Documente Academic
Documente Profesional
Documente Cultură
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
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
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
v
vi Contents
Index 245
Figures
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
1 Complexity in Linguistics
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
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.
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 in Language: A Multifaceted Phenomenon 11
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.
Complexity in Language: A Multifaceted Phenomenon 13
9 Not to be confused with Gell-Mann’s “effective complexity” introduced in the beginning of this
chapter.
14 Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino
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
Complexity in Language: A Multifaceted Phenomenon 15
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.
16 Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino
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
Complexity in Language: A Multifaceted Phenomenon 19
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.
Complexity in Language: A Multifaceted Phenomenon 23
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
24 Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino
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.”
Complexity in Language: A Multifaceted Phenomenon 25
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.
26 Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino
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
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
How to Explain the Origins of Complexity in Language 33
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).
34 Luc Steels and Katrien Beuls
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).
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.
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
40 Luc Steels and Katrien Beuls
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.
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.
42 Luc Steels and Katrien Beuls
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
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)
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).
How to Explain the Origins of Complexity in Language 45
(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
46 Luc Steels and Katrien Beuls
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.
How to Explain the Origins of Complexity in Language 47
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:
Oxford University Press.
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
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 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).
Complexity in Speech: Teasing Apart Culture and Cognition 51
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
Complexity in Speech: Teasing Apart Culture and Cognition 53
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
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.
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.
Vertical Transmission
Horizontal Transmission
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
1s 1 octave
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).
Complexity in Speech: Teasing Apart Culture and Cognition 61
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
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.”
Complexity in Speech: Teasing Apart Culture and Cognition 63
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.
Complexity in Speech: Teasing Apart Culture and Cognition 65
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
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
70 P. Thomas Schoenemann
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
A Complex-Adaptive-Systems Approach to the Evolution of Language 71
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.
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
A Complex-Adaptive-Systems Approach to the Evolution of Language 73
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
74 P. Thomas Schoenemann
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].
76 P. Thomas Schoenemann
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.
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
78 P. Thomas Schoenemann
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.
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
80 P. Thomas Schoenemann
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.
A Complex-Adaptive-Systems Approach to the Evolution of Language 81
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
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.
31 Rose
21 Kelly
17 Liz
16 Kanzi
6 Linda, Sue
4 Matata
3 Panbanisha
2 Karen, Panzee
1 Austin, Krista, Panban, Sherman
A Complex-Adaptive-Systems Approach to the Evolution of Language 83
63 you
45 your
15 me
14 it
5 I
3 her, him, them, we
1 my
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
84 P. Thomas Schoenemann
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
A Complex-Adaptive-Systems Approach to the Evolution of Language 85
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
88 P. Thomas Schoenemann
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
A Complex-Adaptive-Systems Approach to the Evolution of Language 93
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
94 P. Thomas Schoenemann
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.
A Complex-Adaptive-Systems Approach to the Evolution of Language 95
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.
96 P. Thomas Schoenemann
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.
A Complex-Adaptive-Systems Approach to the Evolution of Language 97
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.
A Complex-Adaptive-Systems Approach to the Evolution of Language 99
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.
100 P. Thomas Schoenemann
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
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
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
(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.
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
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
Evolutionary Complexity of Social Cognition, Semasiographic Systems 113
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
Evolutionary Complexity of Social Cognition, Semasiographic Systems 115
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
Evolutionary Complexity of Social Cognition, Semasiographic Systems 117
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
Evolutionary Complexity of Social Cognition, Semasiographic Systems 119
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.
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
Evolutionary Complexity of Social Cognition, Semasiographic Systems 125
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.
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.
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
Evolutionary Complexity of Social Cognition, Semasiographic Systems 131
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.
Oxford: Oxford University Press.
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-
bridge: Cambridge University Press.
Langacker, Ronald W. 1997. Constituency, dependency, and conceptual grouping.
Cognitive Linguistics 8:1–32.
Evolutionary Complexity of Social Cognition, Semasiographic Systems 133
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-
bridge: Cambridge University Press.
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?
1 Theoretical Background
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 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.
138 Christophe Coupé, Egidio Marsico, and François Pellegrino
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
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.
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
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.
To What Extent Are Phonological Inventories Complex Systems? 143
Table 6.2 Observed Frequencies of /a/ and /ã/ Across Languages in UPSID
(Expected Frequencies in Parentheses)
/ã/
Present Absent Total
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
144 Christophe Coupé, Egidio Marsico, and François Pellegrino
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.
The explosion of the number of combinations would also have made the anal-
ysis and presentation of the results significantly more difficult.
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.
148 Christophe Coupé, Egidio Marsico, and François Pellegrino
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.
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.
To What Extent Are Phonological Inventories Complex Systems? 149
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
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
Log 10
ID Seg. 1 Seg. 2 Assoc. Observed Expected (p-value)
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
152 Christophe Coupé, Egidio Marsico, and François Pellegrino
i u ı̃ ũ i: u:
ɪ ʊ
e o ẽ õ e: o:
ɛ ɔ ɛ̃ ɔ̃
a+ a ã a:
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.
To What Extent Are Phonological Inventories Complex Systems? 153
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
154 Christophe Coupé, Egidio Marsico, and François Pellegrino
Log 10
ID Seg. 1 Seg. 2 Assoc. Observed Expected (p-value)
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.
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%
To What Extent Are Phonological Inventories Complex Systems? 155
# 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.
Log 10
ID Feature 1 Feature 2 Assoc. Observed Expected (p–value)
Log 10
ID Feature 1 Feature 2 Assoc. Observed Expected (p-value)
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
Feature 1 Feature 2
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.
To What Extent Are Phonological Inventories Complex Systems? 159
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.
162 Christophe Coupé, Egidio Marsico, and François Pellegrino
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:
Oxford University Press.
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:
Oxford University Press.
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.
164 Christophe Coupé, Egidio Marsico, and François Pellegrino
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
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
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?”
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.
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.
+ Sig. greater than AL = No sig. diff. from AL; – Sig. less than AL
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
178 Barbara L. Davis
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.
A Complexity View of Ontogeny as a Window on Phylogeny 179
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
180 Barbara L. Davis
1.00
0.81
0.80
Ratios
0.00
Place Assimilation
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
0.20
0.06 0.05
0.00
Manner Assimilation
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
182 Barbara L. Davis
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.
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.
184 Barbara L. Davis
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.
186 Barbara L. Davis
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
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.
2.1 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).
Table 8.1 Typical Features of Complex Systems (adapted from San Miguel
et al. 2012)
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
190 Lucía Loureiro-Porto and Maxi San Miguel
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.
Language Choice in a Multilingual Society 191
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.
192 Lucía Loureiro-Porto and Maxi San Miguel
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
Language Choice in a Multilingual Society 193
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.
(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.
3.1 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.
Language Choice in a Multilingual Society 197
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
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).
198 Lucía Loureiro-Porto and Maxi San Miguel
Lattice Network
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).
Language Choice in a Multilingual Society 201
(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.
202 Lucía Loureiro-Porto and Maxi San Miguel
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.
204 Lucía Loureiro-Porto and Maxi San Miguel
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),
Language Choice in a Multilingual Society 205
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.
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.
Language Choice in a Multilingual Society 207
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
208 Lucía Loureiro-Porto and Maxi San Miguel
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.
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.
Language Choice in a Multilingual Society 211
(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
212 Lucía Loureiro-Porto and Maxi San Miguel
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.
Language Choice in a Multilingual Society 213
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.
214 Lucía Loureiro-Porto and Maxi San Miguel
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.
Language Choice in a Multilingual Society 215
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.
216 Lucía Loureiro-Porto and Maxi San Miguel
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.
Language Choice in a Multilingual Society 217
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
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.
Complexity and Language Contact: A Socio-Cognitive Framework 221
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.
222 Albert Bastardas-Boada
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).
Complexity and Language Contact: A Socio-Cognitive Framework 223
Figure 9.1 Main principles of the complexity perspectives in contrast with the
more traditional scientific ones.
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.
224 Albert Bastardas-Boada
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,
Complexity and Language Contact: A Socio-Cognitive Framework 225
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.
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.
226 Albert Bastardas-Boada
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).
228 Albert Bastardas-Boada
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).
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.
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.
Complexity and Language Contact: A Socio-Cognitive Framework 231
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.
3.3.2 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.
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
Complexity and Language Contact: A Socio-Cognitive Framework 235
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
236 Albert Bastardas-Boada
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
Complexity and Language Contact: A Socio-Cognitive Framework 237
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
238 Albert Bastardas-Boada
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.
240 Albert Bastardas-Boada
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
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