Castelló et al. 2013). The model typically uses squares in a computer screen
to depict agents governed by simple rules of conduct they apply in accordance
with any other types of agents with which they come into contact. After a given
number of iterations, the result at the level of language will be the greater or
lesser use of one or another of the coexistent languages. For instance, if one of
the groups of agents is more predisposed to use its second language to speak
with members of the other group than to use its first language, we can see on
screen how such a situation will evolve.
One of the contributions of these types of methodologies is that they enable
us to clearly visualise the bottom-up phenomenon of sociocultural organisation
that emerges from interactions among agents (Holland 1998). We can clearly
1 Although the term has grown in popularity due to the efforts of the Santa Fe Institute, it should
be remembered that the sociologist Walter Buckley was using it as early as 1967.
see how, based on the application of a few stable rules, social behaviours can
shape a society’s customs and other cultural aspects. The complex concept of
self-organisation (Ashby 1962; Solé & Bascompte 2006) is highly useful for
understanding historical developments, including the evolution of language.
This is also true for the evolution of networks (Solé 2009), which enable us
to account for cultural variations that can be produced in accordance with the
configuration of the web of relations that individuals maintain with one another.
Thus, this type of modelling lets us literally see, on screen, in silico, the
process of reduction in the use of a language, showing agents who are applying
rules that they do not suspect will potentially generate negative consequences
over time and ultimately the practical extinction of what was once their first
language. The strategy of simulation, therefore, is a productive one, surpassing
classical statistical tools to the extent that it enables us to control the parameters
of an evolving situation and what emerges out of that situation.2
As for the second set of approaches, ‘general’ complexity (Morin 1992,
1994, 2008) is more heavily committed to an epistemological, multidimen-
sional, integrated and dynamic view of reality: the world is constituted by the
‘emergent’ overlap of different elements that produce new properties or organ-
isations as they complexify at higher levels. And this may go on from initial
physical and genetic elements all the way up to human societies and cultures.
It postulates that in order to gain an adequate understanding of the interwoven
fabric of all these domains in motion, we need to go beyond a way of think-
ing that tends to separate the elements of reality and treat them in isolation. It
pushes us beyond reductionist thinking that prioritises the elementary units and
quantitative aspects of phenomena. It calls for us to think in terms of ‘both/and’
and not ‘either/or’, applying fuzzy logic (Munné 2013) rather than Aristotelian
logic. It demands that we sidestep the pitfalls of dichotomies, and it builds from
the fact that complex thinking is not the ‘opposite’ of simple thinking, but rather
incorporates it3 (Bastardas 2002).
In linguistics, there are also contributions that have arisen between these
two major positions, building on the viewpoint of ‘complex adaptive systems’
from the Santa Fe Institute but seeking to go beyond computational modelling,
2 Not only simulations, but also programmes of this type using real data have been run to validate
a theoretical model. One example is the use of cellular automata to examine the processes of
language shift in Spain in a study devised by the group led by Francesc S. Beltran, using data from
the autonomous community of Valencia (2009 and 2011). The model assumes social pressure –
the number of people in the neighbourhood who encourage one behaviour or another – to be one
of the fundamental variables in the evolution of the sociolinguistic situation, and this allows us
to view the evolution of intergenerational language transmission.
3 It must be conceded that the use of the terms complex and complexity in the vast majority of
publications appearing in English – the most widespread language of science – corresponds much
more to ‘restricted’ complexity than to the more general perspective. For instance, the activity of
researchers at the Santa Fe Institute (Gell-Mann, Holland, etc.) has been immense and extremely
interesting. At present, this approach is also seeing a generous crop of developments in Europe.
toward interesting theoretical propositions like the ones put forward by Hol-
land (2005), The ‘Five Graces’ Group (2009) or Massip-Bonet (2013). Simi-
larly, the ecology-inspired contributions of Mufwene (2001, 2008, 2013) take
a broad perspective consistent with the approaches of general complexity. In
economics, for example, the work of Brian W. Arthur (2013) takes a very open
view also.
The general-complexity perspective aims not so much to devise precise the-
ories about specific phenomena as it sets out to take a comprehensive view of
reality, a view that is holistic and, at the same time, mindful of the autonomous
parts. From this viewpoint, in the words of Castellani and Hafferty:
Social complexity theory is more a conceptual framework than a traditional theory. Tra-
ditional theories, particularly scientific ones, try to explain things. They provide con-
cepts and causal connections (particularly when mathematicised) that offer insight into
social phenomenon ( …) Scientific frameworks, by contrast, are less interested in expla-
nation. They provide researchers effective ways to organize the world; logical struc-
tures to arrange their topics of study; scaffolds to assemble the models they construct
(2009:34).
Certainly, the two perspectives of complexity go together, but they correspond
to distinct levels and emphases, and they need to be complementary and inte-
grated. Recently, authors have taken this task in hand and are able to offer their
reflections to us. This is the case, for example, with Malaina (2012), Rodríguez
Zoya (2013), Roggero (2013), Ruiz Ballesteros (2013), Solana (2013), and
Byrne & Callaghan (2014), who deeply take both traditions into consideration,
integrating and evaluating them. Although computational methods and strate-
gies are useful to illuminate how human situations evolve, including situations
of language contact, it would be hard to go so far as to say that restricted com-
plexity needed to be the foundation on which to build a broad and complex
vision of sociocultural reality. In any event, it should be the other way round.
Our perspective ought to be built on the epistemological and representational
foundations of general complexity, and we need to use the methods, tools and
concepts of restricted complexity on the phenomena and processes that they
can best shed light on.4
Based on general complexity, my approach explores the world in a way that
helps us understand how language contact phenomena unfold. What follows
is my personal synthesis of the main principles of the complexity perspectives
in contrast with the more traditional scientific ones. The concepts listed in the
following two columns are not necessarily opposites.
4 “[W]e do not consider that mathematical representations of the social represent some ideal
towards which social science should be aiming. ( …) Mathematics can be a useful tool for
describing the reality but reality is its messy self, not a higher abstract order existing in math-
ematical form. ( …) When we approach the complex social we need methods which can take
accounts of context, agency and temporality” (Byrne & Gallaghan 2014:257).
Traditional Scientific Perspectives General Complexity Perspectives

conceptual reification there is no science without an observer
(centrality of brain/mind)
territory maps (we see by means of concepts and words)
scientific truth provisional theories
elements elements-and-contexts, interweaving,
figurations, interdependences, networks
objects events and processes
steady-state dynamic flux, change, evolution, development
classical logic fuzzy logic
linear causality circular, retroactive and nonlinear
causality, recursivity
either/or dichotomies both/and; integration and complementarity
top-down and planned creation bottom-up, self-organisation and emergence
unidimensionality inter-influential multidimensionality
‘explicate order’ (things are unfolded and ‘implicate order’ (everything is folded into
each thing lies only in its own particular everything; a hologram: the parts contain
region of space) information on the entire object)
fragmentation of disciplines inter- and transdisciplinarity
structure, code meaningful and emotional interaction
Figure 9.1 Main principles of the complexity perspectives in contrast with the
more traditional scientific ones.
2 A Socio-Cognitive Complexity Perspective on

Language Contact5
In principle, it is not easy to apply the perspective of general complexity to
understanding the co-determinants and evolution of language behaviours in
situations of intense language contact. Quite often, the study of these cases
draws on a sociolinguistic tradition that focuses more on fragmentary aspects –
for example, bilingual competences, code-switching, identities, policies, and so
on – than on a comprehensive view that is dynamic and transdisciplinary. By
contrast, our proposal aims to provide the basis of an integrative focus, from
a perspective of human socio-complexity which draws on the contributions
of traditional approaches to the study of language systems, but goes beyond
them.
One possible way to envisage an integrated study of the complexity of evolv-
ing conditions in situations of language contact is to conceive of them from an
5 In this chapter, I will not go into the specific effects of contact over and above habitual language
forms, such as the phenomenon of interference, language borrowing or change.
ecological point of view (Haugen 1972; Mackey 1979, 1980, 1994). Margalef
(1991:80) calls an ecosystem:
a level of reference formed by individuals together with the materials that are produced
by their activity ( …) and the matrix or physical surroundings in which they are included
and in which they act.
The assumption of the ecosystem concept is that the fate of a particular lin-
guistic variety – that is, its survival, its alteration, or its extinction – depends
basically on the evolution of the sociocultural factors that are involved in its pro-
duction. Its structure, then, is basically governed by the social functions that it
is required to perform. This is, in fact, a characteristic of the complex adaptive
systems. As Holland states, “the context in which a persistent emergent pattern
is embedded determines its function” (1998: 226).
Consequently, this approach sees the relation between languages and linguis-
tic groups as a three-way (rather than a two-way) phenomenon. In conceiving
the relation between two species, for example, the ecological perspective bears
in mind at all times the milieu in which the relation develops. This perspective is
vitally important to understand the impact of migration on language contact, or
the integration of a politically minoritised group in a state, as it underscores the
need to take into account the structure of the broad sociopolitical environment
as well as the groups in question.
Another principle on which this approach is based is that the different orders
and phenomena of the reality make up an interrelated whole, in which there
are not only circular, mutual influences between two variables but also a set of
dynamic interactions that make up the reality, as explained later. Thus mental,
interactional, collective, economic, political, and linguistic phenomena coexist
in such a way that one constitutes the other and vice versa. To express the image,
I use the metaphor of the musical score which enables us to visualise different
planes of the same unitary phenomenon and which exists sequentially, that is,
in time.
The static image of reality is also challenged. Contrary to the traditional
approach, time is an essential, continually present variable. Apparent stabil-
ity is always the result of a dynamic equilibrium that allows the conservation
of the identity of the units even if their elements are changed. More than as
a structure, reality should be seen as a set of events, or, to quote Bohm, as a
“universal flux of events and processes” (1987:31).
From this perspective the fragmentation into disciplines is also questioned.
As reality is multidimensional, an inter- and transdisciplinary focus is neces-
sary, especially in the sociocultural sciences. The new conceptual landscapes
must then allow the integration of perspectives of the different approaches in
a global theorisation which considers simultaneously all the necessary levels
of human beings in an integrated, coordinated way (see Capra 1982, 1996,
2002). In fact, the inspiration for this approach comes not only from outside
but also from inside the sociocultural sciences. According to Bastide (1971:8),
Auguste Comte himself seems to have suggested as much in an era that lacked
the conceptual instruments required for an ecological or complex approach:
‘In the natural sciences the elements exist before the whole; in the human sci-
ences, the whole exists before the parts’ (contemporary physicists make the
same claim for quantum physics). Twentieth-century authors, such as Gregory
Bateson (1972), Norbert Elias (1982, 1990, 1991), Kurt Lewin (1978), and Wal-
ter Buckley (1967, 1968), declare their support for an approach of this kind,
albeit from different angles. The report of the Gulbenkian Commission on the
reorganisation of the social sciences (1996), chaired by Immanuel Wallerstein,
clearly pointed in the same direction, as well as all the works by Edgar Morin.
Uriel Weinreich would agree:
It is in a broad psychological and sociocultural setting that language contact can best be
understood. ( …) On an interdisciplinary basis research into language contact achieves
increased depth and validity (Weinreich 1968: 4).
Noteworthy is also the fact that the orchestral complexity metaphor enables
us to understand and to organise in separate and yet interrelated ways the
dimensions which are most relevant to determining the behaviour of humans
in situations of linguistic diversity and contact. So, as an exploratory exam-
ple, we will construct a pentagram for each of the voices or instruments with-
out forgetting their interrelation with the other pentagrams among which rela-
tions of harmonic interdependence arise.6 For the moment, and in a brief and
simplified image, our score will comprise the following emergent and super-
posing basic parts: the minds, social interaction, human groups, and politi-
cal power (Bastardas 1996). Language varieties ‘live’ and interact with these
dimensions.
3 The Co-Environment of Linguistic Varieties
3.1 The Brain/Mind Complex

As John Holland says, “If we are to understand the interactions of a large
number of agents, we must first be able to describe the capabilities of indi-
vidual agents” (1996:7). Consequently, little can be understood about human
6 In the orchestral score one can see the evolutions of each of the instruments or voices and of the
whole that results from the superimposition of one on the other in the interpretive sequence of
the work. This is no more than applying the vision of systems, where each level forms part of a
whole of multiple interrelated levels, the cooperation of all of which produces the emergence of
a specific behaviour or global product able to be perceived by and to influence another human
being.
behaviour if we do not begin our examination of the metaphorical score by

looking at the ‘brain/mind complex’, since it is at this level that the ultimate con-
trol over human action and understanding lies. Let us, therefore, take the human
being as a bio-social product endowed with a brain/mind that will enable and
regulate the individual’s relationship with the world7 (see Maturana & Varela
1996).
The usual development of the neurocognitive complex occurs – and out of
necessity must occur – in close interrelation with the sociocultural context,
in other words, through interaction with other human predecessors and their
products. Without this requisite activation, during the optimum phase, of the
genetic programming by the stimuli of social activity, no brain can develop
properly or have any chance of recovery during its lifetime. In all probabil-
ity, the interrelationship between the developing brain/mind and the sociocul-
tural phenomenon can be seen more accurately in terms of self-organising sys-
tems rather than in terms of the computer metaphor, with its traditional inputs
and outputs (Varela et al. 1992:157). The fruit of this functional autonomy is
that the brain/mind will construct itself from the perceived cultural artefacts
derived from the social interactions, so that, as Morin suggests, it is a complex
phenomenon formed from inseparable elements – the brain/mind, the individ-
ual and the society/culture – each of which, in its own way, contains the oth-
ers (1986:84). This approach also helps to advance our anthropological under-
standing, given that our cultural knowledge is to be found – as it would appear
to be in reality – at the interface of these elements rather than in one or in all of
them (Varela et al. 1992:178).
In socialisation, the structure of the sociocultural contexts represents a factor
of considerable influence on the final results of this process. If someone comes
across different ways of speaking in the set of contexts in which he or she lives,
the degree and quality of development in each of these linguistic varieties might
differ according to the type and the intensity of exposure or use. The variety or
varieties used in the family setting – in particular that of the mother or per-
son(s) who spend(s) the most time with the child in the first year of his or her
life – will supply the initial elements for the development of comprehension
and expression. It/They will tend to become the code or codes that will form
the base for the conceptual structuring of reality, for the development of the
collective identity of the individual and – if the social context does not impede
it – for the informal linguistic communication ability of the person.
If the remaining social contexts – neighbours, networks of friends, pupils
at school, teachers, the mass media, and so on – confirm the way of speaking
the individual has acquired in the family – and/or at the kindergarten, which
7 Unlike the physical sciences, at the level of human phenomena it is not only the mind of the
observer that we should take into account, but the minds of the subjects of the observation as well.
We should take account of the mind not merely because of its intrinsic importance, but because
it is inside the mind that the great majority of the courses of action of humans are determined.
these days acts in part as a substitute for these contexts – the individual will
gradually expand his competence in this code. He will acquire the registers
corresponding to the different functions and situations and then will use them as
a matter of course in his daily communicative acts with no problem whatsoever.
If, on the other hand, the individual finds other distinct varieties outside the
initial context of socialisation, then he will be presented with a problem of a
different nature with complex causes and consequences. Thus, if, for example,
the variety developed in the family setting is neither spoken nor understood in
all the other contexts, the individual will be obliged to acquire as quickly as
possible the other way(s) of speaking and will become bi- or multilingual –
or bi- or polylectal if they are varieties of the same language.8 In this type of
situation the individual might finally develop a greater degree of competence
in his second variety than in his first – in particular, if exposure to the latter
occurs during the critical period and, if, in addition to the informal contexts, the
new variety is also that of the formal public contexts – the medium of formal
education, of the general street signs, of the media, and so on.
In terms of understanding linguistic behaviour, there are two main interre-
lated functions of the brain/mind complex that would appear to be of particu-
lar relevance: the representation of reality and control over behaviour. It is in
the brain/mind complex where we construct and sustain ideas about the real-
ity that we experience, and from where we activate our motor organs to carry
out specific actions – determined in accordance with the discursive representa-
tions and interpretations of the reality that we make (Van Dijk 2010). And this
we can do, as we shall see, either from the conscience or the ‘subconscience’.
We can hold certain definitions of reality without being conscious of so doing,
and similarly we can undertake certain actions without having been conscious
before, or at the time, of having done so. The conscience, therefore, does not
exhaust the mind. Many of our mental acts are not directly accessible from the
conscience.9
8 Individuals might also find two (or more) systems of linguistic communication within the family
domain. The most typical case is that which occurs where each progenitor uses a different lan-
guage to address the child. In such situations, and if the person-language norm is consistent, the
child should be able to develop more or less equally the bases of two mother tongues (Weinreich
1968:77), and will use them appropriately according to the situation (Fantini 1982:63).
9 In fact, as Popper and Lorenz (1992:30) pointed out, learning includes the effort of consigning
what one has just learnt to the subconscience. Thus, a large part of our behavioural and cognitive
activity is subconscious. The high degree of consciousness that we maintain over each action
when learning to drive a car becomes part of a routine and our subconscience when we have
some experience and we wish to centre our attention on the road. We must conclude, therefore,
that the phenomenon also affects linguistic behaviour and all other human activities. Indeed,
Bateson believes that “the conscience must always be limited to a rather small fraction of mental
process. ( …) The unconsciousness associated with habit is an economy both of thought and of
consciousness; and the same is true of the inaccessibility of the processes of perception. The
conscious organism does not require (for pragmatic purposes) to know how it perceives – only
to know what it perceives” (1972:136).
At the heart of this conception of the human being as a cognitive-interpretive

being is, as maintained by the perspective of symbolic interactionism, the view
that “the meaning does not emanate from the intrinsic makeup of the thing that
has meaning but rather from and through the defining activities of the peo-
ple as they interact” (Blumer 1969:4). Things do not have meaning on their
own; rather, it is human beings that attribute meaning to things, be they physi-
cal objects, words or actions, through the cognitive processing of apprehended
information and internalised interpretive procedures. Facing any perception
and, frequently, from our subconscience, the world is processed and understood
drawing on the available cognitive depository. Any perception that cannot be
recognised and interpreted from the knowledge available at that time will acti-
vate the conscience in order to produce a hypothesis that makes sense, that
might explain what it is that we are perceiving, how it relates with our other
perceptions, what function it performs, for whom, and the like. As Schutz said,
I cannot understand a cultural object without referring it to the human activity in which
it originates. For example, I do not understand a tool without knowing the purpose for
which it was designed, a sign or symbol without knowing what it stands for in the mind
of the person who uses it, an institution, without understanding what it means for the
individuals who orient their behaviour with regard to its existence (1974:41).
While human beings develop direct referential interpretive capacities in rela-

tion to the linguistic structures perceived in their social interactions, they also
develop social evaluative interpretations of these same linguistic structures, in
particular in situations of diversity of ways of talking. Therefore, speaking in
one or another variety, or using each other’s linguistic form, might be socially
significant and have major repercussions on the interaction that develops. Just
as we can assign meanings to the social status of the clothes that we wear, the
linguistic varieties used can also be associated with specific social meanings.
When we interpret our perceptions, we do so polyphonically, multidimension-
ally. Virtually never do we consider one level of meaning in isolation; we inte-
grate within our pertinent perceptions or information, and, what is more, from
within a hierarchical organisation.
The social meanings of the linguistic forms are part of the individual’s
cognitive-interpretive stock and can thus influence the action both of the poten-
tial user and the interlocutor. The latter might, for example, not offer room to
someone who speaks in a way that is considered negative socially. Therefore,
the individual that is perceived negatively might decide, for example, to aban-
don the linguistic variety that is disadvantageous, perhaps especially when it is
also derided.
From out of this context of competences, meanings, habitus and uncon-
scious routines, the individual will choose specific language forms to use in
communication with other people. As we can see, for example, individuals in
situations of social bilingualism will use one code or another in line with the
representation they have of their interlocutor, the available competences (of the
two individuals), the social meanings of the varieties to hand, their previously
established custom, and their cognitive interpretation of the context in which
the interaction takes place. The sum total of behavioural decisions taken by the
individuals of a given society will mark the evolving dynamic of the situation.
3.2 The Level of the Social Interaction of Brains/Minds

Individual brains/minds are also, at the same time, the building blocks of a
higher level of complexity, which emerges naturally from the properties of its
constituents but adds new features that are typical and characteristic of the new
interactional dimension and of the social situations in which this dimension is
produced. Thus, at this level, there are all the elements that we have identified
as belonging to the brain/mind complex as well as all those that arise from the
need for organisation at the level of interaction.
Social interaction must be viewed as an inextricably socio-mental relation
and therefore cognitive and interpretive in nature. If, in normal circumstances,
human action is given meaning by a subject who can actively form interpre-
tations through observation and perception, then any interaction is necessar-
ily mutually significant. Actions, movements, gestures, verbalisations, paraver-
bal elements, the language forms and varieties that are being used, the situa-
tions in which these occur, the biographical precedents of the relation, expected
intentions, and other factors will be a constant source of conscious or uncon-
scious interpretations processed holistically between interacting individuals
(Serrano 1993). Understanding the social organisation of interpersonal rela-
tions becomes crucial to understanding enormously important aspects of lan-
guage behaviour.
In many cases, human interactions are well organised, quite often predictable
within given limits, and meaningful. Daily encounters tend to unfold according
to socially established norms and rituals that coordinate social life so that we do
not have to improvise behaviour each time we come into contact with another
human being and so that we can adequately interpret our interaction. Though
self-organisation constantly applies in daily interactions, it is not a simple fact
or at least a fact easy to describe in detail.
The entire complex of behaviours will be interpreted holistically by the inter-
locutors in terms of the instructions of ‘scripts’ for various social settings,
which determine the extent to which the behaviours fit socially habitual expec-
tations. If the ‘script’ for a given ceremony calls for a given level of formality
of apparel, for example, a person wearing clothing categorised by the social
majority as ‘informal’ may be judged negatively. This negative assessment,
however, will be attenuated or even changed if other significant aspects of the
same person are valued positively, such as his or her way of talking, gestures
or accessories, or if the person has a convincing reason for dressing in this
way.
Once adopted and established sociocognitively, the most daily, regular and
repetitive norms of action generally become subconscious and are followed
routinely, almost automatically in most cases (Nisbet 1977). Such actions
include how to greet someone, what to say and what to do when departing
a place, how to structure a conversation, and also what language or variety to
use when speaking with a particular person.
A relation between individuals typically tends to establish fixed patterns of
behaviour between the two participants. For example, if at some time we adopt
the custom of kissing at each meeting and we repeat this behaviour for a certain
number of days, it is highly likely to become a norm and, therefore, an expec-
tation that must be satisfied at each meeting in the future. Similarly, once we
have, by mutual agreement, adopted a given language behaviour and more or
less confirmed it by periodic repetition, the selection of the variety or language
becomes subconscious and routine; and it will tend to be perpetuated. Indeed, at
some point, changing the variety or language will become extremely difficult.
In social situations involving language diversity, selecting the variety or lan-
guage to use is not a simple action. An initial factor that can influence this selec-
tion is the language competence of the individuals involved. If two individuals
can only understand and speak one variety, they will use it in all likelihood.
However, if they also understand and speak another one, the choice is more
complex. They may choose to use either one. That is, the two individuals are
likely to negotiate the variety to be used, because it is common in communica-
tive relations to prefer the use of one and only one variety by both individuals,
provided that their mastery is sufficient to make this possible (Hamers & Blanc
2000; Hamers 2004). Presumably, if there is a discrepancy, the negotiation is
won by the interlocutor who is more persuasive. If each interlocutor remains
firm in his own position, the ensuing interactions will involve the two interlocu-
tors speaking different languages – what has been called ‘bilingual conversa-
tion’ – or the interlocutors will tend to avoid interactions so as not to reproduce
the conflict each time they communicate.
Although language behaviour tends to be decided subconsciously and rou-
tine, the possibility always exists to bring it back to the conscious level and
control it directly and reflectively by the individual, overcoming the constraints
of competence and habitual behaviour, if desired (Bastardas 1995). Of course,
this also entails social consequences, negative and positive, that may arise from
the individual’s decision. For example, individuals may decide that it is better
for them to change their manner of speaking with specific people or in gen-
eral, using individual words or constructions, a different variety of the same
language or a different language, rather than speaking as they always have.
This fact lies behind many of the ways in which sociolinguistic situations
evolve.
As we shall see next, factors belonging to higher levels of organisation (e.g.,
the group or political level) exert an influence at the interactional level as a
result of the complex interdependence across the various domains of the soci-
olinguistic ecosystem. Thus, an individual’s language behaviour can be affected
not only by elements at the interactional level such as the necessary organisation
of the use of varieties in conversation, but also by elements such as the issue of
identity in relation to the language groups in contact or the regulation of public
uses implemented by the political authorities. Both domains can influence the
language choices made by individuals in their interactions, in accordance with
their representations of reality.
3.3 The Social Group Dimension

3.3.1 The Emergence of Networks and Groups Not only are human soci-
eties organised by interactional pairs but also their members interact with a
greater number of people with whom they co-construct in complex ways a new
level of reality, a ‘groupness’ that shapes many of the aspects of what we call
‘culture’ as opposed to ‘nature’. These individuals-in-society, as Norbert Elias
would say, take advantage of the potentialities of the brain to build networks
and together consolidate forms of sociocultural and communicative organisa-
tion that will adapt to the changing nature of their collective historical experi-
ences. The group, therefore, will usually be the basic unit of survival and social
control, setting standards that will constrain individual and collective action. In
the group, Morin’s idea of recursiveness is at work; that is, the individual makes
up the group that makes up the individual. Once it has formed, a group tends
to persist if it is a functional organisation and it benefits individual members.
Thus, cause and product maintain and change one another.
The existence of groups and limited networks of intense interaction gives rise
to the possibility of cultural diversity. Each network can autonomously create
representations and establish forms and norms of conduct that differ to vary-
ing degrees from those adopted by other collectives. More specifically, groups
differ in the degree of importance they give to specific elements of daily life,
in the behaviours they deem appropriate in various situations, in the language
forms they use or prefer to use, and so on. Cultural and linguistic diversity is a
real, well-established fact.
Sociocultural categorisation plays a significant role in decisions that indi-
viduals make to take joint action. As members of socially, economically and
culturally stratified urban societies, we always interpret others as members of
some social group or category that is the same as or different from our own,
with the ensuing normative and evaluative associations, and we make decisions
about our actions based on such associations. In ethno-linguistic conflicts it is

crucial to distinguish situations characterised by political subordination from
conflicts produced by mere territorial co-existence. In the first case, the con-
flict is typically based on the expansionism of a demographically or militarily
stronger group into a neighbouring territory inhabited historically by other col-
lectives with different cultures. In the second case, groups cohabit regularly in
the same territory and some discord arises from some reason within or outside
the groups. Both cases can generate a high level of awareness of ethnic identity
within the collectives in conflict, and this awareness can have an effect on any
possible inter-group behaviours, because ethnic identity, according to Barth, “is
similar to sex and rank, in that it constrains the incumbent in all his activities,
not only in some defined social situations” (1976:18).
In inter-group relations, the system of linguistic communication that is used
may become highly significant. It may come to act as an ethnic identifier, with
the consequences that that identification entails. Similarly, the overall configu-
ration of ethnicity and inter-group relations may have an effect at the level of
language. For subordinate groups, a positive ethnic consciousness can be the
reason for maintaining their language, while a negative one is the most com-
mon cause for them to abandon it. The differences, for example, in the evolution
of the sociolinguistic situation of Catalonia and the Valencian Community, in
Spain, are probably based on that factor, which itself is due to other elements,
like differences in the history of their economic development (Ninyoles 1978;
Aracil 1982).
Many ethnolinguistic groups that have been incorporated in larger political
units, often forced by historical events, are typically structured according to the
environment of another ethnolinguistic group which can control and ‘patrimo-
nialise’ the political power by virtue of its demography or some other strength.
These groups wind up accepting or rejecting their involvement at the level of
identity in a larger body they are part of, on which they depend economically
and politically, and yet which they perceive as alien to their own self-defined
cultural characteristics. A more or less significant segment of the population
may move toward full acceptance of the superordinate identity, accepting it as
basic, while the original ethnic identity comes to be viewed as secondary. At
the other extreme, another segment may see the matter in the opposite manner:
native cultural traits are fundamental and primary; given that these traits are
typically in decline in the face of traits associated with the polity in which the
ethnic group is found, the cultural forms represented by the state and the state
itself lose legitimacy and are rejected. A third position may also arise in which
individuals contrive a combination of identities and find greater compatibility
between the group categories in conflict. We can see all these evolutions in the
current cases of language contact in Spain (Siguan 1993; Coller 2006; Strubell
& Boix-Fuster 2011).
3.3.2 Groups and the Macrosocial Order The dimension of power, like
that of social category, is constantly present in social life and is an abso-
lutely central element in the changing fortunes of groups. Typically, human
groups do not have identical controls over economic, informational, demo-
graphic, or political resources. An awareness of the differences often comes
with inter-group comparisons. A habitual way of speaking that is associated
with a particular group of high socioeconomic status, for example, may be
admired or rejected according to how it is perceived by other groups it coex-
ists with. Certain ways of speaking among the upper classes can be ridiculed
by members of other social groups and vice versa. In other situations, mem-
bers of social groups at the lower end of the social scale can have self-negating
representations that spur in them the desire to adopt the language traits of eco-
nomically higher social groups in order to raise their own personal prestige and
improve their self-image. Indeed, as Pierre Bourdieu (1980, 1984) points out
very effectively, social positions are closely related to the predispositions – the
representational and behavioural habitus – of the individuals who occupy those
positions. This fact can often be the cause of more or less large-scale shifts in
the language behaviour of individuals aiming to emulate the language forms
that are more closely identified across the entirety of the society as belonging
to affluent, socially dominant groups.
In today’s information society, the greater or lesser possession of cultural or
symbolic goods (e.g. academic training, skills in the arts or high technology
or specialised knowledge of diverse types) is also a factor of social differen-
tiation that is not purely economic in nature. Intellectual elites can also con-
stitute social groups that are perceived and evaluated as having prestige by a
large part of the population, particularly by the middle and working classes.
As a result, they can have a significant influence on the evaluation and use of
specific language forms. Through institutionalised roles occupied by virtue of
political power or social structure, they convey to the population whose forms
shall be considered legitimate and valuable in public discourse. A minority pop-
ulation in a subordinate position can easily shift cultures, a phenomenon that
has occurred frequently in all periods of history.
3.4 Political Power

We have seen how human beings construct new sociocognitive realities basi-
cally in a self-organised manner. They hold representations based on facts of
the world, reaching a consensus on the forms of their interactions and form-
ing cultural and economic groupings. The culmination of contemporary devel-
oped societies is their organisation into states, which are not based specif-
ically on phenomena of self-organisation, but rather on an explicit, formal
act of establishment. At this level, particular individuals – sometimes elected
democratically and sometimes not – are able to exercise significant power over
many aspects of social life. It is a fact that the contemporary state intervenes
more than ever in the life of human communities and exercises an enormous
influence. The level of language not only fails to escape its influence but rather,
contrary to what one may think, can be highly controlled and determined by
political power.
Specifically in the area of language, the impact of political power is both
direct and indirect. Because the state can require the compulsory fulfilment of
its provisions, the explicit or implicit declaration of an ‘official language’ will
result in the codification of the selected variety or varieties, assuming they were
not previously codified. It will also extend their knowledge and use to public
functions across the entire territory where they are named. Typically, even with-
out explicit regulation, the variety selected as an official language will also tend
to be adopted in the remainder of public communication that is not dependent
on the state government. Quite often, it will be the only variety that citizens con-
sciously and reflectively learn and the only one readily available for them to use
in formal speaking and writing. As a result, it will de facto become the language
variety that can be used comprehensively in institutionalised communications
(Corbeil 1983) within the area over which the state exercises sovereignty. It may
even be used in private writing to a great extent. Moreover, as we shall see, the
linguistic characteristics of the official variety may eventually be adopted even
in informal spoken communication, particularly in cities, where the process of
urbanisation also entails both the destruction of local sociocultural ecosystems
and the need to adopt new norms of communication in the complex urban envi-
ronment.
In phenomena of language contact, the state can be particularly critical to
how situations evolve, because the linguistic pressures brought to bear by the
regulation of the official status of languages can play a crucial role in the main-
tenance or abandonment of the varieties in contact. When a state with a multi-
ethnic population identifies itself solely with one of its ethnolinguistic groups,
the situation is a source of potential conflicts between the state and the dominant
ethnolinguistic group, on the one hand, and smaller ethnolinguistic groups, on
the other hand, like in Spain during General Franco’s rule. The political will
of the state to unify its citizens linguistically can crash head on into commu-
nities that often prefer to stick to their own language varieties, varieties whose
structures may differ sharply from those of the official language. Such com-
munities, which may have a historical awareness of collective differentiation,
may not be ready to accept a homogenising policy. If the history also includes
forced annexation, economic or religious differentiation and a policy of national
uniformity pursued at the expense of the languages and cultures of smaller
groups, the conflict can be prolonged and acute. Under these conditions, the
politicisation of the ethnolinguistic reality is inevitable, because the state is the
instrument needed by an ethnolinguistic group wishing to become a ‘nation’

or even simply to protect itself from assimilation. The state’s political sub-
ordination of some of its constituent populations is the fundamental cause
of many processes of language shift, which are nothing more, in such cases,
than the displacement of the traditional language varieties of smaller groups
toward disuse, privileging the official standard varieties sponsored by the
state.
4 Ecosystem Dynamics
So far we have drawn the lines of the orchestral or polyphonic score that can be
used metaphorically to develop a dynamic view of the processes of language
contact from the perspective of complexity. This means that we will understand
the evolving outcome of these types of processes much better if we see them as a
consequence of the mutual influences of the various lines in the score, that is, of
the conflict or reconciliation of different pressures present in the sociocultural
ecosystem (Terborg & García-Landa 2013).
Thus, for example, if we want to gain a holistic understanding of the devel-
opment of sociolinguistic situations such as in Catalonia, we should simulta-
neously consider all levels: their complex interdependences and the changes
occurring over time. We can clearly observe interrelations among the distinct
domains, for example, in the process of restoring the Catalan language under-
taken by the government of Catalonia. While the public school system seeks
to expand knowledge of Catalan (and Spanish) among a population in which
one segment has Catalan as L1, another segment has Spanish as L1, and a fur-
ther growing segment has a mother tongue other than these two, the degree
of success in increasing the interpersonal use of the autochthonous language
is relative. The reason is that there are complex factors at play that have their
own, hard-to-change dynamics. On the interactional level, for example, there
is a very widespread habit among L1 Catalan speakers to accommodate L1
Spanish speakers and not vice versa – a behaviour that is also being taken up
by a large proportion of the younger generations and becoming readily auto-
matic and unconscious. What is occurring at this level would seem to contradict
what is happening at the group level, where there is a growing awareness of
Catalan identity, at least in the group of L1 Catalan speakers. Contrary to what
might be expected, however, this is not carrying over to the level of language
behaviour adopted with L1 Spanish speakers who have already become bilin-
gual – at least within the school context – in Catalan. The brain’s mechanisms
of memory and routine formation appear to play such a fundamental role that
once a language choice for speaking with a person has been made, the choice
is typically repeated in new encounters, even throughout an entire lifetime and
sometimes in contexts that were considered inappropriate. One example of this
can be seen in partners of mixed linguistic origins who get to know each other
in Spanish and continue using Spanish with one another throughout their lives,
even though they speak Catalan to their children. The opposite may also occur,
as we more often see in the Valencian case, when partners speak Catalan to each
other, but speak Spanish to their children (Querol 1990; Conill 2003; Montoya
& Mas 2012).
A complex, multidimensional perspective enables us to grasp much bet-
ter why there are some processes in which state policies prevail and popula-
tions adopt the code that has been declared official, while there are other cases
in which this does not occur so easily. Apart from the previously cited ele-
ments of routinisation and the subconscious, emotional collective identifica-
tions and cognitive self-images also come into play in the acceptance or rejec-
tion of a state language. If a language group views and experiences a state’s
language policy as illegitimate, for example, because its own language is not
given official recognition, it can tend to reject the policy and prefer not to use
the state language. On the other hand, if the group sees an opportunity in the
state language to make social and economic progress and it has negative rep-
resentations of its own code, it will tend to adopt this language much more
readily.
The emphasis placed by the perspective of complexity on the time and pro-
cessuality of phenomena is also highly useful in grasping how different situ-
ations evolve. In the case of Catalonia, for example, it is evident that current
difficulties in further expanding the use of Catalan have their roots in the path
dependence on earlier events. As the Franco dictatorship achieved a high level
of bilingualisation in Spanish among L1 Catalan speakers in a context in which
the public use of Catalan was banned, Catalan speakers largely grew accus-
tomed to using Spanish with L1 Spanish speakers coming to reside in Catalo-
nia, particularly in the city of Barcelona. Now, the established habit is deeply
ingrained and much more difficult to change, despite explicit campaigns with
coherent arguments about the danger posed for the future of Catalan. In addi-
tion, a complex, systemic view can more adequately explain why the group of
L1 Spanish speakers increases their interpersonal use of Catalan less, given the
fact that the earlier adaptation of L1 Catalan speakers leaves them no opportu-
nities to use Catalan, despite government efforts (Boix-Fuster & Farràs 2013).
The biological replacement of populations through generational change is
also of major importance in how processes of contact evolve. New individuals
who replace those who came before can be socialised in a context distinct from
their parents and cognitively adopt new representations that prompt them to
shift to the prevailing behaviours. As complex adaptive systems, the younger
generation may already be more habituated to contact with another group than
their own parents, for example, and they may develop the language skills
needed in a situation of contact earlier on, at a time that is more biocognitively
optimal. Economic contexts, too, can evolve and come to exert influence on the
social meanings and attitudes ascribed to language varieties, which can in turn
lead to the adoption of new behaviours. Viewing these issues from a complex
perspective is not only useful but necessary.
Consequently, it is of special importance to stress that language contact must
be understood as a historical and, therefore, temporal phenomenon, with earlier
events playing a major role in how the phenomenon evolves. In other words,
we need to pay attention not only to the synchronic elements, but also to the
diachronic ones, because the latter may determine the future development of
the phenomenon (Elias 1982).
5 Conclusion
We have seen some examples of how restricted and general complexity per-
spectives can help understand the interwoven mechanisms of sociolinguistic
dynamics at work in cases of contact. Given its holistic view and its mindful-
ness of its own parts, the suggested image of the orchestra that we have used as
a metaphor can enable us to integrate the micro and macro (i.e., respectively,
inter-individual and institutional) dimensions present in human experience and
reflect the mutual interdependencies of the participating levels. What this can
encompass ranges widely from genetic to sociopolitical constraints, and simul-
taneously, it takes into consideration the presence of time as an inescapable
context for the emergence and existence of language forms and varieties.
From the general complexity perspective we must emphasise the cognitive
and emotional uniqueness of human agents and not lose sight of the importance
of this factor in their (inter)actions. In addition, the human social organisation
itself makes linguistic/communicative activity ever present in a variety of areas
and specific contexts that also exert a reverse influence on the linguistic level.
The human linguistic phenomenon is at one and the same time an individual,
social and political fact. As such, its study should bear in mind these complex
interrelations, produced inside the framework of the sociocultural and historical
ecosystem of each human community.
The complexity of these interrelationships, their multidimensionality and the
numerous factors that can affect their evolutionary dynamics will not make
it easy to apply tools in the field of linguistics that are also valid for other,
less complex phenomena. For example, the methods and concepts of restricted
complexity can be used as supplementary strategies that are highly useful in
studying certain characteristics, the stages and speeds of processes of language
contact, but always within the frame of the broader view offered by general
complexity.
Methodologically, this use of computational instruments specific to restricted
complexity ideally needs to be accompanied by the combined use of
qualitative strategies, because without such strategies we cannot gain access

to the sociocultural meanings that individuals confer to their language forms
and to the representations by which they experience the events of their lives
(Mead 1934; Bruner 1990). From the perspective of general complexity, con-
ducting comparative case studies is an interesting strategy in this regard and it
also complements approaches that are based more on statistics and formal mod-
elling (Byrne & Callaghan, 2014). Complexity aims to overcome fragmentary
methodological views. It postulates their harmonious integration in the service
of attaining the deepest possible understanding of phenomena as a whole. As
stated by Byrne and Callaghan:
[W]e see complexity as providing a framing for the unifying of a whole set of opposites
in scientific practice, of quantitative and qualitative research, of analysis and holism as
modes of understanding, and of relativism and hard realism as epistemological position
(Byrne & Callaghan, 2014:255).
In order to develop paradigms, the two major complexity approaches need to

find more common ground and take steps toward a mutual integration based
on the acceptance of the shortcomings of each approach, achieving progress
through a non-contradictory complementarity of perspectives (Heylighen,
Cilliers & Gershenson 2007; Bastardas 2014). It must be conceded that the
practical and methodological applications of basic complexity ideas need to be
developed much farther in order to apply them to specific research. As Roggero
has noted:
[t]oday, there are more experts in formal disciplines taking an interest in the social sphere
than there are sociologists borrowing the techniques of the formal disciplines. If a meet-
ing of minds takes place, it will turn out to be hugely beneficial for both groups. The first
will need to learn sociology’s language and ways of thinking, including the sociologi-
cal culture; the second will have to contend with the formal rigour, the methodological
demands and the utilisation of useful computer tools found in the formal disciplines
(2013:116).
At the same time, the limits of complex adaptive systems as computational

strategies must be accepted in the pursuit of a better understanding of the
evolutionary processes typical of human beings. In the final analysis, models
always have a narrative running behind them that reflects the attempts of a
human being to understand the world, and models are always interpreted on that
basis. This is precisely what Allen and Hoekstra have recognised in the field of
ecology:
Narratives are the bottom line in science. Yes, there are hypotheses, predictions, theories
and models, but all of these devices are in the service of achieving compelling narratives.
( …) The end product of science is a story improved by models and made convincing
by predictions (2015: 310).
REFERENCES
Abrams, Daniel M. & Steven H. Strogatz. 2003. Modelling the dynamics of language
death. Nature 424, 900.
Allen, Timothy F. H. & Thomas W. Hoekstra. 2015. Toward a unified ecology, 2nd edn.
New York: Columbia University Press.
Aracil, Lluís V. 1982. Papers de sociolingüística [Sociolinguistic papers]. Barcelona:
La Magrana.
Arthur, W. Brian. 2013. Complexity economics: A different framework for economic
thought. Available at www.santafe.edu/media/workingpapers/13-04-012.pdf.
Ashby, W. Ross. 1956. An introduction to cybernetics. London: Chapman & Hall.
Ashby, W. Ross. 1962. Principles of the self-organizing system. In Heinz Von Foerster
& George W., Jr. Zopf (eds.), Principles of self-organization: Transactions of the
University of Illinois Symposium, 255–278. London: Pergamon Press.
Ball, Philip. 2005. Critical mass: how one thing leads to another. London: Random
House, Arrow Books.
Barth, Fredrik (comp.). 1976. Los grupos étnicos y sus fronteras. México: Fondo de
Cultura Económica.
Bastardas-Boada, Albert. 1995. Language management and language behavior change:
policies and social persistence. Catalan Review 9 (2), 15–38.
Bastardas-Boada, Albert. 1995b. Cap a un enfocament holístic per a la lingüística
[Towards a holistic approach for linguistics]. In Rosa Artigas et alii (eds.), El
significat textual [Textual meaning], 15–20. Barcelona: Generalitat de Catalunya,
Departament de Cultura.
Bastardas-Boada, Albert. 1996. Ecologia de les llengües. Medi, contacte i dinàmica
sociolingüística [Ecology of languages. Sociolinguistic environment, contact and
dynamics]. Barcelona: Proa.
Bastardas-Boada, Albert. 2002. World Language Policy in the Era of Globalization:
Diversity and Intercommunication from the Perspective of ‘Complexity’, Noves
SL. Revista de Sociolingüística, 1–9.
Bastardas-Boada, Albert. 2013. General linguistics and communication sciences: socio-
complexity as an integrative perspective. In Massip-Bonet & Bastardas-Boada
(eds.), Complexity perspectives on language, communication and society. Berlin:
Springer, 151–173.
Bastardas-Boada, Albert. 2014. Towards a complex-figurational socio-linguistics: Some
contributions from physics, ecology and the sciences of complexity, History of the
Human Sciences 27 (3), 55–75. DOI: 10.1177/0952695114534425.
Bastide, Roger. 1972. Antropología aplicada. Buenos Aires: Amorrortu.
Bateson, Gregory. 1972. Steps to an ecology of mind. New York: Ballantine
Books.
Beltran, Francesc S., Salvador Herrando, Doris Ferreres, Marc-Antoni Adell, Violant
Estrader & Marcos Ruiz-Soler. 2009. Forecasting a language shift based on cellular
automata. Journal of Artificial Societies and Social Simulation 12(3), 1–8.
Beltran, Francesc S., Salvador Herrando, Violant Estrader, Doris Ferreres, Marc-Antoni
Adell, & Marcos Ruiz-Soler. 2011. A language shift simulation based on cellu-
lar automata. In Emmanuel G. Blanchard & Danièle Allard (eds.), Handbook of
research on culturally-aware information technology: perspectives and models,
136–151. Hershey / New York: Information Science Reference.
Bernárdez, Enrique. 1995. Teoría y epistemología del texto. Madrid: Cátedra.

Bertalanffy, Ludwig von. 1969. General system theory. New York: George Braziller.
Blumer, Herbert. 1969. Symbolic interactionism: Perspective and method. Englewood
Cliffs, NJ: Prentice-Hall.
Bohm, David. 1987. La totalidad y el orden implicado. [Spanish translation of Whole-
ness and the Implicate Order. London: Routledge & Kegan Paul, 1980]. Barcelona:
Kairós.
Boix-Fuster, Emili & Jaume Farràs. 2013. Is Catalan a medium-sized language commu-
nity too? In F. Xavier Vila (ed.), 157–178.
Bourdieu, Pierre. 1980. Le sens pratique. Paris: Éditions de Minuit.
Bourdieu, Pierre. 1984. Distinction: A social critique of the judgement of taste. Cam-
bridge: Harvard University Press.
Bruner, Jerome. 1990. Acts of meaning. Cambridge, MA: Harvard University Press.
Buckley, Walter. 1967. Sociology and modern systems theory. Upper Saddle River, NJ:
Prentice Hall.
Byrne, David & Gill Callaghan. 2014. Complexity theory and the social sciences. The
state of the art. London: Routledge.
Capra, Fritjof. 1982. The Turning Point. New York: Simon & Schuster.
Capra, Fritjof. 1996. The web of life. A new synthesis of mind and matter. Hammersmith,
London: HarperCollins, 1996.
Capra, Fritjof. 2002. The hidden connections. New York: Doubleday.
Castellani, Brian & Frederic William Hafferty. 2009. Sociology and complexity science.
Berlin: Springer-Verlag.
Castelló Llobet, Xavier. 2010. Collective phenomena in social dynamics: Consen-
sus problems, ordering dynamics and language competition, doctoral dissertation.
Palma de Mallorca: University of the Balearic Islands.
Castelló, Xavier, Federico Vazquez, Víctor M. Eguíluz, Lucía Loureiro-Porto, Maxi
San Miguel, Laetitia Chapel & Guillaume Deffuant. 2011. Viability and resilience
in the dynamics of language competition. In Guillaume Deffuant & Nigel
Gilbert (eds.), Viability and resilience of complex systems, 39–74. Heidelberg:
Springer.
Castelló, Xavier, Lucía Loureiro-Porto & Maxi San Miguel. 2013. Agent-based models
of language competition. International Journal of the Sociology of Language 221,
21–51.
Cilliers, Paul. 1998. Complexity and postmodernism. Understanding complex systems.
London: Routledge.
Coller, Xavier. 2006. Collective identities and failed nationalism. The case of Valencia
in Spain. Pôle Sud 25, 107–136.
Conill, Josep J. 2003. The situation of Valencian as reported in non-institutional soci-
olinguistic research (1998–2002). Noves SL. Revista de Sociolingüística, 1–9.
Corbeil, Jean-Claude. 1983. Élements d’une theorie de la regulation linguistique. In
Édith Bedard & Jacques Maurais (eds.), La norme linguistique, 281–303. Quebec
& Paris: Conseil de la Langue Française.
Elias, Norbert. 1982. Sociología fundamental. [Spanish translation of Was ist Soziolo-
gie? Munich, Juventa Verlag, 1970]. Barcelona: Gedisa.
Elias, Norbert. 1990. La sociedad de los individuos [Spanish translation of Die
Gesellschaft der Individuen, 1987]. Barcelona: Península.
Elias, Norbert. 1991. The symbol theory. London: SAGE.
Emmert-Streib, Frank. 2010. Statistic complexity: Combining Kolmogorov complexity

with an ensemble approach. PLoS ONE 5(8), 1.
Epstein, Joshua M. 2006. Generative social science: Studies in agent-based computa-
tional modeling. Princeton, NJ: Princeton University Press.
Fantini, Alvino E. 1982. La adquisición del lenguaje en un niño bilingüe, Barcelona:
Herder.
Flos, Jordi (ed.). 1995. Ordre i caos en ecología [Order and chaos in ecology].
Barcelona: Publicacions de la Universitat de Barcelona.
Gell-Mann, Murray. 1994. The Quark and the Jaguar: Adventures in the Simple and the
Complex. New York: Henry Holt and Company.
Hamers, Josiane F. & Michel H. A. Blanc. 2000. Bilinguality and bilingualism. Cam-
Hamers, Josiane. F. 2004. A sociocognitive model of bilingual development.
Journal of Language and Social Psychology 23 (1), 70–98, DOI: 10.1177/
0261927X03261615.
Gershenson, Carlos & Francis Heylighen. 2005. How can we think the complex? In
Richardson, K. (ed.). Managing organizational complexity: Philosophy, theory and
application, vol. 3, 47–62. Charlotte, NC: Information Age Publishing.
Haugen, Einar. 1972. The Ecology of Language. In: Anwar S. Dil (ed.), The Ecology of
Language, 325–339. Stanford: Stanford University Press.
Heylighen, Francis. 2001. The science of self-organization and adaptivity. In L. Douglas
Kiel (ed.), Knowledge management, organizational intelligence and learning, and
complexity. Encyclopedia of life supports systems. Oxford: Eolss Publishers.
Heylighen, Francis, Paul Cilliers & Carlos Gershenson. 2007. Complexity and philos-
ophy. In Jan Bogg & Robert Geyher (eds.), Complexity, science and society, 117–
134. Oxford: Radcliffe Publications.
Holland, John H. 1996. Hidden order. How adaptation builds complexity. New York:
Basic Books.
Holland, John H. 1998. Emergence. From chaos to order. Cambridge, MA: Perseus
Publishing.
Holland, John H. 2005. Language acquisition as a complex adaptive System. In James
W. Minett & William S-Y. Wang (eds.), Language acquisition, change, and emer-
gence, 411–436. Hong Kong: City University Hong Kong Press.
Johnson, Steven. 2002. Emergence. London: Penguin Books.
Jörg, Ton. 2011. New thinking in complexity for the social sciences and humanities.
Heidelberg: Springer.
Mackey, William F. 1979. Toward an ecology of language contact. In William F. Mackey
& Jacob Ornstein (eds.), Sociolinguistic studies in language contact, 453–460. La
Haya: Mouton.
Mackey, William F. 1980. The ecology of language shift. In Peter Hans Nelde (ed.),
Sprachkontakt und Sprachkonflikt, 35–41. Wiesbaden: Franz Steiner Verlag.
Mackey, William F. 1994. La ecología de las sociedades plurilingües. In Albert Bastar-
das Boada & Emili Boix-Fuster (eds.), ?‘Un estado, una lengua? La organización
política de la diversidad lingüística, 25–54. Barcelona: Octaedro.
Malaina, Álvaro. 2012. Le paradigme de la complexité et la sociologie. Possibilité et
limites d’une sociologie complexe. Paris: L’Harmattan.
Margalef, Ramon. 1991. Teoría de los sistemas ecológicos. Barcelona: Publicacions de
la Universitat de Barcelona.
Massip-Bonet, Àngels. 2013. Language as a complex adaptive system: Towards an inte-

grative lingüístics. In Massip-Bonet & Bastardas-Boada (eds.), 35–60.
Massip-Bonet, Àngels & Albert Bastardas-Boada (eds.). 2013. Complexity perspectives
on language, communication and society. Heidelberg: Springer.
Maturana, Humberto. 2002. Autopoiesis, structural coupling and cognition. Cybernetics
& Human Knowing 9(3–4), 5–34.
Maturana, Humberto & Francisco J. Varela. 1996. El árbol del conocimiento. Las bases
biológicas del conocimiento humano. Madrid: Editorial Debate.
McWhorter, John H. 2011. Linguistic simplicity and complexity. Why do languages
undress? Berlin: De Gruyter Mouton.
Mead, George H. 1934. Mind, self and society (Charles W. Morris, ed.). Chicago: The
University of Chicago Press.
Mitchell, Melanie. 2009. Complexity: A guided tour. Oxford: Oxford University
Press.
Montoya Abat, Brauli, & Antoni Mas i Miralles. 2012. Language teaching and family
linguistic transmission: Two correlative factors in the Valencian Region (Catalan
vs. Spanish)? Sociolinguistic Studies 6(1), 45–63.
Morin, Edgar. 1980. La méthode. 2. La vie de la vie. Paris: Éditions du Seuil.
Morin, Edgar. 1986. La méthode. 3. La connaissance de la connaissance. París: Éditions
du Seuil.
Morin, Edgar. 1991. La méthode. 4. Les idées. Leur habitat, leur vie, leurs moeurs, leur
organisation. Paris: Éditions du Seuil.
Morin, Edgar. 1992. Introduction à la pensée complexe. Paris: ESF.
Morin, Edgar. 1994. La complexité humaine. Paris: Flammarion.
Morin, Edgar. 1999. La tête bien faite. Repenser la réforme, réformer la pensée. París:
Éditions du Seuil.
Morin, Edgar. 2001. La méthode. 5. L’humanité de l’humanite. L’identité humaine.
Paris: Éditions du Seuil.
Morin, Edgar. 2007. Restricted complexity, general complexity. (Translated from
French by Carlos Gershenson.) In Carlos Gershenson, Diederik Aerts, & Bruce
Edmonds (eds.), World views, science and us. Philosophy and complexity, 5–29.
Singapore: World Scientific.
Morin, Edgar. 2008. On complexity. Advances in systems theory, complexity, and the
human sciences. New York: Hampton Press.
Mufwene, Salikoko S. 2001. The ecology of language evolution. Cambridge: Cambridge
University Press.
Mufwene, Salikoko S. 2008. Language evolution. Contact, competition and change.
New York: Continuum International.
Mufwene, Salikoko S. 2013. The emergence of complexity in language: An evolution-
ary perspective. In Massip-Bonet & Bastardas-Boada (eds.), 197–218. Heidelberg:
Springer.
Munné, Frederic. 2013. The fuzzy complexity of language. In Massip-Bonet &
Bastardas-Boada (eds.), 175–196.
Newman, Marc, Albert-László Barabási & Duncan J. Watts. 2006. The structure and
dynamics of networks. Princeton, NJ: Princeton University Press.
Nichols, Johanna. 2009. Linguistic complexity: A comprehensive definition and survey.
In Geoffrey Sampson, David Gil & Peter Trudgill (eds.), Language complexity as
an evolving variable, 110–125. Oxford: Oxford University Press.
Ninyoles, Rafael L. 1978. Conflicte linguistic valencià [Valencian linguistic conflict].

València: Eliseu Climent ed.
Nisbet, Robert. 1977. The social bond. New York: Random House.
Popper, Karl & Konrad Lorenz. 1992. El porvenir está abierto. Barcelona: Tusquets Ed.
(Spanish translation of Die Zukunft ist offen. Munic: E. Piper GmbH, 1985.)
Querol, Ernest. 1990. El procés de substitució lingüística: la comarca dels Ports com a
exemple [Language shift process: Els Ports county as an example]. Miscel·lània
89, 87–196. València: Generalitat Valenciana.
Rodríguez Zoya, Leonardo G. 2013. Le modèle épistémologique de la pensée complexe.
Analyse critique de la construction de la connaissance en systèmes complexes.
Unpublished Ph. D. Thesis, Université de Toulouse.
Roggero, Pascal. 2013. Para una sociología según El método. In Ruiz Ballesteros &
Solana Ruiz (eds.), 103–123.
Ruiz Ballesteros, Esteban. 2013. Hacia la operativización de la complejidad en ciencias
sociales. In Ruiz Ballesteros & Solana Ruiz (eds), 137–172.
Ruiz Ballesteros, Esteban & José Luis Solana Ruiz (eds.). 2013. Complejidad y ciencias
sociales. Seville: Universidad Internacional de Andalucía.
Sampson, Geoffrey, David Gil & Peter Trudgill (eds.). 2009. Language complexity as
an evolving variable. Oxford: Oxford University Press.
Schutz, Alfred. 1974. El problema de la realidad social [Spanish translation of Col-
lected papers: I. The problem of social reality. The Hague: Martinus Nijhoff, 1962].
Buenos Aires: Amorrortu.
Schutz, Alfred & Thomas Luckmann. 1977. Las estructuras del mundo de la vida.
Buenos Aires: Amorrortu. (Spanish translation of The Structures of the Life-World.
Evanston, IL: Northwestern University Press, 1973).
Searle, John R. 1985. Mentes, cerebros y ciencia. Madrid: Ed. Cátedra. (Spanish trans-
lation of Minds, Brains and Science. The 1984 Reith Lectures, Cambridge, MA:
Harvard University Press.)
Serrano, Sebastià. 1993. Comunicació, societat i llenguatge [Communication, society
and language]. Barcelona: Empúries.
Siguan, Miquel. 1993. Multilingual Spain. Amsterdam: Swets-Zeitlinger.
Simon, Herbert A. 1962. The architecture of complexity. Proceedings of the American
Philosophical Society 106 (6), 467–482.
Solana Ruiz, José Luis. 2013. El concepto de complejidad y su constelación semántica.
In Esteban Ruiz Ballesteros & José Luis Solana Ruiz (eds.), Complejidad y ciencias
sociales, 19–101. Seville: Universidad Internacional de Andalucía.
Solé, Ricard. 2009. Redes complejas. Del genoma a internet. Barcelona: Editorial Tus-
quets.
Solé, Ricard V. & Jordi Bascompte. 2006. Self-organization in complex systems. Prince-
ton, NJ: Princeton University Press.
Steels, Luc & Rodney Brooks (eds.). 1995. The artificial life route to artificial intelli-
gence. New Haven, CT: Lawrence Earlbaum.
Strubell, Miquel & Emili Boix-Fuster (eds.). 2011. Democratic policies for language
revitalisation: the case of Catalan (Palgrave studies in minority languages and
communities). Basingstoke: Palgrave Macmillan.
Terborg, Roland & Laura García-Landa. 2013. The ecology of pressures: Towards a
tool to analyze the complex process of language shift and maintenance. In Massip-
Bonet & Bastardas-Boada (eds.), 219–239.
The ‘Five Graces’ Group: Beckner, Clay, Richard Blythe, Joan Bybee, Morten H. Chris-
tiansen, William Croft, Nick C. Ellis, John Holland, Jinyun Ke, Diane Larsen-
Freeman & Tom Schoenemann. 2009. Language is a complex adaptive system.
Language Learning 59, 1–26.
Van Dijk, Teun V. 2010. Discourse and context. A sociocognitive approach. Cambridge:
Cambridge University Press.
Varela, Francisco J., Evan Thompson & Eleanor Rosch. 1992. The Embodied Mind.
Cognitive Science and Human Experience. Cambridge, MA: The MIT Press.
Vila, F. Xavier (ed.). 2013. Survival and development of language communities.
Prospects and challenges. Clevedon, UK: Multilingual Matters.
Wagensberg, Jorge. 1985. Ideas sobre la complejidad del mundo. Barcelona: Tusquets
editores.
Wallerstein, Immamnuel (ed.). 1996. Ouvrir les sciences sociales. Rapport de la Com-
mission Gulbenkian. Paris: Descartes & Cie.
Watzlawick, Paul, John H. Weakland & Richard Fisch. 1974. Change. Principles of
Problem Formation and Problem Resolution. New York: Norton.
Weinreich, Uriel. 1968. Languages in contact. The Hague: Mouton.
Wiener, Norbert. 1948. Cybernetics or control and communication in the animal and
the machine. Cambridge & Massachusetts: The MIT Press.
Wolfram, Stephen. 1983. Statistical mechanics of cellular automata. Reviews of Modern
Physics 55(3), 601–644.
Wolfram, Stephen. 2002. A new kind of science. Champaign: Wolfram Media.
Index
Abrams, Daniel M., 190, 196–212, 220 Bolhuis, Johan J., 87, 88, 89, 91
acquisition Bolinger, Dwight, 124
language, 8, 12, 23, 34, 55, 91, 110, Bonabeau, Eric., 39, 168
124–125, 136, 160, 165–183, 212 Boone, Elizabeth Hill, 110
speech, 23, 165–183 bootstrapping, 33, 90
actional basis, 107, 126, 130 borrowing, 59, 60, 61, 223
Allen, Timothy F. H., 238 Bourdieu, Pierre, 233
Anderson, Philip, 188 brain
ape, 74–88, 91, 93, 104, 128 network, 73, 74
Apel, Willi, 119 size, 74, 77
approach temporal lobe, 71, 73, 78
analytical, 218 Bratman, Michael, 104, 127
biological-functional, 23, 171–172 Brighton, Henry, 33, 53, 92
reductionist, 136, 137, 141, 160, 218, 221 Buccellati, Giorgio, 113
Arthur, W. Brian, 222 Burkart, Judith Maria, 74
Australian languages, 153 Bybee, Joan, 1, 10, 13, 53, 67, 93, 94, 102,
autopoiesis, 219 161, 195, 222
Byrne, David, 222, 238
Bagley, Robert W., 113, 114
Baines, John, 112, 115 Callaghan, Gill, 222, 238
Ball, Philip, 187, 190, 194, 198, 220 Camazine, Scott, 39, 168
Barabási, Albert-Lászlo, 192, 193–194, 219, capacity
220 cognitive, 11, 12, 13, 15, 20, 23, 82, 102,
Barsalou, Lawrence W., 72, 73 125, 170
Bastardas-Boada, Albert, 1, 3, 10, 13, 25 language, 12, 23, 51, 101, 121, 165–166,
Bates, Elizabeth, 90, 91 167, 169. See also production system
BEAGLE method, 84–87 neural, 23, 166, 169, 172, 173
Beckner, Clay, 1, 10, 13, 53, 67, 93, 195, 222 perceptual, 169, 170, 172
Bedore, Lisa M., 166, 168 Capra, Fritjof, 219, 224
Bell, Allen, 174 Cassidy, Kimberly, 90
Berwick, Robert C., 52, 87, 88, 89, 91 Castelló, Xavier, 195, 196, 198–209, 211, 220
Beuls, Katrien, 18–19, 20, 36, 42, 44 Catalonia, 232, 235–236. See also Spanish
bilingualisation, 236 language
bilingualism, 23–24, 73, 175, 196–212, 220, causality, 34, 35, 73, 110, 137, 222, 223
223, 229, 230, 235, 236 Chafe, Wallace, 101
biocultural evolutionary process, 68 Changizi, Mark A., 74–77
biological capacity, 165, 166 chaos, 2, 7, 12, 14, 70, 135, 219
Blumer, Herbert, 228 Chater, Nick, 52–53, 68, 94
Blythe, Richard, 1, 9, 10, 13, 53, 67, 93, 195, Cheney, Dorothy L., 79, 88
196, 222 chimpanzee, 21, 50, 74, 75, 77, 79, 83
Boccaletti, Stefano, 194 Chomsky, Noam, 12, 49, 50, 51, 52, 55, 68,
Bohm, David, 224 87, 88, 89, 91, 138, 165, 167–168, 195
245
246 Index
Christiansen, Morten H., 1, 10, 13, 52–53, 67, minimum description length, 5, 138
68, 93, 94, 195, 222 morphosyntactic, 102
Clark, Andy, 170 orchestral, 225
Clark, Herbert H., 104–107 population-level, 30, 33, 38–40
class processing, 30, 32, 38
upper, 233 relative, 11–12, 48
working, 233 restricted, 219–225, 237–238
Clements, George N., 55, 58, 137, 140, science, 1–4, 7, 9, 11, 12, 14, 20, 21, 22–23,
141–142, 144, 159, 160 166, 168–169, 170, 182, 187–188,
code-switching, 205, 223. See also 219–220
bilingualism serial, 182
cognition social, 190, 222
domain-general, 51–53, 58, 61, 62, 93 social-cognitive, 21, 101, 103
grounded, 72–73 theory, 2, 18, 22, 69, 135–136, 138, 139,
language-specific, 49, 51–53, 58, 61, 62 160, 168, 169, 219, 222. See also
social, 104–110, 121, 125, 126–131, complexity, science
223–225, 230, 233 usage, 15
cognitive factor, 51, 53, 56, 58 compositionality, 17, 92, 102
combinability, 101, 111 computational properties, 168
combinatorial structure. See structure computer simulation. See simulation,
common ground, 21, 104–108, 117–122, computer
126–131 conceptualization, 20–21, 67, 69, 70, 71–74,
communication 75, 78–94, 102, 110, 125
domain-specific, 102 constraint, 4, 7, 8, 9, 10, 11, 33, 48, 49, 50,
general-purpose, 22, 102–103, 117, 122, 53–55, 69, 70, 91, 93, 136–137, 139, 146,
124 159, 167, 168, 182, 183, 230, 231, 232,
institutionalized, 234 237
communicative convention, 21, 44, 46, 85, 89, 93, 105–108,
needs, 13, 14, 15, 44 110, 113, 126–127, 128, 130,
pressure, 7, 10, 13, 17, 23 136
technology, 12 co-occurrence
competence, 12, 13, 168, 223, 226–227, pattern, consonant-vowel, 174–178
228–229, 230 phoneme co-occurrence, 4, 141–144,
complex 145–155, 159–160
adaptive system, 1, 10, 14, 17, 67–68, 70, Cooper, Jerrold S., 113–114, 115, 124
93, 94, 195, 220, 221, 224, 236, 238 cooperation, 3, 6, 22, 104, 225
network, 24, 138, 192, 193–196, 198–201, coordination, 2, 22, 56, 57, 104–108, 122, 124,
205–212 126–131, 171
complexity problem, 105–106
absolute, 11 cortical area, 71, 74, 75–78
algorithmic, 5, 11 creole, 7, 9, 16, 18, 19, 35
bit, 4, 16 Croft, William, 1, 9, 10, 13, 14, 20, 21–22,
conceptual, 67, 68, 70–71, 73, 74–78, 53, 67, 93, 101, 102, 123, 130, 195, 196,
83–94, 102 222
effective, 5, 13, 138 cybernetics, 22, 135, 219
form, 30, 31, 32, 41, 42–43
general, 219–225, 237–238 Dale, Rick, 9, 161
hidden, 14 Damasio, Antonio R., 72, 73, 78, 170
interactional, 2, 6 Damasio, Hanna, 72, 73, 78
inventory, 30, 31, 44 Damerow, Peter, 111, 113, 114, 124
Kolmogorov, 5, 138 Davis, Barbara L., 23, 166, 168, 170, 172, 173,
learnability, 11 174–178, 179, 181–182, 183
learning, 30, 32–33, 40–41 de Boer, Bart, 3, 8, 20, 53, 54, 55, 58–60,
linguistic, 4, 5, 6, 8, 10, 11, 12, 13, 14, 16, 136
20, 21, 23, 25, 48, 68, 101, 103 De Landa, Manuel, 35
Index 247
de Waal, Frans, 79, 88 Fauconnier, Gilles, 102

Deacon, Terrence William, 85, 87 feature
Deaner, Robert O., 74 economy, 50, 55–56, 58, 61, 137, 141–142,
DeGraff, Michel, 4, 7 152, 156, 159, 160
Deneubourg, Jean-Louis, 39, 168 maximal use of available features, 137, 152,
Descartes, René, 166–167 159
Diessel, Holger, 123, 129 Fishman, Joshua A., 196
Dobzhansky, Theodoszius, 49 Fitch, W. Tecumseh, 50, 52, 53
Dorogovtsev, Sergey, 138, 194 Five Graces Group, The, 1, 10, 13, 53, 67, 93,
duality of patterning, 101, 102–103, 124, 125, 195, 222
130 frame-content theory, 175
dynamic systems theory, 170–171, 182 Franks, Nigel, 39, 168
Friederici, Angela D., 87, 88, 89, 91
ecology, 3, 9, 21, 25, 219, 222, 238 fuzzy logic, 221, 223
language ecology, 8–9, 13, 15, 17, 196
ecosystem, 24, 189, 193, 224, 231, 234, 235, Gahl, Susanne, 91
237 Galantucci, Bruno, 35, 50, 56–58, 62
Edelman, Gerald M., 170 Ganis, Giorgio, 73
Elias, Norbert, 225, 231, 237 Garcia Casademont, Emilia, 38
Ellis, Nick C., 1, 10, 13, 53, 67, 93, 195, 222 García-Landa, Laura, 235
embodiment theory, 34, 165, 169–170, 171, Gardner, Beatrix T., 79
172, 175, 182–183 Gardner, R. Allen, 79
emergence, 2, 3, 7–8, 9, 10, 12–13, 14, 15, Gell-Mann, Murray, 5, 13, 138, 219, 221
17–23, 34–36, 38, 50, 52, 54–62, 67–68, Gershenson, Carlos, 2, 219, 238
70, 87–94, 104, 107–110, 112, 113, Gibson, Kathleen R., 74
116–117, 119, 124, 126–131, 135, 136, Gildea, Patricia M., 84
139, 141, 166, 169, 172, 173, 183, Gleitman, Lila, 90–91
188–189, 201, 205, 219, 223, 225, 237 Goldberg, Adele E., 48, 91
emergent pattern, 2, 3, 9, 93, 224 Gong, Tao, 3, 92, 138, 191, 194, 195–196,
emergent phenomenon, 10, 17, 20, 23, 187, 197, 211, 212
190 Goodman, Judith C., 90, 91
emergentism, 18, 219 grammatical system, 18, 21, 30, 31, 32, 36, 41,
emotion, 70, 73, 190, 223, 236, 237 88, 89–90, 92, 93, 94, 101, 116, 122, 124,
Englund, Robert, 111–112, 113–114, 124 125, 130, 219
epistemology, 11, 166, 220, 221, 222, 238 grammaticalization, 8, 19, 22, 41, 123
equilibrium, 3, 6, 7, 10, 135, 140, 170, 190, Granovetter, Mark S., 192, 193, 195
224 Greenfield, Patricia M., 91
essence, 167–168, 197 Grenoble, Lenore, 196
ethnic consciousness, 232 Grice, Paul, 105
Evans, Nicholas, 51–53, 68, 139 Griffiths, Thomas L., 51, 56, 58
Everett, Daniel L., 101 group
evolution, 2, 10, 12, 14, 18, 20, 21, 34, 36, 49, minoritized, 224, 233
52, 53, 54, 67, 78, 89, 92, 94, 103, 106, social, 25, 48, 50, 56, 78, 79, 81, 110, 126,
108–131, 190, 221, 223, 224, 232 128–130, 193, 195, 203, 204, 205, 220,
biological, 20–21, 50, 52–54, 62, 68 224, 225, 231–237
co-evolution, 53, 67, 193 Guest, Ann Hutchinson, 120–121
cultural, 20–21, 22, 50, 51–52, 53–54, Gullah, 8
55–62, 67, 68, 70, 89–90, 92–94, 195
language, 8, 9, 13, 18, 19, 20, 21, 23, 24, habitus, 228, 233
35–36, 42, 44, 56, 61, 67, 70, 75, 78, 87, Haiman, John, 88, 102
92–94, 101, 102, 110, 122–131, 141, 165, Halle, Morris, 54–55, 138, 165, 167–168
195, 221, 223 Hawkins, John A., 1, 13–14, 16, 19, 31
exaptation, 12, 13, 21–22, 32, 52, 123, 124 Heine, Bernd, 8, 35
expansionism, 232 helpfulness, 108, 127, 128
experimental cultural learning. See learning Heylighen, Francis, 219, 238
248 Index
Hockett, Charles F., 22, 101 Langacker, Ronald W., 88, 102
Hoekstra, Thomas W., 238 language
holism, 90, 137, 218, 222, 235, 237, 238 as technology, 14
Holland, John, 1, 10, 13, 53, 67, 93, 168, 195, choice, 23–24, 187–188, 189, 191, 194,
219, 220, 221, 222, 224, 225 196–212, 231, 235
hologram, 223 coexistence, 24, 196–212. See also
homunculus, 77 bilingualism
Hooper, Joan B., 174 competition, 9, 12, 23–24, 195–212
Houston, Stephen D., 115 death, 196–212, 224
human infant, 13, 23, 56, 110, 165–183 development, 7, 9, 15, 17, 18, 20, 21, 23, 68,
88, 90, 91, 92, 94, 108, 110, 111, 112,
icon, 21, 85, 102, 111, 113–117, 119, 116, 119, 129, 166, 169, 170, 181, 182,
120–121, 123, 125, 129 190, 220, 223, 226, 232, 235, 237
iconic gesture, 123, 129 diversity, 33, 51, 139–140, 167, 225, 230,
iconicity, 102, 115, 123 231
identity dynamics, 4, 8, 18, 24, 33, 36, 67, 94, 171,
collective, 226 195–196, 201, 205, 209, 211–212, 237
ethnic, 203, 232 extinction, 196–212, 221, 224
idiolect, 10, 11, 19, 22, 33, 136 game, 34–38, 56
index, 85–87, 114, 116, 123, 125, 129 maintenance, 196–212, 234
interaction minoritized, 203–204, 207–209
between components, 12, 16 official, 234–236
constructivist, 171 policy, 25, 223, 234, 236
functional, 182 survival, 24, 187, 190, 196–212, 224
social, 10, 20, 21, 24, 103–104, 108, 128, universal, 1, 45, 48, 51–52, 53, 68, 88–89,
169, 192, 225, 226, 228, 229–231 93, 135, 167, 168, 174, 178
interdisciplinarity, 3, 209, 225 variation, 1, 10, 14, 19, 20, 22, 31, 33,
interpretive procedure, 228–230 38–40, 44, 62, 136
intersyllabic property, 177–178 variety, 7, 9, 19, 21, 209, 224, 226–227,
invisible hand, 2 228–229, 230–231, 234–235, 237
Isler, Karin, 74 Lass, Roger, 123
lateral inhibition, 33, 38–40, 44
Jackendoff, Ray, 68, 70, 89 learning
jaw oscillation, 175, 176, 178, 183 efficiency, 41, 42
Jerison, Harry J., 74 experimental cultural, 50, 54, 56–62
Johnson, Mark, 73, 169 iterated, 56–62, 92, 102
joint Lewis, David, 105–106
action, 21, 22, 103–108, 110, 117, 121, 122, lexigram, 79
125–131, 231 Lieven, Elena V. M., 124
attention, 106, 110, 126, 127, 130 Liljencrants, Johan, 55, 152
salience, 106–108, 126–127, 129 Lindblom, Björn, 55, 135, 136, 152, 170,
Jones, Michael N., 84 174
linear generative phonology, 167
Kanzi, 79–87, 88, 91 Lupyan, Gary, 9, 161
Kauffman, Stuart, 168
Ke, Jinyun, 1, 3, 7, 10, 13, 53, 67, 93, 138, Mackey, William F., 224
191, 194, 195–196, 197, 211, 212, 222 MacNeilage, Peter, 12, 135, 170, 172,
Kern, Sophie, 173, 174, 176, 178 174–178, 181–182, 183
Kilmer, Anne Draffkorn, 117 Maddieson, Ian, 6, 48, 136, 138, 140, 153,
Kinney, Ashlyn, 174 173, 174, 175, 181
Kirby, Simon, 3, 33, 50, 53–54, 56–59, 62, 92, Malaina, Álvaro, 222
102 Margalef, Ramon, 219, 224
Kuteva, Tanja, 8, 35 Marshack, Alexander, 111
Massip-Bonet, Àngels, 1, 3, 10, 13, 222
Lakoff, George, 73, 169 mathematics, 22, 121, 194, 222
Index 249
Maturana, Humberto, 219, 226 musical, 22, 109, 110, 117–119, 120–122
Matyear, Christine, 174, 175, 181–182 stick-figure, 120
maximal perceptual distinctiveness, 55, 170, Nowak, Andrzej, 203, 210
183 number, 110, 111–112, 117, 121, 123
McWhorter, John H., 1, 7, 16 system, 111–112
Mead, George H., 238
Menninger, Karl, 112 Ohala, John J., 55, 58, 137, 159
Milroy, Lesley, 194 ontogeny, 8, 18, 124, 165–166, 171–173,
Minett, James W., 7, 13, 196–198, 210, 182–183. See also phylogeny
211 open system, 31, 139, 170, 172
minimal production difficulty, 170 optimality theory, 138, 167
model, 8, 11, 22, 34, 35, 36, 37, 49, 51, 55, 56, order, 2, 3, 4, 5, 7, 34, 36, 37, 48, 50, 53, 54,
60, 61, 68, 69, 70, 74, 76, 88, 91, 92, 94, 59, 61, 69, 75, 79, 84, 92, 104, 105,
102, 110, 136, 141, 159, 160, 168, 170, 112, 114, 120, 128, 130, 136, 141, 142,
171, 187, 190, 191, 192, 193, 195, 196, 145, 147, 148, 151, 155, 167, 170, 174,
197, 198, 199, 203, 206, 207, 210, 211, 205, 212, 220, 221, 222, 224, 228, 233,
212, 220, 221, 222, 238 238
Abrams-Strogatz, 196–212 explicate, 223
agent-based, 13, 18, 33–46, 55, 92–94, 136, implicate, 223
141, 191–193, 195–212, 220–221 organisation, 5, 7, 12, 13, 50, 55, 61, 135, 136,
computational, 35, 56–57, 94, 101, 168, 138, 139, 151, 155, 160, 172, 178, 188,
187, 192, 195–212, 220–221 189, 218, 219, 220, 221, 228, 229, 231,
modeling, 3, 8, 13, 18, 23–24, 33–46, 56, 131, 233, 237
136, 170, 187–212, 219, 220–221, 238 Oyama, Susan, 171
module, 23, 169, 194
language, 1, 6–7, 12, 16–17, 19, 24, 168 pairwise interaction, 5, 141, 144, 149, 155,
monkey, 73, 78–79, 88 159–160
morality perceptual basis, 126
great ape, 128 Pettersson, John Sören, 111–112
norm-based, 128 PhonBank, 175, 179
second-person, 128–129 phoneme inventory. See phonological,
Morin, Edgar, 219, 220, 221, 225, 226, 231 inventory
Mufwene, Salikoko, 3, 7–10, 13–15, 17, 19, phonological
20, 21, 22, 31, 35, 44, 50, 53, 102, 123, feature, 23, 54–56, 58, 61, 137, 138–139,
130, 196, 203, 209, 210, 222 140–143, 147–160, 201
multimodality, 115, 116, 117, 124, 125 inventory, 7, 8, 14, 20, 23, 48–49, 135–161,
173
nature versus nurture, 51, 166–168 phylogeny, 7–8, 9, 12, 13, 15, 17–18, 21, 23,
neoteny, 166 124, 165–167, 171–172, 182–183
Nettle, Daniel, 191, 195–196, 203, 210 pidgin, 17, 18
network pigeon, 78–79
node degree, 194, 199, 206 political power, 25, 225, 232, 233–235
random, 192, 194, 205–206, 210–211 precedent, 106, 108, 126–127, 128, 129,
regular lattice, 198–201, 205–209, 210–211 130
small-world, 24, 195, 198–201, 205–209, prefrontal lobe, 73, 77, 78. See also brain
210, 211 pre-language, 101, 103, 110, 122–125, 126,
neume, 118–119 129
staffless, 118–119 prestige, 197, 198, 200, 202–204, 205–206,
Newmeyer, Frederick J., 1, 5, 14, 19 210–211, 212, 233
Niger-Congo languages, 153–154 Prigogine, Ilya, 168
Nissen, Hans, 111, 113, 114, 124 Prince, Allen, 138, 165, 167, 171
non-linearity, 23, 139, 140, 159, 160 processing
notation, 109, 110–124 cognitive, 14, 30, 32, 228
dance, 110, 120–124 language, 30, 31, 35, 37, 78
Franconian, 119 neural, 71–78
250 Index
production system, 23, 169, 170, 171, 172 function, 182, 195, 218, 224
psychological reality, 168 institution, 93, 128
meaning, 228–229, 237
recursion, 52, 101, 102, 223, 231 Solana Ruiz, José Luis, 222
reductionism, 188 Spanish language, 16, 175, 203, 204, 205,
regular lattice, 24, 203 235–236
representation, 21, 35, 72–73, 77, 79, 84, 102, speech
109, 111–124, 136, 138, 167, 193, 194, community, 10, 103, 108, 123, 182
197, 211, 222, 227, 229, 231, 233, 236, corpus, 84
238 stability, 39, 42, 52, 53, 62, 129, 170,
representational art, 109 173–179, 183, 193, 221, 224
rhythm, 117–119, 122, 172, 175, 176, 178, statistical control, 144–145
183, 192 Stauder, Andréas, 115
Rissanen, Jorma, 5 Steels, Luc, 3, 8, 14, 18–19, 20, 31, 33–35, 36,
Roggero, Pascal, 222, 238 38, 44–45, 53, 56, 92, 136, 139, 195,
Romaine, Suzanne, 196 220
Rosch, Eleanor, 169, 226 Stengers, Isabelle, 168
Rudman, Peter Strom, 111, 112 Strogatz, Steven Henry, 2, 190, 194, 196–212,
Ruiz Ballesteros, Esteban, 222 220
rule, 4, 9, 10, 11, 14, 15, 25, 33, 58, 61, 88–90, structure
107, 191–192, 204, 219, 220–221 argument, 90–91
combinatorial, 58, 61
San Miguel, Maxi, 3, 6, 23–24, 188–189, 191, conceptual, 21, 87–88, 102–103, 121
192, 193, 195, 196, 197, 198–209, phonotactic, 58
211–212, 220 structuralism, 135
Saussure, Ferdinand de, 1, 7, 10, 135, 167, 195 syntactic, 8, 30, 32–33, 90–91, 102, 105,
scale, 39, 53, 109, 128, 129, 137, 140, 188, 123
195, 233 syllable, open, 174, 182
time, 8, 53–54, 165, 166 symbol, 40, 41, 57, 85–87, 102, 111, 112, 114,
Schelling, Thomas, 188, 191 118, 120, 122, 123–124, 125, 129, 168,
Schmandt-Besserat, Denise, 113 228, 233
Schutz, Alfred, 228 manipulation, 168
Scott-Phillips, Thomas C., 50, 56, 58, 62, 168 symbolic notation, 114, 120
selectional pressure, 102–103, 109–110, 124, synchronisation, 2, 136
127–128 syntactic embedding, 102
self-organisation, 2, 5, 7, 12, 39, 50, 55, 61,
135–136, 188, 189, 218, 219, 221, 223, tablet
226, 229, 233 numerical, 113
semasiographic system, 21–22, 103, 108–125, numero-ideographic, 113, 115, 116
129 protocuneiform, 111, 113–114, 115
shared Taruskin, Richard, 118–119
expertise, 107–108 Terborg, Roland, 235
practice, 107–108, 126 Thelen, Esther, 170–171, 172, 174
signaling system, 130 Theraulaz, Guy, 39, 168
simulation, 73, 93, 94, 191 Thompson, Evan, 169, 226
computer, 34–46, 55, 56–57, 60, 92, 136, Tinbergen, Nicolas, 165–166, 172, 182–183
192, 195–212, 220–221 Toivonen, Riitta, 194, 196, 205, 210, 211
Smith, Kenny, 33, 50, 53, 56–58, 62, 92 Tomasello, Michael, 48, 51, 52, 90, 91,
Smolensky, Paul, 138, 165, 167, 171 104, 106, 110, 124, 126, 127–128,
Sneyd, James, 39, 168 129
social transdisciplinarity, 223
context, 10, 13, 15, 21, 67, 78, 79, 84, 87,
88, 90, 93, 103, 104, 171, 172, 173, Universal Grammar, 52, 68, 88–91
226–227, 229, 235–237 UPSID, 48–49, 139–160
Index 251
Valencian community, 232, 236 Wang, William Shi-Yuan, 3, 7, 13, 77, 138,
Van Trijp, Remi, 31, 36, 45 191, 194, 195–198, 210, 211–212
Varela, Francisco J., 169, 219, 226 Watts, Duncan, 193, 194, 212, 219, 220
Verhoef, Tessa, 54, 58–60 Weinreich, Uriel, 9, 196, 225, 227
Vihman, Marilyn May, 124, 174 Wenger, Étienne, 107
volatility, 197, 198, 200, 202–205, 206, West, Martin Litchfield, 117–118
210–211, 212 Wolfram, Stephen, 220
Wagensberg, Jorge, 219 Zuidema, Willem, 3, 58, 59, 60, 136

Salikoko S Mufwene - Christophe Coupé - François Pellegrino - Complexity in Language - Developmental and Evolutionary Perspectives-Cambridge University Press (2017)

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Salikoko S Mufwene - Christophe Coupé - François Pellegrino - Complexity in Language - Developmental and Evolutionary Perspectives-Cambridge University Press (2017)

Încărcat de

Drepturi de autor:

Formate disponibile

Complexity in Language

Developmental and Evolutionary Perspectives

The question of complexity, as in what makes one language more complex

The contributors include specialists of language acquisition, evolution and

salikoko s. mufwene is Professor of Linguistics and a member of the

Cambridge Approaches to Language Contact is an interdisciplinary series bring-

Cambridge University Press is part of the University of Cambridge.

This publication is in copyright. Subject to statutory exception

Figures page vii

1 Complexity in Language: A Multifaceted Phenomenon 1

8 Language Choice in a Multilingual Society: A View from

2.1 Results of an agent-based simulation page 39

8.2 Square regular lattice of degree k = 4 199

3.1 Languages from UPSID451 (Maddieson, 1984; Maddieson &

6.11 Estimation of pair interactions at the feature level 155

katrien beuls, AI Laboratory, Vrije Universiteit Brussel

Salikoko S. Mufwene, Christophe Coupé,

1.1 Linguistics and the Science of Complexity

(6) Finally, complexity is the peculiarity of emergent patterns in a system in

1.2 Components, Structures, and Domains in Linguistic Complexity

Rather than focusing on (mostly pairwise) interactions between elements,

structure. By contrast, in the case of languages, different modules and levels

entails some cross-linguistic comparison of phonological inventories, morpho-

1.3 From Static to Dynamic Linguistic Systems

to this volume, especially Bart de Boer, P. Thomas Schoenemann, and Luc

1.4 Complexity and Language Ecology

complexity at odds with epistemological models of complex systems in the

A last overarching ecological factor bearing on linguistic complexity is the

1.5 The Debate Over the Equal Complexity of Languages

If languages are complex adaptive systems relative to the communication

2 The 2011 Workshop on Language Complexity

(especially in the system and in forms, among others) emerge in language.

Chapter 3 focuses on the following question: When a language can function

evolutionary perspective, focused on interactions of the biological evolution

Closing the book, Albert Bastardas-Boada approaches linguistic complexity

Blythe, Richard A. & William Croft. 2009. A Usage-Based Account of Constituency

Gershenson, Carlos, ed. 2008. Complexity: 5 Questions. Denmark: Automatic Press.

McWhorter, John H. 2012. Linguistic Simplicity and Complexity: Why do Languages

Luc Steels and Katrien Beuls

Agent-based approaches model a population of linguistic agents engaging

though it is widely accepted in sociology, biology, economics and many other

3 Case Study for Agreement Systems

3.1 The Language Game

3.2 Processing Complexity

with B0 = 0 and B1 = 1. Bn grows exponentially with the number of words. It

3.3 Population-Level Complexity: Language Variation

Figure 2.1 Results of an agent-based simulation, with a population of 10

3.4 Learning Complexity

Figure 2.3 Experiment comparing the learning efficiency Sg between a strat-

3.5 Form Complexity

Figure 2.4 (a) Overall reduction of form complexity in a population of agents

consonants and vowels, without affecting performance. There is greater vari-

3.6 Inventory Complexity

complexity along various dimensions. In particular, we developed explanations

Table 3.1 Languages from UPSID451 (Maddieson, 1984; Maddieson &

Phonemes UPSID Name ISO 639–3 UPSID Family

11 Pirahã Myp South American, Paezan

Other linguists do explain properties of language from language-specific

2 The Controversy Between Biology and Culture

Language and genes are the only

3 Culture and Cognition in Speech

3.1 Cultural and Cognitive Explanations of Speech Sounds

3.2 Experimental Cultural Learning