Models of Language Evolution and Change: Andrew D.M. Smith

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Advanced Review

Models of language evolution


and change
Andrew D.M. Smith∗

In the absence of direct evidence of the emergence of language, the explicitness


of formal models which allow the exploration of interactions between multiple
complex adaptive systems has proven to be an important tool. Computational
simulations have been at the heart of the field of evolutionary linguistics for the
past two decades, particularly through the language game and iterated learning
paradigms, but these are now being extended and complemented in a number of
directions, through formal mathematical models, language-ready robotic agents,
and experimental simulations in the laboratory. © 2014 John Wiley & Sons, Ltd.

How to cite this article:


WIREs Cogn Sci 2014. doi: 10.1002/wcs.1285

LANGUAGE to support particular hypotheses.1 Until the latter


decades of the 20th century, indeed, this meant that
L anguage is often described as the defining
characteristic of humanity: it is certainly the
central foundation of human society, which has
theorizing about the evolution of language was largely
restricted to intuition and extravagant speculation,
yielding only fanciful and effectively unfalsifiable
enabled our unprecedented accomplishments, provid-
theories (see Refs 2, 3 for overviews). To counter this
ing us with an infinitely expressive means of repre-
problem, evolutionary linguists have tried to garner
senting and successfully communicating to others a
research evidence from a wide variety of disciplines,
limitless number of dynamic situations and complex
from archeology and anthropology to genetics and
thoughts. The primary aim of evolutionary linguistics
primatology, the results of which can be seen in
is to explain how this system could have evolved in
many recent collections devoted to the evolution of
humans, how our species made the transition to a
language.4–18
linguistic communication system of the sort we all use
The key characteristics of language, which
today.
distinguish it from less extensive and less powerful
Language exists in two distinct forms, namely
animal communication systems, are often described in
the ephemeral external linguistic behaviors from
terms of the design features of language itemized by
which utterances are assembled (whether in speech or
Hockett.19 Crucial among these are the observations
sign language), and the internal linguistic representa-
that language can be distinguished from animal
tions stored in patterns of neural connections inside
communication in its being a semantic, productive,
individual human brains. Unfortunately, however,
and culturally learnt system: (1) its units are
neither form is durable: no fossil records of sounds
relatively fixed conventional associations between
or brain patterns exist, and written records are very
forms (utterances) and meanings; (2) new utterances
recent indeed, so there is no possibility of directly
can be created and understood by language users with
observing the genesis of language. The consequent
ease; (3) it is learnt through exposure to the linguistic
lack of direct evidence poses a severe methodological
behavior of other language users. These features are
problem for evolutionary linguistics, as there is little
underpinned by four key subsidiary characteristics20 :
agreement on the kind of evidence which can be used

Correspondence to: andrew.smith@stir.ac.uk • the symbolic, arbitrary nature of the associations
Division of Literature and Languages, University of Stirling, Stirling, between form and meaning21 ;
Scotland, UK • the duality of patterning or double articulation
Conflict of interest: The author has declared no conflicts of interest whereby utterances are constructed from small,
for this article.
meaningful units (morphemes) according to

© 2014 John Wiley & Sons, Ltd.


Advanced Review wires.wiley.com/cogsci

morphosyntactic rules, while these morphemes underlying theories can thus be evaluated and later
are themselves made up of individual meaningless revised on the basis of the evidence provided by the
sounds (phonemes), assembled according to a model. From a practical point of view, formal models
different set of phonological rules22 ; have opened up massive parameter spaces for sys-
• recursion, whereby a form can contain embed- tematic exploration and new hypothesis generation,
ded components of the same category as itself,23 particularly in areas such as language evolution, which
famously proposed as the core known com- can profitably be characterized using complex adap-
ponent of humans’ innate language-specific tive systems,29,30 where the large-scale behavior of a
faculty24 ; system emerges from multiple small-scale interactions
among its components, and where our intuitions about
• compositonality, which allows the meaning of a
these complex interactions are notoriously fallible.
complex signal to be derived from its structure
Models necessarily abstract away from some
and the meanings of its component parts,25
details of the system, so there is always a potential
and which is found at many levels of linguistic
danger that vital components might be omitted, but
analysis from morphological paradigms to
the robustness of modeling is that the omissions can
complex syntactic constructions. Even in animal
be explicitly specified and their consequences demon-
communication systems where hierarchical
strated. There is, however, a clear tension in modeling
syntactic structure has been convincingly
between making simple models which focus on few but
established, such as the songs of the Bengalese
potentially crucial aspects of the theory while abstract-
finch26 or the humpback whale,27 there is as yet
ing away from everything else, and more complex
no justification for assuming the existence of the
models which try to represent more aspects of the real-
corresponding complex semantic structures on
world situation which is being modeled. The decision
which compositionality is predicated.
between making a simpler or more complex model
Formal modeling has been at the forefront of is something of a balance between making a model
illuminating the evolutionary history of language; this more realistic and making it more easily analyzable,
paper provides an overview of some of its major ensuring that it is not too simple to ignore key features
contributions over the past two decades or so. I of the problem, yet not so complicated that important
first discuss the nature of formal models in Formal results are obscured. Simple models are frequently crit-
Models. In Levels of Evolutionary Explanation, I icized for being trivial (although such criticism is often
investigate different levels on which evolutionary made with the benefit of hindsight), but more com-
explanations are sought, looking at the differences plex models require much more reflection and nuanced
between genetic and cultural evolution and between interpretation to understand, and therefore lend them-
language evolution and language change. In Types selves less easily to clear and insightful conclusions.28
of Model, I describe various types of models which Modelers therefore need to be flexible, yet careful to
have been utilized, starting with computational ensure that their design and implementation decisions
simulations and then moving onto models grounded are plausible, justifiable and systematically explained.
on robots, mathematical models, and experimental
‘human simulations’. LEVELS OF EVOLUTIONARY
EXPLANATION
FORMAL MODELS Biological Evolution
A formal model makes deliberate theoretical assump- From the historically dominant nativist viewpoint,31
tions, rigorously specifies these assumptions, and then language has traditionally been seen as a biologically-
demonstrates their consequences. The great strength determined, autonomous module in the human brain
of formal modeling is its meticulous and unambiguous (variously dubbed the language ‘organ’ or ‘instinct’)
description of the components of a system and how which strongly determines the arbitrary universal
they interact, such that ‘predictions about the behav- structures hypothesized to underpin all languages
ior of that system can be more or less mechanically (‘Universal Grammar’); differences between individual
produced from the description’ (Ref 28, p. 582). The languages are considered relatively superficial under
explicitness of models allows us to see exactly how this account. The prime motivation for nativism is the
theoretical assumptions have been implemented and famous argument from the ‘poverty of the stimulus’,
can thus prove whether particular mechanisms do which claims that children have insufficient linguistic
yield desired effects, or, conversely, whether a result evidence to acquire the grammar of their language32
can be obtained through different mechanisms. The (although see Ref 33 for a detailed investigation which

© 2014 John Wiley & Sons, Ltd.


WIREs Cognitive Science Language evolution and change

found no convincing empirical support for the nativist Universal Grammar, resulting in very divergent and
position). In a very influential paper which effectively often contradictory interpretations of the biological
served to reignite interest in the field of language basis of language (see Ref 2 for a useful summary).
evolution, Pinker & Bloom provided a theoretical
evolutionary account of the nativist view, pointing out Cultural Evolution
that as linguistic structures are complex and adaptive,
An influential school of thought proposes that, rather
so the organ which determines them must have evolved
than our brains having evolved to learn languages,
biologically through natural selection for language
instead the languages themselves have evolved to being
learning: ‘Evolutionary theory offers clear criteria for
able to be learnt by human brains.39 Languages adapt
when a trait should be attributed to natural selection:
to general, nonlinguistic cognitive constraints which
complex design for some function, and the absence
improve their learnability, such as sequential process-
of alternative processes capable of explaining such
ing or the inference of meaning. Repeated adaptation
complexity’ (Ref 34, p. 707). Early models therefore
to these constraints is reflected in the persistence of
explored the conditions under which language-like
languages with reliable cues to learnability50 and,
bidirectional communication systems could evolve in
more generally, in the universal tendencies we see in
a population, using both computational simulations35 human language.51 Languages need to be learnable
and mathematical models.36,37 because they exist as cultural artifacts among popula-
The evolutionary mechanism to which Pinker & tions of people, and a language can persist over time
Bloom appeal is known as the Baldwin effect, which only if it can be replicated in this cultural environment.
claims that a trait which was originally acquired The continual oscillation of language between its
through learning becomes encoded in the genes, internal representations and external behaviors52 has
because learning itself can change the selective envi- profound implications for the characteristics of the
ronment to make genes for the originally learnt behav- languages which emerge through cultural evolution:
ior more advantageous.38–40 The Baldwin effect has
been demonstrated in models where there is a direct
relationship between genotype and phenotype,41 but 1. Language learners do not have access to
it is suppressed if the genotype–phenotype relation- the entire language they are learning, and
ship is made more biologically plausible by including so must create their internal representations
one-to-many relationships between genotype and based on partial linguistic behavior. This
phenotype such as epistasis and pleiotropy in the transmission bottleneck introduces a pressure
model.42 Interestingly, the Baldwin effect itself can for generalization beyond the data,53 which
result in a weakening of the selective effect of the results in the cultural evolution of structure
original trait43 : once a genetic predisposition for from an initially random language. Models of
language learning has spread through a population, a this process, discussed in more detail in Iterated
Learning, have been convincingly used to claim
process of niche construction44 produces a linguistic
that linguistic structure actually exists because
environment with very little variation, which in
of the poverty of the data available to language
turn can relax the selective pressure to maintain the
learners, not in spite of it.54
original predisposition for language learning.
A major stumbling block to accounts of the 2. Language learners do not have access to
genetic assimilation of arbitrary linguistic properties, the precise meanings of the utterances they
however, is the dynamic and prodigious variation in all encounter, but rather they infer meaning based
language.45,46 Although general principles supporting on the recognition of the common ground they
language learning can indeed be assimilated while the share with the speaker.55–57 This inference of
language is changing, specific linguistic parameters can meaning is an inherently uncertain and vari-
only become canalized if the language being learnt is able process,58 so individual language users
fixed.47 Even under conditions of very slow linguistic inevitably develop different internal represen-
change, biological selection works against those genes tations of language.59 Constant communicative
biased toward a particular linguistic principle and in pressure, however, requires the different internal
favor of neutral genes,48,49 effectively because if the representations to yield external behavior whose
evolutionary target is changing too quickly, biological meanings are reconstructible from context.60
evolution cannot keep up.
Such arguments against its formulation in terms In both cases, the cultural processes of language
of principles and parameters have led over recent acquisition and usage introduce selective pressures
years to increasing scrutiny of the very concept of on language which lead to its cultural evolution:

© 2014 John Wiley & Sons, Ltd.


Advanced Review wires.wiley.com/cogsci

linguistic structures which are easy to acquire, or factors which could influence the system means
are communicatively effective, will tend to persist and that intuitions about their combined effects are very
increase in the language; those which are difficult to liable to be erroneous, but ‘it is exactly in these
learn or hinder communication will tend to be dis- circumstances, when multiple processes have to be
preferred and disappear from the language.46 Natural considered together, that modeling becomes a useful
selection is therefore not the only explanatory can- —and perhaps even necessary— tool’ (Ref 9, p. 13).
didate for the evolution of linguistic structure, rather Language evolution is not a single complex
cultural transmission itself instigates linguistic adapta- adaptive system, however, but rather an interaction
tion. This discovery shifts the burden of evolutionary between three separate complex adaptive systems65
explanation considerably: evolutionary linguists can which all result in different kinds of language
seek to identify the minimal fundamental set of cogni- change, taking place on very different timescales18,66 ;
tive capacities needed to allow cultural conventions to somewhat confusingly, all of these can be referred
be shared in a community, and to document how the to as ‘language evolution’. During an individual’s
processes of cultural evolution can allow language to ontogenetic development, they learn and adapt their
evolve from these underlying mechanisms.20 Impor- linguistic knowledge in response to their environment
tantly, the relevant cognitive capacities, such as the and the linguistic experiences they are exposed to.
assumption of common ground between interlocutors, Such change is most dramatic in children’s initial
and the memorization of experience,56,61 need not language learning, of course, but linguistic knowledge
even be dedicated to language alone, but could plau- is constantly in flux, as people respond to the
sibly have evolved for a different purpose altogether. language used around them by modifying their own
language patterns. Languages themselves change
more slowly, over glossogenetic time,67 through
Complex Adaptive Systems historical processes such as those which have resulted
For a full understanding of language evolution, in the transformation of Old English into modern
therefore, biological and cultural evolution need to English and through which modern English continues
be considered separately, but also in their mutual to develop today. Again, some changes are relatively
interaction: biological evolution provides cognitive small, such as minor changes in pronunciation or
pre-adaptations including those which impact on how meaning, but others can be very substantial, such as
individuals can interact with each other, but these the loss of a case system or the reorganization of a lan-
social interactions themselves can cause the emergence guage’s syntactic structure.68 Yet more slowly, over
of qualitative structural changes in linguistic systems. millennia of phylogenetic time, biological evolution
A useful conceptualization of language, which allows has endowed the human species with the cognitive
the exploration of the role of cumulative cultural capabilities and physical faculties on which the
evolution in explaining its crucial characteristics is continuation of language depends. A full explanation
as a complex adaptive system.62 A complex adaptive of how language evolved and continues to change
system is complex because the behavior of the system therefore requires accounts of the biological evolution
as a whole cannot be projected trivially from the of the cognitive and physical foundations of language
microbehavior of its individual components, and and of the cultural evolution of language in dynamic
adaptive because this microbehavior changes and populations of individuals with these cognitive adap-
develops in response to historical experience and tations. Here too, though, some researchers insist on a
external pressures from the wider environment. A lan- clear distinction between the role of cultural evolution
guage community comprises a multitude of language in the initial emergence of language and its role in
more contemporary historical linguistic change,69
users, whose individual linguistic interactions change
while others appeal to parsimony with an account
according to their previous linguistic experience
based on the same cultural evolutionary processes.56
and their wider social motivations30 ; the linguistic
structures in the community are emergent, deriving
from conventionalized generalized patterns extracted TYPES OF MODEL
from the communicative interactions, but without
any kind of central control, pre-specification, or Computational Simulations
explicit optimisation.63,64 Exploring actual complex Computational simulations have been the bedrock of
adaptive systems is not trivial, of course: it is effec- language evolution models, from the earliest simula-
tively impossible to carry out a sufficiently detailed tions exploring the biological evolution of symbols70
longitudinal study of a community of language users and the ‘critical period’ for language learning,71,72 to
in the real world, and the multitude of different recent simulations investigating the evolution of very

© 2014 John Wiley & Sons, Ltd.


WIREs Cognitive Science Language evolution and change

specific grammatical features.73–76 Many language complete a semiotic cycle: the speaker conceptualizes
processing models use simple recurrent networks77 what needs to be communicated, and produces
to explore the conditions under which language can an utterance to represent this conceptualization;
be learnt from experience. This work has had consid- the hearer receives the utterance, reconstructs
erable success in showing that language acquisition the meaning as far as possible and interprets the
can be accounted for using just general learning meaning based on their own conceptualization of the
strategies which are not specific to language, and have environment.
consequently posed a major challenge to poverty of Wellens88 shows in detail how different kinds of
the stimulus argument and thus nativism. They have language games require different strategies for agents
shown, for instance, that information extracted from to reach their communicative goals. The most basic
transitional probabilities between words can allow language game, the naming game, considers agents
learners to correctly distinguish grammatical auxiliary trying to develop and maintain a set of conventional
inversions from ungrammatical inversions in complex names for objects in their world. Mechanisms
polar interrogative sentences,78 and that networks such as lateral inhibition, which reward successful
can integrate this kind of distributional information words and hinder their competitors, are crucial for
with probabilistic phonological cues to create reliable achieving conventionalization in these circumstances.
and accurate representations of lexical categories.79 If ambiguity is introduced between the word and
Agent-based models of language evolution its referent, in so-called guessing games, then
typically involve a population of simulated individuals agents need additionally to use techniques such as
with some specified cognitive capabilities (such as the cross-situational learning, a mechanism which allows
ability to store their experiences and to generalize over multiple possible meanings to be stored for each
them); the agents interact with each other, adapt their word, and for the meanings of words to be learnt in
linguistic knowledge as a result of these interactions, the face of uncertainty across multiple exposures.89 If
and may be replaced by other agents as the even more meaning uncertainty is facilitated through
simulation unfolds over many hundreds or thousands the introduction of semantic multidimensionality
of interactions. Agent-based models are probably and compositionality, a more flexible semantic
the most commonly used technique in evolutionary representation is required; shared conventions then
linguistic modeling; a formidable range of problems come about through the gradual adaptation of these
from the emergence of phonology and phonological internal representations to reflect usage rather than
structure,80–82 to the evolution of vocabulary70,83,84 through the elimination of competitor words.88
and of syntactic structure53 has been explored using Language games have been devised to explore
models with varying levels of complexity from simple not only the development of shared lexicons as
naming games85 to fully-blown implementations like described above,59,88,90–94 but also the emergence of
Fluid Construction Grammar.86 Two key agent- compositional language,95 and they are increasingly
based paradigms in language evolution research are used to probe the development of specific and
the language game model,85,87 which focuses on more complex grammatical features such as case
the emergence of language through the interactive systems,73 Aktionsart,74 quantifiers75 and gram-
negotiation of co-ordinated behavior, and the more matical agreement.76 In all these cases, the desired
general iterated learning model,65 which focuses on features emerge because the models implement both
the transmission of linguistic behavior from one functional pressures which make the linguistic struc-
individual to another. tures desirable and useful, and appropriate cognitive
mechanisms which allow them to be represented.
Language Games
Language games are based on the idea that goal- Iterated Learning
driven, context-sensitive communication between The iterated learning model65,96,97 is designed to
individuals is a key driver behind the self-organization investigate the relationship between the properties
of language systems; they always therefore contain of individual language learners embedded within a
a population of individual agents within a specified transmission chain,98 and the resultant structure of the
environment, with specific goals to communicate with language they use. Iterated learning is most frequently
each other. The agents have the facility to interact associated with a specific type of vertical transmission
with and perceive the environment, and perhaps to through generations, but is really a much more
build and adapt their own individual, internal models general concept which also encompasses ‘horizontal’
of things they communicate about. In a language negotiation of conventions within a generation as
game, two individuals (a speaker and a hearer) must in the language game model. Iterated learning can be

© 2014 John Wiley & Sons, Ltd.


Advanced Review wires.wiley.com/cogsci

described as ‘a process in which an individual acquires state where every meaning is expressed with the same
a behavior by observing a similar behavior in another utterance.99,107
individual who acquired it in the same way’ (Ref Both the language game and iterated learning
99, p. 10681). This notion is clearly applicable to paradigms have proved very successful, and their
the transmission of human language described above: use in explorations of language evolution and
one individual’s linguistic behavior is the source for change has been substantially broadened from these
others to create their internal representations, and the computational origins into a variety of arenas, the
internal representation which yielded their behavior most important of which may be the implementation
was the result of previous exposure to others’ linguistic of language agents in physical robots, formal math-
behavior. Many computational simulations of iterated ematical models, and the recent use of psychological
learning have been implemented,53,100–103 using a ‘human simulation’ experiments.
wide variety of both different models of learning and
of meaning, but the constant result is clear: the simple
process of repeated learning from others’ behavior is Robotic Agents
sufficient to transform initially random languages into The earliest language games were developed not
stable, complex languages with syntactic structure, in simulations, but on actual physical robots with
demonstrating the key role of cultural evolution in the crude sensorimotor facilities, which moved about
evolution of language. and explored their environment.90,108,109 Later, in the
Kirby’s initial model53 showed how formal famous ‘Talking Heads’ experiment,91 robot cameras
models can yield important knowledge in addition in labs across the world were connected through the
to the simple confirmation or disconfirmation of a Internet and used language games to develop a shared
theoretical proposal: the model was designed as an language with which to describe the colored shapes on
explicit exploration of a somewhat controversial the whiteboards they faced. As technology has con-
theory of the development of protolanguage into tinued to improve, researchers have expanded their
language,104 but turned out not only to confirm experimental platforms to include AIBO robot dogs110
that fractionation was indeed a plausible mechanism and even humanoid robots.111 The main advantage
for this process, but also more importantly that a to using robotic agents over computer simulations
crucial factor in its operation was what has become is realism, particularly in experiments designed to
known as the transmission bottleneck, or the fact explore theories such as embodied cognition.112–114
that learners are required to learn the language from Mobile robots were used, for instance, in a remarkable
a restricted set of data. Learners in the simulation demonstration of the emergence of a communication
were asked to produce utterances for meanings system from noncommunicative behavior.115 The
they had never encountered, and this forced them robots could detect each other’s presence, and were
to generalize from the data they had encountered. evolved simply for their ability to move in the same
This pressure for generalization favors compositional direction as each other; a simple communication
rules with systematic relationships between form and system emerged, however, as they oscillated in and
meaning over holistic rules where the form-meaning out of range of each other, setting up a proposed
relationships are idiosyncratic, so over repeated gen- direction of travel for one robot to follow the other.
erations, the holistic rules in the languages gradually More common, however, is the use of robots in
disappear.53 This finding led to a series of further situations where communicability is already assumed,
computational95,100,105 and mathematical96,106 mod- and modelers explore the conditions favorable to the
els exploring the same basic model under different development of language-like systems. In practice,
parameter settings; these showed that holistic rules however, experiments with robots are very intensive
can only persist in a language if they are very and time-consuming, requiring complex processing
frequent,100 if the meanings expressed in the language of the raw data captured by the robotic sensors to
are unstructured,96 or if the meanings are unrelated abstract information of interest, and considerable
to each other.105 These and similar modifications of hardware maintenance costs116,117 ; many researchers
the assumptions on which the original model was therefore use simulated robotic frameworks118,119
based have also improved our understanding of the to avoid these issues. Robot simulations like this
requirements for the cultural evolution of composi- have been used to demonstrate that cultural imitation
tionality: it is crucial, for instance, that there exists is a much more successful learning strategy than
not only a pressure for generalization but also a bias receiving feedback from the environment,120 once an
against ambiguity as without it the language collapses initial core set of sensorimotor categories has been
into a maximally ambiguous but highly generalizable developed through direct grounding.

© 2014 John Wiley & Sons, Ltd.


WIREs Cognitive Science Language evolution and change

Mathematical Models This simulation result was confirmed by further


Mathematical models have frequently been used to mathematical analysis, which additionally showed,
provide definitive proofs in place of the interpretative however, that mutual exclusivity, the assumption
conclusions derived from simulations, yet simulations that different words have different meanings,127 is
are often also required to supplement mathematical an extraordinarily powerful heuristic which enables
models in cases where the problem cannot be words to be learnt through cross-situational learning
simplified sufficiently to be analytically tractable. Two almost as quickly as they are encountered.128
examples of the interplay between simulations and Mathematical models have also been used to
mathematical models, which also serve to show how illustrate a great many other theories of language
apparently minor changes in the assumptions of a evolution, from theoretical treatments of language
model can have profound effects on its results, can be games,129–131 through dynamical systems models
found in a series of papers exploring the relationship of language acquisition132 and of the emergence of
between biology and culture in language evolution. combinatoriality36 and Universal Grammar,37 to
The first set of models were inspired by models showing that cognitive constraints across
the iterated learning simulations which showed the generations give rise to consistent word order
power of cultural evolution,121 and had learners regularities without Universal Grammar,133 and that
use Bayesian inference to choose a grammar based superficially combinatorial ‘phonological’ structures
on the linguistic data they are exposed to. An can emerge in a single system, when it is optimized
initial mathematical analysis showed that the learners’ for acoustic distinctiveness and robustness against
innate prior bias was the sole factor determining the noise.82 They are very frequently used in models of
languages which emerged from the cultural evolution language change which try to explain specific soci-
process122 : the transmission bottleneck surprisingly olinguistic phenomena such as the selective pressures
had no effect at all. This result, however, was itself behind the diffusion of linguistic variants within an
later shown66,123 to be solely due to an assumption individual’s linguistic representations134 and within
in the model that the selection of a grammar populations,135 and the process of language shift
was proportional to the learner’s confidence in the through competition between languages, which has
grammar having produced the data (i.e. learning been used to support language planning work to
was by sampling); if learners instead select the maintain minority languages.136–138 Perhaps most
grammar which maximizes the posterior probability, notably, Trudgill’s theory of new dialect formation139
the size of the transmission bottleneck is crucial was thoroughly investigated using a mathematical
in determining the stable distribution of languages, model of language change based on Croft’s theory of
as in the simulations. This happens because in the utterance selection46 and real-world data from New
maximizing case, it is only important to know which Zealand140 ; the analysis showed that although the
language is preferred (how much it is preferred over linguistic characteristics of the dialect were indeed
the others does not matter); the resultant distribution consistent with the theory, the timescale required for
of languages therefore reflects the learners’ prior biases the dialect’s emergence made this impossible.
only approximately.20 Further work evaluating these
two models of learning shows that learners who
maximize are more likely to be selected than learners Experimental Approaches
who sample,20 and that very weak general biases are Over the last few years, researchers have developed a
favored over strongly constraining biases. suite of experimental techniques using human partici-
The second body of research explored the power pants to observe the emergence and evolution of new
of cross-situational learning in the rapid acquisition languages and communication systems directly69,141 ;
of large human-size vocabularies of around 60,000 most of these were directly influenced by earlier
words.124 An initial mathematical analysis showed computational models. The first study to take this
that cross-situational learning allows large vocabular- approach was an experiment exploring the invention
ies to be learnt at reasonable speeds,125 even when the and use of a signaling system by pairs of partici-
context of unintended meanings which accompanied pants trying to co-ordinate their movement around
the word’s true meaning was large. Subsequent a simple grid on a computer screen.142 Participants
simulations, however, showed that this result rested were provided with a communication channel which
on the assumption that unintended meanings were cleverly rendered existing communication systems use-
uniformly distributed; under more realistic Zipfean less, yet provided other possibilities; after many trials,
distributions, increases in context size resulted these were exploited by some pairs, who developed
in super-exponential growth in learning times.126 conventionalized and (under certain circumstances)

© 2014 John Wiley & Sons, Ltd.


Advanced Review wires.wiley.com/cogsci

compositional language-like systems. The communi- their answers were used as the training data for
cation channel itself was removed in a fascinating and the next participant.98 In the basic experiment, the
ingenious subsequent experiment,143 where the only resultant languages tend to degenerate, with few
behavior open to participants was their own move- distinct words being used, and whole dimensions
ment around the grid. Communication is extraor- of the underlying meaning structures being lost;
dinarily difficult in these conditions, yet possible if the languages become easy to learn but have little
the pairs had a common understanding of expected expressive power. If this expressive power is explicitly
behavior, and could therefore signal their communica- maintained, however, by ensuring that the same
tive intent by movement which was clearly different word cannot refer to multiple objects, then the
from those expectations. languages change to become easy to learn in a
A related technique which also eschews existing different way, becoming compositional to reflect
communication systems is graphical communication the underlying meaning structure, in a result very
or Pictionary experiments, in which one participant reminiscent of Kirby’s earlier computational model.53
draws and the other guesses their intended The authors argue convincingly that communication
referent.144–147 Initially, the drawings used must be provides the natural pressure for the utility of a
externally motivated in some way if they are to communication system, and that human language
be communicatively useful, but over repeated use is thus the dynamic result of a balancing act
between participants they become simplified and between learnability and expressivity. Experiments
conventionalized, as the information held in them like this have now demonstrated the emergence of
effectively moves from the drawings themselves to communication systems in general,143 and many of
the participants’ memory of the their use in shared the key design features of language such as symbolism
interactions.144,148 Simplification also occurs when and compositionality in particular,144,147 showing
participants make use of pre-existing structure in the that the process of cultural transmission has a key
meaning space from which the referents are taken, and multifaceted explanatory role in the evolution of
and develop systematic, compositional drawings language.
which match this underlying structure.147 In such
experiments, the referents are usually chosen from
a small, pre-defined set, but recent theoretical work CONCLUSION
supported by computational models149 suggests that
Formal models of language evolution and change are
this ‘drift to the arbitrary’150 in shared communication
an invaluable technique which allows theories to be
systems is a much more general process, underpinned
tested, evaluated, and revised in a systematic way.
by the so-called metaphorical use of signs, which are
They are particularly useful in situations where the
used to represent meanings slightly different from
object of study is the result of interactions between
their motivated meaning but still inferable from the
a number of complex adaptive systems. Formal
communicative context.61
models have been used to investigate many different
Experimental approaches have also been used to
aspects of language evolution and change, and have
confirm the results of earlier computational models,
developed from their computational beginnings in
most notably in relation to iterated learning across
both complexity and scope, most notably with the
chains of participants. In the most celebrated of these,
advent of laboratory experiments which have been
participants were asked to learn an ‘alien’ language
used not only to confirm model predictions but also
of random strings paired with a set of structured
to provide additional ecological validity.
meanings; they were tested on their learning, and

REFERENCES
1. Botha RP. Unravelling the Evolution of Language. and Cognitive Bases. Cambridge, MA: Cambridge
Amsterdam: Elsevier; 2003. University Press; 1998.
2. Fitch WT. The Evolution of Language. Cambridge, 5. Knight C, Studdert-Kennedy M, Hurford J, eds.
MA: Cambridge University Press; 2010. The Evolutionary Emergence of Language: Social
3. McMahon A, McMahon R. Evolutionary Linguistics. Functions and the Origins of Linguistic Form.
Cambridge, UK: Cambridge University Press; 2013. Cambridge, UK: Cambridge University Press; 2000.
4. Hurford JR, Studdert-Kennedy M, Knight C, eds. 6. Briscoe E, ed. Linguistic Evolution through Lan-
Approaches to the Evolution of Language: Social guage Acquisition: Formal and Computational

© 2014 John Wiley & Sons, Ltd.


WIREs Cognitive Science Language evolution and change

Models. Cambridge, MA: Cambridge University Press; 25. Krifka M. Compositionality. In: Wilson RA, Keil F,
2002. eds. The MIT Encyclopedia of the Cognitive Sciences.
7. Cangelosi A, Parisi D, eds. Simulating the Evolution Cambridge, MA: MIT Press; 2001.
of Language. London: Springer; 2002. 26. Okanoya K. Sexual display as a syntactic vehicle:
8. Wray A, ed. The Transition to Language. Studies in the evolution of syntax in birdsong and human
the Evolution of Language. Oxford: Oxford University language through sexual selection. In: Wray A, ed. The
Press; 2002. Transition to Language. Oxford: Oxford University
Press; 2002, 46–64.
9. Christiansen MH, Kirby S, eds. Language Evolution.
Oxford: Oxford University Press; 2003. 27. Suzuki R, Buck JR, Tyack PL. Information Entropy
of Humpback Whale Songs. J Acoust Soc Am 2006,
10. Tallerman M, ed. Language Origins: Perspectives on
119:1849–1866.
Evolution. Studies in the Evolution of Language.
Oxford: Oxford University Press; 2005. 28. Smith K. Why formal models are useful for
evolutionary linguists. In: Gibson KR, Tallerman
11. Minett JW, Wang WSY. Language Acquisition,
M, eds. Oxford Handbook of Language Evolution.
Change and Emergence: Essays in Evolutionary
Oxford: Oxford University Press; 2011.
Linguistics. Hong Kong: City University of Hong Kong
Press; 2005. 29. Gell-Mann M. The Quark and the Jaguar. New York:
Freeman; 1994.
12. Cangelosi A, Smith ADM, Smith K, eds. The Evolution
of Language: Proceedings of the 6th International 30. Beckner C, Blythe RA, Bybee JL, Christiansen MH,
Conference. Singapore: World Scientific Press; 2006. Croft W, Ellis NC, Holland J, Ke J, Larsen-
Freeman D, Schoenemann T. Language is a complex
13. Smith ADM, Smith K, Ferrer i Cancho R, eds. The
adaptive system: position paper. Lang Learn 2009,
Evolution of Language (EVOLANG 7). Singapore:
59(S1):1–26.
World Scientific; 2008.
31. Chomsky N. Aspects of the Theory of Syntax.
14. Larson RK, Déprez V, Yamakido H, eds. The Evolu-
Cambridge, MA: MIT Press; 1965.
tion of Human Language: Biolinguistic Perspectives.
Cambridge, UK: Cambridge University Press; 2010. 32. Chomsky N. Rules and Representations. London: Basil
Blackwell; 1980.
15. Smith ADM, Schouwstra M, de Boer B, Smith K,
eds. The Evolution of Language (EVOLANG 8). 33. Pullum GK, Scholz BC. Empirical assessment of
Singapore: World Scientific; 2010. stimulus poverty arguments. Linguist Rev 2002, 19(1-
2):9–50.
16. Tallerman M, Gibson KR, eds. The Oxford Handbook
of Language Evolution. Oxford: Oxford University 34. Pinker S, Bloom P. Natural language and natural
Press; 2011. selection. Behav Brain Sci 1990, 13:707–784.
17. Scott-Phillips TC, Tamariz M, Cartmill EA, Hurford 35. Oliphant M. The dilemma of Saussurean communica-
JR, eds. The Evolution of Language (EVOLANG 9). tion. BioSystems 1996, 37:31–38.
Singapore: World Scientific Press; 2012. 36. Nowak MA, Krakauer D. The evolution of language.
18. Binder PM, Smith K, eds. The Language Phenomenon. Proc Natl Acad Sci USA 1999, 96:8028–8033.
London: Springer; 2013. 37. Nowak MA, Komarova NL, Niyogi P. Evolution of
19. Hockett CF. The origin of speech. Sci Am 1960, Universal Grammar. Science 2001, 291:114–117.
203:88–96. 38. Hinton G, Nowlan S. How learning can guide
20. Smith K, Kirby S. Cultural evolution: implications evolution. Complex Syst 1987, 1:495–502.
for understanding the human language faculty and its 39. Deacon T. The Symbolic Species. London: Penguin;
evolution. Philos Trans R Soc Lond Ser B Biol Sci 1997.
2008, 363:3591–3603.
40. Briscoe T. Grammatical assimilation. In: Christiansen
21. de Saussure F. Course in General Linguistics. New M, Kirby S, eds. Language Evolution. Oxford: Oxford
York: The Philosophical Library; 1959[Translated by University Press; 2003, 295–316.
Wade Baskin].
41. Turkel WJ. The learning guided evolution of
22. Martinet A. La double articulation linguistique. natural language. In: Briscoe E, ed. Linguistic
Travaux du Cercle Linguistic de Copenhague 1949, Evolution through Language Acquisition: Formal and
5:30–37. Computational Models. Cambridge, MA: Cambridge
23. Kinsella AR. Language Evolution and Syntactic University Press; 2002, 235–254.
Theory. Cambridge, MA: Cambridge University Press; 42. Yamauchi H. The Difficulty of the Baldwinian Account
2009. of Linguistic Innateness. In: Kelemen J, Sosík P, eds.
24. Hauser MD, Chomsky N, Fitch WT. The faculty of In Advances in Artificial Life: Proceedings of the 6th
language: what is it, who has it and how did it evolve? European Conference on Artificial Life. Heidelberg:
Science 2002, 298:1569–1579. Springer-Verlag; 2001, 391–400.

© 2014 John Wiley & Sons, Ltd.


Advanced Review wires.wiley.com/cogsci

43. Yamauchi H, Hashimoto T. Relaxation of selection, 60. Smith ADM. Protolanguage reconstructed. Interact
niche construction and the Baldwin Effect in language Stud 2008, 9:100–116.
evolution. Artif Life 2010, 16:271–287. 61. Smith ADM, Hoefler SH. The pivotal role of metaphor
44. Odling-Smee FJ, Laland KN, Feldman MW. Niche in the evolution of human language. In: Vera JED, ed.
construction. Am Nat 1996, 147:641–648. Metaphor and Metonymy through Time and Culture.
45. Trask RL. Historical Linguistics. London: Arnold; Berlin: Mouton de Gruyter; In press.
1996. 62. Smith K. Cultural evolution of language. In: Brown
46. Croft W. Explaining Language Change: An Evolu- K, ed. The Encyclopedia of Language and Linguistics.
tionary Approach. Longman: Longman Linguistics New York: Elsevier; 2006, 315–322.
Library; 2000.
63. Hopper PJ. Emergent grammar. BLS 1987,
47. Munroe S, Cangelosi A. Learning and the Evolution of 13:139–157.
Language: the role of cultural variation and learning
64. Bybee JL. From usage to grammar: the mind’s response
costs in the Baldwin effect. Artif Life 2002, 8:331–339.
to repetition. Language 2006, 82:711–733.
48. Chater N, Christiansen MH. Language acquisition
meets language evolution. Cognit Sci 2009:1–27. 65. Kirby S, Hurford JR. The emergence of linguistic
structure: an overview of the iterated learning model.
49. Chater N, Reali F, Christiansen MH. Restrictions on
In: Cangelosi A, Parisi D, eds. Simulating the
biological adaptation in language evolution. Proc Natl
Evolution of Language. London: Springer Verlag;
Acad Sci USA 2009, 106:1015–1020.
2002, 121–148.
50. Lupyan G, Christiansen MH. Case, Word order
66. Kirby S, Dowman M, Griffiths TL. Innateness and
and language learnability: insights from connectionist
modelling. In: Proceedings of the 24th Annual culture in the evolution of language. Proc Natl Acad
Conference of the Cognitive Science Society. Mahwah, Sci USA 2007, 104:5241–5245.
NJ: Lawrence Erlbaum Associates; 2002, 596–601. 67. Hurford JR. Nativist and functional explanations in
51. Christiansen MH, Chater N. Language as shaped by language acquisition. In: Roca IM, ed. Logical Issues
the brain. Behav Brain Sci 2008, 31:489–508. in Language Acquisition. Berlin: de Gruyter; 1990,
85–136.
52. Hurford JR. Expression/induction models of language
evolution: dimensions and issues. In: Briscoe E, ed. 68. Lass R. Historical Linguistics and Language Change.
Linguistic Evolution Through Language Acquisition: Cambridge, UK: Cambridge University Press; 1997.
Formal and Computational Models. Cambridge, MA: 69. Scott-Phillips TC, Kirby S. Language evolution in the
Cambridge University Press; 2002, 301–344. laboratory. Trends Cogn Sci 2010:898.
53. Kirby S. Syntax without natural selection: how
70. Hurford JR. Biological evolution of the Saussurean
compositionality emerges from vocabulary in a
sign as a component of the language acquisition device.
population of learners. In: Knight C, Studdert-
Lingua 1989, 77:187–222.
Kennedy M, Hurford JR, eds. The Evolutionary
Emergence of Language: Social Function and the 71. Hurford JR. The evolution of critical period for
Origins of Linguistic Form. Cambridge, MA: language acquisition. Cognition 1991, 40:159–201.
Cambridge University Press; 2000, 303–323. 72. Batali J. Innate biases and critical periods: Combining
54. Zuidema WH. How the poverty of the stimulus solves evolution and learning in the acquisition of syntax. In:
the poverty of the stimulus. In: Becker S, Thrun S, Brooks R, Maes P, eds. Artificial Life 4: Proceedings of
Obermayer K, eds. Advances in Neural Information the Fourth International Workshop on the Synthesis
Processing Systems 15 (Proceedings of NIPS ’02). and Simulation of Living Systems. Redwood City, CA:
Cambridge, MA: MIT Press; 2003. Addison-Wesley; 1994, 160–171.
55. Sperber D, Wilson D. Relevance: Communication and 73. van Trijp R. The evolution of case systems for
Cognition. Oxford: Blackwell; 1995. marking event structure. In: Experiments in Cultural
56. Hoefler S, Smith ADM. The pre-linguistic basis of Language Evolution. Amsterdam: John Benjamins;
grammaticalisation: a unified approach to metaphor 2012, 169–206.
and reanalysis. Stud Lang 2009, 33:883–906. 74. Gerasymova K, Spranger M, Beuls K. A language
57. Scott-Phillips TC. The Pragmatic Ape: Human strategy for aspect: encoding Aktionsarten through
Communication and the Evolutionary Origins of morphology. In: Steels L, ed. Experiments in Cultural
Language. London: Palgrave MacMillan; In press. Language Evolution. Amsterdam: John Benjamins;
58. Hurford JR. The Origins of Meaning: Language in the 2012, 257–276.
Light of Evolution. Oxford: Oxford University Press; 75. Pauw S, Hilferty J. The emergence of quantifiers.
2007. In: Steels L, ed. Experiments in Cultural Language
59. Smith ADM. The inferential transmission of language. Evolution. Amsterdam: John Benjamins; 2012,
Adapt Behav 2005, 13:311–324. 277–303.

© 2014 John Wiley & Sons, Ltd.


WIREs Cognitive Science Language evolution and change

76. Beuls K, Steels L. Agent-based models of strategies 94. Steels L, Belpaeme T. Coordinating perceptually
for the emergence and evolution of grammatical grounded categories through language: a case study
agreement. PLoS One 2013, 8:358960. for colour. Behav Brain Sci 2005, 28:469–529.
77. Elman J. Finding structure in time. Cognit Sci 1990, 95. Vogt P. The emergence of compositional structures
14:179–211. in perceptually grounded language games. Artif Intell
2005, 167(1-2):206–242.
78. Reali F, Christiansen MH. Uncovering the richness
of the stimulus: structure dependence and indirect 96. Brighton H. Compositional syntax from cultural
statistical evidence. Cognit Sci 2005, 29:1007–1028. transmission. Artif Life 2002, 8:25–54.
79. Reali F, Christiansen MH, Monaghan P. Phonological 97. Smith K, Kirby S, Brighton H. Iterated Learning: a
and distributional cues in syntax acquisition: scaling framework for the emergence of language. Artif Life
up the connectionist approach to multiple-cue 2003, 9:371–386.
integration. In: Alterman R, Kirsh D, eds. Proceedings 98. Mesoudi A. Using The Methods of Experimental Social
of the 25th Annual Conference of the Cognitive Science Psychology to Study Cultural Evolution. J Soc Exp
Society. Mahwah, NJ: Lawrence Erlbaum Associates; Cult Psychol 2007, 1:35–58.
2003, 970–975. 99. Kirby S, Cornish H, Smith K. Cumulative Cultural
80. de Boer B. The Origins of Vowel Systems. Oxford: Evolution in the Lab: an experimental approach to
Oxford University Press; 2001. the origins of structure in human language. Proc Natl
Acad Sci USA 2008, 105:10681–10686.
81. Oudeyer PY. The Self-Organization of Speech. Studies
in the Evolution of Language. Oxford: Oxford 100. Kirby S. Spontaneous evolution of linguistic structure:
University Press; 2006. an iterated learning model of the emergence of
regularity and irregularity. IEEE J Evol Comput 2001,
82. Zuidema W, de Boer B. The evolution of combinatorial 5:102–110.
phonology. J Phon 2009, 37:125–144.
101. Batali J. Computational simulations of the emergence
83. Steels L, Kaplan F. Bootstrapping grounded word of grammar. In: Hurford JR, Studdert-Kennedy M,
semantics. In: Briscoe E, ed. Linguistic Evolution Knight C, eds. Approaches to the Evolution of
through Language Acquisition: Formal and Computa- Language: social and cognitive bases. Cambridge, MA:
tional Models. Cambridge, MA: Cambridge University Cambridge University Press; 1998, 405–426.
Press; 2002, 53–73.
102. Tonkes B. On the origins of linguistic structure:
84. Smith K. The evolution of vocabulary. J Theor Biol computational models of the evolution of language.
2004, 228:127–142. PhD thesis, University of Queensland, 2002.
85. Steels L. A self-organizing spatial vocabulary. Artif 103. Batali J. The negotiation and acquisition of recursive
Life 1995, 2:319–332. grammars as a result of competition among exemplars.
86. Steels L. Design Patterns in Fluid Construction In: Briscoe E, ed. Linguistic Evolution through
Grammar. Amsterdam: John Benjamins; 2011. Language Acquisition: Formal and Computational
Models. Cambridge: Cambridge University Press;
87. Nowak MA, Plotkin JB, Krakauer D. The evolutionary 2002, 111–172.
language game. J Theor Biol 1999, 200:147–162.
104. Wray A. Protolanguage as a holistic system for social
88. Wellens P. Adaptive Strategies in the Emergence of interaction. Lang Commun 1998, 18:47–67.
Lexical Systems. Brussels: VUB Press; 2012.
105. Smith K, Brighton H, Kirby S. Complex Systems
89. Siskind JM. A computational study of cross-situational in Language Evolution: the cultural emergence of
techniques for learning word-to-meaning mappings. compositional structure. Adv Complex Syst 2003,
Cognition 1996, 61:39–91. 6:537–558.
90. Steels L, Vogt P. Grounding adaptive language 106. Brighton H. Simplicity as a driving force in linguistic
games in robotic agents. In: Husbands P, Harvey I, evolution. PhD thesis, University of Edinburgh, 2003.
eds. Fourth European Conference on Artificial Life. 107. Brighton H, Smith K, Kirby S. Language as an
Cambridge, MA: MIT Press; 1997, 474–482. evolutionary system. Phys Life Rev 2005, 2:177–226.
91. Steels L. The Talking Heads Experiment, Volume I. 108. Billard A, Dautenhahn K. Experiments in learning by
Words and Meanings. Antwerp: Laboratorium; 1999. imitation - grounding and use of communication in
92. Smith ADM. Intelligent Meaning Creation in a robotic agents. Adapt Behav 1999, 7(3/4):415–438.
Clumpy World Helps Communication. Artif Life 109. Vogt P. Lexicon grounding on mobile robots. PhD
2003, 9:175–190. thesis, Vrije Universiteit Brussel, 2000.
93. Vogt P, Coumans H. Investigating social inter- 110. Steels L, Loetzsch M. Perspective alignment in spatial
action strategies for bootstrapping lexicon devel- language. In: Coventry KR, Tenbrink T, Bateman
opment. J Artif Soc Social Simul 2003, 6. JA, eds. Spatial Language and Dialogue. Oxford
http://jasss.soc.surrey.ac.uk/6/1/4.html. University Press; 2009, 70–88.

© 2014 John Wiley & Sons, Ltd.


Advanced Review wires.wiley.com/cogsci

111. Steels L, Spranger M. The robot in the mirror. Connect 124. Bloom P. How Children Learn the Meanings of Words.
Sci 2008, 20:337–358. Cambridge, MA: MIT Press; 2000.
112. Lakoff G. Women, Fire and Dangerous Things: 125. Blythe RA, Smith K, Smith ADM. Learning times
What Categories Reveal About the Mind. Chicago: for large lexicons through cross-situational learning.
University of Chicago Press; 1987. Cognit Sci 2010, 34:620–642.
113. Lakoff G, Johnson M. Philosophy in the Flesh: 126. Vogt P. Exploring the robustness of cross-situational
The Embodied Mind and Its Challenge to Western learning under Zipfian distributions. Cognit Sci 2012,
Thought. New York: Basic Books; 1999. 36:726–739.
114. Pecher D, Zwaan RA, eds. Grounding Cognition: The 127. Markman EM, Wachtel GF. Children’s use of mutual
Role of Perception and Action in Memory, Language exclusivity to constrain the meaning of words. Cogn
and Thinking. Cambridge, UK: Cambridge University Psychol 1988, 20:121–157.
Press; 2005.
128. Reisenauer R, Smith K, Blythe RA. Stochastic
115. Quinn M. Evolving Communication without Dedi- dynamics of lexicon learning in an uncertain and
cated Communication Channels. In: Kelemen J, Sosík nonuniform world. Phys Rev Lett 2013, 110:258701.
P, eds. Advances in Artificial Life: Proceedings of the
129. Lenaerts T, Jansen B, Tuyls K, de Vylder B. The
6th European Conference on Artificial Life, Prague,
evolutionary language game: an orthogonal approach.
September 2001. Heidelberg: Springer-Verlag; 2001,
J Theor Biol 2005, 235:566–582.
357–366.
130. Baronchelli A, Felici M, Loreto V, Caglioti E, Steels
116. Vogt P. Language evolution and robotics: issues on
L. Sharp transition towards shared vocabularies in
symbol grounding and language acquisition. In: Loula
multi-agent systems. J Stat Mech 2006:P06014.
A, Gudwin R, Queiroz J, eds. Artificial Cognition
Systems. London: Idea Group; 2006, 176–209. 131. Baronchelli A, Loreto V, Steels L. In-depth analysis of
the naming games dynamics: the homogenous mixing
117. Steels L. Experiments in Cultural Language Evolution.
case. Int J Mod Phys C 2008, 19:785–812.
Amsterdam: John Benjamins; 2012.
132. Niyogi P, Berwick R. A dynamical systems model of
118. Vogt P. THSim v3.2: the Talking Heads simulation
language change. Linguist Philos 1997:17.
tool. In: Banzhaf W, Christaller T, Ziegler J, Dittrich
P, Kim JT, eds. Advances in Artificial Life: Proceedings 133. Reali F, Christiansen MH. Sequential learning and
of the 7th European Conference on Artificial Life. the interaction between biological and linguistic
Lecture Notes in Computer Science: Lecture Notes adaptation in language evolution. Interact Stud 2009,
in Artificial Intelligence. Heidelberg: Springer-Verlag; 10:5–30.
2003, 535–544. 134. Wedel AB. Exemplar models, evolution and language
119. Tikhanoff V, Cangelosi A, Fitzpatrick P, Metta G, change. Linguist Rev 2006, 23:247–274.
Natale L, Nori F. An open-source simulator for 135. Gong T, Shuai L, Tamariz M, Jäger G. Studying
cognitive robotics research: the prototype of the language change using price equation and Pólya-Urn
iCub Humanoid Robot Simulator. In: Madhavan dynamics. PLoS One 2012, 7:e33171.
R, Messina ER, eds. Proceedings of the IEEE
136. Minett JW, Wang WSY. Modeling endangered
Workshop on Performance Metrics for Intelligent
languages: the effects of bilingualism and social
Systems (PerMIS’08). New York: IEEE; 2008.
structure. Lingua 2008, 118:19–45.
120. Cangelosi A, Harnad S. The adaptive advantage
137. Kandler A, Unger R, Steele J. Language shift,
of symbolic theft over sensorimotor toil: grounding
bilingualism and the future of Britain’s Celtic
language in perceptual categories. Evol Commun
languages. Philos Trans R Soc Lond Ser B Biol Sci
2000, 4:119–144.
2010, 365:3855–3864.
121. Kirby S. Learning, bottlenecks and the evolution
of recursive syntax. In: Briscoe E, ed. Linguistic 138. Bakalis E, Galani A. Modeling language evolution:
Evolution through Language Acquisition: Formal and Aromanian, an endangered language in Greece.
Computational Models. Cambridge, UK: Cambridge Physica A 2012, 391:4963–4969.
University Press; 2002, 173–203. 139. Trudgill P. New Dialect Formation: The Inevitability
122. Griffiths TL, Kalish ML. A Bayesian view of language of Colonial Englishes. Edinburgh: Edinburgh Univer-
evolution by iterated learning. In: Bara BG, Barsalou sity Press; 2004.
L, Bucciarelli M, eds. Proceedings of the 27th Annual 140. Baxter G, Blythe RA, Croft W, McKane A. Modeling
Conference of the Cognitive Science Society. Mahwah, language change: an evaluation of Trudgill’s theory of
NJ: Cognitive Science Society; 2005, 827–832. the emergence of New Zealand English. Lang Variat
123. Griffiths TL, Kalish ML. Language Evolution by Change 2009, 21:257–296.
Iterated Learning with Bayesian Agents. Cognit Sci 141. Galantucci B, Garrod S. Experimental semiotics: a
2007, 31:441–480. review. Front Hum Neurosci 2011, 5:11.

© 2014 John Wiley & Sons, Ltd.


WIREs Cognitive Science Language evolution and change

142. Galantucci B. An experimental study of the emergence 147. Theisen C, Oberlander J, Kirby S. Systematicity and
of human communication systems. Cognit Sci 2005, arbitrariness in novel communication systems. Interact
29:737–767. Stud 2010, 11:14–32.
143. Scott-Phillips T, Kirby S, Ritchie G. Signalling 148. Caldwell CA, Smith K. Cultural evolution and
Signalhood and the Emergence of Communication. perpetuation of arbitrary communicative conventions
Cognition 2009, 113:226–233. in experimental microsocieties. PLoS One 2012,
144. Garrod S, Fay N, Lee J, Oberlander J, MacLeod 7:e43807.
T. Foundations of representations: where might 149. Hoefler S. Modelling the role of pragmatic plasticity in
graphical symbol systems come from? Cognit Sci 2007, the evolution of linguistic communication. PhD Thesis,
31:961–987. University of Edinburgh, 2009.
145. Fay N, Garrod S, Roberts L. The fitness and 150. Tomasello M. Origins of Human Communication.
functionality of culturally evolved communication Cambridge, MA: MIT Press; 2008.
systems. Philos Trans R Soc Lond Ser B Biol Sci
2008, 363:3553–3561.
146. Fay N, Garrod S, Roberts L, Swoboda N. The
interactive evolution of human communication
systems. Cognit Sci 2010, 34:351–386.

© 2014 John Wiley & Sons, Ltd.

You might also like