LINGUIST List 12.1073

Tue Apr 17 2001

Review: Knight, Studdert-Kennedy, & Hurford

Editor for this issue: Terence Langendoen <terrylinguistlist.org>


What follows is another discussion note contributed to our Book Discussion Forum. We expect these discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in. If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for discussion." (This means that the publisher has sent us a review copy.) Then contact Simin Karimi at siminlinguistlist.org or Terry Langendoen at terrylinguistlist.org.

Directory

  1. Andrei Popescu-Belis, Review of The Evolutionary Emergence of Language

Message 1: Review of The Evolutionary Emergence of Language

Date: Tue, 17 Apr 2001 18:30:58 +0200
From: Andrei Popescu-Belis <Andrei.Popescu-Belisissco.unige.ch>
Subject: Review of The Evolutionary Emergence of Language

Knight, Chris, Michael Studdert-Kennedy, and James R. Hurford, ed.
(2000) The Evolutionary Emergence of Language: Social Function
and the Origins of Linguistic Form, Cambridge University Press,
xi+426 pages.

Andrei Popescu-Belis, University of Geneva.


The origins of language have been traditionally a matter of
ceaseless speculation and scientific suspicion. However, the
subject has attracted increasing attention in the past decade,
with contributions from a wide community of researchers from
linguistics, cognitive science and artificial intelligence. The
series of "International Conferences on the Evolution of
Language" (Edinburgh 1996, London 1998, Paris 2000) reveals a
renewed interest in the problem.

The present volume gathers nineteen selected contributions from
the second conference, which represent, as in the preceding
volume (Hurford et al. 1998), extended and revised versions of
the conference papers. The quality of the conference papers
prompted the publication of a second volume, on slightly
different topics (Briscoe, in press). A volume derived from the
latest conference is also underway.


1. Synopsis

Three main themes were identified as crucial, hence the subtitle,
"Social Function and the Origins of Linguistic Form", with two
main aspects of form being discussed, phonetic structure and
syntactic structure. Each of the three parts begins with an
introductory chapter written by an editor, which provides a
summary of the chapters and an analysis of relations among them.
It should be noted from the start that the chapters express
independent contributions from their authors, and often attempt
to summarize extensive research. Here, I will summarize each of
the 23 chapters, and provide comments on some of them right after
their summary. General comments come in the end.

1.1 Opening chapter

The opening chapter sets the stage for the whole book. Signed by
the three editors, it summarizes recent developments through the
work of several major actors of the debate. Chomsky is first
acknowledged for "reopen[ing] language to psychological and
evolutionary study, largely dormant since 'The Descent of Man'
(Darwin 1871)" (p.2). The progress of the contemporary debate on
the origins of language is viewed as an incremental series of
attempts to anchor the debate in the neo-Darwinian framework,
through the works of Bickerton, Pinker and Bloom, Dunbar, and
Deacon among others. These works appear as complementary, since
they tackle different questions that should be answered in an
evolutionary account of language: intermediate phases and
transitions between them (Bickerton), adaptive character of the
transitions (Pinker and Bloom), link with primatology /
paleontology (Dunbar), symbolic and representational function of
language (Deacon, Bickerton). The account of the debate finishes
on, and opens the stage to, the discussion of the social role of
language, drawing inspiration from anthropology but also from
computer simulations of societies. Thus, the chapter that follow
"address the need for a sharply focused attack on the evolution
of language from a post-Chomskyan perspective" (p.10).

1.2 First part

In his introduction to part I, "The Evolution of Cooperative
Communication", Chris Knight provides an introductory and
sequential analysis of the chapters, in about one page per
chapter. The main thread is the analysis of the social role of
language as an epitome of its adaptive character. "Social" refers
here to the small groups of hominids in which language is
supposed to have appeared, including the most granular level, the
speaker/hearer pair. The cohesion of social groups and therefore
the survival of their members, which pass on both genetic and
linguistic material, is threatened by competition for resources.
Selective pressure favors language, as well as other ritual
relations, as a means to moderate competition and consolidate
interdependence links between individuals.

For Robbins Burling, the speaker/hearer relation is at the core
of the origins of meaning, since "communication does not begin
when someone makes a sign, but when someone interprets another's
behaviour as a sign". In Chapter 2, Comprehension, Production and
Conventionalization in the Origins of Language, the emergence of
the first words is viewed as the ritualization of externally
motivated gestures.

Jason Noble uses game-theoretic modeling of a population of
speakers playing communication games to test an hypothesis
proposed by Krebs and Dawkins in 1984 (Chapter 3, Cooperation,
Competition and the Evolution of Prelinguistic Communication).
This hypothesis states that ostensible and costly communication
devices are more likely to occur in competitive situations, when
they are often used to deceive one's competitors, while low-
energy, subtle signaling systems occur in collaborative
situations. The simulations of the global payoffs in both cases
show that communicative acts are rather unlikely to appear at all
when speakers and hearers are mutual competitors.

In the account of the adaptive role of language proposed by
Jean-Louis Dessalles (Chapter 4, Language and Hominid Politics),
the capacity to exchange relevant information is central. The
author simulates the evolution of a population in which two
attitudes towards communication, provide vs. return information,
are genetically transmitted. The results show that stability is
reached when the information exchange is reciprocally relevant.
Cheaters that do not respond with relevant information gradually
die out. In anthropomorphic terms, each group increases its
fitness if it is able to award social status in return for
relevant communication.

Camilla Power makes use of several descriptions of naturally
occurring ritual codes in order to outline the analogies between
their adaptive role, which caused their emergence, and the role
of language in general (Chapter 5, Secret Language Use at Female
Initiation: Bounding Gossiping Communities). According to the
author, social stability is achieved by coercion of new
generations into pre-existing groups (e.g., the group of nubile
women), which in turn is achieved through the use of a common
ritual. Using language to inform is therefore less important than
using the right set of statements at the right moment.

Closing the first part, editor Chris Knight (in Chapter 6, Play
as Precursor of Phonology and Syntax) roots the discrete features
of language into the ability or the interest to perform several
times, in contexts further and further away from the original
one, a ritual or a series of conventional gestures that are
socially relevant. Despite the title, the absence of a formal
description of the proposal weakens the suggested link with the
next two parts.

1.3 Second part

The five chapters of the second part, The Emergence of Phonetic
Structure, share a common approach to the evolution of phonetic
capacities: they examine the constraints set by the phonetic
acquisition capacities of each new generation of speakers. As
Michael Studdert-Kennedy points out in his introduction, this
approach sheds new light on the link between ontogeny and
phylogeny. The former is not viewed as a repetition of the
latter, but as a filter and a constraint on evolution. The
remainder of the introduction summarizes sequentially the
following chapters, which all presuppose the existence of a
symbolic communicative capacity, or at least a propensity for it.

Marilyn M. Vihman and Rory A. DePaolis attempt to compare the
capacities that Donald (1998) considered necessary for language
emergence with those that they identify for language acquisition
(Chapter 8, The Role of Mimesis in Infant Language Development:
Evidence for Phylogeny). The authors hypothesize the existence of
an "articulatory filter" in the child, which matches her
spontaneous phonetic patterns ("vocal motor schemes") with
salient patterns in the spoken input received from adults. This
filter is part of a positive feedback loop involving input/output
sound patterns and understood/expressed meanings. Three
prerequisites for a referential use of articulate words are
identified: (1) capacity for communication using sounds; (2)
representational capacity, to abstract meanings; (3) stable
phonetic/phonologic capacity, which makes use of the articulatory
filter. The ontogeny of the three capacities proceeds at inter-
correlated speeds, despite measurable variation between children.
In support to Donald's theory, the authors suggest that similar
capacities conditioned language emergence, but it was only the
evolution of an articulatory filter that enabled symbolic word
use.

The hypothesized parallelism between language acquisition and
emergence of language is further developed in Chapter 9,
Evolution of Speech: the Relation Between Ontogeny and Phylogeny,
by Peter J. MacNeilage and Barbara L. Davis. It is suggested that
a common description framework, the frame/content theory, can be
applied to both phenomena. They can be conceived as an evolution
of the spontaneous motor capacity to produce cyclical patterns,
into more complex phonetic structures that differentiate the
frames, thanks to self-organization.

In Chapter 10, Michael Studdert-Kennedy develops several
theoretical points that guide a mimetic theory of phonetic
structure (Evolutionary Implications of the Particulate
Principle: Imitation and the Dissociation of Phonetic Form from
Semantic Function). Language, along with other domains such as
physics, chemistry and genetics, is subject to a generative
principle. Entities in these domains are discrete combinations of
elementary constituents, with properties that are not reducible
to those of the constituents. The independence of sound from
meaning isolates the phonetic system, leaving it prone to
generative evolution through a mimetic process. No formal
description of such a process is given, however a similar basis
is hypothesized for writing systems.

Bart de Boer adopts a simplified but computationally tractable
model of the emergence of vowel systems (Chapter 11, Emergence of
Sound Systems Through Self-Organization). His computer agents are
able to generate vowels through the use of articulatory vectors
and F1-F4 formants, to compare perceived and memorized vowels,
and to adjust the memorized vowels according to the perceived
ones. For both humans and agents, "it is easy to calculate the
formant frequencies from the articulatory description, but very
hard to calculate the articulatory description from the acoustic
description" (p.180). The agents engage in imitation games - in
the style of those designed by Luc Steels (1998), in whose
laboratory this research was carried. The agents succeed in
establishing vowel systems that present striking similarities
with naturally occurring ones, and that are also stable with
respect to noise and agent replacement.

In the last chapter of the second part, Daniel Livingstone and
Colin Fyfe describe a computational model which makes use of
neural-network based communicating agents. The reference to
phonetics is quite remote, since agents must only express signals
that are distinct enough to convey elementary meanings. As the
title indicates (Chapter 12, Modeling Language-Physiology
Coevolution), the four described experiments deal with the
advantage to evolve a more complex neural structure, which
facilitates the expression of more meanings, however at a certain
cost.

1.4 Third part and epilogue

The third part and the epilogue amount to almost one half of the
book. Some of the chapters here make use of precise linguistic
facts to hypothesize scenarios for the emergence of syntactic
structures. Some others describe computational models (as were 11
and 12 above). On the whole, the following contributions seemed
to me quite precisely focused, therefore offering a better ground
for discussion.

The introduction by James R. Hurford (Chapter 13, The Emergence
of Syntax) sets the stage in a less sequential manner than the
previous introductions, outlining relations of complementarity or
contrast between proposals. The main ideas of this part are
synthesized: (1) the language faculty is not entirely adaptive;
(2) some of the features of the syntactic capacity are built on
previous non-syntactic, possibly semantic, capacities; (3)
segmentation of and generalization over samples of proto-
linguistic utterances are a significant factor in the emergence
of syntactic structures; (4) computational modeling brings
significant validation to such hypotheses. No contribution
advocates the adaptive role of brain structures as an exclusive
factor in the emergence of syntax, some being even on the
opposite side. Generativist approaches (Lightfoot, Newmeyer) are
distinguished from "broad-brush" theories (Carstairs-McCarthy,
Wray, Bickerton) and from computer modeling (Kirby, Hurford).
 The introduction acknowledges the close relationship between
some of the chapters and the other book originating from the same
conference (Briscoe, in press), but it does not explain the
specificity of each book. For instance, a review article on
computational models is quoted (Hurford, in press), as well as a
new proposal by Batali (Batali, in press) which has strong links
with some of the following chapters (Wray, Kirby, Hurford). But
why were they published separately?

Chapter 14, The Spandrels of the Linguistic Genotype, explicitly
adopts the framework of Universal Grammar (UG), i.e. the
"structured initial state, common to the species" (p.231), which
must be genetically encoded, and from which the individual
grammatical competence develops. The author, David Lightfoot,
proceeds to show that UG cannot be entirely adaptive. First, he
shows strong evidence for a UG principle stating that the
movement of a wh- interrogative pronoun is allowed only if its
trace is the head of a complementizer of a word adjacent to this
trace; the same constraint holds for the deletion of "that".
Second, it is noted that many regularities in the natural world
did not emerge under selective pressure, but followed from
constraints imposed by the laws of physics or chemistry on other
features subject to selection (e.g., scaling factors of body
size). Third, the UG principle described above cannot itself be
adaptive, since it rules out many useful sentences and obliges
speakers to use ad-hoc alternatives. Therefore, against (Pinker
and Bloom 1990), there exist non adaptive principles in the UG,
or "spandrels" (see (Gould and Lewontin 1979) for this metaphor).
 There are at least two difficulties with this kind of
inference. Those who do not accept UG may they explain the above
principle on different grounds, thus completely displacing the
debate. In addition, one inevitably wonders from which other
principle, subject to selection, the above principle can be
derived. If the only candidate is a brain structure that "may
have evolved in a series of explosions and qualitative changes,
and not gradually" (p.234), then this is quite remote from the
identification of evolutionary stages (cf. Bickerton, Wray). As
Bickerton writes about the spandrel metaphor (p.266), "no one has
suggested what might be the equivalent of arches in the case of
syntax".

Also against the view of a strict adaptive role of grammar,
Andrew Carstairs-McCarthy defends a scenario for the emergence
of syntax (without recursion) through the use of a pre-existing,
but not optimal, device: the syllable structure. The progression
of Chapter 15, The Distinction Between Sentences and Noun
Phrases: An Impediment to Language Evolution? is as follows. (1)
The distinction between sentences and noun phrases (or
truthfulness and reference) must have evolutionary origins, since
it is possible to imagine "a kind of syntax" without this
distinction, and philosophers fail to explain it intrinsically.
(2) More precisely, its origins must lie in the syllable
structures, which emerged thanks to articulatory skills (cf.
MacNeilage and Davis) and were subsequently used to organize
messages, with consonants or syllable margins replaced by noun
phrases and syllables replaced by sentences. (3) This way of
expressing complex meanings is not optimal, since no embedding
(recursion) is possible. (4) This limitation explains the long
cultural stagnation of Homo erectus, in agreement with
Bickerton's protolanguage scenario.
 It is to me unclear how the syllable structure could have
been transformed into a proto-syntactic structure. Moreover, if
it grouped several proto-words made of syllables, there must have
been some form of recursion. The mapping between nouns and
consonants, sentences and syllables is problematic, despite
superficial analogies (p.253), since not much is known about
their representation in the mind. Finally, it seems difficult to
explain the opposition between noun phrases and sentences,
between reference and truthfulness, by the margin/syllable one.
This opposition is implicit in most theories of reference, but
has not been analyzed directly by philosophers since it stems
from the irreducible opposition between arguments and predicate-
argument structure.

In Chapter 16, How Protolanguage Became Language, Derek Bickerton
outlines a scenario for the transition from the now famous
intermediate stage to full syntactic language. The author clearly
states three main prerequisites. There is first the capacity to
monitor social alliances between individuals, or "social
calculus", of which syntax is probably an exaptation (use of a
feature previously evolved for a different function). This
capacity explains the familiarity of early human minds with the
Agent- Theme-Goal structure; however, it coexisted for a long
time (two or three million years) with only a rudimentary
protolanguage using small series of words. Second, the capacity
to maintain coherent signals in the brain, thanks to the
recruitment of large neuron sets, made possible the merging of
words into structures exapted from social calculus. Third,
selective pressure and Baldwin effects acting upon the need of a
speaker/hearer compromise led to modern syntactic marking (within
the last 200,000 years). The end of the chapter answers
objections and summarizes the scenario.
 Bickerton's proposal is once again appealing, despite its
brevity (but see Calvin and Bickerton (2000) for details). It is
also open enough to accommodate further findings, especially from
neuroscience, since the concept of signal coherence required to
group arguments is quite unspecified. The reciprocal altruism
(better termed as collective interest) is closely related to
contributions in part one. Bickerton's new scenario brings
tangible ideas about post-protolanguage evolution, which were
highly needed since the protolanguage hypothesis, one of the main
merits of Bickerton's "Language and Species" (1990). The present
scenario convincingly elaborates on the next transition, the
mapping of thematic roles onto syntax. It is hoped that these
elements will constitute an object of further debate and
modeling.

Alison Wray makes one of the most remarkable contributions to the
book in Chapter 17, Holistic Utterances in Protolanguage: the
Link from Primates to Humans. The author attempts to bridge the
gap between the widely accepted conception of language as highly
compositional or "analytic", and a "holistic" protolanguage
(plus primate communication) in which utterances are attached
non-compositionally to meanings. First, Wray argues for the
importance of holistic communication in everyday language,
through "formulaic sequences". Then, she presents a case for the
continuity of the holistic mode from primate communication to
protolanguage to language. Several functions of holistic
utterances, especially social, seem indeed common to all three
stages. If a persistent holistic mode continually served basic
communicative functions, then analytic language may have been
selected for other functions, such as thought and representation.
This must have been preceded by the emergence of phonetic
structure for holistic utterances, not for words. Finally, the
mapping of theta- roles onto phonetic representations led to a
major revolution, the segmentation of holistic utterances into
modern words.
 Wray's proposal constitutes a partial alternative to
Bickerton's, with an encouraging number of common assumptions. As
pointed out by Hurford (p.226), the transition advocated by Wray
is analytic and based on pragmatic utterances, whereas
Bickerton's is synthetic and based on truth-conditional
utterances. These differences have serious implications for
computer modeling, which seems a promising option for both
proposals. Wray's model has obvious connections with Kirby's
simulation (see next chapter), but also with one by Batali (in
press). Another relevant simulation (Nowak et al. 2000) showed
that analytic communication offers selective advantage over
holistic one as soon as a higher number of meanings composed of
independent elements must be expressed.

In the two following chapters, Simon Kirby and James R. Hurford
use computer models of dialogs between communicating agents to
shed light on scenarios of language emergence and evolution.
More precisely, they attempt to model the emergence of
syntactically structured messages describing situations in a
simulated environment. In Chapter 18, Syntax without Natural
Selection: How Compositionality Emerges from Vocabulary in a
Population of Learners, Kirby makes use of a population of
neural-network based agents (Batali 1998). The goal of the agents
is to emit/receive utterances about Agent-Predicate-Patient
situations. Receivers attempt to decompose utterances by
inventing or adapting their vocabulary and their simple
decomposition rules. The result is that the initial holistic
communication conventions (cf. Wray) give way to analytic
conventions (common vocabulary and decomposition rules) despite
the lack of selective pressure to do so. Compositional languages
appear thus as more stable than holistic ones, since they evolve
spontaneously given the implemented individual and social
structures, and the range of occurring meanings. In particular,
frequent meanings continue to be expressed holistically.

In Chapter 19, Social Transmission Favors Linguistic
Generalization, James R. Hurford explores a different hypothesis
for the evolution of compositional languages. In this model, it
is the sender that chooses to encode the situation holistically
or to create or use a composition rule. Each agent infers its
linguistic knowledge (words and rules) from input that it was
exposed to in the acquisition period. After a certain number of
generations (made of five "adults" and one "infant") the results
are: (1) compositional rules supersede holistic words, except (2)
for frequently occurring meanings; (3) general rules appear even
when the agents have a small propensity to generalize; (4) more
general rules (predicates with one, two or three argument slots),
are even more stable. In the end, the model is compared to those
by Kirby (Chapter 18) and Batali (in press).
 Kirby and Hurford describe their models clearly, including the
explicit assumptions and the targeted phenomena. Their conclusion
is that in certain agent-based dynamic systems, given adequate
protocols for dialogs and learning, complex communication codes
can emerge without explicit selective pressure. To my view, the
epistemic value of such models depends on the resemblance between
simulated properties and their counterpart in reality; at
present, the emergent communication codes are still very remote
from natural language. In particular, the semantic
representations closely mimic the desired "grammatical" patterns.
In Kirby's model, the emergent grammar is not recursive.
Moreover, it is hypothesized that receivers always have direct
access to the meaning expressed by senders, as in Oliphant's
(1997) "obverter" learning procedure, acknowledged by the
authors. There are however models based on physical robots that
drop this constraint (Steels and Kaplan, in press).

In Chapter 20, Words, Memes and Language Evolution, Robert P.
Worden relates the emergence of words to more general
considerations on the replication of forms or "memes" (following
Dawkins' terminology). The author adopts the hypothesis of a
lexicalized tree grammar, in which the "words" also contain
feature structures. Their transmission through language learning
is subject to unification and generalization. Unlike the previous
chapters, several factors of selection pressure are identified.
Unfortunately, no idea is given for an implementation that could
somehow validate the proposal, and there is no obvious link to
language emergence.
 Excessive use of analogies with other domains tend to confuse
the reader and fail to bring convincing evidence. Several vague
or unsupported statements weaken the otherwise promising ideas:
"language evolved from primate social intelligence" (p.368); "the
Head Parameter, which has been taken as evidence for innate
language- specific structure in the brain [...] is not evidence
for any innate structure in the brain" (p.366); "the need to
encode some of [the] aspects of the motion arises from the
selection pressure to express useful meanings" (p.364).

In Chapter 21, concluding this part, generative linguist
Frederick J. Newmeyer develops a clearly formulated argument (On
the Reconstruction of 'Proto-world' Word Order). He first defends
the hypothesis of rigid subject-object-verb (SOV) order in the
earliest human languages, then draws from this a statement on the
origins of Universal Grammar (UG). The first point is inferred
from three independent assumptions: (1) OV to VO transitions are
more common than VO to OV, yet SOV is still the most widespread
order; (2) early language probably marked thematic roles (cf.
Bickerton), which are best matched to grammatical roles in SOV
languages; (3) the best way to mark arguments in primitive
argument structures is to delay the appearance of the verb.
 Next, Newmeyer examines three possible explanations of the
origins of UG constraints: (1) big-bang, (2) genetic
assimilation, and (3) "constraints-as- epiphenomena" (p.380; cf.
Bickerton's five scenarios, p.266). The last one (the
"spandrels") is not discussed since it places the explanative
burden on other cognitive processes. The thesis of primitive SOV
order rules out the second hypothesis, since UG principles are
not manifest in SOV languages. Thus, "UG constraints must have
appeared contemporaneously with the appearance of true human
language" (p.384).
 Newmeyer's first thesis is a controversial but explicit
result, very welcome in the debate on early language history. It
is however unclear how directly it supports the second thesis.
The list of language emergence scenarios differs between
Bickerton and Newmeyer, and since little is know about their
details, they may be non-exclusive or they may be blended into
other scenarios. The case for the "big-bang" seems thus untimely:
to the external evidence brought by this chapter, one may oppose
the lack of internal evidence, i.e. our ignorance of the
linguistic big-bang mechanisms.

The epilogue presents a series of simulations on language
evolution by Mark Pagel (Chapter 22, The History, Rate and
Pattern of World Linguistic Evolution). Unlike previous chapters,
this one deals with the birth, death and divergence of individual
languages. The computational models presented here are no longer
based on communicating agents, but on equations involving various
evolution rates. The models are applied to two language families,
and with some parameter tuning, they provide fair approximations
of data from the real world. Several ecological parameters that
influence language evolution are described. The main contribution
of such models to language emergence studies is that they bring
information about the initial states in language evolution (e.g.,
the duration necessary to reach the present state) as well as
geographical clues.


2. Discussion

2.1 General evaluation

>From the overviews above, it has hopefully become clear that the
present book gathers a very wide range of contributions to the
debate on the emergence of language. Therefore, the book is
highly advisable to all those who are in search for a
contemporary snapshot of the research on this problem, since it
brings to them in- depth analyses and a wealth of references. One
of the main merits of the book is thus to provide in a single
volume a gateway to many theories among which the reader can make
her choice and study some in greater detail.

This is not to say that the book develops a completely unbiased
range of theories. The editors well acknowledge that emphasis is
laid on the social grounding of language function and of
linguistic conventions, as opposed for instance to the biological
bases of language. Also, especially in the second half of the
book, a non-adaptive and non- genetic stance prevails. As long as
these positions are explicitly stated, the reader remains free to
refer to quoted literature with other points of view. For
instance, references are made to a complementary collection of
articles (Briscoe, in press).

2.2 The structure of the book

Given the diversity of the contributions, it was probably uneasy
to divide them in three parts. The three themes - Evolution of
Cooperative Communication, Emergence of Phonetic Structure,
Emergence of Syntax - are often intermingled in the chapters,
especially in those attempting to propose global scenarios. For
instance, Bickerton's section on 'social calculus' is strongly
attached to the first theme; Carstairs-McCarthy's 'syntax from
syllable' proposal is at the interface of the second and third;
the chapter by Livingstone and Fyfe is not really about the
second theme, although it ends the second part; etc. The themes
should be taken as global indications, sometimes reflecting the
author's background, but they should not mislead the reader into
avoiding some of the parts.

The chapters of these edited proceedings are largely independent,
which requires some effort from those reading the book in detail.
Each chapter is densely written and builds upon a non-negligible
amount of previous knowledge. Contributors quite often refer to
one or more of their books, in which their arguments are more
extensively developed. The four introductions (the general one
plus one per part) do not really render the book more accessible,
especially when they only summarize the chapters. Brief
explanations as the Appendix on Syntactic Notation (pp. 228-229)
are a good idea, and would have been welcome elsewhere too.

The absence of a part concerned with semantics is intriguing,
especially if the three themes are translated as: pragmatics,
phonetics, syntax. Does this mean that the emergence of semantics
(if this a well-formed question) is no longer problematic? Or,
conversely, that it requires more finesse than the present
models? The pairing of words and meanings was one of the first
successful applications of computer simulations, and is now well
analyzed; the link between meanings and perception in physical
robots is also being studied (Steels and Kaplan, in press). There
are however much more complex semantic properties of language,
e.g., compositionality, or reference. One may attempt to share
the burden with evolutionary semantics, or evolutionary
cognition, or even take the emergence of such capacities for
granted, but such an assumption should be clearly stated.

2.3 Research on language emergence: history vs. methodology

The size and broadness of the book suggest that it attempts to
position itself as a reference collection in the domain. Such a
role, however, would require a more thorough analysis of the
methods and of the assumptions of the field, in other words an
epistemological debate that frames and justifies the
contributions. I will first turn to the general introduction,
which retraces the recent history of the domain, before
commenting on the methodology.

2.3.1 The historical introduction

It is not an easy task to identify debates and schools of thought
in the recent history of a domain that has witnessed so many
developments. The introduction succeeds in giving a synthetic yet
vivid picture of the main contributions and debates, gathering
into a coherent framework proposals from quite different
horizons. It is interesting to note that the most quoted authors
in the book (the references of each chapter) match quite closely
the main names of the introduction and their order: Chomsky,
Bickerton, Pinker (all quoted by about half of the chapters),
Terence Deacon, Robin Dunbar and James Hurford (about 40%).

There are, however, details in the introduction that sometimes
cast doubt on its accuracy. I was surprised to notice two
inexactitudes in the section about Bickerton (all the more that
he contributed to the book). "Roots of Language" is referred to
as Bickerton's "first book" (p.4), while to my knowledge the
first one was "Dynamics of a Creole System" in 1975 (anecdotally,
it seems that it was handed to his thesis committee in lieu of a
dissertation). Second, the hypothesis of a language bio- program,
with all its extreme implications, has been better developed than
in "Roots of Language" in an important target article, not quoted
here, published in Behavioral and Brain Sciences in 1984,
accompanied by peer reactions and by the author's response
(Bickerton 1984). So, before Pinker and Bloom's 1990 article,
this "respected and widely read interdisciplinary journal" (p.5)
had already published crucial articles on language emergence.

And last, while it is difficult to deny the importance of the
contributions quoted in this introduction, I believe that
computational simulations should have been better developed,
although there is no single researcher that could represent this
approach in the debate staged by the introduction. To my view,
the possibility of computer modeling has opened a new dimension
in language emergence research, as some of the chapters here
indeed confirm.

2.3.2 The need for methodology studies

The study of language emergence is once again flourishing - just
as it was two centuries ago in Europe, as we know, when the
French Academy, overwhelmed by memoirs on this topic that could
not be evaluated, decided not to receive them any longer. Are we
today close to a similar reaction from the linguistic community?
Are there ways to prevent such a reaction? The first answer is
fortunately: no, for the time being. Despite the growing number
of books, there are enough consistent contributions that maintain
a high interest for the domain (e.g., most of the present
chapters). As a suggestion for the second, I will stress the
importance of methodological studies, which are strikingly absent
from the present book.

What is, indeed, the goal of the book? What is exactly the
meaning of its title, "The Evolutionary Emergence of Language"?
Quite clearly, the goal is to explain how language emerged (from
a state with no language), using evolutionary concepts, in
agreement with contemporary neo-Darwinian accounts of the
evolution of species. What is, then, such an explanation? How can
such an explanation be evaluated, given that the explained
process left to us only its result? This is the kind of
methodological questions that would ensure the debate remains
truly scientific. It is of course possible to produce a
posteriori analyses (e.g., for the concept of linguistic
explanation, (Newmeyer 1998)) but a priori studies defining the
nature of acceptable theories and arguments would prevent the
risk described above.

Three broad research goals emerge from the book: (1) identify
evidence to be used for explanations (e.g. fossil record) or that
must be explained (e.g. binding phenomena); (2) describe and
justify scenarios of language emergence; (3) run computer models
and use the results to evaluate the scenarios. The first
direction is beyond the scope of the book, while the second is
twice as represented as the third. Regarding the second
direction, the book indirectly poses the problem of partial
scenarios: is it possible to explain the emergence of only one
component? Alternatively, is it possible to explain only one
stage of the emergence process? The titles of the parts suggest
that they present alternative scenarios for a given component,
but their contents show that it is uneasy to isolate components.
I would suggest that agreeing on stages in language evolution,
then attempting to explain each of them, is a more tractable
approach. This is for instance done by Bickerton, who develops
here a scenario of the protolanguage to language transition that
refines a segment of his 1990 scenario.

The task of evaluating a computer model is still more difficult,
since the mirage of novel, unexpected simulation results often
obscures their genuine scientific content. What is for instance a
good computer model in airplane engineering or in epidemiology?
It is a model whose results closely match reality and make
accurate predictions. In language emergence, the "reality" of the
emergence process is still obscure, and its result, language, is
still under study. However, the results of computer simulations
should be more strictly compared with real language, this being
the main argument for validity. Some changes would then occur:
after an initial enthusiasm for simulation results, it would
appear that many of them have little to do with language, and
that others are simple consequences of mathematical models (as
shown by (Nowak et al. 2000)). This in turn would encourage
stricter evaluation and a move towards more complex and
integrated simulations.


3. Conclusion

The book edited by Knight, Studdert-Kennedy and Hurford gathers
contributions that proceed from two opposite directions. Most of
the studies propose scenarios of language emergence, on the basis
of available facts. These are complex scenarios, whose validity
stems from internal coherence and from the number of facts that
are integrated and/or explained. Other studies, aiming to explain
similar facts, design scenarios for computer simulations. These
are much simpler scenarios, whose validity stems from the
comparison of the results with the facts. Both approaches
proceed, presumably, towards the same goal: find more and more
accurate scenarios, which fit the facts, which are clear enough
to be implemented and yield realistic results. The present
contributions attempt to cover partial segments of this grand
pathway. Some of them seem not to far to be joined, some others
still compete. We hope that theorization and modeling will be
brought even closer in the proceedings of the further editions of
the "International Conferences on the Evolution of Language".


4. References

Batali John (1998) : "Computational Simulations of the Emergence
of Grammar", in J.R. Hurford, M. Studdert-Kennedy & C. Knight
(eds.), Approaches to the Evolution of Language, Cambridge, UK,
Cambridge University Press, pp. 405-426.

Batali John (in press) : "The Negotiation and Acquisition of
Recursive Grammars as a Result of Competition Among Exemplars",
in E. J. Briscoe (ed.), Linguistic Evolution Through Language
Acquisition: Formal and Computational Models, Cambridge, UK,
Cambridge University Press.

Bickerton Derek (1975) : Dynamics of a Creole System, Cambridge,
UK, Cambridge University Press.

Bickerton Derek (1984) : "The Language Bioprogram Hypothesis",
The Behavioral and Brain Sciences, vol. 7, n. 2, pp. 173-221
(Target Article: 173-188; Open Peer Commentary: 188-212; Author's
Response: 212-218; References: 218-221).

Briscoe, Edward J. (ed.) (in press) : Linguistic Evolution
Through Language Acquisition: Formal and Computational Models,
Cambridge, UK, Cambridge University Press.

Calvin William H. & Bickerton Derek (2000) : Lingua ex Machina:
Darwin and Chomsky Reconciled, Cambridge, MA, The MIT Press.

Donald Merlin (1998) : "Mimesis and the Executive Suite: Missing
Links in Language Evolution", in J.R. Hurford, M. Studdert-
Kennedy & C. Knight (eds.), Approaches to the Evolution of
Language, Cambridge, UK, Cambridge University Press, pp. 44-67.

Gould Stephen Jay & Lewontin R. (1979) : "The Spandrels of San
Marco and the Panglossian Paradigm: a Critique of the
Adaptationist Programme", Proceedings of the Royal Society of
London B, vol. 205, pp. 581-598.

Hurford James R. (in press) : "Expression/induction Models of
Language Evolution: Dimensions and Issues", in E.J. Briscoe
(ed.), Linguistic Evolution Through Language Acquisition: Formal
and Computational Models, Cambridge, UK, Cambridge University
Press.

Hurford James R., Studdert-Kennedy Michael & Knight Chris (eds.)
(1998) : Approaches to the Evolution of Language - Social and
Cognitive Bases, Cambridge, UK, Cambridge University Press.

Newmeyer Frederick J. (1998) : Language Form and Language
Function, Cambridge, MA, The MIT Press.

Nowak Martin A., Plotkin Joshua B. & Jansen Vincent A. A. (2000):
"The evolution of syntactic communication", Nature, vol. 404
(30/03/2000), pp. 495-498.

Oliphant Michael (1997) : Formal Approaches to Innate and Learned
Communication: Laying the Foundation for Language, PhD Thesis,
University of California San Diego, Cognitive Science Department.

Pinker Steven & Bloom Paul (1990) : "Natural Language and Natural
Selection", Behavioral and Brain Sciences, vol. 13, n. 4, pp.
707-784.

Steels Luc (1998) : "Synthesizing the Origins of Language and
Meaning Using Coevolution, Self-organization and Level
Formation", in J.R. Hurford, M. Studdert- Kennedy et C. Knight
(eds.), Approaches to the Evolution of Language, Cambridge, UK,
Cambridge University Press, pp. 384-404.

Steels Luc & Kaplan Frederic (in press) : "Bootstrapping Grounded
Word Semantics", in E.J. Briscoe (ed.), Linguistic Evolution
Through Language Acquisition: Formal and Computational Models,
Cambridge, UK, Cambridge University Press.

- ------------------------------
Andrei Popescu-Belis is a researcher at ISSCO / TIM, University
of Geneva (ISSCO stands for Institute for Semantic and Cognitive
Studies). After his PhD at the University of Paris XI (LIMSI,
Orsay), he has been a postdoctoral fellow at the University of
California San Diego (Cognitive Science Department). His main
interests are natural language processing (reference resolution,
evaluation, language resources) and multi-agent modeling of
communication. He is involved in the "Co- evolution of Syntax and
Semantics" project (with UCSD, Sony CSL Paris, ISC Lyon). The
URL of his homepage is: http://www.issco.unige.ch/staff/andrei .
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue