3
Polygenesis of Linguistic Strategies:
A Scenario for
the Emergence of Languages
Christophe Coupé and Jean-Marie Hombert
CNRS & Université Lyon 2
Abstract
On the one hand, numerous hypotheses have been put forward to
account for the emergence of language during the last million years
of human evolution. On the other hand, a large majority of linguists
considers that nothing can be said about past languages before
8,000 or 10,000 years in the past, given our current knowledge on
modern languages. A large gap obviously separates such approaches
and conceptions, and has to be crossed to provide a better account
of the development of our communicative system.
To partially bridge the gap between the former domains, we aim
at proposing a plausible scenario for the emergence of languages,
with an emphasis on the development of linguistic diversity. The
present study will address the question of the monogenesis or
polygenesis of modern languages, which is often implicitly biased
toward the first hypothesis. Probabilistic and computational models,
as well as palaeo-demographic data and evolutionary considerations,
will constitute the key points of our proposals.
1
2
Language Acquisition, Change and Emergence
1. The Origin of Languages
1.1
Language capacity and languages
Which definition for language?
Even if “language” is often considered as the capacity that separates
Man from other species, a precise definition remains controversial.
Most of the supposed distinctive features that had for example been
proposed by Hockett (1960) in the 1960s have been put into
question during the last decades by studies on “talking apes”, like
the bonobo named Kanzi (Savage Rumbaugh et al., 1998), or other
animals.
What, among other characteristics, contributes to the specificity
of language lies in the profound unity of its nature in terms of
cognitive or informational features, and at the same time the
extreme diversity of its superficial forms, namely the languages.
Explaining this discontinuity has been one of the major tasks of
linguists, leading to the development of elaborate and highly
detailed constructions like the generative grammars, and various
fields such as typology, the study of language universals, etc. To
what extent some features of language are genetically encoded
remains at the heart of intense debates (Schoenemann, 1999; Enard
et al., 2002; Lai et al., 2001).
While the following study does not deal primarily with this
controversy, we rely on a classical, yet sometimes implicit,
distinction suggested by the following arguments: first, there is a
faculty of language, language-specific or derived from more general
cognitive abilities, which characterizes the human aptitude for its
sophisticated communication. Second, there are instantiations of this
faculty, which are the approximately 6,000 languages spoken today.
As a consequence, we will tend to avoid using the term of
“language” in isolation, and will focus on the similarities or
divergences between the former notions, which are both related to
the notion of linguistic diversity.
Polygenesis of Linguistic Strategies
Emergence of the faculty of language and of modern languages
For the last twenty years, the question of the “origins of language”
has been revitalized by the cooperative efforts of a number of
disciplines. Partly in reaction to the proposal of a
genetically-encoded Universal Grammar (Chomsky, 1975), their
results and paradigms have provided new insights into this topic,
among others:
• Progress in the appreciation of our ancestors’ cognitive
capacities and behaviors (archaeology or palaeoanthropology)
• Discovery of substantial correlations between modern
linguistic and genetic distributions (Cavalli-Sforza, 1994)
• Discovery of plausible neural bases for behaviors that
could be related to the faculty of language, for example
mirror neurons (Rizzolatti et al., 1996; Rizzolati and
Arbib, 1996), etc.
On the basis of these new data, a number of theories have been
put forward to explain when, why and how our communication
system developed to reach its current state. Following the distinction
made earlier, an increasing set of works is dealing on the one side
with the origin of the faculty of language, while on the other side,
another body of research is focusing on the origin of contemporary
languages.
Nevertheless, the temporal gap which separates these two fields
is a large one, and the methods used to gather and analyze the data
in each of them often have few in common. As Figure 1 summarizes,
the emergence of a human-specific capacity of language presumably
happened along our phylogenetic tree somewhere between some tens
of thousands of years and a few million years ago, while the limit for
the validity of reconstructions by historical linguists is most often
assessed to be around 8,000 BP (Before Present). A consensus now
seems to have gained ground which dates the origin of modern
language between 50,000 and 100,000 years ago, in line with the
modern behaviors of our species Homo sapiens. All these dates
differ by several orders of magnitude.
3
4
Language Acquisition, Change and Emergence
Figure 1
The time scale of language evolution
While a number of linguists either try to rebuild the story of
recent human languages or the first steps of the development of the
faculty of language, e.g. Bickerton (1990)’s protolanguage, few of
them have taken an interest in the origin and development of
linguistic diversity per se. This approach differs from both the work
on ancestors of modern linguistic families (e.g. Proto-Indo-European,
Nostratic, Austric, etc.) and the study of the origin of the faculty of
language. Its intrinsic difficulty lies obviously in the absence of clues
from the past (even less than for the faculty of language), but
comparisons with contemporary human societies, models or
relationships between languages and other cultural features may
provide a valuable help.
Considering the evolution of linguistic diversity in itself is not
only interesting because it represents a key event to understand the
duality between the superficial diversity and the deep uniqueness of
the human communication system, but also because its ties with
various variables, for example the size of populations, or with
Polygenesis of Linguistic Strategies
cultural development, can shed light on the history of recent
languages. By knowing if the social or demographic conditions of
our ancestors were likely to correlate with a large or small number
of languages, we may address the plausibility of scenarios put
forward by some linguists about the number of languages that
would have been spoken 10,000 or 50,000 years ago (Pagel, 2000).
For the sake of simplicity, we will often confound linguistic
diversity and diversity of languages, even if one should keep in mind
that several concepts or problems differentiate these two notions,
the most significant actually requesting a precise definition of what a
language is (Nettle, 1999a: 63).
The origins of linguistic diversity as a matter of function or
product (why?) can be partially differentiated from the origins in
space and time (where and when?). Since we will mostly consider
the second notion in this article, we will just briefly summarize
below our position regarding the first issue.
Several theories explain the emergence of the faculty of language
(the “why?” question) by social causes. Dunbar (1996) proposed for
example that this faculty emerged to replace grooming in its
function of preserving social coherency, because of the increase in
social group size. This increase was suggested by the correlation
between the size of the group and the volume of the neo-cortex in
various monkey species (Dunbar, 1993). Another example is
Dessalles’ political theory of language, which focuses on social
aspects in agreement with Darwinian evolution: the development of
the faculty of language would have enabled individuals to better
express their qualities in order to form coalitions (Dessalles, 2001).
Such coalitions are often observed in chimpanzees (De Waal, 1998).
According to a different perspective, but still centered on the
social context of linguistic usage, the data of sociolinguistics have
largely underlined the weight of social and inter-individual
interactions on the evolution of languages and dialects. The social
game is in particular partly accountable for linguistic diversity, as
highlighted by various examples like Labov’s pioneering study of
Martha’s Vineyard (Labov, 1963).
At the crossing of these various hypotheses, we assume that it is
5
6
Language Acquisition, Change and Emergence
reasonable to consider the origin of linguistic diversity and
variability, and therefore of languages, in a concomitant way with
the origin of the function of language. If the development of our
communication system was since the beginning interlaced with the
social life of the communities, and if this link has been preserved
until today, then it seems likely that the social game on linguistic
forms and the resulting diversity of languages were preserved
throughout prehistory. This phenomenon was presumably
modulated by a large number of parameters: size of the communities,
expressiveness of early forms of languages, development of
underlying cognitive capacities, etc. Moreover, the geographic
distribution of populations presumably contributed to a very ancient
diversification of the communication systems, in a similar way with
the evolution of species.
In this conceptual framework, the development of modern
languages has to be integrated into the more general evolution of the
human communication system, and only represents a “last step”. It
therefore becomes interesting to wonder what defines the modernity
of contemporary languages, in other words what differentiates them
from more archaic languages; one may refer here for example to
Bickerton’s notion of proto-language (Bickerton, 1990) or to Coupé
and Hombert (2002)’s proposals regarding “language” in the
context of the first sea-crossings to Australia.
Linguistic components
The diversity of the world’s languages is naturally expressed by the
differences between the structures and elements that compose these
languages. Linguistic typology aims at classifying this variety of
forms, which appear more or less frequently and in more or less
independent ways; linguists often use the terms of (implicational)
universals or tendencies. An obvious method to study the evolution
of linguistic diversity is therefore to rely on a partial individuation of
the linguistic forms, rather than studying languages as monolithic
entities. Following this line, our work relies on the notion of
linguistic items, as defined by Nettle (1999a: 5):
Polygenesis of Linguistic Strategies
A linguistic item is any piece of structure that can be
independently learned and therefore transmitted from
one speaker to another, or from one language to
another. Words are the most obvious linguistic items,
but sounds and phonological processes are items too, as
are grammatical patterns and constructions. . . The
distributions of different items in the world's languages
need not be statistically independent, and indeed very
often are not.
It seems relevant for us to consider linguistic items as
communicative tools. We will use the term linguistic strategies to
reflect the fact that linguistic items before all address functional
needs at a cognitive level: the typological elements represent
different possible solutions or strategies to assemble and bridge
external linguistic projections of mental representations:
• word-order or case-markers to express the thematic
relationships between the syntagms of the sentence;
• phonemes to encode the acoustic forms of words and
overcome the large variability of phonetic forms;
• words as conventions about meanings, etc.
While the term linguistic strategies has been defined and used,
for example by Croft (1990: 27), in a typological context, the notion
of strategy refers for us to the multiplicity of possible functions for
the projection from the private cognitive level to the linguistic level,
and the competition that may exist among them.
In this line of thought, we may also insist on the fact that the
emergence of the cognitive functions themselves has to be considered
together with the corresponding linguistic strategies, and that the
“grain” of these evolutions may not be as coarse as a single division
between a syntactic and a non-syntactic state. Focusing on the
precise linguistic correlates of phenomena like an increase in the size
of the working memory (Baddeley, 1986), the emergence of a theory
of mind or the development of inferential reasoning (Sperber, 1995)
seems highly relevant.
7
8
Language Acquisition, Change and Emergence
From these few comments, it should be clear that we highly
favor a progressive and segmented emergence of the function of
language and languages, rather than a single step from a simpler
stage (iconic, non-syntactic, etc.) to a fully modern one. As we have
mentioned already, this article will mostly be concerned with the
spatial and temporal characteristics of the various steps that have
led to the current linguistic situation, and we will propose plausible
hypotheses regarding these aspects in the final discussion.
1.2
Monogenesis or polygenesis of language
Monogenesis versus polygenesis of an innovation
Two scenarios can be put forward to describe the appearance of any
innovation. The first one is called monogenesis and corresponds to a
single emergence of the innovation, possibly followed by its spread
in a population. Several independent appearances define the second
possible scenario of emergence, namely polygenesis. In this case, the
different emergences take place at several distinct sites, provided
that innovations only appear once at a site.
At least two major cultural innovations of our species seem to
have appeared by polygenesis. First, agriculture is believed to have
emerged independently in different places around 10,000 years ago,
with archaeological proofs offering similar dates but separated by
thousands of kilometers. Various regions, such as Mexico,
New-Guinea, Europe, the Near-East and China display such
evidence. More recently, the development of two or perhaps three
different writing systems also points to a polygenesis: the Chinese
ideographic system around 3.500 years ago, with inscriptions on
bones or turtle shells, and the cuneiform system of the
Mesopotamians around 5,000 years ago (Wang, 1973: 50-52) seem
too distinct and far apart to have originated from a common origin1.
Such questions of course become more difficult to answer for more
1
The possible links between the Egyptian and Mesopotamian systems are hard
to trace.
Polygenesis of Linguistic Strategies
remote innovations, like for example the domestication of fire
around 500,000 years ago.
In the linguistic framework of our topic, it seems natural to
wonder about a possible polygenesis of “language”, especially if one
considers it as a cultural product like agriculture or writing. To be
more precise, it is possible to focus on either the emergence of the
function of language or the different linguistic components. While
we will not further consider the former question and its
archaeological bases, we will try to defend the latter approach as the
most relevant to study the development of linguistic diversity and
languages.
To follow this guideline, we shall begin by introducing some
general ideas about the origin of current languages.
The monogenesis of languages
It is often more or less explicitly admitted that all modern languages
originate from a single original language; this hypothesis is often
described by the term of monogenesis of languages. This proposal
interestingly gathers researchers who oppose themselves on the
possibility to reconstruct languages spoken before 8,000 or 10,000
years ago. Some proponents of strong limitations of the
methodology of historical linguistics do not reject the plausibility of
a single ancestral language, but think that its content is beyond
reach of our current investigations.
This statement is much more a hypothesis than something
strongly demonstrated and validated. To this extent, the following
points should be recalled:
First of all, the principle of reconstruction itself introduces an
implicit bias toward a single ancestor for all contemporary
languages. The forms reconstructed from the current states are,
above all, “terms of abstract comparisons”, as translated from
(Auroux, 2001), and therefore do not necessarily represent the
linguistic reality of the past. It is for example rather difficult and
unreasonable to conclude that Proto-Indo-European was the only
language spoken in Europe or Western Asia around 6,000 years ago.
Such a position would rule out the common dialectal variety of
9
10
Language Acquisition, Change and Emergence
languages, as well as the possible existence of now extinct languages.
If the linguistic context suggested by the reconstructions is not the
one which really took place some thousand years ago, it becomes
dangerous to rely on a recurrent application of the process of
linguistic reconstruction to conclude a decrease in the number of
languages spoken by our ancestors as we go further back in the past.
A smaller size of the meta-population in the past might have resulted
in many fewer languages, but this argument is not sufficient to
conclude the existence of a unique tree for all contemporary
languages.
A crucial event that plausibly separates the more recent
Neolithic period from the Paleolithic situation is the demographic
explosion which took place with the development of agriculture,
after a slow initial growth at the end of the Paleolithic. The notion
of punctuated equilibria, borrowed from the evolutionary theories in
biology and introduced by Dixon (1997) in linguistics, can be
applied to such transitions (ibid: 77). Such approaches point to the
hidden ties between linguistic diversity and demographic contexts.
Some researchers have begun to explore their relationship, either
with computational models studying the role of population size
(Nettle, 1999b), or considerations about the correlations between
densities of speakers and linguistic diversity (Jacquesson, 2001).
A second point is that assimilating a monogenesis of today’s
languages and the existence of a unique ancestor can be viewed as
incorrect. As Figure 2 depicts it, several distinct families could have
indeed appeared independently and evolved before all languages but
one disappeared. The remaining specimen would have given birth to
all modern languages. This scenario is not implausible, since
languages often disappear, for example in case of contacts with
unbalanced social relationships between populations (Nettle and
Romaine, 2000:147-9). Unbalances may have been even stronger in
the past due to the small size of populations (Marsico et al., 2000).
However, given the large areas which have been inhabited by
humans for tens of thousands of years, it seems unlikely that the
descendants of several “initial languages” could have all disappeared
but one. Large areas like Europe have seen the expansion of families
Polygenesis of Linguistic Strategies
Figure 2
Polygenesis of languages and single origin of contemporary languages
like Indo-European, but such developments and replacements did
not reach a world scale in recent times2.
A third argument relies on the link between the origin of
contemporary languages and the origin of our species between
100,000 and 200,000 years ago. Two main hypotheses are still
discussed to account for the origin of modern Man. The first one,
the Out of Africa hypothesis, postulates that the speciation event
which led to our species took place in East Africa, and that our
ancestors subsequently migrated out of this region and colonized the
entire Earth (Lahr and Foley, 1994), replacing all the previous
species that were living in Africa, Europe or Asia. Based on
congruent archaeological and population genetic data (Stringer and
Andrews, 1988; Cann et al., 1987), this scenario gathers the favors
of most scholars in its opposition to the Multiregional Continuity
hypothesis. This second proposal, based on Asian fossils and other
genetic studies, argues that modern humans evolved from
2
One may argue that if Homo sapiens replaced all other Homo species, a
single language could as likely have replaced all other languages. However,
where in the first case the replacement may be due to a physiological or
cognitive advantage, linguists agree that no language is functionally better
than others, which may be partially extrapolated back in the past for modern
humans.
11
12
Language Acquisition, Change and Emergence
pre-sapiens species locally, and that genetic fluxes were dense
enough to preserve a single species despite the large geographical
areas implied (Thorne and Wolpoff, 1992).
In the framework of the Out of Africa theory, an emergence of
modern languages, and some may even say of the true function of
language, due to neuro-physiological changes as part of the
speciation event, would have taken place in the small and
geographically restricted population of “new-born” sapiens. A
monogenesis would then have been more likely, followed by the
spread of languages during the migrations leaving East Africa. On
the other hand, the link between new cognitive abilities and new
forms of language would rather lead to a polygenesis in the case of
the Multiregional Continuity.
“Rethinking” the origins of modern languages
The aim of our work is not to reject the hypothesis of the
monogenesis of modern languages as a whole, but to point at the
complexity of this question, and to show how its nuances may
partially empty the current consensus from its substance. Our goal is
to propose a general sketch of the emergence of linguistic diversity,
which could in particular be applied to the development of modern
languages.
Our initial statement is that it seems often implicitly assumed
that the putative ancestor of modern languages shares with them
their degree of complexity and structures. In opposition to
proponents of a catastrophic emergence of the function of language,
many scholars do not deny the evolution of our system of
communication (e.g. protolanguage), but do not apply this
evolutionary way of thinking to the history of modern languages.
This is of course partly due to the fact that people working on the
origin of “language” often do not tackle the prehistory of modern
languages to this end, and vice-versa.
We aim at bridging this gap by considering that a large number
of differences may have existed between modern languages and their
first ancestors in modern Man. We will defend a scenario where part
of the current linguistic diversity would have emerged during the
Polygenesis of Linguistic Strategies
several tens of thousands of years that have followed the emergence
of our species. The second and third parts of this article will be
devoted to this goal, introducing probabilistic and computational
models and experiments abstracting the appearance of cultural
innovations in group-fragmented populations, as well as
considerations about palaeo-demographic and evolutionary data.
But before beginning to describe our arguments, these introductory
paragraphs shall be concluded with a brief discussion of some of the
palaeo-demographic data on which we will rely hereafter.
1.3
Taking palaeodemography into account
Emergence in a population of individuals
The appearance of any new cultural feature in the human species
must be considered in a realistic framework, partly built on the
specificities of the human population at the period involved. More
precisely, if we want to talk about “sites of emergence”, as in the
definition of polygenesis mentioned above, a clear definition of the
meaning of these words in a demographic context is necessary. We
shall therefore succinctly describe the data and theories about the
societal structures during prehistory.
The structure of the ancestral human population
Several sources of data point at various cues regarding the structure
and the size of past populations.
First of all, palaeo-anthropology and archaeology, by the study
of characteristics of prehistoric living places such as surface,
organization etc., lead us to conclude that the human population
was composed of small groups of some tens of individuals, mostly
between 20 and 50, during most parts of the Paleolithic (Hassan,
1981: 93-94). The number of 25 individuals is regularly quoted in
various studies on prehistoric populations, and appears to be
independent of time, density or type of environment (ibid: 53).
Comparison to recent populations of hunter-gatherers
inhabiting various ecosystems, e.g. Eskimos with either a caribou
13
14
Language Acquisition, Change and Emergence
and sea mammal hunting economy or a caribou hunting and fishing
economy, also gives clues about densities during prehistory. Table 1
is reproduced from (Hassan, 1981: 198) and summarizes Birdsell ‘s
proposals in 1972 for three successive periods of the Paleolithic.
Table 1
Estimates of world prehistoric population,
reproduced from (Hassan, 1981:198)
Pop. Density
(persons/km²)
Area
occupied
(1e6 km)
World pop.
(1e6
persons)
Lower Palaeolithic
0.015
27.0
ca. 0.4
Middle Palaeolithic
0.032
38.3
ca. 1.0
Upper Palaeolithic
0.039
57.5
ca. 2.2
Period
The total surface that was inhabited by the meta-population of
humans can be estimated by the repartition of the prehistoric living
places, as well as the carrying capacities of various environments:
forests, cold or warm deserts etc. (Bocquet-Appel and Demars,
2000). Australia or the Americas were, for example, only very lately
colonized by our own species (and not by former Homo or
pre-Homo species).
Beyond these first data, population genetics studies propose
some evaluations of the global population during the last million
years. All the studies conclude a very small population of one or two
million people at most, and the existence of a genetic “bottleneck”
around 1,800,000 years ago with the speciation leading to Homo
ergaster. However, their analysis of different genetic markers
(mtDNA (Sherry et al., 1994), Alu insertions3 (Sherry et al., 1997),
micro satellites (Zhivotovsky et al., 2000), etc.) feed the debate
about a possible second bottleneck 100,000 years ago and later
3
Alu insertions are primate-specific genetic elements that mobilize via the
process of retroposition. They are believed to be non-coding, and account for
around 5% of the human genome by mass (which represents 500,000 Alu
sequences) (Sherry, 1997).
Polygenesis of Linguistic Strategies
expansions, with the appearance of our species Homo sapiens
(Hawks et al., 2000). These disagreements exemplify the genetic side
of the controversy between the Out of Africa hypothesis and
multiregional continuity.
From these data, it appears clear that the study of the
monogenesis or polygenesis of cultural innovations, whether
linguistic or not, can be based on the human group as a relevant
“base unit”, at least until the appearance of larger communities with
the development of agriculture around 9,000 years ago.
Table 2
Areas of various lands or continents
Area in km²
entire Earth
510,072,200
emerged lands
148,939,800
Asia
44,547,800
Africa
30,043,900
Europe
10,404,000
Table 3
Density of human groups for different areas and population sizes
Macro-population
size
10,000 25,000 125,000 250,000 1,250,000 2,500,000 5,000,000
Surface (km²) / nb.
of groups
400
1,000
5,000
10,000
50,000
1,000,000
4e-4
1e-3
5,000,000
8e-5
2e-4
10,000,000
4e-5
25,000,000
50,000,000
100,000,000
100,000
200,000
5e-3
0.01
0.05
0.1
0.2
1e-3
2e-3
0.01
0.02
0.04
1e-4
5e-4
1e-3
5e-3
0.01
0.02
1.6e-5
4e-5
2e-4
4e-4
2e-3
4e-3
8e-3
8e-6
2e-5
1e-4
2e-4
1e-3
2e-3
4e-3
4e-6
1e-5
5e-5
1e-4
5e-4
1e-3
2e-3
15
16
Language Acquisition, Change and Emergence
To get a better idea of the quantitative values that are involved
in this framework, Tables 2 and 3 report the current sizes of land
and the continents, and computations of the densities of human
groups for different areas and sizes of the macro-population.
In the next experiments and hypotheses, we will mostly
investigate densities of population varying between 0.001 and
0.0001 human groups per km². They correspond to average values
from 0.025 to 0.0025 individuals per km². While they will appear
small compared to the values of Table 1 (especially the lower
bound), we will explain later why we believe such values to be
relevant in some situations.
2. Mathematical Models and Computer Simulations
2.1. A mathematical model to evaluate the probabilities
of monogenesis or polygenesis of language
Quantifying the probabilities of monogenesis or polygenesis
Except in a few cases for which deciding between monogenesis or
polygenesis of an innovation is possible without ambiguity (as we
have seen for the development of agriculture or writing systems), it
becomes harder to estimate the “mode” of emergence of an
innovation when one goes further in the past and clues become rarer.
The main difficulty, as we will see in the next paragraphs, lies in the
possibility of an undetected diffusion of the innovation, which might
lead to the wrong conclusion of a polygenesis.
If concretely distinguishing monogenesis or polygenesis is a
difficult task in some cases, it is however possible to compute the
probabilities of these events. Indeed, if a mathematical model allows
one to conclude that polygenesis of an innovation is much more
likely to have taken place than monogenesis, it becomes relevant for
the theories on this matter to consider the two possibilities, and not
to reject the polygenetic hypothesis without a strong argument. This
conception is even more crucial as definitive proofs are lacking when
one is interested in the origins of the faculty of language or of
languages.
Polygenesis of Linguistic Strategies
Description of Freedman and Wang’s model
Freedman and Wang (1994) have been interested in the possibility
of studying the probabilities of the two scenarios of monogenesis or
polygenesis. To this end, they have proposed a purely mathematical
model that we are now going to describe briefly. One of the targets
of the paper was to reformulate correctly the “folk” intuition which
assumes that if a rare event is already unlikely to happen once, it
will be even less likely to occur twice.
The model focuses on the link which unites the probability p of
the emergence of language4 at one site and the probabilities of no
emergence, monogenesis or polygenesis at n independent sites over a
fixed period of time. The mathematical approach enabling the
calculation of the probabilities depends on Poisson’s probability
distribution, which characterizes the occurrences of rare events, that
is those with a weak probability of occurrence. The values studied
for the probability of emergence at one site p are chosen such that
the expected number of sites at which language emerges is 1, 2, 3,
etc. The probability of emergence is integrated over the entire time
period, and the expected number of sites is therefore equal to p × n.
It should be clear here that this expected number reflects a
statistical approach, i.e. the mean number of sites at which language
emerges if one considers a large number of episodes. By episode, we
mean a concrete instantiation of the model described above, at the
end of which one of the three possible scenarios has occurred: no
emergence, monogenesis or polygenesis. The model does not predict
the outcome of a single episode, but rather, of a large number of
episodes, the percentages of them which end up as no emergence,
monogenesis or polygenesis. The issue is then to investigate how the
probabilities of the three scenarios evolve with the product p × n.
To ease the understanding of the simulations we develop in the
remaining parts of this chapter, we find it useful here to slightly
4
The authors were interested in the emergence of language 1 or 2 million
years ago. This rather relates to the notion of function of language, but the
model may in fact be applied to any cultural innovation.
17
18
Language Acquisition, Change and Emergence
modify Freedman and Wang’s model, in a way that does not change
any result or interpretation: we modify the meaning of the
probability p, which was integrated over a whole time period in the
original model, to introduce the notion of time step. Instead of
considering an indivisible time period, we consider a number T of
time steps, and p the probability of emergence at each time step.
This new approach can therefore be seen as the discrete counterpart
of Freedman and Wang’s continuous model, since the global time
period they considered has been cut into discrete time units. During
an episode, at each time step, a random test is performed against the
probability p to check whether the innovation emerges or not,
provided that it has not emerged before. With this new framework,
it turns out that the relevant parameter, i.e. the expected number of
sites, can be replaced by the mean number of times the test against
the probability p is positive in the n groups during the T time steps.
This new parameter is equal to the product λ = p × n × T. The two
parameters are in fact similar: the percentages of the three scenarios
(no emergence, monogenesis or polygenesis) will be identical
whether the expected number of sites is equal to a value v, or if λ =
p × n × T is equal to this same value.
One should notice that since only the value of the product λ is
relevant, the time variable and the probability of emergence at one
site are not independent; this relationship allows us to focus on the
relevant values of p given T: if the values of p are chosen such that
the product λ is very weak (well below 1.0), then the probabilities of
monogenesis or polygenesis will be insignificant. For values of λ well
above 1.0 (for example larger than 10.0), the probability of
polygenesis will be close to 1.0, and the two other probabilities
insignificant. In these two extreme situations, no qualitative
transitions would be observed in the probabilities of the “modes” of
emergence.
Results
The graph of Figure 3 provides a good understanding of the
situation. For various values of λ = p × n × T, the probabilities that
Polygenesis of Linguistic Strategies
zero, one (monogenesis) or several emergences (polygenesis) take
place are displayed.
Figure 3
Evolution of the probabilities of monogenesis and polygenesis
of an innovation at several sites according to the probability of emergence
at one site; adapted from Freedman and Wang (1996)
The evolution of the curves representing the probabilities of no
emergence, monogenesis or polygenesis can be summarized in the
following way:
• The probability pn of no emergence decreases to 0 as λ
increases according to the relation pn = exp(−λ) ;
• The probability of monogenesis p m first increases for
small values of the product, and then decreases to 0 (bell
curve). p m = λ.exp(−λ) ;
• The probability of polygenesis pp increases as the product
increases, with the relation pp = 1 − (λ + 1).exp(−λ) .
The last two behaviors lead to the existence of a threshold for the
probability of emergence at one site (depending on the number of
sites n and T), over which polygenesis becomes more likely than
monogenesis. In other terms, large values of p, n or T increase the
19
20
Language Acquisition, Change and Emergence
likelihood of a polygenesis of the innovation, and make it more
likely than monogenesis over a given threshold.
As a consequence, it is necessary to reformulate the intuition
according to which a weak probability of emergence at one site
makes the emergence at several sites even less frequent, since this
intuition does not take into account the combinatory function of the
probabilities at several sites.
It should be pointed out here that it is nearly impossible to
estimate the probability of emergence at one site, which widens the
gap between the model and real situations. Nevertheless, the
knowledge a posteriori of the situation may contribute toward a
better estimation of this probability with the help of conditional
probabilities: as the authors suggest it, for small values of p, the
model does not account for the fact that the innovation has emerged.
This is especially meaningful in the case of linguistic components or
the faculty of language, because all human populations possess
languages with some recurrent linguistic items, a situation to be
contrasted with those of agriculture or writing. This may falsely lead
to the rejection of monogenesis; however, it remains impossible to
conclude polygenesis or monogenesis for a single episode (see
definition above), since for a sample of size 1, an unlikely event like
monogenesis cannot be ruled out. The fact that for small values of p
the model does not account for the emergence of the innovation
does not interfere with this issue, but is rather a limitation of the
model itself. As a conclusion, polygenesis appears more likely than
monogenesis for a large range of probabilities of emergence at one
site. It becomes even more likely as the number of sites increases for
a fixed value of p.
In this model, sites remain abstract entities and are, in particular,
totally independent from each other. But, as we have seen, the
reality of prehistory is partly represented by human groups moving
in large geographical areas, and as a consequence able to enter into
contact and transmit cultural innovations. The frequency of such
contacts is hard to estimate intuitively. Freedman and Wang’s model
therefore calls for further enhancements to take this aspect into
account.
Polygenesis of Linguistic Strategies
2.2 Measuring the frequency of contacts
Relevant parameters
Estimating the frequency of contact between human groups
knowing their density in a geographical area is a rather unintuitive
task which asks for precise calculations. Were they meeting every
month, every six months, every ten years or even less often? Of
course, just the density of groups, represented hereafter by the
variable de, is not enough, and other parameters have to be
considered. We shall restrict our attention to three of them:
• The threshold distance d between two groups for a contact
to occur (when does one group detect another one?);
• The “geometrical” features of the groups’ movements;
• The speed of the groups v.
The first parameter may be linked to the surface of the catchment
territory, which may be roughly equated to the region where a
second group could be detected, not taking such clues as distant
sounds or smoke into account. The size of such territories for
hunting or gathering food is estimated to be around several hundred
square kilometers, as indicated by ethnographic studies on various
groups of hunter-gatherers (Biraben, 1997: 46–7). Regarding the
geometrical “features” of the movements, it is reasonable to assume
more or less directional migrations, from fully random
“brownian-like” trajectories to extremely rectilinear motions,
although the latter seem less likely. The speed of displacement may
be approximated to several kilometers per year. Dates of
archaeological sites related to the migration of the first farmers in
Europe 10,000 years ago suggest a speed of 1 kilometer per year for
the population wave (Cavalli-Sforza, 1994: 108–9). However, this
net distance would only equal the average speed of human groups in
the case of unidirectional movement in the same direction as the
wave of migration (Hassan, 1981: 200–1). For more random
movements, the average speed is necessarily higher. Moreover, tribes
of hunter-gatherers were presumably moving faster than farmers
cultivating the ground, given their reduced sedentarization.
21
22
Language Acquisition, Change and Emergence
First theoretical approach
In the case of some specific movements, a useful analogy allows
estimating the frequency of contact between groups. Although we
will not enter here into the details of the mathematical method
(Coupé, 2003), we may just summarize the approach by pointing at
the similarities between the collisions of molecules in a perfect gas
and the contacts between human groups: the diameter of the circular
catchment territory of a human group can indeed be assimilated to
the diameter of a molecule. We rely on this similarity to estimate the
frequency of contacts in the case of pseudo-rectilinear movements
(directional changes only take place after a collision of molecules in
a perfect gas). The frequency of contact f for the molecules is
approximated by the following formula:
f = de × d × v
The numerical application with a density equal to 0.001 groups
per km², a speed of displacement of 4 km/year and a radius of 10
km for the hunting territory leads to a frequency of around one
contact every 12.5 years. The frequency of contact for human
groups seems therefore at first extremely low. This may be
moderated by differences in regional densities, but one decade seems
a reasonable magnitude for the period between two contacts. To
assert whether such values are reasonable, we will further ground
our approach with computer simulations.
Simulations for rectilinear motions
We programmed a simple computational multi-agent model to
measure the frequency of contact experimentally. Agents
representing human groups were allowed to move in a square
bi-dimensional space (with pseudo-rebounds on the frontiers of this
space and an initial random distribution) during a large period of
time; the number of contacts between them was averaged over this
period. The parameters varied as follows:
Polygenesis of Linguistic Strategies
• Number of agents n: 50, 100, 200 and 400. Additional
values were also considered in specific cases (see Figure 5);
• Density of groups: de: 0.001, 0.0005 and 0.0001 groups
per km². The space area was computed given de and n;
• Speed v: 1, 5, 10, 15 or 20 km per year;
• Threshold distance for contact d: 5, 10, 15 and 20 km.
To estimate how close to the theoretical case the experimental
results were, we computed the ratio of the experimental frequency
of contact f over the theoretical frequency f th for various values of
the parameters. Figures 4 displays curves for various values of the
speed and size of the catchment territory, while Figures 5 and 6
allow estimating the impact of the density and number of groups.
In addition to these graphs, various correlations were computed
between series of values of f for different values of the parameters, in
order to discover mathematical relationships. This approach was
applied to all our simulation results and proved to be useful, as will
appear later in this chapter.
Figure 4
Evolution of the frequency of contacts in function of the speed
of the agents and the threshold distance for contact
(speed in horizontal axis)
23
24
Language Acquisition, Change and Emergence
Figure 5
Evolution of the frequency of contacts
in function of the number of human groups
Frequency of contact (density: 0,005 groups/km²,
threshold distance for contact: 15 km, speed: 10 km/year)
Figure 6
Evolution of the frequency of contacts
in function of the density of human groups
Polygenesis of Linguistic Strategies
The following results can be drawn from the figures above,
other graphs that were not included, and analyses of the series of
results:
• The density of groups d does not interact with the other
parameters. Moreover, the curves displayed in Figure 6
show that the ratio of frequencies f / fth remains relatively
stable whatever the value of the density of groups. This
means that the frequency of contact f is linearly related to
the density of groups;
• The number of groups n does not interact with the other
parameters, and the curve in Figure 5 shows that the ratio
of frequency seems to reach an asymptote for large values
of n Since relevant values of n are much larger than 1000,
f can be considered as independent from n in a first
approximation;
• The frequency of contact f can be described by the
following formula expressing the independence or
interactions between the various parameters: f = d x v x de
x g(d, v). For the values of interest to us, g takes values
between 0.5 and 3.
It then appears that the former conclusions should be moderated, in
that the experimental frequencies of contact f may be different from
the theoretical frequency f th with ratios from 0.5 to 3. However,
even with this multiplicative factor, the frequency of contact can be
said to remain very small, especially compared to the life span of an
individual during prehistory.
Three reasons may be invoked to explain the discrepancies
between the theoretical and experimental cases:.
Two reasons may be invoked to explain the discrepancies
between the theoretical and experimental cases:
• The theoretical formula approximates the reality by
considering a molecule moving in a space containing static
elements it can hit; considering a high velocity for these
elements (according to a molecular speed distribution)
25
26
Language Acquisition, Change and Emergence
leads to refining the formula by adding a √2 multiplicative
factor in the right expression: f = √2 × de × d × v. This
specific numerical factor is adapted to the speed
distribution of molecules, but the general principle may
explain why the ratios of frequencies we observed were
mostly greater than 1.0.
• In the theoretical case, molecules move continuously,
where in our case, the groups jump from one location to
the next. For a small speed and a large size of the
catchment territory, the area covered by a group after a
given time span is close to the area covered in the case of a
continuous motion, since the areas covered by the group at
each time step largely overlap. However, for larger speed
and smaller catchment territory, the motion cannot be
assimilated to a continuous one, since the regions covered
at each time step only slightly overlap or do not overlap at
all. Such behavior modifies the frequency of contact, since
on the one hand, a larger area is covered if the catchment
territories overlap less, and on the other hand, close
groups may miss each other in case of distant jumps from
one area to another.
• The theoretical case assumes an infinite space, whereas our
agents were moving in a closed area. For agents close to
the boundaries of this area, the frequency of contact is
smaller than for agents in the center of the space. The
larger the size of the area, the closer the experiments get to
the theoretical case.
Non rectilinear motions
For more random movements, the total area covered by a group
during a period of time is intuitively smaller, since this group will
more frequently revisit the same places, by often changing its
direction. However, it remains unclear whether this affects the
frequency of contact: groups may reduce their chance of meeting
distant groups through more local movements, but this in turn
Polygenesis of Linguistic Strategies
increases their chance of meeting close groups again and again5.
Once again, we relied on computer simulations to evaluate the
situation.
To model the notion of linearity of movement, we introduce an
angle α corresponding to the maximum deviation allowed for the
direction of a group at each time step; an angle equal to 0
corresponds to a linear motion, whereas an angle of 2.π is
equivalent to a Brownian motion: at each time step, agents choose a
deviation between – π and π. In the former experiments with
rectilinear motions, the angle was simply set to 0.
We ran experiments for the same values of parameters as above,
crossing them with the following values of α: 2π, π, π/3, π/6 and 0.
As explained earlier, we computed correlations between the series of
values obtained for these various angles, and compared the
numerical values themselves.
It appears that the frequency of contacts f obeys the behavior
described for rectilinear motions whatever the value of α, and that
this angle in fact plays no role in the frequency of contact. We
interpret this result as a balance between the two phenomena that
we introduced three paragraphs ago (distant versus local contacts).
The frequency of contact can therefore still be computed according
to the following formula: f = d x v x de x g(d, v).
Time to complete a diffusion
The former results may falsely lead to the conclusion that the
randomness of movements plays no role in the impact of contacts on
the emergence of an innovation. However, it is not as much the
frequency of contact as the speed of diffusion of this innovation
which will be relevant in the coming paragraphs. We ran a last series
of simulations to measure the time needed for a diffusion to reach all
5
For a Brownian motion in physics, the study of the intersections of two
trajectories of particles (called Wiener’s sausages) is a hard and still unsolved
problem.
27
28
Language Acquisition, Change and Emergence
the groups, once again according to different sets of values of the
former parameters given above. For each set of parameters, the
average time T for the complete diffusion of the innovation (to all
groups, starting from a single group in possession of the innovation)
was measured over 50 identical simulations.
An additional parameter was added to the model: pt represented
the probability that in case of contact, one agent in possession of the
innovation transmits it to a second agent not in possession of it.
It appears obviously that T is not only dependant on d, de and v,
but also on p t , α and n. Indeed, the local aspect of the contacts in
the case of more random movements decreases the speed of
diffusion, and the more agents for a given density, the bigger the
space to conquer for an innovation. The first phenomenon may be
interpreted as a weak coefficient of diffusion in the modeling of
epidemics (Murray, 1994: 651–5).
According to the analyses of the numerical outputs of the
simulations, and by comparing the values of the frequency of
contact f and the time for complete diffusion T for identical sets of
parameters, the following relationship links T to the other variables
of the model:
T = h1(f, n, p t × exp(α.h2(f, p t )
Figures 7 and 8 illustrate the progressive increase of T as α
increases for two sets of values of f and pt . The different shapes of
the exponential tendency curves and their coefficients clearly point
to the second term of the product in the former expression of T.
The functions h1 and h2 required detailed investigations before
their analytical expressions could be found. First, the analysis of
series of values of T for α=0 led to a further decomposition of the
function h1 as follows:
h1(f, n, p t ) =
−β
η
n δ .(f .p)
β +1
(f .p)
The coefficients η, β and δ were then estimated: η = 1.89, β = 0.17
and δ = 0.42.
Polygenesis of Linguistic Strategies
Figure 7
Evolution of the time for complete diffusion in function
of the directionality of the movements (first set of parameters)
Time to complete diffusion, speed: 10 km per year, Threshold distance
for contact: 10 km, density: 0.001 groups per km², pt = 0.7
Figure 8
Evolution of the time for complete diffusion in function
of the directionality of the movements (second set of parameters)
Time to complete diffusion, speed: 10 km per year, Threshold distance
for contact: 5 km, density: 0.001 groups per km², pt = 0.1
29
30
Language Acquisition, Change and Emergence
Paying attention to the coefficients of the exponential tendency
curves for various values of α further led to the following expression
for h2 (f, pt ):
h2(f, p t ) = ε.f .pθ
The coefficients ε and θ were estimated as η, β and δ previously.
The following values were found: ε = 1.21 and θ = 0.69.
The expression found for T is rather complex, and may take
other simpler forms. It may also be simplified for large values of n,
which were however computationally too expensive to be
investigated with the necessary high number of simulation runs.
Table 4 provides various values of f and T for different values of d,
v, de, n, p and α. Numbers in italic are extrapolations, while other
values are experimental results.
Table 4
Values of f and T for different sets of parameters of the model
v
(km/year)
d
(km)
de
(group/km)
n
10
10
0.0001
258
0
0.7
0.022
810
10
10
0.0001
94
0
0.7
0.022
695
8.81
13.69
0.0034
189
5.40
0.107
0.86
179.88
6.94
10.25
0.001
113
1.57
0.047
0.16
1359.94
10
10
0.001
10000
1.5
0.1
0.24
1263
10
10
0.001
1000
1.5
0.1
0.24
756
10
10
0.001
10000
1.5
0.01
0.24
8953
10
10
0.001
1000
1.5
0.01
0.24
6329
5
10
0.0005
10000
3
0.01
0.065
31055
5
10
0.0005
1000
3
0.01
0.065
23524
α (rad)
p
f
T
(years)
On the nature of contacts between human groups
The opportunities for contact between human groups were
presumably very slight. One question that may be raised is the
possibility for these encounters to have been non-violent and sources
of cultural exchanges or transfers of individuals. To partially answer
Polygenesis of Linguistic Strategies
this question, the exchanges of lithic material (Marwick, 2002) are a
first clue of non-aggressive contacts, unless we always consider
fights for the acquisition of these resources. As it appears in the
analyses of databases of parietal art schemes in France and Spain
during the Upper Paleolithic (50,000–10,000 BP) (Sauvet and
Wlodarczyk, 1995), local stylistic heterogeneities in a globally
homogeneous context for the graphic representations strongly
suggest the existence of inter-groups relationships. These
relationships may be at the origin of both the large-scale
homogenization and the preservation of local diversities
contributing to the social position of a group among others.
Finally, exchanges of genes through the exchanges of individuals
would have usefully preserved the diversity of the gene pool of a
group. Such exchanges between groups, namely exogamy, are still
common in hunter-gatherers like the Australian aboriginals.
Welcoming a man or a woman speaking a different language could
presumably play a significant role in the linguistic evolution of a
group.
Following the previous hypotheses raises the next question:
could the contacts between groups have had an impact on the
monogenesis or the polygenesis of innovations? This point is now
going to be further investigated
2.3 Combining independent discovery
and transmission by contact
Intuitive statements
If one follows the results of sections 2.1 and 2.2, a group may either
discover an innovation by itself, or receive it from another group it
meets. Intuitively, the resulting probabilities of monogenesis or
polygenesis will be the result of the interaction between the diffusion
and the emergence of the innovation among the groups: fast
diffusion may prevent polygenesis from occurring, since all the
groups will be contaminated by the innovation by diffusion before
having the possibility to discover it by themselves. Conversely, a
31
32
Language Acquisition, Change and Emergence
slower diffusion will preserve the possibility of polygenesis and the
probabilities of Freedman and Wang’s model will be more likely to
be relevant for the real situation. The next simulations are aimed at
checking whether this intuition is relevant or not.
Description of the new model
To test the hypothesis of the last paragraph, both aspects of
emergence and diffusion of an innovation among human groups
were combined in a single multi-agent model. Contrary to the
former models, it did not involve a spatial environment, and relied
on a simple discrete probabilistic framework involving two
parameters:
• The parameter r related to the probability for an agent to
receive the innovation from groups already in possession
of it;
• The parameter pc , giving the probability for an agent to
discover by itself the innovation at each time unit.
A total number of agents n and a time limit for the simulation
Tmax were set for each run. The relevant value was as in Freedman
and Wang’s model the product λ = pc × n × T . As a consequence,
different number of agents or time limits led to the same results,
since the value of pc was directly deduced from these parameters.
At each time step, two statistical tests were performed for each
agent to decide whether it would discover the innovation by itself,
or receive it from another group. In the first case, a random number
rn1 between 0 and 1 was compared to pc , while in the second case,
a second random number rn2 in the same interval was compared to
the product r times the percentage p of agents already in possession
of the innovation. An agent not in possession of the innovation was
i) discovering the innovation by itself if rn1 < pc , or ii) receiving the
innovation from other groups if rn2 < p.r. According to this second
condition, no diffusion was taking place before a first agent had
discovered the innovation. After this initial discovery, the diffusion
was taking place according to a logistic growth, similar to what had
Polygenesis of Linguistic Strategies
been observed in the previous model of diffusion. This justified the
use of a simpler and computationally cheaper non-spatial model.
The relation between r and T was explored with additional
experiments, and the following result was established:
r=
2 ln (n)
T
This relation found experimentally can in fact be derived from
the equations describing logistic growth, provided that the number
of human groups n is large enough.
Experiments and results
To test our model, large sets of values for the two parameters r and
λ were built, either by crossing random values for both of them or
by fixing one parameter and choosing random values for the second.
For each set of values of λ and r, 200 runs were performed to
measure the probabilities of no emergence, monogenesis and
polygenesis. A run was stopped i) as soon as two independent
discoveries had taken place, ii) when a single discovery had diffused
to the whole population or iii) if the time limit was reached without
any discovery of the innovation.
An initial series of runs was first computed to check the
conformity of the computer model to the theoretical results
established by Freedman and Wang in the case of no diffusion. A
value of r equal to 0 reproduced the theoretical values, modulated
by a small variability due to the limited number of experiments (an
infinite number of simulations should reproduce the mathematical
laws exactly).
In order to evaluate the impact of diffusion on Freedman and
Wang’s model, we investigated the evolution of the ratios of the
experimental probabilities over the theoretical probabilities for no
emergence, monogenesis and polygenesis, respectively rn =
rm =
pexp
n
′o ,
p the
n
pexp
pexp
p
m
and
r
=
. It appeared that the first ratio was always
p
′o
′o
pthe
pthe
m
p
33
34
Language Acquisition, Change and Emergence
taking values very close to 1, which means that the probability of no
emergence is left unaffected by diffusion. This is rather intuitive
since diffusion can only occur when an emergence has already taken
place. In order to evaluate the evolution of the two other ratios, we
relied on the following intuitions: i) large values of r should decrease
the ratio rp until it eventually reaches 0: an extremely fast diffusion
prevents polygenesis, ii) very weak values of r should give a ratio
close to 1, since there is then virtually no diffusion, iii) the larger the
probability pc and λ, the closer to 1 the ratio rp : groups discover
the innovation by themselves very quickly, which prevents diffusion
to occur, iv) before and after a phase of transition for the value of
the ratio rp , variations in r or λ should have minimal impact. A
possible relationship between them and rp could be:
rp =
1 − exp(−Cste.f (λ, r))
1
= tanh( Cste.f (λ, r))
1 + exp(−Cste.f (λ, r))
2
These hypotheses led us to compute for each set of values the
expression − ln(
1 − rp
1 + rp
) , and compare it with the corresponding values
of r and λ. We found a very strong positive correlation between the
ratio
λ
.and the former expression, and the following relation was
r
established:
rp = tanh(9,32e − 5.λ / r)
Finally, the ratio rm =
pexp
m
′o could be derived simply from the
pthe
m
knowledge of rn and rp .
Analysis
From the previous computational study, we are able to propose a
relation between the probabilities of monogenesis pm and
polygenesis pp , the number of groups N, the expected number of
groups in which the innovation appears when no diffusion occurs, λ,
Polygenesis of Linguistic Strategies
and the time T for the complete diffusion of the innovation in the
population (this last parameter can itself be decomposed as seen
previously):
rp = tanh(9,32e − 5.λ / r)
λ.T
).(1 − exp(−λ) − λ.exp(−λ))
2.ln(n)
λ.T
).(1 − exp(−λ) − λ.exp(−λ))
⇒ pm = 1 − exp(−λ) − tanh(9,32e − 5.
2.ln(n)
⇒ pp = tanh(9,32e − 5.
As we have already made clear, it is often not possible to know
the probability of emergence of an innovation at one site; in the
former expressions, the parameter λ remains an unknown variable.
However, we are able to compute the probabilities of monogenesis
and polygenesis for various values of this parameter, n and T. The
following array summarizes values of the ratio
pp
pm
for various values
of these parameters. The values of T are identical to the values that
were extrapolated in Table 4, and therefore may be related to
precise values of the basic parameters d, v, de, p and α. Tmax was
chosen equal to 50,000 years.
Table 5
pp
Values of the ratio p for various values of λ, n and T
m
T, n / λ
2
5
10
50
100
3.14e-5 0.0027
0.0089
0.032
0.068
0.45
1.29
2.5e-5 0.0021
0.0071
0.025
0.054
0.33
0.89
8953 years, 10.000
2.2e-4
0.019
0.066
0.27
0.74
45.9
4300
6329 years,
1.000
2.1e-4
0.018
0.062
0.25
0.67
35.2
2555
31055 years, 10.000
7.7e-4
0.070
0.26
1.73
11
3.3e6
2.2e13
23524 years,
4.7e-4
0.041
0.15
0.75
2.85
6822
9,3e7
0.1
1263 years, 10.000
756 years,
1.000
1.000
1
As we already mentioned it in section 1.3, we have investigated
values of the density which appear to be small compared to
35
36
Language Acquisition, Change and Emergence
estimations by palaeo-anthropologists. However, a significant
difference between our simple model of diffusion and the reality is
that in the former, agents could move without constraints. In reality,
natural barriers like warm or cold deserts, mountains, oceans, etc.
may have significantly increased the time for diffusion to the whole
population. If one admits that groups were living in restricted areas
(along rivers or lakes or seas, near abundant resources, etc.), the
transmission of innovations between these regions may have taken a
long time6. We simulated this effect by considering smaller densities
of groups. We then propose that the estimated densities for
prehistory correspond to quite long periods for complete diffusion
increases the likelihood of polygenesis.
It appears finally that for values of the parameters that are
congruent with our knowledge of prehistory, the probability of
polygenesis may still be comparable with or higher than the
probability of monogenesis for small values of λ (for example 5
expected sites of emergence with 10,000 human groups). If one
deals with a large number of innovations which value of λ is large
enough, it becomes statistically very likely that some of these
innovations will emerge by polygenesis, and others by monogenesis.
To rephrase this conclusion in our linguistic framework, if we
assume a λ large enough for modern linguistic strategies, the model
predicts that at least some of them appeared by polygenesis, possibly
the majority.
In the third and last part of this article, in order to further assess
the results of our simulations, we will try to adopt a broader
perspective concerning “abstract” models and their implicit
statements, as we will try to place the polygenesis or monogenesis of
linguistic innovations in a context centered on the cognitive
capacities of our ancestors.
6
Extreme cases may be Australia, which required significant sea-crossings to
be reached, or the Americas, which were only accessible through the
northern path between Siberia and Alaska.
Polygenesis of Linguistic Strategies
3. Discussion
3.1. Cognitive potential and structural polygenesis
An implicit assumption behind the former results
The model we have described can be applied to any kind of cultural
innovation. As we have mentioned in the first part of the article, we
think that linguistic items or “linguistic strategies” are the best
candidates when it comes to the evolution of languages and the
development of linguistic diversity. We propose indeed to view them
as cultural innovations, and to apply the results of our model to
write the first draft of a scenario of the development of linguistic
diversity, and of modern languages. However, doing this requires a
better understanding of the probability of emergence at one site for
linguistic components.
A hidden assumption hides behind the former results of the
model: the probability of emergence of the innovation has to remain
constant during all the time period considered. Interestingly,
analyzing this assumption leads to various comments that may
enrich our point of view.
A first aspect is to be related to the putative link we have
already mentioned between the emergence of our species and the
emergence of modern languages. Our entire former discussion
becomes irrelevant if one cannot assume an approximately constant
probability of emergence during the large period of time T
considered (typically tens of thousands of years). This is exactly
what happens if one assumes that the emergence of our species has
to be followed immediately by the emergence of the innovations: all
the strategies appear very shortly after the emergence of our species.
If this emergence is very localized, as in the Out of Africa theory,
then monogenesis is very likely. We will try to refute this hypothesis
by introducing the notion of cognitive potential.
Another objection that may be raised against a constant
probability of emergence at one site is that interactions between
linguistic items create context-dependant probabilities of emergence:
a specific strategy will be more likely to appear in some contexts
made by other pre-existing strategies.
37
38
Language Acquisition, Change and Emergence
The notion of cognitive potential
Evolutionary biologists often quote the following saying: “The
function makes the organ”. This means that an organ will appear to
satisfy a specific functional requirement, and not by random, being
attributed a function. This assumption is of course related to the
Darwinian law of natural selection. However, if a functional
requirement may spawn an organ, this one may also be used for
another function than the one it was first developed for. This
phenomenon, which has received the name of exaptation, is another
possibility to create a link between a physiological device and a
given function. An example is the wing of the bat, which was
formerly an upper leg like for other mammals, but gradually
changed to fill another function. More and more biologists estimate
that most of functions appear as exaptations, rather than with the
primary appearance of an organ. This may especially be the case for
the brain, where neural circuits and structures may have played
different roles during evolution. One may refer here for example to
MacNeilage (1998)’s Frame-Content theory centered on
the
evolution of the Broca’s area.
We may slightly extend this last point of view by saying that the
mechanism of exaptation could be translated to the domain of
higher cognitive functions, and not only restricted to low-level
neural networks. A function could then take advantage of already
existing cognitive mechanisms to become active in the behavior of
an individual. Various authors have already defended this position.
Wang (1991) promoted, for example, the idea of a “mosaic” of
cognitive functions which would have led to the emergence of
language. Writing is a demonstrative example of a cognitive activity
which is not the result of an evolutionary process, but relies on a
collection of cognitive abilities and neural areas which primarily
evolved for other functions.
We propose to use the term “cognitive potential” to describe the
fact that some cognitive functions could potentially emerge given an
organic or cognitive background. At a given time, what would make
them concretely exist or not is a matter of external conditions or
events rather than internal requirements.
Polygenesis of Linguistic Strategies
This notion of potentiality is highly significant for our
discussion, since it implies that the emergence of modern linguistic
strategies did not necessarily take place at the same time as the
emergence of our species. This opens the door to a progressive
emergence of the features of today’s languages, according to cultural
innovations taking place in different human groups. It becomes then
more plausible to assume a relatively constant probability Pc over a
larger period of time, which in turn validates the results and
hypotheses derived from our computational experiments.
What “events” may trigger the emergence of a new linguistic
strategy? This question remains hard to answer, because many
factors have to be taken into account. On the one side, common
linguistic phenomena, such as grammaticalization processes, may
lead to new linguistic forms after a while: primary forms which were
more likely to appear first gradually evolve into more complex states.
Other elements can be related to cultural facts: behaviors like the
need to share cognitive representations about spatially and
temporally distant situations may have led to the emergence of new
linguistic forms, just as many new words have been created
throughout history to name new concepts or tools.
It is interesting to recall here the gap between the emergence of
Homo sapiens and the emergence of behaviors like the first
sea-crossings to Australia around 60,000 BP the religious burial of
the dead 7 or rock painting. If one accepts that several tens of
thousands of years separate the emergence of our species and the
former behaviors, and that such behaviors require specific linguistic
abilities, it is reasonable to assume that linguistic evolution
accompanying these deep changes occurred a long time after the
emergence of our species. It is especially relevant to notice here that
these behaviours appeared when Homo sapiens had already
colonized a significant part of Eurasia and Africa, which implies
7
Archaeologists still debate on the first burials: whether Neanderthals were
burying some of their dead remains very controversial. However, burials
with offerings become obvious during the Upper Palaeolithic (50,000 –
10,000 BP) (Klein, 1999: 468–70, 550–3).
39
40
Language Acquisition, Change and Emergence
large areas and multiple natural barriers that prevent fast diffusions
from occurring.
Structural genesis
Linguistic strategies interact with each other. Some of them are
exclusive or rarely appear together (Greenberg, 1978). One reason
could be that once a cognitive demand is satisfied at a linguistic level,
there is no need to have another strategy for the same purpose.
Case-markers or word-order are such quite redundant strategies.
Others may occur together with a high frequency in today’s
languages.
Some explanations may be found in cognitive constraints or
economy: two strategies may induce the same cognitive operations
and hence save computational time for real speech processing. The
X-bar rule of the Government and Binding Theory, which defines
the position of all heads and specifiers, see (Black, 1999) for
definition, may be explained by such computational savings or costs.
Because of these interactions, all strategies are not as likely to
occur in a context formed by already existing strategies. As a
consequence, instead of considering steady probabilities of
emergence for all strategies during the whole time period,
probabilities at one site should in theory be recomputed at each new
emergence to take the various interactions into account.
Starting from one or several initial forms, families of languages
follow their own pathway, and each new emergence modifies the
possible directions of evolution the system may take. The term
bifurcation, as introduced for example in linguistics by Ehala (1996:
2-3) to describe the evolution of linguistic systems, appropriately
characterizes the fact that each new emergence, according to
external contingent events, forces the system to choose a restricted
path toward some of the initial potential configurations it could
adopt. This is very similar to the evolution of species, where changes
can only occur in the frame defined by the biological characteristics
of the organism (Maturana & Mpodozis 2000).
This specific scheme of interactions between the strategies blurs
Polygenesis of Linguistic Strategies
the situation by adding a lot of conditional relationships between
the various paths that might be taken by a linguistic system from an
initial state. However, one feels intuitively that such structural
constraints do not contradict the main result of the model, which is
that linguistic strategies are likely to emerge according to
polygenesis at various locations. For some of them, the constraints
will increase the probability of emergence at one site, while others
will appear less likely because of pre-existing strategies. We propose
to use the term structural genesis, to summarize the fact that
linguistic innovations appear under structural constraints.
3.2 A scenario for the development of linguistic
diversity and the emergence of modern languages
Given all the previous proposals and hypotheses, we propose the
following sketch for the development of languages:
• Due partially to its social origin, linguistic diversity is as
ancient as what we may call the human function of
language, possibly a few millions of years.
• Since we postulate that linguistic strategies represent
devices to transfer data from a cognitive internal level to
an external and shared one (or vice-versa) relying on
general cognitive abilities, it is reasonable to assume that
these strategies have become increasingly complex with the
evolution of our cognitive capacities in the past, especially
along the several speciation events which constitute the
nodes of our phylogenetic tree.
• Once new cognitive capacities become available, the
linguistic “appropriation” or usage of these capacities is
not immediate. Cognitive potentials appear that may be
instantiated only after a period of time. The evolution of
the probability of emergence at one site and in a given
linguistic state is determined by the nature of the new
cognitive abilities, the benefit for communication of
instantiating the strategy, the structural constraints that
41
42
Language Acquisition, Change and Emergence
weigh on the emergence of the innovation and the cultural
evolution and requirements at a more general level.
• The former assumptions lead to the possibility of
monogenesis or polygenesis of linguistic strategies. Until
recently, around 10,000 years ago, the densities of
population were very low, therefore leading to rather slow
diffusions of linguistic innovations by contact between
human groups. The progressively increasing number of
sites and larger areas colonized seems to favour the
polygenesis of various innovations, as was the case for
both agriculture and writing.
• According to Freedman and Wang’s arguments,
polygenesis is not necessarily less likely than monogenesis
on probabilistic grounds. This proposal has to be revised
because of the possible diffusion of an innovation leading
to the bypass of polygenesis. However, the results of our
experiments and the large number of linguistic strategies
suggest that at least part of these strategies emerged by
polygenesis. In case of a large probability of emergence at
one site, the model predicts that most of the strategies
would have appeared by polygenesis (what the model
cannot predict is exactly what the probability of
emergence at one site is).
If we now turn to the origin of contemporary languages, the
former hypotheses can be reformulated in the following way (if one
assumes the Out of Africa hypothesis):
• Prior to the emergence of our own species, our
predecessors already possessed a function of language.
This function was taking diverse forms at a linguistic level,
according to their cognitive abilities (and physiological
structures). Neither human language nor linguistic
diversity appeared with our species.
• With the emergence of our species, new cognitive
potentials opened the way to new linguistic strategies.
However, these strategies did not all appear right after the
Polygenesis of Linguistic Strategies
speciation event, but during a long time span of several
tens of thousands of years. The various linguistic strategies
took place according to external events, especially the
more general cultural development of our ancestors.
• Whether there were one or several ancestors to modern
languages soon after the first Homo sapiens, these systems
of communication were less complex than today’s
languages. Assuming a single ancestor for all modern
languages makes little sense if this ancestor only shared a
few of the characteristics of contemporary languages, most
likely the simplest ones, and if most of the modern
linguistic features appeared later in various populations
(partially constrained by the structural interactions
between them, and therefore following different paths
starting from the initial emergence). Considering the
polygenesis of numerous linguistic strategies, or their
monogenesis because of diffusion, leads to partially
emptying this hypothesis of its substance and interest.
Moreover, if the transition from pre-modern to modern
humans took place slowly and in a large number of groups,
various sources could have been at the origin of the
linguistic families that later spread all over the world.
• Trying to reconstruct a unique ancestor to today’s
languages on the basis of comparisons of words or
typological structures can only be validated if alternative
scenarios built on the polygenesis of the considered
features are demonstrated as significantly less likely than
the monogenetic hypothesis.
4. Conclusion
The question of the monogenesis or polygenesis of languages is a
difficult one, because it involves numerous factors as diffuse as the
palaeo-demographic conditions of our ancestors, the mechanisms of
a speciation event, the relationships between general cognitive
43
44
Language Acquisition, Change and Emergence
abilities and linguistic “tools,” etc. Most of our proposals are not
firm demonstrations, but aim nevertheless to shed light on elements
that are often forgotten in current debates. We have aimed to
present this question in the more general framework of the evolution
of linguistic diversity. By putting forward a scenario based on the
possible polygenesis of linguistic strategies over long periods of time,
we conclude that the fact of having a single ancestor to all modern
languages makes little sense, since this ancestor was much simpler
than today’s languages, and since many features evolved later
independently in many human groups scattered all over Earth.
Moreover, we strongly believe that language did not emerge with
our species, and that many linguistic strategies, even already
sophisticated, were used by our former ancestors.
This position seems to run counter to the hypotheses of linguists
such as Merritt Ruhlen (1994), who proposed words of the
proto-sapiens, Homo sapiens’ “first” language, or Murray
Gell-Mann, who defends the fact that there are various visible
arrows of time in today’s languages, one of them being word-order
(see contribution in this volume). Despite the fact that alternative
models based on polygenesis have not been taken into account to
contrast their proposals, their positions do not necessarily contradict
our hypotheses, since some (central) features (such as word-order or
core lexicon?) might have already existed in the language of
pre-modern men8. It is actually reasonable for us to assume that
strategies such as word-order already existed long before the last
100,000 years: as experiments on monkeys have shown, sequential
ordering is an ancient ability (Terrace, 2000), and does not seem to
require highly sophisticated cognitive capacities. This has to be
compared with linguistic strategies requiring more complex internal
representation, such as expressions of time, aspect, mode, causation
(Shibatani and Pardeshi, 2001), etc. This view disagrees with some
scholars’ proposal of a catastrophic transition from a
8
We do not enter here into the debate whether word replacement totally
erases the traces of such ancient lexicons or not (Ringe, 1992).
Polygenesis of Linguistic Strategies
proto-language to a fully syntactic one, but a deeper analysis of the
cognitive loads of various linguistic strategies needs to be done.
Contact between small groups of humans and possible linguistic
innovations play a central role in our proposals. Such emphasis will
appear close to the emphasis on contact in historical linguistics.
These elements are often put forward to criticize the simplified, but
useful model of the Stammbaum. We therefore propose to project
the controversies of the recent evolution of languages to their origins.
As well as for recent contacts and consequent language changes,
these considerations may be useful to refine our knowledge and
reconsider well established theories of the prehistory of languages
that may reveal themselves to be implausible in regard of such
arguments.
Acknowledgments
We thank James Minett for his precious help and useful comments to prepare
this paper and improve its quality.
References
Arbib, Michael A. and Rizzolatti, Giacomo. (1996) Neural expectations: a
possible evolutionary path from manual skills to language.
Communication and Cognition 29 (3/4), 393–4242.
Auroux, Sylvain, and Mayet, Laurent. (2001) Entretien avec Sylvain Auroux, Le
mystère des racines. Sciences et Avenir 125, 12–5.
Baddeley, Alan D. (1986) Working memory. Oxford: Clarendon Press.
Bickerton, Derek. (1990) Language and Species. Chicago: The University of
Chicago Press.
Biraben, Jean-Noël, Masset, Claude, and Thillaud, Pierre L. (1997) Le
peuplement préhistorique de l’Europe. In Histoire des populations de
l’Europe, tome 1 — Des origines aux prémices de la révolution
démographique (pp. 39–92). Fayard.
Black, Cheryl A. (1999) A step-by-step introduction to the Government and
Binding theory of syntax. Summer Institute of Linguistics,
http://www.sil.org/mexico/ling/E002-IntroGB.pdf.
45
46
Language Acquisition, Change and Emergence
Bocquet-Appel, Jean-Pierre, and Demars, Pierre-Yves. (2000) Population kinetics
in the upper Palaeolithic in Western Europe. Journal of Archaeological
Science 27, 551–70.
Cann, Rebecca L., Stoneking, Mark, and Wilson, Allan C. (1987) Mitochondrial
DNA and Human evolution. Nature 325, 31–6.
Cavalli-Sforza, Luigi Luca, Menozzi, Paolo, and Plazza, Alberto. (1994) The
history and geography of human genes. Princeton: Princeton University
Press.
Chomsky, Noam. (1975) Reflections on language. New York: Pantheon Books.
Coupé, Christophe and Hombert, Jean-Marie. (2002) Language at 70,000 BP:
Evidence from sea-crossings. In Proceedings of the Fourth International
Conference on the Evolution of Language (p. 27). Harvard.
Coupé, Christophe. (2003) De l’origine du langage à l’origine des langues:
Modélisations de l’émergence et de l’évolution des systèmes linguistiques.
PhD dissertation in Cognitive Sciences, Univ. Lyon 2.
Croft, William. (1990) Typology and Universals. Cambridge Textbooks in
Linguistics. Cambridge: Cambridge University Press.
De Waal, Franz. B. (1998) Chimpanzee Politics: Power and Sex among Apes.
Baltimore: Johns Hopkins University Press.
D’Errico, Francesco, Henshilwood, Christopher, and Nilssen, Peter. (2001) An
engraved bone fragment from c. 70,000-year-old Middle Stone Age levels
at Blombos Cave, South Africa: implications for the origin of symbolism
and language. Antiquity 75 (288), 309–18.
Dessalles, Jean-Louis. (2000) Aux origines du langage — Une histoire naturelle
de la parole. Paris: Hermes Science Publications.
Dixon, Robert M.W. (1997) The rise and fall of languages. Cambridge:
Cambridge University Press.
Dunbar, Robin I. M. (1993) Coevolution of neocortical size, group size and
language in humans. Behavioral and Brain Sciences 16, 681–94.
Dunbar, Robin I. M. (1996) Grooming, Gossip and the Evolution of Language.
London: Farber and Farber.
Ehala, Martin. (1996) Self-organisation and language change. Diachronica 13
(1), 1–28.
Enard, Wolfgang, Przeworski, Molly, Fisher, E. Simon, Lai, Cecilia S. L., Wiebe,
Victor, Kitano, Takashi, Monaco, Tony and Pääbo, Svante. (2002)
Molecular evolution of FOXP2, a gene involved in speech and language.
Nature 418, 869.
Freedman, David. A. and Wang, William S.-Y. (1996) Language polygenesis: A
probabilistic model. Anthropological Sciences 104 (2), 131–8.
Polygenesis of Linguistic Strategies
Greenberg, Joseph H. (1978) Typology and cross-linguistic generalizations. In
Joseph H. Greenberg (Ed.) Universals of human language. Method and
Theory, volume 1 (pp. 33–61). Stanford: Stanford University Press.
Harpending, Henry C., Batzer, Mark A., Gurven, Michael, Jorde, Lynn B.,
Rogers, Alan R., and Sherry, Stephen T. (1998) Genetic traces of ancient
demography. Proceedings of the National Academy of Sciences of the
United States of America 95, 1961–7.
Harpending, Henry C. and Sherry, Stephen T. and Rogers, Alan R. and
Stoneking, Mark. (1993) The genetic structure of ancient human
populations. Current Anthropology 34 (4), 483–96.
Hassan, Fekri A. (1981) Demographic archaeology. Studies in Archaeology.
Academic Press.
Hawks, John, Hunley, Keith, Lee, Sang-Hee, and Wolpoff, Milford. (2000)
Population bottlenecks and Pleistocene human evolution. Molecular
Biology and Evolution 17 (1), 2–22.
Hockett, Charles F. (1960) The origin of speech. Scientific American 203,
88–96.
Jacquesson, François. (2001) Pour une linguistique des quasi-déserts. In
Anne-Marie Loffler-Laurian (Ed.) Etudes de linguistique générale et
contrastive. Hommage à Jean Perrot (pp. 199–216). Paris: Centre de
Recherche sur les Langues et les Sociétés.
Klein, Richard. G. (1999) The human career, human biological and cultural
origins. Chicago; London: The University of Chicago Press.
Labov, William. (1963) The social motivation of a sound change. Word 19,
273–303.
Lahr, Marta Mirazon and Foley, Robert. (1994) Multiple dispersals and modern
human origins. Evolutionary Anthropology 3, 48–60.
Lai, C. S., Fisher, S. E., Hurst, J. A., Vargha-Khadem, F., and Monaco, A.P.
(2001) A forkhead-domain gene is mutated in a severe speech and
language disorder. Nature 413, 519–23.
MacNeilage, Peter F. (1998) The frame/content theory of evolution of speech
production. Behavioral and Brain Sciences 21 (4), 499–511.
Marsico, Egidio. (1999) What can a database of proto-languages tell us about
the last 10,000 years of sound changes? In Proceedings of the XIVth
International Congress of Phonetic Sciences. San Francisco.
Marsico, Egidio, Coupé, Christophe, and Pellegrino, François. (2000) Evaluating
the influence of language contact on lexical changes. In Proceedings of the
third Conference on the Evolution of Language (pp. 154–5). Paris: Ecole
Normale Supérieure des Télécommunications.
Marwick, Ben. (2002) Raw material transportation as an indicator of hominid
47
48
Language Acquisition, Change and Emergence
symbolic linguistic capacity during the Pleistocene. In Proceedings of the
Fourth International Conference on the Evolution of Language (p. 74).
Harvard.
Maturana, Humberto and Mpodozis, Jorge. (2000) The origin of species by
means of natural drift. Revista Chilena de Historia Natural 73, 261–310.
Murray, James D. (1994) Mathematical biology. Second, Corrected Edition.
Berlin Heidelberg New York: Springer Verlag.
Naccache, Albert F. H. (2002) Sociolinguistic approaches to the prehistory of
the Mashriqian (Semitic) languages. In Proceedings of the Fourth
International Conference on the Evolution of Language (p. 80). Harvard.
Nettle, Daniel. (1999a) Linguistic Diversity. Oxford Linguistic. Oxford: Oxford
University Press.
Nettle, Daniel. (1999b) Using social impact theory to simulate language change.
Lingua 108, 95–117.
Nettle, Daniel. (1999c) Is the rate of linguistic change constant? Lingua 108,
119–136.
Nettle, Daniel and Romaine, Suzanne. (2000) Vanishing voices. Oxford: Oxford
University Press.
Nichols, Johanna. (1992) Linguistic diversity in space and time. Chicago;
London: University of Chicago Press.
Pagel, Mark. (2000) The history, rate and pattern of world linguistic evolution.
In Chris Knight, Michael Studdert-Kennedy, and Jim Hurford (Eds.) The
evolutionary emergence of language (pp. 391–416). Cambridge:
Cambridge University Press.
Ringe, Donald. A. J. (1992) On calculating the factor of chance in language
comparison. Transactions of the American Philosophical Society 82 (1),
pp. 1–109.
Rizzolatti, Giacomo, Fadiga, Luciano, Gallese, Vittorio, and Fogassi, Leonardo.
(1996) Premotor cortex and the recognition of motor action. Cognitive
Brain Research 3, 131–41.
Ruhlen, Merritt. (1994) The origin of language. Tracing the evolution of the
mother tongue. New York: John Wiley and Sons.
Sauvet, Georges and Wlodarczyk, André. (1995) Eléments d’une grammaire
formelle de l’art pariétal paléolithique. L’Anthropologie 99 (2–3),
193–211.
Savage-Rumbaugh, Sue, Shanker, Stuart G., and Taylor, Talbot J. (1998) Apes,
language, and the human mind. Oxford: Oxford University Press.
Schoenemann, P. Thomas. (1999) Syntax as an emergent characteristic of the
evolution of semantic complexity. Minds and Machines 9, 309–46.
Polygenesis of Linguistic Strategies
Sherry, Stephen T., Rogers, Alan R., Harpending, Henry, Soodyall, Himla,
Jenkins, Trefor, and Stoneking, Mark. (1994) Mismatch distributions of
mtDNA reveal recent human population expansions. Human Biology 66
(5), 761–75.
Sherry, Stephen T., Harpending, Henry C., Batzer, Mark A., and Stoneking,
Mark. (1997) Alu evolution in human populations: Using the coalescent
to estimate effective population size. Genetics 147, 1977–80.
Shibatani, Masayoshi and Pardeshi, Prashant. (2001) The causative continuum.
The grammar of causation and interpersonal manipulation. In Masayoshi
Shibatani (Ed.) Typological Studies in Language 48 (pp. 85–126).
Amsterdam; Philadelphia: John Benjamins Publishing Company.
Sperber, Dan. (1995) How do we communicate? In John Brockman and Katinka
Matson (Eds.) How things are: A science toolkit for the mind (pp. 191–9).
New York: Morrow.
Stringer, Chris B. and Andrews, Peter. (1988) Genetic and fossil evidence for the
origin of modern humans. Science 239, 1263–7.
Terrace, Herbert S. (2000) Serial expertise and the evolution of language. In
Proceedings of the Third International Conference on the Evolution of
Language (pp. 154–155). Paris: Ecole Nationale Supérieure des Télécommunications, Paris.
Thorne, Alan G. and Wolpoff, Milford H. (1992) The multiregional evolution of
humans. Scientific American 266, 76–83.
Wang, William S.-Y. (1973) Chinese language. Scientific American 228, 50–60.
Wang, William S.-Y. (1991) Explorations in language evolution. In William S.-Y.
Wang (Ed.) Explorations in Language (pp. 105–130). Taiwan: Pyramid
Press.
Zhivotovsky, Lev A., Bennett, Lynda, Bowcock, Anne M., and Feldman, Marcus
W. (2000) Human population expansion and microsatellite variation.
Molecular Biology and Evolution 17 (5), 757–67.
Zubrow, Ezra B. (1989) The Demographic Modelling of Neanderthal Extinction.
In Paul Mellars and Chris Stringer (Eds.) The Human Revolution:
behavioural and biological perspectives on the origins of modern humans
(pp. 212–31). Princeton, New Jersey: Princeton University Press.
49
Christophe Coupé